FOREIGN POLITICAL RISK EXPOSURE,
CAPITAL ALLOCATION, AND PERFORMANCE*
Burcin Col
Art Durnev
Pace University
University of Iowa
Alexander Molchanov
Massey University
Abstract
We argue that international trade is a significant conduit of foreign political uncertainty into U.S.
markets. We find that industries that export considerable shares of their output to countries with
high political risk or countries that hold national elections in a given year experience lower total
factor productivity growth, lower valuation, and worse accounting performance. The key channel
of political uncertainty transmission is disruption of investment efficiency. Our results are not
driven by economic risk or the quality of institutional environment of trading-partner countries,
and they remain robust when we account for potential endogeneity of export flows.
Keywords: Political Risk, International Trade, Investment Efficiency
JEL Classification: F10, G32
*
Art Durnev, Henrie B. Tippie College of Business, University of Iowa, Iowa City, IA, 52242, USA. E-mail: artemdurnev@uiowa.edu., www.artdurnev.com.
Burcin Col, Lubin School of Business, Pace University, New York, NY, 10038, USA. E-mail: bcol@pace.edu.
Alexander Molchanov (corresponding author), School of Economics and Finance, Massey University, Private Bag
102 904, North Shore MSC, Auckland, New Zealand. E-mail: a.e.molchanov@massey.ac.nz.
The authors are grateful for the helpful comments and suggestions by Ben Jacobsen, Francisco Munoz, Felix
Schindler, and seminar participants at Massey University, New Zealand Finance Colloquium, New Economic School,
Australasian Finance and Banking Conference, and European FMA meetings.
0
“A month ago Ahmed Ezz was one of the most powerful businesspeople in Egypt ... Today he is a has-been. Protesters
have demonized him and torched his company headquarters. He is under investigation, his assets have been frozen
and his right to travel has been restricted. Western companies that cultivated Mr. Ezz wasted their time and money.”
“Business people need to think harder about political risk” (The Economist, February 12, p. 75)
1. Introduction
In the 2000s, export flows from the U.S. metal industry to Egypt increased fivefold from 14 to 62
million U.S. dollars.1 While the expansion to that relatively politically-stable region (at the time)
was perceived as a profitable endeavour by U.S. companies, the disruption of export flows caused
by political events in Egypt in 2011 has undermined the potential return from investment into
growth of export channels.2 Although a number of factors could potentially be responsible for
this, it is intuitive to think of high political risk in Egypt as one of the key aspects behind such a
disruption. We posit that export activity is an important conduit of foreign political risk
transmission into performance of domestic economy at the industry level.
A new stream of literature argues that local economic and political risks affect how firms
invest, reallocate capital, and that higher risks deteriorate firm subsequent performance (Bloom
(2009), Bloom et al. (2007), Julio and Yook (2012a), Pástor and Veronesi (2012a, 2012b)). Some
of the previous research also examined how foreign risks affect corporate choices of multinational
corporations.3 This paper adopts an alternative view of exposure to foreign political risks. We
hypothesize that export activity is a channel of foreign political risk transmission into domestic
markets. While it has often been assumed that foreign political risk is of lesser importance to
exporters than to multinationals, as less capital is at stake (Stapenhurst (2002)), the loss of future
revenues from exporting may significantly outweigh the value of expropriated assets (Gillespie
1
Source: WTO Trade Statistics.
See The Economist (2011). This can be viewed as a case of capital misallocation – over-investment to be specific –
which we later define as the deviation of marginal q (the ratio of the change in the market value to the unexpected
change in assets) from the optimal level of one. According to our calculations, over the period of 2000-2005, marginal
q for the metals industry was 0.890, decreasing to 0.602 during 2006-2011, indicating greater over-investment for the
latter period.
3
Specifically, Desai et al. (2004) show that subsidiaries located in politically-risky countries are more highly
leveraged than their counterparts in safer countries. Desai et al. (2008) document more volatile returns on investment
in riskier countries. Henisz (2000) suggests that multinational firms serving politically-risky markets are likely to
share ownership with local partners. Greene et al. (2009) show that U.S. multinational corporations, on average,
exhibit better capital budgeting decisions than companies without foreign operations. Julio and Yook (2012b)
document a drop in FDI flows from U.S. companies to foreign affiliates when there is a national election either in the
U.S. or the destination country.
2
1
(1989)). Outright asset expropriations are far less common now than they were a few decades ago,
and disruptions often come from, e.g., various trade barriers, which may be politically-motivated.4
This paper’s contribution to the existing literature is twofold. First, we explicitly relate foreign
political risk to performance of domestic economy at the industry level. Second, we posit that the
specific channel of such transmission is quality of capital allocation. We claim that when an
industry exports a substantial portion of its output to a politically-risky country, its investment in
export-related projects would be lower than when dealing with a politically-safe importer. Due to
an additional political risk-related discount factor, the number of positive NPV projects decreases.
Thus, performance of such industries will suffer. We further claim that ex post realization of
political risk has an impact on quality of capital allocation. If political risks are not realized in a
given year, under-investment occurs, as project NPV is ex post higher than anticipated. The
opposite happens if anticipated political risks are realized. Thus, even though investment is ex
ante efficient (given the expected political risk reflected in the discount factor), we will observe ex
post inefficiency depending on the realization of political risk. We note that over-investment in
one year will not ‘cancel out’ under-investment in other years – the loss of efficiency from
foregone opportunities is likely to be exacerbated by over-investment in years of realized political
disruptions.
Using a large sample of industry-year observations of the U.S. manufacturing sector, we
provide robust evidence of a detrimental effect of foreign political risk on domestic performance.
Industries with a larger exposure to trading partners’ political environment (measured by political
risk score and occurrence of national elections in a given year) experience significantly lower total
factor productivity (TFP) growth rates, lower valuations, and worse accounting performance
(measured by ROA).5
In order to test the specific channel of foreign political risk into domestic performance, we
explore the quality of capital allocation, which is measured by how close Tobin’s marginal q is to
its optimal level. A larger deviation indicates lower quality of capital allocation. Tobin’s marginal
q, roughly defined as the ratio of the change in total market value to unexpected change in capital
4
For example, the National Association of Manufacturers estimates that, due to the lack of a free trade agreement
with Chile, American exporters lost $800 million in sales in 2007. In the analysis, we distinguish political risk from
the economic risk (e.g., adverse exchange rate movements) and institutional risk (rule of law and corruption).
5
Exposure to trading partners’ political environment is defined as the export-weighted average of trading partners’
political risk scores or national election indicator variables (see Section 3 for a more detailed definition and
examples).
2
stock, is computed using a methodology introduced in Durnev et al. (2004) and further refined in
Ferreira and Laux (2007), Green et al. (2009), Hornstein and Zhao (2011), and Faccio et al.
(2011). Initially, the optimal level of marginal q is set equal to 1.6 We employ a two-stage
estimation procedure. In the first stage, we regress squared distance between marginal q and 1 on
exposure to trading partners’ political environment. We then regress performance measures on
both explained and unexplained capital misallocation. Consistent with our expectations, we
document a significantly negative effect of explained capital misallocation on TFP growth rates,
Tobin’s Q, and ROA. Thus, trade-induced foreign political risk impacts performance of domestic
industries through its effect on capital allocation.7
To further refine our tests, we collect information on actual political crises and divide our
sample according to the realized political crises in trading-partner countries. In the sample of
industry-years with political crises in at least one trading-partner country, marginal q is 0.78,
indicating over-investment (values less than one indicate over-investment, whereas values greater
than one imply under-investment). When no crises occur, marginal q is 1.20. This is consistent
with the hypothesis that over-investment (under-investment) occurs when the political crises are
realized (not realized).
When considering potentially adverse impacts of foreign trade, political, economic, and
institutional risks of trade-partner countries go hand-in-hand (Erb et al. (1996)). Therefore, we
include economic and institutional risk measures as control variables in order to empirically assess
the relative importance of the three types of risks in their impact on capital allocation and
performance. Out of the three types of risks, political and economic risks are statistically
significant; however, political risk has a larger impact than economic risk. For example, in the
case of the metals industry (industry with the median value of exports), when political risk of
trading partners increases by one standard deviation (which is equivalent to switching exports
entirely from Belgium to Argentina), capital allocation quality decreases by 17% (relative to its
average value). A one standard deviation increase in economic risk results in a smaller decrease of
capital allocation efficiency of 4%. Effects of similar magnitude are observed for performance
6
In the latter part of the analysis, we consider factors that may affect the optimal benchmark level and estimate it as
an endogenous parameter.
7
This paper is not the first one to document adverse effects of foreign trade: Newbery and Stiglitz (1984) find that
trade increases uncertainty and income volatility. Di Giovanni and Levchenko (2009) document higher output
volatility in more open industries.
3
variables. TFP growth decreases by 3.3% for a one standard deviation increase in political risk,
and by 1% for economic risk.8
To further address the inter-relatedness of different types of risks we perform two-stage
estimation, regressing political risk on economic and institutional risk scores and using the
unexplained part of the political risk score in our main regressions. Both explained and
unexplained parts of political risk have an adverse effect on performance measures and the quality
of capital allocation, indicating that our results are not driven entirely by economic and
institutional factors.
Certainly, a political environment itself has a substantial effect on export flows. Countries and
industries consider a variety of political factors when developing their trade policies.9 Handley
and Limao (2012) show that policy uncertainty can affect investment and entry decisions in the
context of international trade. Export structure, in this sense, is endogenous to political
environment. Such endogeneity could potentially create a bias in our estimates of political risk
transmission through exports. However, we believe that such a bias works against our potential
findings. Countries and industries adjust their export flows in order to, among other things,
mitigate political uncertainty. Therefore, any statistical significance of foreign political risk we
document, if anything, is likely to be reduced by endogeneity of export flows.
It is, however, possible for a self-selection bias to arise: industries with better performance and
higher quality of capital allocation could increase exports to countries with a stable political
environment and vice versa. We address the issue by instrumenting industry export flows with
such exogenous factors as distance between trading partners, the size of trading partners’
economies, and bilateral trade agreements. Yet another bias may emerge: when dealing with a
country with a stable political environment, a firm may find it advantageous to set up a subsidiary
there, rather than to increase exports. Therefore, the volume of exports to politically-stable
countries would be biased downward. We address this by directly incorporating the sales of
foreign subsidiaries of multinational corporations (MNCs) in a given country into our export
exposure measure.
A possibility remains that differences in performance across industries are driven by
unobserved characteristics, rather than export exposure to political risk. If this is the case, we
8
For ROA, the decrease is 4.3% due to political risk and 0.8% due to economic risk. Valuation decreases by 16.5%
(relative to its mean) due to political risk and 7.3% due to economic risk.
9
See, for example, Mitra et al. (2002), Magee (2003), Egger et al. (2008), and Baier and Bergstand (2007).
4
would observe a negative association between performance and trade-induced political
uncertainty in a sample of firms for which such a link should not exist, namely, firms with no
foreign operations. To address this concern, we run a series of ‘placebo’ tests by constructing
performance and capital allocation measures using a sample of domestic-only firms. We observe
that for these firms neither performance nor capital allocation is related to trade-induced political
uncertainty. Therefore, unobserved industry characteristics are not behind our results.10
We note that this paper does not dispute the well-established benefits of foreign trade, such as
diversification (see, e.g., Hirsch and Lev (1971)). However, as an adverse consequence of
investment efficiency distortions, we document that industries that are more exposed to a trading
partner’s political instability experience lower growth in productivity, worse accounting
performance, and lower valuation. Our results indicate that care must be exercised when directing
export flows to politically-risky trading partners.
The rest of the paper is organized as follows. Section 2 develops testable hypotheses. Section
3 describes the sample and the empirical specification. We present the results in Section 4. Section
5 concludes.
2. Hypotheses
When deciding how much to invest in an export-related project, a firm evaluates its net present
value according to the following equation:
NPV 
 EB   EC 
t
t
(1  d t   t )
t
t
,
(1)
where the difference between expected benefits B and costs C is discounted by a risk-free factor d
and country political risk factor . Clearly, higher political risk is associated with lower NPV,
which would result in worse performance. Moreover, under a high level of political risk,
technology transfer is slower between U.S. industries and trading-partner countries, which is
10
One may argue that we need to run our tests using only firms with foreign operations. We perform the main
analysis using all available firms for three reasons. First, endogenously-determined optimal level of marginal q, which
is, among other factors, a function of investment cycles and taxes, may be biased by foreign political risk in a sample
of exporting-only firms. Second, a large sample of firms is important for reliable estimation of marginal q. Third,
while we employ an extensive algorithm for identifying domestic-only firms, it is imperfect and may misidentify
some companies.
5
likely result in slower growth in industry Total Factor Productivity (TFP).11 Thus, ceteris paribus,
industries that export to politically-risky countries would have fewer positive NPV projects than
their counterparts that export to politically-safe countries. This leads to our first hypothesis:
Industries exporting a substantial share of their output to politically-risky countries exhibit worse
performance and slower growth in TFP.
Political risk discount factor reflects anticipated political risk, and investment decision is
optimal given this risk level. Political risk score is, essentially a probability of a politically-related
disruption. Actual realization of these disruptions, however, has a profound effect on project NPV.
If political risks are not realized in a given year, that period’s NPV is greater than anticipated.
Therefore, investment level was lower than optimal. If a political disruption occurs, NPV is lower
than expected, resulting in over-investment. In either case, investment deviates from its optimal
level. Our second hypothesis is: Industries that export a large share of their output to politicallyrisky countries exhibit suboptimal quality of capital allocation.
Efficient capital allocation is an essential requirement for economic growth. If trade-induced
political uncertainty reduces investment efficiency, capital is not withdrawn from sectors with
poor prospects, and is not invested in profitable sectors. This would have a detrimental effect on
performance, industry valuation, and growth in TFP. Our third hypothesis is: Deviations from
optimal investment due to trade-induced political risk is associated with slower TFP growth and
worse performance.
3. Data, Variables, and Empirical Setup
Our sample is the panel of U.S. manufacturing industries aggregated at the three-digit SIC level
(SIC codes 2000 through 3900). The sample is restricted to manufacturing industries because of
more comprehensive trade and accounting data coverage. The unit of observation is industry-year,
and the scope of the sample is 1990-2006, resulting in 1,399 industry-year observations. We drop
1998 from the sample and do not consider the years after 2006 to ensure that our results are not
driven by high volatility of economic and financial fundamentals during the crisis period.12 On
average, in a given year, the sample contains 93 industries.
11
TFP growth reflects growth in output not caused by inputs, and it can be viewed as a measure of technological
change.
12
Note that the significance and magnitude of the main regression coefficients do not change noticeably if we do not
exclude 1998 from our sample.
6
3.1 Exposure to Foreign Politics
The analysis rests on the premise that foreign political risk is transmitted into domestic
industries through their export activities. First, we define trade exposure of industry i to trading
partner c in year t as in Boutchkova et al. (2012),
TRADE i ,c ,t 
EXPORTS i ,c ,t
SALESi ,t
(2)
.
Data on exports are obtained from the UNCTAD/WTO PC-TAS database compiled by
COMTRADE. Data on industry sales are from the Bureau of Economic Analysis.13 Export data
are classified according to the Standard International Trade Classification (SITC). Sales data are
organized by commodity type using the International Standard Industrial Classification (ISIC).
We convert three-digit ISIC codes to three-digit SITC codes, and then three-digit SITC codes to
three-digit SIC codes.14 We also check the product-industry correspondence manually. To avoid
the impact of outliers, we winsorize the export and sales data at the 1% and 99% levels.
Next, based on the individual trade shares to each trading partner and political variables
pertaining to each foreign trading partner, we compute industry-specific index of exposure to
political risk as
 (TRADE
c
i ,c ,t
 POLITICALc,t )
,
(3)
where POLITICALc ,t is the value of the political variable (political risk or election indicator
variables which are defined below) pertaining to each trading partner c for year t. We call these
variables trade-political risk index and trade-elections index, respectively.
To illustrate, consider the example presented in Figure 1 below. The Primary metals industry’s
(SIC code 3300) top three trading partners in 2000 were Canada, Mexico, and China. In this
industry, 7.18% of its production was exported to Canada, 6.23% to Mexico, and 4.14% to
China.15 In 2000, the political risk variable took values of 13.02 for Canada, 26.18 for Mexico,
and 22.64 for China. The value of the trade-political risk index for the primary metals industry in
2000 is thus equal to 0.072×13.02 + 0.062×26.18 + 0.041×22.64 = 3.489. In 2000, Canada and
13
To construct an alternative measure of trade exposure (discussed in the results section), we add the sales of foreign
subsidiaries of multinational corporations (MNCs) to the industry exports.
14
For this conversion, we apply computer codes provided by Jon Haveman available at http://www.haveman.org.
15
On average, the actual number of trading partners equals 39. Three trading partners are chosen for illustrative
purposes.
7
Mexico held national elections. Therefore, trade-elections index would be equal to 0.072×1 +
0.062×1 + 0.041×0 = 0.134.
Figure 1: Construction of trade-political risk index and trade-elections indexes.
Primary Metals (SIC = 3300)
7.18%
Canada: Political Risk = 13.02
4.14%
6.23%
Mexico: Political Risk = 26.18
China: Political Risk = 22.64
The descriptive statistics of trade exposure (the sum of trade exposures over all trading
partners), trade-political risk, and trade-elections indexes by two-digit SIC codes are presented in
Table IA.16 According to the trade-political risk index, Apparel industry is the riskiest, while the
Stone, Clay, and Glass industry is the safest. Petroleum Refining industry has the highest tradeelections index, while Stone, Clay, and Glass has the lowest one. The correlation coefficient
between the two political risk measures is 0.316 with p-value = 0.00.
The minimum number of trading partners is 25 while the maximum is 49. Not surprisingly,
Canada is the largest trading partner of the U.S. manufacturing industries, with 26% of its exports
directed there. Other top trading partners are Mexico (14%), Japan (9%), UK (5%) and South
Korea (3%).17 Our analysis is based on exports to 49 main trading partners, mainly due to data
availability. Only 4.2% of exports are directed to countries not in our sample.
We use two sets of variables to measure the risk of a political environment: overall political
risk score and whether national elections take place in a given year. Political risk scores are
obtained from the International Country Risk Guide (ICRG), compiled by the Political Risk
16
While the regression analysis is performed using the sample of three-digit SIC industries, the summary statistics in
Table I are reported using two-digit SIC industries to save space.
17
Our results are robust if we exclude Canada and Mexico from the list of trading partners.
8
Service Group. The political risk score is the sum of the following sub-indices: socioeconomic
conditions, investment profile, external conflict, military in politics, religious tensions, and
democratic accountability. The original index ranges from 0 (political instability) to 60 (political
stability). We subtract the original index from 60 so that larger values indicate greater political
risk and expand it to a 1-100 scale.18 Political risks of trading-partner countries are presented in
Table A1 in the Appendix. In the sample, Pakistan is the riskiest country (political risk = 58.736)
while Luxembourg (political risk = 10.011) is the safest.
The national election variable is a dummy variable which equals to one for national election
years (presidential elections in presidential systems and parliamentary elections in parliamentary
systems) and zero otherwise. We use the 2006 edition of the World Bank’s Database on Political
Institutions described in Beck et al. (2010) to obtain election dates. We then cross-reference the
dates with a number of sources, such as Journal of Democracy, Elections around the World,
Election Guide, PARLINE Database on National Parliaments, and CIA Factbook. The sample of
elections is presented in Table A2 in the Appendix. The sample covers 171 national elections with
the average of 3.4 elections per country.
3.2 Dependent Variables
We employ three variables to assess the impact of trade-induced political uncertainty on
performance of domestic industries: accounting performance, industry valuation, and TFP growth.
Accounting performance is measured by ROA (net income over total assets). The data are
obtained from COMPUSTAT. Some descriptive statistics aggregated at a two-digit SIC level are
provided in Panel B of Table I. Tobacco Products (SIC 2100) has the highest ROA in our sample,
while Chemical and Allied Products (SIC 2800) has the lowest ROA.
Industry valuation is measured by Tobin’s average q, which is the ratio of industry value (V) to
its stock of capital (A). The variables V and A are defined in detail in Section 3.3. Petroleum
Refining (SIC 2900) has the highest Tobin’s q, while Fabricated Metal Products (SIC 3400) has
the lowest Q.
18
Full definitions of sub-indexes are available at http://www.prsgroup.com/ICRG_Methodology.aspx. The ICRG
includes two institutional variables (rule of law and corruption) in the calculation of political risk. We exclude them
from the calculation of political risk because we aim to single out the effect of political risk from the quality of
institutional environment. Instead, we explicitly control for the quality of institutional environment in every
regression. The results remain unchanged if we include the rule of law and corruption in the political risk measure.
9
Following Bartelsman and Gray (1996), we define TFP growth as the percentage increase in
gross output less the percentage increase in (weighted) capital, labour, energy, and material inputs.
The TFP growth rates are obtained from NBER database.19 Petroleum Refining (SIC 2900) has
the highest TFP growth rate, while Primary Metal Industries (SIC 3300) has the lowest rate.
3.3 The Quality of Capital Allocation
We measure the quality of capital allocation by the proximity of Tobin’s marginal q to its
optimal level. Durnev et al. (2004) develop a simple and intuitive methodology to estimate
marginal q, which was also used in Ferreira and Laux (2007), Green et al. (2009), Hornstein and
Zhao (2011), and Faccio et al. (2011). Durnev et al. (2004) argue that marginal q is the estimate of
marginal project’s profitability index. Due to declining marginal returns on investment, capital is
invested till the incremental value of a project is equal to its cost, implying an optimal level of
marginal q equal to one. Durnev et al. (2004) define marginal q (denoted as 𝑞̇ ) as the ratio of the
change in the market value of a firm V due to an unexpected unit increase in its stock of capital
goods K, which equals the expectation of profitability index,
q 
V
1   cf t 
ENPV 
 E 
 1
 EPI 
t 
K C  t 1 (1  r ) 
C
.
(4)
In (4), all capital spending is aggregated into a project with the set-up cost C, cf is total cash flow,
r is the discount rate, and E represents investor expectations. It is optimal to invest into projects
with positive NPV, that is, when profitability index PI is greater than 1. Therefore, ignoring taxes
and other complications, firms invest up to the point where marginal q equals 1. Thus, the greater
the distance between marginal q and one, the worse the quality of capital allocation is, and
marginal q greater (lower) than one indicates under-investment (over-investment). Later we
endogenize the optimal level and estimate it in a non-linear regression setting, in order to account
for factors that may shift the optimal level from one.
Marginal q can be expressed as
q j ,t 
V j ,t  V j ,t 1 (1  rˆj ,t  dˆ j ,t )
A  A (1  gˆ  ˆ )
j ,t
19
j ,t 1
j ,t
http://www.nber.org/nberces/nbprod96.htm
10
j ,t
,
(5)
where V j ,t and A j ,t are the market value and stock of capital goods of firm j in year t, rˆj ,t is the
expected return from owning firm j. Variables dˆ j ,t and ˆ j,t represent disbursements to investors
and expected depreciation of capital goods, respectively. Rewriting (5) and normalizing by A j ,t 1
we obtain
V ji,t  V ji,t 1
A ij ,t 1
 q j ,t ( g j ,t   j ,t )  q j ,t
A ij ,t  A ij ,t 1
A ij ,t 1
  j ,t
D ij ,t 1
A ij ,t 1
 r j ,t
V ji,t 1
A ij ,t 1
(6)
,
where i denotes industries a firm j belongs to. In terms of actual estimation, each industry’s
marginal q is represented by coefficient
 0 in the following regression estimated for each three-
digit SIC industry and year,
V ji,t
A ij ,t 1
  j ,t  
i
0 ,t
A ij ,t
A ij ,t 1

i
1,t
V ji,t 1
A ij ,t 1

i
2 ,t
D ij ,t 1
A ij ,t 1
 u ij ,t
(7)
.
In order to calculate marginal q for a given three-digit SIC industry-year, we run panel regression
(7) using quarterly firm-level data. For example, if an industry has 30 firms in a given year, we
collect input variables for four quarters and run a panel using 120 observations to estimate
marginal q for that industry-year. The process is repeated for all 1,399 industry-years. We drop
industries with fewer than 20 firm-quarter observations.20
We estimate V j ,t and A j ,t for firm j in year t as:
V j ,t  Pt (CS j ,t  PS j ,t  LTD j ,t  SD j ,t  STA j ,t )
(8)
A j ,t  K j ,t  INV j ,t
(9)
,
where CS is market value of shares outstanding, PS is estimated market value of preferred shares,
LTD is estimated market value of long-term debt, SD is book value of short-term debt, STA is
book value of short-term assets, P is inflation adjustment using the GDP deflator, K is estimated
market value of plant, property, and equipment, INV is estimated market value of inventories.
The market value of long-term debt is estimated as the value of a 15-year bond issued at par
using book values of debt. The market values of inventories and property, plant, and equipment
20
Our results do not change substantially if we estimate the above regression using annual data. In this case, every
regression is run on fewer observations reducing the efficiency of the estimates.
11
are measured recursively using 10% depreciation rate. We refer to Appendix A in Durnev et al.
(2004) for further details.
Table IB presents the summary statistics for marginal q, aggregated at the two-digit SIC level.
Marginal q is, on average, less than 1 (0.972) indicating slight over-investment over the sample
period. Industrial and Computer Equipment (SIC 3500) exhibits the lowest marginal q (0.130),
while Transportation Equipment (SIC 3700) has the highest (2.200).
The quality of capital allocation is measured by the squared deviation of marginal q from its
optimal level, h, which is initially set equal to 1. According to Table IB, the best quality of
allocation (lowest deviation of marginal q from 1) is observed for the Textile Mill Products
industry (SIC code = 2200), while the Transportation Equipment industry (SIC code = 3700)
exhibits the worst capital allocation.
The optimal value of h can deviate from 1 for a number of reasons, such as taxes, endogeneity
of capital structure and disbursement policies or the low frequency of capital spending disclosure.
Moreover, the change in firm value can arise from past investments or future investment options.
Therefore, in the latter part of the analysis, we relax the assumption of h being equal to 1 and
estimate
(q i  h) 2  bZ i  ui
(10)
,
where Z i represents the list of independent variables, using nonlinear least squares and determine
h and regression coefficients simultaneously. In the case of a squared deviation from h, (10) is
equivalent to
q i2  h 2  2hq i  bZ i  u i
.
(11)
In the nonlinear least squares estimation, the following function is minimized with respect to b:
Qi (b) 
1 I
[ y i  f ( xi ; b)] 2

I i 1
,
(12)
2
2
where y i  q i and f ( xi ; b)  h  2hq i  b Z i .
As a robustness check, we employ an investment efficiency measure developed by Wurgler
(2000) – elasticity of investment with respect to value added (  it ). It is defined as
 I ij ,t 
 Q ij ,t 
i
i


ln i
  t   t ln  i    ij ,i
I

Q

 j ,t 1 
 j ,t 1 
,
12
(13)
where I denotes capital expenditure and Q denotes industry value. Holding everything else equal,
larger values of  it indicate better investment efficiency. Like marginal q,  it is estimated for
every 3-digit industry i and year t using panels of quarterly firm-level data. The elasticity of
investment measure can be viewed as a simplified version of marginal q. We prefer to use the
marginal q measure for the main analysis because the elasticity of investment cannot differentiate
between under-investment and over-investment. However, we note that the two are related. The
correlation coefficient between the squared distance of marginal q from 1 and the elasticity of
investment is 0.418 with p-value = 0.00.
3.4 Control Variables
We include a number of control variables, since the quality of capital allocation may be
influenced by multiple firm, industry, and country factors. For example, liquidity may affect
investment efficiency, as cash-strapped firms may be prone to under-investment. Therefore, not
controlling for liquidity may obscure the relationship between foreign political risk and capital
allocation. In addition, our analysis may suffer from omitted variable bias, as some factors may
have a simultaneous effect on trade exposure and capital allocation quality. For example, more
diversified firms may be more inclined to engage in foreign trade. Also, such firms may exhibit
worse investment efficiency, as they are less focused. Moreover, political risk may be correlated
with economic and institutional risks. To control for these possibilities, we include the following
control variables.
Economic and Institutional risk. Erb et al. (1996) analyse the relative importance of country
political, economic, and institutional risks for portfolio investment. We include economic
(ECONOMIC) and institutional (INST) risks (both in levels and in interactions with export
exposure) to ensure that the main independent variables (trade-political risk and trade-elections
indices) do not pick up economic and institutional factors. The variables are obtained from the
ICRG. Economic risk is based on such variables as GDP per capita, real GDP growth, inflation,
budget balance, and current account. Institutional risk is based on the rule of law and corruption. 21
21
Description of methodology is available at http://www.prsgroup.com/ICRG_Methodology.aspx
13
Economic and institutional risks of trading partners are presented in Table A1. 22 Similar to the
trade-political risk interaction defined in (3), we form the interaction of trade with economic risk
and interaction of trade with institutional risk as
 (TRADE
c
i ,c ,t
 ECONOMIC c,t )
,
(14)
and
 (TRADE
c
i ,c ,t
 INSTc,t ) .
(15)
U.S. political risk and U.S. elections. Domestic political environment may influence both foreign
trade exposure (through a variety of politically-driven trade barriers and/or agreements) and
investment efficiency (through politically-motivated resource allocation). However, one can argue
that more export-dependent industries are less affected by domestic political environment
resulting from the diversification effect of foreign sales. We thus control for the interaction of
industries’ export share with U.S. political risk and with U.S. elections (note that we do not
include U.S. political risk and elections as separate terms, as we control for time fixed effects).
Firm-specific return variation. Durnev et al. (2004) document that the magnitude of firm-specific
return variation is indicative of more informative stock pricing, which in turn results in more
value-enhancing capital budgeting. We measure firm-specific return variation as one minus R2 of
the regression of firm returns on index (S&P 500) returns (aggregated at the three-digit SIC level),
estimated annually.
Average q. This variable serves as a proxy for the presence of intangibles and measures the
importance of growth options, which can affect capital allocation quality. Average q is defined as
the industry total market value V (the sum over firms) scaled by industry stock of capital goods A
defined in (8) and (9), respectively.
Size. Industry size may have a significant impact on capital allocation efficiency. Firms in large
industries may have more cash and fewer growth opportunities, thus making them prone to overinvestment. On the other hand, firms in smaller industries may be more likely to ration capital and
22
Similarly to the political risk index, we rescale the original economic and institutional risk indexes to a 0-100 scale
and subtract them from 100 so that larger values indicate greater risks.
14
under-invest. We measure industry size as the natural logarithm of industry property, plant, and
equipment K.
Liquidity. We conjecture that cash-strapped firms may be prone to under-investment and vice
versa. Liquidity is measured by industry total net current assets over industry property, plant, and
equipment K.
Leverage. Both Jensen (1986) and Myers (1977) argue that the existing capital structure impacts
capital allocation decisions. We define leverage as industry total long-term debt scaled by the
stock of capital goods A.
Diversification. Extensive literature relates corporate diversification to investment efficiency.
While Stein (1997) demonstrates positive effects of diversification, Rajan et al. (2000), among
others, document adverse effects. We measure diversification as an asset-weighted average
diversification level of firms with primary business in a given three-digit SIC industry. Firm
diversification is, in turn, defined as the number of three-digit segments reported in the
COMPUSTAT Industry Segment file.
R&D Expenditures. We include this variable as capital budgeting may be less efficient in
industries with higher intangible asset intensity. R&D expenditures are measured per dollar of the
stock of capital goods A.
In addition, time-specific factors such as U.S. macroeconomic and political conditions are
controlled for by time fixed effects.
3.5 Empirical Specification
We first regress the performance variables (TFP growth, valuation, ROA) on the lagged values
of trade-political risk index (TRADE_POL. RISK) or trade-elections index (TRADE_ELECT), and
lagged values of control variables (CONTROLS). The control variables include interactions of
export exposure with economic and institutional risks defined in (14) and (15). In order to account
for unobserved heterogeneity, we include industry (i) and year fixed effects (i). Every
15
independent variable is lagged by one year to reduce endogeneity. The panel regression equations
(run over the time period from 1990 through 2006, excluding 1998) are estimated as follows:
PERFORMANC Ei ,t   i   t    TRADE _ POL. RISK i ,t 1 or TRADE _ ELECTi ,t 1 
 'CONTROLS i ,t 1   i ,t
,
(16)
where i indexes industries, t indexes years, and  is the vector of coefficients. The main
coefficient of interest (  ) is expected to be negative (Hypothesis 1), and it measures the
transmission of trade-partner countries political risks into performance variables.
Next, to test the second hypothesis, we similarly regress squared deviation of marginal q from
1 on lagged trade-political risk or trade-elections index and controls:
(q  1) i2,t   i   t    TRADE _ POL. RISK i ,t 1 or TRADE _ ELECTi ,t 1 
 'CONTROLS i ,t 1   i ,t
(17)
In the latter part of the analysis, q is benchmarked against an endogenously determined optimal
level h. We expect to (  ) be positive. The standard errors in (16) and (17) are clustered by
industries and years to adjust them for heteroskedasticity, cross-sectional, and time-series
correlation.
Finally, in order to test the specific channel of transmission of trade-induced political risk
(Hypothesis 3), we regress the performance measures on explained and unexplained parts of
squared deviation of marginal q from 1, obtained in (17).
4. Results
4.1 Regression results
Table II presents the results of the main regression analysis. The dependent variables are TFP
growth (Panel A), industry valuation (Panel B) and ROA (Panel C). Specification 1 presents
results for trade-political risk index, while specification 2 reports the results for trade-elections
index.23
23
In specifications 3 and 4, we add the total sales value of MNCs to the calculation of export shares. These results are
described in the robustness section.
16
In Panel A, we observe a significant negative effect of political risk in trading-partner countries
on TFP growth for both trade-political risk and trade-election specifications. The results remain
unchanged in specifications 3 and 4, in which we account for the sales of MNCs. Similar results
are observed when we use industry valuation as a dependent variable (Panel B) – there is a
significantly negative effect in all specifications. As for accounting performance measured by
ROA, we observe a significantly negative effect of trade-political risk and trade-election
interactions when MNC sales are accounted for (specifications 3 and 4).
To gauge the economic significance of the above result, we consider the primary metals
industry (trade exposure = 18.2%). According to the magnitude of the coefficient on the tradepolitical risk interaction (0.020 from specification 1 of Table II), a one standard deviation
increase in political risk of 9 points (equivalent to switching exports from Belgium to Argentina)
reduces growth in TFP by a substantial amount of 3.276% ( = 9 × 0.182 × (0.020)). Similarly, a
one-standard deviation increase in political risk reduces ROA by 4.259% (= 9 × 0.182 × (0.026))
and industry valuation by 0.387 (= 9 × 0.182 × (0.236)), which is a 16% reduction relative to the
sample mean of 2.377.
The impact of elections in the trade-partner countries is also economically significant. If the
fraction of trading partners with elections in a given year increases by 10% (an increase equivalent
to one standard deviation), TFP growth of the median trading share industry, primary metals
industry, is decreased by 0.1 × 0.182 × (1.014) = 1.845%, which is a significant drop relative to
the sample average growth rate of TFP of 2.080%. Similarly, valuation is reduced by 0.1 × 0.182
× (8.143) = 0.148, which is a 6% decrease relative to the sample mean of valuation of 2.377. For
ROA, it is 0.1 × 0.182 × (2.300) = 4.186 % reduction, a large number relative to the mean ROA
of 3.430%
Politically-risky trading partners are likely to be economically unstable with less developed
institutions. However, our findings are not affected by economic risk and institutional risk because
we explicitly control for them in our regressions by forming trade-economic and tradeinstitutional risk interaction terms. A one standard deviation increase in economic risk for the
primary metals industry of 5.152 decreases TFP growth by 5.152 × 0.182 × (0.011) = 1.031%.
Similarly, the decrease in valuation is 5.152 × 0.182 × (0.186) = 0.174 (7.337% relative to the
mean). The decrease in ROA is 5.152 × 0.182 × (0.009) =0.844%.
17
Interaction of trade with U.S. political risk and U.S. elections is insignificant in all
specifications. To account for the possibility that more export-oriented industries exhibit worse
performance regardless of political environment in trade-partner countries, we include industries’
level of trade exposure as one of the control variables. It is insignificant in TFP growth and ROA
specifications. It is positive and significant (albeit marginally) in valuation specifications,
implying that exporting industries have higher valuations. Therefore, trade by itself is not
detrimental to industry performance.
With regard to control variables, average firm size is negative and significant in all
specifications, whereas R&D expenditures are positive and significant across the board. Leverage
is negative and significant in TFP growth and valuation specifications.
Table III presents the results of the regression analysis with dependent variable being squared
deviation of the difference between marginal q (denoted as MQ in the tables) and its optimal level,
which is initially set at 1. Specification 1 presents results for trade-political risk index, while
specification 2 reports the results for trade-elections index. In specifications 3 and 4, we add the
total sales value of MNCs to the calculation of export shares. When trade exposure is interacted
with overall political risk (specification 1), the coefficient on the trade-political index is positive
and highly significant with p-value = 0.00. A similar result is observed in specification 2 that uses
the trade-elections index. Consistent with our hypotheses, industries that are more exposed to
trading partners’ political risk exhibit greater deviation from an optimal investment level.
A question remains, however: does this detrimental effect channel through distortions in
capital allocation efficiency? To address this, we collect explained and unexplained components
of the deviation of capital allocation from its optimal level in equation (17).24 We then regress
performance measures on these explained and unexplained components. The results are presented
in Table IV. We indeed observe that a significant effect is channelled through the explained part
of capital allocation deviation for TFP growth rate (Panel A), industry valuation (Panel B) and
accounting performance (Panel C). We interpret this as evidence of political risk of tradingpartner countries being transmitted into worse performance of domestic industries through its
impact on capital allocation efficiency.
24
We use specification 3 (trade-political risk interaction with MNC sales). Results remain virtually unchanged when
other specifications are used.
18
As noted in the previous section, the optimal level of marginal q can deviate from 1 for a
number of reasons. Therefore, in Table V, we use the deviation of marginal q from its
endogenously estimated optimal level, h (see equations (10) – (12) for estimation details). For the
most part, the results remain qualitatively unchanged. The interactions of trade exposure with the
political environment variables are positive and highly significant. The remaining results remain
largely robust. The endogenously estimated level of optimal capital allocation ranges from 0.780
to 0.836, indicating that the optimal marginal q is less than one, potentially resulting from taxes or
low frequency of capital spending disclosure.
When we use the Wurgler’s (2000) measure of investment efficiency measured as investment
elasticity, the results remain qualitatively unchanged.25 Regression coefficients for trade-political
risk and trade-elections interactions are negative and statistically significant. Therefore, industries
exporting to countries with high political risk scores or countries holding national elections in a
given year exhibit lower investment elasticity and, thus, less efficient investment.
4.2 Addressing Potential Bias in Export Flows
Our results may suffer from the omitted variable and self-selection biases: industries that have
better quality of capital allocation may be exporting mostly to “safer” countries and vice versa.
Therefore, the relationship between capital allocation efficiency and export activity may be
spurious. We address this issue by employing Instrumental Variables (IV) estimation. In the first
stage, we instrument industry trade shares by a set of exogenous parameters:
TRADE c ,i ,t   0  1  DISTANCEc   2  GDPc,t   3  IND _ PRES c,i ,t 
  4  IMPORT _ DEPc,t   5  CAP _ CONTROLS c ,t   6  AGREEMENTc,i ,t   c ,i ,t
.
(18)
In (18), TRADEc,i,t represents exports (scaled by total sales) of U.S. industry i to country c in year
t. DISTANCE is the geographical distance between the U.S. and its trade-partner country,
expressed in the log of miles. Industries are more likely to export to countries that are
geographically closer. The size of a trading-partner country, GDP, is measured by the log of real
GDP. We include this variable because industries are more likely to trade with larger countries.
IND_PRES represents the domestic presence of a given industry in a trade-partner country
25
We do not tabulate these results to save space.
19
(measured by the log of industry sales). Presumably, the demand for imports from a U.S. industry
is lower if a trade-partner country can satisfy domestic demand. IMPORT_DEP represents a
country’s dependence on imports (measured by the ratio of imports to GDP). CAP_CONTROLS is
the index of a trading partner’s capital controls, and it is included in (18) because less trade is
expected with countries that have tighter restrictions on foreign capital.26 Finally, AGREEMENT
is a dummy variable equal to one if there is a trading agreement for industry i with a tradingpartner country c in year t, and zero otherwise.27 The F-test of joint significance is high enough
(17.19) to claim that the instruments are not weak. The instruments also pass the Hansen’s (1982)
J-test of over-identifying restrictions indicating their exogeneity.
In the second stage, we use predicted values from (18) to construct the trade-political risk and
trade-elections indexes. We then run the main performance regression as in (16). The results of
the two-stage estimation are presented in Table VI. With respect to the main independent
variables (trade-political risk and trade-elections interactions), there is no noticeable change in the
magnitude and significance of these variables. Therefore, we conclude that our main results are
not contaminated by the endogeneity of trade flows.
4.3 Accounting for Multinational Corporations
When trading with a politically-stable country, it may be advantageous for an industry to set
up foreign affiliates in that country, which can decrease its exports and thus bias our trade
exposure measure. However, we believe that such a bias would work against statistical
significance of our measure, capturing a significant effect of foreign political risk transmission.
Consider the following example: an industry exports 50% of its output to Canada (average
political risk score of 20.1) and 50% to China (political risk score of 36.59). Our trade measure
would then yield a value of 28.35. If an industry sets up foreign affiliates in Canada and reduces
its exports to 20%, our measure would be equal to 22.31. Therefore, any significant effect of
political risk transmission through exports is captured with a measure that is potentially biased
downward.
26
The index of capital controls is calculated as one minus investability, which is the ratio of the market capitalization
of the IFC Investible index over the market capitalization of the IFC Global index (see Edison and Warnock (2003)
for further details).
27
This variable is taken from the database on trade agreements, www.export.gov.
20
Nevertheless, we explicitly account for activities of foreign subsidiaries by adding sales of
foreign subsidiaries of multinational corporations to export flows. The data are obtained from the
Bureau of Economic Analysis annual survey of U.S. Direct Investment Abroad. The results are
presented in specifications 3 (for trade-political risk index) and 4 (for trade-elections index) in
Tables II-VI. The regression coefficients for the interactions remain statistically significant in all
specifications. In case of ROA, inclusion of MNC sales makes regression coefficients significant.
4.4 Ex Ante versus Ex Post Political Risk
As stated in the introduction, when investing into export channels to politically-risky
countries, industries are expected to exercise caution, and invest less than they would when
exporting to a politically-safe country. Therefore, if in a given year no politically-induced trade
disruptions occur, under-investment is likely to take place. However, if political risk is realized in
a given year, over-investment takes place, as expected income from export operations is
compromised.
In order to directly test this hypothesis, for every sample year and trading-partner country, we
identify countries with government crises using the variable Major Government Crises from the
Cross-National Time Series Data Archive (Banks (2010)). It is defined as “any rapidly developing
situation that threatens to bring the downfall of the present regime”. For every year in the sample,
we identify U.S. industries with a crisis in at least one trading-partner country (63% of the
sample). We then compare marginal q across the subsamples. For the subsample with no crises,
marginal q is 1.2, indicating under-investment. For the crises subsample, marginal q is 0.78,
indicating over-investment. The difference is statistically significant with a p-value = 0.00. This is
consistent with our expectations.
Notice that most industry-years have at least one trading partner with a crisis. Therefore, we
reconstruct our trade-political interactions using a crisis dummy variable, rather than a political
risk score. We then split the sample into industry-years of over-investment and under-investment.
We observe that the trade-crisis interaction is positive and significant for the over-investment
sample, meaning that a greater exposure to crises of trading partners results in greater overinvestment. For the under-investment sample, the interaction is positive, indicating less underinvestment, but statistically insignificant.
21
4.5 Placebo Tests: Calculation of Performance Measures Using Firms without Foreign
Operations
For the main analysis, we compute performance measures and marginal q using all firms in
COMPUSTAT, some of which may have very little exposure to foreign markets. There are three
reasons for this. First, the optimal level of marginal q is determined by aggregate industry
characteristics (e.g., product demand), both domestic and foreign. Second, since marginal q is
calculated as a regression coefficient, its efficiency is greater if we include both types of firms in
the calculation. Finally, it is challenging to clearly identify firms with only domestic operations as
COMPUSTAT does not report such data. While we control for a number of industry
characteristics, a possibility remains that our results are driven by some unobserved differences.
For this purpose, we run a series of ‘placebo’ tests by constructing marginal q and
performance measures using the sample of firms with domestic-only operations. We expect to
observe no relation between the squared deviation of marginal q from its optimal level and tradepolitical risk indexes for domestic-only firms. Similarly, no relationship is expected for
performance variables. To identify (albeit imperfectly) such firms, we design a novel two-step
algorithm combining reported data in COMPUSTAT and a textual analysis of firm 10-K annual
statements. First, we drop companies with missing entries for the “exchange rate effect,” “foreign
currency adjustment,” and “foreign income taxes” items in COMPUSTAT. Then, for the
remaining firms, we manually download 10-K statements using the SEC EDGAR system. We
then use DICTION language recognition software and search for words identifying every country
in the world.28 These procedures eliminate 78% of companies as the ones with foreign operations.
We consider the remaining firms as domestic-only.
We then recalculate marginal q based on this sample and repeat regressions in Tables II, III,
and V (specifications 1 and 2). The analysis includes all previously reported control variables. The
results are presented in Panel A of Table VII. It is evident that the observed relation between
marginal q and trade-political risk and trade-election terms disappears in 9 out of 10
specifications. We conclude that our results are not driven by unobserved industry characteristics
of firms with no foreign operations.
28
A similar procedure was employed by Garcia and Norli (2012) to identify firm operations in different states in the
U.S.
22
Next, we retain the sample of firms with foreign operations and recalculate marginal q. While
marginal q is now based on fewer observations, we expect trade-political risk variables to remain
significant. This is what we observe in Panel B of Table VII. The magnitude of the coefficients
and their significance remains the same in some, and becomes stronger in other specifications in
Panel B.
Finally, we conduct regressions using the difference approach. Specifically, we construct each
variable (except for the trade-political risk measures) using the samples of firms with domesticonly and foreign operations, separately, and take the difference between the two for every industry
and year. This approach eliminates industry characteristics we fail to measure and includes those
that are common for firms with foreign and domestic operations. We then regress the difference in
marginal q on the trade-political risk and trade-elections indexes and differenced control variables.
Panel C of Table VII confirms that as political risk measures increase, the quality of capital
allocation of firms with foreign operations relative to firms with domestic operations becomes
worse.
4.6 Additional Robustness Checks
Regularly scheduled vs. early elections. Elections can be endogenous with respect to economic
performance, especially where there is an option to call early elections. To address this issue, we
subdivide the trade-partner countries in our sample into the ones with fixed and flexible election
timing (i.e. where a chief executive has an option of calling elections ahead of the regularly
scheduled date). A variety of data sources (described in Table A.3) was employed. In countries
with flexible election timing we also identify “early” elections as the ones held more than three
months ahead of the scheduled date (Bialkowski et al. (2008) employ a similar classification). Out
of 183 elections in the sample, 115 are held under flexible electoral systems. Furthermore, 63
elections are classified as “early”. We employ the Wald test of regression coefficient equivalence
between different sets of data and confirm that our results are qualitatively the same regardless of
election classification.
Inter-relatedness between political, economic, and institutional risk. One may argue that political
risk is essentially determined by economic and financial conditions, and it is difficult to isolate the
23
political risk influence. To address this, we perform a two-stage estimation. In the first stage, we
regress political risk scores on economic and institutional risks, collecting the residual
(unexplained) parts. In the second stage, we use the unexplained component of political risk. The
coefficients on trade-political risk interaction remain qualitatively unchanged.
Delayed reaction to investment. It is possible that the benefits from investment in export channel
expansion could be delayed, rather than having an immediate impact on a firm’s value. In other
words, an increase in value may not be realized until a number of years after the investment has
taken place. In order to address such a possibility, when computing marginal q, we lag the change
in value by one, two, three, and four years. Our results are robust to measuring value change with
such lags.
Over/under-investment. The main analysis relies on the magnitude, rather than the direction of
deviation of marginal q from its optimal level. We have shown that larger deviation is associated
with lower subsequent TFP growth. When we split the sample according to industries that overinvest (those with marginal q lower than 1) and industries that under-invest (those with marginal q
larger than 1), we find that the drop in TFP growth is observed for either type of industries.
Therefore, both types of capital misallocation (under-investment and over-investment) triggered
by political risk reduce industry growth in TFP.
Exports to Canada and Mexico. Not surprisingly, Canada and Mexico are dominating trading
partners for virtually all industries in our sample. Our results remain robust after we exclude
Canada and Mexico from the trade-political risk and trade-elections indexes.
Controlling for competition. As an additional control variable, we include competition within an
industry (measured by the Herfindahl index). Firms in competitive industries are likely to invest
more efficiently. At the same time, competition can affect firms’ incentives to export. The
coefficients on the main independent variables remain qualitatively unchanged when the
Herfindahl index is included in the regressions.
5. Conclusion
24
We argue that U.S. industries that export to politically-risky countries exhibit worse performance
and allocate capital less efficiently. The volume of exports to a particular country acts as a conduit
of transmission of that country’s political risk into performance through the quality of domestic
capital allocation. We show that even modest changes in a trade-partner country political
environment can have detrimental consequences for domestic industries.
For our empirical analysis, we construct a measure of foreign political risk sensitivity, which
is essentially an index of political risks of trading partners or occurrence of national elections,
weighted by relative export volumes of particular industries. We show that industries with greater
sensitivities to political risk experience significantly worse capital allocation, lower TFP growth,
lower accounting performance, and valuation.
Our findings are consistent with the notion that industries exercise caution when exporting to
politically-risky countries. This results in under-investment when such risks are not realized, and
over-investment otherwise. We control for a number of factors to ensure that our results are not
driven by omitted variables, such as the presence of MNCs and economic uncertainty. In addition,
we use the two-stage estimation to show that the results are not driven by endogeneity of export
flows.
By applying a novel methodological approach to political risk transmission analysis, this paper
makes a contribution to our understanding of capital allocation by showing that export exposure
could be detrimental to industries in the long-run. Along with gains from export activity, such as
diversification, managers and policymakers should also consider the drawback of capital
misallocations and slower growth.
Clearly, this paper is not the final word on the subject. Our measure of political risk
transmission captures only one dimension of risk, and future research may build upon our results
to capture alternative transmission mechanisms. Moreover, one may develop a unified theoretical
model that endogenizes political risk, trade policy, and performance.
25
References
Baier, S., and J. Bergstrand, 2007, Do free trade agreements actually increase members'
international trade? Journal of International Economics 71, 72-95.
Banks, A., 2010. Cross-national Time-series Data Archive. (Electronic resource).
Bartelsman, E., and W. Gray,1996, The NBER manufacturing productivity database, NBER
manual.
Beck, T., G. Clarke, A. Groff, P. Keefer and P. Walsh, 2010, New tools in comparative political
economy: The database of political institutions. World Bank Economic Review 15, 165-176.
Bialkowski, J., K. Gottschalk, and T. Wisniewski, 2008, Stock market volatility around national
elections, Journal of Banking and Finance 32, 1941–1953.
Bloom, N., 2009, The impact of uncertainty shocks, Econometrica 77, 623685.
Bloom, N., S. Bond, and J. Van Reenen, 2007, Uncertainty and investment dynamics, Review of
Economic Studies 74, 391415.
Boutchkova, M., A. Durnev, H. Doshi, and A. Molchanov, 2012, Precarious politics and return
volatility, Review of Financial Studies 25, 1111-1154.
Desai, M., C. Foley, and J. Hines, 2004, A multinational perspective on capital structure choice
and internal capital markets, Journal of Finance 59, 2451-2487.
Desai, M., C. Foley, and J. Hines, 2008, Capital structure with risky foreign investment, Journal
of Financial Economics 88, 534-553.
Di Giovanni, J., and A. Levchenko, 2009, Trade openness and volatility, Review of Economics
and Statistics 91, 558-585.
Durnev, A., R. Morck, and B. Yeung, 2004, Value-enhancing capital budgeting and firm-specific
return variation, Journal of Finance 59, 65-105.
The Economist, 2011, Business people need to think harder about political risk, February 12, p.
75.
Edison, H., and F. Warnock, 2003, A simple measure of intensity of capital controls, Journal of
Empirical Finance 10, 81-103.
Egger, H., P. Egger, and D. Greenaway, 2008, The trade structure effects of endogenous regional
trade agreements, Journal of International Economics 74, 278-298.
Erb, C., C. Harvey, and T. Viskanta, 1996, Political risk, economic risk, and financial risk,
Financial Analysts Journal 52, 29-46.
Faccio, M., M-T. Marchica, and R. Mura, 2011, CEO gender, corporate risk-taking, and the
efficiency of capital allocation, working paper, Purdue University.
Ferreira, M., and P. Laux, 2007, Corporate governance, idiosyncratic risk, and information flow,
Journal of Finance 52, 951-989.
Garcia, D., and O. Norli, 2012, Geographic dispersion and stock returns, Journal of Financial
Economics, forthcoming.
Gillespie, K., 1989, Political risk implications for exporters, contractors, and foreign licensors: the
Iranian experience, Management International Review 29, 41-52.
Greene, W., A. Hornstein, and L. White, 2009, Multinationals do it better: Evidence on the
efficiency of corporations’ capital budgeting, Journal of Empirical Finance 16, 703-720.
26
Handley, K., and N. Limao, 2012, Trade and investment under policy uncertainty: Theory and
firm evidence, working paper, University of Michigan.
Hansen, L., 1982, Large sample properties of generalized method of moments estimators,
Econometrica 50, 1029-1054.
Henisz, W., 2000, The institutional environment for multinational investment, Journal of Law,
Economics, and Organization 16, 334-364.
Hirsch, S., and B. Lev, 1971, Sales stabilization through trade diversification, Review of
Economics and Statistics 53, 270-77.
Hornstein, A., and M. Zhao, 2011, Corporate capital budgeting decisions and information sharing,
Journal of Economics and Management Strategy 20, 1135-1170.
Jensen, M., 1986, Agency costs of free cash flows, corporate finance and takeovers. American
Economic Review 76, 323-329.
Julio, B., and Y. Yook, 2012a, Political uncertainty and corporate investment cycles, Journal of
Finance 67, 45-84.
Julio, B., and Y. Yook, 2012b, Policy uncertainty, irreversibility, and cross-border flows of
capital, working paper, London Business School.
Magee, C., 2003, Endogenous preferential trade agreements: An empirical analysis, Contributions
to Economic Analysis and Policy 2, Article 15.
Mitra, D., D. Thomakos, and M. Ulubasoglu, 2002, Protection for sale in a developing country:
Democracy vs. dictatorship, Review of Economics and Statistics 84, 497-508.
Myers, S., 1977, Determinants of corporate borrowing, Journal of Financial Economics 5, 147175.
Newbery, D., and J. Stiglitz, 1984, Pareto inferior trade, Review of Economic Studies 51, 1-12.
Pástor, L., and P. Veronesi, 2012a, Uncertainty about government policy and stock prices, Journal
of Finance, forthcoming.
Pástor, L., and P. Veronesi, 2012b, Political uncertainty and risk premia, working paper,
University of Chicago.
Rajan, R., H. Servaes, and L. Zingales, 2000, The cost of diversity: The diversification discount
and inefficient investment, Journal of Finance 55, 35-80.
Stapenhurst, F., 2002. Political Risk Assessment around the North Atlantic. (St. Martin’s Press,
New York, NY).
Stein, J., 1997, Internal capital markets and the competition for corporate resources, Journal of
Finance 52, 111-133.
Wurgler, J., 2000, Financial markets and the allocation of capital, Journal of Financial Economics
58. 187-214.
27
Table I
Panel A: Descriptive Statistics of Trade Exposure, Trade-political Risk Index, and Trade-elections Index by
Industry.
This table reports 2-digit SIC industry average values of trade exposure, trade-political risk index, and trade-elections index for U.S.
manufacturing industries. The average values are calculated across the panel of 1,399 three-digit SIC manufacturing industry-year
observations. Trade exposure is calculated as the ratio of the value of exports to the value of production in an industry. The sample
years are from 1990 through 2006 (excluding 1998). Trade-political risk index is the weighted average of trading-partner countries
political risk with weights equal to trade exposure. Trade-elections index is the weighted average of trading-partner countries
national election indicators with weights equal to trade exposure.
industry
Food Products
Tobacco Products
Textile Mill Products
Apparel
Lumber And Wood Products
Furniture And Fixtures
Paper And Allied Products
Printing And Publishing
Chemicals And Allied Products
Petroleum Refining
Rubber And Plastics Products
Leather And Leather Products
Stone, Clay, And Glass
Primary Metal Industries
Fabricated Metal Products
Industrial And Computer Equipment
Electronic And Electrical Equipment
Transportation Equipment
Measuring Instruments
Miscellaneous Industries
Average
sic
code
2000
2100
2200
2300
2400
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
3600
3700
3800
3900
28
trade
exposure ,
%
17.443
6.972
24.816
23.777
24.130
12.116
20.536
22.369
20.939
4.363
22.568
10.643
6.651
18.193
11.376
32.502
31.769
28.150
13.790
21.856
18.748
tradepolitical risk
index
3.900
1.053
4.600
7.916
5.047
1.858
5.743
5.303
4.680
1.595
5.127
2.282
0.549
3.526
2.379
6.943
5.440
6.514
3.570
4.678
4.135
tradeelections
index
0.106
0.083
0.100
0.121
0.081
0.084
0.122
0.109
0.089
0.123
0.098
0.069
0.062
0.086
0.083
0.092
0.111
0.104
0.089
0.107
0.096
Table I
Panel B: Descriptive Statistics of Capital Allocation and Performance Measures by Industry.
This table reports two-digit SIC industry average values of the number of firms, marginal q, squared deviation of marginal q from 1,
industry valuation, accounting performance (ROA), and total factor productivity (TFP) growth for U.S. manufacturing industries. The
average values are calculated across the panel of 1,399 three-digit SIC manufacturing industry-year observations. The sample years
are from 1990 through 2006 (excluding 1998). Marginal q is calculated as the elasticity of investment with respect to change in firm
value as in Durnev et al. (2004). The squared deviation of marginal q from 1 is the measure of capital allocation quality with larger
values indicating worse capital allocation.
industry name
Food Products
Tobacco Products
Textile Mill Products
Apparel
Lumber And Wood Products
Furniture And Fixtures
Paper And Allied Products
Printing And Publishing
Chemicals And Allied Products
Petroleum Refining
Rubber And Plastics Products
Leather And Leather Products
Stone, Clay, And Glass
Primary Metal Industries
Fabricated Metal Products
Industrial And Computer Equipment
Electronic And Electrical Equipment
Transportation Equipment
Measuring Instruments
Miscellaneous Industries
Average
Total
sic
code
2000
2100
2200
2300
2400
2500
2600
2700
2800
2900
3000
3100
3200
3300
3400
3500
3600
3700
3800
3900
number
of firms
145
10
30
63
32
37
71
85
517
46
72
21
43
105
83
382
539
131
388
79
144
2,879
marginal
q
1.241
1.480
1.018
1.690
0.692
1.108
0.740
1.738
1.245
1.174
0.422
0.580
0.700
0.611
0.319
0.130
0.309
2.200
0.530
1.517
0.972
29
(marginal q – 1)2
0.058
0.230
0.000
0.476
0.095
0.012
0.068
0.545
0.060
0.030
0.334
0.176
0.090
0.151
0.464
0.757
0.477
1.440
0.221
0.267
0.298
valuation,
q
2.492
4.398
1.485
1.495
1.919
1.666
1.868
2.212
4.052
5.539
1.374
1.249
1.844
1.611
1.209
3.126
3.578
1.631
3.230
1.560
2.377
accounting
performance
, ROA, %
4.113%
13.167%
-1.367%
5.281%
4.253%
5.203%
3.261%
3.439%
-4.981%
8.054%
2.350%
5.490%
3.896%
1.656%
3.063%
2.383%
2.533%
3.965%
1.690%
1.178%
3.430
TFP growth,
%
0.865%
4.850%
4.513%
2.384%
1.331%
2.685%
-0.294%
1.844%
0.738%
6.140%
0.547%
3.491%
1.916%
-0.469%
0.095%
2.769%
2.922%
1.012%
2.594%
1.672%
2.080
Table II
The Effect of Political Risk on Performance of Trade-dependent Industries.
This table reports the results of OLS panel regressions of growth in Total Factor Productivity (TFP) growth (Panel A), valuation (Panel B),
and return on assets (Panel C) on trade-political risk index, trade-elections index, interactions of trade exposure with economic risk,
institutional risk, U.S. political risk, and U.S. elections. Additional variables are: trade exposure, average levels of political risk, economic
risk, and institutional risk of trading partners, size, leverage, and R&D expenditures. The regressions are run using the panel of three-digit
SIC U.S. manufacturing industries spanning years from 1990 through 2006 (excluding 1998). Every independent variable is lagged by one
year. In specifications 3 and 4, we add sales of multinational corporations to industry exports. Every regression includes industry and year
fixed effects. Numbers in parentheses are probability levels at which the hypothesis of zero coefficient can be rejected. The coefficients
significant at the 10% level (based on a two-tailed test) or higher are in bold face. *, **, *** indicate significance at the 10%, 5%, and 1%
levels, respectively. Standard errors are clustered at the industry level to adjust them for heteroskedasticity and time-series correlation.
Panel A
DEPENDENT VARIABLE
SPECIFICATION
Trade-political risk index
Trade-elections index
Interaction of trade with economic risk
Interaction of trade with institutional risk
Interaction of trade with U.S. political risk
Interaction of trade with U.S. elections
Trade exposure
Political risk
Economic risk
Institutional risk
Size
Leverage
R&D expenditures
Regression R2 –adj.
Number of observations
TFP growth
1
without MNC
-0.020***
(0.00)
-0.011**
(0.04)
-0.002
(0.19)
0.013
(0.14)
0.068
(0.22)
-0.001
(0.16)
-0.003***
(0.00)
0.003
(0.23)
-0.806***
(0.00)
-0.928***
(0.00)
4.611***
(0.00)
0.328
1,399
30
TFP growth
2
without MNC
-1.014**
(0.03)
-0.007
(0.16)
-0.001
(0.12)
-0.961
(0.34)
0.021
(0.11)
0.002*
(0.10)
-0.003***
(0.01)
0.000
(0.48)
-0.851***
(0.00)
-0.911***
(0.00)
2.300***
(0.00)
0.330
1,399
TFP growth
3
with MNC
-0.016***
(0.00)
-0.009**
(0.05)
-0.004
(0.14)
-0.011
(0.18)
0.068
(0.22)
0.001
(0.15)
-0.002***
(0.00)
0.003
(0.21)
-0.718***
(0.00)
-0.921***
(0.00)
3.827***
(0.00)
0.336
1,399
TFP growth
4
with MNC
-0.920**
(0.02)
-0.004
(0.18)
-0.006
(0.16)
-0.960
(0.32)
0.028
(0.11)
0.001
(0.12)
-0.003***
(0.00)
0.004
(0.14)
-0.771***
(0.00)
-0.904***
(0.00)
2.116***
(0.00)
0.332
1,399
Panel B
DEPENDENT VARIABLE
SPECIFICATION
Trade-political risk index
Trade-elections index
Interaction of trade with economic risk
Interaction of trade with institutional risk
Interaction of trade with U.S. political risk
Interaction of trade with U.S. elections
Trade exposure
Political risk
Economic risk
Institutional risk
Size
Leverage
R&D expenditures
Regression R2 –adj.
Number of observations
valuation
1
without MNC
-0.236***
(0.00)
-0.186*
(0.10)
-0.003
(0.24)
0.014
(0.20)
0.250**
(0.05)
-0.003
(0.11)
-0.035***
(0.00)
0.002
(0.18)
-0.349***
(0.00)
-0.129***
(0.0)
7.349***
(0.00)
0.321
1,399
31
valuation
2
without MNC
-8.143***
(0.00)
-0.116*
(0.10)
-0.002
(0.23)
-0.306
(0.12)
0.224*
(0.10)
0.002
(0.15)
-0.035***
(0.00)
0.002
(0.20)
-0.834***
(0.00)
-0.140***
(0.00)
7.100***
(0.00)
0.334
1,399
valuation
3
with MNC
-0.482***
(0.00)
-0.127***
(0.01)
-0.002
(0.12)
-0.010
(0.44)
0.328*
(0.10)
0.000
(0.12)
-0.036***
(0.00)
0.002
(0.14)
-0.917***
(0.00)
-0.132***
(0.00)
7.912***
(0.00)
0.380
1,399
valuation
4
with MNC
-8.759**
(0.01)
-0.014
(0.20)
-0.003
(0.15)
-0.280
(0.21)
0.334*
(0.08)
0.000
(0.16)
-0.038***
(0.00)
0.003
(0.17)
-0.662***
(0.00)
-0.134***
(0.00)
7.494***
(0.00)
0.330
1,399
Panel C
DEPENDENT VARIABLE
SPECIFICATION
Trade-political risk index
Trade-elections index
Interaction of trade with economic risk
Interaction of trade with institutional risk
Interaction of trade with U.S. political risk
Interaction of trade with U.S. elections
Trade exposure
Political risk
Economic risk
Institutional risk
Size
Leverage
R&D expenditures
Regression R2 –adj.
Number of observations
ROA
1
without MNC
ROA
2
without MNC
-0.021
(0.27)
-
-
-0.011**
(0.04)
-0.002
(0.19)
0.013
(0.14)
0.068
(0.22)
-0.001
(0.16)
-0.003***
(0.00)
0.003
(0.23)
-0.806***
(0.00)
-0.928
(0.27)
4.611***
(0.00)
0.216
1,399
32
-2.270
(0.23)
-0.007
(0.16)
-0.001
(0.12)
-0.961
(0.34)
0.021
(0.11)
0.002*
(0.10)
-0.003***
(0.01)
0.000
(0.48)
-0.851***
(0.00)
-0.911
(0.24)
2.300***
(0.00)
0.220
1,399
ROA
3
with MNC
-0.026***
(0.00)
-0.009**
(0.05)
-0.004
(0.14)
-0.011
(0.18)
0.068
(0.22)
0.001
(0.15)
-0.002***
(0.00)
0.003
(0.21)
-0.718***
(0.00)
-0.921
(0.30)
3.827***
(0.00)
0.225
1,399
ROA
4
with MNC
-2.300**
(0.00)
-0.004
(0.18)
-0.006
(0.16)
-0.960
(0.32)
0.028
(0.11)
0.001
(0.12)
-0.003***
(0.00)
0.004
(0.14)
-0.771***
(0.00)
-0.904
(0.32)
2.116***
(0.00)
0.218
1,399
Table III
The Effect of Political Risk on Capital Allocation Quality of Trade-dependent Industries. OLS
estimation.
This table reports the results of OLS panel regressions of the measures of capital allocation quality (squared deviation of marginal q
from 1) on trade-political risk index, trade-elections index, interactions of trade exposure with economic risk, institutional risk, U.S.
political risk, and U.S. elections. Additional variables are: trade exposure, average levels of political risk, economic risk, and
institutional risk of trading partners, scaled firm-specific return variation, average q, diversification, size, liquidity, leverage, and
R&D expenditures. The regressions are run using the panel of three-digit SIC U.S. manufacturing industries spanning years from
1990 through 2006 (excluding 1998). Every independent variable is lagged by one year. In specifications 3 and 4, we add sales of
multinational corporations to industry exports. Every regression includes industry and year fixed effects. Numbers in parentheses are
probability levels at which the hypothesis of zero coefficient can be rejected. The coefficients significant at the 10% level (based on a
two-tailed test) or higher are in bold face. *, **, *** indicate significance at the 10%, 5%, and 1% levels, respectively. Standard
errors are clustered at the industry level to adjust them for heteroskedasticity and time-series correlation.
DEPENDENT VARIABLE
SPECIFICATION
Trade-political risk index
Trade-elections index
Interaction of trade with economic risk
Interaction of trade with institutional risk
Interaction of trade with U.S. political risk
Interaction of trade with U.S. elections
Trade exposure
Political risk
Economic risk
Institutional risk
Scaled firm-specific return variation
Average q
Diversification
Size
Liquidity
Leverage
R&D expenditures
Regression R2 – adj.
Regression
R2 –adj.
Number
of observations
(MQ-1)2
1
without
MNC
0.032***
(0.00)
0.016**
(0.05)
0.010
(0.21)
0.023*
(0.10)
-0.130
(0.17)
0.004*
(0.07)
0.006***
(0.00)
-0.005
(0.18)
-0.013**
(0.03)
-0.254
(0.12)
-0.414
(0.12)
-0.575***
(0.00)
-0.980
(0.29)
-1.336***
(0.00)
-1.602***
(0.00)
0.441
00.441
1,399
33
(MQ-1)2
2
without MNC
5.106***
(0.00)
0.011
(0.19)
-0.012
(0.28)
-1.150
(0.30)
-0.114
(0.11)
0.005**
(0.04)
0.006***
(0.01)
-0.004
(0.26)
-0.013
(0.12)
-0.377
(0.14)
-0.390
(0.22)
-0.667***
(0.00)
-1.061
(0.38)
-1.416***
(0.00)
-1.576***
(0.00)
0.416
0.416
1,399
(MQ-1)2
3
with MNC
0.044***
(0.00)
0.018**
(0.06)
0.010
(0.24)
0.026*
(0.10)
-0.126
(0.12)
0.004**
(0.05)
0.007***
(0.00)
-0.001
(0.13)
-0.019***
(0.01)
-0.360
(0.11)
-0.408
(0.14)
-0.551***
(0.00)
-0.971
(0.34)
-1.316***
(0.00)
-1.599***
(0.00)
0.438
0.438
1,399
(MQ-1)2
4
with MNC
5.118***
(0.00)
0.010
(0.22)
-0.013
(0.21)
-1.142
(0.23)
-0.155
(0.15)
0.004**
(0.03)
0.007***
(0.01)
-0.001
(0.16)
-0.010
(0.12)
-0.377
(0.17)
-0.320
(0.15)
-0.638***
(0.00)
-1.049
(0.31)
-1.395***
(0.00)
-1.546***
(0.00)
0.420
1,399
Table IV
Performance and Capital Allocation Quality Decomposition.
This table reports the results of OLS panel regressions of TFP growth (panel A), valuation (panel B) and return on assets (panel C) on
explained and unexplained levels of the squared deviation of marginal q from 1. To obtain the explained and unexplained parts, we first
regress the squared deviation of marginal q from 1 on all variables as in Table II. The regressions are run using the panel of three-digit
SIC U.S. manufacturing industries spanning years from 1990 through 2006 (excluding 1998). Every independent variable is lagged by
one year. In specifications 3 and 4, we add sales of multinational corporations to industry exports. Every regression includes industry
and year fixed effects. Numbers in parentheses are probability levels at which the hypothesis of zero coefficient can be rejected. The
coefficients significant at the 10% level (based on a two-tailed test) or higher are in bold face. *, **, *** indicate significance at the
10%, 5%, and 1% levels, respectively. Standard errors are clustered at the industry level to adjust them for heteroskedasticity and timeseries correlation.
Panel A
DEPENDENT VARIABLE
SPECIFICATION
Explained values of deviation from optimal
capital allocation
Residual values of deviation from optimal capital
allocation
Regression R2 –adj.
Number of observations
TFP growth
TFP growth
TFP growth
TFP growth
1
without MNC
-0.315***
(0.00)
2
without MNC
-0.380***
(0.00)
3
with MNC
-0.321***
(0.00)
4
with MNC
-0.331***
(0.00)
-0.212
(0.65)
0.210
1,399
-0.210
(0.50)
0.208
1,399
-0.119
(0.46)
0.261
1,399
-0.102
(0.52)
0.264
1,399
valuation
valuation
valuation
valuation
1
without MNC
-2.104**
(0.05)
2
without MNC
-2.368**
(0.02)
3
with MNC
-2.907*
(0.06)
4
with MNC
-2.000
(0.13)
0.083
(0.66)
0.217
1,399
0.020
(0.52)
0.221
1,399
0.032
(0.46)
0.224
1,399
0.020
(0.59)
0.233
1,399
ROA
ROA
ROA
ROA
1
without MNC
-2.077***
(0.00)
2
without MNC
-1.714***
(0.00)
3
with MNC
-1.293***
(0.00)
4
with MNC
-1.016***
(0.00)
0.309
(0.66)
0.162
1,399
0.202
(0.52)
0.180
1,399
0.183
(0.46)
0.189
1,399
0.114
(0.59)
0.215
1,399
Panel B
DEPENDENT VARIABLE
SPECIFICATION
Explained values of deviation from optimal
capital allocation
Residual values of deviation from optimal capital
allocation
Regression R2 –adj.
Number of observations
Panel C
DEPENDENT VARIABLE
SPECIFICATION
Explained values of deviation from optimal
capital allocation
Residual values of deviation from optimal capital
allocation
Regression R2 –adj.
Number of observations
34
Table V
The Effect of Political Risk on Capital Allocation Quality of Trade-dependent Industries. Non-linear LS
Estimation Using Endogenously Determined Optimal Level of Marginal q.
This table reports the results of non-linear LS panel regressions of the measures of capital allocation quality (squared deviation of
marginal q from endogenously estimated optimal level h) on trade-political risk index, trade-elections index, interactions of trade
exposure with economic risk, institutional risk, U.S. political risk, and U.S. elections. Additional variables are: trade exposure, average
levels of political risk, economic risk, and institutional risk of trading partners, scaled firm-specific return variation, average q,
diversification, size, liquidity, leverage, and R&D expenditures. The regressions are run using the panel of three-digit SIC U.S.
manufacturing industries spanning years from 1990 through 2006 (excluding 1998). Every independent variable is lagged by one year.
In specifications 3 and 4, we add sales of multinational corporations to industry exports. Every regression includes industry and year
fixed effects. Numbers in parentheses are probability levels at which the hypothesis of zero coefficient can be rejected. The coefficients
significant at the 10% level (based on a two-tailed test) or higher are in bold face. *, **, *** indicate significance at the 10%, 5%, and
1% levels, respectively. Standard errors are clustered at the industry level to adjust them for heteroskedasticity and time-series
correlation.
DEPENDENT VARIABLE
SPECIFICATION
Endogenously estimated optimal level h
Trade-political risk index
Trade-elections index
Interaction of trade with economic risk
Interaction of trade with institutional risk
Interaction of trade with U.S. political risk
Interaction of trade with U.S. elections
Trade exposure
Political risk
Economic risk
Institutional risk
Scaled firm-specific return variation
Average q
Diversification
Size
Liquidity
Leverage
R&D expenditures
Regression R2 –adj.
Number of observations
(MQ-h)2
1
without MNC
0.812***
(0.00)
0.020***
(0.00)
0.011
(0.15)
0.002
(0.21)
0.020
(0.20)
-0.230
(0.25)
0.010**
(0.03)
0.009***
(0.00)
-0.005
(0.17)
-0.031***
(0.00)
-0.255
(0.29)
-0.434
(0.14)
-0.316***
(0.00)
-1.113**
(0.05)
-1.000***
(0.00)
-2.075***
(0.00)
0.518
1,399
35
(MQ-h)2
2
without MNC
0.836***
(0.00)
3.225***
(0.00)
0.014
(0.12)
0.001
(0.21)
1.014
(0.14)
-0.190
(0.13)
0.007**
(0.05)
0.009***
(0.00)
-0.005
(0.20)
-0.020*
(0.10)
-0.132
(0.32)
-0.396
(0.28)
-0.497***
(0.00)
-1.211**
(0.03)
-1.214***
(0.00)
-2.404***
(0.00)
0.496
1,399
(MQ-h)2
3
with MNC
0.815***
(0.00)
0.022***
(0.00)
0.014
(0.13)
0.005
(0.24)
0.015
(0.21)
-0.211
(0.14)
0.001
(0.13)
0.007***
(0.00)
-0.005
(0.21)
-0.032***
(0.00)
-0.251
(0.29)
-0.414
(0.16)
-0.572***
(0.00)
-1.239**
(0.05)
-1.047***
(0.00)
-2.073***
(0.00)
0.531
1,399
(MQ-h)2
4
with MNC
0.780***
(0.00)
3.220***
(0.00)
0.016*
(0.10)
0.001
(0.22)
1.016
(0.21)
-0.198
(0.12)
0.003**
(0.05)
0.009***
(0.00)
-0.004
(0.14)
-0.022*
(0.10)
-0.120
(0.30)
-0.316
(0.20)
-0.511***
(0.00)
-1.269**
(0.04)
-1.292***
(0.00)
-2.869***
(0.00)
0.502
1,399
Table VI
The Effect of Political Risk on Performance of Trade-dependent Industries. Instrumental Variable
Estimation.
This table reports the results of the IV panel regressions of the measures of growth in Total Factor Productivity (TFP) growth (Panel
A), valuation (Panel B), and return on assets (Panel C) on trade-political risk index, trade-elections index, interactions of trade
exposure with economic risk, institutional risk, U.S. political risk, and U.S. elections. Additional variables are: trade exposure, average
levels of political risk, economic risk, and institutional risk of trading partners, size, leverage, and R&D expenditures. Unlike in Table
III, the regressions are estimated using the IV approach. The variables for the first stage are: log of distance between U.S. and a
trading-partner country, industry size in the U.S., trading-partner industry size, trading-partner import dependence, trading-partner
index of capital controls, and a dummy variable for bi-lateral trade agreements. The regressions are run using the panel of three-digit
SIC U.S. manufacturing industries spanning years from 1990 through 2006 (excluding 1998). Every independent variable is lagged by
one year. In specifications 3 and 4, we add sales of multinational corporations to industry exports. Every regression includes industry
and year fixed effects. Numbers in parentheses are probability levels at which the hypothesis of zero coefficient can be rejected. The
coefficients significant at the 10% level (based on a two-tailed test) or higher are in bold face. *, **, *** indicate significance at the
10%, 5%, and 1% levels, respectively. Standard errors are clustered at the industry level to adjust them for heteroskedasticity and timeseries correlation.
Panel A
DEPENDENT VARIABLE
SPECIFICATION
Trade-political risk index
Trade-elections index
Interaction of trade with economic risk
Interaction of trade with institutional risk
Interaction of trade with U.S. political risk
Interaction of trade with U.S. elections
Trade exposure
Political risk
Economic risk
Institutional risk
Size
Leverage
R&D expenditures
Regression R2 –adj.
Number of observations
TFP growth
1
without MNC
-0.016***
(0.00)
-0.016**
(0.04)
-0.001
(0.24)
0.016*
(0.10)
0.067
(0.25)
-0.001
(0.20)
-0.003***
(0.00)
0.004
(0.28)
-0.807***
(0.00)
-0.936***
(0.00)
4.620***
(0.00)
0.380
1,399
36
TFP growth
2
without MNC
-1.026**
(0.05)
-0.015
(0.20)
-0.001
(0.11)
-0.957
(0.40)
0.026
(0.16)
-0.002*
(0.10)
-0.003***
(0.01)
0.001
(0.42)
-0.856***
(0.00)
-0.921***
(0.00)
2.345***
(0.00)
0.336
1,399
TFP growth
3
with MNC
-0.018***
(0.00)
-0.019**
(0.03)
-0.006
(0.16)
-0.012
(0.11)
0.048
(0.24)
-0.001
(0.12)
-0.002***
(0.00)
0.003
(0.21)
-0.721***
(0.00)
-0.924***
(0.00)
4.850***
(0.00)
0.345
1,399
TFP growth
4
with MNC
-1.138*
(0.06)
-0.014
(0.18)
-0.006
(0.16)
-1.126
(0.33)
0.021
(0.11)
-0.001
(0.16)
-0.003***
(0.00)
0.004
(0.14)
-0.612***
(0.00)
-0.961***
(0.00)
2.222***
(0.00)
0.371
1,399
Panel B
DEPENDENT VARIABLE
SPECIFICATION
Trade-political risk index
Trade-elections index
Interaction of trade with economic risk
Interaction of trade with institutional risk
Interaction of trade with U.S. political risk
Interaction of trade with U.S. elections
Trade exposure
Political risk
Economic risk
Institutional risk
Size
Leverage
R&D expenditures
Regression R2 –adj.
Number of observations
valuation
1
without MNC
-0.245***
(0.00)
-0.186*
(0.10)
-0.003
(0.24)
0.014
(0.20)
0.250**
(0.05)
-0.003
(0.11)
-0.035***
(0.00)
0.002
(0.18)
-0.349***
(0.00)
-0.148***
(0.0)
7.349***
(0.00)
0.329
1,399
37
valuation
2
without MNC
-6.413***
(0.00)
-0.116*
(0.10)
-0.002
(0.23)
-0.306
(0.12)
0.224*
(0.10)
0.002
(0.15)
-0.045***
(0.00)
0.003
(0.20)
-0.834***
(0.00)
-0.140***
(0.00)
9.100***
(0.00)
0.316
1,399
valuation
3
with MNC
-0.311***
(0.00)
-0.127***
(0.01)
-0.002
(0.12)
-0.010
(0.44)
0.417*
(0.10)
0.001
(0.12)
-0.033***
(0.00)
0.002
(0.14)
-0.917***
(0.00)
-0.132***
(0.00)
7.949***
(0.00)
0.384
1,399
valuation
4
with MNC
-11.344**
(0.00)
-0.014
(0.20)
-0.003
(0.15)
-0.284
(0.26)
0.332*
(0.10)
0.001
(0.16)
-0.021***
(0.00)
0.014*
(0.10)
-0.677***
(0.00)
-0.148***
(0.00)
8.519***
(0.00)
0.314
1,399
Panel C
DEPENDENT VARIABLE
SPECIFICATION
Trade-political risk index
Trade-elections index
Interaction of trade with economic risk
Interaction of trade with institutional risk
Interaction of trade with U.S. political risk
Interaction of trade with U.S. elections
Trade exposure
Political risk
Economic risk
Institutional risk
Size
Leverage
R&D expenditures
Regression R2 –adj.
Number of observations
ROA
1
without MNC
ROA
2
without MNC
-0.030
(0.34)
-
-
-0.020*
(0.01)
-0.001
(0.34)
0.016
(0.12)
0.073
(0.16)
-0.001
(0.14)
-0.003***
(0.00)
0.002
(0.21)
-0.815***
(0.00)
-0.884
(0.45)
4.617***
(0.00)
0.208
1,399
38
-1.116
(0.40)
-0.007
(0.16)
-0.001
(0.12)
-0.961
(0.34)
0.021
(0.11)
0.002*
(0.10)
-0.003***
(0.01)
0.001
(0.48)
-0.836***
(0.00)
-0.911
(0.24)
2.300***
(0.00)
0.215
1,399
ROA
3
with MNC
-0.020*
(0.09)
-0.014**
(0.03)
-0.003
(0.17)
-0.015
(0.23)
0.062
(0.25)
0.001
(0.15)
-0.002***
(0.00)
0.003
(0.28)
-0.742***
(0.00)
-0.926
(0.30)
3.019***
(0.00)
0.221
1,399
ROA
4
with MNC
-2.377*
(0.10)
-0.004
(0.16)
-0.005
(0.22)
-1.121
(0.15)
0.028
(0.11)
0.001
(0.12)
-0.003***
(0.00)
0.004
(0.18)
-0.622***
(0.00)
-0.904
(0.32)
2.225***
(0.00)
0.203
1,399
Table VII
Placebo Tests: Recalculating dependent variables using firms with only domestic operations
This table reports the coefficients on trade-political risk index and trade election index from regressions in Tables II, III and V
(specifications 1 and 2). Unlike in Tables II, III and V, in Panel A, marginal q and performance variables are calculated using
companies with domestic operations only. In Panel B, marginal q and performance variables are calculated after dropping
companies with domestic operations only. In Panel C, every variable (except for the political risk indexes interacted with trade
exposure) is calculated as the difference of measures based on firms with foreign and domestic operations and firms with domestic
operations only. A company is classified as one with domestic operations if (i) the company does not report a foreign segment in
the COMPUSTAT Segments file, (ii) COMPUSTAT reports a missing entry for the “exchange rate effect,” “foreign currency
adjustment,” and “foreign income taxes” items, and (iii) a company’s 10K annual statement filed with the Securities and Exchange
Commission does not mention a country other than the U.S. The sample starts in 1994, the year with available 10-K annual
statements in electronic form. The proportion of companies with only domestic operations (relative to the total sample) is 29%.
Panel A: Dependent variables are calculated using firms with domestic operations only
TABLE
SPECIFICATION
Trade-political risk index
TABLE
SPECIFICATION
Trade-elections index
Table II
(Panel A)
Table II
(Panel B)
Table II
(Panel C)
Table III
Table V
1
0.028
(
(0.30)
1
0.114
(
(0.13)
(0.13)
1
-0.010
(0.60)
1
0.015
(0.14)
1
0.090
(0.43)
Table II
(Panel A)
2
-0.890*
(0.10)
Table II
(Panel B)
2
1.281
(0.58)
Table II
(Panel C)
2
1.117
(0.20)
Table III
Table V
2
2.128
(0.16)
2
-0.810
(0.21)
Panel B: Dependent variables are calculated after dropping firms with domestic operations only
TABLE
SPECIFICATION
Trade-political risk index
TABLE
SPECIFICATION
Trade-elections index
Table II
(Panel A)
Table II
(Panel B)
Table II
(Panel C)
Table III
Table V
1
-0.047***
(0.00)
1
-0.232***
(0.00)
1
-0.017
(0.27)
1
0.027***
(0.00)
1
0.722***
(0.00)
Table II
(Panel A)
2
-2.017**
(0.05)
Table II
(Panel B)
2
-6.228***
(0.00)
Table II
(Panel C)
2
-2.480
(0.23)
Table III
Table V
2
4.188***
(0.00)
2
3.302***
(0.00)
Panel C: Variables are expressed as difference between measures calculated using firms with domestic and foreign
operations and firms with domestic operations only
TABLE
SPECIFICATION
Trade-political risk index
TABLE
SPECIFICATION
Trade-elections index
Table II
(Panel A)
Table II
(Panel B)
Table II
(Panel C)
Table III
Table V
1
-0.041***
(0.00)
1
-0.228***
(0.00)
1
-0.004
(0.42)
1
0.024***
(0.00)
1
0.718***
(0.00)
Table II
(Panel A)
2
-2.030**
(0.05)
Table II
(Panel B)
2
-5.418***
(0.00)
Table II
(Panel C)
2
-2.016
(0.51)
Table III
Table V
2
3.222***
(0.00)
2
4.017***
(0.00)
39
Appendix: Risk Indexes of Trading Partners, Political Systems, and Electoral Timing.
Table A1: Descriptive Statistics of Trading Partners.
This table reports average values of political risk, economic risk, and institutional risk variables for trading partners. The data source is International Country Risk Guide
(ICRG). Average values are calculated using quarterly data for years from 1990 through 2005 (excluding 1998). Political risk is based on government stability (0-12 scale),
socioeconomic conditions (0-12), investment profile (0-12), internal conflict (0-12), external conflict (0-12), military in politics (0-6), religious tensions (0-6), ethnic tensions
(0-6), and democratic accountability (0-6). Economic risk is based on GDP
per capita (0-5), real GDP growth (0-10), annual inflation rate (0-10), state budget balance
(0-10), current account (0-10), state foreign debt (0-10), foreign debt service (0-10), international liquidity (0-5), and exchange rate stability (0-10). Institutional risk is based
on the rule of law (0-6) and corruption (0-6). Larger values for economics risk, institutional risk, and political risk indicate greater risks. The three indexes are brought to the
common 1-100 scale.
Country
Political
Risk
Economic
Risk
Institutiona
l Risk
Country
Political
Risk
Economic
Risk
Institutiona
l Risk
Political
Risk
Economic
Risk
Institutional
Risk
Argentina
32.314
55.017
55.785
Hungary
24.149
51.228
29.361
Poland
24.743
50.353
36.381
Australia
18.304
49.547
11.949
India
47.191
46.439
Austria
17.883
46.925
12.868
Indonesia
51.913
52.506
61.658
Portugal
17.711
51.603
24.153
74.908
Russia
43.471
48.445
66.985
Belgium
21.976
46.975
25.608
Ireland
14.817
47.466
25.098
Singapore
18.676
42.986
24.204
Brazil
36.841
53.906
65.130
Israel
46.953
49.194
36.101
South Africa
33.251
50.400
60.458
Canada
20.100
49.273
4.264
Chile
26.651
46.426
38.730
Italy
24.311
47.589
39.778
South Korea
26.574
46.545
43.786
Japan
21.294
42.100
27.063
Spain
26.227
48.803
30.944
China
36.593
52.081
51.291
Luxembourg
10.011
46.039
6.025
Sri Lanka
52.173
53.339
63.879
Colombia
Czech
Republic
50.297
54.234
83.819
Malaysia
29.126
46.728
50.578
Sweden
17.779
47.392
2.503
Denmark
24.070
49.953
33.451
Mexico
30.439
50.055
70.159
Switzerland
15.033
43.828
11.209
17.839
46.256
1.889
Morocco
34.810
48.994
41.615
Taiwan
23.709
44.570
43.123
Egypt
41.291
50.814
63.700
15.934
49.111
3.166
Thailand
35.969
48.450
55.939
Finland
13.746
46.934
0.000
Netherlands
New
Zealand
17.857
56.050
5.489
47.664
56.970
56.705
France
24.914
47.900
27.778
Norway
17.611
42.528
6.434
20.439
48.442
14.450
Germany
20.826
47.811
15.865
Pakistan
58.736
51.555
76.236
Turkey
United
Kingdom
United
States
21.897
50.534
18.178
Greece
26.424
51.106
37.429
Peru
44.360
49.850
67.709
Venezuela
47.403
50.550
62.424
Philippines
38.671
50.800
69.955
Zimbabwe
55.919
64.539
75.980
40
Country
Table A2. Political System, Party Orientation, and Elections.
This table lists the type of a political system (presidential or parliamentary), the chief executive’s party orientation during the sample period (left, right, or center), and years
of the elections of the chief executive based on the World Bank Database of Political Institutions. We cross-check the election data with data reported by International
Institute for Democracy and Electoral Assistance, Center on Democratic Performance, Journal of Democracy, Elections around the World, Election Guide, The CIA World
Factbook, the PARLINE Database on National Parliaments, and Keesing’s Record of World Events. The political system is classified as presidential when (i) the chief
executive is not elected or (ii) presidents are elected directly or by an electoral college in the event there is no prime minister. In systems with both a prime minister and a
president, exact classification depends on the veto power of the president and the power of the president to appoint a prime minister and dissolve parliament. Systems in
which the legislature elects the chief executive are classified as parliamentary. Election year is the year of presidential election for presidential systems and of parliamentary
elections for parliamentary systems. Party orientation is determined according to the party of chief executive using the following rule: right for parties that are defined as
conservative, Christian-Democratic, or right-wing; left for parties that are defined as communist, socialist, social-democratic, or left-wing; center for parties that can be best
described as centrist. “NA” appears for cases when the exact party orientation cannot be determined. Refer to Beck et al. (2001) for further details. The sample beginning
years for every country correspond to the availability of return series. Notes: *Pakistan had a parliamentary system until 1999. In 1999, the system changed to presidential
after a military coup d'état.
country
system
Argentina
Presidential
Australia
Parliamentary
Austria
Belgium
Parliamentary
Parliamentary
party type
year
1990-1995:R
1996-1999:R
2000-2001:C
2002-2003:R
2004-2006:L
1990-1993:L
1994-1996:L
1997-1998:R
1999-2001:R
2002-2004:R
2005-2006:R
1990-1994:L
1995-1995:L
1996-1999:L
2000-2002:R
2003-2006:R
1990-1991:R
1992-1995:R
1996-1999:R
2000-2003:R
2004-2006:R
1995
1999
2003
1990
1993
1996
1998
2001
2004
1990
1994
1995
1999
2002
1991
1995
1999
2003
country
system
Brazil
Presidential
Canada
Parliamentary
Chile
Presidential
China
Colombia
NA
Presidential
Czech Rep.
Parliamentary
party type
1990-1994:R
1995-1998:L
1999-2002:L
2003-2006:L
1990-1993:R
1994-1997:L
1998-2000:L
2001-2004:L
2005-2006:L
1990-1993:R
1994-2000:R
2001-2005:R
2006-2006:R
1990-2006:L
1992-1994:C
1995-1998:C
1999-2002:R
2003-2006:NA
1994-1996:R
1997-1998:R
1999-2002:L
2003-2006:L
41
year
1994
1998
2002
1993
1997
2000
2004
1993
2000
2005
1994
1998
2002
1996
1998
2002
country
system
Denmark
Parliamentary
Egypt
Parliamentary
Finland
Parliamentary
France
Parliamentary
Germany
Parliamentary
party type
1990-1994:R
1995-1998:L
1999-2001:L
2002-2005:R
2006-2006:R
1995-2000:NA
2001-2005:NA
2006-2006:NA
1990-1991:R
1992-1995:C
1996-1999:L
2000-2003:L
2004-2006:C
1990-1993:L
1994-1997:R
1998-2002:L
2003-2006:R
1990-1994:R
1995-1998:R
1999-2002:L
2003-2005:L
2006-2006:R
year
1990
1994
1998
2001
2005
1995
2000
2005
1991
1995
1999
2003
1993
1997
2002
1990
1994
1998
2002
2005
Table A2 continued
country
Greece
system
party type
year
Parliamentary
1990-1990:L
1991-1993:R
1994-1996:L
1997-2000:L
2001-2004:L
2005-2006:R
1991-1994:R
1995-1998:R
1999-2002:L
2003-2006:L
1990-1991:L
1992-1996:L
1997-1999:L
2000-2004:R
2005-2006:L
1990-1992:NA
1993-1997:NA
1998-1999:NA
2000-2004:NA
1990-1992:C
1993-1997:R
1998-2002:C
2003-2006:C
1990-1992:R
1993-1996:L
1997-1999:R
2000-2001:R
2002-2006:R
1990-1992:C
1993-1994:L
1995-1996:R
1997-2001:C
2002-2006:R
1990
1993
1996
2000
2004
1994
1998
2002
1991
1996
1999
1998
2004
1992
1997
1999
2004
1992
1997
2002
1992
1996
1999
2001
1992
1994
1996
2001
Table A1 continued.
Hungary
Parliamentary
India
Parliamentary
Indonesia
Parliamentary
Ireland
Parliamentary
Israel
Parliamentary
Italy
Parliamentary
system
party type
year
Japan
country
Parliamentar
y
Luxembourg
Parliamentar
y
Malaysia
Parliamentar
y
Mexico
Presidential
Morocco
Netherlands
NA
Parliamentar
y
New Zealand
Parliamentar
y
1990-1993:R
1994-1996:L
1997-2000:R
2001-2003:R
2004-2005:R
1991-1994:C
1995-1999:C
2000-2004:C
2005-2006:C
1990-1995:NA
1996-1999:NA
2000-2004:NA
2005-2006:NA
1990-1994:L
1995-1997:L
1998-2000:L
2001-2006:R
1993-2006:NA
1990-1994:R
1995-1998:L
1999-2002:L
2003-2003:L
2004-2006:R
1990-1993:L
1994-1996:R
1997-1999:R
2000-2002:L
2003-200r5:L
2006-2006:L
1990
1993
1996
2000
2003
1994
1999
2004
1990
1995
1999
2004
1994
1997
2000
1994
1998
2002
2003
1990
1993
1996
1999
2002
2005
42
country
system
Norway
Parliamentary
Pakistan
Parliamentary*
Peru
Presidential
Philippines
Presidential
Poland
Parliamentary
Portugal
Parliamentary
Russia
Presidential
party type
year
1990-1993:L
1994-1997:L
1998-2001:R
2002-2005:R
2006-2006:L
1900-1990:R
1991-1993:R
1994-1997:L
1998-2006:NA
1991-1995:R
1996-2000:R
2001-2001:R
2002-2006:C
1990-1992:NA
1993-1998:C
1999-2004:C
2005-2006:C
1990-1991: NA
1992-1993: NA
1994-1997: NA
1998-2001: NA
2002-2004: NA
2005-2006: NA
1990-1991:R
1992-1995:R
1996-1999:L
2000-2002:L
2003-2005:R
2006-2006:L
1994-1996:NA
1997-2000:NA
2001-2004:NA
2005-2006:NA
1993
1997
2001
2005
1990
1993
1997
1995
2000
2001
1992
1998
2004
1991
1993
1997
2001
2005
1991
1995
1999
2002
2005
1996
2000
2004
Table A2 continued.
system
party type
year
Singapore
country
Parliamentary
South Africa
Parliamentary
South Korea
Presidential
1990-1991:NA
1992-1997:NA
1998-2001:NA
2002-2006:NA
1990-1994:R
1995-1999:L
2000-2004:L
2005-2006:L
1990-1992:R
1993-1996:R
1997-2000:C
2001-2006:C
1990-1993:L
1994-1996:L
1997-2000:R
2001-2004:R
2005-2006:L
1990-1994:C
1995-1999:L
2000-2005:L
2006-2006:NA
1990-1991:L
1992-1994:R
1995-1998:L
1999-2002:L
2003-2006:L
1990-1991:NA
1992-1995:NA
1996-1999:NA
2000-2003:NA
2004-2006:NA
1991
1997
2001
1994
1999
2004
1992
1996
2000
1993
1996
2000
2004
1994
1999
2005
1991
1994
1998
2002
1991
1995
1999
2003
Spain
Parliamentary
Sri Lanka
Presidential
Sweden
Parliamentary
Switzerland
Parliamentary
System
party type
year
Taiwan
country
Parliamentary
Thailand
Parliamentary
Turkey
Parliamentary
UK
Parliamentary
U.S.
Presidential
1996
2000
2004
1992
1995
1996
2001
2005
1991
1995
1999
2002
1992
1997
2001
2005
1992
1996
2000
2004
Venezuela
Presidential
1990-1996:R
1997-2000:R
2001-2004:R
2005-2006:R
1990-1992:R
1993-1995:R
1996-1996:R
1997-2001:R
2002-2005:NA
2006-2006:NA
1990-1991:R
1992-1995:R
1996-1999:R
2000-2002:L
2003-2006:NA
1990-1992:R
1993-1997:R
1998-2001:L
2002-2005:L
2006-2006:L
1990-1992:R
1993-1996:L
1997-2000:L
2001-2004:R
2005-2006:R
1990-1993:R
1994-1998:NA
1999-2000:NA
2001-2006:NA
1990-1995:NA
1996-2000:NA
2001-2006:NA
Zimbabwe
Presidential
43
1993
1998
2000
1990
1996
2002
Table A3. Classification of Electoral Timing (Robustness).
This table presents the classification of elections according to electoral timing. The countries are classified as having flexible electoral timing if the national leader or legislative body has the option to call an
election before the regularly scheduled date (Julio and Yook (2012a)). An election is classified as ‘called’ if it took place more than three months before the regularly scheduled date. Bialkowski, Gottschalk, and
Wisniewski (2008) employ a similar classification. A variety of additional data sources was consulted, such as International Institute for Democracy and Electoral Assistance, Center on Democratic
Performance, Journal of Democracy, Elections around the World, Election Guide, The CIA World Factbook, the PARLINE Database on National Parliaments, and Keesing’s Record of World Events. Notes: *
1999 Indonesian election is classified as ‘called’ even though Indonesia is classified as having fixed election timing. It was an irregular election following the fall of Suharto administration; ** the 2001 Peruvian
election is classified as ‘called’ as it was an irregular election following the fall of Fujimori administration.
Country
Argentina
Argentina
Argentina
Australia
Australia
Australia
Australia
Australia
Australia
Austria
Austria
Austria
Austria
Austria
Belgium
Belgium
Belgium
Belgium
Brazil
Brazil
Brazil
Canada
Canada
Canada
Canada
Chile
Chile
Chile
Colombia
Colombia
election
1995
1999
2003
1990
1993
1996
1998
2001
2004
1990
1994
1995
1999
2002
1991
1995
1999
2003
1994
1998
2002
1993
1997
2000
2004
1993
2000
2005
1994
1998
electoral
timing
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Fixed
Fixed
called
election
No
No
No
Yes
No
No
Yes
No
No
No
No
Yes
No
Yes
No
Yes
Yes
No
No
No
No
No
Yes
Yes
Yes
No
No
No
No
No
country
Colombia
Czech Rep
Czech Rep
Czech Rep
Denmark
Denmark
Denmark
Denmark
Denmark
Egypt
Egypt
Egypt
Finland
Finland
Finland
Finland
France
France
France
Germany
Germany
Germany
Germany
Germany
Greece
Greece
Greece
Greece
Greece
Hungary
election
2002
1996
1998
2002
1990
1994
1998
2001
2005
1995
2000
2005
1991
1995
1999
2003
1993
1997
2002
1990
1994
1998
2002
2005
1990
1993
1996
2000
2004
1994
electoral
timing
Fixed
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
44
called
election
No
No
No
No
Yes
Yes
No
Yes
Yes
No
No
No
No
No
No
No
No
No
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
No
No
country
Hungary
Hungary
India
India
India
India
India
Indonesia
Indonesia
Indonesia*
Indonesia
Ireland
Ireland
Ireland
Israel
Israel
Israel
Israel
Italy
Italy
Italy
Italy
Japan
Japan
Japan
Japan
Japan
Luxembourg
Luxembourg
Luxembourg
election
1998
2002
1991
1996
1998
1999
2004
1992
1997
1999
2004
1992
1997
2002
1992
1996
1999
2001
1992
1994
1996
2001
1990
1993
1996
2000
2003
1994
1999
2004
electoral
timing
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
called
election
No
No
Yes
No
Yes
Yes
Yes
No
No
Yes
No
Yes
Yes
No
Yes
Yes
Yes
Yes
No
Yes
Yes
No
Yes
Yes
Yes
Yes
Yes
No
No
No
Table A3 continued.
country
Malaysia
Malaysia
Malaysia
Malaysia
Mexico
Mexico
Mexico
Netherlands
Netherlands
Netherlands
Netherlands
New Zealand
New Zealand
New Zealand
New Zealand
New Zealand
New Zealand
Norway
Norway
Norway
Norway
Pakistan
Pakistan
Pakistan
Peru
Peru
Peru**
Philippines
Philippines
Philippines
Poland
election
year
1990
1995
1999
2004
1994
1997
2000
1994
1998
2002
2003
1990
1993
1996
1999
2002
2005
1993
1997
2001
2005
1990
1993
1997
1995
2000
2001
1992
1998
2004
1991
electoral
timing
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Fixed
Fixed
Fixed
Flexible
called
election
Yes
Yes
Yes
No
No
No
No
No
No
No
Yes
No
No
No
No
No
No
No
No
No
No
Yes
Yes
Yes
No
No
Yes
No
No
No
Yes
country
Poland
Poland
Poland
Poland
Portugal
Portugal
Portugal
Portugal
Portugal
Russia
Russia
Russia
Singapore
Singapore
Singapore
S. Africa
S. Africa
S. Africa
S. Korea
S. Korea
S. Korea
Spain
Spain
Spain
Spain
Sri Lanka
Sri Lanka
Sri Lanka
Sweden
Sweden
Sweden
election
year
1993
1997
2001
2005
1991
1995
1999
2002
2005
1996
2000
2004
1991
1997
2001
1994
1999
2004
1992
1996
2000
1993
1996
2000
2004
1994
1999
2005
1991
1994
1998
electoral
timing
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
45
called
election
Yes
No
No
No
No
No
No
Yes
Yes
No
No
No
Yes
No
Yes
Yes
No
No
No
No
No
Yes
Yes
No
No
No
Yes
No
No
No
No
country
Sweden
Switzerland
Switzerland
Switzerland
Switzerland
Taiwan
Taiwan
Taiwan
Thailand
Thailand
Thailand
Thailand
Thailand
Turkey
Turkey
Turkey
Turkey
UK
UK
UK
UK
U.S.
U.S.
U.S.
U.S.
Venezuela
Venezuela
Venezuela
Zimbabwe
Zimbabwe
Zimbabwe
election
year
2002
1991
1995
1999
2003
1996
2000
2004
1992
1995
1996
2001
2005
1991
1995
1999
2002
1992
1997
2001
2005
1992
1996
2000
2004
1993
1998
2000
1990
1996
2002
electoral
timing
Fixed
Fixed
Fixed
Fixed
Fixed
Fixed
Fixed
Fixed
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Flexible
Fixed
Fixed
Fixed
Fixed
Fixed
Fixed
Fixed
Fixed
Fixed
Fixed
called
election
No
No
No
No
No
No
No
No
Yes
Yes
Yes
No
No
Yes
Yes
Yes
Yes
No
No
Yes
Yes
No
No
No
No
No
No
No
No
No
No