Empirical strategy

advertisement
Running for the Exit
International Banks and Crisis Transmission
Ralph De Haas (EBRD)
Joint with Neeltje Van Horen (DNB)
17th Dubrovnik Economic Conference
June 2011
1
Introduction: aim of the paper
• Cross-border bank lending singled out as a key channel of crisis
transmission
• Lehman Brothers collapse: syndicated cross-border lending declined by
53 per cent on average compared to pre-crisis levels…
• … but some countries suffered more than others from a ‘sudden stop’
Introduction
Distribution of post-Lehman ‘sudden stop’
Number of destination countries
19
20
15
12
11
10
5
4
5
4
2
0
le nding 75-99% 50-75% 25-50%
s to p
de c re a s e in le nding
0-25%
1
2
0
0-25% 25-50% 50-75% 75-99%
inc re a s e in le nding
> 100%
Introduction: aim of the paper
•
We look at differences across banks to explain this cross-country
heterogeneity in the sudden stop (keeping all else equal)

•
Specifically: does access to borrower information affect stability of bank
lending?
Using loan-level data, we find that cross-border lending is more stable if:

Destination country is geographically close

Bank has a network of domestic co-lenders in destination country

Bank has prior experience in destination country

Bank has subsidiary in (EM) destination country
Introduction: contribution of the paper
•
Contribution to literature on transmission of current financial crisis



Role US$ funding vulnerability banking systems (Cetorelli and Goldberg 2010)
Role average profitability banking systems (McGuire and Tarashev 2008)
Role stock-market performance banking system (Herrmann and Mihaljek 2010)
•
Previous work based on (bilateral) BIS data
•
We are first to use loan-level data
Introduction: background literature

Screening and monitoring varies across borrowers
– opaque borrowers rationed more (Stiglitz and Weiss 1981)

Screening and monitoring varies over time
– marginal benefit increases during crisis or recession (Ruckes 2004) when
agency problems increase as net worth of firms declines (Rajan 1994)

So: opaque borrowers are rationed disproportionally during an adverse
shock (‘flight to quality’; Bernanke et al. 1996)
Introduction

We expect a more severe ‘sudden stop’ when banks are unable to
sufficiently increase the screening of foreign borrowers:
1.
Distance between international bank and the borrower
2.
Presence of a subsidiary of the international bank
3.
Cooperation of international bank with domestic banks
4.
Experience of the international bank in a country
1. Distance
• Theory

Information costs increase with distance, in particular for ‘soft’ info (Stein 2002)

Screening and monitoring more difficult when distance increases: geographical
credit rationing (Jaffee and Modigliani 1971)
• Empirical literature

Negative relationship between geographical distance and amount of lending (Buch
2005; Portes et al. 2001)

Negative relationship between cultural distance and amount of lending (Giannetti
and Yafeh 2009)

Negative relationship between distance and pricing power of banks (spatial price
discrimination, Degryse and Ongena 2005)
• Impact on the stability of bank lending?
2. Presence of a local subsidiary
• Theory

Local subsidiary reduces distance between loan officer and borrower (Mian 2006)

Local subsidiary improves collection and processing of soft information

But establishing a local subsidiary creates ‘functional’ distance between loan officer
and HQ (Aghion and Tirole 1997)

New problem: transmitting ‘soft’ info from subsidiary to HQ… Involves not only
transportation costs but also intrabank agency costs (Rajan et al. 2000)
• Empirical literature

Greater functional distance reduces credit availability (Alessandrini et al. 2009)
• Impact on cross-border lending stability?
3. Cooperation with domestic banks
• Theory

Domestic banks may have a comparative advantage in reducing information
asymmetries vis-à-vis local firms (Mian 2006, Carey and Nini 2007)

Repeated co-lending with domestic banks may allow foreign banks to increase
local know-how as well
• Empirical literature

(Contemporaneous) local bank participation leads to larger, longer and cheaper
syndicated loans (Nini, 2004)
• Impact on lending stability?
4. Previous lending experience

Theory


Empirical literature


Repeated interaction reduces information asymmetries and agency problems
Repeat lending reduces information asymmetries in the syndicated loan market
(De Haas and Van Horen, 2010)
Impact on lending stability?
Required characteristics of data
1.
Loan flows:
•
From individual banks…
•
… to individual countries…
•
… over a prolonged period of time
2.
Lending by one bank to various countries (exploit within-bank
variation)
3.
Lending by multiple banks to one country (control for credit
demand)
4.
Information about the underlying individual deals
5.
Important market (to generalize results)
Syndicated loan data have all of these characteristics
Data
Novel dataset: sample of 118 largest international banks
•
Only commercial, savings, cooperative or investment banks
•
Each covers at least 0.01% of the cross-border syndicated loan market
•
Participated in at least 20 cross-border loans in 2006
Banks from 36 countries (43 banks from emerging markets)
•
Lending cross-border to 60 advanced and emerging countries
•
2,146 bank-country pairs
Data: calculation of cross-border lending flows
•
We download all syndicated loans to private borrowers between
January 2005 and October 2009
•
Each loan has multiple lenders, so we determine for each bank the share
of the loan it provided:
•

± 25% sample: we have data on loan distribution

± 75% sample: we assume equal loan distribution (and show robustness tests)
Identify all the loan portions that are ‘cross-border’

•
Cross border means: nationality of bank (parent) is different from nationality of borrower)
Result: per bank, per month, total cross-border lending to each country
Example
Loan to US borrower signed
in October 2008
Syndicate members:
1.
2.
3.
4.
Source: Dealogic Loan Analytics
Citigroup
Deutsche Bank
Nomura
Erste Group
Number of loans: 23,237
Number of loan portions: 108,530
Citigroup
Deutsche Bank
Nomura
Erste Group
Erste Group
Erste Group
Erste Group
Erste Group
+
Total cross-border lending Erste in October 2008 to U.S.
Empirical strategy
•
We compare lending from bank i to country j in two periods: PostLehman (Oct 08-Oct 09) versus pre-crisis (Jan 05-Jul 07)
•
Dependent variables
1. Change in cross-border lending volume from bank i to country j
2. Change in cross-border number of loans from bank i to country j
3. Sudden stop dummy: loan volume decline <-75 per cent
Empirical strategy
•
Information variables

Distance: Km distance (in logs) between the country of bank i and
borrower country j

Subsidiary: Presence

Domestic lenders: Number of different domestic lenders with
whom bank i participated in loans to country j since 2000 (as a %
of all domestic lenders)

Experience: Number of loans by bank i to country j since 2000
that had matured by September 2008
Empirical strategy
•
Challenge is to control properly for changes in credit demand
•
Khwaja & Mian (AER, 2008) technique:
•

Multiple banks lending to one firm: use firm fixed effects to
control for credit demand at firm level

In our case, multiple banks lending to same country: use
country fixed effects to control for credit demand at the hostcountry level (cf. Cetorelli & Goldberg 2010)

Banks active in multiple countries: we can also use bank fixed
effects (or bank-specific controls)
OLS (logit for SS dummy) with standard errors clustered by bank
Empirical strategy
Lij    I ij    X i   j  ij
'
'
Controls
1. Bank solvency – 2006 and Δ(2009-2006)
2. Bank liquidity – 2006 and Δ(2009-2006)
3. Bank size
4. Pre-crisis exposure to country j
5. State support (‘financial protectionism’)
Empirical strategy
Lij    I ij    X i   j  ij
'
'
Lij    I ij   i   j  ij
'
In sum: we control for time invariant country variables, changes in credit
demand, and bank-specific variables
Allows us to focus on pairwise bank-country determinants
Empirical results: baseline results
Volume
Subsidiary
0.122*** 0.117**
[0.006]
Distance
[0.013]
-0.043** -0.073***
[0.021]
Domestic lenders
[0.000]
0.362*** 0.369***
[0.000]
Experience
0.066
0.056
[0.131]
[0.210]
-0.016
-0.048**
[0.376]
[0.016]
0.281*** 0.264***
[0.000]
0.051*** 0.059***
[0.000]
[0.000]
[0.000]
[0.002]
0.011
0.014
[0.466]
[0.379]
Economic impact
• Distance: 19% higher reduction lending for borrowers at mean distance compared to
borrowers at minimum distance
• Domestic lenders: 9% lower reduction lending to country with mean level of cooperation
compared to country without domestic bank network
Empirical results: baseline results (II)
Sudden stop
Subsidiary
-0.138*** -0.146***
[0.000]
[0.000]
Distance
0.084*** 0.110***
[0.000]
-0.053
[0.309]
[0.226]
0.047** 0.072***
[0.000]
Domestic lenders
-0.040
[0.012]
-0.570*** -0.588***
[0.000]
-0.371*** -0.360***
[0.000]
Experience
[0.002]
[0.000]
[0.000]
-0.096*** -0.117*** -0.041** -0.054**
[0.000]
[0.000]
[0.027]
[0.018]
Observations
2026
1960
2026
1960
1998
1934
2026
1960
1998
1934
Pseudo R2
0.168
0.226
0.176
0.235
0.188
0.244
0.181
0.238
0.196
0.255
Empirical results: baseline results (III)
Numbers
Subsidiary
0.041*** 0.035***
[0.000]
0.029*** 0.024***
[0.000]
Distance
-0.013*** -0.015***
[0.001]
[0.001]
Domestic lenders
0.090*** 0.079***
[0.000]
[0.007]
-0.006
-0.010**
[0.131]
[0.030]
0.071*** 0.056***
[0.000]
Experience
[0.001]
0.011*** 0.012***
[0.000]
[0.003]
-0.001
0.001
[0.003]
[0.006]
[0.787]
[0.822]
Observations
2075
2100
2075
2100
2047
2072
2075
2100
2047
2072
R-squared
0.273
0.344
0.269
0.344
0.28
0.353
0.268
0.342
0.285
0.359
Results: controls
•
Some evidence banks retrenched from non-core (emerging)
markets
•
Banks that reduced lending the most:




•
Supported
Small
Low solvency (2006)
Banks that had to increase liquidity
But economic effect limited compared to information variables
Robustness checks
Sudden stop
Base
1 year
change
Extensive
margin
Alternative rule
Model
Extreme
distribution
-0.053
-0.033
-0.079
-0.068
-0.078*
-0.048
[0.226]
[0.453]
[0.116]
[0.116]
[0.056]
[0.269]
0.072***
0.053**
0.071***
0.078***
0.077***
0.059***
[0.002]
[0.017]
[0.003]
[0.000]
[0.000]
[0.001]
-0.360***
-0.357***
-0.242**
-0.342***
-0.368***
-0.226**
[0.000]
[0.000]
[0.031]
[0.002]
[0.000]
[0.024]
-0.054**
-0.070***
-0.143***
-0.057**
-0.055**
-0.03
[0.018]
[0.005]
[0.000]
[0.014]
[0.015]
[0.120]
0.447
3.690*
1.812
0.992
0.873
-1.33
[0.835]
[0.084]
[0.389]
[0.629]
[0.700]
[0.333]
Observations
1,934
1,809
2,077
1,921
1,924
1,913
(Pseudo) R-squared
0.255
0.260
0.287
0.260
0.258
0.232
Subsidiary
Distance
Domestic banks
Experience
Exposure
Empirical results: What is distance?
All countries
Distance
0.072***
0.067**
0.072***
0.074***
0.070***
[0.002]
[0.005]
[0.002]
[0.001]
[0.003]
Common language
-0.118** -0.085*
[0.012]
[0.078]
Colonial links
-0.055
-0.056
[0.491]
[0.486]
Credit info
-0.020
-0.026
[0.363]
[0.230]
Legal difference
Subsidiary
Domestic banks
0.086** 0.088**
[0.034]
-0.053
-0.060
-0.049
-0.067
-0.053
-0.069
-0.055
-0.062
-0.050
[0.226]
[0.167]
[0.260]
[0.121]
[0.222]
[0.112]
[0.205]
[0.149]
[0.254]
-0.360***-0.394***-0.361***-0.391***-0.354***-0.392***-0.353***-0.380***-0.347***
[0.000]
Experience
[0.019]
[0.000]
[0.000]
[0.000]
[0.000]
[0.000]
[0.001]
[0.000]
[0.001]
-0.054***-0.069***-0.054***-0.071***-0.054***-0.072***-0.055***-0.069***-0.053***
[0.018]
[0.002]
[0.019]
[0.002]
[0.018]
[0.001]
[0.016]
[0.002]
[0.020]
0.447
0.214
0.65
-0.157
0.438
-0.355
0.192
0.12
0.662
[0.835]
[0.921]
[0.759]
[0.943]
[0.839]
[0.872]
[0.929]
[0.956]
[0.754]
Observations
1,934
1,934
1,934
1,934
1,934
1,934
1,934
1,934
1,934
Pseudo R-squared
0.255
0.251
0.256
0.25
0.255
0.25
0.256
0.252
0.257
Exposure
Results: endogeneity
•
Omitted variables that are correlated both with info variables and
stability of lending?
•
But problem not as pronounced as
•
Control for all unobserved country variables (e.g. growth potential country)
•
Control for all unobserved bank variables (e.g. bank strategy)
•
Main bank-country pair variables already included in model
Strategy to test if results are biased
•

Only very weak bank-country pair instruments, so leave IV

Control for additional bank-country pair variables: trade, (banking) FDI, differences in
supervisory power, and stringency of capital regulation

Findings: results unchanged when adding these variables to the model
Empirical results: endogeneity
All countries
Subsidiary
Distance
-0.053
-0.052
-0.052
-0.053
-0.053
-0.051
[0.226]
[0.236]
[0.236]
[0.226]
[0.226]
[0.245]
0.072*** 0.070*** 0.066*** 0.072*** 0.072*** 0.065***
[0.002]
Domestic banks
Exposure
[0.005]
[0.002]
[0.002]
[0.006]
-0.360*** -0.364*** -0.631*** -0.360*** -0.360*** -0.364***
[0.000]
Experience
[0.002]
[0.000]
-0.054*** -0.054**
[0.000]
[0.000]
[0.000]
[0.000]
-0.052**
-0.054**
-0.054**
-0.052**
[0.018]
[0.018]
[0.022]
[0.018]
[0.018]
[0.022]
0.447
0.482
1.287
0.447
0.447
1.296
[0.835]
[0.821]
[0.552]
[0.835]
[0.835]
[0.545]
T rade
-0.018
-0.014
[0.337]
[0.434]
Bank FDI
-0.049***
-0.047**
[0.008]
[0.010]
Supervisory power
0.025***
0.023***
[0.000]
[0.001]
Capital regulation
0.036***
0.000
[0.000]
[0.993]
Observations
1,934
1,934
1,934
1,934
1,934
1,934
Pseudo R-squared
0.255
0.255
0.258
0.255
0.255
0.259
Results: extensions
First-time vs repeat borrower
•

Impact of access to borrower information is same for repeat and first-time borrowers

Except Experience which is particularly important for first-time borrowers

Probability of Sudden stop higher for lending flows to first-time borrowers
Bank vs non-bank borrower
•

Access to borrower information had no impact on stability of lending to bank borrowers

Agency problems and mistrust in inter-bank market were too large

Probability of Sudden stop higher for bank borrowers
Conclusions

We know little about what affects the stability of cross-border lending.
Especially not about banks’ behavior across different countries
•
Our results suggest that information asymmetries not only affect the
level but also stability of cross-border lending
•
Resilience cross-border lending depends on ability of banks to limit
increase in agency problems
•
Even in a ‘hard information’ market access to (supplementary) ‘soft
information’ matters
•
Specific role for distance to borrower, cooperation with domestic banks,
presence subsidiary (in EMs), and lending track-record
Policy implications
•
Banks further away from customers may be less reliable sources of
funding especially when they have no local presence
•
Suggests that countries that want to open up their economy to crossborder lending flows

Should consider to also allow foreign subsidiaries and branches

Attract debt funding from lenders that are geographically close (or at least not
only from remote lenders)

Also develop the domestic banking system to not become completely reliant on
the kindness (and stability) of strangers…
This figure compares the change in cross-border syndicated lending to a country (horizontal axis) with the change in total syndicated lending
(cross-border plus domestic syndicated lending) in that country. Lending change is the percentage change in average monthly lending in the precrisis period compared to the post-Lehman period. The pre-crisis period is defined as January 2005 to August 2007 and the post-Lehman period
as October 2008 to October 2009. The left-hand pane shows all 60 destination countries included in our dataset whereas the right-hand pane
zooms in on those countries that experienced a decline in both cross-border and total syndicated lending. Countries that experienced a percentage
change in domestic lending that was exactly equal to the percentage change in cross-border lending are on the 45º line. Countries where domestic
lending shrank faster (slower) than cross-border lending are to the right (left) of this line.
Finally: domestic syndicated lending was unable to cushion
much of the decline in cross-border inflows…
200%
150%
0%
IND
SVN
Total lending
PAN
-20%
LUX
-40%
CHN
J PN
PER
VNM
50%
-60%
POL
KAZ
ISLPHL
NGA
0%
-80%
-50%
TWN
BRA
SGP
AUS
CAN
M EX HRVNZL
ITA
AZE CHE
USA
CZE EGY
ZAF BHR
ESP
DEU
IRL THA
QAT
BM U
KWT KOR
TUR
ARE
M YS
HKG NLD GBR
NOR
FRA DNK
BGR
RUSBEL
GIB
SWE
ROU
PRT
100%
-100%
-100%
CHL NGA
FIN
GRC
AUT
IRN UKR ARG
J EY HUN
LBR OM N
LVA SAU
-50%
0%
50%
100%
150%
-100%
-100%
Cross-border lending
-80%
-60%
-40%
-20%
0%
Thank you!
37
Download