ARCHNES
MASSAC USETTS INSTITCITE
OF rEcHNOLOLGY
Essays in Financial Economics
JUN 02 2015
by
LIBRARIES
Hong Ru
B.A., Fudan University (2008)
M.Fin., Massachusetts Institute of Technology (2010)
Submitted to the the Alfred P. Sloan School of Management
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June 2015
@ Massachusetts Institute of Technology 2015. All rights reserved.
Author .....
Signature redacted
the Alfred P. Sloan School of Management
April 29, 2015
Certified by....
Signature redacted
I
Antoinette Schoar
Michael Koerner '49 Professor of Entrepreneurial Finance
Thesis Supervisor
Accepted by ...
Signature redacted
Ezra Zuckerman
Director, Sloan School of Management PhD Program
2
Essays in Financial Economics
by
Hong Ru
Submitted to the the Alfred P. Sloan School of Management
on April 29, 2015, in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy
Abstract
This thesis considers three empirical essays on financial economics. The first chapter examines the effect of government credit on firm investment, employment, debt,
profitability, and survival by using unique data from the China Development Bank
(CDB). I explore the different effects of various types of government credit (credit to
infrastructure vs. credit to state-owned enterprises (SOEs)). I also trace the effect of
government credit across different levels of the supply chain. I find that CDB SOE
industry loans crowd out private firms in the same industry but crowd in private firms
in downstream industries. I also find private firms benefit from CDB infrastructure
loans. I use the exogenous timing of municipal political leaders' turnover as an instrument for CDB loans to cities. The second chapter, joint with Antoinette Schoar,
analyzes pricing and advertising strategies of credit card offers. We show that credit
cards which have reward programs have lower regular APR but rely more heavily on
backward loaded and more hidden payment features. Issuers target different reward
programs at different types of the population: Programs such as miles, cash back and
points are offered to richer and more educated customers, while low intro APR offers
are offered to poorer and less educated customers. Our results also suggest that card
features that are mainly demanded by sophisticated consumers cannot be shrouded
and need to be priced upfront. Finally, using shocks to the credit worthiness of customers, we show that card issuers rely more heavily on backward loaded credit terms
when customers are more protected. The third chapter studies the effects of privatization on both SOEs and privately-owned firms in China. Using political turnover as an
instrument variable for privatization, I find that after privatization, the productivity
of SOEs and private firms increases by 50% and 100%, respectively. Moreover, every
100 workers got fired by SOEs come with a 169 increase from private firms' hiring
in the same industry and same province. I also find that politicians' fixed effect on
SOEs is significant. Moreover, corrupt politicians make SOEs less efficient but more
powerful in the market.
Thesis Supervisor: Antoinette Schoar
Title: Michael Koerner '49 Professor of Entrepreneurial Finance
3
4
Acknowledgments
When I applied PhD programs five years ago, many people asked me: Are you sure
this is what you want to do? Today, I can tell them: Yes, it is one of the best choices
that I have ever made in my life. Along this exciting and arduous path, I am grateful
to so many people for their support and encouragement.
The first one I want to thank is Antoinette Schoar: she guides me step by step
and teaches me how to think as a researcher in finance. I benefit a lot from our many
long and engaging discussions. Her advice, support, and comments are invaluable. I
feel so lucky to have her as my mentor, and I can't thank her enough.
I am truly grateful to Nittai Bergman, Andrey Malenko, and Robert Townsend.
They always push me to think harder and deeper. They make me a better researcher,
and I am indebted to them. I also benefit a lot from discussions with Jean-Noel
Barrot, Asaf Bernstein, Yan Ji, Mark Kritzman, Deborah Lucas, Eric Maskin, Tran
Ngoc-Khanh, Stewart Myers, Stephen Ross, Felipe Severino, Daan Stuyven, Richard
Thakor, Wei Wu, Yu Xu ,and Luo Zuo.
My research in China has been benefited from many people; I learn a lot from
working with Gao Jian whose comments about China are extremely insightful. He
guides me toward the important research questions in China. I am deeply grateful
to have him as my mentor and friend. I also want to thank Jinglu Feng, Yong Liu,
Tianshan Meng, and Yue Wu from the China Development Bank for their generous
supports.
My family is always there for me. I want to thank my parents, Jianping Ru and
Weiping Xie, for giving me the support that I needed to chase my dream. They
guided me to where I am today. I also want to thank my son, Will, for all the joys
you bring to my life. He is truly the best thing that has ever happened to me.
Last, but definitely not least, I want to thank my lovely wife Wei Chen. Thank
you for always being there along the way and helping me to be all that I can be.
Without you, I couldn't achieve what I have today. You make me a better man.
5
6
Contents
1
Government Credit, a Double-Edged Sword: Evidence from the
China Development Bank
13
1.1
1.2
1.3
1.4
1.5
1.6
1.7
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Literature review ........
.............................
Background: the China Development Bank and Local Government
Financing in China . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3.1 History of the China Development Bank . . . . . . . . . . . .
1.3.2 Local Government Financing . . . . . . . . . . . . . . . . . . .
Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4.1 D ata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.4.2 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . .
Empirical Analysis and Results . . . . . . . . . . . . . . . . . . . . .
1.5.1 Firms' Response to CDB Loans . . . . . . . . . . . . . . . . .
1.5.2 Instrument: City Secretary's Turnover Timing . . . . . . . . .
1.5.3 Second Stage: CDB City Level Loan Analysis . . . . . . . . .
1.5.4 Politician's Other Channels to Affect Local Economy . . . . .
1.5.5 CDB Industry Loan Analysis . . . . . . . . . . . . . . . . . .
1.5.6 Overall Effects of the Government Credit from CDB . . . . .
1.5.7 Politicians' Incentives Behind the Borrowing Patterns . . . . .
C onclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2 Do Credit Card Companies Screen for Behavioral Biases?
2.1
2.2
2.3
2.4
2.5
2.6
Introduction . . . . . . . . . . . . . . . . . . . . . . . . .
Literature Review . . . . . . . . . . . . . . . . . . . . . .
Data and Summary Statistics . . . . . . . . . . . . . . .
2.3.1 Data Description . . . . . . . . . . . . . . . . . .
2.3.2 Descriptive Statistics . . . . . . . . . . . . . . . .
R esults . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.1 Customer Characteristics and credit card features
2.4.2 Pricing of Credit Cards . . . . . . . . . . . . . . .
2.4.3 Trade-off between Regular APRs and Late Fees .
2.4.4 Unemployment Insurance . . . . . . . . . . . . .
C onclusion . . . . . . . . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . .
7
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
13
17
19
19
21
22
22
23
26
26
27
29
32
34
37
39
43
44
77
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
77
82
83
83
85
87
87
89
95
95
97
98
3 Privatization, Politics, and Corruption
3.1
3.2
117
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Literature review and economic reform in China . . . . . . . . . . . .
3.2.1 Literature review . . . . . . . . . . . . . . . . . . . . . . . . .
117
119
119
3.3
3.2.2 History of Chinese SEOs' reform . . . . . . . . . . . . . . . .
Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
121
122
3.4
3.3.1 D ata . . . . . . . . . .
3.3.2 Matching Politicians to
3.3.3 Summary Statistics . .
Empirical Analysis . . . . . .
.
.
.
.
122
123
123
124
Privatization's effect on SOE itself . . . . . . . . . . . . . . .
Privatization's effect on private firms . . . . . . . . . . . . . .
124
126
3.4.1
3.4.2
3.5
3.6
3.7
3.8
. . . .
Firms
. . . .
. . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
128
3.5.1
3.5.2
3.5.3
Privatization's effects for SOEs . . . . . . . . . . . . . . . . .
Privatization's effect on private firms . . . . . . . . . . . . . .
Examination of exclusion condition and robustness check . . .
128
129
130
Influence of Politician in China . . . . . . . . . . . . . . . . . . . . .
3.6.1 Policitian's fixed effects . . . . . . . . . . . . . . . . . . . . . .
131
131
3.6.2
3.6.3
Corruptions and firm's performance . . . . . . . . . . . . . . .
Politicians' career path . . . . . . . . . . . . . . . . . . . . . .
132
133
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
134
135
8
List of Figures
CDB Outstanding Loan Amount and New Issuance
Political Turnover in China . . . . . . . . . . . .
Local Government Borrowing
. . . . . . . . . .
Manufacturing Firms in China
. . . . . . . . . .
SOEs' Fixed Asset Patterns
. . . . . . . . . .
SOEs' Employment Patterns
. . . . . . . . . .
Shifts of CDB Industry Loan Over Time . . . . . .
2-1
2-2
2-3
2-4
2-5
2-6
2-7
2-8
Monthly Time Trend of Reward
Monthly Time Trend of Size . .
Monthly Time Trend of Color
Monthly Time Trend of Picture
Waterfall Regression Coefficients.
Waterfall Regression Coefficients.
Simple Visual Credit Card Offer
High Visual Credit Card Offer
101
102
103
104
105
106
107
108
3-1
3-2
3-3
Manufacturing Firms in China.
137
138
139
.
.
.
.
.
.
.
.
.
.
.
1-1
1-2
1-3
1-4
1-5
1-6
1-7
.
.
Turnover and Privatization . . .
Estimated Hazard Functions . .
9
48
49
50
51
52
53
54
10
List of Tables
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
Summary Statistics Data . . . . . . . . . . . . . . . . . . . . . . . . .
CDB Loan's Effect on Firms: Evidence from OLS Regressions . . . .
Exogeneity of City Secretary Turnover Timing . . . . . . . . . . . . .
City Secretary Turnover Timing's Effect on Borrowing from CDB . .
City Secretary's Turnover Effects on Firms . . . . . . . . . . . . . . .
Opposing Effects of CDB City Infrastructure Loan and Industry Loan
(Use Secretary Year in Office as Instrument) . . . . . . . . . . . . . .
CDB Province Industry Loan's Effect on Firms within Same Industry
55
56
57
58
59
(2SLS)..........
62
...................................
61
Upstream Industry Loan's Effect on Firms in Downstream Industry
(2SL S) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.9
1.10
1.11
1.12
1.13
1.14
1.15
63
What Private Firms Benefit from CDB Upstream Industry Loans (2SLS)
Variables' Definition and Construction . . . . . . . . . . . . . . . . .
City Secretary's Turnover and Borrowing from CDB(off national-cycle)
What Firms are Crowded Out in Private Sector . . . . . . . . . . . .
CDB Loan's Effect on Firms (Use Secretary Year in Office as Instrument)
Turnover Timing's Effects on SOEs under Different CDB Loan Amount
CDB Loan and Firm Exit&Enter (Use Secretary Year in Office as Instrum ent) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.16 Politician's Other Channels to Affect Local Economy . . . . . . . . .
64
65
66
67
68
69
1.17 Where Does the CDB Industry Loan Go? . . . . . . . . . . . . . . . .
72
70
71
1.18 CDB Province Industry Level Loan and City Secretary Turnover (First
Stage)
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
1.19 City Secretary's Promotion and GDP Performance . . . . . . . . . .
1.20 City Secretary's Characteristics and Borrowing from CDB . . . . . .
1.21 City Secretary's Borrowing from CDB and City's Debt Level . . . . .
74
75
76
2.1
2.2
2.3
2.4
2.5
Summary Statistics Data . . . . . . . . . . . . . . . . . . . . .
Descriptive Statistics for Format Design of Credit Card Offers
Credit Card Features and Demographics . . . . . . . . . . . .
Relationship Between APRs/Fees and Reward Program . . . .
Mileage Program vs. Zero Introductory APR Program . . . .
2.6
2.7
Relationship Between APRs/Fees and Credit Card Offer Design . . .
Regular APR vs. Late Fees . . . . . . . . . . . . . . . . . . . . . . .
114
115
2.8
Unemployment Insurance and Credit Card Features . . . . . . . . . .
116
11
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. 109
. 110
. 111
. 112
. 113
3.1
Summary Statistics Data . . . . . . . . . . . . . . . . . . . . . . . . .
140
3.2
Hazard Ratio Analysis of Politicians' Turnover . . . . . . . . . . . . .
141
3.3
Does Privatization Improve SOEs' Efficiency? (OLS)
142
3.4
Effects of Privatization on SOEs' Efficiency (Using Political Turnover
Timing as Instruments) . . . . . . . . . . . . . . . . . . . . . . . . . .
143
3.5
3.6
Does Privatization Improve Private Firms' Efficiency? (OLS) . . . . .
Does Bigger Privatization Have Bigger Effects? . . . . . . . . . . . .
144
145
3.7
3.8
DID Regressions Results(Estimated Privatization) . . . . . . . . . . .
Exclusion Examination and Robustness Check . . . . . . . . . . . . .
146
147
3.9 2SLS Robustness Check . . . . . . . . . . . . . . . . . . . . . . . . .
3.10 Do Politicians' Individual Fixed Effects Matter? . . . . . . . . . . . .
148
149
. . . . . . . . . . . .
150
. . . . . . . . . . . . . . . . . . . . . . . . . .
151
3.11 Do Bad Politicians Extract Rents From Firms?
3.12 Politician Career Path
12
. . . . . . . . .
Chapter 1
Government Credit, a Double-Edged
Sword: Evidence from the China
Development Bank
1.1
Introduction
Government credit programs are pervasive in many countries around the world and
play an important role in capital allocation, especially to infrastructure investment
and state-owned enterprises (SOEs). For example, the United States established its
earliest federal credit program - the Farm Credit System - in 1916 to provide credit to
agricultural and rural America. This was followed by a surge in additional government
credit programs after the Great Depression. Some credit programs, such as the Rural
Electrification Administration and the Small Business Administration, are still in
place.' In 2010, the U.S. Government's outstanding commitments for loans and
guarantees totaled approximately $2.3 trillion. This was roughly one-third the size of
the loans of all the U.S. banks combined (Elliott (2011)). Outside the United States,
many countries have development banks that typically issue government credit. 2
The literature has outlined two opposing views of the effects of government di'For detailed histories of the Farm Credit Council, please go to www.fccouncil.com documents.
Rural Electrification Administration (REA) was the former agency of the U.S. Department of Agriculture that administered loan programs for electrification and telephone service in rural areas.
Reconstruction Finance Corporation (RFC) was a U.S. government corporation started in 1932.
It provided financial support to state and local governments, and made loans to banks, railroads,
mortgage associations, and other businesses. RFC was the predecessor of the Small Business Administration. Small Business Administration (SBA) was founded in 1953 to provide loans, loan guarantees, contracts, counseling sessions, and other forms of assistance to small businesses. In 2007, the
National Infrastructure Reinvestment Bank was proposed, later backed by President Obarma.
2
For example, Germany formed KfW Bankengruppe in 1948, a German government-owned development bank based in Frankfurt. Korea founded Korea Development Bank in 1954, which is a
state-owned policy bank in South Korea. There are also many multilateral development banks such
as the African Development Bank, the Asian Development Bank, the European Bank for Reconstruction, and Development and the Inter-American Development Bank Group. In the U.S., the
National Infrastructure Reinvestment Bank was proposed in 2007.
13
rected credit. On the one hand, government credit can be justified by the existence
of credit market failures. Private banks may not allocate funds to high social return
projects with positive externalities if the returns are difficult to be captured (Stiglitz
(1993)). Prime examples are infrastructure investments such as highways or airport
constructions. Private firms could benefit from these projects' positive externalities.
On the other hand, government credit could crowd out private sector investment, especially when the credit is given at below market rates or to firms that have distorted
incentives (e.g., Demirguc-Kunt and Maksimovic (1998), King and Levine (1993a,
1993b), Rajan and Zingales(1998)). A concern is that inefficient established firms
are subsidized (e.g., SOEs), while more efficient firms are crowded out (e.g., private
firms), which harms the economy in the long run.
Empirical studies have shown mixed results on whether government credit 3 crowds
out private sector investment and growth or whether it encourages private sector
(crowd in) (e.g., Blanchard and Perotti (2002), Cohen et al. (2011), Gale (1991)). Due
to limited data, these studies usually explore only the effects of aggregate government
credit. In this paper, I am able to add to the prior literature by analyzing the role
of different types of government credit (infrastructure loans vs. industry loans to
SOEs) in the context of China. In addition, I am able to trace the effect of credit
across SOEs and private firms, as well as across different levels of the supply chain.
I find that government loans to SOEs crowd out private firms in the same industry,
but interestingly they crowd in private firms in downstream industries. Moreover,
infrastructure loans appear to have positive effects on private firms. These opposing
effects may shed light on the mixed results from previous studies.
I study this issue based on a unique industry-level loan data set from the China
Development Bank (CDB) for the period 1998 to 2013. I combine it with a firm-level
panel data set from the Chinese Industry Census, which contains all manufacturing
firms with annual sales of more than 5 million RMB ($700K) from 1998 to 2009. The
China Development Bank is one of three policy banks with a mandate to provide
credit support to infrastructure projects and SOEs in basic industries, primarily by
lending to local governments. 5 The CDB loan data contain city-level, aggregate, outstanding loan amounts and issuances. Loans are categorized into infrastructure loans
and industry loans to SOEs. The data set also contains province-level, aggregate,
outstanding loan amounts and issuances, which are categorized into 95 industries.6
I first use a simple OLS regression framework. I find that CDB industry loans
are associated with larger investment in assets, greater employment, and more debts
3 Government
credit can also be viewed as government spending (Lucas (2012b)). CDB loans are
subsidized. In China, CDB lending can be viewed as an extension of government fiscal policy.
'I am the visiting scholar in the China Development Bank and have access to their internal data
system. This internal data is compiled in CDB headquarter from detailed monthly loan reports of
each CDB branch.
5The CDB usually coordinates with local governments and lends to infrastructure projects and
SOEs in the province or city. Most of CDB's loans are lent via local governments, which are implicitly
or explicitly responsible for these loans. See more details in Section 1.3.
6
These industries include both infrastructure sectors (e.g., road transportation, water supply,
public facilities) and industry sectors (e.g., agriculture, tobacco, software, oil refining, textile). See
more details in section 1.4.
14
.
of SOEs that receive these CDB loans. On the other hand, the amount of CDB
industry loans to SOEs is negatively correlated with investment, employment, total
sales ,and sales per worker of private firms in the same industry7 . In contrast to the
industry SOE loans, the CDB city infrastructure loan amount is positively correlated
with private firms' assets, debt, total sales, and sales per worker. Doubling the CDB
industry loans is associated with a 0.6% increase in SOEs' assets and a 0.3% decrease
in private firms' assets
While the effects are significant, there is a potential endogeneity concern. Government credit might flow into areas or industries with especially high or low growth
potential. Moreover, in China, it is the local government that borrows from the CDB
for infrastructure projects and SOEs. The local governments, which enjoy closer relationships with the CDB, may be granted more loans. To address these concerns, I
use the exogenous timing of political turnovers as a way to identify the causal effects
of government directed credit.'
In particular, I use the timing of municipal government turnovers as an instrument
for CDB loans. In China, the city secretary (equivalent to a mayor in the U.S.) is
replaced every five years on average. This turnover is decided not by an election but
by a higher level communist party official. About half of the cities have the same
five-year turnover cycle as the National Congress of the Communist Party. But some
cities have different turnover cycles if they were newly founded during the 1990s and
started on different cycles due to the year they were founded9. The city secretary
plays a key role in CDB loan allocations. Infrastructure loans are usually lent to
municipal governments directly. Industry firm loans are often lent to SOEs, and
municipal governments are deeply involved in and are responsible for arranging these
loans.
To verify the instrument, I begin by testing the exogeneity of city secretaries'
turnover timings. I find that turnover timing does not correlate with past economic
performance. I conduct first-stage regressions by regressing cities' borrowing from
the CDB on the number of years the city secretary has been in the city. I control
for city-fixed and politician-fixed effects to mitigate concerns that the CDB might
lend more to the cities or politicians that have better political connections. I find
that most cities borrow significantly more during the first year of the secretaries'
terms, and monotonically decrease borrowing during later years. The borrowing of
the city goes up again when a new secretary comes in. On average, secretaries reduce
total loan amounts by 36.4% each year during their tenure in office. I find similar
borrowing patterns in both infrastructure and industry loans. Moreover, I also control
7The majority of the CDB industry credit goes
to SOEs. These loans are subsidized. On average,
the interest rates of the CDB loans are 100bps below the market rates.
8
Previous studies use the exogeneity of political variables (e.g., political cycle, political competitiveness and politicians' seniority) to overcome endogeneity of the government credit or spending
(e.g., Bertrand, Kramarz, Schoar and Thesmar (2007), Carvalho (2013), Cohen et al. (2011), Dinc
(2005)), Sapienza (2004)).
9
More than 100 new cities were converted from villages during the 1990s. For example, Fuyang,
a city in northwestern Anhui province in China, was founded in 1996. Its turnover cycle started
from 1996, and the next secretary began in 2000. This is dissimilar from the national cycle.
15
for year-fixed effects to take out the macro time trend of the national turnover cycle.
When I select the "off-national cycle" cities 0 , I find that these borrowing patterns
still exist. This further shows that the part of the variation from turnover timing is
due to the different cycles among different cities, rather than the national turnover
cycle only. I also find the secretaries' chances of promotion are associated with more
borrowing from the CDB in their early terms. This is in line with the hypothesis that
the political ambitions of the secretaries drive these borrowing patterns.
In the second-stage regressions, I regress firm-level dependent variables on the
estimated CDB city-level loan amounts. Consistent with OLS results, I find that
increasing CDB city-level, aggregate, outstanding loans led to increases in SOEs'
fixed assets, employment, and debt. Conversely, increasing CDB industry SOE loans
led to decreases in private firms' fixed assets, sales, and sales per worker. Moreover,
when CDB loans increase, the private sector has fewer firms," and higher ROA
private firms are crowded out. On the other hand, I find that increasing CDB city
infrastructure loans led to increases in private firms' fixed assets, employment, debt,
and total sales.
The CDB loans' opposing effects on private firms and SOEs also alleviate the
concern that the timing of city secretary turnovers is driven by changing investment
opportunities in a city. I further document that other channels through which the
city secretary may influence the business in a city are not correlated with the turnover
cycles (e.g., enforcing tax treaties better, raising money by requesting more transfers,
and selling more land). Moreover, I find that for cities with no CDB loans, turnover
timing has no effect on firms. These findings support my hypothesis that turnover
timing provides exogenous variation of borrowing timing. The findings also help
reduce the concern that alternative channels may play a major role.
To explore the heterogeneous effects of government credit on private firms at different levels of the supply chain, I analyze province-level loan data in 41 manufacturing
industries. I again use city-level turnover timing as the exogenous shock and match
it at the province level. I identify the number of years the city secretary has been
in the city and match this data to the city's biggest SOE industry at province level.
If the city secretary is in the early term (within three years of entering the office), I
consider it a shock to CDB province-level loans to the industry (i.e., the largest SOE
industry of the city). In the first-stage regression, I find that province-level CDB
loans in an industry are 33.2% higher if the corresponding city secretary is in the
first three years of a term. In the second-stage regressions, I find that increases in
CDB loans led to decreases in fixed assets, employment, debts, and sales of private
firms within the same industry and same province. SOEs in the same industry and
province experience increases in fixed assets, employment, debts, and sales. Therefor,
CDB industry loans crowd out private firms in the same industry and crowd in SOEs.
10I exclude the cities whose
turnover are at the same time as the national turnover cycle(year
1998, 2003 and 2008). The remaining "off national cycle" cities have different turnover cycles than
the national cycle.
" Firm-level data used in this paper record all manufacturing firms with annual sales of more than
5 million RMB ($700K). If a firm drops from the sample, it could be a bankruptcy or the firm's
sales was below the threshold. It is also the same for firm enters.
16
Finally, I study CDB industry loan effects on upstream and downstream industries,
and I use the input-output matrix to identify inter-industry relationships. For each
industry, I pick its largest intermediate input from other industries as an upstream
industry. I find that increasing CDB loans to the upstream industry led to increases
in downstream private firms' fixed assets, debts, and sales. Evidence also suggests
that private firms with better political connections benefit significantly more from
these CDB upstream industry loans. In sum, although CDB industry loans crowd
out private firms within the same industry, they crowd in private firms in downstream
industries.
To calculate the overall effects of the CDB loans on individual firms, I multiply
the growth of different types of CDB loans (e.g., infrastructure loans, industry loans,
and upstream loans) by the estimated coefficients. The total effect of CDB loans
(including both infrastructure and industry credit) on annual asset growth of private
firms is 3.4% on average. Annual asset average growth of private firms was 15.4%
from 1998 to 2009. Therefore, credit from the CDB contributes 22% of the growth
in the private sector. Moreover, the total CDB credit increases private firms' sales
per worker by 10.8% annually. However, CDB industry credit to SOEs has negative
overall impacts on private firms. On average, a $1 million increase in the CDB's
outstanding industry loan amount led to a $0.52 million decrease in private firms'
total assets.
The remainder of the paper is organized as follows: Section 1.2 is the literature
review. Section 1.3 provides the history of the CDB and local government debt in
China. Section 1.4 describes the data. Section 1.5 gives the empirical analysis and
presents the results. Section 1.6 concludes.
1.2
Literature review
This paper relates to literature that examines whether government credit and spending crowds in or out the private sector. On the one hand, Atkinson and Stiglitz
(1980) suggests the "social view" that SOEs can be justified under market failures
(e.g., Stiglitz and Weiss (1981) and Greenwald and Stiglitz (1986)). Stiglitz (1993)
argues that government directed credit can fund projects with high social returns
where private banks might not allocate the funds. On the other hand, government
credit could crowd out private sector investment, especially when the credit is given
at below market rates or to firms that have distorted incentives. King and Levine
(1993a, 1993b) explore the relationship between financial development and growth.
They find that credit to SOEs is associated with lower growth of GDP, capital stock,
investment, and lower efficiency. Rajan and Zingales(1998) find the fraction of domestic credit going to the private sector is strongly correlated with market capitalization
to GDP. One explanation is that government involvement in industry crowds out the
private sector. Bertrand, Schoar and Thesmar (2007) suggests that government intervention in banking may create implicit barriers to entry and exit in product markets
by subsidizing poorly performing established firms. Many studies shows mixed evidence on whether government credit crowds out or crowds in the private sector. Gale
17
(1991) numerically estimates the effects of federal lending, suggesting credit subsidies
are costly and raise private investment slightly. A survey from Schwarz (1992) shows
limited evidence regarding the impact of U.S. credit programs on growth. Craig et
al. (2007) find a small correlation between SBA loan guarantees and local economic
growth. Shaffer and Collender (2009) find that total aggregate federal funding associates with significantly faster employment growth, but also with volatile incomes.
In this paper, government credit from the China Development Bank can be treated
as fiscal spending. Lucas (2012b) views federal credit programs as fiscal policy since
credit subsidies are costly, and affect pricing and allocation in credit markets. For
government spending, there has been a long debate between Keynesian and neoclassical' 2 theories. Vector auto regression on macro data is the standard method to
study the effects of government spending shocks (e.g., Blanchard and Perotti (2002),
Caldara and Kamps (2008), Fatas and Mihov (2001), Gali et al. (2007), Rotemberg
and Woodford (1992), Ramey (2008)). The literature also explores exogenous shocks
to government spending. Ramey and Shapiro (1998) use U.S. military buildups as
exogenous shocks, finding that product and consumption wages fall after a military
buildup. Burnside et al. (2004) use exogenous changes in military purchases as a
fiscal shock, suggesting real wages decline and tax rates increase after the shock, and
investments rise in the short term. This paper is closely related with Cohen et al.
(2011). They use changes in congressional committee chairmanships as an exogenous
shock to federal expenditure in states, finding that increasing fiscal spending makes
firms reduce investments in new capital and R&D, and increase pay outs to shareholders. They aggregate government spending at the state level. However, we still know
little about the effects of government credit or spending. One problem is that there is
no detailed analysis on the impact of government credit in individual industries. This
paper is based on far more detailed industry categories, which include 95 industries
in China (e.g., farming, livestock, food, beverage, tobacco, non-ferrous metals mining
and processing, software). I am able to separate the crowding out and crowding in
effects of government credit by looking at different levels of the supply chain.
This paper also relates to literature that studies political influences on government spending or credit. The political view assumes politicians have political and
personal goals that conflict with social welfare maximization (Kornai (1979), Shleifer
and Vishny (1994)). Political business-cycle literature started with Nordhaus (1975)
and McRae (1977), and was followed by Alesina and Sachs (1988) to model how the
government uses economic policies to influence elections under democratic political
systems. Many recent empirical studies support this view. Sapienza (2004) finds
that Italian state-owned banks charge low interest rates in a province in which the
associated party is stronger. Dinc (2005) uses cross-country data to demonstrate
that government-owned banks increase lending during election years in comparison
to private banks. Dinc and Gupta (2011) show that in India, the government delays
privatization of SOEs in regions with more political competition. Carvalho (2013)
2
Aschauer and Greenwood (1985), Barro (1981, 1989), Baxter and King (1993), Finn (1995), Hall
(1980), and Mankiw (1987) use dynamic general equilibrium mode to study the effects of government
spending.
18
uses Brazilian data and finds that politicians influence elections with bank lending. Bertrand, Kramarz, Schoar and Thesmar (2007) show that politically connected
CEOs in France create more jobs in politically contested areas, especially during election years. Khwaja and Mian (2005) find that government banks in Pakistan favored
politically connected firms by providing greater access to credit. Connected firms
received 45% more loans and had 50% higher default rates on these loans. Private
banks show no such political bias. However, there is little empirical evidence of the
political view in countries without elections. In these countries, politicians are usually much more powerful. In China, government credit affects firms through local
government debt instead of a firm's direct political connections with CDB. Municipal
government plays a key role in CDB loan allocations.
1.3
1.3.1
Background: the China Development Bank and
Local Government Financing in China
History of the China Development Bank
The China Development Bank was founded in 1994 from six SPC Investment Corporations 13. The CDB, along with Export-Import Bank of China and Agricultural
Bank, were assigned as three policy banks during financial system reform in 1994.
CDB investment covers basic industries and the infrastructure sector. In 2008, the
CDB became a corporation with Ministry of Finance (MOF) and China Investment
Corporation14 as its two shareholders. Although the CDB is now a corporation, it
can still be viewed as an extension of the government's fiscal function.
CDB's funding is largely from bond issuances. At its establishment in 1994, the
CDB had $6.3 billion capital. The CDB is entitled to receive disbursements from
stage budgets, repayments of principal and interest, and fiscal subsidies arranged
from the state budget to national projects. In addition, the CDB was entitled to
issue financial debentures to state-owned financial institutions. Commercial banks in
China are the largest buyers of CDB bond. The total volume of financial debentures
issued by the CDB are decided jointly by PBOC and SPC based on yearly credit
and fixed-asset investment plans. Interest rates of financial debentures are decided by
PBOC in consultation with SPC and MOF. The CDB had explicit guarantees from
the central government until 200816. The guarantee gives the CDB the advantage to
"State Planning Commission (SPC) is a macroeconomic management agency under the Chinese
State Council that has broad administrative and planning control over the Chinese economy. These
six Investment Corporations were policy institutions, established in late 1980s, affiliated directly with
the State Planning Commission and functioning as long-term investment instruments on behalf of
the government.
"China Investment Corporation is a sovereign wealth fund responsible for managing part of the
People's Republic of China's foreign exchange reserves.
"The People's Bank of China is the central bank of the People's Republic of China, with the
power to control monetary policy and regulate financial institutions in mainland China.
16
The guarantee became implicit in 2008 when the CDB changed its corporate governance from
SOE to a corporation. However, the interest rate and the issuance volume is unaffected by this
19
finance the bond market cheaply, and its bond interest rate is slightly higher than
the treasuries in China. By the end of 2013, the CDB had approximately $1.3 trillion
outstanding loans, and most are financed through bond issuance. As shown in Figure
1-1, CDB loans started to increase in 2003, and grew over time, especially from 2008
through 2010. At the end of 2013, the CDB has 5.8 trillion RMB ($0.95 trillion)
outstanding domestic loans and 1 trillion RMB ($0.16 trillion) overseas loan.
The bank's lending rates were regulated by the Central Bank before 2008. Between
2008 and 2013, the lending rate was required to sit within a range of a referred rate
set by the Central Bank. This range was expanded over the years until mid-2013.
After July 20, 2013, the lending rate was liberalized. However, the deposit rate is
still controlled by the government. The CDB has been regulated like other stateowned commercial banks, but in practice, CDB's long-term loan rates have been
lower than those of state-owned commercial banks, and much lower than private or
shareholding commercial banks, because 1) the CDB is less profit driven, and 2)
CDB's administrative costs are lower than commercial banks". CDB's loans can be
viewed as subsidized loans or government spending.
The CDB is fully state-owned, just like other state-owned commercial banks such
as ICBC, CCB, BOC and ABC 18 , but its behavior is different. CDB's business
usually covers infrastructure sectors and uncontested markets in which other stateowned commercial banks have little interest. This phenomenon might due to three
reasons. First, CDB's policy mandate pinpoints the bank in such policy-related areas.
Second, the CDB finances its loans by issuing long-term bonds with sovereign ratings,
whereas other state-owned commercial banks rely primarily on short-term deposits.
Therefore, the CDB conducts long-term lending that not only caters to infrastructuresector requirements, but also matches its assets and liabilities durations. Third,
CDB's managers and its elites used to work in central government agencies, such as
NDRC'" or SPC, MOF, PBOC, etc., whereas those of commercial banks are lifelong
bankers. Such career background disparities might led to differences in respective
visions, political awareness, and behavioral preferences, which affect a bank's business.
Generally, local governments initiate loan applications by submitting project proposals to the CDB. The CDB decides whether to accept or reject them. Allocation of
the CDB loans depends on many determinants. First, based on the central government's yearly credit plan, the CDB gives each province an annual credit quota. For
example, if the total credit amount increases by 20%, each province can also increase
change. The CDB is still considered an extension of the government's fiscal function.
1 7 Unlike commercial bank, which usually have branches in all cities and villages in China, the
CDB has branches only at the province level. One reason is that the CDB does not need local
branches to attract depositors. Most of CDB's money is from bond issuances. The CDB usually
focuses only on projects at city or province levels, and it does not invest often at village level or
below.
8
1 ICBC stands for Industrial and Commercial Bank of China; CCB for China Construction Bank;
BOC for Bank of China; ABC for Agricultural Bank of China.
19The National Development and Reform Commission (NDRC) of the Government of the People's
Republic of China, formerly State Planning Commission and State Development Planning Commission, is a macroeconomic management agency under the Chinese State Council, which has broad
administrative and planning control over the Chinese economy.
20
loans by 15% from the previous year. Second, the CDB keeps part of the quota,
and has flexibility to allocate credit based on other determinants such as supporting
national projects, political connections with local governments, how hard local governments lobby, etc. Local governments usually compete for CDB loans, especially
in recent years. Although there is a general rule on credit allocations, final loan
issuances do not necessarily accord with initial plans.
1.3.2
Local Government Financing
Since 1989, local governments in China have been prohibited from incurring debt by
budgetary law. As a result of tax-sharing system reform between local and central
governments in 1994, the central government takes the majority (around 70%) of tax
revenue. For example, the central government takes 60% of personal and corporate
income taxes and 75% of value-added tax. At the same time, local governments bear
the responsibility of infrastructure development but do not have the money to do so.
With a thorough understanding of local governments' desires to boost local economies
and develop local infrastructures, the CDB established direct connections with local
governments and helped them create borrowing platforms by creating 100% stateowned companies. Local governments are then able to use these companies to borrow
from banks and issue bonds legally. The CDB usually commits a certain amount of
loans to fund infrastructure development within a certain period (normally 5 years).
The CDB has several advantages to support local governments. First, it is mandated
to invest in infrastructure sectors and pillar industries. Second, different from other
banks, the CDB has special long-term funding resources through CDB bond issuances
in domestic bond markets.
Before 2008, the CDB was the primary resource for local-government financing,
especially concerning long-term borrowing. In November 2008, along with 4 trillion
RMB ($586 billion) stimulus plan2 0 , commercial banks started to lend to local governments aggressively. These were usually short-term loans (1 to 3 years). In 2010,
the central government decided to pull back the stimulus plan. As a result, many
commercial banks either stopped lending or rolled loans over to local governments.
However, local governments' investments are usually long-term (such as infrastructure
investments), and they needed loans to continue projects begun under the stimulus
program. Therefore, after 2010, many local governments started issuing bonds and
borrowing from shadow banking systems in China. The CDB is still a long-term, stable finance resource for local governments. This paper focuses on the period from 1998
to 2009, which overlaps with the stimulus plan by only one year. During the sample
period, the CDB played the most important role in local government borrowing.
20
The 4 trillion (US$ 586 billion) stimulus plan was announced by the State Council of the People's Republic of China on November 9, 2008 to minimize the impact of the global financial crisis.
The central government also ordered financial institutions (mainly commercial banks) to lend certain amounts within a limited period. Commercial banks started to increase lending dramatically,
including lending to local governments.
21
1.4
1.4.1
Data Description
Data
This paper uses three data sets: (1) The China Development Bank, (2) The Chinese
Industry Census (CIC), and (3) The Zechen and Baidu Encyclopedia Database.
CDB data contain both city- and province-level loan data. At the city level,
they record yearly aggregate CDB outstanding loan amounts and loan issuances to
both infrastructure projects and industry SOEs from 1998 to 2010, across 310 cities
in China2 1 . City-level data were collected by a CDB internal survey in 2010, on
which branch managers at the province level manually categorized projects into various cities. At the province level, the CDB data set contains monthly aggregate
CDB outstanding loan amounts and loan issuances in 95 industries for each of the
31 provinces 22 from 1998 to 2013. The industries include infrastructure sectors (e.g.,
road, air, rail transportation, water supply, public facilities, etc.) and industry sectors (e.g., agriculture, tobacco, software, oil refining, textile, etc.). In total, there
are 27 infrastructure sectors and 68 industry sectors. Table 1.1 Panel B lists some
large infrastructure sectors such as road construction, railway, water systems, and
telecommunications. City-level loan data do not include province-level projects such
as highways. CDB has only provincial branches, and each branch is required to report
project information to headquarters at the end of each month. A CDB central server
compiles the provincial data and updates them monthly. I use annual data during
analysis. City- and province-level economic variables (e.g., GDP, income per capita,
total employment, and fiscal income) are from the China Statistical Yearbook.
The Chinese National Bureau of Statistics (NBS) collected the Chinese Industry
Census (CIC) data, including all manufacturing firms in China with annual sales over
5 million RMB (about $700,000) from 1998 to 2009. It has detailed annual accounting
data and firm characteristics such as number of workers, industry categories, locations,
registration types, political hierarchies, government subsidies, wages, etc. In total,
there are 711,892 firms in China. CIC appears to be the most detailed database
on Chinese manufacturing firms, and the content and quality of the database are
sufficient. Using firm registration type from CIC data, I classify firms as SOE, private,
or foreign firms. Location data in CIC is an 11-digit number that locates the firm
at the street level. I cut the first 4 digits to identify the city. Industry codes are
the standard 4-digits, and I cut the first 2 to match CDB industry codes. There was
a change in industry codes by Chinese National Bureau of Statistics in 2002, so I
adjusted the industry codes to 2002 standard.
Regarding the data set from politician profiles, I manually collected it from the
Zechen Database, which records all mayors and secretaries of municipal committees
in each city from 1949 through 2013. I collected a name list of mayors and secretaries,
1They do not include Beijing, Shanghai, Tianjing, and Chongqing, which are classified as
provinces.
22
In China, there are 27 provinces plus Beijing, Shanghai, Tianjing, and Chongqing that are under
direct control of the central government. Among these 95 industries, CDB added 11 new industries
in 2005 and doesn't have data before 2005.
22
and their terms in office, at the monthly level. I also collected data for members of
provincial committees of the Communist Party of China. These data cover all the
334 cities and 31 provinces in China. Based on the name list, I searched politicians'
profiles from the Baidu Encyclopedia database, a Chinese-language, collaborative,
web-based encyclopedia provided by the Chinese search engine Baidu. The Encyclopedia is the best Chinese online encyclopedia, and generally provides clear profiles
of prominent people, better than official public profiles of politicians. However, the
quality of politicians' profiles varies among cities, especially for small cities. To compensate, I uses Xinhua News 2 3 as a supplementary source to crosscheck data from the
Baidu Encyclopedia. Final profile data includes 1,227 city secretaries and 97 provincial governors. Each profile had a politician's gender, age, and birthplace. Some
politicians had the same name, which is common in China. To overcome this limitation, I conducted a thorough double check for politicians with the same name, and
distinguished them with a separate ID number. When I merged the CDB city-level
data with politicians' profiles, there were 310 cities in total (the remaining 24 cities
did not have CDB loans).
1.4.2
Summary Statistics
Table 1.1 presents summary statistics, and Table 1.10 in the Appendix contains a
detailed definition and construction of each variables. CDB city-level loan data include 310 cities from 1998 to 2010. The data separate loans into two categories:
infrastructure and industry. The top 2 panels of Figure 1-1 show that the total citylevel, outstanding loan amount increased from 321 billion RMB ($40 billion) in 1998
to 2,811 billion RMB ($433 billion) in 2010. Total loans for infrastructure increased
from 27.4 billion RMB in 1998 to 1,143 billion RMB in 2010. Industry loans increased
from 293.6 billion RMB in 1998 to 1,659 billion RMB in 2010. Overall, industry loans
are bigger than infrastructure loans at the city level, but infrastructure loans grew
faster. The bottom two panels of Figure 1-1 shows the ratio of city level infrastructure
loan and industry loan to total loan amount respectively. For infrastructure loan, the
ratio was almost 0 from 1998 to 2000 which is consistent with the top 2 panels of
Figure 1-1. Moreover, the gap between top and bottom quartiles enlarged from 1999
to 2003 and closes a bit after 2003. It means that different cities have very different
combinations of infrastructure and industry SOE loans. One reason is that city secretaries tend to help the local SOEs to borrow from CDB. I find industry loan amounts
are significantly higher in cities with more SOEs. These cities with big state-owned
sector borrow relatively less in infrastructure. For new loan issuances at the city level,
the amount increased from 98 billion RMB in 1998 to 1,066 billion RMB in 2010. On
average, annual loan issuances were about one-third of the outstanding loan amount.
One feature of CDB loans is that both cash inflows and outflows are huge because
investment projects are huge, and loans are often rolled over.
"Xinhua News is the official press agency of the People's Republic of China, and the largest center
that collects information and press conferences in China. It is also the largest news agency in the
country.
23
CDB province-level loan data covers 31 provinces and 95 industries from 1998 to
2013. The middle 2 panels of Figure 1-1 show that total province-level outstanding
loan amounts increased from 474 billion RMB ($59 billion) in 1998 to 5,888 billion
RMB ($935 billion) in 2013. Total loans for infrastructure increased from 124 billion
RMB in 1998 to 3,569 billion RMB in 2013, and total loans for industry increased from
326 billion RMB in 1998 to 2,205 billion RMB in 2013. Overall, industry loans were
larger than infrastructure loans in 1998, but infrastructure loans grew much faster
and surpassed industry loans. For new loan issuances, the total amount increased
from 142 billion RMB in 1998 to 1,796 billion RMB in 2013. There was a drop in
loan issuances in 2012. In comparison to city-level loans, province-level loans were
larger than infrastructure projects. If a loan was for a provincial project such as
highway construction, CDB did not break it down and assign it to various cities.
Instead, it recorded it at the province level. From middle right panel of Figure 1-1,
there was a large jump in infrastructure new loan issuances from 2008 to 2009, due
to a four trillion RMB ($586 billion) stimulus package beginning in November 2008.
After 2010, there was a decrease in infrastructure loan issuances, due primarily to a
pull-back of the package.
Chinese Industry Census data contain 711,892 firms in the manufacturing sector.
Each has a unique registration number and company name. The registration number is a unique number to a company from SAIC 21. The registration number of a
company that no longer exists is recycled. I use registration numbers as IDs for the
companies. Unfortunately, CIC data do not record registration IDs for 2008 and 2009.
I used a name-matching algorithm2 5 and recovered 90% of registration IDs. CIC also
has registration types for the company, which depends on a company's shareholders.
I used registration type to categorize companies into 3 groups: state-owned enterprises(SOEs), private companies, and foreign companies. SOEs are defined as stateowned or collective-owned enterprises. Private firms are defined as private-owned
enterprises, private partnership enterprises, private limited liability company, or private limited company. If a firm's shareholders are mainly foreigners, it is classified as
foreign firms. I exclude the firms from CIC data set if it has mixed ownerships (e.g.,
half private owned and half state owned). From Table 1.1, the average ROA was 9%,
and the average number of employees was 108. Average ROA for SOEs was 5.2%
and the employee number was 353. In China, SOEs are less efficient and have more
employees. Tax_Corp is the effective annual corporate tax for each firm, missing if a
firm's income before tax was negative. TaxVAT is the effective annual value-added
tax rate of each firm. CIC recorded value-added tax in only 1998, 1999, 2000, 2003,
2005, 2006, and 2007. The average corporate tax rate was 19.41% and value-added
tax rate was 15.09%, consistent with real tax rates 26. Figure 1-4 shows the time
2
1State
Administration for Industry and Commerce of the People's Republic of China
matched the name of each company in 2008 and 2009 with company names from 1998 to 2007
that had registration IDs. I grouped companies by city and industry to increase matching probability
and accuracy.
26
The current corporate tax rate is 25%. For small businesses, it is 20%. For high-technology
firms or other firms supported by the government, the corporate tax rate is 15%. Value-added tax
rate varies between 13% and 17%.
21I
24
.
trend of firm number and aggregate assets for SOEs, private firms, and foreign firms.
The number of private firms increased dramatically from 1998 to 2009. The number
of SOEs decreased over time, due primarily to privatization. Foreign firm numbers
also increased slightly. For aggregate assets, the private sector again increased most,
surpassing foreign and SOE sectors. SOEs' assets decreased from 1998 to 2002 and
increased gradually after 2002. From 1998 to 2009, the private sector contributed
greatly to China's double-digit economic growth, and on average, it had smaller firm
sizes than SOEs and foreign companies. After matching CDB province industry data
with CIC firm data, there were 41 industries in the manufacturing sector.
In China, the political leader in a municipal government is called the Secretary of
Municipal Committee of the Communist Party of China (equivalent to the mayor in
the U.S.). For political turnover, the top panel of Figure 1-2 is a histogram of city
secretaries' term lengths from 1997 to 2013, among the 334 cities. I rounded monthly
turnover data into year frequency using June as the cutoff". Forty-three percent of
city secretaries left the city in their fifth year. In some cases, city secretaries left before
the fifth year, and in fewer cases, they remained. The national political turnover cycle
is also 5 years, and it occurs around the National Congress of the Communist Party
of China (CPC). Four national CPC congresses occurred in the data from 1997 to
2013: 1997, 2002, 2007, and 2012. Since the National Congresses of the Communist
Party usually occur at the end of these years, the turnover could occur the next year.
The middle panel of Figure 1-2 is a histogram of turnover years, concentrated in years
1998, 2003, 2008, and 2013. These years are one year after the National Congress of
the Communist Party. However, some special cases appeared such that turnover did
not occur every 5 years, and did not occur during a National Congress. The reason
is because China experienced fast urbanization, and many cities were incorporated
during 1990s. A city secretary's first tenure is usually 5 years, but the starting year
might not have coincided with the national cycle. The bottom panel of Figure 12 shows the sub-sample of secretaries whose term lengths were not 5 years (e.g., 4
or 6 years). Turnover more likely occurred around National Congresses, suggesting
provincial governments in China wanted to bring cities with different cycles back to
the national cycle, so a few cases had 4- and 6-year terms. One explanation is that the
turnover of provincial governors accorded with national cycle. New governors usually
want to have their own people to be city secretaries. In the politicians' profile data,
the average age of city secretaries is 50 years; the youngest was 32 and the oldest
62. The legal retirement age for males is 60, and 55 for females. "Promotion" is a
dummy variable for whether a city secretary was promoted during turnover. 38% of
city secretaries were promoted after their terms ended. Promotion is defined as being
appointed to a higher political hierarchy in the government 28
2
1f secretaries left before June and their successors succeeded them before June, I considered the
successor the city secretary for the entire year. If the secretaries left after June, I considered them
the city secretary for the entire year.
28
City secretaries' political hierarchies vary across cities. Secretaries from Beijing, Shanghai,
Tianjing, and Chongqing are at ministerial level. Fifteen cities in China are at vice-ministerial level:
Dalian, Qingdao, Ningbo, Xiamen, Shenzheng, Haerbing, Changchun, Shenyang, Jinan, Nanjing,
Hangzhou, Guangzhou, Wuhan, Chengdu, and Xian. Others are at the departmental level. I do not
25
1.5
1.5.1
Empirical Analysis and Results
Firms' Response to CDB Loans
There has been a long debate on the effects of government credit, especially for
the private sector. Public goods such as infrastructure help economic growth since
firms benefit from the infrastructures around them. However, government credit also
crowds private sectors through tax channel, interest rate channel, etc. In China, tax
and interest rate channels are shut off; the cities and even the provinces cannot change
tax rates or categories. Only the central government can determine taxes. Although
some tax treaties are flexible and local governments use them differently, the central
government sets the rules for tax treaties, and manipulation is limited. Interest rates
were not liberalized until July 20, 2013, after which banks determined their own
interest rates on loans, but deposit rates were still controlled by the central bank
strictly. The primary channel of crowding in China comes from competition between
SOEs and the private sector. Chinese SOEs are important to many industries, even
after the privatization wave from 1998 to 2005. Even today, SOEs control significant
shares of assets (Figure 1-4). The majority of CDB's industry loans go to SOEs, and
only a small share goes to the private sector, and then only to large, powerful private
firms. It is rare that CDB lends money to small private firms such as entrepreneurs.
I begin the analysis by OLS regressions and explore the correlations between
CDB loan amount and a firm's asset investment, employment, borrowing as well as
firm's performances such as ROA, total sales and sales per worker. I use the total
outstanding loan amounts instead of new issuances because many new loan issuances
are for old loans' rollover. I match province industry outstanding loan amounts with
manufacturing firms in the same province and same industry. The regression is:
Y,k,p,t = a - ,3 x LogLoan_ Pk,,,t + Controlp,to + YearFE + FirmFE+ El,(1.1)
where Y,k,p,t is the dependent variables of firm 1 in industry k province p in year t
such as logarithm of total assets, fixed assets, number of employees, total debts, ROA,
sales per worker, and total sales. I control local economic condition variables, and
year-fixed and firm-fixed effects. Panel A of Table 1.2 shows the regression results
for SOEs. CDB industry firm loans had significantly positive correlations with SOEs'
total assets, employment and debts. Panel B of Table 1.2 shows the regression results
for private firms. CDB industry firm loans had significantly negative correlations with
private firms' total assets, fixed assets, employment, total sales and sales per worker.
These results show that CDB industry loan amounts may have positive impacts on
SOEs' asset investment, employment and borrowing, and have negative impacts on
private firms' asset investment, employment and sales. The CDB industry loans may
crowd out private sector and crowd in SOEs within the same industry. Moreover,
I also study the infrastructure loans. Panel C of Table 1.2 shows the regression
include ministerial-level cities during city-level analyses since they are at the same level as provinces.
I define promotion based on varying political hierarchies accordingly.
26
results for city-level infrastructure loans. It shows that private firms' total assets,
fixed assets, debt, total sales and sales per worker are positively correlated with CDB
infrastructure loans.
However, there is a potential endogeneity concern that CDB credit allocations
are endogenous. CDB lends to infrastructure projects and SOEs via local governments. The local governments with better relationships with CDB may borrow more.
Furthermore, CDB is a policy bank with a mandate to provide credit supports to infrastructure and pillar industries, especially in undeveloped areas in China. In order
to fix this problem, I use municipal government turnover timing as the instrument
for CDB loans.
1.5.2
Instrument: City Secretary's Turnover Timing
I first test the exogeneity of city secretaries' turnover timing. Mentioned above, city
secretaries' terms are 5 years, and each city has its own turnover cycle. Although the
types of turnover might be endogenous such as promotion, the timing of turnover is
exogenous. I only use the variation from turnover timing. I use the Cox proportional
hazard model, which includes politicians' demographics, economic performance, and
time of turnover. The hazard rate of turnover is given by:
h(t) = ho(t) exp(ixi +
2
x 2 + - - - + /kXk),
(1.2)
where X 1 ... Xk are politician's ages and gender, economic performance in the city
such as GDP, income per capita, city government fiscal income, and city employment. I also include dummy variable NationalCycle to account for whether it a
National Congress Party year. I have both time-varying and time-invariant variables,
and I assume the time-varying variables are constant throughout a year. I followed
Wooldridge(2002) to construct a hazard model, and report the coefficients and hazard
ratios from the estimation in Table 1.3. I include previous CDB outstanding loans to
test whether the timing of turnover is affected by existing CDB loans. In column 1
to 4 of Table 1.3, NationalCycle had a positive effect on city secretaries' turnover,
consistent with patterns in Figure 1-2 since many city turnovers are in the national
turnover years. Column 5 of Table 1.3 reports hazard ratios, and on average, the
turnover probability increased by 39.3% during national turnover years. Age of a
secretary had positive effects on the timing of turnover because when city secretaries
get older, they are more likely to retire. The timing of turnover did not depend on a
city's past economic performance and CDB loans. I include economic variables and
city loans in two-year lags and find nothing significant. In columns 1 and 3 of Table
1.3, I exclude Age and Gender since they have some missing values. Again, timing
of turnover did not depend on a city's economic performance and CDB loans.
Then, I explore borrowing patterns over various periods of a city secretary's term.
The CDB's primary lending method is to coordinate with local governments and
support both infrastructure projects and industry firms, which are usually SOEs.
The city secretary is the top-ranking politician in the city, and usually plays a large
role during the lending process. The regression is:
27
LogLoant =
a + / x PoliticianYeari,
3 ,t + Controlj,t-i + YearFE
+cityFE + SecretaryFE+ sj,t,
(1.3)
where LogLoant is the CDB loan variables in city j in year t. I used the logarithm
of CDB outstanding loan amounts and new issuance for infrastructure, industry, and
total loans as dependent variables. PoliticianYearjj,t is the number of years that
secretary i stayed in city j in year t. Controlj,t_1 are variables for economic conditions
such as GDP, urban income per capita, fiscal income, and the working population. I
also include city fixed effects, year fixed effects, and secretaries' personal fixed effect
since cities have varying situations and secretaries have their own investment styles.
Standard errors are clustered at the city level. In Table 1.4 Panel A, the coefficient
of PoliticianYear was -0.364 at 1% significance in column 1; on average, if a city
secretary stayed one more year, borrowing from CDB decreased by 36.4%. If I break
loans into infrastructure and industry loans, the effects are stronger for industry loans.
In columns 3 and 5 of Table 1.4 Panel A, the coefficient of PoliticianYearwas -0.122
for infrastructure outstanding loans and -0.338 for industry loan. Instead of using
PoliticianYearjj,t, I use Year_ 1 ij,t, ... , Year_ 6 i,j,t, which are dummy variables for
years that secretary i stayed in city j in year t. Year_ li,j,t equals 1 if it was the first
year secretary i stayed in city j and zero otherwise. Year 2 i,j,t equals 1 if it was the
second year secretary i stayed in city
j
and zero otherwise. Year_ 3i,j,t to Year_ 6 i,j,t
were constructed similarly.
LogLoanj,t
=
a+
/1 x
Year_ li,j,t +
2
x Year_ 2i,j,t +
/3
x Year_ 3i,j,t
x Year_4ijt + /35 x Year_ 5 i,j,t + f36 x Year_ 6 3,,t
+Control3 ,t-j + YearFE + cityFE + SecretaryFE+ Ej,t (1.4)
+34
Results are shown in Table 1.4 Panel B. Year_1 is the missing category. Consistent with results in Panel A, Year_2's coefficient was -0.386 at 1% significance on
total CDB outstanding loan; on average, city secretaries borrowed 38.6% less during
their second years than first years. In column 1 of Table 1.4 Panel B, Year_3's coefficient was -0.749, Year _4's coefficient -1.071, Year_5's -1.429, and Year _6's -1.9.
Borrowing from CDB decreased monotonically when a city secretary stayed longer in
a city. In column 3 of Table 1.4 Panel B, CDB loan amount in a secretary's second
year was 10.9% less than the first year, and decreased monotonically over time. This
pattern was also true for infrastructure and industry loans, confirming results in Table
1.4 and suggesting that city secretaries borrowed more soon after they took the office,
and slowed borrowing monotonically over their terms. This pattern was also true for
infrastructure and industry loans. I exclude cities with the same turnover cycles as
the national cycle, and perform regressions with equations (1.3) and (1.4) again. Table 1.11 in the Appendix shows that for these off-national-cycle cities, city secretaries
had the same borrowing patterns; they borrowed more in the first year and decreased
monotonically over time. For these cities, the secretaries also borrowed more during
28
early terms. The variation was not only from 5-year national turnovers, but also from
these off-cycle cities. I find the similar patterns in CDB loan new issuances.
Moreover, in Figure 1-3, I plot the average logarithm of CDB city total loan
amounts after taking out the year fixed effects, city fixed effects and politician fixed
effects. There are 3 national turnover cycles during my sample period: 1998 to 2002,
2003 to 2007, and 2008 to 2010. I cluster the cities by PoliticianYear for these 3
cycles respectively and calculate the average logarithm of CDB city total loan amounts
for each bin in PoliticianYear. Figure 1-3 shows the "Zig-Zag" pattern during these
3 cycles. On average, city secretaries borrowed significantly more during their first
year in office and monotonically decreased the borrowing over time. When a new city
secretary came in, the borrowing spiked again. It verifies the results in Table 1.4. I
also look at the borrowing pattern in each city. Most of them follow this "Zig-Zag"
pattern. It alleviates the concern that certain cities with extreme values drive the
results in Table 1.4. The next logical questions are: why city secretaries intended
to borrow more during the early periods of terms? What are the incentives behind
these patterns? In China, promotion is one of the most important career aspects to
a politician. It is known well that city secretaries' and mayors' promotions in China
depend heavily on local GDP growth. One way to boost up local GDP in short term
is to borrow from CDB and invest as early as possible. See section 1.5.7 for more
detailed discussions.
1.5.3
Second Stage: CDB City Level Loan Analysis
In the second-stage regressions, I begin by studying city secretaries' turnover effects on
firm decisions. To study heterogeneous impacts on various types of firms, I separate
firms into 2 major categories: SOEs and private firms. SOEs' primary shareholders
are state and collective owners. Private firms' shareholders are private investors
such as individuals and institutions. Moreover, there is another type of firms with
shareholders from foreign countries. These foreign firms can be considered as the
subgroup of the private sector. Foreign firms in China usually have weaker connections
with government than domestic private firms, and may suffer more from government
credit than private firms. I match city secretaries' turnover data with manufacturing
firms in the same city. The regression is:
Y,t
a+
#1 x
Year
1
j,t
+34 x Year_ 4 j,t +
+
/2
x Year _2, + /33 x Year_ 3 ,t
/35 x Year_ 5j,t +
/6
x Year _6,t
+Contro1 ,t-1 + YearFE + FirmFE+ SecretaryFE+ Elt,
(1.5)
where Y,t is the dependent variables of firm I in year t such as logarithm of total
assets, fixed assets, number of employees, total debts, etc. Year_ 1 j,t, ... , Year_6j,t
are dummy variables for years secretaries stayed in city j in year t. For example,
Year_2j,t equals 1 if it was the second year that a secretary stayed at city j and zero
otherwise. Controlj,ti are variables for economic conditions such as GDP, urban
income per capita, fiscal income, and working population. I control the year fixed
29
effects, firm fixed effects, and politician personal fixed effects. In the regressions,
Year_1 is the missing category. Table 1.5 Panel A shows results for SOEs. In
column 1, assets of SOEs, on average, were 1.9% smaller in the second year of a city
secretary's term versus the first year. The coefficients for year 2 to year 6 were also
negative, and decreased monotonically. This pattern is also true for fixed assets and
debts (column 2 and 4). In column 5, in comparison to a city secretary's first year,
ROA increased by 1% in the city secretary's second year, and monotonically in later
years. In column 6 Table 1.5 Panel A, LogNumber is the logarithm of the total
number of SOEs in each city annually, and controls for a year fixed effect, city fixed
effects, and a secretary's personal fixed effect. There was no clear pattern of SOE
numbers in various years during city secretaries' terms. In short, SOEs had more
assets and debts in a city secretary's first year, and decreased overtime.
Table 1.5 Panel B is for private firms. In column 1, assets of private firms, on
average, were 1.5% smaller during the second year of a city secretary's term versus
the first year. Coefficients for years 3 through 5 were negative, and decreased monotonically. This pattern was also true for debts in column 4. However, the number
of private firms was smaller in the first year of a city secretary's term. CIC data
contain all manufacturing firms with annual sales greater than $700,000. If a firm
was dropped from the data, it either went bankrupt or had annual sales smaller than
$700,000. Results in Table 1.5 Panel B suggest that, on average, private firms were
crowded during the earlier periods of city secretaries' terms, but remaining firms grew.
These different effects from turnover might suggest that different types of CDB credit
might have different effects on private firms. CDB credit might also affect private
firms differently at different levels of the supply chain.
To explore which private firms were crowded, I separate firms into two group:
those characterized by low and high ROAs. The dummy variable LowROA equaled
1 if a firm's ROA was below the median. In column 1 Table 1.12, the interaction
term PoliticianYearLowROA had a negative coefficient, which means low ROA private firms increased assets more than efficient private firms during the early stages
of city secretaries' terms. In columns 2 and 3, PoliticianYearLowROA had negative
coefficients, which means low ROA private firms also increased fixed assets and employment more than efficient private firms during the early stages of city secretaries'
terms.
To link city secretaries' borrowing patterns and CDB loans' economic impacts, I
use Year_ 1 i,j,t,..., Year_ 6i,j,t as instrumental variables to measure CDB outstanding
loan amounts, and perform 2 two-stage, least-squares regressions. I include economic
control variables such as GDP, urban income per capita, fiscal income, and working
population, and firms' fixed effects, year fixed effects, and politicians' fixed effect as in
the first-stage regressions. Table 1.13 shows 2SLS regression results. Consistent with
previous evidence, CDB loans increased the assets, employment, and debts of SOEs,
and decreased foreign firms' assets and employment. For SOEs, doubling CDB outstanding loans increased SOEs' fixed assets by 6.2%, increased employment by 5.8%,
and increased debts by 7%29. For foreign firms, doubling CDB outstanding loans
29
1n Table 1.13, coefficients in columns 2 to 4 are 0.089, 0.083, and 0.101, respectively. Doubling
30
decreased the fixed asset by 11.8%, and decreased employment by 8.6%. From Table
1.5 Panel B and C, there were fewer private and foreign firms when city secretaries
were in their early terms. To explore changes to firm numbers further, I calculate new
firm entry and old firm exit for SOEs, private-sector firms, and foreign-sector firm.
Again, exit does not necessarily mean a firm went bankrupt. It could be that the
firm's annual sales were lower than 5 million RMB($700,000), so CIC did not record
it. Enter does not necessarily mean the firm is new. It could be that the firm's annual
sales were higher than 5 million RMB($700,000). Table 1.15 shows 2SLS regression
results on number of firms that exited or entered CIC data by using PoliticianYear
as the instrument for CDB loans. Columns 1 to 2 show that private firms exited more
and entered less when CDB loans increased. Columns 3 to 4 show that foreign firms
also exited more and entered less when CDB loans increased. Columns 5 to 6 show
that SOEs exited less and entered more when CDB loans increased, which provides
more evidence that private and foreign firms are crowded by CDB loans, and SOEs
become stronger.
However, various types of government credit should have different effects on private
and foreign sectors. I separate CDB loans into infrastructure- and industry-firm
loans. I perform 2SLS, and use Year_ li,j,t, ... , Year_6ijt as instrumental variables
for both CDB infrastructure and industry loans. Table 1.6 shows 2SLS regression
results. Interestingly, CDB infrastructure loans helped private and foreign sectors by
increasing their assets, and industry loans crowded both private and foreign sectors.
In column 1 to 7 in Table 1.6 Panel A, for private firms, CDB infrastructure loan
increases increased assets, fixed assets, number of workers, debts, sales per worker,
and total sales. CDB industry loan increases decreased assets, fixed assets, and
sales per worker. In column 1 to 7 in Table 1.6 Panel B, for foreign firms, CDB
infrastructure loan increases increased assets, debts, and sales per worker. CDB
industry loan increases decreases assets, fixed assets, debts, and sales per worker of
foreign firms. From these results, infrastructure loans supplemented both private
and foreign sectors, but industry loans, which usually go to SOEs, hurt private and
foreign firms. In sum, city secretaries borrowed more during early years of their
terms, and during these years, SOEs increased their assets and debts, and became
less efficient. For the private sector, there were fewer private firms, and the efficient
ones were crowded. Moreover, infrastructure loans helped all SOEs, private firms, and
foreign firms. Industry loans crowded both private and foreign sectors. Foreign firms
decreased their assets and employment. The number of foreign firms also decreased
during early years of secretaries' terms. Foreign firms can be viewed as the subgroup
of the private sector but with weaker connections. It may explain why the crowding
out effect is bigger on the foreign firms.
In China, it is debatable whether political turnover correlates with other macro
variables that influence local economies. The variation I use is from the varying
political cycles among cities. I control year fixed effects to remove effects from the
CDB city loans means Log(LoanCity) increases by Log(2), which equals 0.693. I multiply these
coefficients by 0.693 to calculate the percentage increase of the dependent variables. All following
analysis are based on the same calculation.
31
national turnover cycle. Moreover, I find disparate CDB loans' effects on private
and SOEs from 2SLS regressions. It helps mitigate the concern that turnover timing
of a city secretary correlates with demand changes or investment opportunities in a
city. Disparate effects between infrastructure and industry loans in Table 1.6 further
support the credit story instead of demand story in the city.
1.5.4
Politician's Other Channels to Affect Local Economy
Another concern is whether CDB loans are the only way a city secretary can affect
the economy; does the exclusion condition hold that local political turnover only affects the economy through CDB borrowing? When new city secretaries come to their
cities, they usually have their own plans or preferences to develop local economies, and
they have several tools to do it. For example, a secretary can build business districts
to attract investment, speed approvals of city projects, provide better government
services, etc. However, the biggest constraint is limited fiscal income. Local governments in China only share 20% to 30% of the tax revenue, and are responsible for
infrastructure buildup. City secretaries have many good projects piled on the desks,
but they require financial resources. There are three common ways to raise money:
borrowing from CDB, selling more land, and asking for more transfers. In China,
there are many pro-economic policies that are determined by the central government
such as export tax rebates, corporate tax breaks for foreign companies and export
companies, etc. City secretaries enforce these policies disparately. For example, they
can simply give tax breaks to more firms.
To rule out these channels, I regress the variable PoliticianYear on developed
land, export amounts, fiscal income, fiscal transfers, average effective corporate taxes,
and average effective value-added tax for each city to explore correlations between
political turnover and these variables. Table 1.16 shows there were no correlations.
However, it might take some time for a politician to act and effect local economies. To
address this concern, I use a lagged PoliticianYearinstead and find no correlations,
suggesting city secretaries do not sell more land to raise money, encourage exportation,
increase fiscal income, ask for more fiscal transfers, or lower tax rates 0 during their
early terms.
I plot the time trend of SOEs' fixed assets and employment to demonstrate the
effect of political turnover. The top 2 panels of Figure 1-5 examine SOEs' fixed-asset
patterns for high CDB industry loan cities and low CDB industry loan cities. I define
high and low categories as the top and bottom quartile of CDB city-level outstanding
industry loan distributions among 310 cities annually. The plots show a clear effect of
political turnover cycles on SOEs' assets in high CDB loan cities, but not in low CDB
loan cities. SOE assets were higher during early periods of city secretaries' terms
(i.e., 1 to 3 years) than during late periods (i.e., 4 to 6 years) in high CDB loan cities
after 2003. In low CDB loan cities, this pattern is not evident, consistent with 2SLS
regression results, which show that CDB loans increase SOE assets. One reason there
30
Export tax rebates offer a huge benefit for export firms in China, and the government gives part
of the value-added tax or sales taxes back to firms. This is reflected by TaxVAT in Table 1.16
32
are no differences in SOE assets between early and late periods of a secretary's term
in high CDB loan cities before 2003 is that before 2003, overall CDB lending size was
very small in comparison to later years (Figure 1-1). Even in these high CDB loan
cities, CDB did not lend much before 2003.
The middle two panels of Figure 1-5 show stratification on high and low average
effective corporate tax rates in the cities. I define high and low categories as the top
and bottom quartile of average effective corporate tax rate distributions among 310
cities annually. SOE fixed assets have no clear difference between early and late periods of city secretaries' terms in either high or low corporate tax cities, disqualifying
the channel with which city secretaries can offer more tax breaks to influence local
economies. Another way to influence a local economy is through selling land. Land
reserves play two roles in China. First, new business districts require land on which
to build. Second, revenue from selling land goes to local governments whose fiscal
income is limited. Selling land is a common way for a city to raise money, supplement
fiscal income, and support local developments. The bottom two panels in Figure 1-5
examine SOE fixed-asset patterns for cities with large and small land developments.
I define large and small categories as the top and bottom quartile of developed land
distributions among 310 cities annually. The reason I do not use total land is that
only developed land can be used to develop new districts or be sold. There is much
deserted land in the western part of the country, and I exclude it during analysis.
The bottom two panels of Figure 1-5 show no clear difference in SOE fixed assets
between early and late periods of city secretaries' terms in the cities that have either
large or small amounts of developed land.
Figure 1-6 shows results from the same analysis as in Figure 1-5 but for number
of employees, another important variable. Only the top panel for high CDB loan city
shows a clear pattern that SOEs had higher fixed assets during early periods of city
secretaries' terms (i.e., 1 to 3 years) than during late periods (i.e., 4 to 6 years) after
2003. All other panels do not have similar clear patterns. This confirms the exclusion
condition of 2SLS.
Third, I explore various effects of turnover timing in cities with different CDB
loan levels. In Panel A of Table 1.14, I interact PoliticianYearwith dummy variable
HighCDB for whether a CDB loan amount was above the median of all CDB loans
in 310 cities from 1998 to 2009. I regress them on SOE variables. The interaction
term PoliticianYearand HighCDB had negative coefficients with total asset, fixed
assets, employment, debts, and sales per worker, suggesting turnover timing had
larger effects when CDB loans were high. The main effects of PoliticianYeardid not
have significant coefficients, so most of the effects of turnover timing were from high
CDB loan areas. Panel B in Table 1.14 is restricted to SOEs with zero CDB loans in
the city. There are no significant coefficients with PoliticianYearin these zero CDB
loan areas. These results further support the exclusion conditions of turnover timing.
33
1.5.5
CDB Industry Loan Analysis
Industry loan's effect on firms in the same industry
Section 1.5.1 and 1.5.3 shows that at the city level, CDB loans for infrastructure
help private and foreign firms grow. However, industry firm loans crowd private
and foreign sectors. This section uses province-industry level loan data and explores
more industry loan effects. The data offer the advantage of analyzing the effects of
government credit on industries, and effects on related industries. Most extant studies
are based on aggregate government credits or spending.
Most CDB industry firm loans go to SOEs; approximately, 90% of total industry
loans are for them. I regress the lagged aggregate assets of SOEs and private firms
on CDB loans in the same province and industry. Table 1.17 Panel A shows that
only SOEs' assets had positive coefficients. In column 1, the coefficient for LagAsset
of SOEs was 0.058 at 1% significance. The unit of LagAsset and Loan _ P1 are
both thousands of RMB; when a province has 1000 RMB more total assets of SOEs,
the industry in this province can have 58 RMB more CDB outstanding loans the
following year. Private-sector assets did not have significant coefficients, verifying
that CDB industry firm loans usually go to SOEs. In Table 1.17 Panel B, I regress
aggregate lagged assets of SOEs and private firms on aggregate infrastructure loans.
Columns 1 and 2 of Table 1.17 Panel B are for province-level infrastructure loans,
and columns 3 and 4 of Table 1.17 Panel B are for city-level infrastructure loans.
Lagged SOE assets had no effects, but lagged, private-sector assets had positive effects
on infrastructure loans, verifying that industry loans go primarily to SOEs. This
disqualifies the alternative explanation that industry loans go to places with more
investment opportunities, which happen to have more SOEs because infrastructure
loans do not depend on aggregate SOEs' assets. Infrastructure loans tend to be
lent to places with more private firms. The reason could be that areas with more
private firms are often big cities or provinces that also require greater infrastructure
investments.
In China, cities in the same province usually focus on different industries, especially for SOEs that adjust slower than those in the private sector. I aggregate total
assets of SOEs at the city-industry level each year and pick out the largest industry
in each city. On average, only two cities in the same province focused on the same
industry, and the city usually stuck to the same industry over time. Among 310 cities,
42% did not change industries from 1998 to 2009. 40% changed once, 14% twice, and
only 3% more than twice.
Based on these findings, I again use city-level turnover as a shock to CDB industry
loans at the province level. I mark each city with its largest SOE industry annually.
If city secretaries were in their earlier terms and based on previous conclusions, the
city usually borrowed more industry firm loans for SOEs from CDB. I consider it a
shock to a province-level CDB loan on the industry that is the largest SOE industry
in the city. For example, city A in province B focuses on industry C. If city A's
current secretary is in an earlier term, I consider it a shock to CDB loans of industry
C in province B that year. Since cities in the same province usually focus on different
34
industries and the largest SOE industry in a city does not change often over time,
if a city borrows more for its SOEs, it should be reflected in the province-level CDB
loan in the industry. Formally, the regression is:
LogLoanPIkp,,
= a + 01 x Firstk,p,t + 132 x Secondk,,, + /33 x Thirdk,p,,
+04 x Fourthk,p,t + 35 x Fifthk,p,t + 6 x Sixthk,,,t
+Controlp,t-I + YearFE + provinceFE
+IndustryFE+ Ek,p,t,
(1.6)
where LogLoanPIk,p,t is the logarithm of CDB outstanding loan amount in
industry k, province p in year t. Firstk,p,t is a dummy variable that equals 1 if there
is a city that focuses on industry k in province p that has a secretary who is in her
first year. Secondk,p,, is a dummy variable that equals 1 if a city focuses on industry
k in province p that has a secretary who is in her second year. Thirdkp,, to Sixthk,p,t
are defined similarly. Controlp,ti includes lagged GDP, urban income per capita,
fiscal income, and working population in province p. I control year, province, and
industry fixed effects. Regression results are shown in Table 1.18. In the first column
of Table 1.18, CDB industry loans were larger if a city that focused on this industry
had a secretary who was in her early term; Firstk,p,t , Secondk,p,,, and Thirdk,,, had
positive coefficients. I also combine the first two or three years and create dummy
variables First- Secondk,,, and First- Thirdk,p,t. In columns 2 and 3 of Table 1.18,
these two variables also had positive coefficients. On average, a province borrowed
33.2% more from CDB for an industry that focused on cities that had secretaries in
the first three years of their terms. City secretaries borrowed more for a city's top
SOE industry during their early terms, consistent with previous results that suggest
secretaries borrowed more during earlier term.
Next, I use Firstk,,, to Sixthk,p,t to instrument LogLoanPIk,p,t and perform
2SLS. I include all control variables such as economic variables Controlpt_1, yearfixed effects, and firm-fixed effects during first-stage regressions. The second-stage
regression is:
Yl,k,p,t
-- a+ /3 x LogLoan PIkpt+Control,ti + YearFE + FirmFE+
6
i,t,
(1.7)
where Yl,k,p,t is the dependent variables of firm I in industry k province p in year t
such as logarithm of total assets, fixed assets, number of employees, total debts, ROA,
sales per worker, and total sales. Again, I control local economic condition variables,
and year-fixed and firm-fixed effects. Panel A of Table 1.7 is the 2SLS regression
results for private firms. In columns 1 to 4 and 6 to 7, CDB industry firm loans
had negative effects on private firms' total assets, fixed assets, employment, debts,
sales per worker, and total sales. When industry loans doubled, private firms in the
same industry, on average, decreased fixed assets by 14.8%, decreased employment by
9.7%, decreased debts by 20.1%, decreased sales per worker by 17.4%, and decreased
total sales by 26.3%. In column 5 of Table 1.7 Panel A , CDB industry firm loans
35
had positive effects on private firms' ROAs. Panel B of Table 1.7 is 2SLS regression
results for SOEs. Again, industry loans helped SOEs increase total assets, fixed assets,
employment, debts, sales per worker, and total sales. When industry loans doubled,
SOEs in the same industry, on average, increased fixed assets by 19.3%, increased
employment by 13.0%, increased debts by 30.4%, increased sales per worker by 2.3%,
and increased total sales by 12.6%. In sum, CDB industry loans make SOEs grow
larger and sell more. Contrarily, private firms shrink in size and sell less. Since CDB
industry firm loans usually go to SOEs, it is unsurprising that SOEs become stronger
and crowd the private sector. The results from 2SLS are consistent with the OLS
results in Table 1.2.
Industry loan's effect on firms in the related industries
It is well known that China's economy grew dramatically during the last two decades,
and the private sector was the primary driver of this growth(Figure 1-4). Although
government credit crowds the private sector in the same industry, it might complement the private sector in related industries. CDBaAZs strategy is to aid basic
industries such as energy and mining to help related industries. For this sector, I use
an input-output matrix to identify inter-industry relationships and study spillover
effects of government credit. I use the national input-output matrix from 2007 from
the National Bureau of Statistics of China to define upstream and downstream industries 3 '. The input-output matrix has 42 industries, and CDB classifies its loans
into 95 industries, which is more detailed. I match these two industry classifications
by combining CDB industries. For each industry, I pick the largest intermediate input from other industries as the upstream industry. At the firm level, I match each
firm with its upstream industry CDB loan in the same province. After the merger,
there were 25 industries in the manufacturing sector. Again, I use city-level turnover
dummy variables to instrument UpstreamLoan and perform 2SLS. The first-stage
regression is:
LogUpstreamLoani,k1,,,
=
z+
1
x Firstk',,t +
/2
x Secondk',,,+ /33 x Thirdk',P,,
+34 x Fourthk,t + 35 x Fifthk,t +
6
x Sixthk,p,t
+Control,t_1 + YearFE + FirmFE+ ejt,
(1.8)
where LogUpstreamLoanI,kI,,, is the logarithm of CDB outstanding loan amount
in the upstream industry of firm 1 in industry k, province p in year t. k' indexes the
upstream industry of k. Firstk',,, is a dummy variable that equals 1 if there was a
city that focused on industry k' in province p and that had a secretary who was in the
first year. Second1,,, to Sixthk',,t are defined similarly. The second-stage regression
is:
31I also use other years' input-output matrices to double check the definition of upstream and
downstream industries and assess the same inter-industry relationships that do not change much
over time.
36
Y,k,p,t
= a + 3 x LogUpstreamLoanlpk,
+ Controlp,t-I + YearFE + FirmFE+ eI,t,
(1.9)
where Y,k,p,t are dependent variables of firm I at year t, which is in industry k,
province p. LogUpstreamLoank,,,t is the estimated CDB loan to upstream industry
k'. k' indexes the upstream industry of k. Panel A of Table 1.8 Panel A shows
results for private firms. Generally, CDB loans to firms' upstream industry helped
the private sector. In columns 1, 2, 4, 6, and 7 of Panel A Table 1.8, the upstream
CDB industry firm loan had positive effects on private firms' total assets, fixed assets,
debts, sales per worker, and total sales. When the upstream industry loan doubled,
private firms in the downstream, on average, increased fixed assets by 9.4%, increased
debts by 3.2%, increased sales per worker by 4.4%, and increased total sales by 4.9%.
However, there was no change in employment or ROA. Moreover, in Table 1.9 Panel
A, I interact the LogUpstreamLoanwith dummy Connected for whether the private
firms' political hierarchy is above the city level or not. In China, every firm(include
private firm) has a political hierarchy. It defines which level of the government the
firm needs to report to. In another word, it determines which level of the government
the firm is affiliated with. For example, city level firms means that the firms are
under city governments and report to city governments. From the regression results
in Table 1.9 Panel A, the private firms with better political connections can benefit
significantly from the upstream loans. It means that although the CDB upstream
loans have positive effects on downstream private firms but the connected private
firms can benefit significantly more. Moreover, in order to check whether firm size
is correlated with political hierarchy and drives the results, I control the interaction
terms between firm total assets and the instrumental variables in Panel B Table 1.9.
It turns out that the results in Panel B are very similar to Panel A. This confirms
that political connection matters instead of the firm size. Table 1.8 Panel B shows
results for SOEs. Generally, CDB loans to firms' upstream industry also helped
SOEs. In column 1 to 4 and 7 of Table 1.8 Panel B, the upstream CDB industry
firm loans had positive effects on SOEs' total assets, fixed assets, employment, debts,
and total sales. When the upstream industry loan doubled, SOEs in the downstream,
on average, increased fixed assets by 9.2%, increased employment by 8.7%, increased
debts by 13.2%, and increased total sales by 5.8%. However, SOEs' sales per worker
decreased. In column 5 and 6 of Table 1.8 Panel B , ROA and sales per worker both
decreased. Although both the private sector and SOEs grew, unlike private firms,
SOEs hired more workers and experienced lower efficiency.
1.5.6
Overall Effects of the Government Credit from CDB
From the analyses above, CDB infrastructure loans help all types of firms: SOEs,
private, and foreign. CDB industry loans that usually go to SOEs expand SOEs but
crowd private firms in the same industry. Upstream CDB industry loans help private
firms in downstream industries. What is the overall effects of government credit from
CDB?
37
One component of CDB's mandate is to help basic industries. Although CDB
industry loans crowd the private sector in the same industry, they help private firms
in downstream industries grow. I use estimated coefficients and various types of CDB
credits to study overall effects. In column 1 of Table 1.6 Panel A, the coefficient of
CDB city-level infrastructure loans on private firms' total assets was 0.246, which
means one unit increase in the logarithm of CDB city infrastructure loans increases
the logarithm of each private firm's assets by 0.246 in the same city. For industry
loans, column 1 of Table 1.7 Panel A shows that one unit increase in the logarithm
of CDB province industry loans decreases the logarithm of each private firm's assets
by 0.246 in the same province and industry. Column 1 of Table 1.8 panel A shows
that one unit increase in the logarithm of CDB province upstream industry loans
increases the logarithm of each private firm's assets by 0.092 in the same province and
industry. Based on these coefficients and changes to CDB loans in infrastructures and
industries, I find that, on average, total CDB credit increased private firms' assets by
3.4% annually from 1998 to 2009. During the same period, the average annual growth
rate of private firms' assets was 15.4%. Therefore, CDB loans contributed about 22%
to private-sector growth from 1998 to 2009. On average, a $1 million increase in
CDB total outstanding loan amount led to a $0.33 million increase in private firms'
total assets. If I exclude infrastructure loans, then from 1998 to 2004, CDB industry
loans increased private firms' assets from 2% to 7% annually and a $1 million increase
in CDB outstanding industry loan amount led to a $0.43 million increase in private
firms' total assets. From 2005 to 2009, CDB industry loans decreased private firms'
assets from 2% to 5% annually and a $1 million increase in CDB outstanding industry
loan amount led to a $0.89 million decrease in private firms' total assets. Overall,
from 1998 to 2009, a $1 million increase in CDB outstanding industry loan amount
led to a $0.52 million decrease in private firms' total assets.
Sales per worker is an important efficiency measurement in China, especially in the
manufacturing sector. An abundant labor supply made labor costs cheaper in China,
one of the most important reasons for dramatic growth in exports and the economy
as a whole. Most manufacturing firms in China are labor intensive. Higher sales per
worker means a firm can do more with fewer workers. From results in Tables 1.6,
1.7, and 1.8, CDB infrastructure loans increased sales per worker for private firms,
and CDB industry loans decreased sales per worker for private firms in the same
industry. CDB loans to upstream industries increased sales per worker for private
firms in downstream industries. Based on these coefficients, I find that, on average,
total CDB credits increased private firms' sales per worker by 10.8% annually from
1998 to 2009. During the same period, the average sales per worker annual growth
rate of private firms was 20%. Therefore, CDB loans contributed more than 50% to
the growth of the private sector's sales per worker from 1998 to 2009. If I exclude
infrastructure loans, then from 1998 to 2004, CDB industry loans increased private
firms' sales per worker from 2% to 9% annually. From 2005 to 2009, CDB industry
loans decreased private firms' assets from 6% to 9% annually.
I further explore the reasons behind disparate credit effects during various periods. Figure 1-7 plots total CDB loan issuances from 1998 and 2010 for the top 10
industries. In 1998, electric power supply and coal mining were the top two industries,
38
followed by petroleum and natural gas extraction, oil processing and refining, chemical products, etc. Except transportation equipment manufacturing, all of these were
upstream industries. Manufacturing firms were usually direct downstream industries.
The dominant weight on upstream industries can have large positive spillover effects
on downstream industries, one reason industry credit can have positive effects on the
private sector during earlier years. In 2010 and after 12 years, the top industries
of CDB loans changed. Electric power supply is still the top industry of CDB loan
issuances. However, three of the top 10 industries are in manufacturing sectors. Electronic equipment manufacturing ranks third, which was not included in the top 10
industries back to 1998. Special equipment manufacturing12 ranks fourth, and transportation equipment manufacturing ranks sixth. This could lead to bigger crowding
effects in manufacturing industries, and smaller spillover effects for downstream industries. This might explain why CDB industry loans had positive effects on private
sectors during earlier years, and negative effects during later years. During the past
20 years, China had dramatic GDP growth, and there were many shortages such as
energy supply and mining. CDB loans in upstream industries helped solve these demand constraints, possibly explaining why CDB industry loans can help the private
sector grow faster and become more efficient during earlier years. However, in later
years, CDB focused less on basic industries, shifting to other industries such as electronic equipment. CDB loans to infrastructure always helped a firm grow and improve
efficiency. Although there are increasingly new, modern cities in China, urbanization
is far from over.
1.5.7
Politicians' Incentives Behind the Borrowing Patterns
Section 1.5.2 shows that city secretaries borrow significantly more during their first
year in office and decrease the borrowing monotonically overtime. The next logical
questions are: why city secretaries intended to borrow more during the early periods
of terms? What are the incentives behind these patterns? The political view suggests
politicians had personal goals that did not necessary accord with social goals. In
China, promotion is one of the most important career aspects to a politician. Li and
Zhou (2005) find that the likelihood of promotion of provincial leaders increased with
economic performance in China between 1979 and 1995. It is known well that city
secretaries' and mayors' promotions in China depend heavily on local GDP growth.
Although other determinants also matter such as political background and personal
connections, GDP is one of few aspects that can be quantified. To verify the hypothesis, I calculated GDP growth from various periods and performed the regression:
pomotionij
=
a + 01 x GPD_IIncreaset,,
3 +
+3
x
agej +
4
$2
x genderi + Ej
x relationi,
(1.10)
Each observation represents one city secretary in one city 3 3 , where pomotioni is a
3
Special equipment includes equipment for mining, agriculture, medical, clothing, etc.
"Some cases exist such that the same city secretaries were in different cities. The observation
39
dummy variable for whether city secretary i in city j got promoted during turnover.
GPDJIncreaset,ijis the GDP increase from secretary i 's first year in city j in year
t. I set t = 1, 2, 3, 4, 5 to examine effects from GDP increased in various stages of
a city secretary's term. relationij is a dummy variable for whether city secretary i
in city j was from the same hometown as provincial governor. ageij is the age of
secretary i in city j during the turnover year. genderi is a dummy variable for whether
a city secretary i was female. Standard errors clustered at the city level. Table 1.19
shows probit regression results. GDP increases increased city secretaries' promotional
chances. GDP increases during the first couple of years of a city secretary's term had
larger effects. In column 1 of Table 1.19 Panel A, the coefficient for GDP increases in
the first year of a city secretary's term was 0.975. It decreases monotonically when
I included more years (column 2 to 5). age had negative coefficients because older
secretaries were more likely to retire and a rule states that to get promoted, a city
secretary's age should be lower than 55 years.
I also study CDB loan increase impacts on promotion. The probit model becomes:
pomotionij
=
a + 31 x Loan_ Increaset,ij +
02
+/33 x age,3 + /4 x genderi + Ej,
x relationi,
(1.11)
where Loan_Increaset,ij is the logarithm of CDB outstanding loan increase from
secretary i's first year in city j in year t. I set t = 1, 2, 3, 4, 5 again. Table 1.19
Panel B shows that CDB loan increases had positive effects on promotion. This effect
is primarily from loan increases during the first two years of secretaries' terms. In
column 1 and 2 of Table 1.19 Panel B, coefficients for loan increases were 0.102 and
0.087, both significant. When I included later years(column 3 to 5), the coefficients
are lower and less significant. relation had positive coefficients (column 1 to 5 of
Table 1.19 Panel B), which makes sense since promotion decisions were usually made
by provincial governors. Again, age has negative effects on promotion.
In short, GDP growth had positive effects on city secretaries' promotions when
their terms ended and earlier years' GDP growth had bigger effects. CDB loans
also had positive effects on promotion probability, and loan increases during first two
years of secretaries' terms had larger impacts on promotions. In China, borrowing
from CDB is the primary method city secretaries use to boost local GDP during the
last 15 years, and the loans take time to help the economy, so city secretaries should
intend to borrow from CDB as early as possible. One concern is that politicians
in China are assigned by the Communist Party instead of elected by voters. Moreconnected politicians can be assigned to better cities and can borrow more from CDB.
Consequently, these politicians have greater chances of promotion. To deal with this,
I focus on borrowing patterns from the timing of turnover. Most city secretaries'
terms are 5 years, and based on duration model results, the timing of the turnover
is unaffected by economic conditions and CDB loan amounts. Second, I control city
is at the city/secretary level. For example, if one city secretary stayed in two cities, there are two
observations for that secretary.
40
secretaries' personal fixed effects during regression to remove politicians' personal
time-invariant effects.
The borrowing patterns and promotions of a city secretary accord with the hypothesis that promotion is an important goal for city secretary, and promotions depend
heavily on local GDP growth. One way to increase short-term GDP growth is to
borrow from CDB and invest, explaining why city secretaries tend to borrow significantly more during earlier terms and slow borrowing later. However, there are several
alternative explanations for this pattern. First, it could be that a city secretary's goal
is to maximize social welfare, and the secretary wants to explore investment opportunities as early as possible during a term. Since city secretaries invest in good projects,
GDP increases and they are promoted. Second, more-connected city secretaries have
greater chances to get promoted and assigned to a better city with more economic
growth potential.
To separate the hypothesis from these alternatives, I include a dummy variable
for city secretaries whose expected terms were fewer or equal to two years when they
became secretaries. The bottom panel of Figure 1-2 shows a tendency to bring city
turnover cycles back to national cycle. A city secretary would have known that there
was a greater probability of turnover during these national turnover years. For example, if a city secretary began her term in 2011, she knew that the next turnover
could be in 2013, a national turnover year, and there was only two years remaining.
Retirement age is another cause. In China, the retirement age for male city secretary
is 60, and 55 for female. I made dummy variable ShortTerm equal to 1 if national
turnover or retirement was within two years when secretaries begin their terms. These
secretaries expected their terms to be fewer than or equal to two years. If the secretaries' goal was to maximize social welfare, they should have borrowed and invested
regardless of how long they expected to stay in the city. If their goal was promotion,
they had less incentive to borrow if there were only two years remaining. In Table
1.20 columns 1 and 2, ShortTerm had negative effects on both CDB outstanding loan
amounts and new issuances. The coefficient for CDB outstanding loans was -2.064.
On average, for city secretaries who expected their terms to be less than or equal
to two years from the beginning of their terms, they borrowed 50% less. If a city
secretary wanted to explore good investment opportunities as soon as possible, they
should still have borrowed and invested even if they expected short tenure.
Second, I include dummy variable Younger for whether city secretaries were
younger than 50 years when their terms began. In China, if city secretaries are
more than 55 years old, they do not get promoted during turnover, and are usually
assigned to a same-level, though less important, position until retirement. The hypothesis is that if city secretaries know there is a large chance to turnover again in the
near future (e.g., 2 years), they have fewer incentives to borrow since a loan requires
time to operate, and it will not help their promotions much. However, if city secretaries care about social welfare, they want to invest in good projects even if they will
not help with promotions. Table 1.20 columns 3 and 4 show that if city secretaries
were from the same home town as provincial governors, they borrowed 400% more.
Again, if city secretaries wanted to explore good investment opportunities as soon as
possible, they should still have borrowed and invested even if they were older and had
41
few chances at promotions.
Third, I include dummy variable relation for whether a city secretary was from
the same hometown as a provincial governor. Provincial governors make promotional
decisions for city secretaries, and it is difficult identifying personal relationships between them. City secretaries are not necessarily from the same cities they rule, and
neither are provincial governors. In the data, 94.64% of city secretaries were from
other cities. The assumption here is that if a city secretary and provincial governor
are from the same hometown, they are more likely to know each other, to have worked
together previously, or for the governor to bring a city secretary with him/her during
turnover. This cronyism is common in China. In the data, 8.42% of city secretaries
were from the same hometowns as their governors. Table 1.20 column 5 shows that
if city secretaries came from the same hometowns as the governors, they borrowed
18.9% more from CDB. If city secretaries or provincial governors cared about social welfare, they should have borrowed money based on project quality rather than
relationships between them.
A question that follows naturally is what prevented city secretaries from borrowing as much as possible during later terms? Although CDB loans during city
secretaries' later terms do not help promotions, they can still borrow as much as
they can unless there is a cost of doing so. Distress costs from high leverage is
an obvious possibility. If a city already had many loans when a secretary began a
term, the secretary should have known approximately the total borrowing during the
term. Were the city constrained, the secretary had more incentives to move borrowing forward, and vice versa. To test this hypothesis, I use CDB debt-to-GDP
ratio to measure constraints. City average CDB debt-to-GDP ratio was 2.4%, and
I define a city as contained if the ratio were more than 6.9%(top quartile). Overall
debt-to-GDP appeared low, but in China, a city shares about 20% to 30% of tax
revenue. The fiscal income they can use to repay a loan is even smaller. In the top
25% of cities with a debt-to-GDP ratio greater than 6.9%, the median CDB debtto-fiscal income ratio was 219%; these cities are constrained. I used dummy variable
Constrained for whether a city had a CDB debt-to-GDP ratio greater than 6.9%
before the city secretary began a term. I also used the interaction of Constrained
and PoliticianYear, with results in Table 1.21. PoliticianYear still had negative
coefficients from columns 1 to 6. Constrainedhad positive coefficients on CDB outstanding loans for both infrastructure and industry, which is mechanical. For CDB
loan new issuances (columns 2, 4, and 6), Constraineddid not have an effect, meaning constrained cities did or could not borrow more to invest or repay old loans. The
interaction term PoliticianYearConstrained
had negative coefficients. For example,
in Table 1.21 column 1, the coefficient for PoliticianYearConstrainedwas -0.167
and the main effect of PoliticianYear was -0.274, suggesting that for constrained
cities, secretaries tended to move more CDB borrowing forward. This supports the
hypothesis that CDB loans are costly, and if city secretaries know they can borrow
only a limited amount during their terms, they tend to borrow as early as possible.
In sum, city secretaries borrow significantly more during early terms, and they
borrow less if the expected turnover is in fewer than or equal to two years. City
secretaries who are older than 50 when they begin their terms borrow significantly less.
42
A secretary who is from the same place as a provincial governor borrows significantly
more. These results support the hypothesis that city secretaries want to borrow as
early as possible to increase GDP and enhance promotion chance. Results are also
against the hypothesis that city secretaries want to invest in good projects as soon
as possible. However, instead of promotions, there could be other explanations for
why short-term secretaries borrow less and younger secretaries borrow more. Perhaps
short-term secretaries just want to live a quiet life and do not want to make huge
changes during their short terms. Young secretaries are more ambitious and might
have different educations than older generations. I do not disqualify these alternatives,
but instead offer evidence that is consistent with the political view. These alternatives
do not affect my identifications in the primary results.
1.6
Conclusion
This paper explores the different effects of various types of government credit (infrastructure vs. industry credit). It traces the effect of government credit across different
levels of the supply chain. Using unique industry, detailed loan data from China Development Bank, I find that government credits to infrastructure help firms expand
in size, debt, employment, and sales per worker. Government credits to industries,
which usually go to SOEs, help SOEs expand, but crowd private firms in the same
industry. However, these industry loans help private firms in downstream industries.
Overall, government credit from CDB supplemented the private sector with a contribution of approximately 22% to private sector growth from 1998 to 2009. These
results shed light on the prior mixed empirical findings. I use municipal politicians'
turnover timing as an instrument for loans from the CDB. I find that city secretaries
borrowed more during their early terms, and I provide evidence that accords with the
hypothesis that promotion is the incentive behind politicians' borrowing patterns.
Although direct costs of government credit is essential (Lucas and Moore (2010),
Lucas (2012a)) to evaluating government credit programs, they are beyond the scope
of this paper. In future research, it is important to value the costs of these government loans to evaluate overall costs and benefits. Local-government indebtedness has
a direct impact on housing prices and the shadow banking system in China. How does
government credit affect households in China? What is the relationship between government credit and China's shadow banking system? Answers to these questions will
elucidate China's government credit and the larger picture of the country's economy.
43
1.7
Bibliography
Allen, Franklin, Jun Qian, and Meijun Qian. "Law, finance, and economic growth
in China." Journal of financial economics.77.1 (2005): 57-116.
Aschauer, David Alan, and Jeremy Greenwood. "Macroeconomic effects of fiscal
policy." Carnegie-Rochester Conference Series on Public Policy. Vol. 23. (1985).
Atkinson, Anthony Barnes, Joseph E. Stiglitz. "Lectures on public economics.",
London, McGraw Hill (1980).
Banerjee, Abhijit V. "A theory of misgovernance." The Quarterly Journal of
Economics 112 (1997): 1289-1332.
Banerjee, Abhijit, Esther Duflo, and Nancy Qian. "On the road: Access to
transportation infrastructure and economic growth in China." No. w17897. National
Bureau of Economic Research (2012).
Barro, Robert J. "Output Effects of Government Purchases." Journal of Political
Economy 89 (1981): 1086-1121.
Barro, Robert J. "Government Spending in a Simple Model of Endogenous
Growth." Journal of Political Economy 98 (1990): S103-25.
Baxter, Marianne, and Robert G. King. "Fiscal policy in general equilibrium."
The American Economic Review (1993): 315-334.
Bertrand, Marianne, Francis Kramarz, Antoinette Schoar and David Thesmar.
"Politicians, Firms and the Political Business Cycle: Evidence from France" Working
Paper, Chicago Booth (2007).
Bertrand, Marianne, and Antoinette Schoar. "Managing with style: The effect
of managers on firm policies." The Quarterly Journal of Economics (2003): 1169-
1208.
Bertrand, Marianne, Antoinette Schoar, and David Thesmar. "Banking deregulation and industry structure: Evidence from the French banking reforms of 1985."
The Journal of Finance 62.2 (2007): 597-628.
Besanko, David, and Anjan V. Thakor. "Competitive equilibrium in the credit
market under asymmetric information." Journal of Economic Theory 42.1 (1987):
167-182.
Blanchard, Olivier, and Roberto Perotti. "An Empirical Characterization Of
The Dynamic Effects Of Changes In Government Spending And Taxes On Output."
The Quarterly Journal of Economics 117.4 (2002): 1329-1368.
Blanchard, Olivier, and Andrei Shleifer. "Federalism With and Without Political Centralization: China versus Russia, in Transition Economies: How Much
Progress?" IMF Staff Papers (2001): 171-179.
Burnside, Craig, Martin Eichenbaum, and Jonas DM Fisher. "Fiscal shocks and
their consequences." Journal of Economic theory 115.1 (2004): 89-117.
Caballero, Ricardo J., Takeo Hoshi, and Anil K. Kashyap. "Zombie lending
and depressed restructuring in Japan." The American Economic Review 98.5 (2008):
1943-1977.
Caldara, Dario, and Christophe Kamps. "What are the effects of fiscal shocks?
A VAR-based comparative analysis." No. 877. European Central Bank, (2008).
44
Carvalho, Daniel. "The Real Effects of Government-Owned Banks: Evidence
from an Emerging Market." The Journal of Finance 69.2 (2014): 577-609.
Cetorelli, Nicola, and Philip E. Strahan. "Finance as a barrier to entry: Bank
competition and industry structure in local US markets." The Journal of Finance
61.1 (2006): 437-461.
Chaney, Paul K., and Anjan V. Thakor. "Incentive effects of benevolent intervention: The case of government loan guarantees." Journal of Public Economics 26.2
(1985): 169-189.
Chen, Ye, Hongbin Li, and Li-An Zhou. "Relative performance evaluation and
the turnover of provincial leaders in China." Economics Letters 88.3 (2005): 421-425.
Claessens, Stijn, Erik Feijen, and Luc Laeven. "Political connections and preferential access to finance: The role of campaign contributions." Journal of Financial
Economics 88.3 (2008): 554-580.
Clarke, George RG, and Robert Cull. "Political and Economic Determinants of
the Likelihood of Privatizing Argentine Public Banks." Journalof Law and Economics
45.1 (2002): 165-197.
Cohen, Lauren, Joshua D. Coval, and Christopher Malloy. "Do powerful politicians cause corporate downsizing?" Journal of Political Economy 119.6 (2011): 10151060.
Cole, Shawn."Fixing market failures or fixing elections? Elections, banks and
agricultural lending in India. American Economic Journal: Applied Economics 1 (2009):
219-250.
Craig, Ben R., William E. Jackson, and James B. Thomson. "Small Firm Finance, Credit Rationing, and the Impact of SBA-Guaranteed Lending on Local Economic Growth." Journal of Small Business Management 45.1 (2007): 116-132.
Demirguc-Kunt, Ash, and Vojislav Maksimovic. "Law, finance, and firm growth."
The Journal of Finance 53.6 (1998): 2107-2137.
Dinc, I. Serdar. "Politicians and banks: Political influences on governmentowned banks in emerging markets." Journal of Financial Economics 77.2 (2005):
453-479.
Dinc, I. Serdar, and Nandini Gupta. "The decision to privatize: Finance and
politics." The Journal of Finance 66.1 (2011): 241-269.
Dyck, Alexander, and Luigi Zingales. "Private benefits of control: An international comparison." The Journal of Finance 59.2 (2004): 537-600.
Elliott, Douglas J. "Uncle Sam in Pinstripes: Evaluating U.S. Federal Credit
Programs." Washington, D.C.: Brookings Institution (2011).
Fatas, Antonio, and Ilian Mihov. "The effects of fiscal policy on consumption
and employment: theory and evidence." Centre for Economic Policy Research Vol.
2760 (2001).
Faulkender, Michael, and Rong Wang. "Corporate financial policy and the value
of cash." The Journal of Finance 61.4 (2006): 1957-1990.
Gale, William G. "Federal lending and the market for credit." Journalof Public
Economics 42.2 (1990): 177-193.
Gale, William G. "Economic effects of federal credit programs." The American
Economic Review (1991): 133-152.
45
Gali Jordi, J. David Lopez-Salido, and Javier Valles, "Understanding the effects of government spending on consumption." Journal of the European Economic
Association 5.1 (2007): 227-270.
Gerschenkron, Alexander. "Economic backwardness in historical perspective."
Cambridge, MA: Harvard University Press (1962).
Hanson, Samuel G., David S. Scharfstein, and Adi Sunderam. "Fiscal Risk
and the Portfolio of Government Programs." HarvardBusiness School working paper
(2014).
Hart, Oliver, Andrei Shleifer, and Robert W. Vishny. "The proper scope of government: theory and an application to prisons." The Quarterly Journal of Economics
112 (1997): 1127:1162.
Hsieh, Chang-Tai, and Peter J. Klenow. "Misallocation and manufacturing TFP
in China and India." The Quarterly Journal of Economics 124.4 (2009): 1403:1448.
Julio, Brandon, and Youngsuk Yook. "Political uncertainty and corporate investment cycles." The Journal of Finance 67.1 (2012): 45-83.
Khwaja, Asim Ijaz, and Atif Mian. "Do lenders favor politically connected
firms? Rent provision in an emerging financial market." The Quarterly Journal of
Economics (2005): 1371-1411.
King, Robert, and Ross Levine. "Finance and growth: Schumpeter might be
right." The Quarterly Journal of Economics 108 (1993a): 717-738.
King, Robert, and Ross Levine. "Finance, entrepreneurship and growth." Journal of Monetary Economics 32 (1993b): 513-542.
Kornai, Janos.
"Resource-constrained versus demand-constrained systems."
Econometrica: Journal of the Econometric Society (1979):801-819.
La Porta, Rafael, Florencio Lopez-de-Silanes, and Andrei Shleifer. "Government
ownership of banks." The Journal of Finance 57.1 (2002): 265-301.
La Porta, Rafael, Florencio Lopez-de-Silanes, and Guillermo Zamarripa. "Related Lending." The Quarterly Journal of Economics 118.1 (2003): 231.
Li, Hongbin, and Li-An Zhou. "Political turnover and economic performance:
the incentive role of personnel control in China." Journal of public economics 89.9
(2005): 1743-1762.
Lucas, Deborah and Damien Moore. "Guaranteed Versus Direct Lending: The
Case of Student Loans" Measuring and Managing Federal Financial Risk, edited by
D. Lucas, University of Chicago Press (2010).
Lucas, Deborah. "Credit Policy as Fiscal Policy," MIT manuscript (2012b).
MacRae, C. Duncan. "A political model of the business cycle." The Journal of
PoliticalEconomy (1977): 239-263.
Mankiw, N. Gregory. "The optimal collection of seigniorage: Theory and evidence." Journal of Monetary Economics 20.2 (1987): 327-341.
Maskin, Eric, Yingyi Qian, and Chenggang Xu. "Incentives, information, and
organizational form." The Review of Economic Studies 67.2 (2000): 359-378.
Nordhaus, William D. "The political business cycle." The Review of Economic
Studies (1975): 169-190.
Rajan, Raghuram, and Luigi Zingales. "Financial dependence and growth." The
American Economic Review 88 (1998): 559-586.
46
Ramey, Valerie A., and Matthew D. Shapiro. "Costly capital reallocation and
the effects of government spending." Carnegie-RochesterConference Series on Public
Policy. Vol. 48. (1998).
Ramey, Valerie A. "Identifying government spending shocks: it's all in the timing." working paper, National Bureau of Economic Research, (2009).
Rotemberg, Julio J., and Michael Woodford. "Oligopolistic pricing and the effects of aggregate demand on economic activity." Journal of political Economy (1992):
1153-1207.
Sachs, Jeffrey, and Alberto Alesina. "Political Parties and the Business Cycle in
the United States, 1948-1984." Journal of Money, Credit, and Banking 20 (1988):6382.
Sapienza, Paola. "The effects of government ownership on bank lending." Journal
of FinancialEconomics 72.2 (2004): 357-384.
Schwarz, Anita M. "How Effective are Directed Credit Policies in the United
States?: A Literature Survey." Vol. 1019. World Bank Publications (1992).
Shaffer, Sherrill, and Robert N. Collender. "Federal Credit Programs and Local
Economic Performance." Economic Development Quarterly 23.1 (2009): 28-43.
Shleifer, Andrei. "State vs. Private Ownership." Journal of Economic Perspectives (1998): 133-150.
Shleifer, Andrei, and Robert W. Vishny. "Politicians and firms." The Quarterly
Journal of Economics (1994): 995-1025.
Shleifer, Andrei, and Robert W. Vishny. "The grabbing hand: Government
pathologies and their cures." Cambridge, MA: Harvard University Press (2002).
Stiglitz, Joseph E., and Andrew Weiss. "Credit rationing in markets with imperfect information." The American Economic review (1981): 393-410.
Townsend, Robert M. "Optimal contracts and competitive markets with costly
state verification." Journal of Economic Theory 21.2 (1979): 265-293.
Williamson, Stephen D. "Do informational frictions justify federal credit programs?." Journal of Money, Credit and Banking (1994): 523-544.
Wooldridge, Jeffrey M. "Econometric Analysis of Cross Section and Panel Data."
Cambridge, MA: MIT Press (2002).
47
Figure 1-1: CDB Outstanding Loan Amount and New Issuance
CDB Total City Loan Amount
CDB City Loan New Issuance
S0~
8
8
0:
5-
1998 1999
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Year
Total outstanding Loan
Loan for Industry Firms
-
1998 1999
Loan for Intastructure
4
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Year
Total Issuance
Issuance for Industry Firms
-0-
CDB Province Loan New Issuance
CDB Total Province Loan Amount
-
S
Issuance for Infrastruclure
4
8
S0~
0
S0~
0-
0
1997
1999
2001
2003
2005
Year
Total Outstanding Loan
Loan for Industry Firms
-6---
2007
2009
2011
2013
1997
Loan for Intastructure
-
2001
2003
2005
Year
Total Issuance
Issuance for Industry Firms
---
Ratio of Infrastructure Loan to Total Loan Amount
/
1999
2007
2009
2011
2013
Issuance for Infrastructure
-
Ratio of Industry Loan to Total Loan Amount
'---a-.
-C
9
A
-A
-
04
C
0-1
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Year
--
a--
----
Top 25 Percentile
4
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Year
Top 25 Percentile
----
Median Ratio
Bottom 25 Percentile
-
--
-+-
Median Ratio
Bottom 25 Percentile
Figure 1-1 includes the plots of CDB outstanding loan amount and new issuance. The loans can
be separated into infrastructure loan and the loan for industry firms. Infrastructure includes transportation(e.g.,road, railway, airport, bridge, tunnel), water supply, energy supply(e.g., gas,electric),
telecommunication and public service(e.g., Sewage discharge) The top two panels are the loans at
city level from 1998 to 2010. City level loan doesn't include the province level projects even though
part of the project is located in the city such as high way. The middle 2 panels are the loans at
province level from 1998 to 2013. The unit is billion RMB. The bottom 2 panels are ratios of infrastructure loan and industry loan to the total city level loan amount respectively. The solid lines
are median ratio among 310 cities and dash lines are top and bottom quartile of ratios among 310
cities each year.
48
Figure 1-2: Political Turnover in China
Histogram of City Secretaries' Tenure
0
0
-
LL
1
2
3
4
5
6
Length of Tenure
7
8
9
10
Histogram of City Secretaries' Turnover Year
0
U-
1997
1999
2001
2003
2005
2007
Turnover Year
2009
2011
2013
City Secretaries' Turnover Year(Tenure is not 5 years)
-
C1.
1997
1999
2001
2007
2003
2005
Turnover Year
2009
2011
2013
Figure 1-2 includes the plots of city level political turnover in China. The top panel is the histogram of city secretaries
term length from 1997 to 2013 among 334 cities and 1,227 city secretaries. I round the monthly turnover data into
year frequency by using June as the cutoff. 43% of the city secretaries leave the city in their fifth year. The middle
panel is the histogram of the turnover year of all city secretaries from 1997 to 2013 among 334 cities and 1,227 city
secretaries. The bottom panel is the histogram of the turnover year of the sub sample of city secretaries whose term
is not 5 years. It is also from 1997 to 2013 among 306 cities and 689 city secretaries. The bottom penal shows that if
the tenure is not 5 years (e.g, 4 or 6 years), the turnover usually happens in national turnover years.
49
Figure 1-3: Local Government Borrowing
City Secretary's Borrowing Pattern
0
C-4
a)C-
I
I
0)-
0
5
PoliticianYear
Cycle
10
II
CD0
0
E
M
0
-j
LO
0D
15
----- Loan Amount w/o Fixed Effects
Figure 1-3 is the plots of logarithm of CDB city total loan amounts after taking out the year fixed
effects, city fixed effects and politician fixed effects. The horizontal axis is the cycle of city secretary
turnover. There are 3 cycles in total and every cycle has 5 years. From 1998 to 2010, there are
3 national turnover cycles: 1997 to 2002, 2003 to 2007, and 2007 to 2012. The left vertical axis
PolitianYearis the number of years that city secretary has been staying in the city, which is from 1
to 5 years. For example, the first cycle (1 to 5 on horizontal axis) is from 1998 to 2002. I cluster the
cities by PolitianYear(1to 5 years) from 1998 to 2002 and plot the average CDB city loan amounts
for each bin of PolitianYear. I do the same thing for the second cycle from 2003 to 2007 which is
from 6 to 10 in the horizontal line. For the third cycle is from 2008 to 2010 which is from 11 to 15
in the horizontal line.
50
Figure 1-4: Manufacturing Firms in China
Number of Firms in Chinese Manufacturing Sector
C~)
-
0
LL 0
0 C
00
z
I
I
I
I
1998 1999 2000 2001
I
I
2002 2003 2004 2005 2006 2007 2008 2009
Year
SOEs Number
Foreign Firm Number
0
A
*
Private Firm Number
Aggregate Assets in Chinese Manufacturing Sector
C%1
.
0
C'-
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Year
S SOEs Asset
A
Private Firm Asset
Foreign Firm Asset
Figure 1-4 includes the plots of aggregate firm assets and firm numbers in manufacturing
sector of China. The data is the Chinese Industry Census(CIC) from 1998 to 2009 which includes
all the manufacturing firms with annual sales more than 5 million RMB($ 700,000). The top
panel is the firm numbers of SOEs, private firms and foreign firms from 1998 to 2009. The unit is
thousand. The bottom panel is the aggregate firm assets of SOEs, private firms and foreign firms
from 1998 to 2009. The unit is trillion RMB.
51
Figure 1-5: SOEs' Fixed Asset Patterns
Logarithm of Fixed Asset, High CDB Loan
Logarithm of Fixed Asset, Low CDB Loan
a,-
Nl
OD
1998 1999 2006 260
2002 20
2Cr04 2005 2006 2007 2008 2D'09
Year
Early Years(1 to 3)
0.
-
--
1998 1999 2000 2001
Later Years(4 to 6)
2002
2003 2004 205 2006
Year
Early Years(1 to 3)
Logarithm of Fixed Asset, High Corp Tax Rate
--
2007 208 2009
Later Years(4 to 6)
0--
Logarithm of Fixed Asset, Low Corp Tax Rate
Wt.
A
~
,/
M
4k/
1998
1999 2000
2001
~-..
2002 2003
Earty Years(1 to 3)
2004 2005 2006 2007
Year
--
2008 2009
,-
,-
1998 1999 2000 2001
&r-Later Years(4 to 6)
S
Logarithm of Fixed Asset, Large Developed Land
~
20
-- 4
-
I
2003 2004 2005
Year
Early Years(1 to 3)
-- 4
-
--
2006 2007 200
2009
Later Year( o6
Logarithm of Fixed Asset, Small Developed Land
al~
,
0
re
1968 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Year
--e
Earty Years(1 to 3)
---
1998
Later Years(4 to 6)
1999 2000
2001
2002 2003 2004 2005
Early Years(1
Year
to 3)
----
2006 2007 2008
2009
Later Years(4 to 6)
Figure 1-5 includes the plots of patterns of average SOEs' fixed assets from 1998 to 2009 in the cities
where secretaries are in the early terms and late terms. The vertical axis is the average logarithm
of SOEs' fixed assets. The top 2 panels are stratified on CDB loan amount. The left panel is the
pattern for high CDB industry loan cities and the right panel is the pattern for high CDB industry
loan cities. "high" and "low" categories are defined as the top and bottom quartile of CDB city level
outstanding industry loan amount distribution among 310 cities each year. The middle 2 panels are
stratified on average effective corporate tax rate in the city. The left panel is the pattern for high
effective corporate tax rate cities and the right panel is the pattern for low effective corporate tax rate
cities. "high" and "low" categories are defined as the top and bottom quartile of average effective
corporate tax rate distribution among 310 cities each year. The bottom 2 panels are stratified on
city developed land amount. The left panel is the pattern for cities with high developed land and
the right panel is the pattern for the cities with low developed land. "high" and "low" categories
are defined as the top and bottom quartile of developed land amount distribution among 310 cities
each year. The dash lines are for the cities where secretaries are in the early terms(1 to 3 year). The
solid lines are for the cities where secretaries are in the late terms(4 to 6 year).
52
,
Figure 1-6: SOEs' Employment Patterns
Logarithm of Employment, Low CDB Loan
Logarithm of Employment, High CDB Loan
N
Wi
*
I'
-
In
/
~6---
p
I
/
I
/
I
4
I
I
-I
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Year
1998 1999 2000 201 2002 2003 2004 2005 2006 2007 2008 2009
Year
Early Years(1 to 3)
Later Years(4 to 6)
6-+-
Early Years(1
----
to 3)
Later Years(4 to 6)
-----
Logarithm of Employment, Low Corp Tax Rate
Logarithm of Employment, High Corp Tax Rate
N.
Ln
C4
W)
9|
6
\k
/
\
ID
4,
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Year
1998 1999 2000 2001 2 02 2003 2004 2005 206 2007 2008 2009
Year
---
Early Years(1 to 3)
--
Early Years(1 to 3)
Later Years(4to6)
6--
Logarithm of Employment, Large Developed Land
ci.
*0
Later Years(4 to 6)
Logarithm of Employment, Small Developed Land
A ,,
--
*
0
/
,
-- 0--
,/
I
/
/
/
E
I
/
-
~0
I
/
I
Q
1W
\~
U
I
/
I
6
2005 2006 2007 2008 2009
1998 1999 2000 2001 2002 2003 2004
1998 1999 2000 201 2 02 2003 200 2005 2 06 2 07 2 08 2009
Year
Early Years(1 to 3)
----
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Year
Later Years(4 to 6)
Early Years(1 to 3)
--
6--
Later Years(4 to
6)
Figure 1-6 includes the plots of patterns of average SOEs' number of employees from 1998 to 2009
in the cities where secretaries are in the early terms and late terms. The vertical axis is the average
logarithm of SOEs' employee numbers. The top 2 panels are stratified on CDB loan amount. The
left panel is the pattern for high CDB industry loan cities and the right panel is the pattern for high
CDB industry loan cities. "high" and "low" categories are defined as the top and bottom quartile
of CDB city level outstanding industry loan amount distribution among 310 cities each year. The
middle 2 panels are stratified on average effective corporate tax rate in the city. The left panel is the
pattern for high effective corporate tax rate cities and the right panel is the pattern for low effective
corporate tax rate cities. "high" and "low" categories are defined as the top and bottom quartile of
average effective corporate tax rate distribution among 310 cities each year. The bottom 2 panels are
stratified on city developed land amount. The left panel is the pattern for cities with high developed
land and the right panel is the pattern for the cities with low developed land. "high" and "low"
categories are defined as the top and bottom quartile of developed land amount distribution among
310 cities each year. The dash lines are for the cities where secretaries are in the early terms(1 to 3
year). The solid lines are for the cities where secretaries are in the late terms(4 to 6 year).
53
Figure 1-7: Shifts of CDB Industry Loan Over Time
CDB Outstanding Loan Issuance in 1998 (BIllion
E eipowwem
nd
ha
RMB)
WOW pvOmton udWWPy
cow Mk" WWdDneoftg
Parolimm Idnaua eaIrmdo
Opr
an rdrkv
-
roduttsnUme0w0r
w
cheicatmetaM
Tr.60owrbda0qudmut
nuaamt
T40u3wr proztionanda
dn-mean-
*
cO.i~ua..IfnnuJ~uIra
m'' 1 pgdu
0
100
00
200
5O
400
600
CDB Outstanding Loan issuance in 2010 (Billion RMB)
Ekwtric pou,
om
anrd hot wam producton nd gappIV
PffostjbOurdmntr
Nfinram60Nfr*mm
Tmn-traus
N"
M6*ude wpmrV4
COMMOee
prduandsapmef
0
Me
4W0
6no
60
I
100
140
160
180
Figure 1-7 are plots of top 10 industries which have the biggest loan issuance from CDB. Data is
restricted to CDB province level industry loan data. There are 41 manufacturing industries in total
among 31 provinces in China. The top penal shows the CDB's biggest 10 industries in 1998. The
amount of each industry is the sum of all CDB loan issuance amounts from all 31 provinces in China.
The bottom penal shows the CDB's biggest 10 industries in 2010. The amount of each industry is
the sum of all CDB loan issuance amounts from all 31 provinces in China. The unit is billion RMB.
54
Table 1.1: Summary Statistics Data
Panel A: City Data
Variable
LoanCity
IssuanceCity
Loan _ INF _City
IssuanceINF_City
LoanINDCity
Issuance _IND_ City
N
3,605
3,605
3,605
3,605
3,605
3,605
3,587
3,568
3,587
3,579
3,605
3,325
3,605
3,158
Min
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
1.00
32.00
0.00
0.00
Max
1,268.20
0.01
0.08
S.D.
69.01
26.76
30.37
12.21
51.54
18.77
782.80
8,047.79
60.98
18,000.38
1.32
4.22
0.10
0.28
3,605
0.38
0.48
0.00
1.00
N
44,733
44,733
Mean
8.09
2.81
S.D.
43.33
16.23
Min
0.00
0.00
Max
1,369.09
1,004.12
LoanRoad
Loan Rail
496
496
154.18
51.44
163.63
102.32
0
0
744.87
1260.99
Loan_ Water
496
12.60
23.95
0
203.29
Loan Tel
496
8.55
41.03
0
419.93
Panel C: Firm Data
Variable
N
Mean
S.D.
Min
Max
LogAsset
LogFixAst
2,949,514
2,933,323
9.72
8.46
1.48
1.77
0.00
0.00
20.16
19.11
LogWorker
LogDebt
2,944,543
2,930,818
4.69
8.98
1.17
1.73
0.00
0.00
12.58
19.32
ROA
Log(Sales/Worker)
LogSales
2,949,502
2,918,158
2,931,478
0.09
5.25
9.95
0.20
1.24
1.46
-0.76
-8.12
0.00
2.44
17.38
19.24
Tax_ Corp
1,520,597
19.41
10.24
0.00
61.54
Tax VAT
1,356,623
15.09
13.92
0.00
86.91
GDP
AvgIncome
FiscalIncome
Employment
PoliticianYear
Age
Gender
Relation
Promotion
Panel B: Province Data
Variable
Loan_PI
Issuance_PI
Mean
31.97
11.31
10.92
4.20
21.34
7.13
559.46
10,799.02
30.43
1,108.81
2.49
50.11
428.79
593.74
195.03
847.77
284.22
10,604.48
139,574.00
1,138.31
329,858.30
6.00
62.00
1.00
1.00
Panel A restricted to 310 cities from 1998 to 2010. LoanCity and IssuanceCity are the city level total CDB
outstanding loan amount and new issuance. LoanINFCity
and Issuance INFCity
are the city level total
CDB outstanding loan amount and new issuance for infrastructure. LoanINDCity
and IssuanceIND_City
are the city level total CDB outstanding loan amount and new issuance for industry firms. The unit of LoanCity to
IssuanceIND_City is 100 million RMB. GDP, AvgIncome, FiscalIncome and Employment are city level GDP,
income per capita, fiscal income and total employment respectively. Age is the age of the city secretary. Gender
is the dummy for whether the city secretary is a female. Relation is the dummy for whether the city secretary
is born at the same place as the province governor. Promition is the dummy for whether the city secretary is
promoted or not at the end of the term. Panel B restricted to 31 provinces from 1998 to 2013 among 95 industries.
Loan_PI and Issuance_PIare the province level CDB industry outstanding loan amount and new issuance. There
are 31 provinces and 95 industries in total. The total observation number is 44,733 which covers all 95 industries.
Among these 95 industries, CDB added 11 new industries in 2005 and doesn't have data before 2005. LoanRoad,
LoanRail, LoanWater, and LoanTel are the province level CDB industry outstanding loan amounts for road
construction, railway construction, water system, and Telecommunication. The number of observations is 496 which
covers 31 provinces from 1998 to 2013. The unit is in 100 million RMB.Panel C restricted to all manufacturing firms
in Chinese Industry Census data from 1998 to 2009. LogAsset is the logarithm of each firm's total asset. Tax Corp
is the corporate tax rate of each firm every year. TaxVAT is the value-added tax rate of each firm every year.For
more detailed variable definition and construction, please see Appendix Table 1.10.
55
Table 1.2: CDB Loan's Effect on Firms: Evidence from OLS Regressions
Panel A: SOEs
Dependent Variable
Log(Loan_ PI)
GDPt_
1
AvgIncomet-
1
FiscalIncomet-1
Employmentt- 1
Fixed Effects
Observations
R-squared
Panel B:Private Firms
Dependent Variable
Log(LoanPI)
GDPt-1
AvgIncomet -1
FiscalIncomet -I
Employmentt_1
Fixed Effects
Observations
R-squared
Panel C:Private Firms
Dependent Variable
Log(Loan INF City)
GDPt_1
AvgIncomet-i
FiscalIncomet
Employment_
Fixed Effects
Observations
R-squared
-1
(1)
LogAsset
(4)
(3)
(2)
LogFixAst LogWorker LogDebt
0.006***
(0.002)
-0.055***
(0.014)
0.002
(0.001)
0.156
(0.161)
0.001
(0.001)
Yes
231,682
0.093
0.003
(0.003)
-0.117***
(0.020)
0.001
(0.001)
0.393*
(0.220)
(1)
LogAsset
(5)
ROA
(6)
(7)
Log(Sales/Worker) LogSales
-0.002***
(0.001)
0.011***
(0.003)
0.000
(0.000)
-0.174***
(0.038)
-0.000
(0.000)
Yes
231,682
0.008
-0.010***
(0.003)
-0.010
(0.015)
0.000
(0.001)
-0.570***
(0.174)
0.002*
(0.001)
Yes
225,214
0.189
(4)
(3)
(2)
LogFixAst LogWorker LogDebt
(5)
ROA
(6)
(7)
Log(Sales/Worker) LogSales
-0.003**
(0.001)
-0.013**
(0.006)
0.000
(0.000)
-0.324***
(0.062)
0.001**
(0.000)
Yes
756,826
0.219
-0.007***
(0.002)
-0.049***
(0.009)
-0.002***
(0.001)
-0.423***
(0.086)
0.000
(0.001)
Yes
753,444
0.107
-0.004***
(0.000)
-0.001
(0.001)
-0.000
(0.000)
-0.209***
(0.012)
0.000
(0.000)
Yes
755,947
0.033
-0.016***
(0.001)
0.025***
(0.006)
-0.001**
(0.000)
-1.197***
(0.058)
0.000
(0.000)
Yes
756,820
0.225
(1)
LogAsset
(4)
(3)
(2)
LogFixAst LogWorker LogDebt
(5)
ROA
(6)
(7)
Log(Sales/Worker) LogSales
0.013***
(0.003)
-0.046***
(0.006)
-0.000
(0.000)
-0.123**
(0.055)
0.694
(0.688)
Yes
1,022,595
0.215
0.012***
(0.004)
-0.041***
(0.008)
-0.001***
(0.000)
-0.294***
(0.078)
2.804***
(0.912)
Yes
1,017,890
0.108
0.002***
(0.001)
-0.014***
(0.002)
-0.000
(0.000)
-0.136***
(0.012)
1.092***
(0.156)
Yes
1,022,586
0.058
0.041***
(0.003)
-0.139***
(0.006)
0.000
(0.000)
-0.290***
(0.055)
3.445***
(0.627)
Yes
1,021,059
0.245
-0.001*
(0.001)
Yes
230,283
0.040
0.010***
0.017***
(0.002)
-0.046***
(0.013)
0.001*
(0.001)
0.309**
(0.140)
-0.001
(0.001)
Yes
232,003
0.028
(0.003)
-0.106***
(0.018)
0.002*
(0.001)
0.215
(0.203)
0.000
(0.001)
Yes
229,696
0.054
-0.007***
(0.001)
0.001
(0.005)
0.000
(0.000)
-0.306***
(0.054)
0.001**
(0.000)
Yes
757,033
0.013
-0.010***
(0.002)
0.004
(0.005)
-0.000
(0.000)
-0.350***
(0.050)
1.295**
(0.544)
Yes
1,022,954
0.034
0.003*
(0.002)
-0.015*
(0.008)
0.002***
(0.001)
-0.036
(0.079)
0.002***
(0.001)
Yes
752,871
0.081
0.009**
(0.004)
-0.063***
(0.008)
0.000
(0.000)
0.078
(0.072)
-0.360
(0.945)
Yes
1,016,623
0.091
-0.001
(0.003)
-0.060***
(0.018)
0.001
(0.001)
-0.311
(0.197)
0.001
(0.001)
Yes
227,342
0.152
-0.022***
(0.001)
0.025***
(0.007)
-0.000
(0.000)
-1.499***
(0.067)
0.001**
(0.000)
Yes
755,687
0.273
0.032***
(0.003)
-0.137***
(0.007)
-0.000
(0.000)
-0.627***
(0.059)
4.630***
(0.702)
Yes
1,021,245
0.281
Note. In Penal A and B, Loan_ PI is at provincex industryx year level. It is the CDB industry loan (to SOEs)
amount at each of the 31 provinces and 41 manufacturing industries. In Panel A, data is restricted to SOEs in CIC
data from 1998 to 2009. It shows the OLS regression results by estimating equation 1.1. In Panel B, data is restricted
to private firms in CIC data from 1998 to 2009. GDP, AvgIncome, FiscalIncome and Employment are province
level GDP, income per capita, fiscal income and total employment respectively. All columns are controlled by year
fixed effect and firm fixed effect. Standard errors are clustered at firm level.In Panel C, LoanINFCity is the
CDB infrastructure loan amount at city x year level. Data is restricted to private firms in CIC data from 1998 to
2009 among 310 cities. GDP, AvgIncome, FiscalIncome and Employment are city level GDP, income per capita,
fiscal income and total employment respectively. All columns are controlled by year fixed effect and firm fixed effect.
Standard errors are clustered at firm level.
56
Table 1.3: Exogeneity of City Secretary Turnover Timing
NationalCycle
(1)
Coefficient
(2)
Coefficient
(3)
Coefficient
(4)
Coefficient
(5)
Hazard Ratio
0.366***
(0.061)
0.340***
(0.064)
0.019**
(0.008)
0.363***
0.332***
1.393
(0.063)
0.087
(0.462)
(0.065)
0.018**
(0.008)
0.133
(0.297)
0.044
(0.459)
-0.164
-0.118
(0.536)
0.002
(0.011)
-0.003
(0.012)
(0.537)
-0.000
(0.011)
-0.000
(0.012)
0.769
2.167
2.755
(0.857)
(4.034)
(3.936)
-1.658
-2.457
(4.720)
-0.008
(0.010)
0.009
(0.010)
0.028
(0.043)
-0.025
(0.041)
(4.634)
-0.006
(0.009)
0.007
(0.009)
0.033
(0.048)
-0.030
(0.045)
2,878
36.51
2,652
Age
Gender
0.123
GDPt_1
(0.293)
-0.065
(0.067)
-0.058
(0.064)
GDPt-2
-0.001
(0.003)
AvgIncomet-1
-0.001
(0.004)
AvgIncomet-2
0.807
(0.837)
FiscalIncomet -1
FiscalIncomet -2
0.000
(0.001)
Employmentt-1
0.000
(0.001)
Employment t-2
Log(LoanCityt_1)
0.001
(0.019)
2
)
Log(LoanCityt-
0.002
(0.018)
Observations
3,012
Chi Squared
36.84
2,775
38.07
34.91
1.019
1.142
1.045
0.888
1.000
1.000
15.722
0.086
0.994
1.007
1.033
0.970
2,652
34.91
Note. The regressions are estimated at city x year level. Data restricted to 1,106 city secretaries
among 310 cites from 1998 to 2011. Table shows the results from Cox proportional hazard regression by following Wooldridge(2002). Estimated coefficients are reported in column 1 to 4. Age
is the city secretary's age. Gender is the dummy for whether the city secretary is female or not.
NationalCycle is the dummy for whether it is the year of national congress party(1998, 2003 and
2008). GDP, AvgIncome, FiscalIncome and Employment are city level GDP, income per capita,
fiscal income and total employment respectively. Log(Loan_ City) is the logarithm of CDB city level
total outstanding loan amount. Column 1 and 2 only controls GDP, AvgIncome, FiscalIncome
and Employment one year ago. Column 3 and 4 controls GDP, AvgIncome, FiscalIncome
and Employment in the last two years. The time-varying variables are Age, NationalCycle,
GDAvgIncome, FiscalIncome, Employment and Log(LoanCity). I assume these time-varying
variables are constant during one year. Gender is time-invariant. Column 2 and 4 includes all the
time-varying and time-invariant variables. Column 1 and 3 excludes Gender and Age of the city
secretaries since these two variables has many missing values. Column 5 is the estimated hazard
ratio from column 4. Standard errors are clustered at city secretary's level.
57
Table 1.4: City Secretary Turnover Timing's Effect on Borrowing from CDB
Panel A
Total Loan
(1)
(2)
Dependent Variable LogLoanCity LogIssueCity
PoliticianYear
GDPt-1
AvgIncomet -1
FiscalIncomet -1
Employmentt-i
Fixed Effects
Observations
R-squared
-0.364***
(0.020)
-0.269*
(0.138)
-0.004**
(0.002)
0.874
(1.197)
-0.954
(5.875)
Yes
3,127
0.853
-0.308***
(0.029)
0.018
(0.167)
-0.001
(0.007)
-1.040
(1.428)
11.682**
(5.870)
Yes
2,795
0.685
Loan for Infrastructure
Loan for Industry Firms
(3)
(4)
LogLoanCity LogIssueCity
(6)
(5)
LogLoanCity LogIssueCity
-0.122***
(0.013)
-0.222*
(0.123)
0.004
(0.004)
-0.449
(0.817)
-2.004
(5.254)
Yes
2,449
0.884
-0.165***
(0.025)
0.091
(0.296)
-0.001
(0.007)
-2.549
(2.250)
3.287
(6.935)
Yes
2,227
0.588
-0.338***
(0.028)
-0.156
(0.145)
-0.005**
(0.002)
0.837
(1.309)
-9.066
(6.758)
Yes
2,727
0.756
-0.252***
(0.037)
-0.011
(0.205)
-0.004
(0.007)
1.552
(2.031)
-30.014***
(10.539)
Yes
2,323
0.597
Panel B
Total Loan
(1)
(2)
Dependent Variable LogLoanCity LogIssueCity
Year_2
Year_3
Year_4
Year_5
Year_6
GDPt-1
AvgIncomet 1
FiscalIncomet _i
EmploymenttFixed Effects
Observations
R-squared
1
-0.386***
(0.031)
-0.749***
(0.042)
-1.071***
(0.063)
-1.429***
(0.108)
-1.900***
(0.144)
-0.264*
(0.138)
-0.004**
(0.002)
0.859
(1.196)
-1.205
(5.867)
Yes
3,127
0.854
-0.357***
(0.057)
-0.660***
(0.074)
-0.912***
(0.089)
-1.318***
(0.168)
-1.581***
(0.196)
0.022
(0.167)
-0.001
(0.007)
-1.089
(1.432)
-11.934**
(5.917)
Yes
2,795
0.685
Loan for Infrastructure
Loan for Industry Firms
(3)
(4)
LogLoan City LogIssueCity
(6)
(5)
LogLoanCity LogIssueCity
-0.109***
(0.028)
-0.240***
(0.035)
-0.351***
(0.041)
-0.467***
(0.081)
-0.691***
(0.110)
-0.223*
(0.123)
0.004
(0.004)
-0.439
(0.836)
-1.976
(5.271)
Yes
2,449
0.884
-0.167***
(0.063)
-0.298***
(0.070)
-0.509***
(0.082)
-0.691***
(0.172)
-0.870***
(0.218)
0.087
(0.300)
-0.001
(0.007)
-2.556
(2.277)
3.167
(6.973)
Yes
2,227
0.588
-0.330***
(0.044)
-0.703***
(0.064)
-1.001***
(0.086)
-1.341***
(0.151)
-1.668***
(0.204)
-0.154
(0.145)
-0.005**
(0.002)
0.856
(1.316)
-8.969
(6.773)
Yes
2,727
0.756
-0.327***
(0.081)
-0.540***
(0.099)
-0.731***
(0.123)
-1.238***
(0.216)
-1.337***
(0.240)
-0.014
(0.202)
-0.004
(0.007)
1.461
(2.011)
-30.738***
(10.516)
Yes
2,323
0.599
Note. The regressions are estimated at city x year level. Data restricted to CDB city level loan to 310 cites from
1998 to 2010. Panel A shows the results from OLS regression from equation 1.3. PoliticianYear is the number
of years that the secretary has been staying in the city. Column 1 and 2 are the regressions on the logarithm
of total CDB loan outstanding amount and total issuance respectively. Column 3 and 4 are the regressions on
the logarithm of CDB infrastructure loan outstanding amount and new issuance respectively. Column 5 and 6 are
the regressions on the logarithm of CDB industry firm loan outstanding amount and new issuance respectively.
GDP, AvgIncom, FiscalIncome and Employment are city level GDP, income per capita, fiscal income and total
employment respectively. All columns are controlled by year fixed effect, city fixed effect and politician personal fixed
effect. Standard errors are clustered at city level. Panel B shows the results from OLS regression from equation 1.4.
Year_2 is the the dummy for whether it is the second of the secretary in the city. Year_3 is the the dummy for
whether it is the third of the secretary in the city. Year_4 to Year_6 are defined in the same way. The dummy for
year 1 is the missing category. All columns are controlled by year fixed effect, city fixed effect and politician personal
fixed effect. Standard errors are clustered at city level.
58
Table 1.5: City Secretary's Turnover Effects on Firms
Panel A: SOEs
Dependent Variable
Year_2
Year_3
Year_4
Year_5
Year_6
GDPt_ 1
AvgIncomet - 1
FiscalIncomet - I
Employmentt -1
Year Fixed Effect
Firm Fixed Effect
Politician Fixed Effect
City Fixed Effect
Observations
R-squared
(1)
LogAsset
(2)
LogFixAst
(3)
LogWorker
(4)
LogDebt
(5)
ROA
(6)
LogNumber
-0.019***
(0.007)
-0.032**
(0.013)
-0.047**
(0.020)
-0.075***
(0.027)
-0.091***
(0.034)
-0.021
(0.013)
0.000
(0.001)
-0.243**
(0.112)
-0.008**
(0.004)
Yes
Yes
Yes
No
382,317
0.108
-0.023***
(0.009)
-0.042**
(0.017)
-0.056**
(0.026)
-0.098***
(0.035)
-0.115***
(0.044)
-0.095***
(0.019)
0.000
(0.001)
0.031
(0.160)
-0.009
(0.007)
Yes
Yes
Yes
No
379,632
0.0523
-0.006
(0.006)
-0.021*
(0.012)
-0.025
(0.017)
-0.031
(0.023)
-0.044
(0.030)
-0.022*
(0.013)
0.000
(0.001)
0.197*
(0.112)
-0.002
(0.002)
Yes
Yes
Yes
No
381,720
0.0642
-0.028***
(0.010)
-0.043**
(0.019)
-0.065**
(0.028)
-0.097**
(0.038)
-0.128***
(0.048)
-0.081***
(0.018)
0.000
(0.001)
-0.105
(0.151)
-0.010
(0.006)
Yes
Yes
Yes
No
378,567
0.0647
0.010***
(0.002)
(0.007)
0.035***
(0.008)
0.003
(0.004)
-0.000
(0.000)
-0.107***
(0.027)
-0.001
(0.002)
Yes
Yes
Yes
No
382,318
0.0437
0.019
(0.015)
0.056***
(0.018)
0.076***
(0.024)
0.035
(0.049)
0.070
(0.050)
-0.035
(0.094)
0.004**
(0.002)
1.813
(1.278)
-0.003***
(0.001)
Yes
No
Yes
Yes
3,250
0.963
(1)
LogAsset
(2)
LogFixAst
(3)
LogWorker
(4)
LogDebt
(5)
ROA
(6)
LogNumber
-0.015***
(0.005)
-0.032***
(0.009)
-0.058***
(0.014)
-0.096***
(0.019)
-0.081***
(0.024)
-0.047***
(0.006)
-0.000
(0.000)
-0.132***
(0.051)
-0001
(0.001)
Yes
Yes
Yes
No
1,171,115
0.234
-0.008
(0.006)
-0.011
(0.012)
-0.022
(0.018)
-0.060**
(0.025)
-0.008
(0.031)
-0.039***
(0.008)
-0.001***
(0.000)
-0.257***
(0.073)
0.001
(0.001)
Yes
Yes
Yes
No
1,165,459
0.117
-0.004
(0.004)
-0.005
(0.008)
-0.013
(0.011)
-0.025*
(0.015)
0.002
(0.019)
0.003
(0.005)
-0.001**
(0.000)
-0.269***
(0.047)
-0.000
(0.001)
Yes
Yes
Yes
No
1,171,449
0.0317
-0.012*
(0.007)
-0.038***
(0.013)
-0.070***
(0.019)
-0.095***
(0.025)
-0.111***
(0.032)
-0.064***
(0.008)
0.000
(0.000)
0.072
(0.067)
0.000
(0.001)
Yes
Yes
Yes
No
1,163,749
0.103
-0.003***
(0.001)
-0.003
(0.002)
0.002
(0.003)
0.006
(0.004)
0.008
(0.005)
-0.008***
(0.001)
-0.000
(0.000)
-0.128***
(0.012)
-0.001*
(0.001)
Yes
Yes
Yes
No
1,171,106
0.0625
0.085***
(0.019)
0.144***
(0.021)
0.177***
(0.030)
0.212***
(0.056)
0.283***
(0.053)
-0.091
(0.115)
0.003
(0.004)
2.394
(1.690)
-0.001**
(0.001)
Yes
No
Yes
Yes
3,233
0.974
0.017***
(0.003)
0.024***
(0.005)
0.028***
Panel B: Private Firms
Dependent Variable
Year_2
Year_3
Year_4
Year_5
Year_6
GDPt_ 1
AvgIncomet 1
FiscalIncomet -1
Employmentt -1
Year Fixed Effect
Firm Fixed Effect
Politician Fixed Effect
City Fixed Effect
Observations
R-squared
59
Table 1.5: City Secretary's Turnover Effects on Firms(Continue)
Panel C: Foreign Firms
Dependent Variable
Year_2
Year_3
Year_4
Year_5
Year_6
(1)
LogAsset
(3)
(4)
(2)
LogFixAst LogWorker LogDebt
(5)
ROA
(6)
LogNumber
0.020***
(0.007)
0.028**
(0.014)
0.043**
(0.021)
0.059**
(0.029)
0.017*
(0.010)
0.034*
(0.019)
-0.001
(0.002)
-0.001
(0.003)
0.000
(0.005)
0.008
(0.007)
0.016**
(0.008)
-0.001
(0.002)
0.000*
(0.000)
-0.121***
(0.016)
-0.001
(0.001)
Yes
Yes
Yes
No
256,887
0.0292
0.063***
(0.020)
0.120***
(0.027)
0.127***
(0.040)
0.141*
(0.083)
0.311***
(0.064)
-0.093
(0.116)
0.001
(0.001)
1.952
(1.597)
-0.003***
(0.001)
Yes
No
Yes
Yes
2,976
0.968
0.110***
(0.036)
GDPt- 1
-0.019***
(0.007)
AvgIncomet-1j
0.000
(0.000)
FiscalIncomet _1
-0.485***
(0.062)
Employmentt_1-0.002
(0.007)
Year Fixed Effect
Yes
Firm Fixed Effect
Yes
Politician Fixed Effect Yes
City Fixed Effect
No
Observations
256,887
R-squared
0.175
0.056*
(0.029)
0.075*
(0.039)
0.164***
(0.050)
0.000
(0.010)
-0.000
(0.001)
-0.652***
(0.085)
-0.003
(0.008)
Yes
Yes
Yes
No
255,976
0.0541
0.009
(0.007)
0.039***
(0.013)
0.043**
(0.019)
0.081***
(0.026)
(0.033)
0.047***
(0.007)
0.001*
(0.000)
0.008
(0.011)
-0.001
(0.022)
-0.001
(0.032)
-0.003
(0.044)
0.018
(0.055)
-0.059***
(0.012)
0.001
(0.001)
-0.262***
-0.334***
(0.064)
-0.002
(0.005)
(0.094)
-0.002
(0.009)
Yes
Yes
Yes
No
255,513
0.0601
0.141***
Yes
Yes
Yes
No
256,652
0.0673
Note. Data restricted to manufacturing firms in Chinese Industry Census data from 1998 to 2009 among 310 cities.
Year_2 to Year_6 are at the city x year level. Panel A is for SOEs. Panel B is for private firms and Panel C is for
foreign firms. Column 1 is the logarithm of total asset of the firm. Column 2 is the logarithm of fixed asset of the firm.
Column 3 is the logarithm of total number of workers of the firm. Column 4 is the logarithm of total debt of the firm.
Column 5 is the ROA of the firm. Year_2 is the the dummy for whether it is the second of the secretary in the city
where the firm locates. Year_3 is the the dummy for whether it is the third of the secretary in the city. Year 4 to
Year__6 are defined in the same way. The dummy for year 1 is the missing category. GDP, AvgIncome, FiscalIncome
and Employment are city level GDP, income per capita, fiscal income and total employment respectively. Column
1 to 5 are controlled by year fixed effect, firm fixed effect and politician personal fixed effect. Standard errors are
clustered at firm level. Column 6 is the logarithm of total number of firms in each city every year. Column 6 is
controlled by year fixed effect, city fixed effect and politician personal fixed effect. Standard errors are clustered at
firm level.
60
Table 1.6: Opposing Effects of CDB City Infrastructure Loan and Industry Loan (Use
Secretary Year in Office as Instrument)
Panel A: Private Firms
Dependent Variable
Log(Loan _IND_ City)
Log(LoanINF
City)
GDPt_1
AvgIncomet_1
FiscalIncomet-1
Employmentt_
1
Year Fixed Effect
Firm Fixed Effect
Politician Fixed Effect
Observations
Wald F-stat
(1)
LogAsset
(4)
(3)
(2)
LogFixAst LogWorker LogDebt
(5)
ROA
(7)
(6)
Log(Sales/Worker) LogSales
-0.073***
(0.016)
0.246***
(0.033)
0.019*
(0.010)
-0.001***
(0.000)
-0.191***
(0.050)
0.814
(0.725)
Yes
Yes
Yes
851,720
474.3
-0.165***
(0.022)
0.236***
(0.046)
0.026*
(0.014)
-0.003***
(0.000)
-0.434***
(0.070)
3.296***
(0.959)
Yes
Yes
Yes
847,260
472.1
0.001
(0.005)
-0.023**
(0.009)
-0.017***
(0.003)
0.000
(0.000)
-0.148***
(0.011)
1.434***
(0.157)
Yes
Yes
Yes
851,712
474.4
-0.134***
(0.018)
0.544***
(0.037)
0.013
(0.011)
-0.002***
(0.000)
-0.526***
(0.052)
3.664***
(0.707)
Yes
Yes
Yes
850,374
478.0
(1)
LogAsset
(4)
(3)
(2)
LogFixAst LogWorker LogDebt
(5)
ROA
(7)
(6)
Log(Sales/Worker) LogSales
-0.104***
(0.023)
0.062**
(0.031)
0.015
(0.014)
-0.000
(0.000)
-0.404***
(0.073)
-3.892***
(1.169)
Yes
Yes
Yes
194,732
288.4
-0.146***
(0.034)
0.059
(0.046)
0.033
(0.020)
-0.001
(0.001)
-0.574***
(0.101)
-3.104*
(1.670)
Yes
Yes
Yes
194,108
284.1
-0.011
(0.007)
-0.001
(0.009)
-0.001
(0.004)
0.000*
(0.000)
-0.148***
(0.019)
0.312
(0.316)
Yes
Yes
Yes
194,732
288.4
-0.109***
(0.026)
0.167***
(0.040)
-0.022
(0.018)
0.002***
(0.000)
-0.459***
(0.080)
3.241**
(1.471)
Yes
Yes
Yes
194,313
282.4
-0.021
(0.014)
0.052*
(0.028)
0.021**
(0.009)
-0.001***
(0.000)
-0.366***
(0.045)
1.358**
(0.579)
Yes
Yes
Yes
852,039
477.2
0.005
(0.023)
0.161***
(0.047)
-0.027*
(0.015)
-0.000
(0.000)
0.137**
(0.066)
-1.474
(0.972)
Yes
Yes
Yes
846,406
471.5
-0.146***
(0.019)
0.576***
(0.039)
0.028**
(0.012)
-0.002***
(0.000)
-0.869***
(0.055)
4.944***
(0.784)
Yes
Yes
Yes
850,546
475.8
Panel B: Foreign Firms
Dependent Variable
Log(Loan _IND_ City)
Log(Loan
INF _City)
GDPt-1
AvgIncomet-_1
FiscalIncomet 1
Employmentt_1
Year Fixed Effect
Firm Fixed Effect
Politician Fixed Effect
Observations
Wald F-stat
0.038*
(0.022)
-0.176***
(0.032)
0.009
(0.014)
0.000
(0.000)
-0.230***
(0.076)
-2.718**
(1.156)
Yes
Yes
Yes
194,545
280.6
-0.100***
(0.036)
0.092*
(0.052)
-0.021
(0.024)
0.000
(0.001)
-0.285***
(0.109)
-2.293
(1.944)
Yes
Yes
Yes
193,752
283.4
-0.070**
(0.029)
-0.011
(0.040)
-0.014
(0.017)
0.003***
(0.001)
-0.680***
(0.091)
0.339
(1.492)
Yes
Yes
Yes
194,583
289.1
Note. Table 1.6 are the two stage least squares results by using Year_ 1 to Year_6 as instrumental variable for
logarithm of CDB loan for infrastructure and industry SOEs at cityxyear level. Log(Loan_INF_City) is the
logarithm of the aggregate loan for infrastructure at city x year level. Log(Loan_IND_City) is the logarithm of
the aggregate loan for industry firms at city x year level. Panel A restricted to the private firms in CIC from 1998 to
2009 among 310 cities. Panel B restricted to the foreign firms in CIC from 1998 to 2009 among 310 cities. In column
5, LogSales/Worker is the logarithm of Sales per worker. GDP, AvgIncome, FiscalIncome and Employment are
city level GDP, income per capita, fiscal income and total employment respectively. All columns are controlled by
year fixed effect, firm fixed effect and politician personal fixed effect. Standard errors are clustered at firm level.
Cragg-Donald Wald F-statistics for weak identification tests are reported.
61
Table 1.7: CDB Province Industry Loan's Effect on Firms within Same Industry
(2SLS)
Panel A: Private Firms
Dependent Variable
Log(LoanPI)
GDPti
AvgIncomet - 1
FiscalIncomet-1
Employmentt- 1
Year Fixed Effect
Firm Fixed Effect
Observations
Wald F-stat
(1)
LogAsset
(3)
(4)
(2)
LogFixAst LogWorker LogDebt
(5)
ROA
(6)
(7)
Log(Sales/Worker) LogSales
-0.246***
(0.034)
-0.040***
(0.006)
0.001***
(0.000)
0.475***
(0.097)
0.000
(0.001)
Yes
Yes
693,782
52.15
-0.213***
(0.044)
-0.073***
(0.007)
-0.001***
(0.000)
0.317**
(0.125)
-0.000
(0.001)
Yes
Yes
690,071
52.71
0.020**
(0.009)
0.003***
(0.001)
-0.000
(0.000)
-0.256***
(0.024)
0.000*
(0.000)
Yes
Yes
693,774
52.11
-0.251***
(0.037)
0.014***
(0.005)
-0.000
(0.000)
-0.511***
(0.099)
-0.000
(0.000)
Yes
Yes
692,560
52.58
(1)
LogAsset
(4)
(3)
(2)
LogFixAst LogWorker LogDebt
(5)
ROA
(6)
(7)
Log(Sales/Worker) LogSales
0.196***
(0.017)
-0.050***
(0.012)
0.001
(0.001)
0.027
(0.138)
0.002
(0.001)
Yes
Yes
230,336
88.78
0.279***
(0.022)
-0.109***
(0.017)
0.001
(0.001)
0.207
(0.188)
0.001
(0.001)
Yes
Yes
228,818
87.91
-0.001
(0.004)
0.011***
(0.003)
0.000
(0.000)
-0.173***
(0.030)
-0.000
(0.000)
Yes
Yes
230,335
88.23
0.033*
(0.019)
-0.011
(0.012)
-0.000
(0.001)
-0.562***
(0.139)
0.002**
(0.001)
Yes
Yes
222,680
84.03
-0.140***
(0.027)
-0.016***
(0.004)
0.001***
(0.000)
0.168**
(0.077)
0.000
(0.000)
Yes
Yes
693,966
51.31
-0.290***
(0.047)
-0.043***
(0.007)
0.002***
(0.000)
0.803***
(0.131)
0.001
(0.001)
Yes
Yes
689,685
51.22
-0.380***
(0.042)
-0.002
(0.006)
0.000
(0.000)
-0.370***
(0.116)
0.000
(0.001)
Yes
Yes
692,874
52.39
Panel B: SOEs
Dependent Variable
Log(LoanPI)
GDPt-1
AvgIncomet -1
FiscalIncomet -1
Employmentt-1
Year Fixed Effect
Firm Fixed Effect
Observations
Wald F-stat
0.187***
(0.016)
-0.045***
(0.011)
0.001
(0.001)
0.197
(0.121)
0.000
(0.001)
Yes
Yes
229,774
91.84
0.304***
(0.024)
-0.097***
(0.016)
0.002*
(0.001)
0.024
(0.175)
0.002
(0.002)
Yes
Yes
228,127
87.17
0.181***
(0.020)
-0.059***
(0.015)
0.000
(0.001)
-0.397**
(0.166)
0.002**
(0.001)
Yes
Yes
225,486
81.87
Note. Loan_PI is the outstanding loan amount at province x industry x year level. In Panel A, data is restricted
to private firms in CIC data from 1998 to 2009. Panel A shows the two stage least squares regression results by
using Firsth to Sixth as instrumental variable for logarithm of CDB province level outstanding loan amount in 41
manufacturing industries and 27 provinces (exclude Beijing, Shanghai, Tianjing, and Chonqing). GDP, AvgIncome,
FiscalIncome and Employment are province level GDP, income per capita, fiscal income and total employment
respectively. All columns are controlled by year fixed effect and firm fixed effect. Standard errors are clustered at
firm level. Cragg-Donald Wald F-statistics for weak identification tests are reported.
In Panel B, data is restricted to SOEs in CIC data from 1998 to 2009. Panel B shows the two stage least squares
regression results by using First to Sixth as instrumental variable for logarithm of CDB province level outstanding
loan amount in 41 manufacturing industries and 27 provinces (exclude Beijing, Shanghai, Tianjing, and Chonqing).
Loan_PI is the outstanding loan amount at province-industry level every year. GDP, AvgIncome, FiscalIncome
and Employment are province level GDP, income per capita, fiscal income and total employment respectively. All
columns are controlled by year fixed effect and firm fixed effect. Standard errors are clustered at firm level. CraggDonald Wald F-statistics for weak identification tests are reported.
62
Table 1.8: Upstream Industry Loan's Effect on Firms in Downstream Industry (2SLS)
Panel A: Private Firms
Dependent Variable
Log(Upstream _Loan)
GDPt-
1
AvgIncomet-
1
FiscalIncomet- 1
Employmentt-
1
Year Fixed Effect
Firm Fixed Effect
Observations
Wald F-stat
(1)
LogAsset
(2)
(3)
(4)
LogFixAst LogWorker LogDebt
(5)
ROA
(6)
(7)
Log(Sales/Worker) LogSales
0.092***
(0.011)
-0.011**
(0.005)
0.000
(0.000)
-0.394***
(0.057)
0.001***
(0.000)
Yes
Yes
804,718
581.3
0.136***
(0.015)
-0.034***
(0.006)
-0.001
(0.003)
0.004***
(0.001)
0.000
(0.000)
-0.235***
(0.013)
0.000**
(0.000)
Yes
Yes
804,709
580.0
0.064***
(0.012)
(0.004)
-0.001***
(0.000)
-1.333***
(0.053)
0.001**
(0.000)
Yes
Yes
803,296
580.2
(1)
LogAsset
(2)
(3)
(4)
LogFixAst LogWorker LogDebt
(5)
ROA
(6)
(7)
Log(Sales/Worker) LogSales
0.122***
(0.012)
-0.067***
(0.012)
0.001
(0.001)
0.091
(0.135)
0.002**
(0.001)
Yes
Yes
285,544
620.1
0.132***
(0.015)
-0.109***
(0.015)
0.001
(0.001)
0.209
(0.174)
-0.000
(0.001)
Yes
Yes
283,591
613.0
-0.009***
(0.003)
0.006**
(0.003)
0.000
(0.000)
-0.137***
(0.030)
-0.000
(0.000)
Yes
Yes
285,544
617.9
-0.033**
(0.014)
-0.023**
(0.012)
-0.001
(0.001)
-0.496***
(0.135)
0.001
(0.001)
Yes
Yes
275,484
577.8
-0.002***
(0.000)
-0.670***
(0.077)
0.001
(0.001)
Yes
Yes
800,058
573.7
0.007
(0.009)
0.005
(0.004)
0.000
(0.000)
-0.337***
(0.047)
0.001***
(0.000)
Yes
Yes
804,853
581.3
0.046***
(0.015)
-0.019***
(0.006)
0.002***
(0.000)
0.051
(0.073)
0.002***
(0.000)
Yes
Yes
799,002
569.9
0.037***
0.071***
(0.012)
0.042***
(0.005)
-0.001***
(0.000)
-1.672***
(0.060)
0.001***
(0.000)
Yes
Yes
803,712
579.0
Panel B: SOEs
Dependent Variable
Log(Upstream_ Loan)
GDPt-
1
AvgIncomet-
1
FiscalIncometEmploymentt-
1
1
Year Fixed Effect
Firm Fixed Effect
Observations
Wald F-stat
0.125***
(0.011)
-0.047***
(0.010)
0.001
(0.001)
0.227**
(0.115)
-0.000
(0.001)
Yes
Yes
283,980
607.8
0.191***
(0.017)
-0.106***
(0.015)
0.003***
(0.001)
0.181
(0.171)
0.002
(0.001)
Yes
Yes
282,245
602.5
0.084***
(0.015)
-0.073***
(0.014)
-0.000
(0.001)
-0.347**
(0.155)
0.000
(0.000)
Yes
Yes
279,226
590.6
Note. Upstream Loan is the loan amount in firm's most related upstream industry at province x industry x year
level. In Panel A, data is restricted to private firms in CIC data from 1998 to 2009. Panel A shows the two stage
least squares regression results by using First to Sixth as instrumental variable for logarithm of CDB province level
outstanding loan amount in 25 manufacturing industries (collapsed from 41 CIC manufacturing industries) and 27
provinces (exclude Beijing, Shanghai, Tianjing, and Chonqing). Upstream_Loan is the outstanding loan amount in
firm's most related upstream industry. GDP, AvgIncome, FiscalIncome and Employment are province level GDP,
income per capita, fiscal income and total employment respectively. All columns are controlled by year fixed effect and
firm fixed effect. Standard errors are clustered at firm level. Cragg-Donald Wald F-statistics for weak identification
tests are reported.
In Panel B, data is restricted to SOEs in CIC data from 1998 to 2009. Panel B shows the two stage least squares
regression results by using First to Sixth as instrumental variable for logarithm of CDB province level outstanding
loan amount in 25 manufacturing industries (collapsed from 41 CIC manufacturing industries) and 27 provinces
(exclude Beijing, Shanghai, Tianjing, and Chonqing). Data restricted to SOEs in CIC data from 1998 to 2009.
Upstream Loan is the outstanding loan amount in firm's most related upstream industry. GDP, AvgIncome,
FiscalIncome and Employment are province level GDP, income per capita, fiscal income and total employment
respectively. All columns are controlled by year fixed effect and firm fixed effect. Standard errors are clustered at
firm level. Cragg-Donald Wald F-statistics for weak identification tests are reported.
63
Table 1.9: What Private Firms Benefit from CDB Upstream Industry Loans (2SLS)
Panel A
Dependent Variable
Log(Upstream _Loan)
Log(UpstreamLoan)*
Connected
GDPt_1
AvgIncomet-1
FiscalIncomet-1
Employmenti1
Year Fixed Effect
Firm Fixed Effect
Observations
Wald F-stat
Panel B
Dependent Variable
Log(Upstream_ Loan)
Log(Upstream_ Loan)*
Connected
GDPt-i
Avgncomet _ 1
Fiscallncomet_1
Employmentti1
SizexIV
Year Fixed Effect
Firm Fixed Effect
Observations
Wald F-stat
(5)
ROA
(7)
(6)
Log(Sales/Worker) LogSales
0.045***
(0.015)
0.120***
(0.024)
-0.000
(0.003)
-0.026***
(0.003)
0.066***
(0.011)
-0.050***
(0.014)
0.072***
(0.012)
0.016
(0.019)
-0.019***
(0.006)
0.002***
(0.000)
0.059
(0.073)
0.002***
(0.000)
Yes
Yes
799,002
1431
0.004***
(0.001)
0.000
(0.000)
-0.237***
(0.013)
0.000**
(0.000)
Yes
Yes
804,709
1456
0.037***
(0.004)
-0.001***
(0.000)
-1.340***
(0.053)
0.001**
(0.000)
Yes
Yes
803,296
1457
0.043***
(0.005)
-0.001***
(0.000)
-1.674***
(0.060)
0.001***
(0.000)
Yes
Yes
803,712
1454
(5)
ROA
(7)
(6)
Log(Sales/Worker) LogSales
0.039**
(0.015)
0.123***
(0.025)
-0.000
(0.003)
-0.026***
(0.003)
0.064***
(0.011)
-0.050***
(0.015)
0.068***
(0.012)
0.020
(0.020)
-0.020***
(0.006)
0.002***
(0.000)
0.080
(0.073)
0.016***
(0.000)
Yes
Yes
Yes
799,002
1434
0.004***
(0.001)
0.000
(0.000)
-0.239***
(0.013)
0.000**
(0.000)
Yes
Yes
Yes
804,709
1460
0.037***
(0.004)
-0.001***
(0.000)
-1.334***
(0.053)
0.001**
(0.000)
Yes
Yes
Yes
803,269
1460
0.042***
(0.005)
-0.001***
(0.000)
-1.661***
(0.060)
0.013***
(0.000)
Yes
Yes
Yes
803,681
1457
(1)
LogAsset
(4)
(3)
(2)
LogFixAst LogWorker LogDebt
0.091***
(0.011)
0.073***
(0.019)
0.135***
(0.015)
0.059**
(0.023)
0.006
(0.009)
0.067***
(0.013)
-0.011**
(0.005)
0.000
(0.000)
-0.390***
(0.057)
0.001***
(0.000)
Yes
Yes
804,718
1451
-0.034***
(0.006)
-0.002***
(0.000)
-0.666***
(0.077)
0.001
(0.001)
Yes
Yes
800,058
1441
0.005
(0.004)
0.000
(0.000)
-0.333***
(0.046)
0.001***
(0.000)
Yes
Yes
804,853
1459
(1)
LogAsset
(4)
(3)
(2)
LogFixAst LogWorker LogDebt
0.085***
(0.011)
0.082***
(0.021)
0.129***
(0.015)
0.065***
(0.025)
0.006
(0.009)
0.071***
(0.014)
-0.012**
(0.005)
0.000
(0.000)
-0.370***
(0.057)
0.001***
(0.000)
Yes
Yes
Yes
804,718
1459
-0.035***
(0.006)
-0.002***
(0.000)
-0.648***
(0.077)
0.001
(0.001)
Yes
Yes
Yes
800,058
1444
0.005
(0.004)
0.000
(0.000)
-0.334***
(0.046)
0.001***
(0.000)
Yes
Yes
Yes
804,226
1462
Note. Data is restricted to private firms in CIC data from 1998 to 2009. Table 1.9 Panel A shows the two
stage least squares regression results by using First to Sixth as instrumental variable for logarithm of CDB
province level outstanding loan amount in 25 manufacturing industries (collapsed from 41 CIC manufacturing industries) and 27 provinces (exclude Beijing, Shanghai, Tianjing, and Chonqing). Upstream Loan is the outstanding loan amount in firm's most related upstream industry. Connected is the dummy for whether the private firm's political hierarchy is above the city(province level and national level) or not. In order to include
the interaction term Log(Upstream Loan) * Connected in 2SL, I follow Wooldridge (2002). I first regress instrumental variables(First to Sixth) on Log(Upstream_Loan) with all exogenous variables and get fitted value
Log(Upstream Loan). Second, I use Log(Upstream_Loan) and Log(Upstream_Loan) * Connected as instrumental variables for Log(Upstream_Loan) and Log(Upstream_Loan) * Connected, and perform the standard 2SLS
again. GDP, AvgIncome, FiscalIncome and Employment are province level GDP, income per capita, fiscal income
and total employment respectively. Panel B is the robustness check by controlling the interaction terms between firm
size(total asset) and the instrumental variables. All columns are controlled by year fixed effect and firm fixed effect.
Standard errors are clustered at firm level. Cragg-Donald Wald F-statistics for weak identification tests are reported.
64
Table 1.10: Variables' Definition and Construction
Loan _City
IssuanceCity
LoanINFCity
IssuanceINFCity
LoanINDCity
IssuanceIND _City
GDP
AvgIncome
FiscalIncome
Employment
PoliticianYear
Age
Gender
Relation
Promotion
Loan_PI
Issuance_PI
LogAsset
LogFixAst
LogWorker
LogDebt
ROA
Log(Sales/Worker)
LogSales
TaxCorp
TaxVAT
Outstanding total CDB city level loan from 1998 to 2010. It covers 310 cities in China.
The unit is 100 million RMB.
Total CDB new city level loan issuance from 1998 to 2010. It covers 310 cities in China.
The unit is 100 million RMB.
Outstanding CDB city level loan for infrastructure projects from 1998 to 2010. It covers
310 cities in China. The unit is 100 million RMB.
Total CDB city level loan issuance for infrastructure projects from 1998 to 2010. It covers
310 cities in China.The unit is 100 million RMB.
Outstanding CDB city level loan for industry firms from 1998 to 2010. It covers 310 cities in
China. The unit is 100 million RMB.
Total CDB city level loan issuance for industry firms from 1998 to 2010. It covers 310 cities
in China. The unit is 100 million RMB.
City GDP amount from 1998 to 2010. It covers 310 cities in China. The unit is 100 million
RMB. Data is from National Bureau of Statistics of China.
Income per capita in the city from 1998 to 2010. It covers 310 cities in China. The unit is
RMB. Data is from National Bureau of Statistics of China.
City fiscal income amount from 1998 to 2010. It covers 310 cities in China. The unit is 100
million RMB. Data is from National Bureau of Statistics of China.
City total number of workers from 1998 to 2010. It covers 310 cities in China. The unit is
100 thousand. Data is from National Bureau of Statistics of China.
Number of years that the city secretary has been staying in the city. It covers 310 cities in
China from 1998 to 2010.
The age of the city secretary. It covers 310 cities in China from 1998 to 2010.
Dummy for whether the city secretary is a female or not. It covers 310 cities in China from
1998 to 2010.
Dummy for whether the city secretary is born at the same city as the province governors.
It covers 310 cities in China from 1998 to 2010.
Dummy for whether the city secretary is promoted to a higher position in the government or
not in the end of her term. It covers 310 cities in China from 1998 to 2010. In China,
different cities have different political hierarchy. For example, Shanghai, Beijing, chonqing,
and Tianjing are in the ministry level (same level as province). I take this into account when
define "Promotion".
Outstanding total CDB province loan for each of the 95 industries from 1998 to 2013. It
covers all 31 provinces in China. The unit is 100 million RMB.
Total CDB province loan new issuance for each of the 95 industries from 1998 to 2013. It
covers all 31 provinces in China. The unit is 100 million RMB.
Logarithm of the total asset of the firm. The unit of asset is in thousand RMB. It covers all
711,892 firms in CIC data from 1998 to 2009.
Logarithm of the fixed asset of the firm. The unit of fixed asset is in thousand RMB. It
covers all 711,892 firms in CIC data from 1998 to 2009.
Logarithm of the employee number of the firm. It covers all 711,892 firms in CIC data
from 1998 to 2009.
Logarithm of the total debt of the firm. The unit of the total debt is in thousand RMB. It
covers all 711,892 firms in CIC data from 1998 to 2009.
Contemporaneous ROA. It is calculated by dividing a firm's annual earnings by its
total asset in the same year. It covers all 711,892 firms in CIC data from 1998 to 2009.
Logarithm of the sales per employee. It covers all 711,892 firms in CIC data from 1998
to 2009.
Logarithm of the total sales. The unit of the total sales is in thousand RMB. It covers all
711,892 firms in CIC data from 1998 to 2009.
The corporate tax rate of each firm by dividing corporate tax payable amount by the income
before tax every year. Tax Corp is a missing value when income before tax is negative. It
covers all 711,892 firms in CIC data from 1998 to 2009.
The value-added tax rate of each by dividing value-added tax payable amount by the
production value-added every year. TaxVAT is a missing value when production
value-added is negative. It covers all 554,882 firms in CIC data in 1998, 1999, 2000, 2003,
2005, 2006 and 2007. CIC data doesn't record the value-added tax payment in 2001, 2002,
2004, 2008 and 2009.
65
Table 1.11: City Secretary's Turnover and Borrowing from CDB(off national-cycle)
Panel A
Total Loan
Dependent Variable
PoliticianYear
GDPt-1
AvgIncomet -1
FiscalIncomet -1
Employmentt
-1
Year Fixed Effect
City Fixed Effect
Politician Fixed Effect
Observations
R-squared
Panel B
Dependent Variable
Year_2
Year_3
Year_4
Year_5
Year_6
GDPt-1
Avglncomet -1
Fiscallncomet -1
Employmentt -1
Year Fixed Effect
City Fixed Effect
Politician Fixed Effect
Observations
R-squared
Loan for Infrastructure
(1)
(2)
(3)
(4)
LogLoan_ City LogIssueCity LogLoanCity LogIssueCity
-0.078**
(0.031)
-0.234
(0.157)
-0.005***
(0.002)
0.023
(1.302)
-1.083
(7.736)
Yes
Yes
Yes
1,686
0.809
-0.190***
(0.023)
0.062
(0.194)
-0.003
(0.008)
-1.655
(1.604)
-1.815
(8.707)
Yes
Yes
Yes
1,542
0.639
-0.124***
(0.017)
-0.228*
(0.135)
0.003
(0.004)
-0.652
(0.965)
0.539
(3.062)
Yes
Yes
Yes
1,387
0.855
-0.161***
(0.032)
0.200
(0.379)
-0.004
(0.007)
-3.266
(2.811)
5.085
(6.218)
Yes
Yes
Yes
1,278
0.553
Loan for Industry Firms
(6)
(5)
LogLoanCity LogIssueCity
-0.099**
(0.041)
-0.180
(0.201)
-0.007***
(0.002)
-0.535
(1.634)
-6.179
(9.291)
Yes
Yes
Yes
1,495
0.706
-0.242***
(0.033)
-0.121
(0.233)
-0.005
(0.008)
1.678
(2.265)
-23.345***
(6.817)
Yes
Yes
Yes
1,293
0.514
Total Loan
Loan for Infrastructure
Loan for Industry Firms
(2)
(1)
LogLoanCity LogIssueCity
(3)
(4)
LogLoanCity LogIssueCity
(5)
(6)
LogLoanCity LogIssueCity
-0.086*
(0.047)
-0.137**
(0.063)
-0.181**
(0.091)
-0.314**
(0.154)
-0.579***
(0.195)
-0.224
(0.162)
-0.005***
(0.002)
-0.096
(1.327)
-1.193
(7.917)
Yes
Yes
Yes
1,686
0.810
-0.174**
(0.073)
-0.350***
(0.067)
-0.527***
(0.085)
-0.891***
(0.181)
-1.050***
(0.194)
0.063
(0.201)
-0.003
(0.008)
-1.815
(1.612)
-11.330
(9.092)
Yes
Yes
Yes
1,542
0.640
-0.119***
(0.042)
-0.228***
(0.045)
-0.358***
(0.055)
-0.487***
(0.095)
-0.728***
(0.133)
-0.224
(0.137)
0.003
(0.004)
-0.676
(1.022)
0.616
(3.043)
Yes
Yes
Yes
1,387
0.856
-0.161*
(0.095)
-0.261***
(0.083)
-0.537***
(0.114)
-0.624***
(0.201)
-0.791***
(0.279)
0.196
(0.384)
-0.004
(0.007)
-3.229
(2.852)
5.065
(6.169)
Yes
Yes
Yes
1,278
0.554
-0.043
(0.063)
-0.179**
(0.083)
-0.207*
(0.119)
-0.415*
(0.217)
-0.611**
(0.278)
-0.169
(0.210)
-0.007***
(0.002)
-0.737
(1.727)
-5.658
(9.528)
Yes
Yes
Yes
1,495
0.707
-0.257**
(0.111)
-0.472***
(0.097)
-0.644***
(0.130)
-1.229***
(0.242)
-1.316***
(0.231)
-0.122
(0.234)
-0.005
(0.009)
1.385
(2.283)
-23.103***
(6.509)
Yes
Yes
Yes
1,293
0.517
Note. Data restricted to CDB city level loan from 1998 to 2010. I exclude the city secretaries whose turnover are at the
same time as the national turnover cycle(year 1998, 2003 and 2008). Panel A shows the results from OLS regression
from equation 1.3. PoliticianYear is the number of years that the secretary has been staying in the city. Column
1 and 2 are the regressions on the logarithm of total CDB loan outstanding amount and total issuance respectively.
Column 3 and 4 are the regressions on the logarithm of CDB infrastructure loan outstanding amount and new issuance
respectively. Column 5 and 6 are the regressions on the logarithm of CDB industry firm loan outstanding amount
and new issuance respectively. GDP, AvgIncome, FiscalIncome and Employment are city level GDP, income per
capita, fiscal income and total employment respectively. All columns are controlled by year fixed effect, city fixed
effect and politician personal fixed effect. Standard errors are clustered at city level.Panel B shows the results from
OLS regression from equation 1.4. Year 2 is the the dummy for whether it is the second of the secretary in the city.
Year_3 is the the dummy for whether it is the third of the secretary in the city. Year_4 to Year_6 are defined in
the same way. The dummy for year 1 is the missing category. All columns are controlled by year fixed effect, city
fixed effect and politician personal fixed effect. Standard errors are clustered at city level.
66
Table 1.12: What Firms are Crowded Out in Private Sector
(2)
(1)
(3)
(4)
Dependent Variable
LogAsset LogFixAst LogWorker LogDebt
PoliticianYear
-0.015***
-0.004
0.002
-0.018**
(0.005)
(0.007)
Low ROA
0.148*** 0.089***
(0.004)
(0.006)
PoliticianYear*Low ROA -0.011*** -0.010***
(0.001)
(0.002)
GDPt-j
-0.052*** -0.045***
(0.006)
(0.009)
AvgIncomet - 1
-0.000 -0.001***
(0.000)
(0.000)
(0.005)
-0.052***
(0.004)
-0.010***
(0.001)
-0.002
(0.006)
-0.000
(0.000)
(0.008)
0.187***
(0.006)
-0.003
(0.002)
-0.067***
(0.008)
0.000
(0.000)
FiscalIncomet_
-0.299***
-0.091
(0.050)
(0.070)
1
-0.248*** -0.317***
(0.053)
Employmentt_
1
Year Fixed Effect
Firm Fixed Effect
Politician Fixed Effect
Observations
R-squared
(0.077)
-0.001
0.001
-0.000
0.001
(0.001)
(0.001)
(0.001)
(0.002)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
899,857
895,698
900,086
893,378
0.261
0.133
0.0436
0.117
Note. Data restricted to private firms in Chinese Industry Census data from 1998 to 2009. Column
1 is the logarithm of total asset of the firm. Column 2 is the logarithm of fixed asset of the firm.
Column 3 is the logarithm of total number of workers of the firm. Column 4 is the logarithm of total
debt of the firm. PoliticianYearis the number of years that the secretary has been staying in the
city. GDP, AvgIncome, FiscalIncome and Employment are city level GDP, income per capita,
fiscal income and total employment respectively. All columns are controlled by year fixed effect, firm
fixed effect and politician personal fixed effect. Standard errors are clustered at firm level.
67
Table 1.13: CDB Loan's Effect on Firms (Use Secretary Year in Office as Instrument)
Dependent Variable
Log(LoanCity)
SOEs
Foreign Firms
(4)
(3)
(2)
(1)
LogAsset LogFixAst LogWorker LogDebt
(8)
(7)
(6)
(5)
LogAsset LogFixAst LogWorker LogDebt
0.069**
0.089**
(0.042)
(0.031)
-0.013
GDPti
-0.096***
(0.011)
(0.016)
Avglncomet-1
0.001
0.001
(0.001)
(0.001)
Fiscallncomet -1
-0.333***
-0.088
(0.141)
(0.099)
-0.583
1.249*
Employmentti(0.748)
(0.519)
Yes
Year Fixed Effect
Yes
Firm Fixed Effect
Yes
Yes
Politician Fixed Effect
Yes
Yes
Observations
305,516
303,351
Wald F-stat
307.8
300.5
0.083***
(0.029)
-0.007
(0.012)
0.000
(0.001)
0.109
(0.098)
-2.325***
(0.634)
Yes
Yes
Yes
304,861
302.2
0.101**
(0.044)
-0.064***
(0.015)
0.001
(0.001)
-0.250*
(0.134)
-1.626**
(0.744)
Yes
Yes
Yes
302,458
308.0
-0.106***
(0.033)
-0.015*
(0.008)
-0.000
(0.000)
-0.491***
(0.072)
-1.133
(0.943)
Yes
Yes
Yes
236,430
801.2
-0.170***
(0.046)
-0.001
(0.011)
-0.001**
(0.001)
-0.706***
(0.100)
0.381
(1.299)
Yes
Yes
Yes
235,517
789.3
-0.124***
(0.030)
0.056***
(0.008)
0.000
(0.000)
-0.272***
(0.075)
-2.809***
(0.889)
Yes
Yes
Yes
236,222
795.8
-0.051
(0.048)
-0.057***
(0.013)
0.001
(0.001)
-0.367***
(0.107)
-0.093
(1.444)
Yes
Yes
Yes
235,144
797.3
Note. Table 1.13 are the two stage least squares regression results by using Year_1 to Year_6 as instrumental
variable for logarithm of CDB city level outstanding loan amount. Column 1 to 4 are based on SOEs in CIC from
1998 to 2009. Column 5 to 8 are based on foreign firms in CIC from 1998 to 2009. GDP, AvgIncome, FiscalIncome
and Employment are city level GDP, income per capita, fiscal income and total employment respectively. All columns
are controlled by year fixed effect, firm fixed effect and politician personal fixed effect. Standard errors are clustered
at firm level. Cragg-Donald Wald F-statistics for weak identification tests are reported.
68
Table 1.14: Turnover Timing's Effects on SOEs under Different CDB Loan Amount
Panel A: All Cities
Dependent Variable
(2)
(1)
(3)
(4)
(5)
LogAsset LogFixedAst LogWorker LogDebt Log(Sales/Worker)
PoliticianYear
-0.007
-0.013
0.000
(0.007)
(0.009)
(0.006)
PoliticianYear*HighCDB -0.013*** -0.010***
-0.012***
(0.003)
(0.003)
(0.003)
HighCDB
0.025***
0.023**
0.055***
(0.009)
(0.011)
(0.009)
GDPt- 1
-0.008
-0.082***
-0.007
(0.013)
(0.019)
(0.014)
AvgIncomet -1
0.000
0.000
0.000
(0.001)
(0.001)
(0.001)
FiscalIncomet 1
-0.237**
0.031
0.181
(0.111)
(0.160)
(0.112)
Employmentt-1
-0.008**
-0.009
-0.002
(0.004)
(0.007)
(0.002)
Year Fixed Effect
Yes
Yes
Yes
Firm Fixed Effect
Yes
Yes
Yes
Politician Fixed Effect
Yes
Yes
Yes
Observations
382,317
379,632
381,720
R-squared
0.108
0.0523
0.0643
Panel B: Zero CDB Loan
(1)
(2)
(3)
Dependent Variable
LogAsset LogFixedAst LogWorker
PoliticianYear
-0.073
(0.157)
-0.062
GDPt- 1
(0.214)
AvgIncomet-1
0.000
(0.001)
FiscalIncomet-1
2.092
(1.946)
-0.007
Employmentt_
1
(0.005)
Year Fixed Effect
Firm Fixed Effect
Politician Fixed Effect
Observations
R-squared
Yes
Yes
Yes
78,512
0.0391
0.003
(0.220)
0.121
(0.287)
-0.001
(0.003)
-0.288
(2.731)
-0.009
(0.009)
Yes
Yes
Yes
77,919
0.0237
-0.182
(0.142)
0.505**
(0.228)
0.003*
(0.002)
0.225
(2.091)
-0.002
(0.003)
Yes
Yes
Yes
78,261
0.0635
-0.018*
(0.010)
-0.007**
(0.004)
0.006
(0.012)
-0.074***
(0.018)
0.000
(0.001)
-0.094
(0.151)
-0.010
(0.006)
Yes
Yes
Yes
378,567
0.0647
0.014*
(0.008)
-0.010***
(0.004)
0.013
(0.012)
-0.029*
(0.017)
0.000
(0.001)
-0.365**
(0.148)
-0.004
(0.005)
Yes
Yes
Yes
369,479
0.178
(4)
(5)
LogDebt Log(Sales/Worker)
-0.025
(0.230)
-0.385
(0.317)
-0.001
(0.002)
3.216
(2.701)
-0.010
(0.007)
Yes
Yes
Yes
77,643
0.0305
0.130
(0.210)
0.828***
(0.319)
-0.001
(0.002)
-4.682
(3.084)
-0.002
(0.006)
Yes
Yes
Yes
75,212
0.0534
Note. Panel A is restricted to SOEs in CIC data from 1998 to 2009. PoliticianYearis the number of years that the
secretary has been staying in the city. HighCDB is the dummy for whether the city's CDB loan amount is above
the median of all CDB loan in 310 cities from 1998 to 2009. GDP, AvgIncome, FiscalIncome and Employment are
city level GDP, income per capita, fiscal income and total employment respectively. All columns are controlled by
year fixed effect, firm fixed effect and politician personal fixed effect. Standard errors are clustered at firm level.Panel
B is restricted to SOEs in CIC data from 1998 to 2009 with 0 CDB loan amount in the city. PoliticianYear is the
number of years that the secretary has been staying in the city. GDP, AvgIncome, FiscalIncome and Employment
are city level GDP, income per capita, fiscal income and total employment respectively. All columns are controlled
by year fixed effect, firm fixed effect and politician personal fixed effect. Standard errors are clustered at firm level.
69
Table 1.15: CDB Loan and Firm Exit&Enter (Use Secretary Year in Office as Instrument)
Private Firm
Foreign Firms
SOEs
(2)
(1)
LogExit LogEnter
(4)
(3)
LogExit LogEnter
(6)
(5)
LogExit LogEnter
0.383*** -0.270***
(0.097)
(0.067)
0.082
0.427
GDPt 1
(0.114)
(0.271)
-0.027
0.002
AvgIncomet _1
(0.004)
(0.025)
2.075
-1.673
FiscalIncomet_1
(1.583)
(2.393)
14.465 9.460***
Employmentt_ 1
(11.330) (3.505)
Yes
Yes
Year Fixed Effect
Yes
Yes
City Fixed Effect
Yes
Yes
Politician Fixed Effect
2,632
2,152
Observations
68.35
62.43
Wald F-stat
0.321*** -1.123***
(0.150)
(0.067)
-0.046
0.593*
(0.318) (0.251)
-0.006
-0.047
(0.042) (0.003)
-1.361
3.169
(2.225)
(2.237)
18.686
-18.981
(11.957) (19.411)
Yes
Yes
Yes
Yes
Yes
Yes
2,076
1,521
58.34
155.2
-0.405*** 0.310***
(0.106)
(0.117)
0.306**
-0.290
(0.129)
(0.185)
0.012***
0.008
(0.002)
(0.006)
1.390
1.853
(1.107)
(1.451)
28.998*** 17.325
(10.529) (10.925)
Yes
Yes
Yes
Yes
Yes
Yes
2,444
2,434
65.00
40.16
Dependent Variable
Log(Loan_ City)
Note. Table 1.15 are the two stage least squares regression results by using PoliticianYear as
instrumental variable for logarithm of CDB city level outstanding loan amount. Data restricted
to manufacturing firms in CIC data from 1998 to 2009. GDP, AvgIncome, FiscalIncome and
Employment are city level GDP, income per capita, fiscal income and total employment respectively.
All columns are controlled by year fixed effect, firm fixed effect and politician personal fixed effect.
Standard errors are clustered at firm level. Cragg-Donald Wald F-statistics for weak identification
tests are reported.
70
Table 1.16: Politician's Other Channels to Affect Local Economy
Panel A: Current
(1)
(2)
(3)
(4)
(5)
(6)
Dependent Variable DevelopedLand Export FiscalIncome FiscalTransfer TaxCorp TaxVAT
PoliticianYear
0.139
(0.092)
-0.298
(0.383)
0.473
(0.715)
-0.411
(0.368)
-0.066
(0.050)
0.019
(0.078)
Observations
R-squared
3,401
0.001
3,588
0.000
3,587
0.000
3,378
0.000
3,263
0.001
1,873
0.000
Panel B: Lagged
(1)
(2)
(3)
(4)
(5)
(6)
Dependent Variable DevelopedLand Export FiscalIncome FiscalTransfer TaxCorp TaxVAT
PoliticianYear -1
0.157
(0.110)
-0.270
(0.356)
0.280
(0.998)
-0.349
(0.341)
-0.069
(0.049)
-0.012
(0.079)
Observations
R-squared
3,110
0.001
3,279
0.000
3,277
0.000
3,091
0.000
2,956
0.001
1,601
0.000
Note. Data restricted to 310 cites from 1998 to 2010. In column 1, Developed_Land is the total
developed land in each city every year. In column 2, Export is the total export amount in each
city every year. In column 3, Fiscal__Income is the total tax fiscal income amount in each city
every year. In column 4, Fiscal_Transfer is the total central government transfer amount in each
city every year. In column 5, Taxcorp is the average manufacturing firm effective corporate tax
rate in each city every year.In column 6, Tax__VAT is the average manufacturing firm effective
value-added tax rate in each city every year. This table mainly explore the correlations between city
secretary turnover and other channels that city secretary can influence the local economy. Panel A
is for current turnover PoliticianYearand Panel B is for the lagged turnover PoliticianYearLag.
Standard errors are clustered at city level.
71
Table 1.17: Where Does the CDB Industry Loan Go?
Panel A: IND Loan
Dependent Variable
AssetSOEt_
1
(1)
Loan_PI
(2)
Issuance_PI
0.058***
(0.009)
0.016***
(0.002)
0.003
(0.007)
Yes
Yes
Yes
6,518
0.640
Assetprivatet-1
Year Fixed Effect
Province Fixed Effect
Industry Fixed Effect
Observations
R-squared
Panel B: INF Loan
Dependent Variable
AssetSOEt_
1
Assetprivatet1
Year Fixed Effect
Province Fixed Effect
City Fixed Effect
Observations
R-squared
Yes
Yes
Yes
7,334
0.687
(3)
Loan_PI
Yes
Yes
Yes
7,334
0.372
(4)
(5)
(6)
Issuance_PI LoanPI Issue_PI
0.002
(0.003)
Yes
Yes
Yes
6,518
0.344
0.051*** 0.013***
(0.002)
(0.008)
0.002
0.003
(0.003)
(0.007)
Yes
Yes
Yes
Yes
Yes
Yes
6,230
6,230
0.386
0.735
(4)
(3)
(2)
(1)
LoanINFP IssueINFP LoanINFC IssueINF_C
0.015
(0.042)
0.063***
(0.014)
Yes
Yes
No
297
0.905
0.007
(0.019)
0.025***
(0.005)
Yes
Yes
No
297
0.843
0.009
(0.011)
0.035**
(0.015)
Yes
No
Yes
3,257
0.619
0.009
(0.007)
0.017**
(0.009)
Yes
No
Yes
3,257
0.561
Note. In Panel A,data restricted to CDB province level industry loan data from 1998 to 2009 in 31
provinces and 41 manufacturing industries. Loan_PI is the outstanding loan amount at provinceindustry level every year. Issue_PI is the loan new issuance amount at province-industry level
every year. The unit is 1000 RMB. Asset_SOE is the total asset of SOEs in each province and
industry every year. Asset_private is the total asset of private firms in each province and industry
every year. The unit is in 1000 RMB. All columns are controlled by year fixed effect, province fixed
effect and industry fixed effect. Standard errors are clustered at province level.
In Panel B, column 1 and 2 are based on province level aggregate infrastructure loan data from 1998
to 2009 in 31 provinces. LoanINFPis the outstanding infrastructure loan amount at province
level every year. IssuanceINFPis the infrastructure loan new issuance amount at province
level every year. The unit is 1000 RMB. AssetSOE is the total asset of SOEs in each province
every year. Asset_private is the total asset of private firms in each province every year. The unit
is in 1000 RMB. Column 1 and 2 are also controlled by year fixed effect and province fixed effect.
Standard errors are clustered at province level. Column 3 and 4 in Panel B are based on city level
aggregate infrastructure loan data from 1998 to 2009 in 310 cities. Loan_INF_ C is the outstanding
infrastructure loan amount at city level every year. IssuanceINFC is the infrastructure loan
new issuance amount at city level every year. The unit is 1000 RMB. Asset_ SOE is the total asset
of SOEs in each city every year. Assetprivate is the total asset of private firms in each city every
year. The unit is in 1000 RMB. Column 3 and 4 are also controlled by year fixed effect and city
fixed effect. Standard errors are clustered at city level.
72
Table 1.18: CDB Province Industry Level Loan and City Secretary Turnover (First
Stage)
Dependent Variable
First
(1)
Log(LoanPI)
Second
0.268***
(0.090)
0.366***
(0.099)
Third
0.251***
(3)
Log(LoanPI)
(0.086)
0.128
(0.082)
0.057
(0.140)
-0.097
(0.120)
Fourth
Fifth
Sixth
First-Second
0.339***
(0.103)
First-Third
GDPt
(2)
Log(LoanPI)
1
AvgIncomet-
1
FiscalIncomet -1
-0.022
(0.022)
0.190**
(0.083)
0.005
(0.199)
Employmentt_ 1
0.953
Year Fixed Effect
Province Fixed Effect
Industry Fixed Effect
Observations
R-squared
Yes
Yes
Yes
-0.021
(0.022)
0.184**
(0.083)
-0.000
(0.199)
0.916
(2.101)
Yes
Yes
Yes
4,197
0.535
(2.094)
4,197
0.538
0.332***
(0.099)
-0.020
(0.022)
0.181**
(0.083)
-0.011
(0.200)
0.977
(2.074)
Yes
Yes
Yes
4,197
0.536
Note. The regressions are estimated at province x industry x year level. Data restricted to CDB
province level industry loan data from 1998 to 2009 in 31 provinces and 41 manufacturing industries.
Log(LoanPI) is logarithm of the outstanding loan amount at province-industry level every year.
First is the dummy for whether the city secretary is in the first year of the term and the firm is
in the city's largest SOE industry. Second is the dummy for whether the city secretary is in the
second year of the term and the firm is in the city's largest SOE industry. The dummies from
Third to Sixth are defined in the same way. GDP, AvgIncome, FiscalIncome and Employment
are province level GDP, income per capita, fiscal income and total employment respectively. All
columns are controlled by year fixed effect, province fixed effect and industry fixed effect. Standard
errors are clustered at province level.
73
Table 1.19: City Secretary's Promotion and GDP Performance
Panel A: Promotion and GDP
(5)
(4)
(3)
(2)
(1)
Promotion Promotion Promotion Promotion Promotion
Dependent Variable
GDP Increase in 1st year
0.975**
(0.481)
0.597***
(0.215)
GDP Increase in 1st-2nd year
0.407***
(0.148)
GDP Increase in 1st-3rd year
0.348***
(0.126)
GDP Increase in lst-4th year
0.318***
GDP Increase in 1st-5th year
(0.121)
0.270
(0.196)
-0.079***
(0.011)
0.185
(0.333)
Relation
Age
Gender
0.295
(0.194)
-0.081***
(0.011)
0.191
(0.333)
0.299
(0.194)
-0.080***
(0.011)
0.198
(0.331)
0.299
(0.194)
-0.079***
(0.011)
0.203
(0.330)
0.303
(0.194)
-0.079***
(0.011)
0.204
(0.330)
1,023
1,024
1,028
1,034
1,027
Observations
62.48
63.65
64.12
64.86
59.73
Chi squared
Panel B: Promotion and CDB Loan
Promotion Promotion Promotion Promotion Promotion
Dependent Variable
Log(Loan Increase in 1st year)
0.102***
(0.035)
0.087**
(0.034)
Log(Loan Increase in 1st-2nd year)
Log(Loan Increase in
1st-3rd year)
0.055*
(0.033)
0.062*
(0.032)
Log(Loan Increase in 1st-4th year)
Log(Loan Increase in
Relation
Age
Gender
Observations
Chi squared
1st-5th year)
0.430*
(0.237)
-0.084***
(0.013)
0.617
(0.388)
691
54.58
0.457**
(0.218)
-0.078***
(0.013)
0.663*
(0.388)
758
48.75
0.431**
(0.211)
-0.074***
(0.012)
0.477
(0.364)
792
42.96
0.394*
(0.207)
-0.074***
(0.012)
0.410
(0.381)
810
43.85
0.057*
(0.033)
0.426**
(0.205)
-0.074***
(0.012)
0.407
(0.381)
812
43.70
Note. Data restricted to 1,066 city secretaries among 310 cites from 1998 to 2010. Panel A shows the results from
Probit regression. Promition is the dummy for whether the city secretary is promoted or not at the end of the term.
GDP Increase in 1st year is the GDP increase in the first year of the city secretary's tenure. GDP Increase in lst-2nd
year is the GDP increase in the first two years of the city secretary's tenure. The unit of GDP is 100 billion RMB.
Panel B shows the results from Probit regression with CDB loan. Log(Loan Increase in 1st year) is the logarithm of
CDB outstanding loan amount increase in the first year of the city secretary's tenure. Log(Loan Increase in lst-2nd
year) is the logarithm of CDB outstanding loan amount increase in the first two years of the city secretary's tenure.
Standard errors are clustered at city level.
74
Table 1.20: City Secretary's Characteristics and Borrowing from CDB
Dependent Variable
PoliticianYear
(1)
(2)
(3)
(4)
(5)
(6)
LogLoanCity LogIssue _City LogLoanCity LogIssue City LogLoanCity LogIssueCity
-0.364***
(0.020)
-2.064***
(0.086)
ShortTerm
-0.308***
(0.029)
-1.919***
(0.142)
-0.366***
(0.021)
-0.300***
(0.030)
4.042***
(0.298)
2.026***
(0.212)
Younger
Relation
GDPt-1
AvgIncomet-1
FiscalIncomet-1
Employmentt-
1
Year Fixed Effect
City Fixed Effect
Politician Fixed Effect
Observations
R-squared
-0.269*
(0.138)
-0.004**
(0.002)
0.874
(1.197)
-0.953
(5.875)
Yes
Yes
Yes
3,127
0.853
0.018
(0.167)
-0.001
(0.007)
-1.040
(1.428)
-11.682**
(5.870)
Yes
Yes
Yes
2,795
0.685
-0.283**
(0.138)
-0.004**
(0.002)
0.954
(1.189)
-1.802
(5.659)
Yes
Yes
Yes
2,921
0.851
-0.007
(0.170)
-0.002
(0.007)
-1.126
(1.433)
-10.212*
(5.803)
Yes
Yes
Yes
2,644
0.676
-0.380***
(0.023)
-0.299***
(0.032)
0.189***
(0.072)
-0.271**
(0.133)
-0.004***
(0.001)
1.006
(1.181)
-2.728
(5.179)
Yes
Yes
Yes
2,782
0.857
0.085
(0.144)
-0.002
(0.169)
-0.002
(0.007)
-1.344
(1.453)
-10.223*
(5.817)
Yes
Yes
Yes
2,523
0.671
Note. The regressions are estimated at city x year level. Data restricted to CDB city level loan to 310 cites from 1998
to 2010. Column 1 to 6 are the regressions on the logarithm of total CDB loan outstanding amount and total issuance
respectively. PoliticianYearis the number of years that the secretary has been staying in the city. ShortTerm is
the dummy for whether the secretaries expect their terms to be less or equal than 2 years when they begin their
terms. Younger is the dummy for whether the city secretaries are younger than 50 years old when they begin the
terms. Relation is the dummy for whether the city secretaries are from the same place as the province governors.
GDP, AvgIncome, FiscalIncome and Employment are city level GDP, income per capita, fiscal income and total
employment respectively. All columns are controlled by year fixed effect, city fixed effect and politician personal fixed
effect. Standard errors are clustered at city level.
75
Table 1.21: City Secretary's Borrowing from CDB and City's Debt Level
Loan for Infrastructure
Total Loan
Dependent Variable
PoliticianYear
Constrained
PoliticianYear*
Constrained
GDPt_1
AvgIncomet -1
FiscalIncomet
-1
Employmentt
-1
Year Fixed Effect
City Fixed Effect
Politician Fixed Effect
Observations
R-squared
(1)
(2)
(4)
(3)
LogLoanCity LogIssue City LogLoanCity LogIssueCity
Loan for Industry Firms
(6)
(5)
LogLoanCity LogIssue_ City
-0.274***
(0.025)
0.726***
(0.128)
-0.167***
(0.031)
-0.283***
(0.039)
0.125
(0.179)
-0.137***
(0.048)
-0.112***
(0.015)
0.203*
(0.119)
-0.051
(0.036)
-0.154***
(0.029)
-0.039
(0.218)
-0.041
(0.060)
-0.234***
(0.034)
0.806***
(0.161)
-0.158***
(0.041)
-0.203***
(0.053)
0.243
(0.250)
-0.189***
(0.067)
-0.311**
(0.142)
-0.004*
(0.002)
1.561
(1.178)
-2.665
(5.189)
Yes
Yes
Yes
3,127
0.859
-0.083
(0.180)
-0.001
(0.007)
0.000
(1.462)
-12.442**
(5.926)
Yes
Yes
Yes
2,795
0.689
-0.234*
(0.126)
0.004
(0.004)
-0.181
(0.856)
-2.304
(5.299)
Yes
Yes
Yes
2,449
0.884
0.059
(0.296)
-0.000
(0.007)
-2.209
(2.259)
3.272
(6.911)
Yes
Yes
Yes
2,227
0.588
-0.205
(0.146)
-0.006**
(0.003)
1.410
(1.303)
-11.028*
(6.127)
Yes
Yes
Yes
2,727
0.764
-0.171
(0.223)
-0.005
(0.007)
2.811
(2.043)
-31.956***
(11.404)
Yes
Yes
Yes
2,323
0.604
Note. Data restricted to CDB city level loan to 310 cites from 1998 to 2010. Table shows the results from OLS
regression from equation 1.3. Column I and 2 are the regressions on the logarithm of total CDB loan outstanding
amount and total issuance respectively. Column 3 and 4 are the regressions on the logarithm of CDB infrastructure
loan outstanding amount and new issuance respectively. Column 5 and 6 are the regressions on the logarithm of CDB
industry firm loan outstanding amount and new issuance respectively. PoliticianYear is the number of years that
the secretary has been staying in the city. Constrainedis the dummy for whether CDB debt to GDP ratio is at the
top quartile among 310 cities. GDP, AvgIncome, FiscalIncome and Employment are city level GDP, income per
capita, fiscal income and total employment respectively. All columns are controlled by year fixed effect, city fixed
effect and politician personal fixed effect. Standard errors are clustered at city level.
76
Chapter 2
Do Credit Card Companies Screen
for Behavioral Biases?
2.1
Introduction
Retail financial products have grown in the heterogeneity and complexity of the terms
they offer over the last two decades. A few recent papers have documented this
growth of the retail financial sector, see for example Tufano (2003), Phillipon (2012)
and Greenwood and Scharfstein (2013). But there are widely diverging views about
the reasons behind this proliferation of new products and services. On the one side is
a "functionalist" view as suggest by Merton (1992) or Miller (1993) or more recently
Cochrane (2013). Greater product heterogeneity is seen as a reflection of increased
financial innovation, which allows companies to better accommodate consumer demands and tailor products to their preferences. On the other side is a concern that
the proliferation of new products and features aims to exploit consumers' behavioral
biases and add complexity, which could make it difficult for consumers to compare
prices and across products. This view has been summarized in Thaler and Sunstein
(2008) or Campbell et al. (2011).
While a large empirical literature over the last thirty years has documented different ways how consumers might make mistakes in choosing financial products, much
less work has focused on how firms respond to these biases when designing product
features and optimal pricing strategies. Recent advances on the theory side explore
optimal supply side responses and equilibrium implications, if firms face consumers
with behavioral biases. For an overview of the theoretical approaches see for example,
Dellavigna (2009) or Koszegi (2013).
In this paper we focus on the credit card industry to shed light on the pricing
and product design strategies that credit card issuers use to attract and screen customers. We show that credit card issuers use different credit card features to separate
more sophisticated from less sophisticated customers. For example, low introductory
rates are offered to less educated and poorer customers, while miles are offered to
more educated and richer customers. Conditional on the reward program, the pricing of the cards then differs significantly. Reward programs that are offered to less
77
educated consumers, have more expensive, backward loaded and hidden pricing features, compared to those offered to more sophisticated customers. In addition, cards
with reward programs respond less to changes in the banks' cost of funds, which
suggests that consumers are less sensitive to the pricing of these cards. Overall our
findings suggest that credit card issuers screen for behavioral biases of customers in
order to maximize the rents they extract from (unsophisticated) the customers. But
we also document that card companies take into account the interaction between
a customer's credit risk and heavily relying on a person's behavioral biases. Using
exogenous shocks to the credit worthiness of customers via increases in state level
unemployment insurance, we show that card issuers rely more heavily on backward
loaded credit terms when customers are more protected.
Our results draw on the recent theory literature that explores optimal supply
responses when consumers have behavioral biases. Nonlinear pricing models under
adverse selection a la Mussa and Rosen (1978) or Maskin and Riley (1984) cannot
explain the typical three-part tariffs, which are prominent for credit cards contracts:
the typical card has a low APR or even lower introductory APR and backward loaded
fee structures with very high late fees and overdraft fees.1 These models predict
marginal cost pricing for the last unit of consumption and suggest that the highest
demand consumer will pay the lowest marginal price. The intuition is that under
adverse selection firms do not want to distort the quantity chosen by the highest
demand users in order to maximize the infra-marginal rents they can extract from
these customers. Backward loaded credit card features with high late fees can only
be optimal if customers do not understand their actual cost of credit, since in that
case they will demand credit as if they were facing only the low APR. The two most
prominent approaches of modeling behavioral biases in the credit card market focus
either on self-control and time inconsistency problems of customers, or alternatively
on myopic consumers who for example are not able to value more complex product
features.
An application of the first set of biases is the work by Heidhues and Koszegi
(2010) or Grubb (2009) that derives the optimal credit contract if borrowers have
self-control issues but naively underestimate their likelihood to be tempted in the
future. The profit-maximizing contract for the issuer uses a three-part tariff with low
introductory rates to make it look more attractive to consumers who underestimate
their propensity to have self-control issues in the future. This contract maximizes
the consumer's mistake, since it entices consumers to take on more debt than they
optimally would. In contrast, sophisticated consumers will correctly forecast their
propensity to have self-control issues and thus choose a less backward loaded contract.
Similarly, Grubb (2009) suggests that three part tariff are an optimal response for
firms that face consumers who are overoptimistic about how well they can forecast
the variance of their future demand when choosing a card.
A prominent example of firm responses to myopic consumers is the influential
paper by Gabaix and Laibson (2006) on shrouded attributes. It suggests if myopic
consumers have difficulty understanding the price of add-on features of a product,
'See for example Grubb (2009) for further discussion of this point.
78
companies can attract these consumers by charging a low base price or offering other
enticing features, which are very visible to consumers. But then charging them high
prices via hidden features. In the credit card context that would translate into features such as late fees, high default APRs or overdraft fees. As a result myopic or
unsophisticated consumers transfer rent to the firm and also to more sophisticated
consumers who avoid the expensive add-on features. The theory also suggests that
products or features that are mainly demanded by sophisticated consumers are difficult to shroud and have to be priced at cost. 2
However, there are two additional implications of the model that have not been
discussed in the literature, but as we will show, are important for the credit card
industry. First, Gabaix and Laibson (2006) model a pooling equilibrium, since their
firms have no other tool to separate customers. However, in their model firms would
like to separate the sophisticated from the unsophisticated consumers, since part of the
rents extracted from unsophisticated consumers go to the sophisticated consumers,
since they avoid the costly add-ons and therefore access the product at a price below
cost. We show that credit card companies seem to take into account that cross
subsidy and design reward programs to separate sophisticated from unsophisticated
customers.
Second, the model does not take into account credit risk, since it was written
for one shot product market interactions. However, in credit contracts an excessive
reliance on shrouding and backward loading of payments, could change the credit
risk of customers. In the extreme these pricing strategies could attract customers
who cannot afford the products that are offered, since they do not understand the
features. Or the backward-loaded nature of the contract could entice them to take
on too much leverage that they ultimately cannot afford. This creates an endogenous
limit in how heavily lenders can rely on these strategies. Our analysis suggests that
banks are aware of this trade-off when designing credit card offers and rely more
heavily on these backward loaded features when the credit risk of borrowers improves.
To test the importance of these models for the retail finance market we use data
on preapproved credit card offers and their contract features from the US credit card
industry. Credit cards are an ideal testing ground to observe whether and how firms
use communication and product features to target different customers, since the majority of credit cards in the US are sold via pre-approved credit card solicitations done
by mail. This means that the information that customers get is also observable to
researcher, once we obtain the card solicitation that customers received. In contrast
in almost all other retail financial areas customer choices are intermediated by advisors who might change the information or even product features in ways that are
unobservable to the research. The data for this study are obtained from Compremedia (Mintel) a company that collects monthly information on all credit card mailers
sent to a set of about 7000 representative households, which work with Compremedia
across the US. These clients are chosen to represent the demographic and economic
2
Carlin (2013) suggests a related model where heightened product complexity increases the market power of financial institutions because it prevents some consumers from becoming knowledgeable
about prices. Here complexity works as a (negative) externality on all customers, rather than being
targeted at particular subsets of the population.
79
distribution of the US credit card owning population. Customer characteristics are
parallel with the type of information that credit card issuers observe when targeting
customers. This data allows us to observe the supply side of the credit card market,
i.e. the type of offers that customers receive. Our data set covers the time period
between 1999 and 2011.
Based on the pdf data that we receive from Compremedia, we created algorithms
to extract card information and the features of the offer. We can classify the hard
information of the offer such as APRs, fees, and reward programs. But we also can
observe what we call the "soft" features of the offer, for example the use of photos,
color, font size or amount of information provided in the mailer that the customer
receives.3
We show that credit card issuers use different card features to target especially
educated versus uneducated customers, and rich versus poor customers. Less educated customers and poorer customers receive more card offers with backward loaded
payment features: Low introductory APRs but high late fees, penalty interest rates
and over-limit fees. In contrast richer and more educated people receive almost no
offers with introductory low APRs, and instead are more likely to receive card offers
with cash back programs and points. And especially miles programs are targeted
at more educated customers: Only 8% of offers have a miles program and they are
exclusively offered to the highest income and educational groups.
We then analyze the pricing structure of credit cards conditional on having different reward programs. First, we find that cards that either have cash back or points
programs have lower regular APR. But at the same time these cards display much
higher late fees as well as higher over limit fees. The same relationship holds when
looking at cards that have introductory APR programs as rewards. Regular APRs are
negatively associated with late fees, but cards that have an intro APR program show
a significantly larger trade-off between regular APRs and late fees. These results
go through when controlling for person specific and even bank fixed effects, which
holds constant the credit risk of the person and the cost of credit for the bank. In
this regression we de facto rely on the variation that comes from the fact that banks
randomly send several offers to the same type of households to find out whether the
customer responds to the backward loaded or shrouded features of the card. Therefore, we can estimate how banks on average change the pricing strategy of a card
when using rewards programs such as miles, cash back and others.
Following the approach in Ausubel (2001) we also show that these card features
are associated with lower sensitivity of regular APR of credit cards to changes in fed
fund rate. However, we find that late fees and default APR are more sensitive to
changes in the Fed fund rate. These results suggest reward features might be used
to lower the sensitivity of customers to the price of the cards, either by screening
for customers with behavioral biases or by shrouding the less attractive features of a
3
Since financial institutions in the US have to follow TILA (the Truth in Lending Act) we know
that all the information concerning the card have to be on the pre-approved mailer. In addition
the mandatory Schumer box regulates the disclosure of most of the main card features that have to
be included in the letter. However, issuers can choose how they display the information that they
highlight in the main part of the text.
80
credit card offer.
One interesting exception are miles programs, which we show are mainly offered to
the most educated and richer groups of the population: These cards have significantly
higher regular APR but lower late fees, and show a much smaller trade-off between
the regular APR and backward loaded fees. We argue that this finding is in line with
the idea in Gabaix and Laibson (2006) that product features which are demanded
by more sophisticated (in our case measured as more educated) consumers cannot be
easily shrouded and thus have to be priced upfront.
Finally, we test whether credit card issuers rely more heavily on backward loaded
or shrouded features, when the ultimate credit risk of customers is lower and thus the
borrower's mistake in picking the high-priced product, does not significantly affect
credit risk. For that purpose we look at exogenous shocks to customer creditworthiness, in particular changes in state level unemployment insurance (UI) in the US over
the last decade. 4 UI was increased in staggered fashion across several US states over
the last decade. These changes all went in the direction of providing higher levels of
unemployment insurance as well as longer time period. By reducing the impact on
consumer cash flows in the case of negative shocks, it reduces also a lender's exposure to one of the largest negative economic shock that customers might suffer. This
allows us to run a standard Difference in Difference estimator to regress changes in
card features on UI changes across states and across income groups. Our results show
that he credit conditions of the borrower indeed affect the willingness of card issuers
to rely on shrouding and backward loaded features. We find that increases in the UI
levels leads to an increase in the fraction of offers that have low intro APRs and also
an increase in other reward programs. But at the same time we see an increase in
the late fees and default APR. Taken together these results suggest that credit card
companies realize that there is an inherent trade off in the use of backward loaded
features of credit card offers: They might help in inducing customers to take on more
(expensive) credit, but at the same time they expose the lender to people who pose
a greater risk.
One separate reason why card issuers might use rewards programs which aim to
increase the transaction volume of the card (e.g., miles, cash back or points) is that
card issues get paid by the payment processors (visa and master card) based on how
much vendor fees a card generates. While the incentive related to rewards programs
surely are important, our results suggest that they are orthogonal to the channel
we document in the paper. While one might argue that providing consumers with
rewards (that are valuable) justifies an increase in the overall fees of the card, it
does not explain why the form of the price increase should be via backward-loaded
features, as we find in the case of points, cash back or introductory APR. In addition,
as discussed before, we do see that reward programs that target more educated and
richer people, like miles, have the opposite effect on the shape of the fee structure.
The rest of the paper is structured as follows. Section 2.2 provides a detailed
literature discussion. In section 2.3 we lay out the data used in the study as well
4We follow
Agrawal and Matsa (2013) in using changes in the state level unemployment insurance
limits as a source of variation in employees risk exposure.
81
as variables we constructed for the paper and the design of the sample. Section 2.4
summarizes our results and Section 2.5 concludes.
2.2
Literature Review
Our paper builds on a large literature in economics and marketing that has looked
at how individuals respond to information about product features is displayed when
choosing between complex contracts such as retail financial products, medical insurance contracts or even cell phone plans. For example, Lohse (1997) demonstrates
in an eye-tracking study that colored Yellow Pages ads are viewed longer and more
often than black-and-white ads. Similarly, Lohse and Rosen (2001) suggests that the
use of color and photos or graphics increases the perception of quality of the products that are being advertised and enhances the credibility of the claims made about
the products when compared with non-color advertisements. Heitman et al. (2014)
documents that the way prices and add-on features are displayed, significantly affects
how well people choose between products. Besheres, Choi, Laibson and Madrian
(2010) show that even when subjects are presented with very transparent and easy
to digest information about different mutual funds, they select dominated savings
vehicles. Bertrand et al. (2010) show that the advertising content indeed can have
significant impact on product take up and even willingness to pay. They set up a
field experiment with a consumer lender's direct mailing campaign in South Africa
and find that advertising content which appeals to emotions (such as a woman's versus a man's face) or a simpler display of choices leads people to accept much more
expensive credit products. We build on this earlier literature by analyzing if firms
deliberately incorporate these behavioral biases when designing credit card offers.
There is also a growing literature in household finance that has looked at credit
card usage of borrowers to document that people make non-optimal choices. Agarwal
et al. (2008) analyze more than 4 million transactions of credit card customer to
show that customers on average pay significant fees (late payment and penalty fees)
of about $14 per month, which does not include interest payments. These results
confirm that fees indeed have significant bite and customers are not able to optimally
avoid all the negative features of their cards. The paper also shows that customers
seem to learn to reduce fees over time. But this learning is relatively slow, payments
fall by about 75 percent after four years of account life. Using a similar data set,
Gross and Souleles (2000) show that consumers respond strongly to an increase in
their credit limit and especially to interest rate changes such as low introductory teaser
rates. The long-run elasticity of debt to the interest rate is about -1.3 of which more
than half reflects net increases in total borrowing (rather than just balance switches).
In a related work, Agarwal et al. (2010) document that consumers who respond to
the inferior offers of a lender have poorer credit characteristics ex ante and also end
up defaulting more ex post. Similarly, Agarwal et al. (2009) show that over the lifecycle middle-aged households get the best credit terms, while older customers select
worse credit. The authors conjecture that deterioration in cognitive abilities could
82
be a reason why older people choose worse terms.' These papers provide important
confirmation that credit cards with disadvantageous features are being taken up and
have a significant impact on borrowers cost of capital.
Finally, we relate to a number of papers, which have documented large heterogeneity in the pricing or retail financial products even in the face of increasing competition.
See for example the seminal paper by Ausubel (1991) which documents that credit
card companies have very low pass through rates of any changes in their cost of capital. Hortacu and Syverson (2004) or Bergstresser et al. (2009) show wide dispersion
in fees for the mutual fund industry that is related to changes in the heterogeneity
of the customer base. More recently Sun (2014) and Celerier and Vallee (2014) show
that even with the introduction of increased competition price dispersion does not go
down and product complexity might go up. Similarly, Hastings, Hortacsu and Syverson (2012) look at the introduction of individual savings accounts in Mexico and show
that firms that invested more heavily in advertising had both high prices and larger
market shares, since customers seem to not be sufficiently price sensitive. Similarly,
Gurun, Matvos and Seru (2014) show that areas with large house price increases
and expansion of mortgage originations, saw an increase in marketing expenses and
amounts of marketing solicitations being sent out. These results suggest that firms
seem to compete on nonfinancial dimensions such as advertising to substitute for price
competition.
A paper that uses a very similar data set is Han, Keys and Li (2013), but focus
on a complementary topic. The authors use Mintel data between 2007 and 2011 to
document the large expansion in the supply of credit card debt in the time period
leading up to the financial crisis and after the crisis. The results show that the
expansion prior to crisis was particularly large for consumers with medium credit
scores as supposed to sub-prime customers. In addition they show that even customers
who had previously gone through a bankruptcy still have a high likelihood of receiving
offers, but that these offers are more restrictive.
2.3
2.3.1
Data and Summary Statistics
Data Description
We use a comprehensive dataset from Mintel Comperemedia that contains comprehensive information on the types of credit card offers that customer with different
characteristics receive in the US. This data is based on panel conduced with more
than 4000 households monthly where the household collects all mailings of direct
(preapproved) credit mailing and send this original information to Mintel. For this
data collection effort Mintel selects the households based on their demographic and
5Hastings and Mitchell (2011) use a large-scale, nationally representative field survey from Chile
to directly relate impatience and financial literacy relate to poor financial decisions within a savings
context. The results show that impatience is a strong predictor of wealth. Financial literacy is
also correlated with wealth though it appears to be a weaker predictor of sensitivity to framing in
investment decisions.
83
economic characteristics in order to be representative of the distribution of the US
credit card owning population. For each household, Mintel collects detailed demographic information including the age and education of household head, and household income, composition, races, zip code, etc. Each month, Mintel receives from
the household all credit solicitation mails, such as credit card, home equity loan and
mortgage offers that these selected households received during the month. We only
observe offers to the entire households usually to the head of the household.
After gathering the physical credit card offers from the households, Mintel manually scans the actual mailers in PDF format and electronically enters some key information which usually are contained in the Schumer box: regular purchasing APRs,
balance transfer APRs, cash advantage APRs, default APRs, maximum credit limits,
annual fees, late fees (penalty fees), over limit fees and so on. We manually check the
quality of the dataset and find that all the variables are very well collected except
default APRs which has many missing values.
The data sample ranges from March 1999 to February 2011. For each month,
there are about 4,000 households and 7,000 credit card mail campaigns on average.
In total, there are 1,014,768 mail campaigns which consist of 168,312 different credit
card offers. This is because credit card companies usually issue the same offer to
many households at the same time. Our second data resource from Mintel is the
scanned images of all pages of the credit card offers. However, Mintel only keeps
scanned images of about 80% of the credit card offers. Therefore, 803,285 out of the
total 1,014,768 credit card offers have complete scanned images of each page.
We extract information of reward programs and any soft information about the
design of the mailer itself from these scanned images. First, we use OCR (Optical
character recognition) software to transfer all the images into Word documents. The
OCR software we use is OmniPage Professional version 18.0. It is one of the leading
document imaging software which is accurate and fast. The OCR software separates the characters and graphics/background patterns from the original documents
(scanned credit card offers), recombines them together based on original digital documents' design and turns it into editable Word documents. After that, we use keyword
searching algorithm to search the reward programs in each offer. We are able to identify 8 commonly used reward programs: cash back, point reward, flight mileage, car
rental insurance, purchase protection, warranty protection, travel insurance and zero
introductory APRs.
Moreover, because we keep format information of each character in the offers, we
can also get the format design of these reward programs. By using Word application
in VBA, we are able to identify the font of the characters. We collect the size and color
of each reward program when they were mentioned in the offer as well as whether they
were highlighted with bold or italic. Also, we count the number and size of pictures
on each page. To check the quality of the OCR and keyword searching algorithm,
we randomly select some offers to manually check the accuracy which turns out to
be over 90%. As we mention before, there are some missing values for default APRs
from Mintel's hand collect database. To deal with it, we also use keyword searching
algorithm to search the default APRs stated in the offers. Usually, the Schumer box
contains default APRs which sometimes called penalty APRs. We extract default
84
APRs for all credit card offers with scanned images by using our algorithm and
compare them with the ones collected by Mintel. The accuracy of our algorithm is
about 98%. In this way, we are able to add back some of the missing values to almost
complete the default APRs data. Because we only have 80% samples with scanned
offers, our variables for reward programs and format are limited to these 80% sample.
2.3.2
Descriptive Statistics
Table 2.1 is the summary statistics. FFR is the monthly average federal fund rate
from January 1999 to December 2011. We merge FFR into our credit card dataset
by month. In Table 2.1, variables from APR to Intro APR-cash are based on our
entire 1,014,768 mail campaigns from Mintel. APR is the regular purchasing APRs
in the credit card offers. Sometimes, regular APR is a range and we pick the middle
point as the APR. The mean of APR is 12.64% among 982,796 total mailings received
by consumers. The APRs for balance transfer has a mean of 11.33% and standard
deviation of 3.34%. The cash advance APR has a mean of 19.88% and the standard
deviation is 4.28%. For default APRs, the mean is much 26.51% which is higher
than all other APRs. The credit card companies charge very high default APRs
which may be applied to all outstanding balances of a credit card if a consumer pays
the monthly bill late (usually in 60 days). All these APRs are monthly couponed.
IntroAPRregular, IntroAPRbalance and IntroAPRcash are the dummies
of whether the offer has 0% introductory APR for regular purchase, balance transfer
and cash advance respectively. "Max Card limit" is the nature log of the maximum
credit card limit stated in the offers. We only have 526,949 observations for "Max
Card limit" since many credit card offers don't have it, especially during later years.
Credit cards also have a number of different fee types, the dimensions that we
observe in the data are annual fee, late fee and over limit fee in our sample. Annual
fee on average are $12.28 with a standard deviation of 31.99. The distribution of
annual fee in our sample is pretty skewed. 81.5% of the mailing offers have zero
annual fee and the maximum annual fee is $500. Typically the types of cards that
have annual fees associated with them, offer mileage programs and other expensive
value added services. Late fee is the one-time payment if the consumer misses paying
at least their minimum monthly payment by the due date. In our sample, late fee
has a mean of $33.83 and a standard deviation of 6.17. It is much less skewed than
annual fee. About 90% of the credit card offers have late fee from $29 to $39. Since
this is a fixed monthly fee that comes due if the minimum payment has not been,
one can imagine that the cost of this fee structure can be very high especially when
customers carry small balances.
Finally, over limit fee is the fee charged when consumers' credit card balance goes
over the card limit. The mean of over limit fee is $29.7 with a standard deviation of
10.16. The distribution of over limit fee is also concentrated: about 87% of the cards
has over limit fee from $29 to $39. Although credit card companies usually charge
zero annual fee, they do charge much more from late payment and over borrowing.
In Table 2.1, variables from "Size" to "Purchaseprct" are from 80% of 1,014,768
total mail campaigns which have scanned images of credit card offers. "Size" is the
85
maximum size of the reward programs minus the average size of all characters in
every pages of each credit card offer. For example, if the letters "cash back" appear
3 times in the offer, we pick the largest one. "Size" equals this largest number minus
the average size of all characters on the same page. The unit of size is directly from
Word document. The variable "Size" has 4.71 mean and 5.49 standard deviation. The
maximum value of Size is 143.6 because some offers will print very large characters
to highlight reward programs. The 90th percentile of variable Size is 10.99. We
use this relative size measurement because credit card companies tend to enlarge
the reward program characters' size relatively to the paragraphs nearby in order to
highlight the reward programs. The size differences between them should be the
measure of highlight. Moreover, "Color" is the dummy of whether reward programs
in the offer use color other than black/white. We only focus on the characters of
reward programs rather than the entire offer since almost every credit card offers use
some colors, especially during later years. 6 "Bold" is the dummy of whether the offer
use bold to highlight reward programs.
"Picture" is the file size of each page of the offer which is the measurement of how
many or how "fancy" the pictures are in the offer. We don't use actual count of the
pictures nor the size of the pictures because our algorithm considers the background
of the page as a big picture as well (usually it is just a big plain color picture). Using
storage size of each Word document, we can approximate how complicated the design
of the page is. Other information such as characters also use some storage. However,
Pictures in Word documents usually take most of the storage room. We think that
file storage size is a good measurement of the pictures in the credit card offers. The
variable "Picture" is the file storage size and the unit is megabyte (MB). The mean
of "Picture" is 0.23MB with 0.26MB standard deviation. In the appendix, we plot
two samples of the credit card offers. Figure 2-7 is the sample of simple visual offer,
which doesn't use big font, flashy colors, or pictures to highlight the reward programs.
Figure 2-8 is the sample of high visual offer with many fancy designs.
Moreover, we define "Reward" as the number of reward programs of CASH,
POINT and Car rental insurance in each offer. We choose these three reward programs because they are similar and most commonly used. CASH, POINT, MILE,
Carrental, Purchaseprct are dummies of whether the offer has these reward programs
respectively.
Table 2.2 summarize the design of the credit card offers. All credit card offers state
late fees, default APRs, over-limit fees, and annual fees. Only 5.8% of the credit card
offers mention late fees in the first page of the offers. 4.97% of the credit card offers
mention default APRs in the first page of the offers. 6.96% of the credit card offers
mention over-limit fees in the first page of the offers. Basically, credit card offers
usually don't mention backward loaded terms in the first page of the offer. On the
other hand, 79.28% of the credit card offers put annual fees information on the first
page which is considered to be the front loaded term. Moreover, reward programs
6To construct the format variables such as Size, Color, and Bold, we only focus on the reward
programs fonts which include cash back, point rewards, mileage, car rental insurance, purchase
protection, and low intro APR programs.
86
are usually mentioned in the first page of the offers; 100% of the cash back program
and mileage programs are mentioned in the first page. For point reward, car rental
insurance, and 0 introductory APRs, the changes to be put on the first pages are
93.51%, 80.48%, and 91.04% respectively. Panel B of Table 2.2 compares the credit
card terms when they are mentioned on the first page or not. It is clear that late fee,
over-limit fee, and annual fee are much lower in the offers when they are mentioned
on the first page than in the offers when they are mentioned in the back.
Timetrend: In figure 2-1 to figure 2-4, we plot the monthly time trend of reward
programs and variables of credit card design from March 1999 to February 2011. As
we can see, the number of reward programs increases overtime and appears cyclical.
For example, there is a drop of reward program numbers during the recent crisis. The
uses of size, color and pictures in credit card offers also increase overtime and have
cyclical patterns.
2.4
2.4.1
Results
Customer Characteristics and credit card features
We start by analyzing how the features of offers vary with the characteristics of
the customers. The analysis allows us to understand whether specific credit card
features are used to target different client groups. But the exercise also allows us
to see if the relationships in the data are intuitive. For that purpose in Table 2.3
we run a simple regression model of card features, such as card APR, late fees or
reward program on customer characteristics. The characteristics of interest for us
are the education levels of customers, which are measured as six distinct educational
achievement dummies ranging from some high school to graduate school. And nine
different income groups ranging from the lowest income group of less than $15,000 to
the highest of over $200,000. In these regressions we also control for the age group
fixed effects of the customer, the state fixed effect, dummies for household composition
and credit card company fixed effects. Standard errors are clustered at the month
level. It is important to note that these customer characteristics are also the same
ones that credit card companies observe when they send out mailing campaigns.
In column 1 of Table 2.3 we start with the mean APR as the dependent variable
and report the coefficients on the education and income bins. The results show
that regular APR decreases significantly for higher income groups and the results
are relatively monotonically going up with increasing income. The magnitude of the
effect is quite large. Between the lowest and the third highest income groups there
is a difference in mean APR of almost 0.561 percentage points which is a significant
difference. The relationship between APR and income levels off a little for the highest
two income groups, but we will show that these groups also have different product
features. In contrast there is no significant relationship between the regular APR and
education. The estimated coefficients are all close to zero and insignificant. We repeat
the same regression for APR on balance transfers and APR on cash withdrawals and
get very similar results. The regressions are not reported but can be obtained from
87
the authors. These results intuitively suggest that higher income customers are better
credit risk and as a result enjoy a lower cost of capital. Interestingly the same is not
true for more educated customers.
We now use the logarithm of the maximum credit balance as the left hand side
variable and repeat the same regression set up as in column (1). We see that limits
increase with higher education bins, but the increase is even steeper with income
categories. It is intuitive to think that credit card companies are more comfortable
increasing their exposure to customers with higher income and better educational
attainment since this might be correlated with increased future earnings potential.
Interestingly in columns (3) and (4) these results reverse when we look at Default APR and late fee. These are backward loaded fees which become due when the
customer is either 30 days late or becomes more than 90 days delinquent. Very surprisingly we find that late fees and default APR increase significantly with customer
income, but drop with higher educational attainment. For example the difference in
default APR between the lowest and the highest income group is about 0.543 percentage points. So customers with higher income actually face higher default interest
rates than those with lower income. The same pattern holds true for late fees. In
contrast, customers with higher education receive card offers that have smaller late
fees and lower default APR. This result is very puzzling: If default APR or late fees
are a means to manage customer default risk, it is not clear why customers with
higher income would be a higher default risk conditional on falling late, compared
to customers with lower income. These results might be a first indication that these
interest rates are set with a different set of considerations in mind.
In a next step we look at how reward programs are offered to customers. In column
(5) the dependent variable is a dummy equal one if the credit card offer contains a
cash back program. We see that there is a strong positive correlation with income,
between the highest and lowest income group there is a 4 percent difference that a
card has cash back program. This is economically very substantive since only 21
percent of card offers contain a cash back program. In contrast we do not see any
relationship between educational levels and the likelihood of receiving a credit card
offer with a cash back program. In column (6) we see a very similar result for points
programs. Again there is a statistically and economically significant increase in the
likelihood of receiving an offer with a points program for households that have higher
income levels. But there is no relationship with educational levels.
We observe a very different relationship when looking at miles programs. In
column (7) we show that the likelihood of receiving a card offer with a miles program
increases significantly with the education level of the household. Households in the
second to last highest income group are more than percent more likely to receive
an offer with a miles program compared to a household in a the lowest educational
bin. Since only eight percent of credit card offers include a miles program in the
first place, education seems to be a very important dimension in receiving miles
programs. We also see that miles programs increase with the income level of the
customer. The final reward program we look at are low introductory APR offers.
These usually expire after a few months (customarily between six to 12 months) and
then a higher interest rate kicks in. In column (8) we see that introductory APR
88
programs are predominantly offered to less educated and lower income customers. A
similar relationship holds for introductory APR rates on balance transfers.
Moreover, in Table 2.3 Column 9, "Format" is the first principal component of
reward programs' size, color, bold and the picture sizes on the credit card offers.
We show that more educated households or high income households can get fancier
designed credit card offers which usually use bigger font size, more colors, more bold,
and more pictures to emphasize the reward programs.
As discussed above, the detailed customer information in the Mintel data allows
us to analyze how credit card issuers target customers with different characteristics,
for example across the income and educational attainment distribution. However,
one dimension that we do not have in our data, are the FICO scores for individual
borrowers. To analyze if the lack of FICO scores in our data is a significant limitation,
we obtained Mintel data via the CFPB. While the data covers a shorter time period
than ours, starting only from 2007, it has the advantage of containing also FICO
scores.
The idea it to see if credit card features differ significantly by FICO scores, after
controlling for all the other observable characteristics of the customer. This is equivalent to asking whether card issuers use FICO scores to screen a different dimension
of the borrowers from all the other characteristics. For that purpose we repeat our
waterfall regressions where we regress card features on the different customer characteristics and then add FICO scores as an additional explanatory variable. Adding
the customers' FICO scores does not add any additional explanatory power to the
regression. The adjusted R-squared of the regressions are unchanged and none of the
coefficients on other RHS variables change when including the FICO scores. So, overall it appears that the dimensions spanned by the FICO scores are jointly spanned
by the other observable characteristics. These results reduce the concern that we are
missing an important, and separate screening dimension.
Taken together these results suggest that different reward programs are used to
target different customers groups. Introductory APR offers are primarily offered to
less educated and poor clients. In contrast points and cash back programs are offered
to richer customers but independent of their educational level. Finally, miles is the
only reward program that are predominately targeted to richer and importantly to
more educated customers. We plot the coefficients from Table 2.3 in Figure 2-5 and
2-6 to make the patterns more clear. Figure 2-5 plots the estimated coefficients of the
education on credit card terms and reward programs. Figure 2-6 plots the estimated
coefficients of the income level on credit card terms and reward programs.
2.4.2
Pricing of Credit Cards
In a next step we now want to understand how the pricing of credit card offers
changes with reward programs. The idea is to test if these reward programs are
used in combination with more backward loaded credit card features or different
pricing levels. If indeed rewards are used to attract customers who are more prone
to shrouding we would expect more pricing of hidden features and more backward
loading. For that purpose we will investigate the different reward programs separately.
89
We follow the general idea pioneered in Ausubel (1991) to assume that APR should
be very sensitive to the Fedfundrate since this is the rate at which the banks can raise
capital. Parallel to Ausubel (1991) we will interpret any elasticity between the APR
and FFR that is much below 1 as an indication that credit card issuers are insulated
from competition. But in our case we want to understand whether the use of reward
programs allows issuers to react less strongly to changes in the FFR and also use
more backward loaded fee structures. Our first regression specification is
Yi,jt =
#1 x
FFRM + /2 x RPj,t + FEi,jt + cij,t,
(2.1)
where Yi,jt indexes the dependent variables we are interested in such as regular purchasing APRs, default APRs, late fee and over limit fee. For example, APRi,j,t is
the regular purchasing APR of the credit card offer issued by company i to consumer
j at time t. FFRM indexes the federal fund rate at month M. RPjt indexes the
dummy of certain reward program in the credit card offer such as cash back, point
reward, flight mileage and zero introductory APRs. We also control the fixed effects
such state fixed effects, bank fixed effects, and household demographic fixed effects. 7
t is at daily frequent.
Also, we explore the sensitivity between APRs and FFR by adding interaction
terms of FFR and reward programs:
Yi,j,t = #1 x FFR + 02 x RPjt +
/3
x RPi,jt x FFR + FEi,jt + (i,j,t,
(2.2)
We cluster the standard errors at the cell level.
Points Program
In Table 2.4 Panel A, we focus on the use points programs. Ex ante there is no
obvious reason why a credit card offer should even be linked with a points program,
since the credit card company could rather offer a lower interest rate. In column
(1) we regress the regular APR on the FFR and an indicator for whether the credit
card contains a points program. In this first column we only control for state fixed
effects but no person specific characteristics. We see that the coefficient on FFR
is only 0.326, while the coefficient is highly significant it indicates that there is less
than perfect pass through of the cost of capital to customers. We also see that those
cards that have a points program have lower APR. But in this baseline specification
the negative coefficient could be driven by a composition effect, since was we saw
in the prior Table only higher income customers receive credit cards with points.
Therefore, in column (2) we include cell fixed effects in the regression. This means
we control for every consumer cell, which is a unique combination of the customers'
income bin, educational attainment, household composition, age and the state that
she lives in. This specification allows us to test how credit card offers that have points
programs are priced compared to those without a points program holding constant
'We construct the household demographic cells by age, education, income, household composition,
and the states.
90
the customer characteristics. As we see in column (2) the coefficient on Points stays
negative and only drops by about 10 percent, which means that it cannot be driven
by the customer's observable credit type but has to be specific to the pricing strategy
of a given card.
In column (3) of Table 2.4 Panel A we now add bank fixed effects to the regression.
This allows us to control for the differences in pricing strategies between banks. We
see that with bank fixed effect the coefficient on Points drops by a huge factor, almost
ten! This result suggests that banks differ significantly in their use of point programs
and those issuers which use point programs extensively are also those that charge lower
APR rates. However, in this most stringent specification, we still find a negative and
significant coefficient on the POINT dummy. This means that when two credit cards
are sent to the same person by the same bank, but one card offers a reward program
with points to the customer and the other does not, the one with the points program
has a lower APR affiliated with it. This is very surprising since the production of
points is most likely not costless and thus we need to ask how the card company will
break even on the production of such points.
In the remaining columns of the table we now analyze how this relationship correlates with other card features. First in column (4) we follow Ausubel (1991) and
interact Points x FFR to test if credit cards that offer points are less sensitive to
changes in the FFR. The results suggest that this is indeed true, since the coefficient
on the interaction term is minus 0.244 and significant at the one percent level. This
means that credit cards that have points programs do not need to adjust as quickly to
changes in the FFR. However, maybe more interestingly in the context of our analysis
in column (5) we now use the Default APR as a dependent variable. We repeat the
same regression as in columns (2) regressing Default APR on FFR and Points controlling for cell fixed effects. We show that Default APR increases significantly with
Points, again keep in mind that this is holding constant all the characteristics of the
recipient of the offer. When we now interact Points x FFR in column (6) we again find
that the there is a large and significantly negative coefficient suggesting that Points
programs allow credit card issuers to maintain higher default APRs and not pass on
any changes in their cost of capital to consumers. And finally we repeat the same
regression specifications in columns (8) to (10) but using late fees as the dependent
variable. Again we see a very strong increase in late fees for cards that have a point
program. But in this case the reward program is associated with higher sensitivity of
late fees to underlying changes in the FFR. This could indicate that late fees are an
important way to make adjustments for the banks changing cost of capital.
Cash Back Program
In Table 2.4 Panel B, we focus on the use cash back programs. In column (1) we
regress the regular APR on the FFR and an indicator for whether the credit card
contains a cash back program. As before, we first only control for state fixed effects
but no person specific characteristics. We see that the coefficient on FFR is only
0.312, while the coefficient is highly significant it indicates that there is less than
perfect pass through of the cost of capital to customers. We also see that those cards
91
that have a cash back program have lower APR. Parallel to before, in column (2)
we include cell fixed effects in the regression. This specification allows us to test
how credit card offers that have cash back programs are priced compared to those
without a cash back program holding constant the customer characteristics. As we
see in column (2) the coefficient on cash back stays negative and only drops by about
8 percent. In column (3) of Table 2.4 Panel B we now add bank fixed effects to the
regression. This allows us to control for the differences in pricing strategies between
banks. We see that with bank fixed effect the coefficient on cash back drops by a
factor of three. This result suggests that banks differ significantly in their use of cash
back programs and those issuers which use cash back programs extensively are also
those that charge lower APR rates.
We now analyze how this relationship correlates with other card features. First in
column (4) as before we follow Ausubel (1991) and interact cash back x FFR to test if
credit cards that offer cash back are less sensitive to changes in the FFR. The results
suggest that this is indeed true, since the coefficient on the interaction term is minus
0.363 and significant at the one percent level. However, maybe more interestingly
in the context of our analysis in column (5) and (6) we now use the Default APR
as a dependent variable. When we interact cash back x FFR in column (6) we again
find that the there is a large and significantly negative coefficient suggesting that
cash back programs allow credit card issuers to maintain higher default APRs and
not pass on any changes in their cost of capital to consumers. And finally we repeat
the same regression specifications in columns (8) to (10) but using late fees as the
dependent variable. Again we see a very strong increase in late fees for cards that
have a cash back program. But in this case the reward program is associated with
higher sensitivity of late fees to underlying changes in the FFR. This could indicate
that late fees are an important way to make adjustments for the banks changing cost
of capital.
Mileage Programs
In Table 2.5 Panel A, we focus on the use of mileage programs. In column (1) we
regress the regular APR on the FFR and an indicator for whether the credit card
contains a mileage program. Again, we only control for state fixed effects but no
person specific characteristics. Moreover, we see a very strong increase in APR for
cards that have a mileage program. The coefficient on MILE is 1.698, significant at
one percent level. In column (2) we again include cell fixed effects in the regression.
As we see in column (2) the coefficient on mileage stays significantly positive and
increases by about 24 percent.
In column (3) of Table 2.5 Panel A, we now add bank fixed effects to the regression.
This allows us to control for the differences in pricing strategies between banks. We
see that with bank fixed effect the coefficient on mileage stays almost the same. This
result suggests that banks don't differ significantly in their use of mileage programs.
In the remaining columns of the table we now analyze how this relationship correlates with other card features. First in column (4) we follow Ausubel (1991) and
interact mileage x FFR to test if credit cards that offer mileage are less sensitive to
92
changes in the FFR. The results suggest that this is indeed true, since the coefficient
on the interaction term is minus 0.363 and significant at the one percent level. In
column (5) and (6) we now use the Default APR as a dependent variable. We find
similar results as regular APRs in column (1) and (2). Finally we repeat the same
regression specifications in columns (8) to (10) but using late fees as the dependent
variable. Again we see a very strong increase in late fees for cards that have a mileage
program. But in this case the reward program is associated with higher sensitivity of
late fees to underlying changes in the FFR. This could indicate that late fees are an
important way to make adjustments for the banks changing cost of capital.
Zero Introductory APR
In Table 2.5 Panel B, we focus on the use zero introductory APRs programs. In
column (1) we regress the regular APR on the FFR and an indicator for whether the
credit card contains a zero introductory APRs program. Again, we only control for
state fixed effects but no person specific characteristics. We see that the coefficient
on FFR is only 0.397, while the coefficient is highly significant it indicates that there
is less than perfect pass through of the cost of capital to customers. We also see that
those cards that have a zero introductory APRs program have lower APR. But in
this baseline specification the negative coefficient could be driven by a composition
effect, since was we saw in the prior Table only higher income customers receive credit
cards with zero introductory APRs. Therefore, in column (2) we include cell fixed
effects in the regression. This specification allows us to test how credit card offers that
have zero introductory APRs programs are priced compared to those without a zero
introductory APRs program holding constant the customer characteristics. As we
see in column (2) the coefficient on zero introductory APRs stays negative and only
drops by about 4 percent, which means that it cannot be driven by the customer's
observable credit type but has to be specific to the pricing strategy of a given card.
In the remaining columns of the table we now analyze how this relationship correlates with other card features. First in column (3) we follow Ausubel (1991) and
interact zero introductory APRs x FFR to test if credit cards that offer zero introductory APRs are less sensitive to changes in the FFR. The results suggest that this is
indeed true, since the coefficient on the interaction term is minus 0.566 and significant
at the one percent level. However, maybe more interestingly in the context of our
analysis in column (4) and (5) we now use the Default APR as a dependent variable.
When we interact zero introductory APRs x FFR in column (6) we again find that
the there is a large and significantly negative coefficient suggesting that zero introductory APRs programs allow credit card issuers to maintain higher default APRs
and not pass on any changes in their cost of capital to consumers. And finally we
repeat the same regression specifications in columns (8) to (10) but using late fees
as the dependent variable. Again we see a very strong increase in late fees for cards
that have a zero introductory APRs program. But in this case the reward program
is associated with higher sensitivity of late fees to underlying changes in the FFR.
This could indicate that late fees are an important way to make adjustments for the
banks changing cost of capital.
93
Format of the Reward Programs
In Table 2.6, we focus on the design of the credit card offers. We use the font size,
color, and bold of the reward programs on the offers. We also use the picture sizes to
measure the design of the credit card offers. "Format" is the first principal component
of reward programs' size, color, bold and the picture sizes on the credit card offers.
In column (1) we regress the regular APR on the FFR and Format. Again, we only
control for state fixed effects but no person specific characteristics. We see a very
strong increase in APR for cards with fancier design. The coefficient on Format is
0.374, significant at one percent level. In column (2) we again include cell fixed effects
in the regression. As we see in column (2) the coefficient on Format stays significantly
positive and increases by about 2 percent.
In column (3) of Table 2.6, we now add bank fixed effects to the regression. This
allows us to control for the differences in pricing strategies between banks. We see
that with bank fixed effect the coefficient on Format stays almost the same. This
result suggests that banks don't differ significantly in their use of mileage programs.
In the remaining columns of the table we now analyze how this relationship correlates with other card features. First in column (4) we follow Ausubel (1991) and
interact Format x FFR to test if credit cards with fancier design are less sensitive to
changes in the FFR. The results suggest that the sensitivity doesn't change significantly with Format. In column (5), (6) and (7) we now use the Default APR as a
dependent variable. We find similar results as regular APRs in column (1) and (3).
However, in column (7), we find that default APRs of credit cards with fancier design
are significantly less sensitive to changes in the FFR. Finally we repeat the same
regression specifications in columns (8) to (10) but using late fees as the dependent
variable. Again we see a very strong increase in late fees for cards with fancier design.
Format again is associated with lower sensitivity of late fees to underlying changes in
the FFR. This could indicate that fancy designs such as big colorful font, can make
back loaded terms less sensitive to banks cost of capital.
In sum these results suggest that the majority of reward programs are designed
to allow the company to backward load a lot credit card payments to a later time
period when credit terms are very expensive. For example through the use of late
fees and default APRs. At the same time credit cards with point programs are
even in the current time less sensitive to the ups and downs of the FFR. A similar
relationship holds for cash back program. It is interesting that both of these programs
are used in combination with backward loading of payments, since schemes like points
or cashback are not inherently depending on the time factor. This might point to the
idea that these reward programs are used to shroud the more expensive aspects of a
card program. Similarly cards with introductory APR offers have almost by definition
more backward loaded payment features.
The one big exemption are cards with mileage programs. These cards seem to be
predominately targeted at more educated customers. These cards are associated with
a higher mean APR rate but much lower late fees and default APR.
94
2.4.3
Trade-off between Regular APRs and Late Fees
From above, we see that credit card companies use different reward programs with
different pricing strategies and target different consumer groups. We explore the
pricing trade-offs between font loaded terms such as regular APR and back loaded
terms such as late fees. In Table 2.7 column (1), we first look at the trade-offs between
regular APRs and late fees by regressing regular APRs on late fees. We only control
state fixed effects in column (1). We find that a $1 increase in late fee is associated
with a 0.06% increase in regular APRs. When we control for cell fixed effects and
bank fixed effects in column (2) and (3), we still find that regular APRs are negatively
associated with late fees. This suggests that credit card companies usually choose to
decrease regular APRs and increase late fees to make the terms more backward loaded
or choose to increase regular APRs and decrease late fees to make the terms front
loaded.
Moreover, we interact the reward programs (cash back, point, and car rental insurance programs) with late fees in column (4) and find significantly negative coefficient
on the interaction term. This suggests that more reward programs led to significantly
more trade-offs between late fees and regular APRs. Then, we compare the heterogeneous effects of mileage program and intro APR program on credit card pricing
strategy. In Table 2.7 column (5) and (6), we find that intro APR programs led to
significantly more trade-offs between late fees and regular APRs. Interestingly, in
Table 2.7 column (7) and (8), we find that mileage programs led to significantly fewer
trade-offs between late fees and regular APRs. This is consistent with our previous
findings; Intro APR programs led to more backward loaded terms but not the mileage
program which usually targets the well-educated consumers.
2.4.4
Unemployment Insurance
We now analyze the effect of changes in the unemployment insurance's effects on
credit card terms and reward programs. The idea is to use an exogenous shock to the
credit worthiness of customers, in particular their risk of default.
UI was increased in staggered fashion across several US states over the last decade.
These changes all went in the direction of providing higher levels of unemployment
insurance as well as longer time period. By buffering consumer cash flow in the
case of negative shocks, it reduces also a lender's exposure to one of the largest
negative economic shock that customers might suffer. We obtain data on the level
of unemployment insurance (UI) from the U.S. Department of Labor for each state.
Based on this information we calculate annual changes in UI at the beginning of each
year from 1999 to 2012 and match them into our credit card dataset. Following,
Hsu, Matsa and Melzer (2012) we use the maximum UI benefits as the measure of
unemployment protection. We define the maximum Ul benefits as the product of
the maximum weekly benefit amounts (WBA) and the maximum number of weeks
allowed. For example, in January 2000, Alabama allows a maximum of 26 weeks
unemployment insurance during 52 week period and the maximum weekly benefit
amounts (WBA) is $190. We use $4,940 (26 weeks times $190 WBA) as the level of
95
UI. For each state, we then calculate the annual percentage increase of UI. We use
10% annual growth as the cut-off and define a UI "jump" if it increases more or equal
to 10% within a year.
This allows us to run a standard Difference in Difference estimator to regress
changes in card features on UI changes across states and across time. We use one
year before and after the UI jumps to perform the Difference in Difference regressions.
The reason to use this short cut off is that some states have a large increase in UI in
one year and then small changes in follow on year. So we did not want to confound
the impact of the UI change with small subsequent changes. Table 2.8 Panel A is
the one year Diff-in-Diff regression results across our entire sample period (1999 to
2011). In column 1, the coefficient of UI jump is not significant. This suggests that
an increase in the UI doesn't have significant effects on regular APRs. But in column
(2) and (3) we see a greater reliance on backward-loaded payment features such as
a very strong increase in the late fees and balance transfer APRs. Column 4 is for
annual fees. UI doesn't have significant effects on annual fees. In the last column,
we also look at whether credit card issuers use more intro APR programs when UI
increases. For that purpose we build a dummy variable "Intro_APR_ All" for whether
the credit card offer has either zero intro APR for regular purchases, balance transfer,
or cash advance. We see in column (5) that indeed card issuers use more intro APR
programs after UI increases have been implemented. We control year fixed effects,
cell fixed effects, and bank fixed effects. We also repeated this regression set up using
other time windows, e.g., two year windows around the change and the results are
qualitatively and quantitatively very similar.
We now repeat the same analysis in Table 2.8 Panel B but from 1999 to 2007. A
lot of UI increases happened between 2008 and 2010. Since many other things also
happened during this crisis period, there is a concern that other hidden variables drive
our results. In order to mitigate this concern, we drop years after 2007 and only focus
the period from 1999 to 2007 in Panel B. The regression results in Panel B are very
similar with Panel A; Balance transfer APR, late fee, and zero intro APR increase
after the UI jump. Moreover, when we drop the bank fixed effects, the regression
results are quite similar with Table 2.8. This means that the results are not driven
by banks differentially selecting to offer credit cards in states with UI changes. Our
results are driven by the variation within bank decisions to change pricing policies
based on the UI changes.
Taken together these results suggest that credit card companies realize that there
is an inherent trade off in the use of backward loaded or shrouded features of credit
card offers: They might help in inducing customers to take on more (expensive)
credit, but at the same time they expose the lender to people who pose a greater risk.
Therefore we observe a greater reliance on these features when the customer pool has
an (exogenous) improvement in credit quality.
96
2.5
Conclusion
The paper shows that credit card issuers use different card features to separate customer groups with higher or lower propensity for behavioral biases. We show that
less educated and poorer customers receive more card offers with backward loaded
payment features, and they are also less likely to receive rewards programs that are
targeted at richer and more educated people, especially miles but also points and cash
back programs. In contrast, richer and more educated people receive more card offers
with miles, cash back and points, but are much less likely to receive offers with low
intro APR. This latter customer group gets on average better terms: lower interest
rate and fees. Interestingly we find that cards with rewards that are predominantly
offered to richer and more educated people do not show backward-loaded pricing
structures. These results are in line with the insight of Gabaix and Laibson (2006)
that suppliers will not offer shrouded terms on products which are mainly demanded
by sophisticated consumers, since they can undo the shrouding and as a result hurt
the profits of the firm.
Finally our analysis highlights and important dimension of the use of shrouding
and backward loading that has previously been ignored in the literature. It is beneficial for banks to maximize shrouding and backward loading of payments if this does
not change the credit risk of customers. However, at an extreme these pricing strategies could attract customers who cannot afford the products that are offered. Using
shocks to unemployment insurance, which reduce the credit risk of especially poorer
customers, we show that banks are well aware of this trade off. They are willing
to extend more backward loaded or "shrouded" credit to these customers when their
overall credit risk is lower.
The results in this paper provide evidence that credit card companies do screen for
behavioral biases via the type of reward programs and low introductory APRs that
are offered to customers. However, the interaction between behavioral screening and
classic adverse selection is much more complex than the prior theory literature has
taken into account. There appears to be an inbuilt trade off between the immediate
benefits from using shrouded terms to charge higher cost of capital (via a combination
of interest rates and fees) and their impact on increasing the credit risk of the customer
pool by drawing in customers who do not understand the credit terms that they are
offered and thus have a higher chance that they can ultimately not afford them.
97
2.6
Bibliography
Agarwal, Sumit, John C. Driscoll, Xavier Gabaix, and David Laibson. "Learning
in the credit card market", NBER Working Paper (2008).
Agarwal, Sumit, John C. Driscoll, Xavier Gabaix, and David Laibson. "The Age
of Reason: Financial Decisions over the Lifecycle with Implications for Regulation",
Brookings Papers on Economic Activity, Vol. Fall,(2009):51-117.
Agarwal, Sumit, Souphala Chomsisengphet, and Chunlin Liu. "The Importance
of Adverse Selection in the Credit Card Market: Evidence from Randomized Trials
of Credit Card Solicitations, Journal of Money, Credit, and Banking, 42.4 (2010):
743-754.
Agrawal, Ashwini K., and David A. Matsa."Labor unemployment risk and corporate financing decisions." Journal of FinancialEconomics 108.2 (2013): 449-470.
Ausubel, Lawrence M. "The failure of competition in the credit card market."
The American Economic Review (1991): 50-81.
Bergstresser, Daniel, John MR Chalmers, and Peter Tufano. "Assessing the
costs and benefits of brokers in the mutual fund industry." Review of FinancialStudies
22, no. 10 (2009): 4129-4156.
Bertrand, Marianne, Dean S. Karlan, Sendhil Mullainathan, Eldar Shafir, and
Jonathan Zinman. "What's advertising content worth? Evidence from a consumer
credit marketing field experiment." Quarterly Journal of Economics, 125(1) (2010):
263-306.
Beshears, John, James J. Choi, David Laibson, and Brigitte C. Madrian. "Simplification and saving." Journal of economic behavior & organization 95 (2013): 130-
145.
Campbell, John Y., Howell E. Jackson, Brigitte C. Madrian, and Peter Tufano.
"Consumer financial protection." The journal of economic perspectives, 25.1 (2011):
91-114.
Carlin, Bruce I. "Strategic price complexity in retail financial markets." Journal
of FinancialEconomics, 91.3 (2009): 278-287.
Cochrane, John H. "Finance: Function matters, not Size." Journal of Economic
Perspectives, 27 (2013): 29-50.
Vigna, Stefano Della, and Ulrike Malmendier. "Contract design and self-control:
Theory and evidence." The Quarterly Journal of Economics (2004): 353-402.
DellaVigna, Stefano. "Psychology and economics: Evidence from the field."
Journal of Economic Literature 47.2 (2009): 315-372.
Gabaix, Xavier, and David Laibson. "Shrouded attributes, consumer myopia,
and information suppression in competitive markets." The Quarterly Journal of Economics 121.2 (2006): 505-540.
Greenwood, Robin M., and David S. Scharfstein. "The growth of modern finance." Journal of Economic Perspectives. (2012).
Gross, David B., and Nicholas Souleles. "consumer response to changes in credit
supply: evidence from credit card data." Center for Financial Institutions Working
Papers 00-04, Wharton School Center for Financial Institutions, University of Pennsylvania (2000).
98
Grubb, Michael D. "Selling to overconfident consumers." The American Economic Review, 99.5 (2009): 1770-1807.
Gurun, Umit G., Gregor Matvos, and Amit Seru. "Advertising expensive mortgages." National Bureau of Economic Research working paper (2013).
Han, Song, Benjamin J. Keys, and Geng Li. "Unsecured Credit Supply over
the Credit Cycle: Evidence from Credit Card Mailings." unpublished working paper
(2013).
Hastings, Justine S., and Olivia S. Mitchell. "How financial literacy and impatience shape retirement wealth and investment behaviors.", National Bureau of
Economic Research (2011).
Hastings, Justine S., Ali HortaAgsu, and Chad Syverson. "Advertising and
competition in privatized social security: The case of Mexico.", National Bureau of
Economic Research (2013).
Heidhues, Paul, and Botond Koszegi. "Exploiting naivete about self-control in
the credit market." The American Economic Review (2010): 2279-2303.
Heitmann, M., Johnson, E. J., Herrmann, A. and Goldstein D. G. "Pricing
Add-Ons as Totals: How Changing Price Display can Influence Consumer Choice."
Working Paper (2014).
Hortacsu, Ali, and Chad Syverson. "Product differentiation, search costs, and
competition in the mutual fund industry: a case study of the S&P 500 Index Funds."
Quarterly Journal of Economics,119 (2004): 403-456.
Hsu, Joanne W., David A. Matsa, and Brian T. Melzer. "Unemployment Insurance and Consumer Credit." Unpublished Working Paper (2013).
Koszegi, Botond. "Behavioral contract theory." Journal of Economic Literature.
Forthcoming (2013).
Lohse, Gerald L. "Consumer eye movement patterns on yellow pages advertising." Journal of Advertising, 26. 1 (1997): 61-73.
Lohse, Gerald L., and Dennis L. Rosen. "Signaling quality and credibility in
yellow pages advertising: the influence of color and graphics on choice." Journal of
advertising, 30.2 (2001): 73-83.
Maskin, Eric, and John Riley. "Monopoly with incomplete information." The
RAND Journal of Economics, 15.2 (1984): 171-196.
Merton, Robert C. "Financial innovation and economic performance." Journal
of Applied Corporate Finance, 4.4 (1992): 12-22.
Miller, Merton H. "Financial innovation: The last twenty years and the next."
Journal of Financialand Quantitative Analysis, 21.4 (1986): 459-471.
Mussa, Michael, and Sherwin Rosen. "Monopoly and product quality." Journal
of Economic theory, 18.2 (1978): 301-317.
Shui, Haiyan, and Lawrence M. Ausubel. "Time inconsistency in the credit card
market." In 14th Annual Utah Winter Finance Conference (2004).
Thaler, Richard H., and Cass R. Sunstein. "Nudge: Improving decisions about
health, wealth, and happiness." Yale University Press (2008).
Tufano, Peter. "Securities innovations: a historical and functional perspective."
Journal of Applied Corporate Finance 7.4 (1995): 90-104.
99
Tufano, Peter. "Financial innovation." Handbook of the Economics of Finance
1 (2003): 307-335.
100
Figure 2-1: Monthly Time Trend of Reward
14
1.2
1
0.8
0.6
OA
0.2
0
-oe-Raward
Figure 2-1 is the plot of monthly time trend of variable "Reward". Reward equals to how many
reward programs the offer has out of Cash back, point and car rental insurance program. For each
month, we calculate the average "Reward" of the credit card offers.
101
Figure 2-2: Monthly Time Trend of Size
9
8
7
6
5
4
S
2
1
0
-v-Ske
Figure 2-2 is the plot of monthly time trend of variable "Size". Size is the maximum size of the
reward programs minus the average size of all characters in every pages of each credit card offer.
For each month, we calculate the average "Size" of the credit card offers.
102
.
.....
..
-.: .
.......
....
Figure 2-3: Monthly Time Trend of Color
0.7
0.6
0.4
OA
0.3
0.2
0.1
0
-.- Cobr
Figure 2-3 is the plot of monthly time trend of variable "Color". "Color" is the dummy of whether
reward programs in the offer use color other than black/white. For each month, we calculate the
average "Color" of the credit card offers.
103
Figure 2-4: Monthly Time Trend of Picture
0.35
0.3
0.25
0.2
0.15
0.1
--. Picture
Figure 2-4 is the plot of monthly time trend of variable "Picture". "Picture" is the file storage size
of scanned images of credit card offers. The unit is megabyte (MB). For each month, we calculate
the average "Picture" of the credit card offers.
104
Figure 2-5: Waterfall Regression Coefficients
APRs and Fees by Education
0.10
0.00
High ?ol
Some College
Post Colege
ate
-0.10
-0.20
-0.30
-040
-0.50
-0.60
-@-APR
-- +-ARPDefault
-4--LateFee
Reward Programs by Education
0.08
0.06
0.04
0.02
0.00
High School
e
College
-0.02
Post College
Graduate
-0.04
-0.06
-"-CASH
-- &-POINT ---
MILE
-*--IntroAPR
Figure 2-5 plot the estimated coefficients on the education from a regression where we regress
individual card features on dummies for different education levels (as provided by Mintel). The
regression results are reported in table 2.3. We omitted the highest education bin (graduate school)
since it is very rare and noisy.
105
Figure 2-6: Waterfall Regression Coefficients
APRs and Fees by Income
0.80
0.60
OAO
0.20
0.00
-020
-0.40
-0.60
-0.80
-*-APR
-- o-ARP_Default
-- a-LateFee
Reward Programs by Income
0.15
0.10
0.05
0.00
-0.05
-0.10
--
e-CASH
-- e-POINT
--
e-MILE
--- IntroAPR
Figure 2-6 plot the estimated coefficients on the income from a regression where we regress individual
card features on dummies for different income levels (as provided by Mintel). The regression results
are reported in table 2.3.
106
.......
....
Figure 2-7: Simple Visual Credit Card Offer
XYZ Bank Credit Card
0intro
APR
The new standardof excellence.
Dear Sir/Madam,
Congratulations! You're Pre-Approved for this Credit Card! It is time to get the card you deserve! And you
pay $0 Annual Fee.
Get Rewards on Every Purchase.
Your Credit Card rewards you more by earning:
* 2% reward points for every dollar spent on restaurants, airfare and gas
* 1% reward points on all other purchases
Points add up fast - make your purchases work for you.
Get the Privileges. Save the Fee.
With no annual fee, the value keeps coming. You will also receive a 0% introductory APR on purchases
and balance transfers for the first 12 months. Additionally, we offer a Balance Transfer Fee that is only
3% of the amount.
The Credit Card is designed for people who are smart with their money, and want to enjoy the benefits.
Once you have the card, you can:
- Get 24/7 online account access
- Ask for e-alerts to make sure you don't forget when payment is due
- Call us anytime with questions or concerns
Request your card today, and let the rewards begin.
Sincerely,
David Hughes
Senior Vice President
Figure 2-7 is the sample of simple visual credit card offers. It has relatively small font size to
emphasis the reward programs. It doesn't have many colors or flashy pictures.
107
lire 2-8: Higrh Visual Crpdit Crd Offer
Dear Sir/Madam,
You're Pre-Approvedfor a MisterGold Cardwith a Credit Line
up to $3,000.
Isn't it time you get the credit you deserve? Your credit history shows that
you're a perfect match for this card. We offer you unmatched convenience, an
exclusive rewards program, no annual fee and superb client service.
Enjoy Premium0% Intro APR for the First12 Months.
Enjoy a 0% introductory APR for 12 months on purchases and balance transfers
after your account is opened - after that, a variable APR, currently 18.99%.
That's a year of savingsl
Enjoy the Benefits of Being a MisterGoldCard Member.
Earn one point for every dollar you spend on purchases. You can redeem points
for a statement credit towards any travel purchase you have made on the Card.
It gets better. With the MisterGold card, there is no annual fee and you have
the flexibility to pay for your purchases over time.
Act Now and Get Your MisterGoldCard.
Don't miss out on this exceptional opportunity to enjoy the benefits and buying
power of your MisterGold card with a credit line up to $3,000.
We look forward to welcoming you as a new XYZ Bank member.
Sincerely,
Julia Squire
Senior Vice President
Figure 2-8 is the sample of high visual credit card offers. It has relatively big font size to emphasis
the reward programs. It also has many colors and flashy pictures to draw consumers' attention.
108
Table 2.1: Summary Statistics Data
Variable
FFR
APR
Max Card limit
APRBalance
APRCASH
DefAPR
Annualfee
Latefee
overlimitfee
IntroAPR regular
IntroAPRbalance
IntroAPRcash
Size
Color
Bold
Picture
Reward
CASH
POINT
MILE
Carrental
Purchaseprct
N
156
982767
526949
749264
942430
721393
1003977
1001221
898636
1014768
1014768
1014768
644865
644865
644865
803285
803285
803285
803285
803285
803285
803285
Mean
2.676708
12.64551
10.05231
11.33406
19.88759
26.51097
12.28505
33.8348
29.74222
0.4674211
0.4739418
0.0562621
4.709489
0.3210936
0.355845
0.229046
0.6762071
0.2105218
0.2382828
0.0878804
0.2274025
0.2340713
S.D.
2.137708
4.180521
1.366279
3.341435
4.282504
3.970656
31.99156
6.165273
10.15561
0.4989377
0.4993208
0.2304273
5.491711
0.466897
0.4787689
0.2638539
0.767116
0.4076795
0.4260333
0.283121
0.4191549
0.4234173
Min
0.0706452
0
6.214608
0
0
0
0
0
0
0
0
0
0
0
0
0.001715
0
0
0
0
0
0
Max
6.544516
79.9
15.42495
29.9
79.9
41
500
85
79
1
1
1
143.6293
1
1
4.10319
3
1
1
1
1
1
Note: FFR is the federal fund rate at monthly frequency. Other variables are based on Mintel's
credit card's direct mail campaigns from March 1999 to February 2011. Variables from "Size" to
"Purchaseprct" are from 80% of 1,014,768 total mail campaigns which have scanned images of credit
card offers. Size is the maximum size of the reward programs minus the average size of the whole
page in credit card offer. Color is the dummy of whether reward programs in the offer use color other
than black/white in the offer. Bold is the dummy of whether the offer use bold to highlight reward
programs. If there is no reward programs in the offer, we put missing value to Size, Color, and Bold.
Picture is the file size of each page of the offer which is the measurement of how many or how large
are pictures in the offer. Reward is the number of reward programs of CASH POINT and Car rental
insurance in each offer. CASH, POINT, MILE, Carrental, Purchaseprct are dummies of whether
the offer has these reward programs respectively. Intro APRregular, Intro_APRbalance and
IntroAPRcash are the dummies of whether the offer has 0% introductory APR for regular purchase, balance transfer and cash advance respectively. APR is the regular purchase APR of the
credit card offer which is the middle point if APR is a range in the offer. Card Limit is the log of
maximum credit card limit stated in the offer. Annual fee, late fee and over limit fee are fees charged
by credit card company which usually are in shumerbox.
109
Table 2.2: Descriptive Statistics for Format Design of Credit Card Offers
Penal A
Percentage of cards that
have this term
Term mentioned on 1st
page
Font size of term if mentioned on 1st page
Font size of CC term if
NOT mentioned on first
page
Font color of CC term if
mentioned on first page
Font color of CC term if
NOT mentioned on first
page
Font bold of CC term if
mentioned on first page
Font bold of CC term if
NOT mentioned on first
page
# Obs
Late fee
DefAPR
100.00%
MILE
100.00%
Over
Annual CASH POINT
limit fee fee
100.00% 100.00% 21.05% 23.83%
8.79%
Carrental Intro
APR
51.64%
22.74%
5.80%
4.97%
6.96%
79.28%
100%
93.51%
100%
80.48%
91.04%
9.49
9.28
9.80
13.24
11.16
11.47
14.12
10.27
11.27
9.57
9.63
9.50
13.76
10.62
10.80
9.91
10.04
10.62
33.98%
37.88%
27.73% 66.86%
40.13%
42.84%
47.12%
24.34%
32.28%
24.67%
26.19%
27.73% 44.35%
37.24%
38.45%
29.47%
23.31%
32.29%
38.59%
27.77%
35.07%
79.01% 47.24%
43.90%
56.34%
10.56%
53.15%
49.00%
19.59%
34.53%
53.20%
29.97%
18.08%
13.08%
39.99%
776,624
776,624
776,624 776,624 803,285 803,285
803,285
803,285
776,624
Late fee
Default
APR
28.20%
27.01%
Over
Annual
limit fee fee
27.58632 7.691491
30.11493 33.22181
36.58%
Penal B
if term is on first page
29.38247
if term is in the back 35.10307
(schumer box)
Note: The dataset is based on Mintel's credit card's direct mail campaigns from March 1999 to February 2011.
Descriptive statistics are based on 80% of 1,014,768 total mail campaigns which have scanned images of credit card
offers. Penal A is the descriptive statistics of format information of credit card terms and reward programs. In Penal
A, late fee, default APR. over limit fee and annual fee appears in 776,624 offers since we have missing pages of Schumer
box where these terms usually appear. IntroAPRs contains all introductory APR programs: regular intro APR,
balance transfer Intro APR and cash advance Intro APR. Size is the maximum size of the reward programs in credit
card offer. Color is the dummy of whether reward programs in the offer use color other than black/white in the offer.
Bold is the dummy of whether the offer use bold to highlight reward programs. Picture is the file size of each page
of the offer which is the measurement of how many or how large are pictures in the offer. Penal B is the descriptive
statistics of credit card terms when they mentioned on the first page or not. "First page" includes the envelop and
the first page letter of credit card offers.
110
Table 2.3: Credit Card Features and Demographics
Variable
FFR
Education_2
Education_3
Education_4
Education_5
Education_6
Income_2
Income_3
Income_4
Income_5
Income_6
Income_7
Income_8
Income_9
Age FE
State FE
HH Compo FE
Bank FE
Observations
R-squared
(1)
APR
(3)
(2)
CardLimit DefAPR
(4)
Late Fee
(5)
CASH
(6)
POINT
(7)
MILE
0.352***
(0.076)
-0.039
(0.029)
0.033
(0.037)
-0.067
(0.045)
0.003
(0.047)
0.043
(0.047)
-0.227***
(0.036)
-0.348***
(0.050)
-0.451***
(0.056)
-0.530***
(0.070)
-0.606***
(0.078)
-0.561***
(0.084)
-0.419***
(0.095)
-0.390***
(0.096)
0.005
(0.010)
0.095***
(0.008)
0.103***
(0.011)
0.206***
(0.013)
0.242***
(0.015)
0.003
(0.013)
0.118***
(0.011)
0.178***
(0.013)
0.228***
(0.014)
0.301***
(0.015)
0.361***
(0.016)
0.380***
(0.017)
0.405***
(0.018)
0.419***
(0.018)
0.882***
(0.096)
-0.009
(0.030)
-0.031
(0.031)
-0.026
(0.040)
-0.111***
(0.039)
-0.009
(0.047)
0.092***
(0.029)
0.149***
(0.035)
0.132***
(0.040)
0.225***
(0.048)
0.263***
(0.060)
0.356***
(0.070)
0.431***
(0.081)
0.543***
(0.087)
-0.242*
(0.133)
-0.104**
(0.041)
-0.310***
(0.051)
-0.264***
(0.055)
-0.528***
(0.078)
0.086
(0.067)
0.133*
(0.074)
0.135**
(0.066)
0.322***
(0.077)
0.431***
(0.088)
0.462***
(0.098)
0.581***
(0.110)
0.612***
(0.131)
0.556***
(0.139)
-0.012***
(0.004)
0.015***
(0.002)
0.010***
(0.002)
0.018***
(0.003)
0.011***
(0.003)
0.003
(0.004)
0.020***
(0.003)
0.026***
(0.003)
0.030***
(0.003)
0.044***
(0.004)
0.048***
(0.005)
0.048***
(0.005)
0.047***
(0.006)
0.040***
(0.006)
0.010***
(0.003)
0.008***
(0.002)
0.004
(0.003)
0.009***
(0.003)
0.004
(0.004)
0.005
(0.005)
0.013***
(0.002)
0.022***
(0.003)
0.026***
(0.003)
0.035***
(0.003)
0.042***
(0.003)
0.047***
(0.004)
0.049***
(0.005)
0.051***
(0.005)
0.008*** -0.026***
(0.002)
(0.005)
0.013*** -0.002
(0.002)
(0.002)
0.019*** -0.015***
(0.001)
(0.003)
0.046*** -0.026***
(0.003)
(0.004)
0.064*** -0.049***
(0.004)
(0.004)
0.003
-0.003
(0.003)
(0.005)
0.015*** 0.002
(0.002)
(0.004)
0.022*** -0.002
(0.002)
(0.004)
0.028*** -0.006
(0.002)
(0.005)
0.045*** -0.014***
(0.003)
(0.005)
0.059*** -0.021***
(0.003)
(0.006)
0.076*** -0.035***
(0.004)
(0.006)
0.100*** -0.051***
(0.006)
(0.008)
0.117*** -0.066***
(0.008)
(0.006)
-0.014
(0.014)
0.069***
(0.007)
0.075***
(0.008)
0.159***
(0.011)
0.190***
(0.014)
-0.003
(0.013)
0.079***
(0.010)
0.111***
(0.011)
0.137***
(0.011)
0.200***
(0.013)
0.247***
(0.014)
0.291***
(0.016)
0.336***
(0.020)
0.380***
(0.022)
Y
Y
Y
Y
942,397
0.253
Y
Y
Y
Y
496,063
0.607
Y
Y
Y
Y
713,882
0.310
Y
Y
Y
Y
961,247
0.157
Y
Y
Y
Y
777,192
0.248
Y
Y
Y
Y
777,192
0.262
Y
Y
Y
Y
777,192
0.075
Y
Y
Y
Y
629,637
0.080
(8)
(9)
IntroAPR Format
Y
Y
Y
Y
972,260
0.159
Note: OLS regressions to estimate relationship between credit card features and consumer's demographics. Data
is restricted to offers we have scanned pictures from column 5,6,7, and 9. Format is the first principal component
of reward programs' size, color, bold and the picture sizes on the credit card offers.Income_ 2 is the dummy for
households whose annual income is from 15k to 25K. Income_3 is for 25k to 35k. Income.Income_4 is for 35k to 50k.
Income_5 is for 50k to 75k. Income_6 is for 75k to 100k. Income_7 is for 100k to 150k. Inocme_8 is for 150k to
200k. Income_9 is for over 200k. Education_2 is dummy for household head whoes highest education is high school.
Education_3 is for some college. Education 4 is for graduated college. Education_5 is for post college graduate.
Standard errors in parentheses are clustered by month. Regressions are controlled by age fixed effects, household
composition fixed effects, state fixed effects and bank fixed effects.
111
Table 2.4: Relationship Between APRs/Fees and Reward Program
Panel A
Variable
FFR
POINT
(1)
APR
(2)
APR
(3)
APR
(4)
APR
(5)
DefAPR
(6)
(7)
DefAPR DefAPR
(8)
Late Fee
(9)
Late Fee
(10)
Late Fee
0.326***
(0.004)
-0.844***
(0.013)
0.314***
(0.005)
-0.671***
(0.013)
0.258***
(0.005)
-0.062***
(0.012)
0.759***
(0.005)
0.750***
(0.013)
-0.160***
(0.007)
2.063***
(0.018)
No
Yes
Yes
753,690
0.214
No
Yes
No
616,957
0.116
No
Yes
Yes
616,957
0.313
No
Yes
No
769,923
0.0226
-0.277***
(0.009)
0.869***
(0.024)
0.434***
(0.010)
No
Yes
No
769,923
0.0258
-0.133***
(0.007)
1.510***
(0.015)
No
Yes
No
753,690
0.0218
0.808***
(0.006)
1.341***
(0.025)
-0.197***
(0.007)
No
Yes
No
616,957
0.118
0.733***
(0.005)
0.429***
(0.012)
Yes
No
No
778,497
0.0282
0.373***
(0.005)
0.007
(0.023)
-0.244***
(0.007)
No
Yes
No
753,690
0.0241
(1)
APR
(2)
APR
(3)
APR
(4)
APR
(5)
DefAPR
(6)
DefAPR
(7)
DefAPR
(9)
(10)
(8)
Late Fee Late Fee Late Fee
0.312***
(0.004)
-0.528***
(0.012)
0.301***
(0.005)
-0.451***
(0.013)
0.255***
(0.005)
-0.165***
(0.012)
0.784***
(0.005)
0.002
(0.015)
-0.120***
(0.008)
1.573***
(0.019)
No
Yes
No
753,690
0.0189
No
Yes
Yes
753,690
0.214
0.846***
(0.006)
0.621***
(0.027)
-0.238***
(0.008)
No
Yes
No
616,957
0.111
0.748***
(0.005)
0.587***
(0.015)
Yes
No
No
778,497
0.0236
0.395***
(0.006)
0.472***
(0.020)
-0.363***
(0.006)
No
Yes
No
753,690
0.0242
No
Yes
Yes
616,957
0.315
No
Yes
No
769,923
0.0127
POINT*FFR
State FE
Cell FE
Bank FE
Observations
R-squared
No
Yes
Yes
769,923
0.227
Panel B
Variable
FFR
CASH
CASH*FFR
State FE
Cell FE
Bank FE
Observations
R-squared
No
Yes
No
616,957
0.109
-0.139***
(0.008)
1.381***
(0.026)
0.076***
(0.011)
No
Yes
No
769,923
0.0128
-0.109***
(0.007)
0.845***
(0.022)
No
Yes
Yes
769,923
0.221
Note: Panel A shows OLS regressions to estimate relationship between point reward programs and credit card APRs
and fees. Panel B shows OLS regressions to estimate relationship between cash back reward programs and credit card
APRs and fees. Data is restricted to offers we have scanned pictures. Regressions in column 1 is controlled by state
fixed effects. Regression 2 to 8 are controlled by demographic cell fixed effects based on states, age, income, education
and household composition. Regressions in column 3,7 and 10 are controlled by bank fixed effects. POINT is the
dummy of whether the credit card offer has point reward program or not. CASH is the dummy of whether the credit
card offer has cash back reward program or not. Standard errors in parentheses are clustered by cells.
112
Table 2.5: Mileage Program vs. Zero Introductory APR Program
Panel A
Variable
FFR
(1)
APR
(2)
APR
(3)
APR
0.314*** 0.294*** 0.241***
(0.004)
(0.005)
(0.005)
1.698***
1.938*** 1.971***
(0.018)
(0.019)
(0.020)
(4)
APR
(5)
DefAPR
(6)
DefAPR
(7)
DefAPR
(9)
(10)
(8)
Late Fee Late Fee Late Fee
0.783***
(0.005)
0.332***
(0.024)
0.742***
(0.005)
0.317***
(0.017)
No
Yes
No
616,957
0.110
0.774***
(0.005)
0.058
(0.044)
0.100***
(0.012)
No
Yes
No
616,957
0.110
No
Yes
Yes
616,957
0.312
-0.125*** -0.028*** -0.107***
(0.007)
(0.007)
(0.007)
-2.724*** 0.128
-1.507***
(0.063)
(0.090)
(0.051)
-0.992***
(0.034)
No
No
No
Yes
Yes
Yes
No
No
Yes
769,923
769,923
769,923
0.0176
0.0249
0.223
Yes
No
No
778,497
0.0350
No
Yes
No
753,690
0.0368
No
Yes
Yes
753,690
0.234
0.278***
(0.005)
1.448***
(0.028)
0.170***
(0.010)
No
Yes
No
753,690
0.0373
(1)
APR
(2)
APR
(3)
APR
(4)
APR
(5)
DefAPR
(6)
DefAPR
(10)
(9)
(8)
(7)
DefAPR Late Fee Late Fee Late Fee
0.397***
(0.004)
IntroAPR -0.950***
(0.012)
IntroAPR*FFR
0.391***
(0.005)
-0.990***
(0.012)
0.326***
(0.004)
-0.740***
(0.013)
0.891***
(0.005)
0.183***
(0.011)
No
Yes
No
942,397
0.0452
No
Yes
Yes
942,397
0.238
0.774***
(0.005)
0.058
(0.044)
0.100***
(0.012)
No
Yes
No
713,882
0.127
0.859***
(0.005)
State FE
Cell FE
Bank FE
Observations
R-squared
0.664***
(0.006)
0.560***
(0.022)
-0.566***
(0.006)
No
Yes
No
942,397
0.0613
MILE
MILE*FFR
State FE
Cell FE
Bank FE
Observations
R-squared
Panel B
Variable
FFR
Yes
No
No
982,736
0.0486
No
Yes
No
713,882
0.126
(0.011)
-0.286***
(0.007)
1.035***
(0.016)
No
Yes
Yes
713,882
0.300
No
Yes
No
961,247
0.0148
0.571***
-0.479***
(0.009)
-0.082***
(0.025)
0.411***
(0.010)
No
Yes
No
961,247
0.0185
-0.239***
(0.007)
1.105***
(0.017)
No
Yes
Yes
961,247
0.209
Note: Panel A shows OLS regressions to estimate relationship between mileage reward programs and credit card
APRs and fees. Panel B shows OLS regressions to estimate relationship between zero intro APR reward programs
reward programs and credit card APRs and fees. Data is restricted to offers we have scanned pictures in Panel A.
Panel B includes the entire credit card offer sample with and without scanned pictures. Regressions in column 1 is
controlled by state fixed effects. Regression 2 to 8 are controlled by demographic cell fixed effects based on states,
age, income, education and household composition. Regressions in column 3,7 and 10 are controlled by bank fixed
effects. MILE is the dummy of whether the credit card offer has mileage reward program or not. IntroAPR is the
dummy of whether the credit card offer has 0 intro APR for regular purchase or not. Standard errors in parentheses
are clustered by cells.
113
Table 2.6: Relationship Between APRs/Fees and Credit Card Offer Design
Variable
FFR
Format
(1)
APR
(2)
APR
(3)
APR
(4)
APR
(5)
DefAPR
(7)
(6)
DefAPR DefAPR
(10)
(9)
(8)
Late Fee Late Fee Late Fee
0.229***
(0.004)
0.374***
(0.004)
0.232***
(0.005)
0.393***
(0.005)
0.198***
(0.005)
0.426***
(0.004)
0.797***
(0.006)
0.274***
(0.005)
0.738***
(0.005)
0.208***
(0.005)
-0.087***
(0.008)
0.160***
(0.009)
-0.060***
(0.007)
0.009
(0.009)
Yes
No
No
623,476
0.0297
No
Yes
No
608,946
0.0321
No
Yes
Yes
608,946
0.177
0.198***
(0.005)
0.420***
(0.007)
0.002
(0.002)
No
Yes
No
608,946
0.177
No
Yes
No
510,788
0.131
No
Yes
No
510,788
0.310
No
Yes
No
627,119
0.00176
No
Yes
No
627,119
0.186
FormatFFR
State FE
Cell FE
Bank FE
Observations
R-squared
0.743***
(0.005)
0.333***
(0.008)
-0.048***
(0.002)
No
Yes
Yes
510,788
0.311
-0.057***
(0.007)
0.077***
(0.010)
-0.026***
(0.005)
No
Yes
Yes
627,119
0.186
Note: OLS regressions to estimate relationship between credit card offer design and credit card APRs and fees. Data
is restricted to offers we have scanned pictures and reward programs. Regressions in column 1 is controlled by state
fixed effects. Regression 2 to 8 are controlled by demographic cell fixed effects based on states, age, income, education
and household composition. Regressions in column 3,7 and 10 are controlled by bank fixed effects. Format is the first
principal component of reward programs' size, color, bold and the picture sizes on the credit card offers. Standard
errors in parentheses are clustered by cells.
114
Table 2.7: Regular APR vs. Late Fees
Variable
FFR
LateFee
(1)
APR
(2)
APR
(3)
APR
(4)
APR
(5)
APR
(6)
APR
(7)
APR
(8)
APR
0.571***
(0.015)
-0.066***
(0.001)
0.568***
(0.013)
-0.057***
(0.001)
0.459***
(0.011)
-0.016***
(0.001)
0.809***
(0.010)
-0.024***
(0.002)
1.945***
(0.056)
-0.065***
(0.002)
0.493***
(0.013)
-0.029***
(0.002)
0.424***
(0.011)
0.030***
(0.002)
0.761***
(0.011)
-0.077***
(0.002)
0.666***
(0.009)
-0.017***
(0.002)
0.837***
(0.073)
-0.059***
(0.002)
2.278***
(0.071)
-0.094***
(0.002)
-2.193***
(0.092)
0.115***
(0.003)
Yes
No
Yes
No
749,983
0.168
-0.906***
(0.086)
0.080***
(0.002)
Yes
No
Yes
Yes
749,983
0.347
Reward
LateFee*Reward
IntroAPR
LateFee*IntroAPR
MILE
LateFee*MILE
Year FE
State FE
Cell FE
Bank FE
Observations
R-squared
Yes
Yes
No
No
975,486
0.165
Yes
No
Yes
No
936,641
0.150
Yes
No
Yes
Yes
936,641
0.332
Yes
Yes
No
No
773,694
0.172
Yes
No
Yes
No
936,641
0.173
Yes
No
Yes
Yes
936,641
0.348
Note: OLS regressions to estimate relationship between regular APR and late fees in credit card offers. Data is
restricted to offers we have scanned pictures in column 2, 7 and 8. Regressions in column 1 and 2 are controlled by
state fixed effects. Regression in column 3 to 8 are controlled by demographic cell fixed effects based on states, age,
income, education and household composition. Regressions in column 5, 6 and 8 are controlled by bank fixed effects.
Reward is the number of reward programs of CASH POINT and Car rental insurance in each offer. MILE is the
dummy of whether the credit card offer has mileage reward program or not. IntroAPR is the dummy of whether
the credit card offer has 0 intro APR for regular purchase or not. All regressions are controlled by year fixed effects.
Standard errors in parentheses are clustered by cells.
115
Table 2.8: Unemployment Insurance and Credit Card Features
Panel A
Variable
(1)
APR
(2)
APRBalance
(3)
Late Fee
(4)
AnnualFee
(5)
IntroAPRAll
FFR
0.237***
UI
(0.029)
0.030
(0.047)
0.160*
(0.088)
Yes
Yes
Yes
128,778
0.265
0.169
(0.266)
0.292
0.459***
(0.104)
0.107***
(0.039)
Yes
Yes
Yes
63,134
0.190
129,344
0.237
0.005*
(0.003)
Yes
Yes
Yes
131,025
0.133
Variable
(1)
APR
(2)
APRBalance
(3)
Late Fee
(4)
AnnualFee
(5)
IntroAPRAll
FFR
0.223***
UI
(0.028)
-0.014
(0.036)
0.537***
(0.148)
0.102**
(0.043)
Yes
Yes
Yes
57,510
0.197
0.179*
(0.092)
Yes
Yes
Yes
122,439
0.258
-0.016
(0.347)
0.008***
Yes
Yes
Yes
122,965
0.245
Yes
Yes
Yes
124,616
0.132
Year FE
Cell FE
Bank FE
Observations
R-squared
Yes
Yes
Yes
127,697
Yes
Yes
Yes
Panel B
Year FE
Cell FE
Bank FE
Observations
R-squared
Yes
Yes
Yes
121,467
0.297
(0.003)
Note. OLS regressions to estimate unemployment insurance effects on credit card features. Panel A
includes the credit card offers from 1999 to 2011. Panel B is from 1999 to 2007. All regressions are
controlled by year fixed effects, bank fixed effects and cell fixed effects based on states, age, income,
education and household composition. UI is the dummy which equals 1 if unemployment insurance
increases by more than 10% in this year and 0 in the year before the increase. IntroAPRAll is
the dummy for whether the credit card offers have any of introductory APRs for regular purchase,
balance transfer or cash advance. Standard errors are clustered at the state level.
116
Chapter 3
Privatization, Politics, and
Corruption
3.1
Introduction
In the 1990s, a huge wave of privatization swept across the world, especially in such
countries as Mexico, Central Europe, the Soviet Union, and China. In 1978, SOEs in
China controlled 99% of the production. Thirty years later, in 2008, this figure had
decreased dramatically to 32%.1 There has been much debate around the reasons why
governments want to privatize state-owned enterprises (SOEs) and what the effects
of such privatization are?
On the theory side, there has been a long debate why SOEs exist. There are
three main views about SOEs: social view, agency view and political view. The
social view claims that government's object is to maximize social welfare and private
firms can't internalize the positive externalities (e.g., Stiglitz (1993)). Agency view
argues that managers of SOEs have less incentives (Tirole(1994)), because SOEs have
more objectives than private firms and need to serve multiple masters (Dixit(1997)).
Political view assumes that politicians have their personal goals which are conflicted
with social welfare maximization (Shleifer and Vishny(1994)). SOEs are captured by
politicians and pursue their self-interested objects. I provide the suggestive evidence
of political view in this paper.
I investigate the impacts of privatization on SOEs and private firms in order
to understand the influence of privatization on the economy. I use detailed firmlevel data on the Chinese manufacturing sector that include both SOEs and private
firms from 1998 to 2005 (The Chinese Industry Census Dataset (CIC)). This dataset
contains detailed financial information as well as other firms' variables such as their
registration type, percentage state ownership, number of workers, average wages and
so on. That allows me to study both partial privatization (selling part of the assets
to private owners) and full privatization (changing registration types).
'Thirty Years of Phenomenal Achievements, National Bureau of Statistics, China Statistics Press,
2008.
117
It is well known that decisions regarding privatization are not random. 2 Government
usually picks some SOEs for privatization in preference to others based on firms' financial characteristics or other reasons. To deal with this selection bias, I use political
turnover timing as an instrument variable for privatization. Politics has a big influence on government-owned firms. Shleifer and Vishny (1994) argue that politicians
capture SOEs to derive their own benefit from them. Dinc and Gupta (2011) find
evidence that government's decision to privatization in India also depends on political
costs and benefits.
To investigate the role of politicians in China, I hand-collected every city mayor's
profile from 1990 to 2011. Combined with CIC firm data, I find that politicians delay
privatization during turnover years. The number of privatized SOEs and the amount
of privatized assets are significantly lower when a city changes mayors. This evidence
is consistent with the argument that negative effects of privatization, such as layoffs,
may cost government the political supports of its people3 . It is a similar story in
China. When new mayors take over the city, they want to build their reputation and
gain support from citizens or their superiors. Doing a lot of privatizations may hurt
them. Another story could be that mayors have greater incentives to privatize SOEs
before turnover since they can usually gain extra rents from privatization. There
are many cases that state-owned assets were sold at significantly lower prices and
politicians got their cuts from the deal. Based on these instances, I use political
turnover timing as an instrumental variable for privatization.
To verify this instrumental variable's quality, first I use the Cox proportional
hazard model to study the turnover of mayors. The only variable that shows a
significant effect on turnover is age. In China, the retirement age is 60 for males and
55 for females. Moreover, age is a constraint on promotion. For example, city mayors
are usually between 40 and 50 years old. Those who are too young or too old cannot
be promoted as city mayors. This mitigates the concern that provincial politicians
might select younger/older mayors in cities where they want less/more privatizations.
More importantly, cities' economic performance, such as average ROA and total net
profits from firms, don't have significant effects on the hazard ratio either. It turns
out that the length that a mayor has been in a city has strong predictive powers on
turnover. I also use the probit model to double check these results and find that
turnover again is not dependent on economic conditions. This result can mitigate
the concern that the timing of turnover might be endogenous, since politicians are
assigned by government instead of voted for in China. Second, for the exclusion
condition, turnover should affect SOEs' performance only through privatization. To
examine this, I pick a sub-sample of SOEs who are 100% state-owned during my
entire sample period and estimate whether turnover has effects on these SOEs. The
results show that turnover doesn't have significant effects on 100% state-owned SOEs.
To study privatization's effect on SOEs, I use both actual political turnover and
years the mayors have been in the cities to instrument privatizations and perform
2Gupta,
Ham and Svejnar (2008) argues that more profitable firms got privatized first in Czech.
Government's decision of privatization is endogenous and there is selection bias of using privatization
directly to study its effects.
3Lindbeck and Weibull(1987), Dixit and Londregan (1996)
118
2SLS. For privatization's effects on private firms, I employ difference-in-differences
identification by estimating the difference between the industry in a certain province
that has privatized SOEs and the ones that have not. I found that privatization
can increase SOEs' efficiency by 50% and increase private firms' efficiency by 100%.
Moreover, SOEs cut 64% of their workers after being privatized. Private firms increase
16.3% employment if there are privatizations around. Overall, every 100 workers got
fired by SOEs come with an 169 increase in private sector hiring in the same industry
and same province.
I also look at the politicians' individual effects on firm performance and other
activities by estimating politicians' fixed effect. In my dataset, 60% of the politicians
have been mayors in more than two cities. I am able to catch this part of the variations
and find that politicians have a large influence on SOEs under their jurisdiction. But
they have less influence on SOEs that have higher political hierarchies (provincial
level and central government level).
Moreover, to study the heterogeneity of politicians' individual effects further, I
identify "bad" politicians if they have been in jail because of corruption. The results
show that SOEs performed worse if there were corrupt politicians in the city. Also,
SOEs under "bad" politicians tend to have more market power, an effect that is not
significant for private firms. This evidence can be explained best by the political view
of SOEs. 4
Overall, this paper documents the evidence of privatization's effect on China's
economy. The results also support the political view of SOEs and suggest that politicians are motivated by self-interested benefits and capture SOEs to achieve their own
goals.
The paper is organized as follows. The next section presents the literature review
and history of economic reform in China. Section 3.3 describes the data. Section 3.4
presents the empirical methodology, and Section 3.5 shows the results. Section 3.6
examines politicians' individual effects and corruption. Section 3.7 concludes.
3.2
3.2.1
Literature review and economic reform in China
Literature review
There are three main views about SOEs: social view, agency view and political view.
The social view claims that government's object is to maximize social welfare. Private
firms have different goals: maximize their own profits which sometimes has negative
externalities and we call them market failures. Government can cure market failures
by intervening in the case of negative externalities such as pollution, regulating natural
monopolies and providing public goods. One way of doing it is to set up SOEs. For
example, Stiglitz (1993) argues that private banks may not give money to high social
returns projects and we need state owned banks to do it instead. Sappington and
4Shleilfer and Vishny (1994) argues
that politicians' are self-interested and they are able to
capture SOEs to achieve their personal goals. And the only way to fix it is to fully privatize SOEs.
119
Stiglitz (1987) argues that SOEs can reduce government intervention transaction cost
to do social optimal projects.
Agency view is based on the same assumption as social view that government
wants to maximize social welfare. However, managers of SOEs have less incentives
(Tirole (1994)), because SOEs have more objectives than private firms and need
to serve multiple masters (Dixit(1997)). Moreover, there is no owner with strong
incentives to monitor managers because they bear all the monitoring costs but only
benefit a fraction (Alchian(1965)). Vickers and Yarrow(1988, 1991) also point out
that asymmetric information can give managers bargaining power to oversees. Under
this view, SOEs intend to do socially beneficial projects but managers would put
less efforts. Government needs to strike the balance between social welfare gains and
internal inefficiencies.
Political view assumes that politicians have their personal goals which are conflicted with social welfare maximization (Shleifer and Vishny (1994)). SOEs are captured by politicians and pursue their self-interested objects such as political supports
from people, direct benefits to friends or even putting money into their own pockets.
In return, politicians give privileges to SOEs via government subsidy, tax treatment
and market power. Under this view, politicians use tax payers money to capture SOEs
and force them to achieve their own goals. Shleifer and Vishny (1994) and Boycko,
Shleifer , and Vishny (1996b) argue that full privatization is the only sure way to
solve this political inefficiency because privatizations increases the intervention costs
to politicians.
Social and agency view share the same key assumption that government's goal
is to maximize the social welfare. Political view assumes that politicians are selfinterested. Under all these three views, privatization would increase the SOEs' efficiency. However, the first two views predict that private firms around privatized SOEs
will suffer since after privatization, SOEs' goal become the same as private firms(e.g.
stop provide public goods) and there are more competition in the market. Under
political view, privatization of SOEs will benefit private firms because politicians will
"bribe" SOEs less(e.g. less protection from competition) and give more opportunities
to private firms.
On empirical side, efficiency improvement during privatization is well documented.
La Porta and Les-de-Silanes (1999) study Mexico privatizations between 1983 and
1991 which suggests that privatizations have an average of 54 percent gain in output,
half worker being laid off with increased wage, and productivity surge by 64 percent
after privatization. Frydman, Gary, Hessel and Rapaczynski (1999) consider the similar objective on central Europe countries' privatization tide around 1994. They argue
that privatized SOEs perform better than before, but the performance improvement
is only significant on the revenue measure for the firms sold to outside owners. No
obvious sign of productivity jump if SOEs were privatized to inside owners, for example, the former managers. There are many recent empirical works have been done
on political view of SOEs. Dinc and Gupta (2011) look at Indian political election
voting shares of different parties to define the political competition and use it as IV of
privatization. The evidence suggest that government delays privatization where the
political party faces more competition and privatizations improve SOEs' efficiency.
120
To my knowledge, it is probably the most similar work with this paper.
3.2.2
History of Chinese SEOs' reform
Economic reform and the privatization of SOEs in China started in 1978. This reform
and opening-up policy was initiated by Xiaoping Deng, the successor of Chairman
Mao Zedong. There were three main stages of privatization. The first period was
from 1978 to 1984, when SOEs started to pay tax instead of giving profits directly to
the government. This was not successful because the tax rate was too high (around
55%) and many SOEs were unable to pay. The second period was from 1986 to 1993,
when SOEs could be contracted out, which gave the managers and contractors more
incentives. However, since ownership did not change and the regulations were not
good, contractors extracted rents as much as possible, which harmed state owned
assets. The third stage was from 1993 to 2005. During this stage, SOEs have sold
to other investors, such as institution investors, private investors and so on. In this
paper, I focus only on the third stage because it is true "privatization" and is also
the most influential reform among SOEs. After 2005, State-owned Assets Supervision
and Administration Commission put more strict rules of privatization which lowered
down the privatization process significantly.
A competitive market is one of the main forces that has accelerated the privatization of Chinese SOEs. Yao (2005) uses survey data of 800 SOEs from 1995 to 2001 to
show that insolvency was the big problem among many SOEs, and government and
state banks did not want to support these firms anymore, so privatization became the
most efficient way out. By 1988, more than 40,000 SOEs had been privatized or reorganized. Yao's study shows that the new reform between 1997 and 1998 improved
SOEs' efficiency a lot.
In China, SOEs, especially the big ones, have many privileges and resources from
their political connections. For example, SOEs control a large variety of industries
including oil, electricity, telecommunications, finance, and so on. The big names
among China's SOEs include China National Petroleum Corporation, China Mobile,
and China Telecom. Moreover, banks are more willing to loan money to SOEs because
of their state backing. In a few cases, the state has allowed SOEs to go bankrupt
and has bailed out some that have been in trouble. Many top managers of these big
SOEs end up being government officials at either central or state government level.
More interestingly, there is a party committee in each SOE that exerts considerable
influence over a firm's daily activities and big decisions. Accordingly, Chinese SOEs
are very closely connected with politicians and government. SOEs also comprise a big
part of the Chinese economy. The CIC dataset reveals that in 1998 SOEs controlled
42% of assets in the economy. This had dropped to 18% by 2005.
121
3.3
3.3.1
Data Description
Data
The data for this paper come from two sources: (1) The Chinese Industry Census
(CIC) from 1998 to 2005 , (2) the Zechen Database and (3) the Baidu Encyclopaedia
Database.
The CIC dataset, which is collected by the Chinese National Bureau of Statistics
(NBS), includes any manufacturing firm in China with annual sales of more than
RMB 5 million (around US$700,000). It has detailed annual accounting data and firm
characteristics such as the number of workers, industry category, location, registration
type, political hierarchy, government subsidy, wages, and so on. In total, there are
about 380,000 firms comprising about 40% of industrial output in China. To my
knowledge, CIC is the most detailed database on Chinese manufacturing firms and
the content and quality are very good. One potential concern is that SOEs might over
report their performance to impress the government. SOEs' managers' promotions
depend on firms' economic performance. This could lead to a downward bias of
privatization's effects, and the real effects should be more positive.
Using firm registration type from the CIC dataset, I can classify a firm as an SOE
or a private firm. Moreover, I also have the shareholders' data, which has enabled me
to calculate the percentage share-holdings by the state and other investors. This is
the main measurement used this paper since many privatizations in China are partial,
which means the state sells part of the SOEs to private firms, and it usually takes
several years to fully privatize a SOE. The location data in the CIC dataset is an
11-digit number that can locate the firm at street level. I cut the first four digits to
identify the city and use it to match the politician dataset.
My second data source is the Zechen Database, which provides the names of all
the mayors and secretaries of municipal committee in each city from 1949 to 2011. I
manually collected the list of mayors' and secretaries' names as well as their terms
of office on a monthly basis. I also collected data for members of the provincial
committees of the Communist Party of China. In sum, this dataset covers all 334
cities and 31 provinces in China.
Based on the lists constructed from the Zechen Database, I also hand collect
these politicians' profiles from the Baidu Encyclopaedia database, a Chinese language
collaborative Web-based encyclopaedia provided by the Chinese search engine Baidu.
Baidu Encyclopaedia is the number 1 online Chinese encyclopaedia and generally
provides very good profiles of famous people (better than the public official profile
for politicians). However, the quality of politicians' profile still varies among different
cities, especially small ones. In order to fix this, I also use the Xinhua News Website as
a supplementary source to cross check the data from the Baidu Encyclopaedia. Xinhua
News is the official press agency of the People's Republic of China and the biggest
center for collecting information and press conferences in China. It is also the largest
news agency in China, and has complied profile files for some politicians. The final
5
This dataset has been used in many other papers such as Chang and Klenow(2009), Huang, Jin
and Qian(2012), Dougherty and Herd (2005), Geng (2006)
122
profile dataset includes 2,227 city-level politicians (both city mayors and secretaries)
and 574 provincial-level politicians. Among them, I am able to get about 1,674 city
level politicians' profiles. Each profile has the politician's gender, age, education, and
place of birth. For education, I have used a dummy variable of whether the politician
has a college degree (or a higher qualification) or not. Place of birth is a four-digit
city ID. Moreover, we have some politicians with the same name (very common in
China). In order to fix this, I double-checked the politicians who have the same name
and distinguished by their different ID numbers.
3.3.2
Matching Politicians to Firms
The CIC data include names and addresses of all firms. I use the first four-digit city
code to identify the firm's location. For the politicians' data, I also have the same
four-digit city code for each one. Based on this location information, I merge the CIC
dataset with the politicians' dataset at city level. As a result, 95% of the CIC data
matches up with the politicians' dataset. Table 3.1 presents the summary statistics
of key variables in these datasets. In China, there are several political hierarchy
levels headed by the central government in Beijing. Under the central government
are 31 provinces and four municipalities: Beijing, Tianjing, Shanghai, and Chongqing.
Beneath those are about 300 cities, which are what this paper looks at. There are also
thousands of counties in China controlled by cities. Since the firms in the CIC dataset
are the biggest in China, many are located in cities. In case where they are located
in small counties, they still fall under the jurisdiction of the city. Moreover, the CIC
dataset also contains information about firms' political hierarchy level. In China,
each firm needs to answer to a certain level of government. From this information, I
can determine a politician's influence on a firm. For example, firms whose hierarchy
is at central government level need to listen to the central government and may not
need to answer to the city government where they are actually located.
3.3.3
Summary Statistics
Table 3.1 presents summary statistics for variables of interest for the CIC dataset
and the data relating to politicians. To analyze privatization's effects on SOEs and
its externalities on private firms, I mainly use percentage of private share to measure
privatization. In China, most of the SOEs' privatizations are partial, which means
government usually sells part of the assets to private owners. I also use the changes
in firms' registration types to construct the dummy to capture whether the firm was
transferred from a SOE to private ownership. The top panel of Figure 3-2 plot the
number of SOEs' privatization from 1999 to 2005 in China. It shows that, generally,
privatization increased over time except during 2005. Moreover, 2002 was a "quiet"
year and there were only about 65% volume compared with 2003 and 2004.
On the political side, the top panel of Figure 3-2 is the time trend of political
turnover. There are two spikes in 1998 and 2002/2003. This is mainly because in
China, we have the National Congress of the Communist Party of China (CPC)
every five years and there was one in September 1997 and another in November 2002.
123
Most of the politicians' turnover, such as transfer to another city with the same level
of promotion and retirement, would have been accomplished during these periods.
City mayors are usually replaced every five years, which is the same cycle served as
members of the Congress of the Communist Party. In turnover, politicians would
either get promoted, be transferred to another city, or retire. There is a very clear
cycle of political turnover in China.
3.4
3.4.1
Empirical Analysis
Privatization's effect on SOE itself
First, I want to study the effects of privatization on SOEs. In this analysis, I use both
percentage private share of SOEs and the registration type change(from state owned
to private owned) as independent variables to measure privatization. For dependent
variables, I use ROA, OROA, sales per worker(labor productivity) to measure the
firm's efficiency. I also study privatization's effect on firm's market share, government
subsidy, average workers' wage and number of workers. The basic specification I use
to test is based on panel data:
Yt= a
+
at + aj +,31
- private_sharejt+ y1 - Xi,t + Eit
(3.1)
where Yi,t is one of the measures of efficiency or other independent variables I discussed
above. a, is the firm fixed effect (84,661 categories), at is the year fixed effect (8
categories) and aj is the industry fixed effect (40 categories). In total, I have 84,706
dummies to control these fixed effects. The reason I add industry fixed effect is that
some firms changed industry in the sample and it is not collinear with firm fixed
effect. private_sharei,t is the percentage share of private owners of firm i at time
t. #1 in (1) is the coefficient of interest that captures privatization's effect. Xjt are
controls for firms' characteristics such as total sales, total assets and so on.
Moreover, I also use the registration type to create the dummy private which
equals 1 if the SOE is registered as a private firm and 0 otherwise. It is a before/after
privatization dummy. The regression becomes:
Yi,t= ai + at + a + #1 -privatej,t + 7y - Xi,t + Eit
(3.2)
Equation 3.1 studies efffects of 1% increase of SOE's private share while 3.2 estimates
the effects of changing registration type from SOE to private firm. On average,
private share will increase around 40% when SOE changes its registration type. Two
specifications above should give the similar results on privatization's effect because
most of the state-owned shares are sold when change registration type.
There are several concerns about this basic identification. First, as I mentioned
before, the decision of privatization is endogenous. Gupta, Ham, and Svejnar(2008)
looks at SOEs' privatization in Czech and find evidence that government chose to
privatize more profitable SOEs first. Including firm fixed effect can alleviate the
problem since it controls firms' time invariant part of selection bias. However, the
124
government may determine the privatization on time varying variables such as firms'
economic performance, market share and so on. Second, their might be an omitted
variable bias if there is any missing depended variables which are correlated with both
privatization and firm's performance.
Based on these concerns, I use political turnover timing as instrumental variable to
privatization. Dinc and Gupta(2011) looked at Indian political election voting shares
of different parties to define the political competition and use it as IV of privatization.
The main idea behind it is that politicians may lose votes if they privatize SOEs which
would come with lay-offs and temporary unemployment. In more competitive area,
the government are more reluctant to do privatizations. In China, there is no voting
system and I use the political turnover as IV instead. In Figure 3-1 Panel C and
D , privatization decreased a lot in 2002 and in Panel E, it is clear that political
turnover mainly happens during National Congress of CPC (1998 and 2002). In
aggregate level, during turnover years, there were less privatizations. One story can
be that the government doesn't want to have many changes and negative side-effects
of privatizations during these "sensitive" times and new mayors want to build good
reputations and gain support from citizens. Another possibility is that during these
turnover periods, politicians are busy with other things such as their promotions
and moves. SOEs' privatizations may be not on top of their lists. About exogenous
condition of this IV, National Congress of CPC happens exactly every 5 years and the
tenure of city mayor is between 3 to 5 years. In another word, mayors are "forced"
to change cities every 5 years. In order to test the exogeneity of political turnover,
I use the Cox proportional hazard model since it incorporates both turnover of the
politicians' demographic and the time of turnover. The hazard rate of turnover is
given by:
h(t) = ho(t) exp( 1 zxi + 2 x 2 + - - + #Xik),
(3.3)
where Xi ... Xk are politician's demographic (age, gender, education and place of birth)
and average ROA and total net profits from the firms under the mayors' jurisdiction.
I have both time-varying and time-invariant variables and I assume that time-varing
variables are constant during one year. I followed Wooldridge(2001) to do this proportional hazard model and report the coefficients as well as hazard ratios from the
estimation. I included the economic performance variables: average ROA and total
net profits to test whether the turnovers are determined by politicians' performance.
Li and Zhou(2005) found that the likelihood of promotion of provincial leaders increases with their economic performance in China between 1979 and 1995. Moreover,
it is well known that the city mayors' promotion in China are heavily depend on
local GDP. The variation I am using here is from the timing of turnover rather than
the turnover's type which can be promotions, transfer to another city or retirement.
However, it is still a concern that economic performance can also affect the timing
of the turnover. Using (3), I can test whether it is the case or not. I use both
contemporaneous average ROA and total profits as well as lagged average ROA and
total profits to make sure that past economic performance are also counted here. In
section 6, I will also use probit model to study the politician's promotion, retirement
and corruption.
125
After the verification of IV, I use 2SLS to perform the IV regressions. I use
how many years have the politicians(mayors and party leaders) stayed at the city
as IV rather than the actual turnover. Although, from hazard ratio analysis and
probit analysis, it shows that turnover doesn't depend on politicians' demographic
and the economic performance, it is still hard believe that political turnover is totally
exogenous. Moreover, the length that the politicians have stayed has significant
predictability of turnover and it is more exogenous. The first stage regression is:
private_shareit= ai + at
+ a j + F1 -Politician1lt+F2 - Politician2,t+ Y -Xi,t + Ei,t
(3.4)
where F, is the coefficient vector of Politicianl,t which are the dummies of the years
the mayor of the city I stayed at time t: D11 ,t, D2,,,...,D9,, since the longest year is 9
in my sample. And F2 is the coefficient vector of Politician2,t which are for secretary
of municipal committee. It also includes 9 dummies as mayors. The second stage is:
Yt = ai + at + a, + 1
private sharei,t + 'y - Xi,t + Ei,t
(3.5)
Then, I also use the actual turnover in the first stage regressions first to double
check the results.
private_sharei,t= ai + at + a, + A - TurnoveriT, +
Yi,t = ai + at + a +)31
'Y1 - Xi,t
+
Ei,t
- private sharei,t + yi - Xi,t + Ei,t
(3.6)
(3.7)
where Turnoverl,t is how many turnovers in city 1 at year t. In order to test the
weak instrument variable problem, I also calculate the F-stats of IVs in first stage
regression.
3.4.2
Privatization's effect on private firms
After study privatization's effects on SOEs which has a lot of well documented evidence, it is more interesting to look at the effects on private firms surround privatized
SOEs. It would give us the whole picture of the effects from privatizations on entire
economy. I estimate this effect by looking at the differences between the industry in a
certain province who has privatized SOEs and the ones who don't have. The hypothesis is that SOEs have a lot of privileges and government protections over private
firms. Privatized SOEs suppose to lose these benefits and the private firms would
gain from it because they can compete with SOEs more fairly. Many Chinese SOEs
have artificial monopoly powers. Here, I employ the difference in difference model:
Yi,t = ai + at + a, + ak + 01 - privatizedSOEj,k,t +-1 - Xi,t +
Ei,t
(3.8)
where Yi,t is as the same as in equation 3.1 which is the efficiency measures and other
independent variables of private firms. Again, ai is the firm fixed effect , at is the
year fixed effect, a, is the industry fixed effect and ak is the province fixed effect.
privatizedSOEj,k,t is the dummy whether there was any privatization in province k
and industry j at time t. Xi,t are controls for firms' characteristics. The assumption
126
of this DID specification is that the trends after taking out these fixed effects are
the same for all private firms. In another word, all the other effects are level effects.
Under this assumption, I can compare the private firms who have privatized SOEs in
the same province and industry with private firms without any privatized SOEs.
One concern is that the assumption of same trend may be not valid. I add 2 year
and 3 year pre-trend dummies privatizedSOEj,k,t+rin equation 3.8. Moreover, I also
analyse the long-term effects of privatized SOEs on private firms by adding lagged
dummies(post trend) privatizedSOEj,k,et-. T = 2, 3. the specification becomes:
yi,t
~
+ Ct + a3 + ak + 01 -privatizedSOEj,k,t +32
+3
privatizedSOEj,k,t+ + 7h - Xi,t + Ei,t
i
- privatizedSOE,k,,
(3.9)
All the regressions above only show the correlations between the SOE's privatizations
and the private firms surround them. In order to estimate the causality between
them, I use political turnover again to instrument privatization. The 2SLS are:
1
privatizedSOEj,k,t - ai+at+aj+ak+Fl.Politician
p,t +F 2 .Politician2p,t+-i1-Xit+Ei
(3.10)
= Cti
ai,t
+t + 'j +
&k
+
s1 - privatizedSOEj,k,t +
Y1
- Xi,t +
Eit
(3.11)
where F 1 is the coefficient vector of Politicianlp,t which are the dummies of the
average years the city mayors have stayed in cities under province p at time t: D1 ,t,
1
D21,t,...,D9,,t since the longest year is 9 in my sample. And F 2 is the coefficient
vector of Politician2pt which are for secretary of municipal committee. It includes 7
dummies because the longest average tenure of secretaries of municipal committee in
a province is 7 years.
The difference in difference specifications above don't consider the magnitude of
privatization. Bigger privatizations suppose to have bigger impacts on private firms.
To verify it, I use privatized Assets to replace dummy of whether there is privatized
SOE or not in (8). And I use the sample condition on existence of privatization. The
specification becomes:
,t
ai + at - aj + ak + /1 - privatizedAssetsj,k,t + 71 - Xi,t + Ei,t
(3.12)
where privatizedAssetsj,k,t is the log of total asset got privatized in province k and
industry j at time t. And the sample size is conditioned on the province and industry who has privatization happened. It can give us the sense whether bigger firms'
privatization has bigger effects to further test my hypothesis.
To further verify my story that private firms gain from privatizations because they
can compete more fairly with privatized SOEs, I use number of firms in each province
and industry to measure the competitiveness. The hypothesis is that privatization's
effects on private firms should be smaller in more competitive market. It is because
that if the market is already very competitive, one more privatization should have less
effect. I interact privatizedSOEj,k,t with FirmNumberjk,t-1 which is the number of
127
firms in industry
Yi,
=
j province k at time t - 1.
ai + at + a3 + ak +
+32
/1
- privatizedSOEj,k,t
privatizedSOEj,k,t * FirmNumber,k,t_1 + Y1 - Xi,t -+Ei,t (3.13)
The prediction should be that more firms in a certain industry and province (more
competitions) should lead to less impact of privatization.
3.5
3.5.1
Results
Privatization's effects for SOEs
Before doing 2SLS regressions, it would be good to see the validity of instrument
variable. Table 3.2 shows the results of estimating equation 3.3. The results for Cox
proportional hazard model are in column (1) and (2). In column (1), except age has
significantly positive effects on hazard ratio(turnover), other things don't have effects
including average ROA and total profit from the firms under this politicians' jurisdiction. Column (2) uses lagged ROA and total profit. Again, there is no significant
results for them. Column (3) and (4) use Weibull distribution to double check the
results and there is no significant coefficients at all. Two hazard ratio specifications
have pretty consistent results. I also plot the hazard functions in Figure 3-2 and it is
clear that probability of turnover increases when politicians stay longer.
Table 3.3 panel A shows the results of estimating equation 3.1 for all efficiency
measures and other key variables. It is OLS regression and the standard errors are
clustered by firm.
Column (1) to (3) presents evidence for efficiency gain of SOEs via privatization. ROA and OROA significantly increase 0.3% when the SOEs are fully privatized(private share changes from 0% to 100%). Average SOEs' ROA is around 5%
so that SOEs' gains efficiency by 6% after privatization. At the same time, sales per
worker increase by 4.1% after privatization. Moreover, the coefficient of privatized
share on market share is significantly positive because that privatized SOEs become
more competitive. Column (6) and (7) analyse the privatization's effects on employment. Both number of workers and wages increase after privatization by 1.5% and
1.1% after full privatization. For government subsidy, I don't find any significant result in this specification. In Panel B, I use registration type change from state-owned
to private-owned to measure privatization. Use before/after dummy Privatize, I get
the similar results as in Panel A except OROA which doesn't have significant effect.
Table 3.4 panel A shows the results from 2SLS for equation 3.4 and 3.5. First stage
is strong, F-stats for excluded instruments(dummies for politicians' tenure) is 28.14
and privatizations are less when politicians stayed longer time which predicts higher
turnover probability. In the second stage, from column (1) to (3), coefficients are still
positive and larger than panel A. ROA significantly increases 2.3% when the SOEs
are fully privatized which is about 50% efficiency gain. For subsidy in column (5),
coefficient is now significantly negative and SOEs' probability of getting government
128
subsidy decreases about 10% after privatization. For employment in column (6) and
(7), the coefficients are all negative now. It makes more sense because one of the
SOEs social object is to increase employment and usually over pay workers in China.
In Table 3.4 panel B, I use actual turnover as IV. First stage is very strong and
there are less privatizations when the city is under political turnover. Coefficient of
turnover is -0.007. The results on seven variables are consistent with Panel A.
In sum, privatizations increases SOEs profitability and efficiency and government
can subsidize them less.
3.5.2
Privatization's effect on private firms
In order to understand the total effects of SOEs, we turned to private sectors here.
Table 3.5 panel A are the results of estimating equation 3.8. Column (1) indicates
that ROA of private firms increase significantly when there are SOE privatizations in
the same industry same province. The magnitude is 0.1% increase of ROA which is
about 2% efficiency gain. From column (3), labour productivity also increase by 0.7%.
For employment part, private firms will hire 0.5% more workers and decrease wages
by 0.4%. Combined with the results from Table 3.3, it seems that workers who were
lay-off by SOEs transfer to private firms instead. Bigger supply of labour decreases
the wages. In panel B, by adding 2 year or 3 year pre-trend dummies and lagged
dummies, private firms who have privatized SOEs around them did slightly worse
before privatizations. The coefficient for 2 year pre-trend at column (1) to (3) are
insignificant. In column (4), there is a slightly negative pre-trend for ROA 3 years before. It could be that these private firms were doing worse because of the SOEs in the
neighbour. For example, SOEs have priory or unique access to the market, funding
and government projects. For the lagged dummies, effects seems to last during couple
of years after privatizations. Table 3.6 panel A are the results from equation 3.12 to
check whether bigger privatizations have bigger effects. All the samples in panel A
are private firms who have privatized SOEs around them. For measures of profitability, Column (1) and (3) have significantly positive coefficients. It means that bigger
privatizations increase private firms' efficiency more. And bigger privatizations make
private firms to hire more workers. These results has the same direction as Table 3.5
panel A and it verifies the results in panel A as well. To further verify my story here
that private firms gain from privatizations because that they can compete with privatized SOEs more fairly, I put interaction term between dummy privatizedSOE and
number of firms in this industry and province. In Table 3.6 panel B, privatizedSOE
still have significantly positive effect on productivity measurement. The coefficient of
privatizedSOE * FirmNumber is significantly negative. It means that when there
were more firms in this industry(more competitions), privatization has less effect because private firms can gain less from the market that was already competitive.
Table 3.7 are the results from 2SLS in equation 3.10 and 3.11. Again, first stage
is very strong with F-stats 124.3. In the second stage, ROA of private firms will
increase 6.3% if there are SOEs in the same province same industry get privatized.
The magnitude is big which double the efficiency. From Column (2) to (4), all other
129
profitability measures have big and significantly positive coefficients. For subsidy,
coefficient is also significantly positive and bigger than OLS regression. Number
of workers increases as well. The only different thing here is wages and benefits.
Column(6) tells that wage significantly increases for private firm after SOEs' privatization.
Moreover, I calculated the overall effects of the privatization on employment. Based
on the coefficients I estimate in Table 3.4 and 3.7, SOEs cut 64% of their workers after
being privatized. Private firms increase 16.3% employment if there are privatizations
around. Overall, every 100 workers got fired by SOEs come with an 169 increase in
private sector hiring in the same industry and same province.
The privatization may also affect the firms in the related industries such as downstream industries. In Chapter one, I find that the government credit from CDB has
positive spillover effects on downstream industries. Again, I use the input-output matrix to define the industry supply chain. However, I don't find any significant effects
of privatizations on downstream industries. The reason could be that privatizations
in China from 1998 to 2005 were mainly focus on the manufacturing industry such as
textile. State still controls the strategic industries such as energy and mining sector
which are usually at the upstream level in the supply chain. In the CIC data, only
2.7% of the privatizations are in the upstream industries and 97.3% of the privatizations are in the downstream industries. This could explain why I couldn't find
significant impacts of privatizations on downstream industries.
Based on these results, private firms benefit from SOEs privatizations as well and
hire the workers lay-offed by privatized SOEs. It could be the case that private firms
now can compete with SOEs on more fair basis. Politicians stop or do less "bribe"
to privatized SOEs because it is more costly to do so. However, to direct test this
channel, the evidence above is not enough. In the following sector, I will estimate
politicians' effects on both private and state owned firms.
3.5.3
Examination of exclusion condition and robustness check
To examine the exclusion condition of IV, I selected the SOEs who are 100% owned
by state during the entire sample period(1998-2005). The hypothesis is that the
only channel for political turnover to affect firms is through privatization. For SOEs
who have never been privatized should not be affected by political turnover. There
are 2,711 SOEs in this sub sample and I simply regress actual turnover on firms'
performance measurement and employment and control firm fixed effect, year fixed
effect, industry fixed effect and other firms' characteristics. Table 3.8 Panel A are
the regression results. There is no significant effect from political turnover to 100%
owned SOEs. This evidence also support the exclusion assumption of the instrumental
variable.
People may argue that government may privatize SOEs based on their industry
performance or prospectives. To check the robustness of the results in Table 3.3,
3.4 and 3.7, I re-run the regressions in Table 3.3 Panel A, Table 3.4 Panel A and
130
Table 3.7, adding industry trend control(interaction between industry dummies and
year dummies). Table 3.8 panel B and Table 3.9 are the results and it is almost the
same as Table 3.3, 3.4 and 3.7. It means that privatization's effects are not driven by
industry specific trends.
3.6
Influence of Politician in China
A natural question followed is whether politicians have real influence on firms? If yes,
How do politicians put influence? Bertrand and Schoar (2003) studies the managers
fixed effects matter for many firms' operations and performance. Here, I use the
politicians fixed effects to study how much do individual politicians matter for firms
performance. Then, I also classify politicians' "type" into two groups: Good one and
Bad one. I use politicians' profile data which contains whether the politician has been
to jail or not. I pick the politicians who went to jail because of the corruptions such as
taking bribe and I mark them as "Bad guys". From my database, 95% of politicians
who went to jail are due to corruption. It is also the most common reason to charge
politicians in China. I can then estimate the bad politicians' effect on economy.
3.6.1
Policitian's fixed effects
In order to estimate politicians' individual effects on the firms, I controlled firm fixed
effect, time fixed effect, industry fixed effect and city fixed effect. I also controlled
firms' other characteristics. By controlling these variations, I can estimate whether
politicians can have their own effects on the firms. The variation of politicians' fixed
effect comes from that there are 60% politicians in the sample have been to different
cities as mayors or party leaders at different times. In provincial level, 30% in the
sample have been to different provinces. Firm fixed effect and city or province fixed
effect can't capture this part of variation. Moreover, the CIC dataset has each firm's
political hierarchy and the hypothesis is that politicians should have more effects on
the firms under their jurisdiction and less effect on the firms above their levels even
these firms are located in their city or province. Then, I repeat the same specification
in equation 3.14 on different political levels of the firm. The specification is:
yi,t = ai + at + a3 + ah + -Y1 - Xi,t +
Apoliticians
+ Ei't
(3.14)
where Yi,t again stands for one of the firm economic variables, a, is the firm fixed
effect , at is the year fixed effect, a3 is the industry fixed effec and ah is the city fixed
effect. Apoliticians are fixed effects for 1,116 politicians and it is a group of dummies of
whether the firm is under the each politician's jurisdiction or not. It is the same for
provincial level politicians and I have 171 of them.
The results are in Table 3.10. It is clear that generally politicians' fixed effects
are significant. For provincial level SOEs, city level politicians don't have effect
on firms' performance(ROA) but they have significant effect on market shares and
employment. On the other hand, provincial politicians have effects on ROA, market
131
shares and number of workers. For the city level SOEs, both city level and provincial
level politicians have significant effects as well as on county level SOEs.
3.6.2
Corruptions and firm's performance
To further examining the channel of political influence, I classify the politicians into
good or bad ones. It allows me to study the differences between firms under good
politicians' jurisdictions and bad ones. Again, I look at firm's performance, market
share, government subsidy, employment, wages and so on. The specification is:
Y,t = ai + at + a3 + yi - Xi,t + Jaili,t + ej,t
(3.15)
where Jaili,t is the dummy whether the firms i is under bad politicians' jurisdiction
at time t. Other independent variables are the same as before. And I only use the
sample of SOEs since it is easier for bad politicians to extract rents from SOEs. I
also use the sample of private firms to check this hypothesis.
One of the concerns here is that politicians' corruption is endogenous and it is
highly depended on the economic performance of the city. The politicians in rich areas
suppose to have more opportunities to take bribe and easier to become corrupted. By
controlling firm fixed effect and city fixed effect, I can mitigate this concern. Another
concern is that in China, the corrupted politicians may not end up in Jail. There
is a saying in China that most corrupted politicians are usually most powerful ones
and they can always avoid prisons. This problem may be less severe since I am using
city level politicians and usually they are not so powerful compared with provincial
and central level politicians. Moreover, communist party passed a rule to regulate
the politicians and fight corruptions at 1997. After that, there has been couple of
thousand politicians sentenced to jail during last two decades. Based on that, my
measurement here has less selection bias.
The results in Table 3.11 panel A show that corrupted politicians will significantly decrease the economic performance of SOEs. ROA decreases by 0.4% that
is about 10% efficiency loss. OROA and labour productivity also decrease significantly(column(2) and (3)). However, market share of these SOEs increases which
means that although these SOEs becomes less competitive but they gain more market powers from the government. Moreover, number of workers significantly increases
by 4.7% and wages decreases by 1.8%. These results are consistent with Shilefer
and Vishny(1994) that politicians extract the rents from SOEs and force SOEs to
hire more workers to gain more political support. In return, they give SOEs more
privileges such as market power.
In panel B Table 3.11, it is the same specification in equation 3.15 and use private
firms instead. From Column(1) to (4), corrupted politicians have no or less effect
on private firms performance except labour productivity. The interesting thing is
that employment increases and wage decrease like SOEs do but with smaller magnitude.The decrease in labour productivity is probably because of the increase of
number of workers.
With all the concerns I discussed before, based on these results in Table 3.11, it
132
seems the case that corrupted politicians extract the rents or personal interests from
SOEs and force them to hire more workers and pay less wages while they compensate
SOEs with more market powers. This effect is less or insignificant for private firms
since it is more costly for politicians to do that on private firms.
3.6.3
Politicians' career path
Although the timing of political turnover is exogenous based on the evidence I showed.
It is interesting to study where do politician go when they leave the old positions. It
is well documented that politicians' promotion is highly depend on economic performance. I have the politicians' profile data which can help me to test it. I classify the
politicians' turnover into 5 categories: promoted, go to jail, retire, die and move to
another city. I only look at the first 4 categories and focus on city level politicians
here.
First, I want to re-test the exogeneity of political turnover. I use probit model to
test:
T urnoveri,t = at + F - StayedYeari,t + yi - Xi,t + Eit
(3.16)
where at is the year fixed effect, StayedYeari,t is the group of dummies of the years
the Politician i stayed at his or her positions at time t. Again, the longest year is 9
years. Xi,t are the politicians' demographics such as age, gender, education, place of
born and the economic performance in his or her jurisdiction such as average ROA
and net profit of the firms in that city. Then, I control politicians' personal effects to
double check the results.
Condition on turnover, I use probit model again to study the career path of politicians,
the specification is:
Careeri =
1
ROAi +
/2
- profiti + Y1 - Xi + Ei
(3.17)
where Careeri is the dummy of 4 categories I mentioned before: promotion, Jail,
Retire and Death of politician i. ROAi and profiti are the average ROA and net
profit of the firms under politician i's jurisdiction. They are the measurement of
politicians' economic performance.
The results are in Table 3.12. In column (1) and (2), again, turnover does not
depend on economic performance and politicians' demographics. And the longer
the politician stayed, the higher probability of turnover. It is consistent with the
results from hazard ratio estimation. In column (3) to (6), promotions are highly
positively correlated with economic performance as well as Jail. It shows again that
corruptions are highly correlated with economic environment. Also, retirement is
positively depended on age and death is totally random. In column (4), I also find
that the probability of going to jail is higher if the politicians were born in rural area.
It might be the case that these politicians don't have as many connections as the
one born in cities. It is difficult for them to avoid jail. It also might be that these
politicians grew up in relatively poor area and this early life experience affects their
behaviours later on.
133
3.7
Conclusion
The main objective of this chapter is to document privatization's effects on the Chinese economy, which is the second-largest economy in the world. Using a detailed
manufacturing firms' dataset (CIC) and the politician profile dataset in China, I find
that more political turnovers lead to fewer privatizations. I use political turnover
as an instrumental variable and find that privatization significantly increases the efficiencies of both SOEs and private firms in the same industry and area. Bigger
privatizations have larger impacts. Privatization benefits all firms in the economy
due to more market competitions and fewer government interventions. For employment, SOEs cut 64% of their workers after being privatized. Private firms increase
16.3% employment if there are privatizations around. Overall, every 100 workers got
fired by SOEs come with an 169 increase in private sector hiring in the same industry
and same province. The aggregate effect of privatization on employment is positive.
While the timing of political turnover is exogenous from my tests, the heterogeneity across politicians is a very important factor for the economy. I find that the
fixed-effect of politicians can affect SOEs' economic performance and employment
significantly. Moreover, I also use politicians' profiles to find that politicians who are
more corrupt will extract more rents from SOEs and give them more market power
in return. Corrupt politicians have smaller effects on private firms. This evidence
mainly supports the political view of SOEs (Shilefer and Vishny (1994)). After all,
privatization is one of the most important factors in the future growth of China.
In the World Bank "China 2030" report, it writes:
"China's rapid growth, particularly since 2003, benefited from SOE restructuring
and expansion of the private sector.. .Relative to the private sector, state-owned enterprises (SOEs) consume a large proportion of capital, raw materials and intermediate
inputs to produce relatively small shares of gross output and value added. A large
share of state enterprise profits comes from a few state enterprises where profitability is often related to limits on competition and access to cheaper capital, land, and
natural resources.. .Privatization and market reform generated vibrant competition in
most manufacturing sectors."
134
3.8
Bibliography
Alchian, Armen Albert. "Some economics of property rights." IL Politico 30
(1965): 816-829.
Bertrand, Marianne, and Antoinette Schoar. "Managing with Style: The Effect
of Managers on Firm Policies." The Quarterly Journal of Economics 118.4 (2003):
1169-1208.
Boycko, Maxim, Andrei Shleifer, and Robert W. Vishny. "A theory of privatisation." The Economic Journal 106 (1996): 309-319.
Dinc, Serdar, and Nandini Gupta. "The Decision to Privatize: Finance, Politics
and Patronage." The Journal of Finance LXVI (2011): 241-270.
Dixit, Avinash, and John Londregan. "The Determinants of Success of Special
Interests in Redistributive Politics." The Journal of Politics 58.4 (1996): 1132-1155.
Dixit, Avinash. "Power of Incentives in Private versus Public Organizations."
The American Economic Review 87.2 (1997): 378-382.
Frydman, Roman, Cheryl W. Gray, Marek Hessel, and Andrzej Rapaczynski.
"When Does Privatization Work? The Impact of PrivateOwnership on Corporate
Performance in Transition Economies" The Quarterly Journal of Economics 114(4)
(1999): 1153-1191.
Gupta, Nandini. "Partial Privatization and Firm Performance." The Journalof
Finance 60.2 (2005): 987-1015.
Gupta, Nandini, Jhon C. Ham, and Jan Svejnar. "Priorities and Sequencing
in Privatization: Theory and Evidence from the Czech Republic." The European
Economic Review 52 (2008): 183-208.
Khwaja, Asim, and Atif Mian. "Do Lenders Favor Politically Connected Firms?
Rent Provision in an Emerging Financial Market." The Quarterly Journal of Economics 120.4 (2005): 1371-1411.
Lindbeck, Assar, and Jorgen W. Weibull. "Balanced-budget redistribution as
the outcome of political competition." Public Choice 52.3 (1987): 273-297.
Megginson, William, and Jeffry Netter. "From State to Market: A Survey of
Empirical Studies on Privatization." Journal of Economic Literature 39.2 (2001):
321-389.
Megginson, William L. "The financial economics of privatization." New York:
Oxford University Press, 2005.
Porta, Rafael La, and Florencio de Silanes. "The Benefits of Privatization:
Evidence from Mexico." The Quarterly Journal of Economics 114.4 (1999): 11931242.
Sapienza, Paola. "The Effects of Government Ownership on Bank Lending."
Journal of FinancialEconomics 72(2) (2004): 357-384.
Sappington, David E. M., and Joseph E. Stiglitz. "Privatization, information
and incentives." Journal of Policy Analysis and Management 6.4 (1987): 567-585.
Shleifer, Andrei, and Robert Vishny. "Corruption." The Quarterly Journal of
Economics 108.3 (1993): 599-617.
Shleifer, Andrei, and Robert Vishny. "Politicians and Firms." The Quarterly
Journal of Economics 109.4 (1994): 995-1025.
135
Silanes, Florencio, and Andrei Shleifer. "Privatization in the United States."
Cambridge, MA: National Bureau of Economic Research, 1995.
Tirole, Jean. "The Internal Organization of Government." Oxford Economic
Papers 46.1 (1994): 1-29.
Vickers, John, and George K. Yarrow. "Privatization: an economic analysis."
Cambridge, MA: The MIT Press, 1988.
Vickers, John, and George Yarrow. "Economic Perspectives on Privatization."
The Journal of Economic Perspectives 5.2 (1991): 111-132.
Wooldridge, Jeffrey M. "Econometric analysis of cross section and panel data."
Cambridge, Mass.: MIT Press, 2002.
Yao, Yang. "Chinese Privatization: Causes and Outcomes." China and World
Economy 13 (2005): 66-80.
136
Figure 3-1: Manufacturing Firms in China
Number of Firms in Chinese Manufacturing Sector
00
C%
Q
0
-
LO
z
1998
2000
1999
2001
SOEs Number
S
Year
2003
2002
2004
2005
Private Firm Number
----
Aggregate Assets in Chinese Manufacturing Sector
-
CO
0
1998
2000
1999
-
2001
SOEs Asset
Year
-
2003
2002
2004
2005
Private Firm Asset
In Figure 3-1, the top Panel is the plot for numbers of SOEs and private firms in China from 1998
to 2005. The bottom Panel is the total asset value of SOEs and private firms. Units in the bottom
Panel is 1 trillion RMB which is about 142 billion US dollars.
137
Figure 3-2: Turnover and Privatization
Privatization Number
5000
4500
4000
,
3500
3000
2500
2000
1500
Em
M m
1000
500
--
0
1999
-i
2000
2001
2002
2003
2005
2004
Year
Pliticians' Turnover
%Turnover
70
60
50 +40
30
20
10
0
1998
In Figure
1999
2000
2001
2002
2003
2004
2005
Year
3-2, the top Panel is the number of total SOEs who were privatized each year. The bottom
Panel the percentage turnover of city level politicians each year from 1998 to 2005.
138
Figure 3-3: Estimated Hazard Functions
Cox proportional hazards regression
Weibull regression
-
N4
U,
C
..I
-
a
2
4
6
8
10
0
5
Swfiw
10
Figure 3-2 is the estimated hazard functions from estimating equation 3.3. Left side panel is from Cox
proportional hazard regression and right side panel is from Weibull distribution hazard regression.
The horizontal line is in years.
139
Table 3.1: Summary Statistics Data
Panel A: CIC Variables
Variable
ROA
OROA
Log(Sales/Worker)
Subsidy
Market share
Log(Wage)
Log(Worker)
Log(Assets)
Log(Sale)
Leverage
Obs
1,041,841
1,041,841
1,036,390
1,041,841
1,041,841
1,033,384
1,036,390
1,041,841
1,041,841
1,041,841
Mean
0.06
0.06
5.05
0.10
0.01
2.23
4.71
9.59
9.76
0.54
S.D.
0.10
0.23
1.07
0.30
0.04
0.69
1.11
1.42
1.33
0.25
Obs
1,420
1,674
1,674
1,343
1,674
1,674
1,674
1,674
Mean
46.16
0.81
0.03
0.25
0.30
0.06
0.01
0.03
S.D.
6.05
0.39
0.16
0.43
0.46
0.24
0.12
0.17
Panel B: Politicians' profile Variables
Variable
Age
Education
Gender
Born at rural
Promotion
Jail
Dead
Retire
Variable ROA= NneT
TotalWage
Wage=
W
of t
NumberofWorkers,
R
, Worker is the total number of employees,
Leverae TotalLiability. Age is the politiTotalAsset
r
NumberofWorkers'
Operati
rofit
TotalBenefit
cian's age in 1998, Education is a dummy of whether politician has college degree or above. Born at
rural is a dummy of whether politician was born in a rural area(not city). Promotion is a dummy of
whether the politician has ever been promoted in turnover. Jail is a dummy of whether the politician
has ever been in Jail. Dead is a dummy of whether the politician was dead during turnover or not.
Retire is a dummy of whether the politician retired in turnover. In total, there is 1674 politicians
and for some of them, age and place of born are missing.
140
Table 3.2: Hazard Ratio Analysis of Politicians' Turnover
Dependent Variable
Age
Education
Gender
Born at rural
Average ROA
Average Total Profit
Hazard Ratio
(1)
Cox
(2)
Cox
(3)
Weibull
(4)
Weibull
0.015**
(0.006)
0.206
(0.447)
0.181
(0.193)
-0.038
(0.085)
-1.223
(1.109)
0.000
(0.000)
0.017**
(0.007)
0.352
(0.488)
0.175
(0.193)
-0.020
(0.090)
-0.002
(0.007)
0.466
(0.538)
0.224
(0.222)
-0.058
(0.100)
-0.509
(1.215)
0.000
(0.000)
-0.000
(0.007)
0.665
(0.574)
0.189
(0.222)
-0.037
(0.105)
Lagged Average ROA
-0.309
(1.208)
0.000
(0.000)
Lagged Average Total Profit
Constant
Observations
Chi squared
2,914
7.900
2,587
7.286
-3.925***
(0.641)
0.331
(1.366)
0.000
(0.000)
-4.339***
(0.664)
2,914
3.596
2,587
2.966
Data restricted to city level mayors and secretary municipal committee of the CPC(party leaders). Robust standard errors reported in parentheses. Column (1) and (2) are the results from Cox
proportional hazard regression. Column (3) and (4) are from Weibull distribution hazard regressions. AverageROA AverageTotalProfit are the means of ROA and net profit from firms which
are under the politicians' jurisdiction each year.
141
Table 3.3: Does Privatization Improve SOEs' Efficiency? (OLS)
Panel A: OLS Regressions use privatized share
(1)
ROA
Variable
(2)
OROA
(3)
(4)
(6)
(5)
(7)
Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
Privatized Share 0.003*** 0.003***
(0.001)
(0.001)
LogSale
0.029*** 0.036***
(0.000)
(0.001)
LogAsset
0.041***
(0.005)
0.001***
(0.000)
0.216***
(0.007)
Constant
2.246***
0.031***
(0.002)
0.019*** -0.251*** 1.982***
(0.007)
(0.046)
(0.109)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
228,230 228,230 224,037
0.057
0.009
0.104
Firm FE
Time FE
Industry FE
Observations
R-squared
-0.235*** -0.293***
(0.014)
(0.016)
Yes
Yes
Yes
Yes
Yes
Yes
228,230 228,230
0.067
0.030
(0.174)
Yes
Yes
Yes
225,471
0.132
-0.001
(0.002)
0.011**
(0.005)
0.015***
(0.004)
5.012***
(0.121)
Yes
Yes
Yes
225,471
0.004
Panel B: OLS Regressions use registration type change
Variable
Privatize
Logsale
(1)
(2)
ROA
OROA
0.003***
-0.001
(0.001)
(0.003)
0.028*** 0.035***
(0.000)
(0.001)
Logasset
(3)
(4)
0.025***
(0.006)
0.002***
(0.000)
0.228***
-0.222*** -0.274***
(6)
(7)
0.006
(0.004)
0.012***
(0.005)
0.059***
(0.005)
0.022***
(0.007)
Constant
(5)
Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
(0.002)
2.208***
0.014**
-0.139***
1.940***
4.976***
(0.011)
(0.014)
(0.149)
(0.007)
(0.034)
(0.086)
(0.096)
Firm FE
Time FE
Industry FE
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
R-squared
262,622
0.064
262,622
0.012
259,796
0.122
262,622
0.056
262,622
0.054
258,169
0.093
259,796
0.005
Data restricted to SOEs both got privatized and remain state-owned. Standard errors in parentheses
are clustered by firm. All the regressions control firm fixed effect, year fixed effect, industry fixed
effect. Panel A are OLS regression results from estimating equation 3.1 using the share owned by
private firms as right hand side variable to measure privatization. Panel B are OLS regression results
from estimating equation 3.2 using Privatize which is the dummy of registration type change from
state-owned to private-owned.
142
Table 3.4: Effects of Privatization on SOEs' Efficiency (Using Political Turnover
Timing as Instruments)
Panel A: 2SLS Regressions (IV: mayor's Tenure)
Dependent Variable
Privatized Share
LogSale
(7)
(6)
(5)
(4)
(3)
(2)
OROA LogSales/worker Mkt share Subsidy LogWage LogWorker
(1)
ROA
0.023**
0.009
(0.009) (0.013)
0.030*** 0.037***
(0.001) (0.001)
LogAsset
Firm fixed effect
Yes
Time fixed effect
Yes
Industry fixed effect
Yes
Observations
184,203
R-squared
0.058
F-stats of first stage 28.14
Yes
Yes
Yes
184,203
0.031
28.14
0.148*
(0.080)
0.010***
(0.003)
-0.105*** -0.227***
(0.036)
(0.069)
-0.041
(0.063)
Yes
Yes
Yes
184,203
0.044
20.68
0.033***
(0.002)
Yes
Yes
Yes
184,203
0.01
28.23
Yes
Yes
Yes
181,913
0.002
28.39
0.212***
(0.007)
Yes
Yes
Yes
181,913
0.127
28.23
Yes
Yes
Yes
180,615
0.082
28.14
Panel B: 2SLS Regressions (IV: mayor's Turnover Dummy)
First Stage
Dependent Variable Privatized Share
Turnover
(1)
ROA
(2)
OROA
(3)
(4)
(5)
(6)
(7)
LogSales/worker Mkt share Subsidy LogWage LogWorker
-0.007***
(0.001)
Privatized Share
0.078**
0.115
(0.040) (0.072)
0.028*** 0.034***
(0.001)
(0.002)
Logsale
Logasset
Firm fixed effect
Time fixed effect
Industry fixed effect
Observations
R-squared
Yes
Yes
Yes
200,827
0.087
Yes
Yes
Yes
200,827
0.077
Yes
Yes
Yes
200,827
0.057
1.276***
(0.418)
0.203***
(0.009)
Yes
Yes
Yes
198,246
0.307
0.025
(0.015)
-0.255*
(0.155)
0.094
(0.325)
-0.640**
(0.280)
Yes
Yes
Yes
200,827
0.043
0.034***
(0.003)
Yes
Yes
Yes
200,827
0.101
Yes
Yes
Yes
196,843
0.101
Yes
Yes
Yes
198,246
0.245
Data restricted to SOEs both got privatized and remain state-owned. Standard errors in parentheses are clustered
by firm. All the regressions control firm fixed effect, year fixed effect, industry fixed effect. Panel A are results of
second stage regression of 2SLS described in equation 3.4 and 3.5. Instrument variables are the length of how many
years have politicians stayed in a city. Weak instrumental variable test statistics(F-test) from first stage regressions
are reported as well. Panel B are are results of second stage regression of 2SLS use actual political turnover as IV.
Turnover is the dummy whether the city is in the turnover year or not. First stage regression result is in Panel B as
well.
143
Table 3.5: Does Privatization Improve Private Firms' Efficiency? (OLS)
Dependent Variable
Privatized SOE
Logsale
Panel A: Before and After Privatization Analysis
and After Privatization Analvsis
(2)
(3)
(4)
(1)
(5)
(6)
(7)
ROA
OROA Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
0.001***
(0.000)
-0.000
(0.001)
0.042*** 0.049***
(0.000)
(0.001)
Logasset
0.007***
(0.002)
0.000
(0.001)
0.209***
0.029***
(0.004)
2.444***
(0.387)
-0.000
(0.000)
-0.004**
(0.002)
0.005**
(0.002)
Privatized SOE
LogSale
-
0.001***
-0.000
(0.000)
(0.001)
0.042*** 0.049***
(0.000)
(0.001)
LogAsset
2 year pre trend
2 year post trend
-0.001
(0.000)
0.001***
(0.000)
-0.001
(0.001)
0.002***
(0.001)
3 year pre trend
0.007***
(0.002)
-0.332*** -0.332***
(0.050)
(0.125)
Firm fixed effect
Yes
Yes
Time fixed effect
Yes
Yes
State fixed effect
Yes
Yes
Industry fixed effect
Yes
Yes
Observations
485,452 485,452
R-squared
0.094
0.050
0.001***
-0.000
(0.000)
(0.001)
0.042*** 0.049***
(0.000)
(0.001)
0.209***
(0.004)
0.004
(0.003)
-0.007***
(0.003)
3 year post trend
Constant
(
(0.001)
-0.333*** 0.333***
-0.383*** -0 .060** 1.713*** 4.830***
(0.050)
(0.125)
).028) (0.291)
(0.136)
(0.259)
Firm fixed effect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Time fixed effect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
State fixed effect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Industry fixed effect
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Observations
485,452 485,452
484,255
485,452 4 85,452 483,423
484,255
R-squared
0.094
0.050
0.124
0.053
.101
0.090
0.028
Panel
Before and After Privatization Analysis
(2)
(3)
(4)
(1)
(5)
(6)
VARIABLES
ROA
OROA Log(Sales/worker)
ROA
0 ROA Log(Sales/worker)
Constant
2.437***
(0.387)
Yes
Yes
Yes
Yes
484,255
0.124
0.008***
(0.002)
0.209***
(0.004)
-0.001**
-0.001
(0.001)
(0.001)
0.001* 0.002***
(0.000)
(0.001)
-0.332*** -0.331***
(0.050)
(0.125)
0.019***
(0.005)
-0.007**
(0.003)
2.420***
(0.388)
485,452
485,452
484,255
0.094
0.050
0.124
Data restricted to private firms who are not converted from SOEs. Standard errors in parentheses
are clustered by firm. Panel A is the DID regression in equation 3.8. PrivatizedSOEis the dummy
of whether the private firm has any privatized SOEs in the same industry and same province. Panel
B is the results from equation 3.9 by adding 2 and 3 year pre/post trend dummies.
144
Table 3.6: Does Bigger Privatization Have Bigger Effects?
Panel A: Privatization's Effects on Private Firms Close-by
Variable
(3)
(4)
(5)
(6)
(7)
(2)
OROA Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
(1)
ROA
LogPrivatizedAsset 0.001**
-0.000
(0.000) (0.000)
Logsale
0.043*** 0.051***
(0.001) (0.001)
Logasset
0.003***
(0.001)
0.002*** -0.000**
(0.001)
(0.000)
0.202***
(0.005)
Constant
0.029***
(0.002)
-0.220***
(0.064)
Firm FE
Time FE
State FE
Industry FE
Observations
R-squared
-0.378***
(0.024)
Yes
Yes
Yes
Yes
269,376
0.096
-0.170
3.178***
(0.256)
Yes
Yes
Yes
Yes
269,376
(0.215)
Yes
Yes
Yes
Yes
269,312
0.055
0.135
-0.001
(0.001)
0.002 2.147***
(0.005) (0.104)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
269,376
0.051
269,376
0.025
Yes
Yes
Yes
Yes
269,023
0.100
0.001*
(0.001)
4.500***
(0.123)
Yes
Yes
Yes
Yes
269,312
0.025
Panel B: Heterogeneous Effects of Different Market Competitiveness
Variable
(1)
ROA
(2)
OROA
PrivatizedSOE 0.001***
0.000
(0.000)
(0.001)
PrivatizedSOE -0.002** -0.006***
Firm#
(0.001)
(0.001)
Logsale
0.043*** 0.051***
(0.000)
(0.001)
Logasset
Constant
Firm FE
Time FE
State FE
Industry FE
Observations
R-squared
-0.299*** -0.310**
(0.128)
(0.032)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
433,052 433,052
0.096
0.052
(6)
(7)
(3)
(4)
(5)
Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
0.015***
(0.003)
-0.048***
(0.006)
-0.015***
(0.000)
0.012***
(0.001)
-0.002
0.003
(0.002)
(0.001)
0.005 -0.058***
(0.004)
(0.004)
0.000
(0.002)
0.015***
(0.005)
0.202***
(0.004)
1.901***
(0.169)
0.031***
(0.001)
-0.380*** -0.398*** 1.750***
(0.072)
(0.085)
(0.106)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
433,052 433,052 432,169
0.042
0.034
0.101
5.288***
(0.121)
Yes
Yes
Yes
Yes
432,907
0.024
Yes
Yes
Yes
Yes
432,907
0.135
Data restricted to private firms who are not converted from SOEs. Standard errors in parentheses
are clustered by firm. Panel A is the results from equation 3.12 and the sample is restricted to the
private firms who have privatized SOEs around. Panel B is is the results from equation 3.13 with
interactions between PrivatizedSOEand the number of firms in the industry and province. Firm#
are in unit of 1,000.
145
Table 3.7: DID Regressions Results(Estimated Privatization)
Dependent Variable
PrivatizedSOE
Logsale
(1)
ROA
0.063***
(0.005)
0.041***
(0.000)
(3)
(4)
(5)
(6)
(7)
(2)
OROA Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
0.104***
(0.007)
0.048***
(0.001)
Logasset
Firm fixed effect
Yes
Time fixed effect
Yes
State fixed effect
Yes
Industry fixed effect
Yes
Observations
408,348
R-squared
-0.042
F test of first stage
124.3
0.323***
(0.034)
0.000
(0.001)
0.142*** 0.087***
(0.016)
(0.030)
Yes
Yes
Yes
Yes
408,348
0.101
125.1
0.030***
(0.001)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
408,348 406,451
0.005
0.084
125.1
126.5
0.209***
(0.004)
Yes
Yes
Yes
Yes
408,348
-0.088
124.3
Yes
Yes
Yes
Yes
407,254
0.065
126.6
0.163***
(0.026)
Yes
Yes
Yes
Yes
407,254
0.002
126.6
Data restricted to private firms who are not converted from SOEs. Standard errors in parentheses
are clustered by firm. The results are from second stage regressions by estimating equation 3.10 and
3.11. The instrumental variables are group of dummies of years the mayors stayed and F-statistics
from first stage regression are reported.
146
Table 3.8: Exclusion Examination and Robustness Check
Panel A: Exclusion Examination
(1)
ROA
Variable
Turnover
(2)
OROA
(7)
(6)
(3)
(4)
(5)
Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
0.001
-0.000
(0.001)
(0.001)
0.021*** 0.023***
(0.002)
(0.003)
Logsale
Logasset,
Constant
-0.203*** -0.232***
(0.036)
Firm FE
Time FE
Industry FE
Observations
R-squared
Yes
Yes
Yes
16,263
0.062
(0.044)
Yes
Yes
Yes
16,263
0.038
0.000
-0.001
(0.010)
0.007
(0.009)
-0.004
(0.007)
0.354***
(0.134)
0.043***
1.087 1.125***
(1.516)
(0.017)
(0.235)
5.681***
(0.424)
(0.001)
0.390***
(0.031)
0.010
(1.305)
Yes
Yes
Yes
15,997
0.105
-0.026
(0.053)
Yes
Yes
Yes
16,263
Yes
Yes
Yes
16,263
0.097
0.064
Yes
Yes
Yes
15,812
0.071
Yes
Yes
Yes
15,997
0.037
Panel B: OLS industry trend robustness check
Variable
(1)
ROA
(2)
OROA
PrivatizedShare 0.002*** 0.003**
(0.001)
(0.001)
Logsale
0.029*** 0.036***
(0.000)
(0.001)
Logasset
Constant
Firm FE
Time FE
Industry FE
Industry trend
Observations
R-squared
-0.253*** -0.296***
(0.013)
(0.016)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
228,230 228,230
0.077
0.035
(7)
(6)
(5)
(4)
(3)
Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
0.042***
(0.005)
0.001***
0.015***
(0.005)
0.017***
(0.004)
0.214***
(0.007)
2.164***
(0.171)
Yes
Yes
Yes
Yes
225,471
0.142
0.031***
(0.002)
0.022*** -0.207*** 2.009***
(0.007)
(0.047)
(0.109)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
228,230
228,230 224,037
0.148
0.014
0.111
5.057***
(0.124)
Yes
Yes
Yes
Yes
225,471
0.010
(0.000)
-0.001
(0.002)
Panel A use SOEs who are 100% state-owned from 1998 to 2005 and use actual political turnover as
independent variable to exam the exclusion condition of using political turnover as IV. Panel B is
the robustness check by re-estimating the results in Table 3.3 Panel A and add interactions between
industry dummies and time dummies. Standard errors in parentheses are clustered by firm.
147
Table 3.9: 2SLS Robustness Check
Panel A: Industry trend robustness check
Dependent Variable
Privatized Share
(6)
(7)
(4)
(5)
(3)
(2)
OROA Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
(1)
F OA
0.026***
(0.010)
0.173**
(0.082)
Firm fixed effect
Yes
Time fixed effect
Yes
Industry fixed effect
Yes
Industry time trend
Yes
Observations
184,203
R-squared
0.064
F-stat of first stage
27.07
0.210***
(0.007)
Yes
Yes
Yes
Yes
181,913
0.136
27.27
Logsale
0.012
(0.014)
0.029*** 0.036***
(0.001) (0.001)
Logasset
Yes
Yes
Yes
Yes
184,203
0.035
27.07
0.007***
(0.003)
-0.095*** -0.209***
(0.037)
(0.070)
-0.052
(0.064)
Yes
Yes
Yes
Yes
184,203
0.164
27.31
0.033***
(0.002)
Yes
Yes
Yes
Yes
184,203
-0.000
27.34
Yes
Yes
Yes
Yes
181,913
0.008
27.22
Yes
Yes
Yes
Yes
180,615
0.092
26.55
Panel B: Industry trend robustness check
Dependent Variable
Privatized SOEs
Logsale
(1)
ROA
(7)
(5)
(6)
(4)
(3)
(2)
OROA Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
0.063*** 0.103***
(0.005) (0.008)
0.041*** 0.048***
(0.000) (0.001)
Logasset
Firm fixed effect
Yes
Yes
Time fixed effect
Yes
Yes
Industry fixed effect
Yes
Yes
Industry time trend
Yes
Yes
Observations
408,348 408,348
R-squared
-0.035 -0.078
F-stat of first stage
105.9
105.9
0.317***
(0.036)
0.206***
(0.004)
Yes
Yes
Yes
Yes
407,254
0.079
107.8
0.000
(0.001)
0.134*** 0.079**
(0.017) (0.032)
0.160***
(0.028)
Yes
Yes
Yes
Yes
408,348
0.134
106.5
0.030***
(0.001)
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
408,348 406,451
0.015
0.090
106.5
107.6
Yes
Yes
Yes
Yes
407,254
0.010
107.8
Table 3.9 is the robustness check of privatization's effect on SOEs and private firms using years
that mayor stayed as IV. Panel A is the results from re-do Table 3.4 Panel A and add interactions
between industry dummies and time dummies. Panel B is the results from re-do Table 3.7 and add
interactions between industry dummies and time dummies. All standard errors are clustered by
firm.
148
Table 3.10: Do Politicians' Individual Fixed Effects Matter?
Panel A: Province level firm
City level politicians
Variable
F-stat P-value Constraints
ROA
0.9238 0.9425
Provincial level politicians
N
F-stat P-value Constraints
N
863
18,503 1.5242 <0.0001
171
19,202
Market Share 3.8817 <0.0001
863
18,503 7.3476 <0.0001
171
19,202
LogWorker
857
18,257 2.0118 <0.0001
169
18,951
1.4655 <0.0001
Panel B: City level firm
City level politicians
Variable
Provincial level politicians
F-stat P-value Constraints
ROA
1.1230 0.0030
Market Share 3.8185 <0.0001
2.5496 <0.0001
LogWorker
1,098
1,098
1,090
N
F-stat P-value Constraints
41,623 2.0445 <0.0001
41,623 7.9435 <0.0001
41,106 4.6190 <0.0001
168
168
166
N
42,533
42,533
42,010
Panel C: County level firm
City level politicians
Variable
Provincial level politicians
F-stat P-value Constraints
ROA
1.8432 <0.0001
Market Share 2.7664 <0.0001
4.4848 <0.0001
LogWorker
1,116
1,116
1,109
N
F-stat P-value Constraints
71,588 4.4796 <0.0001
71,588 6.4467 <0.0001
70,649 9.1727 <0.0001
166
166
164
N
72,514
72,514
71,570
Data restricted to state-owned firms. Reported in the table are the results from fixed effects panel
regressions in equation 3.14. Standard errors are clustered by firm. Panel A is sample of SOEs at
provincial political hierarchy. Panel B is sample of SOEs at city level political hierarchy. Panel C is
sample of SOEs at county level political hierarchy. F-statistic and P-value in the table are calculated
from politicians' fixed effects and N is the number of observations. In total, there are 1,116 city level
politicians and 171 provincial level politicians who changed the job during sample period.
149
Table 3.11: Do Bad Politicians Extract Rents From Firms?
Panel A: OLS regressions on SOEs
Variable
Jail
Logsale
(1)
(2)
ROA
OROA
-0.004*** -0.003***
(0.001)
(0.001)
0.028*** 0.035***
(0.000)
(0.001)
Logasset
(4)
(3)
(5)
(6)
(7)
Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
-0.027***
(0.006)
0.001***
(0.000)
-0.031*** -0.019***
(0.003)
(0.005)
0.014*
(0.007)
Yes
Yes
Yes
260,871
0.056
0.022***
(0.002)
-0.139*** 1.941***
(0.034)
(0.086)
Yes
Yes
Yes
Yes
Yes
Yes
260,871 256,419
0.056
0.093
0.231***
Constant
-0.223*** -0.274***
(0.011)
(0.014)
Firm FE
Yes
Yes
Time FE
Yes
Yes
Industry FE
Yes
Yes
Observations 260,871 260,871
R-squared
0.064
0.012
(0.007)
2.186***
(0.149)
Yes
Yes
Yes
258,045
0.122
0.029***
(0.005)
4.964***
(0.095)
Yes
Yes
Yes
258,045
0.004
Panel B: OLS regressions on private firms
Variable
Jail
Logsale
(1)
(2)
ROA
OROA
-0.001
-0.000
(0.001)
(0.001)
0.040*** 0.050***
(0.000)
(0.001)
(3)
(4)
-0.027***
(0.007)
0.000
(0.000)
Logasset
0.180***
Constant
-0.002
(0.012)
Firm FE
Time FE
Industry FE
Observations
R-squared
(5)
(6)
(7)
Log(Sales/worker) Mkt share Subsidy LogWage LogWorker
-0.012*** -0.026***
(0.003)
(0.006)
0.017***
(0.006)
0.023***
-0.253***
(0.017)
-0.179
(0.132)
(0.005)
3.357***
(0.154)
Yes
Yes
Yes
319,506
Yes
Yes
Yes
319,506
Yes
Yes
Yes
319,248
Yes
Yes
Yes
319,506
Yes
Yes
Yes
319,506
Yes
Yes
Yes
318,803
Yes
Yes
Yes
319,248
0.088
0.049
0.138
0.118
0.051
0.099
0.032
(0.001)
-0.208*** 2.128***
(0.050)
(0.099)
4.030***
(0.121)
Data restricted to SOEs in panel A and private firms in panel B. Standard errors in parentheses are
clustered by firm. The results are from estimating equation 3.15. Jail is the dummy of whether the
politicians are bad or good."bad" means he or she went to Jail eventually.
150
Table 3.12: Politician Career Path
Dependent Variable
Average ROA
Total Profit
Age
Education
Gender
Born at rural area
Year_2
Year_3
Year_4
Year_5
Year_6
Year 7
Year_8
Year_9
Constant
Person fixed effect
Observations
(1)
Turnover
-1.298
(2.334)
-0.000
(0.000)
0.010
(0.007)
-0.931**
(0.453)
-0.237
(0.199)
0.003
(0.081)
1.125***
(0.126)
1.851***
(0.125)
2.247***
(0.129)
2.567***
(0.136)
3.204***
(0.145)
3.424***
(0.163)
3.815***
(0.191)
3.984***
(0.246)
-1.011
(1.459)
NO
6,509
(2)
Turnover
(3)
(4)
Promotion Jail
-5.137
(5.485)
0.000
(0.000)
4.428***
(1.317)
(5)
Retire
(6)
Dead
1.186
(2.677)
-0.003
(0.025)
0.033**
(0.015)
-1.701
(3.901)
-0.047
(0.071)
0.026
(0.028)
0.226*
(0.130)
-0.172
(0.192)
0.595
(0.474)
-0.023
(0.249)
-2.517**
-3.652***
(0.763)
NO
(0.998)
NO
NO
-3.504**
(1.440)
NO
1,167
1,140
1,130
1,157
0.026**
(0.011)
-0.063***
(0.009)
0.308
(0.467)
5.078***
(1.911)
0.027*
(0.014)
0.018
(0.015)
-0.304
(0.581)
0.252
(0.250)
-0.064
(0.089)
1.445***
(0.143)
2.625***
(0.155)
3.543***
(0.172)
4.508***
(0.194)
5.925***
(0.225)
7.079***
(0.268)
8.276***
(0.328)
9.466***
(0.424)
-1.021
(1.294)
YES
7,446
2.377***
(0.664)
Data in column (1) and (2) is panel data for each politicians' turnover. In total, there are 1,167
politicians and 7 years. Column (3) to (6) are the politicians' career path condition on turnover.
Promotion is the dummy of whether the politicians got promoted or not. Jail is the dummy of
whether the politicians went to jail or not. Retire and Dead are for retirement and death respectively.
All regressions are probit model and robust standard errors are reported in parentheses.
151