Chen et al “the effect of the political connections of government bank CEOs on bank performance during the financial crisis”, j financial stability (2018)

Accepted Manuscript
Title: The Effect of the Political Connections of Government
Bank CEOs on Bank Performance during the Financial Crisis
Authors: Hung-Kun Chen, Yin-Chi Liao, Chih-Yung Lin,
Ju-Fang Yen
PII:
DOI:
Reference:

S1572-3089(17)30525-9
/>JFS 609

To appear in:

Journal of Financial Stability

Received date:
Revised date:
Accepted date:

26-7-2017
9-12-2017
28-2-2018

Please cite this article as: Chen, Hung-Kun, Liao, Yin-Chi, Lin, Chih-Yung, Yen,
Ju-Fang, The Effect of the Political Connections of Government Bank CEOs
on Bank Performance during the Financial Crisis.Journal of Financial Stability
/>This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process

errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.


The Effect of the Political Connections of Government Bank CEOs on
Bank Performance during the Financial Crisis✩
Hung-Kun Chen
Department of Banking and Finance
Tamkang University
Email:

SC
R

IP
T

Yin-Chi Liao
Department of Management and Marketing
Western Illinois University
Email:

N
A
M

ED

Ju-Fang Yen
Department of Statistics

National Taipei University
Email:

U

Chih-Yung Lin*
College of Management
Yuan Ze University
Email:

✩

PT

* Corresponding author. Tel.: (886) 3-463-8800#6370; Fax: +886-3-4557040.
E-mail address: (C.-Y. Lin).

A

CC
E

We are especially grateful for constructive comments from Yan-Shing Chen, Yehning Chen, Tse-Chun
Lin, Yanzhi Wang and seminar participants at National Taiwan University for helpful comments and
suggestions. Chih-Yung Lin appreciates financial support from the Taiwan Ministry of Science and
Technology. Any remaining errors are ours.

1

Abstract

SC
R

IP
T

This study investigates how the political connections of government bank CEOs
affected their banks’ performance during the 2007-2009 financial crisis. Examination
of global data shows that government banks with politically connected CEOs
experienced significantly higher loan default rates and worse operating performance
during the crisis than those without politically connected CEOs. However, these
politically connected CEOs were less likely than others to be penalized for the poor
performance of their banks. Our evidence suggests that politically connected CEOs of
government banks can influence a bank’s lending decisions by using their political
power and influence to relax lending standards and to reap private benefits that thus
raise their banks’ sensitivity to a crisis.
Keywords: Political connections, government banks, financial crisis, institutional
ownership, country corruption and governance.

U

JEL classification: G01, G21, G28, G34

N

1. Introduction

A

In August 2007, the credit markets froze after two hedge funds run by New York-

M

based Bear Stearns Co. collapsed because of the plummeting values of their subprime
mortgage holdings. The inability to set a price on such securities paralyzed the market.

ED

At that time, most government-owned banks were encouraged by their governments to
increase lending to prevent the collapse of business and to stabilize and promote

PT

economic recovery (Laeven and Valencia, 2010, 2013), which resulted in banks

CC
E

acquiring many nonperforming loans that weakened their capital reserves.1 However,
not all government banks suffered equally. While researchers have been intrigued by
the heterogeneous performance of banks during the most recent financial crisis (2007

A

to 2009) (e.g., Fahlenbrach and Stulz, 2011; Beltratti and Stulz, 2012; Berger and
Bouwman, 2013; Ellul and Yerramilli, 2013; Ho, Huang, Lin, and Yen, 2016), we
contribute to the literature by providing a new perspective to explore why some
government banks’ performance was worse than others during the crisis period.

1

For instance, Iannotta, Nocera, and Sironi (2013) find that some European government banks became
insolvent following the onset of the global financial crisis.
2


We propose that the political connections of government banks CEOs contributed
to risky lending decisions by their banks that influenced the banks’ subsequent
performance during the crisis. We focus on CEOs because they are the primary
influence on government bank lending standards, which affect the banks’ sensitivity to
a crisis (Sapienza, 2004; Khwaja and Mian, 2005; Shen and Lin, 2012). In a theoretical

IP
T

model, Acharya and Naqvi (2012) also show that a manager’s incentives to take
excessive risks can induce over-lending decisions and thus sow the seeds of an

SC
R

impending crisis.

We follow the previous literature and define government banks as banks with at

U

least 20 percent government ownership (La Porta, Lopez-de-Silanes, and Shleifer,

N

2002). We divide these banks into two groups by year: we designate banks with CEOs

A

who served as politicians as political banks and those without these types of executives

M

as non-political banks (Faccio, Masulis, and McConnell, 2006; Fan, Wong, and Zhang,
2007). CEOs with political backgrounds may retain their political connections, even as

ED

executives. To pursue these political affiliations, such as a future political career,
politically connected CEOs of government banks tend to follow the interests of other

PT

politicians. Further, with the support that politically connected CEOs obtain from other

CC
E

politicians, they may ignore market pressures to report low-quality accounting
information or poor operating performance (Chaney, Faccio, and Parsley, 2011). Hence,
we conjecture that political banks may have carried more low-quality loans because of

A

political connections than did non-political banks, either before or during the crisis (the
political-connection hypothesis).
To investigate the issue, we compare the loan quality and performance of political
and non-political government banks during the global financial crisis by using data
from 41 countries. We obtain these data from Bankscope. The period of the global
3


financial crisis ran from 2007 to 2009 and the period from 2004 to 2006 we term the
pre-crisis period (Ivashina and Scharfstein, 2010; Beltratti and Stulz, 2012).
Empirically, the results support the political-connection hypothesis. Political
banks significantly approved more low-quality loans than did non-political banks

IP
T

before or during the crisis, such that they were confronted with a higher ratio of
nonperforming to gross loans during the crisis. This ratio indicates that politically

SC
R

connected banks became increasingly inefficient and pursued riskier lending behavior.

Furthermore, these lower-quality loans caused significant underperformance, as
measured by return on assets, return on equities, net interest income to total assets, and

U

the cost-to-income ratio during the crisis years.

N

Our research design, based on the recent global financial crisis, can mitigate

A

endogeneity concerns. The financial crisis represents an exogenous shock with a

M

negative effect on all individual firms. The crisis resulted in a systematic decrease in
loan quality and in the subsequent performance of the banks and thus allows us to

ED

employ a difference-in-differences (DiD) analysis. We also include bank fixed effects

PT

in our regression models to control for any potential endogeneity concerns arising from
omitted variables or measurement error (Roberts and Whited, 2013). However,

CC
E

government banks may choose politically connected CEOs for unobserved bank
characteristics or political reasons (Cooper, Gulen, and Ovtchinnikov, 2010), leading to
a self-selection bias.2 We use Heckman’s (1979) two-stage approach to address the

A

self-selection bias that might result from a ruling party choosing politically connected
CEOs for government banks (Cooper, Gulen, and Ovtchinnikov, 2010). As a result, our
main results remain unchanged after controlling for the potential selection bias.

2

In another view, the ruling party can choose politically connected CEOs for government banks (Shen
and Lin, 2012), whereby hiring politically connected CEOs for government banks is an endogenous
assignment. We report the results in Table 5 based on the endogenous assignment assumption.
4


We also find that the negative influence of political connections on government
banks, which we term the PC effect, can be reduced by the presence in those countries
of institutional ownership and superior institutional factors. We find that the PC effect
is diminished if a government bank has institutional ownership. We also find that the
underperformance of political banks is not observed in countries with strong

IP
T

governance systems or low levels of corruption, which is consistent with the findings
of previous studies that institutional factors in countries have a significantly positive

SC
R

influence on the lending behavior of banks (Qian and Strahan, 2007; Bae and Goyal,

2009; Haselmann, Pistor, and Vig, 2009). This evidence also shows that these

U

institutional factors may in fact exercise sufficient influence to protect banks from

N

political intervention. Thus, the inefficient allocation of resources by political banks

A

can be partly controlled by institutional ownership and the specific country’s

M

institutional factors.

Moreover, we investigate possible motives for the performance-destructive

ED

behavior of government banks with politically connected CEOs, i.e., whether these
CEOs grant more low-quality loans for their own benefit, such as enhancing their own

PT

future political careers. By studying employment renewal as well as whether the CEOs

CC
E

of government banks were offered a political position after the crisis, we find positive
evidence of such motives. We find that 22.50% of politically connected CEOs remained
at the same government bank from 2010 to 2013, as compared to only 8.79% of non-

A

politically connected CEOs; 28.75% of the politically connected CEOs of government
banks were offered a political position from 2010 to 2013, as compared to only 4.40%
of non-politically connected CEOs. Therefore, the CEOs of government banks with
political connections and poor operating performance are less likely to be penalized by
the bank or by the political system. These CEOs even had the potential for a successful

5


political career after the crisis. This evidence is consistent with our political-connection
hypothesis, that the CEOs of government banks used their political power and influence
to relax lending standards and to reap private benefits.
Research studies have typically used a macro-level measure, election years, to

IP
T

represent the political factor and to analyze the influence of political ties (Sapienza,
2004; Dinỗ, 2005; Brown and Dinỗ, 2005; Micco, Panizza, and Yaňez, 2007; Shen and

SC
R

Lin, 2012; Jackowicz, Kowalewski, and Kozłowski, 2013). The literature shows that
politicians obtain more benefits during major elections (Dinỗ, 2005; Micco, Panizza,
and Yaez, 2007; Iannotta, Nocera, and Sironi, 2013). Different from previous studies,

U

we focus on a bank-level political factor by considering the previous role of a CEO as

N

a politician and investigate whether the PC effect is stronger in a major election year.

A

We do not find evidence showing that a major countrywide election aggravates the

M

negative influence of political connections on government banks’ performance,
however, indicating that the PC effect still exists after controlling for a major election

ED

year.

PT

The contributions of our study to the literature are threefold. First, we complement
the literature on political connections of bank CEOs by investigating their negative

CC
E

influence on government banks. The influence of political connections on corporate
finance has recently attracted critical attention within industrial firms. Most of this
literature indicates that the political connections of CEOs add value to firms.3 While

A

these studies have shown that borrowing firms usually use their own political
connections to attract favorable loans from government banks (Sapienza, 2004; Dinỗ,

3

That is, politically connected firms are more likely to obtain preferential treatment when applying for
bank loans (Khwaja and Mian, 2005; Charumilind, Kali, and Wiwattanakantang, 2006) to gain an
increase in stock returns during the elections (Goldman, Rocholl, and So, 2009; Cooper, Gulen, and
Ovtchinnikov, 2010), to be informed in advance on future policy directions (Belo, Gala, and Lin, 2013),
to be the first to be bailed out (Faccio, Masulis, and McConnell, 2006), and so on.
6


2005; Carvalho, 2014), we provide evidence that the political connections of lenders
could also affect their lending decisions. 4 Using global data, we propose a novel
viewpoint to show the dark side of political connections from the perspective of the
supply side of the financial system.5 This paper thus complements the literature by
showing that government banks with politically connected CEOs suffer from lower

IP
T

lending standards.6

SC
R

Second, we relate political connections to the banking literature on corporate

governance and institutional factors. We show that the PC effect can be partly
eliminated when government banks have institutional ownership. This finding

U

contributes to the field of corporate governance in which the presence of institutional

N

ownership reduces the agency problem (e.g., Weisbach 1988; Bhojraj and Sengupta,

A

2003; Henry, 2008). In addition, previous studies reveal the far-from-ideal track record

M

of government banks regarding the efficiency of their capital allocations (Sapienza,
2004; Khwaja and Mian, 2005; Iannotta, Nocera, and Sironi, 2007; Ho, Chen, Lin, and

ED

Chi, 2016). We find that the influence of the political connections of government bank

4

PT

CEOs is not as strong in countries with better governance and lower corruption levels.

A

CC
E

Government banks tend to charge lower interest rates to firms associated with the ruling party than to
those without such an affiliation (Sapienza, 2004). In addition, politicians can use government banks to
distribute incentives to their supporters by increasing lending during election periods (Dinỗ, 2005) and
to use lending to expand employment in politically attractive regions (Carvalho, 2014).
5
Only two studies show the dark side of political connection, that is, Fan, Wong, and Zhang (2007) and
Chaney, Faccio, and Parsley (2011). They find that politically connected firms underperform in
comparison with non-politically connected firms in terms of post-IPO stock returns and the quality of
accounting information reports, respectively. Our study complements Fan, Wong, and Zhang (2007) and
Chaney, Faccio, and Parsley (2011) by showing that banks with political CEOs would perform worse.
6
Hung, Jiang, Liu, Tu, and Wang (2017) find that banks with politically connected CEOs outperform
their non-connected counterparts, which is in contrast with ours. Although both studies focus on the
political view of lenders, we use a sample of global government banks whereas Huang et al. (2017) use

a sample of commercial banks in China, which includes both government-owned and privatelyowned
banks. Our results are in line with prior literature, which has argued that political connections hurt the
value of government-owned banks (Sapienza, 2004; Khwaja and Mian, 2005; Iannotta, Nocera, and
Sironi, 2007; Shen and Lin, 2012; Shen, Hasan, and Lin, 2014). Hung et al. (2017) is consistent with the
literature that political connections enhance the value of privately owned firms (Cooper, Gulen, and
Ovtchinnikov, 2010).
7


These findings complement those in recent studies that investigate the important role
of corruption in bank lending (Beck, Demirgỹỗ-Kunt, and Levine, 2006; Barth, Lin,
Lin, and Song, 2009; Houston, Lin, and Ma, 2011).
Third, this paper complements the literature on bank lending during global

IP
T

financial crises (e.g., Ivashina and Scharfstein, 2010; Puri, Rocholl, and Steffen, 2010;
Acharya and Naqvi, 2012; Chen, Chen, Lin, and Sharma, 2016). Although these studies

SC
R

show a substantial decline in the loan supply during crisis periods, little attention has
been given to the lending behavior of government banks. During the 2007-09 crisis

period, we find that the political connections of government banks led to the

U

deterioration of their lending quality and a consequent decline in their operating

N

performance. These findings are consistent with the view that the political intervention

A

of politicians leads to inefficient lending by government banks (Dinỗ, 2005; Shen and

M

Lin, 2012; Iannotta, Nocera, and Sironi, 2013).

The remainder of this paper is organized as follows. We develop our hypotheses

ED

in Section 2 and describe our data and present basic statistics in Section 3. We examine

PT

the relationship between the PC effect and bank performance in Section 4 and in Section
5 we examine that between the PC effect and institutional factors. Section 6 presents a

CC
E

discussion of the results, and Section 7 concludes the paper.

2. Hypotheses development

A

Government banks played an important role in preserving the stability of financial

markets during the global financial crisis (Laeven and Valencia, 2010, 2013). However,
not all government banks suffered equally during the crisis. This difference raises the
question of why some government banks performed worse than others during crisis
period.
8


The political view suggests that the operations of government banks are constantly
being used by politicians to pursue their individual political goals, such as the provision
of jobs, resources, or subsidies to their friends and supporters (Shleifer and Vishny,
1998). Therefore, the maximization of the political benefits to politicians becomes their
main objective. For instance, several studies have shown that industrial firms often use

IP
T

their political connections to attract favorable loans from government banks (Sapienza,
2004; Dinỗ, 2005; Faccio, Masulis, and McConnell, 2006; Carvalho, 2014). No studies

SC
R

have as yet taken into account, however, whether the types of political connection of
some government bank CEOs might differ from others, so that some banks are aligned

U

with politicians’ interests, but others are not.

N

When government bank CEOs have previously served as politicians, they can

A

retain their political ambitions. On the one hand, politically connected bank CEOs are

M

willing to align themselves and their banks with the interests of other politicians to
facilitate their future political careers. To ensure the success of these careers, politically

ED

connected CEOs can grant more low-quality loans that lead to poor operating

PT

performance, especially during financial crises.
On the other hand, politically connected CEOs of government banks are not

CC
E

penalized for high loan default rates as long as their politician friends can protect them
from market pressures. To some extent, politicians provide protection to the companies
with which they associate by preventing them from being penalized for low-quality

A

accounting information (Chaney, Faccio, and Parsley, 2011). If politically connected
government bank CEOs care less about loan quality, they are more likely to lend money
to low-quality borrowers to gain political influence, resulting in an increase in default
loan rates and a reduction in operating performance during crisis years.
Non-politically connected CEOs of government banks tend to care more about
9


loan quality, which may not align with politicians’ interests. These CEOs do not readily
take risks by engaging in low-quality loans because they do not have strong support
from political friends. Moreover, non-politically connected CEOs are not concerned as
much as their politically connected peers about a future political career. Instead, they

decisions. Therefore, we present our first two hypotheses as follows:

IP
T

may prefer to maintain their current position and carefully participate in lending

SC
R

Hypothesis 1: Political banks made more low-quality loans than did non-political

banks before or during the global financial crisis.

Hypothesis 2: Political banks performed worse than non-political banks during the

U

crisis because of the low-quality loans made before or during the crisis.

N

Several studies have shown that the presence of institutional ownership can

A

mitigate agency problems by providing efficient monitoring of managers and by

M

reducing information asymmetry between a firm and its lenders. Greater institutional

ED

ownership is associated with lower bond yields and higher credit ratings on new bond
issues (Bhojraj and Sengupta, 2003). Firms with higher levels of institutional ownership

PT

are more likely to terminate poorly performing CEOs and are associated with better
valuation (Henry, 2008; Aggarwal, Erel, Ferreira, and Matos, 2011). Some studies also

CC
E

show that the proportion of institutional ownership increases with the quality of the
governance structure (Aggarwal et al., 2011; Chung and Zhang, 2011). Therefore, if the
presence of institutional ownership has a well-established governance mechanism to

A

discipline managers, it should also reduce the negative influence of political
connections. Therefore, our third hypothesis is as follows:
Hypothesis 3: The presence of institutional ownership can eliminate the PC effect on
bank performance.

10


Studies have also observed that a country’s institutional factors have a significantly
positive influence on the lending behavior of banks (Qian and Strahan, 2007; Bae and
Goyal, 2009; Haselmann, Pistor, and Vig, 2009). For example, high-quality country
governance (e.g., regulatory quality and government effectiveness) is also related to the
improved performance of government banks (Boehmer, Nash, and Netter, 2005; Shen

IP
T

and Lin, 2012).

SC
R

In addition, in countries with high corruption levels, the performance of

government banks is severely hampered by political interference (Boehmer, Nash, and
Netter, 2005; Shen and Lin, 2012; Shen, Hasan, and Lin, 2014) that impedes banks

U

capital efficiency (Beck, Demirgỹỗ-Kunt, and Levine, 2006; Houston, Lin, and Ma,

N

2011). Overall, these previous studies suggest that government banks in countries with

A

superior governance and low corruption levels and are less likely to lend to low-quality

M

borrowers. Thus, our fourth hypothesis is as follows:

3. Data

PT

bank performance.

ED

Hypothesis 4: Countries’ superior institutional factors can eliminate the PC effect on

CC
E

3.1. Sample collection and identification of political banks
Our initial sample is obtained from Bankscope, which contains ownership and

accounting data of banks worldwide. To avoid the policy effect of government banks,

A

our sample includes only bank holding companies, commercial banks, and savings
banks. We then restrict our sample to government banks or banks that have greater than
20 percent government ownership (La Porta, Lopez-de-Silanes, and Shleifer, 2002). We
check individual bank websites and other publications to identify and verify
government ownership.
11


Similar to the literature on political connection measures (Faccio, Masulis, and
McConnell, 2006; Fan, Wong, and Zhang, 2007), we classify government banks with
CEOs or presidents who have served as politicians as political banks; otherwise they
are classified as non-political banks. We acquire the names of the CEOs or presidents
for each government bank from 2004 to 2009 through Bankscope to determine any

IP
T

political bank connections by year. We then manually collect background information

on the CEOs or presidents from bank websites, the Wall Street Journal, and the Factiva

SC
R

database to confirm their political experience.

We also limit our sample to countries with both political and non-political banks

U

from 2004 to 2009 to avoid sample selection issues. The financial performance of all

N

political and non-political banks in the same country are subsequently assessed. The

A

final sample comprises 207 government banks in 41 countries from 2004 to 2009. Table

M

1 presents the definitions for all variables used in this study.

ED

[Insert Table 1 here]

Table 2 shows the sample distribution of the selected government banks in 2006.7

PT

Most of the sampled countries have only from two to four government banks. India,
Argentina, and China have the highest numbers of government banks at 20, 12, and 12,

CC
E

respectively. Our sample comprises 96 political banks and 111 non-political banks, as
reported in the fourth and the fifth columns of Table 2. China, Taiwan, and the United

A

Arab Emirates have the most political banks at ten, eight, and seven, respectively.
[Insert Table 2 here]

3.2. Loan quality
Studies have found that political connections have a significantly negative

7

For brevity, we do not show the dynamic government bank data of Table 2.
12


influence on lending quality. Khwaja and Mian (2005) find that government banks
differentially favor politically connected firms by granting them greater access to credit.
Such preferential treatment damages the lending quality of these banks. Research has
focused on the political connections between borrowing firms and lenders. In the

present study, we investigate the influence of political connections on lenders and their

IP
T

loan quality.

SC
R

We use the ratio of nonperforming loans to total gross loans (NPL) as a proxy for
loan quality. This proxy measures the portion of banks’ loan portfolios that are in default
or are close to default. Nonperforming loans are those with payments of interest and

U

principal that are more than 90 days overdue. Various studies have used this proxy as a

N

non-discretionary measure of loan quality (Wahlen, 1994; Liu, Ryan, and Wahlan, 1997;

A

Liu and Ryan, 2006). The higher the ratio of nonperforming loans to total gross loans,

M

the higher are future loan default rates, and the more low-quality loans that are made
ex ante by banks.

ED

3.3. Summary statistics

PT

Table 3 shows the summary statistics for our sample during the crisis and pre-crisis
periods (from 2004 to 2009). Panel A presents the descriptive statistics. All accounting

CC
E

data are winsorized at the top and bottom one percent to avoid potential inference bias
because of outliers. First, we find that 46.58% of our sample is composed of political
banks (PB). The loan quality, measured by the NPL, on average is 6.22%, with a

A

standard deviation of 7.21%. The mean values of the performance measures are 1.46%
(ROA), 13.87% (ROE), 4.27% (NIM), and 58.22% (C/I).
We then focus on bank characteristics and macroeconomic variables. The mean
values of the natural logarithm of total assets (Asset), debt-to-equity ratio (D/E), loan-

13


to-deposit ratio (LOANDEP), and ratio of current assets to total assets (LIQUID) are
8.51, 12.30, 0.8180, and 0.2280, respectively. The mean log of GDP per capita (GDP),
GDP growth rate (GDP growth), Budget surplus, Inflation rate, and exchange rate

changes (Exchange rate) are 8.3064, 0.36%, 17.39%, 6.15%, and 0.24%, respectively.
These values are consistent with studies on global banking that examine bank

IP
T

characteristics. The first-quarter, median, third-quarter, and standard-deviation values

of all variables are also given in Panel A of Table 3. Panel B of Table 3 presents the

SC
R

correlation coefficient matrix of the variables. The results show that correlations
between variables in general are very small, making multicollinearity less of a concern.

U

[Insert Table 3 here]

A

N

4. CEO political connections and bank operating performance

M

4.1. Univariate analysis

Table 4 presents a comparison of the characteristics and performance of political

ED

and non-political banks during the pre-crisis period (2004–2006) and during the crisis
period (2007–2009), with Panel A focusing on the bank characteristics and Panel B on

PT

bank performance. Panel A of Table 4 shows that political and non-political banks have

CC
E

similar characteristics during the pre-crisis period, which confirms that non-political
banks are appropriate benchmarks for political banks. The differences in Asset, D/E,
LOANDEP, and LIQUID are also insignificant.

A

[Insert Table 4 here]
Panel B of Table 4 shows that the differences in performance between political

and non-political banks are all insignificant during the pre-crisis period. The NPLs for
political banks, however, are significantly higher than those of non-political banks
during the crisis period. Given the same bank characteristics and loan change rates in
14


both pre-crisis and crisis periods, we can infer that political banks lowered their loan

quality ex ante such that their NPLs became higher than those of non-political banks
when the market crashed. Thus, non-political banks outperformed political banks to
some extent during the crisis period. For example, the ROEs of political banks
deteriorated during 2008 and 2009.8 Similar results are obtained for the ROA and NIM

IP
T

confirming that the political connections of government banks resulted in poor bank
performance during the crisis. Finally, the performance differences of C/I are

SC
R

insignificantly positive during the crisis. Therefore, the PC effect of C/I is weaker than
that of the other three variables. In sum, the univariate analysis results align with our

U

Hypotheses 1 and 2, that political banks made more low-quality loans and performed

N

worse than non-political banks during the global financial crisis.9

A

4.2. Multivariate analysis

M

In this subsection, we conduct a multivariate analysis to re-examine Hypothesis 1
and to control for some variables that might influence bank performance. We employ a

ED

DiD analysis to measure the influence of political connections on government banks.

PT

The econometric model is as follows:

CC
E

𝑃𝐸𝑅𝐹𝑂𝑅𝑀𝑖𝑗𝑡 = 𝛼1 + 𝛼2 𝑃𝐵𝑖𝑗 + 𝛼3 𝑃𝐵𝑖𝑗 × 𝐶𝑟𝑖𝑠𝑖𝑠𝑡 + 𝛼4 𝐶𝑟𝑖𝑠𝑖𝑠𝑡
+𝜷′ 𝒁𝒊𝒋𝒕−𝟏 + 𝜈𝑖 + 𝜀𝑖𝑗𝑡 ,

(1)

where PERFORMijt is substituted by one loan-quality measure (NPL), three

A

8

The lower ROE of political banks is not derived from their highly leverage effects because political
banks tended to be low-leveraged banks during the crisis period. As shown in Panel A of Table 4, the
differences between in leverage (D/E) between political banks and non-political banks during the crisis
period are negative, while the value in 2009 is significant with a 10% level.

9
It is possible that these banks make low quality loans because they had a higher interest burden to cope
with compared to other banks, but not necessarily because they were lending to firms that favored by
politicians. However, as shown in Panel B of Table 4, the difference in ratio of net interest income to
total assets (NIM) between political banks and non-political banks is insignificant for each year from
2004 to 2006. Based on the results of NIM, we find no evidence that political banks make low quality
loans in exchange for higher interest.
15


profitability measures (ROA, ROE, and NIM) and one cost measure (C/I), PBij is a
dummy variable that equals one if bank i in country j is a political bank, and zero for
all other cases, 𝐶𝑟𝑖𝑠𝑖𝑠𝑡 is a dummy variable that equals one if the end of a bank’s fiscal
year, t, is within the crisis period (2007–2009), and zero otherwise, and
𝒁𝒊𝒋𝒕−𝟏 represents a vector of control variables containing bank characteristics and five

IP
T

macroeconomic variables from bank i in country j at year t-1. The control variables for

bank characteristics are Asset, D/E, LOANDEP, and LIQUID; and the five

SC
R

macroeconomic variables are GDP, GDP growth, Budget surplus, Inflation rate, and
Exchange rate. All control variables in this study are suggested by Dinỗ (2005),

U

Iannotta, Nocera, and Sironi (2007), and Shen and Lin (2012). In the regression

N

analyses, the t-statistics based on standard errors are adjusted for heteroskedasticity and

A

are clustered at the country level (White, 1980; Petersen, 2009). Moreover, 𝜈𝑖 captures

bank dummies are not reported.

M

bank fixed effects, while 𝜀𝑖𝑗𝑡 is the random error. To save space, the coefficients of the

ED

Table 5 shows the multivariate analysis results. In Model (1), we examine whether
political banks approved more low-quality loans than did non-political banks before or

PT

during the crisis by focusing on the interaction term between 𝑃𝐵𝑖𝑗 and 𝐶𝑟𝑖𝑠𝑖𝑠𝑡 .

CC
E

Sapienza (2004) finds that government banks charge lower interest rates to firms

associated with the ruling party than to others. Dinỗ (2005) also shows that, unlike
private banks, government banks increase their lending during election periods. The PC

A

effect therefore emerges from corruption or related lending behaviors. We thus
conjecture that political banks tend to increase their loans to politicians, related parties,
or their party supporters with loose lending standards, making them more vulnerable to
the shock of a crisis. Hence, the nonperforming loans of political banks could have

16


increased more than those of non-political banks during the crisis period.10
[Insert Table 5 here]
Consistent with our conjecture, we find that the coefficient of the interaction term
𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) has a significantly positive association with NPL, which shows that

IP
T

political banks, on average, experienced 1.2948% higher nonperforming loans than
non-political banks during the crisis period. It implies that political banks significantly

SC
R

approved more low-quality loans than did non-political banks before or during the crisis

(H1). As the average total loan amount in our sample is about $31.049 million dollars,

political connection of government banks led to an increase of $0.4020

U

(0.4020=31.049×1.2948%) million dollars in nonperforming loans during the crisis.

N

The finding indicates that the PC effect is both statistically significant and economically

A

meaningful.

M

In Models (2) to (5), we further determine whether political banks that approved

ED

more low-quality loans caused the poor performance of political banks during the crisis
period. We find that the coefficients of the interaction term 𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) have

PT

a significantly negative association with ROA, ROE, and NIM, and the t-statistics are 2.83, -2.50, and -2.09, respectively. For example, the coefficient on the interaction term

CC
E

𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) with respect to ROE, -4.6081, shows that political banks, on
average, experienced a reduction of 4.6081% in ROE over non-political banks during
the crisis period. Although the 𝑃𝐵𝑖𝑗 (𝛼2 ) coefficient on C/I is not significant, the sign

A

remains positive. These findings support our Hypothesis 2, that the political
connections of government banks worsened their performance during the financial

10

Wahlen (1994) finds that a high ratio of NPL is associated with low loan quality. Liu, Ryan, and
Wahlan (1997) and Liu and Ryan (2006) indicate that NPL information is contemporaneous and less
discretionary about loan defaults because banks are required to disclose NPL. Hence, the ratio of NPL is
the most appropriate and available measure for overall loan quality.
17


crisis. The coefficients of the bank characteristics and macroeconomic variables are
also consistent with our expectations.
In sum, a synthesis of the results confirms our conjecture that political banks
increasingly introduced low-quality lending before or during the crisis, which resulted

IP
T

in poor operating performance. Our results thus show that inefficient lending behavior
stemming from political influence explains the PC effect.

SC

R

4.3. Self-selection bias

The president or ruling party of a country generally makes decisions regarding the
hiring of the top managers of government banks. Politicians can therefore provide jobs

U

or direct resources to their friends or party supporters by recommending them to

N

government banks. However, some government banks choose politically connected

A

CEOs while others do not. This choice introduces a possible self-selection bias into our

ED

Gulen, and Ovtchinnikov, 2010).

M

sample, which may lead to inconsistent estimates and a spurious interpretation (Cooper,

To address the potential self-selection bias, we use Heckman’s (1979) two-stage

PT

approach. In the first stage, we perform a probit regression using 𝑃𝐵𝑖𝑗 as the
dependent variable. Four bank characteristics (Asset, D/E, LOANDEP, and LIQUID)

CC
E

and country dummies are used as independent variables to assess the possible motives
for government banks to build political connections. The resulting inverse Mill’s ratio

A

(IMR) is inserted into the second-stage regression to correct for any potential bias.
Table 6 presents the results of the second-stage regression based on Equation (1).

The results in Table 6 are broadly consistent with those in Table 5. For example, the
coefficients of the interaction term 𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) have a significantly positive
association with NPL and a significantly negative association with ROA, ROE, and NIM.

18


These findings confirm that the low loan quality of political banks led to their
underperformance during the crisis. This evidence indicates that our main results are
not driven by a self-selection bias.
[Insert Table 6]

IP
T

4.4. Robustness: Control government ownership
Several prior studies have confirmed that bank performance is worse when

SC
R

government control is more pervasive (e.g., greater than 50 percent). Thus, it is
reasonable to assume that politicians are appointed as CEOs when the government stake

is particularly high (50 percent or more). If this is the case, most of the results in this

U

study could be explained by the size of the government stake (as opposed to CEOs’

N

political connections). Therefore, in this section, we additionally control for

M

A

government ownership to rule out the latter’s influence.

We employ a DiD analysis, which is designed to measure the influence of political

ED

connections on government banks by controlling for government ownership. The

econometric model is as follows:

PT

𝑃𝐸𝑅𝐹𝑂𝑅𝑀𝑖𝑗𝑡 = 𝛼1 + 𝛼2 𝑃𝐵𝑖𝑗 + 𝛼3 𝑃𝐵𝑖𝑗 × 𝐶𝑟𝑖𝑠𝑖𝑠𝑡 + 𝛼4 𝐶𝑟𝑖𝑠𝑖𝑠𝑡
+𝛼5 𝐺𝑂𝑖𝑗 + 𝜷′ 𝒁𝒊𝒋𝒕−𝟏 + 𝜈𝑖 + 𝜀𝑖𝑗𝑡 ,

CC
E

(2)

where PERFORMijt is substituted by NPL, ROA, ROE, and NIM, and C/I, PBij is a

A

dummy variable that equals one if bank i in country j is a political bank, and zero for
all other cases, 𝐶𝑟𝑖𝑠𝑖𝑠𝑡 is dummy variable that equals one if the end of a bank’s fiscal
year, t, is within the crisis period (2007–2009), and zero otherwise, 𝐺𝑂𝑖𝑗 is
government ownership of the bank i in country j, and 𝒁𝒊𝒋𝒕−𝟏 represents a vector of
control variables.

19


Table 7 presents the results based on Equation (2). After controlling for
government ownership, the results in Table 7 are broadly consistent with those in Table
5. For instance, the coefficients of the interaction term 𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) have a
significantly positive association with NPL and a significantly negative association with
ROA, ROE, and NIM again. This evidence indicates that our main results are not driven

IP
T

by size of the government’s stake.

SC
R

[Insert Table 7]

5. Institutional ownership, institutional factors, and CEO political

U

connections

N

5.1. Institutional ownership and the PC effect

A

Because institutional ownership can provide an efficient monitoring mechanism

M

for the self-interest of managers (Henry, 2008; Aggarwal et al., 2011; Chung and Zhang,
2011), we hypothesize that the PC effect should be diminished if a government bank

ED

has more than ten percent private institutional ownership.11 To test this issue, we divide
our sample into two subgroups, banks without and with institutional ownership, and

PT

reexamine the PC effect by applying Equation (1).

CC
E

Table 8 presents the results regarding the influence of private institutional
ownership on the PC effect.12 Panels A and B present the samples of banks without
and with institutional ownership, respectively. The coefficients of the interaction term

A

𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) in Panel A have a significantly negative association with ROA,
ROE, and NIM and a significantly positive association with C/I. By contrast, in Panel
B, the coefficients of the interaction term 𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) are all insignificant.
11

La Porta, Lopez-de-Silanes, and Shleifer (1999) also use 10% as cut-off point to define a significant
owner in a company because this amount provides a significant threshold of votes.
12
For the sake of space, we also do not report the coefficients of the control variables hereafter.
20

These results confirm our Hypothesis 3, that the presence of institutional ownership can
mitigate or eliminate the negative impacts of political connections in government banks.
Therefore, our results are aligned with the literature that institutional ownership can
provide an external monitoring mechanism to reduce the self-interest of managers.13

IP
T

[Insert Table 8 here]
Greater government ownership is associated with low institutional ownership,

SC
R

which could influence the results of Hypothesis 3. Thus, we add a control for
government ownership in Table 8 to rule out the latter’s influence. For brevity, we do
not report the estimated results; however, they remain similar to those reported in Table

U

8, which confirms that the PC effect is diminished in banks with institutional ownership.

A

N

5.2. Institutional factors and the PC effect

M

Scholars suggest that institutional factors affect the behavior of banks across
countries. Using country-level institutional factors, such as strength of the country

ED

governance system and corruption level, we examine whether countries’ superior
institutional factors can eliminate the negative influences of political connections on

PT

government banks.

CC
E

We initially apply the Worldwide Governance Indicators (WGI) compiled by
Kaufmann, Kraay, and Mastruzzi (2007) as our country governance index. 14 We
classify the sample countries as having either strong or weak governance based on a

13

A

The question may be raised as to why institutional investors want to invest in government banks, given
that some of those banks may make poor lending decisions due to political connections. One potential
motivation is because institutional investors are potentially controlling shareholders when government
banks are undergoing privatization (Megginson, 2005; Boehmer, Nash, and Netter 2005; Bonin, Hasan,
and Wachtel, 2005; Boubakri, Cosset, Fischer, and Guedhami, 2005; Clarke, Cull, and Shirley, 2005).
Other than the role of corporate governance, however, the motivation for institutional investors to invest
in government banks could be an interesting topic for further research.

14
Considering the increasing importance of country governance, Kaufmann, Kraay, and Mastruzzi
(2007) updated their WGI. The WGI involves six dimensions of governance, namely, regulatory quality,
rule of law, control of corruption, political stability, absence of violence, voice and accountability, and
government effectiveness, and covers 212 countries from 1996 to 2006.
21


zero value. We expect the PC effect to be weaker in countries with strong governance.
Table 9 shows the PC effect results in countries with different country governance
situations. Panels A and B represent the weak and strong governance countries,
respectively. The coefficients of the interaction term 𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) in Panel A

IP
T

have a significantly negative association with ROA, ROE, and NIM and a significantly
positive association with C/I. By contrast, in Panel B, the coefficients of the interaction

SC
R

term 𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) are insignificant for ROE and NIM and have a significantly
negative association with C/I, which rejects the existence of the PC effect. This
evidence confirms that the PC effect is diminished in countries with strong governance,

U

confirming that superior institutional factors can eliminate the negative influence of

N

political connections on government banks.

A

[Insert Table 9 here]

M

We then use a corruption index for each country to investigate the linkage between

ED

the PC effect of political banks and the country’s corruption level. The corruption index
is obtained from the International Country Risk Guide. The scale of the index is from

PT

0 to 6, in which a high number implies a low corruption level. We then divide our
sample into two subgroups: countries with corruption values of less than three are

CC
E

classified as having high corruption; countries with values above three have low
corruption levels.

A

Table 10 shows the PC effect results in countries with different corruption

situations. Panels A and B represent the high- and low-corruption countries,
respectively. The coefficients of the interaction term 𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) in Panel A
have a significantly negative association with ROA, ROE, and NIM and a significantly
positive association with NPL and C/I. By contrast, in Panel B, the coefficients of the

22


interaction term 𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) are all insignificant, except for the coefficient on
the ROA measure. These findings support our Hypothesis 4, that superior institutional
factors can eliminate the negative influence of political connections on government
banks. High corruption levels in countries increase the PC effect, thus this result
supports the argument of Barth et al. (2009) that corruption in bank lending is a

IP
T

particularly serious problem in developing countries. This evidence complements and

is consistent with the recent studies of Beck, Demirgỹỗ-Kunt, and Levine (2006) and

SC
R

Houston, Lin, and Ma (2011), which suggest that high corruption levels of countries
lead to bank lending inefficiencies.

U

[Insert Table 10 here]

N

In sum, superior institutional factors in countries can eliminate the PC effect. Thus,

A

political banks do not underperform in countries with strong country governance or low

M

levels of corruption. Hence, the “inefficient or wasted” lending problem can be at least

ED

partly controlled by superior institutional factors.
5.3. Robustness: Election years and CEO political connections

PT

During major elections, politicians can obtain greater benefits that lead to the
underperformance of government banks (Sapienza, 2004; Dinỗ, 2005; Micco, Panizza,

CC
E

and Yaňez, 2007). In this subsection, we examine whether the poor operating
performance of political banks emerges from the political connection measure after

A

controlling for the election factor. Major elections include presidential elections in
countries that elect a president and the highest parliamentary elections in countries that
elect a prime minister (see Persson and Tabellini, 2003 for the details).
Following the studies of Dinỗ (2005) and Micco, Panizza, and Yaňez (2007), we
create a dummy variable, ELECTIONjt, that equals one if a major election is occurring

23


in country j at year t, and zero if otherwise. We add the ELECTIONjt dummy and the
interaction terms ( 𝑃𝐵𝑖𝑗 × 𝐸𝐿𝐸𝐶𝑇𝐼𝑂𝑁𝑗𝑡 and 𝑃𝐵𝑖𝑗 × 𝐸𝐿𝐸𝐶𝑇𝐼𝑂𝑁𝑗𝑡 × 𝐶𝑟𝑖𝑠𝑖𝑠𝑡 ) to
Equation (1). The econometric model is then expressed as
𝑃𝐸𝑅𝐹𝑂𝑅𝑀𝑖𝑗𝑡 = 𝛼1 + 𝛼2 𝑃𝐵𝑖𝑗 + 𝛼3 𝑃𝐵𝑖𝑗 × 𝐶𝑟𝑖𝑠𝑖𝑠𝑡 + 𝛼4 𝑃𝐵𝑖𝑗 × 𝐸𝐿𝐸𝐶𝑇𝐼𝑂𝑁𝑗𝑡

+𝛼7 𝐶𝑟𝑖𝑠𝑖𝑠𝑡 + 𝜷′ 𝒁𝒊𝒋𝒕−𝟏 + 𝜈𝑖 + 𝜀𝑖𝑗𝑡 ,

IP
T

+α5 𝑃𝐵𝑖𝑗 × 𝐸𝐿𝐸𝐶𝑇𝐼𝑂𝑁𝑗𝑡 × 𝐶𝑟𝑖𝑠𝑖𝑠𝑡 + 𝛼6 𝐸𝐿𝐸𝐶𝑇𝐼𝑂𝑁𝑗𝑡

(3)

SC
R

where PERFORMijt is substituted by NPL, ROA, ROE, NIM, and C/I; PBij is a dummy

variable that equals one if bank i in country j is a political bank and zero, if otherwise;

U

𝐶𝑟𝑖𝑠𝑖𝑠𝑡 is a dummy variable that equals one if the end of a bank’s fiscal year, t, is

N

belong to crisis period (2007–2009), and zero otherwise; ELECTIONjt is a dummy

A

variable that equals one if there is a major election in country j at year t, and zero for

M

all other cases. Major elections include presidential and highest parliamentary elections
(as defined above); Zijt-1 represents a vector of control variables from bank i in country

ED

j at year t-1, 𝜈𝑖 capture the bank fixed effect; and 𝜀𝑖𝑗𝑡 is the random error.
Table 11 presents the results on the influence of political connections on

PT

government banks after controlling for the election factor. First, the coefficients of the

CC
E

interaction term 𝑃𝐵 × 𝐶𝑟𝑖𝑠𝑖𝑠 (𝛼3 ) have a significantly positive association with
NPL and a significantly negative association with ROA, ROE, and NIM, which reconfirms our PC effect. In addition, the coefficients of the interaction term 𝑃𝐵 ×

A

𝐸𝐿𝐸𝐶𝑇𝐼𝑂𝑁 (𝛼4 ) have a significantly negative association with ROA and ROE. The
results support the conjecture that major countrywide elections aggravate the
underperformance of government banks. This finding is consistent with those in the
election literature on government banks (Sapienza, 2004; Dinỗ, 2005; Khwaja and Mian,
2005; and Micco, Panizza, and Yaňez, 2007). Finally, the coefficients of the interaction
24