Banking on elections. An empirical analysis of political influences
on bank lending over the state electoral cycle
Aldo L´eon Scheffers∗ MSc Thesis Finance University of Groningen† Supervisor: A. Rauf, CFA, Ph.D
June 2, 2020
Abstract
Politicians have incentives to exploit banks for their own political benefit. Using annual aggregated bank lending data, this paper provides empirical evidence for the presence of political influences in the Indian banking system over the period 2002-2019. To circumvent the potential problem of endogeneity of the election cycle, this paper employs an instrument variable that depends on the fact that early called elections are not pre-determined. I find that total bank lending increases by 1.7% to 2.1% in state elections years compared to non-election years with a decrease in lending in the years prior to an non-election. The bank credit distortion in election years is not equally distributed among sectors. Moreover, I find an increase in lending for government-owned banks in elections years, while this effect is absent for private banks. Furthermore, I do not find evidence for an increase in bank lending towards the priority sector in election years. These results indicate that the distortion in bank lending is driven by government-owned banks.
Word count: 11,199
Key words: Banking, Politics, Elections, Bank lending, Government-owned banks, India
1
Introduction
Across many countries in the world, government ownership of banks is prevalent and sub-stantial. Consequently, in addition to their regulatory role, governments also control financial resources through ownership of banks (Baum et al., 2010). La Porta et al. (2002) and Meggin-son and Netter (2001) argue that government ownership is substantial in developing countries, where government-owned banks are viewed as necessary to bolster economic growth. However, developing countries are generally characterized as having underdeveloped financial systems and inefficient governments, which set the stage for political capture of the banks. It can be argued that the problem of political influences at banks will be more severe than for any other government-owned enterprise.1 First of all, banks are active across the whole economy, allowing politicians to strategically target the most politically attractive areas. Second, political moti-vations behind lending are relatively easy to hide for the population with the cost of politically motivated loans only becoming transparent at the maturity of the loan (Din¸c, 2005).
There is an emerging literature on how politicians misuse government-owned banks to enrich themselves or to gain voter’s support for political elections. Most of the literature is focused on the inefficiencies and the misallocation of resources of government-owned banks and the costs of politically motivated loans (Khwaja and Mian, 2005; Berger et al., 2009; Carvalho, 2014; Kumar, 2019). However, what remains relatively unexplored is how bank lending comoves with the electoral cycle and how politically motivated loans are distributed across sectors. Therefore, this paper seeks to fill this gap and contribute to the existing literature by examining the behavior of bank lending over the electoral cycle. Furthermore, this paper will support the above-mentioned by examining how politically motivated loans are distributed among different sectors. Additionally, this paper examines how bank lending over the electoral cycle differs between government-owned and private banks and between credit lend towards the priority sector and non-priority sector.
This paper analyzes the above-mentioned on state elections within one of the most popu-lous developing democratic countries in the world. Specifically, it examines the effect of state electoral cycles on bank lending in 16 major states in India over the period 2002-2019 using aggregated annual bank lending data at the district-level.2 To isolate the political influences, this paper employs a panel data setting with fixed effects to control for annual regional macroe-conomic fluctuations and time invariant district characteristics. One of the issues that arises in analyzing political cycles is the problem of potential endogeneity of elections. The incumbent government has the power to call an election early, which casts doubt on whether the timing of elections are not endogenous. The identification strategy employed to alleviate the endogeneity concern is by defining an instrument variable for the timing of elections that depends on the
1See Shleifer and Vishny (1994) for a general theory on the behavior of government-owned and private
enter-prises.
2
fact that early held elections are not pre-determined.
India provides a particular interesting laboratory to test for the presence of political bank lending cycles in developing countries for several reasons. First of all, state elections in India are not synchronized, which means that within variation in India can be exploited, and therefore variation in business cycles can be excluded as a possible explanation of the occurrence of bank lending cycles. Second, due to the branch expansion program in India, bank branches are uniquely spread out across India, providing politicians with ample opportunities for state capture (Burgess and Pande, 2005). Third, the central bank of India, the Reserve Bank of India, requires all banks to regularly file their data on among else credit outstanding by district and by sector, creating a valuable and detailed database for conducting this research (Cole, 2009).
In this paper, I provide evidence of the presence of political influences on bank lending in India over the state electoral cycle. I find that total bank lending increases by 1.7% to 2.1% in election years compared to non-election years. Additionally, this paper shows that the increase in election years coincides with a decrease in bank lending in non-election years. Furthermore, to my knowledge, this is the first paper that examines how this increase in bank lending is distributed among different sectors. This paper shows that the credit distortion in election years is not equally distributed among sectors. The agriculture, personal loans, trade and finance sector experience a statistically significant increase during election years, while the industry sector experiences a decline in bank lending.
Furthermore, this paper provides evidence that the political influences present in bank lend-ing in India is driven by government-owned banks. Government-owned banks, which in the sample period studied represent on average 71% of credit outstanding, experience a 2.6% to 2.9% increase in bank lending in election years compared to non-election years, while for pri-vate banks there is no evidence of an increase in bank lending in election years. Moreover, I find no evidence that the magnitude of the bank lending increase for government-owned banks is dependent upon whether the ruling party of the state government is affiliated with the ruling party in the center. At last, I do not find a significant increase in bank lending to the prior-ity sector in election years. The results provided in this paper are robust for both the panel regression with fixed effects as well as for the fixed effects two-stage least squares (2SLS).
2
Background on banking and state elections in India
2.1 Banking in India
The banking system in India is to a great extent influenced by government policies. In the 1950s, the banking system was rather liberal with limited control by the government. This changed in 1969 when the government nationalized the 14 largest private banks with the 1969 Bank Nationalization Act. A second nationalization wave occurred in 1980, when the gov-ernment nationalized an additional six major commercial banks. Although the nationalized banks remained corporate entities, the board of directors was replaced by appointees of the government (Das and Ghosh, 2006). Furthermore, in 1969, India launched a branch expansion program to improve access for people in rural areas to bank credit. In 1977, the Indian Central Bank introduced a branch licensing policy, which obliged banks to open four bank branches in unbanked areas for every branch they opened in an area where they already had an existing branch. Due to this branch expansion policy, India achieved a unique scale in terms of bank branches located in every district throughout the country (Burgess and Pande, 2005). As a result, banks were present in every corner of India, which made them prone to state capture, especially since the majority of the banks were under control of the government (Kumar, 2019). Moreover, the Indian government exerted social control over banks after the nationalization in 1969. One of the arguments for government-owned banks is to finance projects that would otherwise not be financed by private banks (La Porta et al., 2002). The government took measures to prevent that bank lending is not only driven by profit maximization incentives, but also to contribute towards the economic development of India by requiring banks to lend credit to the priority sector. The priority sector includes those sectors and activities that the Indian government views as important from a socioeconomic standpoint. The primary categories included in the priority sector are agriculture, micro, small and medium enterprises, export, education and housing loans (Roy, 2006).3 Banks are required to direct 40% of their net bank credit outstanding to the priority sector. Of this 40%, at least 10% has to be directed to weaker sections, which includes among else, small and marginal farmers, distressed farmers, disabled individuals and self-help groups. Banks are required to periodically submit a filing with data regarding the proportion lend to the priority sector. Oversight on the coordination of lending to the priority sector on a state level is governed by the State Level Bankers’ Committee (Kumar, 2019).
In 1992, India implemented a set of bank reforms, also known as the Liberalization of the Indian Banking System, with the aim to strengthen the banking sector in the country. The objective of these reforms was to stimulate the market mechanism in allocating resources and setting prices and to increase the role of the private sector (Das and Ghosh, 2006). Until 1992, all government-owned banks were fully in hands of the government. Since the reforms, several acts enabled government-owned banks to raise capital from the public up to 49% (Ahluwalia, 2002).
3The category agriculture is not exactly similar as the sector agriculture in section 6.2, since almost half of
To further strengthen the banking sector, India initiated the second phase of liberalization reforms in 1998. These reforms shared the same objective as those of 1992 (Bhattacharyya and Pal, 2013). These reforms have resulted in a proportionate increase of the private sector and foreign banks in India measured by total bank assets, which came at the share of government-owned banks. In 1990 – 1991, the share of government-government-owned bank’s credit as of total bank credit was 92% (Das and Ghosh, 2006). Government-owned banks still represent the majority of the banking sector, however, due to the reforms, both private and foreign banks are gradually increasing their share.
2.2 State elections in India
India has a federal structure with assemblies on both the national and state level. The national elections are held every five years. The state level stands between the central gov-ernment and the voters in the state. State elections are held for 28 states and seven union territories. State government elections are similarly to the national elections held every five years. An important feature of state elections is that they do not occur simultaneously across all states. Hence, despite the fact that every state has a five-year electoral cycle, each stage follows its own cycle with every year at least a couple of state elections across India. The party that has won the majority of the seats is invited to form the state government. If no party achieved a majority, a coalition between parties will be formed (Kumar, 2019). Elections can be called early as the central government has the power to remove a state government, known as the Presidential Rule. Khemani (2004) argues that the enforcement of the Presidential Rule is motivated by partisan motives.
States in India are composed of constituencies. On average, there are 136 constituencies per state. Voters elect their local representative, known as the Member of Legislative Assembly (MLA), at the constituency level. The elections follow a first-past-the-post system, where the candidate who received the most votes is declared the winner. The party with a majority of MLAs is elected the winner of the state election and is invited to form the government of the state (Kumar, 2019). The Lok Sabha, the Indian parliament, is composed of representatives of the state governments. In turn, the Lok Sabha appoints the prime minister who leads the central government (Arulampalam et al., 2009). This illustrates the importance of state elections, especially for populous states. The distribution of the seats of the Lok Sabha is dependent upon the size of the population of the state. The more populous a state, the more important it is for a political party to become the victor of the state election and form the state government. For example, the state Uttar Pradesh fills 80 of the 545 seats in the Lok Sabha (Verma, 2004).
2.3 Intersection of banking and politics
government-owned. By virtue of the government’s majority ownership, they have the ability to exert in-fluence at the banks. The government has the power to appoint key senior management and the majority of the board members (Kumar, 2019). Following the political view as discussed in La Porta et al. (2002), the government can control the banks to gain political support, for ex-ample by increasing bank lending to politically attractive areas and sectors prior to an election. Because of the branch expansion program in 1969, the government-owned banks are present in every corner of India, providing an attractive platform for political manipulation.
The state government does not have formal authority over banks, however, they could have the ability to influence bank lending through another mechanism. The state government has the power to appoint members to the State Level Bankers’ Committee (SLBC). The SLBC coordinates the lending policies and practices and in particular lending to the priority sector, on the state level. Next to state government representatives, the committee consists out of representatives from government-owned and private banks and from the Reserve Bank of India. The committee can direct banks to increase lending to categories in the priority sector to meet the priority lending target. On the district-level, oversight on the priority lending target is done by the District Consultative Committee (DCC) that collaborates with the SLBC. In both the committees key posts are held by state representatives. Membership of the committee naturally changes when the state government changes (Kumar, 2019; Cole, 2009).
3
Literature review and hypotheses development
3.1 Literature review
lending to stimulate firm decisions and increase employment in politically attractive regions. Additionally, the paper present evidence that the closeness of the election result is a strong predictor of the politically attractiveness of an area. A finding that is shared by Cole (2009) and Kumar (2019), who both, similar like this paper, examined political influences on banks in India. Kumar (2019) examined the real cost of political interference in bank lending in India using annual bank-branch level data and finds that when banks are forced to increase lending to the agriculture sector, this comes at the expense of the manufacturing sector. The focus of Kumar (2019) is primarily around the agriculture and manufacturing sector and finds that in election years, agriculture lending crowds out manufacturing. The finding that agriculture credit increases in election years, is also observed by Cole (2009) who focused on the agriculture credit market using annual district-level data in India and finds that government-owned banks lend 5%-10% more agricultural credit in election years than in the years following an election over the period 1992-1999. Furthermore, the author presents evidence of a political cycle for agriculture credit with government-owned banks in India over the period 1992-1999.
It is important to note that the period covered by Cole (2009) represents a different banking system in India than the period studied in this paper. As noted in section 2, in 1992, 92% of bank credit was from government-owned banks and in that year India launched its banking liberalization program with a second phase in 1998. Although the majority of the credit out-standing remains to be originated from government-owned banks, it has decreased substantially. Furthermore, Akhmedov and Zhuravskaya (2004) find that cycles get smaller over time and at-tributes this primarily to the development of a democracy with an increase in informational symmetry, literacy, voter awareness and the level of regional democracy.
3.2 Hypothesis development
The related literature is primarily concerned with the political influences via government-owned banks. This finding is especially relevant for identifying the effect of state elections on bank lending in India, as over the sample period studied 71% of the aggregated bank credit is provided by government-owned banks. Moreover due to the branch expansion program in India in the previous century, bank branches are uniquely spread out across India, making them a perfect target for state capture. Additionally, the state government in India may have the power to influence both public and private banks through the SLBC by directing banks to increase lending to categories within the priority sector. Whether Indian banks actually increase their lending in state election years is an empirical question. The hypothesis is focused on total bank credit with the credit divided into several sector categories to support the interpretation of the results and to be able to analyze how the potential credit distortion is distributed among the different sectors. The formal hypothesis is as follows:
H1: Bank lending in India increases in state election years.
Akhmedov and Zhuravskaya (2004), and Din¸c (2005) find that incumbent politicians have an incentive to induce good economic conditions prior to elections to improve the probability of being re-elected. According to Martinez (2009), the rationale of this is that re-election concerns only start to play a role at the end of the term, consequently resulting in political cycles. The theory of political cycles would predict that bank lending increases in election years would be accompanied with bank lending decreases in non-election years (Martinez, 2009). Thus, in a political lending cycle, an increase in bank lending in election years will coincide with a decrease in lending in non-election years. The hypothesis is the following:
H1a: Bank lending in India follows a political lending cycle.
Hypothesis 1 provides an interesting foundation to further explore the mechanism through which politicians can influence bank lending for their own benefit. As noted in section 2, one of the mechanisms through which this influence can be exercised is via government-owned banks. Most of the banks in India are majority-owned by the government. This means that the government has the power to select senior management and members of the board. Through this power, the government can effectively control the bank and its lending decisions, while this power of control is absent for private banks. If government-owned banks follow a political lending cycle, while private banks do not, this would provide evidence for that the results in hypothesis 1 are driven by government-owned banks and the presence of private banks diminishes these results. Hence:
H2: The increase in bank lending in state election years is larger for government-owned banks than for private banks.
The central government has a controlling stake in the government-owned banks and as a result can influence bank lending decisions for its own political benefits. One way to do this is by expanding credit prior to elections in states where the government is aligned with the central government. If the central government influences bank lending to increase the probability of the incumbent state government of getting re-elected, naturally only state governments that are affiliated with the central government are expected to experience an increase in lending. Kumar (2019) finds a significant effect on the interaction term of election year and affiliation using bank-branch level data for agriculture credit. To test whether the occurrence of political affiliation increases bank lending of government-owned banks, the following hypothesis will be tested:
H2a: When the state government is affiliated with the central government, bank lending from government-owned banks increases more in election years than when there is no affiliation with the central government.
influences in the coordination of the priority sector lending. On the contrary, an argument can be made that the priority sector in India includes plausible politically attractive areas. For example agriculture, housing, education and loans to small businesses. This makes the priority sector an attractive area for carrying out political influences via bank lending. Following this line of argumentation, the specific hypothesis to test is as follows:
H3: Bank lending towards the priority sector increases in state election years.
4
Methodology
Din¸c (2005) argues that there are two issues that arise when researching and isolating po-litically motivated actions by banks. First, an event that induces politicians to influence banks must be identified. In this paper I use state elections as event. Second, the regression must control for differences across countries. This is no issue, since this study focuses solely on India, with fixed effects employed to control for local differences within India.
The ideal data set to identify political incentives in bank lending and the mechanism behind it is one that exhibits a district-wise distribution, since this allows for district fixed effects.4 Since state is time invariant within district, the district fixed effects includes the state fixed effects. Therefore, the district fixed effects model is a more robust specification than the state fixed effects model. District fixed effects controls for more unobservable factors, among them, time invariant state-level unobservable factors. Although this is available for aggregated bank credit and for the split of total credit among several sectors, hypothesis 1, it is not available for testing hypotheses 2 and 3. Data that allows for a split between government-owned and private banks and between credit to the priority sector and the non-priority sector are only available at the state-level. The strategy employed in this paper to deal with this complication, is by specifying separate models for hypotheses 2 and 3 using a state-level distribution instead of the level distribution. Although a data set on state-level is less robust than on district-level, given the 18-year time dimension there still remain 270 annual bank credit observations. Furthermore, the hypotheses are focused on comparing the different effects of state elections within the same model. However, a cautionary note should be provided that the coefficients are not exactly comparable to those from hypotheses 1, due to the different specifications in the regressions.5
An important feature that arises when studying political cycles is the problem of potential endogeneity of elections (Khemani, 2004). Section 4.2 discusses this problem in greater detail along with the instrument employed to address this problem.
In the following subsections, the regression specifications employed in this paper will be
4
Since on average every district consists out of 136 constituencies, an analysis on the constituency level would be more robust than on a district level. However, it is not possible to match the credit data to the constituency level, since there is, to my knowledge, no constituency-wise bank credit data available.
5For the interested reader, Appendix B discusses the diagnostic tests have been conducted to support the
described in detail and also how these specifications isolate the political influences and test the hypotheses developed in the previous section. Limitations and shortcomings will be discussed, as well as the techniques applied to address these.
4.1 Regression analyses using district-level data
To test whether state elections in India have an effect on bank lending, the credit outstanding in an election year will be compared to the credit outstanding in non-election years. The idea here is to isolate the political influences using econometrical techniques in a panel setting with fixed effects. The regression estimation includes district fixed effects to control for time-invariant characteristics in a district that affects credit.6 Additionally, region-year fixed effects are added to the model to control for macroeconomic changes over time and across regions. A conscious decision has been made to use region-year fixed effects instead of year fixed effects. While year fixed effects would also control for macroeconomic changes over time, region-year fixed effects allows me to control for different time trends across different geographical regions.7 For example, it controls for differences in rainfall per region over time, as agricultural credit will be impacted by the level of rainfall. One can argue to simply use rainfall data per state as a control variable. This would be a logical avenue and Cole (2009) included annual rainfall per state to control for this, however, this data is not publicly available and I do not have access to a database containing this data. The best alternative to account for this are the region-year fixed effects, which are included in the regression. Formally the model is specified as:
ydst = αdst+ βEst+ λrt+ dst (1)
Where ydst represents the real log level of credit, αdst represents the district fixed effects, Est is the dummy that equals 1 when there is a state election and 0 otherwise and λrt is the variable representing the region-year fixed effects. To focus solely on the actual growth in credit and eliminate inflationary factors, the credit variables are adjusted for inflation using the CPI. To identify the presence of a political lending cycle, a slight modification is needed to Eq. (1). Given that the timing of mid-term elections is generally unforeseen, the specification of the political lending cycle is based on the pre-determined scheduled election years. Following Cole (2009), I define Sst−k, k=0,. . . 4, as dummies that equals 1 if the upcoming scheduled election for state st is in k years. For example, if Karnataka had elections in 2004, 2008 and 2013, Sst−4 would equal 1 in 2005, 2009 and 2014, while Sst−1 would only be 1 in 2003 and 2012. Formally, the regression is given by:
ydst= αdst+ β−4Sst−4+ β−3Sst−3+ β−2Sst−2+ β−1Sst−1+ λrt+ dst (2)
6The specification of the regression has been inspired by the model developed by Cole (2009). 7
4.2 Using the scheduled electoral cycle as an instrument to overcome the endogeneity problem
A methodological issue that arises is the problem of potential endogeneity of elections. Al-though state elections are scheduled to occur every five years, early elections are not uncommon. In the data set of this study, two out of the 56 elections were held prior to the scheduled elec-tion date. Early elecelec-tions can be triggered by a change in coalielec-tion leadership, the incumbent government that wants an early election or if the central government enforces the Presidents Rule (Kumar, 2019). Incumbents could call elections early when economic conditions are pros-perous, potentially causing spurious correlation between election years and credit (Cole, 2009). The identification strategy employed to deal with this potential problem of endogeneity is to define an instrument for the timing of elections that is plausibly exogenous to a change in bank lending and is correlated with the actual occurence of elections. Following Khemani (2004), I define an instrument cycle for the timing of the scheduled elections. The instrument is a dummy variable, Sst0, that equals 1 if five years have elapsed since the previous election.8 However, the instrument takes on a value of 0 if an election has been held early, as is the case in 2008 in the state Karnataka. The instrument electoral cycle follows the pre-determined five-year scheduled election cycle, but starts over after an election was held early. This instrument deals with the problem of incumbents tactically choosing to intervene and call an election early when eco-nomic conditions are prosperous. The method used to include the instrument variable is the fixed effects 2SLS regression. The first stage regression is given by the following equation:
ˆ
Est= αdst+ βSst0 + λrt+ dst (3)
The fitted value, ˆEst, or the electoral cycle instrument, derived from Eq. (3), is subsequently plugged in the second stage regression, which is given by:
ydst = αdst+ β ˆEst+ λrt+ dst (4)
The validity of the instrument can be analyzed by examining two conditions that need to be fulfilled, those of instrument relevance and of instrument exogeneity (Stock and Watson, 2007). The relevance of the instrument is supported by the notion that the instrument is correlated with the actual occurrence of elections.9 Elections that occur according to the pre-determined schedule, concur exactly with the instrument electoral cycle. Instrument exogeneity can be a reasonable assumption, because the instrument electoral cycle follows the scheduled electoral cycle that is pre-determined by constitutional arrangements. Note that the instrument variable is not used for Eq. (2), because the dummy variables for the non-election years are based on the pre-determined scheduled election years. Therefore, by construction omitted variables with respect to the timing of elections do not influence the electoral cycle.
8
The superscript denotes the sum of years until the upcoming election.
9
4.3 Regression analyses using state-level data
To identify the effect of political influences on government-owned and private banks in election years, essentially the same regression estimations, Eq. (1), Eq. (2) and Eq. (4) are used, with the difference that district-fixed effects is replaced by state-fixed effects. Following the methodology of Din¸c (2005), I examine and compare the different effects of state elections on bank lending for government-owned and private banks.
To examine the effect of affiliation of the state government with the central government, the credit outstanding in an election year where the state government is aligned with the party in the center is compared to an election year where there is no affiliation. Following Kumar (2019), I extend the panel with fixed effects regression with a dummy variable representing the political affiliation and an interaction effect between election year and affiliation:
yst= αst+ β1Est+ β2Ast+ β3Est∗ Ast+ λrt+ st (5)
Where yst is the log real level of credit of government-owned banks, Astis a dummy variable for affiliation which takes on the value of 1 if the party in charge of the state government at time t is affiliated with the party in the central, and 0 otherwise. β3 captures the interaction-term between election year and affiliation. The state government is only considered to be affiliated with the central government if they have the same ruling party. Coalitions are not considered, because it is unknown how much influence each party can exercise within a coalition.
For testing whether priority lending is more influenced by politicians than non-priority lending, I employ the same strategies as for comparing the effect of elections years on bank lending of government-owned and private banks.
5
Data
The sample period of this study covers an 18-year period from 2002 up to and including 2019, which is the longest time panel as the data would allow. This period covers 56 state elections.10 One of the limitations of Cole (2009) was the relatively short time panel of eight years, which does not cover two state electoral cycles. By using the 18-year time period, this limitation is taken into account. The required data points necessary to be able to conduct this research are bank level data and political data. The data used in this paper are primarily derived from two sources.
The bank level data is collected from the Reserve Bank of India, which contains a library of data on the Indian economy and banks in specific. The Reserve bank of India aggregates every loan made by every bank in India. All banks have to submit regulatory filings, containing detailed information on among else, bank credit. The filing requirement frequency varies from monthly to annually, but all banks have to submit their filings at least on an annual basis. For this paper, time-series bank data is collected on a yearly frequency basis. For testing of
10
evidence of the effects of state elections on bank lending and the political lending cycle, district-wise and sector-district-wise data is collected as published in RBI’s “Basic Statistical Returns”. As noted in section 4, one of the data limitations is that data on the split between government-owned and private banks and between priority and non-priority sector lending is only available on state level. As discussed in the same section, this does not invalidate the regressions used for hypotheses 2 and 3. Bank-group wise data is published in RBI’s “Basic Statistical Returns” and data on priority lending is published in RBI’s “Statistical Tables Relating to Banks in India”.
The political data is collected from the Election Commission of India, which encompasses detailed information about state elections in India. This includes party identity, party affiliation, the number of seats won for every participating party and the polling date.
5.1 Sample construction
To prevent sample selection bias in the selection of the states for the study, this paper follows Khemani (2007), Cole (2009), and Kumar (2019) and focus only on the major states. India consists out of 28 states. Following Khemani (2007), nine states were removed from the sample, because they are designated as special states.11 States are designed the special state status, largely because of separatist tensions, economic backwardness, non-viable nature of state finance and are provided with special transfers from the center (Khemani, 2007). In addition, the state Telangana is removed, since it only came into existence during the later years of the sample period studied. At last, small states with a share of the total Indian population of less than 1% are excluded, since the number of members represented in the Lok Sabha is based on the state’s population and such small states are from a political view relatively not interesting. Subsequently the entire sample contains 16 major states in India.12 These 16 states account for 90% of the total population of India.
State elections follow the calendar year and are held throughout the year, while bank level data is reported with a fiscal year end date of March 31. For example, fiscal year 2019 is from the beginning of April 2018 to the end of March 2019. If elections take place late in the calendar year, the election related increase in bank lending might occur in the following bank fiscal year (Din¸c, 2005). To match the political data with the bank data, all the analyses are performed using the bank data fiscal year. This requires an adjustment to the years in which the elections are held.13 Following Kumar (2019), I define an election year when the election is held after September 30 of the fiscal year and before October 1 of the following fiscal year. This ensures that the fiscal year captures the largest share of the twelve months preceding a state election, when the effects should have the largest magnitude. For the exact data of the state election,
11Khemani (2007) is the only paper that discusses the sample construction of states in detail.
12These 16 states are: Andhra Pradesh, Assam, Bihar, Gujarat, Haryana, Jharkhand, Karnataka, Kerala,
Madhya Pradesh, Maharashta, Odisha, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh and West Bengal.
13
the state election reports of the Election Commission of India are collected, in which for each election the polling data is provided.
The final sample of hypotheses 1 contains 8,781 annual bank credit observation from 16 distinct states and 540 unique districts. The number of observations can differ between sectors, as for some districts there is no bank lending activity to a certain sector in a given year. The sample of both hypotheses 2 and 3 contain 270 annual bank credit observations from 16 distinct states.14 To minimize the effect of outliers, the credit variables are winsorized by year at the 1% and 99% tails.
5.2 Summary of the sample
Table 1 reports the summary statistics of the data set used in this paper. Panel A shows the credit variables, credit share and political variables that are used for analyzing the pres-ence of an electoral cycle. The unit of observation of this panel is district-year. The sample contains 540 unique districts. Panel A displays how the credit outstanding is distributed per sector. The largest share of credit outstanding is to the industry sector, representing 38% of credit outstanding over the sample period 2002-2019. Followed by personal loans with 20% and agriculture with 14%. Transport operators is the smallest sector in terms of credit outstanding, with around 2%.
Panel B reports the credit and political variables related to the analysis of bank lending in election years for government-owned and private banks. The sample contains 16 states over the time period 2002-2019. The unit of observation for panel B is state-year. Over the sample period studied, on average government-owned banks represented 71% of outstanding credit and private banks 21%. In 10 out of the 56 elections, the ruling party at the state government was affiliated with the central government.
Panel C reports the credit variables, credit share and political variables that are used for the analysis of bank lending towards the priority versus the non-priority sector. Similar to panel B, the sample consist out of 16 states over the time period 2002-2019. The unit of observation of this panel is state-year. Over the sample period studied, on average 41% of credit outstanding has been lend to the priority sector. The largest categories within the priority sector are agriculture, small business and small scale industries, export, education and housing loans.
As mentioned in section 4, all credit variables are adjusted for inflation using the CPI index.15 In line with previous studies such as Carvalho (2014), Cole (2009), and Kumar (2019), the log is subsequently taken from the real credit variables. Taking the log of the credit variables improves the statistical inference of the models. Since the dependent variable is in log, the point estimate can be interpreted as a percentage change in bank credit in election years.
14
Singleton observations within a group are removed.
15
Table 1
Summary statistics
Table 1 reports the summary statistics of the data set used in this paper. The level of observation in Panel A is district-year. The level of observation in Panel B and Panel C is state-year. Panel A is an unbalanced panel and Panel B and Panel C are strongly balanced.
Panel A: Summary statistics for bank lending across the electoral cycle
Obs. Mean Std. Dev.
Credit Variables
Log real Total bank credit 8.791 16.84 1.35
Log real Agriculture credit 8.787 15.58 1.24
Log real Industry credit 8.791 12.38 1.95
Log real Transport operators credit 8.791 12.38 1.68
Log real Professional and other services credit 8.791 13.36 1.72
Log real Personal loans credit 8.790 15.36 1.39
Log real Trade credit 8.791 14.54 1.33
Log real Finance credit 8.704 10.95 2.34
Credit Share
Agriculture 0.142
Industry 0.381
Transport operators 0.023
Professional and other services 0.071
Personal loans 0.202
Trade 0.101
Finance 0.080
Political Variables
Election year 8.791 0.199 0.399
Scheduled election year in 4 years 8.791 0.198 0.399
Scheduled election year in 3 years 8.791 0.202 0.401
Scheduled election year in 2 years 8.791 0.206 0.405
Scheduled election year in 1 year 8.791 0.188 0.390
Scheduled election year (instrument) 8.791 0.192 0.394
Panel B: Summary statistics for government-owned banks versus private banks
Obs. Mean Std. Dev.
Credit Variables
Log real Public credit 288 13.89 1.00
Log real Private credit 288 12.55 1.84
Political Variables
Election year 288 0.194 0.396
Scheduled election year (instrument) 288 0.188 0.391
Affiliation 56 0.179 0.386
Panel C: Summary statistics for the priority and non-priority sector
Obs. Mean Std. Dev.
Credit Variables
Log real Priority sector credit 288 11.02 1.02
Log real Non-priority sector credit 288 11.32 1.13
Political Variables
Election year 288 0.194 0.396
6
Results
This section discusses the results of the study. Section 6.1 covers the main focus of this paper, how bank lending behaves across the state electoral cycle in India. Subsequently, section 6.2 and 6.3 deconstruct these results by analyzing two mechanisms through which political influences could flow into the banking system. Section 6.2 examines the political influences through the government’s majority ownership in banks by examining and comparing bank lending behavior in election years for government-owned and private banks. Section 6.3 examines the political influences from the mechanism of the state government’s influence on coordinating lending policies to the priority sector by analyzing and comparing bank lending behavior to the priority and non-priority sector in election years.
6.1 Bank lending across the electoral cycle
For testing hypothesis 1, the credit outstanding in an election year is compared to the credit outstanding in non-election years. Table 2 reports the results of this analysis. As noted in section 3, the hypothesis is concerned with the coefficient of total bank credit with credit divided into sector-wise categories to provide further information about the distribution of the changes in bank lending. Both panel A and panel B report strong evidence that total bank lending increases in state election years compared to non-election years.
Panel A reports the results of the panel with fixed effects estimation. I find that total bank credit increases by 2.1% in election years, compared to non-election years.16 This coefficient is statistically different from zero at the 1% significance level. Panel B reports the results of the fixed effects 2SLS regression. Similar as to in panel A, the coefficient on total bank credit is positive and is significant at the 1% significance level, albeit with a lower coefficient of 1.7% compared to 2.1% in panel A. In terms of statistical and economical significance, the results are robust for both the panel with fixed effects as well as for the fixed effects 2SLS with both providing evidence supporting the first hypothesis that bank lending in India increases in state election years. This result is in line with the existing literature as discussed in section 3. Despite India’s initiatives to liberalize the banking system in the ninety’s and to make it more competitive, the results indicate that political incentives still find its way through the banking system.
One of the contributions of this paper is to examine how the credit distortions in election years differs by sector. Table 2 shows that in both panels credit linked to agriculture, personal loans, trade and finance sector increases and credit linked to industry sector decreases in election years. The finding that agriculture credit increases in election years in India is documented in existing literature by Cole (2009) and Kumar (2019). The increase can be explained by the fact that agriculture plays an important role in the Indian economy and that 42% of the population is working in this sector, making it an attractive target for politicians. The decrease in industry
16
Table 2
Bank lending in state election years
Table 2 reports the results of the regression of state election years on bank lending for total bank credit and for different sectors. The unit of observation is district-year and the time period is 2002-2019. For both panels the dependent variable is the real log level of credit. Panel A reports the results of the panel with fixed effects regression. In Panel A, Election year is a dummy variable for the election year. Panel B reports the results of the fixed effects 2SLS regression. In Panel B, Election year is a dummy variable for election year instrumented with the scheduled election cycle. Both panels include district and region-year fixed effects. The heteroskedasticity-robust standard errors are clustered at the district level and are reported in parenthesis. Significance of the parameters are indicated as follows: * p < 0.10, ** p < 0.05, *** p < 0.01.
Panel A: Panel with fixed effects
Total bank credit
Agriculture Industry Transport operators Professional and other services Personal loans Trade Finance Election year 0.021*** 0.016*** -0.029** -0.004 0.023** 0.017*** 0.017** 0.062* (0.006) (0.006) (0.012) (0.013) (0.011) (0.005) (0.007) (0.033)
District Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
Region-Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
Observations 8.781 8.777 8.781 8.781 8.781 8.780 8.781 8.694
Adjusted R2 0.963 0.958 0.916 0.883 0.933 0.965 0.936 0.747
Panel B: Fixed effects 2SLS
First stage coefficient
Total bank credit
Agriculture Industry Transport operators Professional and other services Personal loans Trade Finance
Scheduled Election year (instrument) 1.000*** (0.001)
Election year 0.017*** 0.012* -0.032** -0.005 0.014 0.012** 0.019** 0.077**
(0.006) (0.006) (0.013) (0.013) (0.011) (0.005) (0.008) (0.034)
District Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Region-Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations 8.781 8.781 8.777 8.781 8.781 8.781 8.780 8.781 8.694
Adjusted R2 0.970
credit is in line with the finding of Kumar (2019), that an increase in agriculture crowds out lending to the industry sector. The increase in personal loans is not yet explained by the existing literature, however, a plausible argument can be made that since personal loans consists for the majority out of housing, education and vehicle loans, it is a logical sector to indirectly (via the banks) influence the well-being of the state’s population. Theoretical explanations for the effect on bank lending for the remaining sectors are still absent.
Table 3 extends Table 2 by examining how bank lending comoves with the electoral cycle. The regressions are run with election years as a reference, hence the results are interpreted as the percentage change relative to an election year. The results of Table 3 present evidence in support of hypothesis 1a, that in India political lending cycles are present. For all the four years preceding a scheduled state election, the coefficients for total bank credit are negative and statistically significant. The coefficients for total bank credit in non-election years range between -0.026 to -0.021, implying that bank lending in non-election years is 2.1% to 2.6% lower than in election years. To compensate for the boom in lending in election years, banks need to reduce their lending in non-election years, which is what is shown in Table 3. This result is in line with the literature as discussed in section 3 that politicians have an incentive to influence good economic conditions (i.e., by increasing bank lending) only prior to elections, since voters are primarily influenced just prior to elections. A similar cycle is observed for agriculture. For industry this cycle is reversed, with increases in two of the four years prior to the scheduled election, which potentially can be explained by the rebound of the decrease in the election year.
6.2 Government-owned banks versus private banks
Table 4 reports the results of the analysis of bank lending during election years for owned and private banks. Both panel A and B provide evidence that bank credit of government-owned banks increases in election years, while this evidence is absent for private banks.
Table 3
Political bank lending cycle
Table 3 reports the results of the regressions of the state election cycle on bank lending for total bank credit and for different sectors. The unit of observation is district-year and the time period is 2002-2019. The dependent variable is the log real level of credit. The variables denote the number of years until the upcoming scheduled election. Each column represents a regression that includes district and region-year fixed effects. The heteroskedasticity-robust standard errors are clustered at the district level and are reported in parenthesis. Significance of the parameters are indicated as follows: * p < 0.10, ** p < 0.05, *** p < 0.01.
Total bank credit
Agriculture Industry Transport operators Professional and other services Personal loans Trade Finance
Scheduled election in 4 years -0.021** -0.032*** 0.030 0.011 -0.017 -0.005 0.009 -0.061
(0.008) (0.008) (0.019) (0.017) (0.014) (0.008) (0.011) (0.043)
Scheduled election in 3 years -0.026*** -0.024*** 0.030* 0.012 -0.031** -0.013* -0.035*** -0.073*
(0.008) (0.008) (0.016) (0.018) (0.014) (0.007) (0.009) (0.042)
Scheduled election in 2 years -0.026*** -0.022*** 0.029 0.020 -0.008 -0.017** -0.014 -0.143***
(0.008) (0.008) (0.018) (0.017) (0.015) (0.007) (0.010) (0.042)
Scheduled election in 1 year -0.023*** -0.029*** 0.042** 0.049*** -0.005 -0.013* 0.017 -0.124***
(0.008) (0.008) (0.018) (0.017) (0.013) (0.008) (0.010) (0.042)
District Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
Region-Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
Observations 8.781 8.777 8.781 8.781 8.781 8.780 8.781 8.694
Adjusted R2 0.963 0.958 0.916 0.883 0.933 0.965 0.936 0.747
Table 4
Bank lending in state election years for government-owned versus private banks
Table 4 reports the results of the regressions of the state election years for bank lending for government-owned and private banks. The unit of observation is state-year and the time period is 2002-2019. The dependent variable is the log real level of credit. Panel A reports the results of the panel with fixed effects regressions. In Panel A, Election year is a dummy variable for the election year. Panel B reports the results of the fixed effects 2SLS regressions. In Panel B, Election year is a dummy variable for election year instrumented with the scheduled election cycle. Each regression includes state and region-year fixed effects. The heteroskedasticity-robust standard errors are clustered at the state level and are reported in parenthesis. Significance of the parameters are indicated as follows: * p < 0.10, ** p < 0.05, *** p < 0.01.
Panel A: Panel with fixed effects
Government-owned banks Private banks
Election year 0.029** 0.008
(0.013) (0.054)
State Fixed Effects Yes Yes
Region-Year Fixed Effects Yes Yes
Observations 270 270
Adjusted R2 0.981 0.946
Panel B: Fixed effects 2SLS
First stage coefficient Government-owned banks Private banks Scheduled election year (instrument) 1.000***
(0.001)
Election year 0.026* 0.014
(0.015) (0.050)
State Fixed Effects Yes Yes Yes
Region-Year Fixed Effects Yes Yes Yes
Observations 270 270 270
Adjusted R2 0.945
Additional support for evidence of the presence of political influence by government-owned banks is provided in panel A in Table D.1 in Appendix D. This panel reports bank lending of government-owned and private banks across the electoral cycle. Table D.1 provides evidence of the existence of a political lending cycle for government-owned banks, while a cycle is not observed for private banks. For government-owned banks three out of the four years prior to the scheduled elections are statistically significant with a coefficient ranging between -0.047 to -0.032. This indicates that in non-election years bank lending decreases by between 3.2% to 4.7% compared to election years.
shows a coefficient of -0.056, however, the coefficient is not statistically significant. This coeffi-cient indicates that when in an election year the state government is affiliated with the central government, the occurrence of the affiliation provides an additional decrease in bank lending of 5.6% on the 3.9% increase in election years. Both the sign of the coefficient and the statistical in-significance of the interaction term fails to support hypothesis 2a. This result is in contrast with Kumar (2019), who documents an increase in agriculture bank lending when the governments are affiliated. However, a crucial difference exists between the model specifications, where the author conducts his analysis using bank-branch data, this paper employs state-level data. The state-wise data distribution substantially decreases the observations of states that in election years are affiliated with the central government. In only ten out of the 56 state elections in the studied time period, there was a political affiliation. Hence, the analysis lacks statistical power. Moreover, Kumar (2019) accounted for coalitions, while in this paper a deliberate decision has been made not to do so.
Table 5
Political affiliation between state and central government on bank lending of government-owned banks in elections years Table 5 reports the results of the regressions of the political affiliation between state and central government in elections years on bank lending for government-owned banks. The unit of observation is state-year and the time period is 2002-2019. The dependent variable is the log real level of credit. Election year is a dummy variable for the election year. Affiliation is a dummy variable that takes on the value of 1 if the ruling parties in the state and central are affiliated. Election year*Affiliation is an interaction term for when in an election year the parties are affiliated. The regression includes state and region-year fixed effects. The heteroskedasticity-robust standard errors are clustered at the state level and are reported in parenthesis. Significance of the parameters are indicated as follows: * p < 0.10, ** p < 0.05, *** p < 0.01.
Government-owned banks Election year 0.039* (0.022) Affiliation 0.076 (0.086) Election year*Affiliation -0.056 (0.044)
State Fixed Effects Yes
Region-Year Fixed Effects Yes
Observations 270
Adjusted R2 0.981
6.3 Bank lending towards the priority sector versus non-priority sector
sector increases in state election years. Furthermore, Table D.1 in Appendix D shows that for both bank lending towards the priority as the non-priority sector a cycle in bank lending is absent.
One interpretation of these results is that the influence of state representatives is limited, preventing state representatives to influence bank lending to the priority sector prior to elections. This result may appear puzzling, when compared to the results in Table 2, which showed that agriculture and personal loans are both positive and significant. The priority sector consists out of agriculture, housing and education loans, where housing and education loans both are components of the sector personal loans. However, as noted in section 2, the agriculture category in the priority sector is not vis-`a-vis comparable to the agriculture sector in Table 2, because the agriculture category in priority lending is more weighted to the “weaker section”, with an emphasis to small and marginal farmers. Additionally, education and housing loans are only two components of the personal loans sector. Moreover, agriculture, education and housing loans represent less than 60% of the priority sector. Furthermore, since the monitoring of the priority lending target is not only done on the state-level, but also on the district-level, a district-wise data distribution could provide a more robust insight into this analysis.
Table 6
Priority versus non-priority lending
Table 6 reports the results of the regressions of state election years on bank lending towards the priority and the non-priority sector. The unit of observation is state-year and the time period is 2002-2019. The dependent variable is the real log level of credit. Panel A reports the results of the panel with fixed effects regression. In Panel A, Election year is a dummy variable for the election year. Panel B reports the results of the fixed effects 2SLS regression. In Panel B, Election year is a dummy variable for election year instrumented with the scheduled election cycle. Both panels include state and region-year fixed effects. The heteroskedasticity-robust standard errors are clustered at the state level and are reported in parenthesis. Significance of the parameters are indicated as follows: * p < 0.10, ** p < 0.05, *** p < 0.01.
Panel A: Panel with fixed effects
Priority sector Non-priority sector
Election year 0.014 0.024
(0.010) (0.026)
State Fixed Effects Yes Yes
Region-Year Fixed Effects Yes Yes
Observations 270 270
Adjusted R2 0.984 0.938
Panel B: Fixed effects 2SLS
First stage coefficient Priority sector Non-priority sector
Scheduled election year (instrument) 1.000*** (0.001)
Election year 0.007 0.028
(0.013) (0.027)
State Fixed Effects Yes Yes Yes
Region-Year Fixed Effects Yes Yes Yes
Observations 270 270 270
7
Conclusion
This paper provides empirical evidence about the political influences on banks in India over the time period 2002-2019. The paper focuses on state elections and examines their effect on bank lending. To isolate the political influences, this paper employs a panel regression setting with district and region-year fixed effects to control for time invariant district characteristics and for annual regional macroeconomic changes. To add robustness to the results, all the analyses concerning a change in bank lending in election years are tested using two specifications, the panel with fixed effects and fixed effects 2SLS. Where the latter includes an instrument variable to deal with the potential problem of endogeneity.
This paper shows that banks in India increase their lending in election years. This effect of state elections on bank lending is not due to inflationary fluctuations, since the credit amounts are adjusted for inflation, nor is it due to time-invariant characteristics in a district, which would be absorbed by the district fixed effects, nor can it be attributed to regional annual macroeco-nomic fluctuations, which would be absorbed by the region-year fixed effects. Moreover, this paper provides evidence that in all the four years prior to an election year bank lending is lower than in election years. In addition, I find that the increase in bank lending is not equally dis-tributed among sectors. Specifically, an increase in bank lending is observed in the agriculture, personal loans, trade and finance sector, while for the industry sector bank lending decreases in election years and increases in the years prior to the election. Furthermore, this paper shows that government-owned banks experience an increase in bank lending in election years, while this evidence is absent for private banks. This indicates that the increase in bank lending for all banks is driven by government-owned banks that in the sample period studied represent 71% of all outstanding bank credit. Additionally, I do not find evidence of an increase in bank lending of government-owned banks in election years when the ruling party in the state is affiliated with the central party. At last, I do not find supporting evidence for a significant increase in bank lending towards the priority sector in election years, which could indicate that the level of power of state government representatives in the SBLC, the committee that coordinates the priority lending, to influence bank lending is rather limited.
Implications of the findings reported in this paper are primarily of concern to policy makers and bank oversight officials. This paper shows that bank behavior changes with the electoral cycle and that it distorts bank lending across different sectors. Political influences are espe-cially present with government-owned banks. Since banks form a corner stone of economic and financial development of a country, especially in a developing country, bank lending distortions driven by political influences are undesirable. The large presence of government-owned banks in India is not uncommon for developing countries, as government-owned banks are very common in developing countries (La Porta et al., 2002). Therefore, the implications of this paper are not only limited to India, but to all developing countries with a high share of government-owned banks.
issues. First of all, data on the type of bank and on priority lending is only available at state-level data. As discussed in the paper, district fixed effects is more robust than state fixed effects, making the results across hypothesis 1 with those of 2 and 3 not directly comparable. Second, the state-level data limits the number of values of political affiliation for testing hypothesis 2a. Third, since electoral competition occurs at the constituency level, analyses related to examine politically attractive areas based on the competitiveness of the prior election are ruled out given that the credit data is either district-wise or state-wise distributed.
Appendix A
Data description
Variable Description
Credit variables
Total bank credit Aggregated outstanding bank credit of all scheduled commercial banks in India.
Source: Reserve Bank of India’s “Basic Statistical Returns”
Agriculture Aggregated outstanding bank credit related to the agriculture sector of all scheduled commercial banks in India.
Source: Reserve Bank of India’s “Basic Statistical Returns”
Industry Aggregated outstanding bank credit related to the industry sector of all scheduled commercial banks in India. The industry sector includes among else, manufacturing, mining, textiles, chemicals, basic metals, petroleum, and construction.
Source: Reserve Bank of India’s “Basic Statistical Returns”
Transport operators Aggregated outstanding bank credit related to the transport sector of all scheduled commercial banks in India.
Source: Reserve Bank of India’s “Basic Statistical Returns” Professional and
other services
Aggregated outstanding bank credit related to the professional services sector of all scheduled commercial banks in India. The professional and other services sector includes professional services, tourism, hotel and restaurants, recreation services, and IT and telecommunication.
Source: Reserve Bank of India’s “Basic Statistical Returns”
Personal loans Aggregated outstanding bank credit related to the personal loans sector of all scheduled commercial banks in India. The personal loans sector includes housing, educational, consumer, and vehicle loans.
Source: Reserve Bank of India’s “Basic Statistical Returns”
Trade Aggregated outstanding bank credit related to the trade sector of all scheduled commercial banks in India. This includes credit towards wholesale and retail trade.
Source: Reserve Bank of India’s “Basic Statistical Returns”
Finance Aggregated outstanding bank credit related to the finance sector of all scheduled commercial banks in India.
Source: Reserve Bank of India’s “Basic Statistical Returns” Government-owned
bank credit
Aggregated outstanding bank credit of all scheduled commercial government-owned banks in India.
Source: Reserve Bank of India’s “Basic Statistical Returns”
Source: Reserve Bank of India’s “Basic Statistical Returns” Priority lending
credit
Aggregated outstanding bank credit to the priority sector, which in-cludes agriculture, educational loans, housing loans, micro, small and medium enterprises and export credit.
Source: Reserve Bank of India’s “Statistical Tables Relating to Banks in India”
Non-priority lending credit
Total aggregated outstanding bank credit that is not lend out to the priority sector. Specified as total aggregated bank credit minus priority sector bank credit.
Source: Reserve Bank of India’s “Statistical Tables Relating to Banks in India”
Political variables
Election year Dummy variable that takes on the value of one if in a specific year in a particular state, a state election took place. To match the credit data, the election year is based on the fiscal year of the banks.
Source: Election Commission of India’s “Results and Statistics” Scheduled election in
4 years
Dummy variable that takes on the value of one if in a specific year in a particular state, a state election is scheduled to occur in 4 years. Source: Election Commission of India’s “Results and Statistics” Scheduled election in
3 years
Dummy variable that takes on the value of one if in a specific year in a particular state, a state election is scheduled to occur in 3 years. Source: Election Commission of India’s “Results and Statistics” Scheduled election in
2 years
Dummy variable that takes on the value of one if in a specific year in a particular state, a state election is scheduled to occur in 2 years. Source: Election Commission of India’s “Results and Statistics” Scheduled election in
1 year
Dummy variable that takes on the value of one if in a specific year in a particular state, a state election is scheduled to occur in 1 year.
Source: Election Commission of India’s “Results and Statistics” Affiliated Dummy variable that is equal to one when the party leading the state
government is the same as the party in the central government. Data on the leading party per state after every election is collected from publica-tions per selection containing information with the seats won per party and the leading party.
Appendix B
Diagnostic tests
Table B.1 reports the results of the diagnostic tests performed to help determine the specifi-cation of the models employed in this paper. Panel A reports the results of the modified Chow test for the validity of pooled OLS in this study. The results in panel A show that this test is rejected for all models, indicating that a pooled OLS is invalid and hence a panel regression is required.
Panel B reports the results of the modified Hausman test, that allows for heteroskedasticity-robust clustered standard errors. The Hausman test tests whether the errors are uncorrelated with the independent variables. All the regressions fail to reject the null hypothesis, which indicates that the difference in coefficients is not systematic. The test would suggest to use the random effects model, since it is more efficient. However, as Wooldridge (2010) argues, the Hausman test is not flawless, as it for example is unable to incorporate time fixed effects, which are present in the panel estimations. Therefore, to determine the panel estimator approach, this paper follows previous literature on isolating political influences, such as Din¸c (2005), Sapienza (2004), Carvalho (2014), Cole (2009) and Kumar (2019) and employs fixed effects in the panel estimations. The economic argument behind choosing fixed effects over random effects is that the goal of this paper is to isolate the effect of state elections on bank lending in a district or state, which is not affected by information of the other districts or states in the sample.
Panel C reports the result of the Hausman test for time fixed effects. It is a joint test that tests whether the dummies for all years are equal to zero. For all the models, the null hypothesis is rejected, indicating that time-fixed effects are needed. Additionally, as discussed in section 4, this paper uses region by year fixed effects instead of year fixed effects.
Panel D reports the results of the modified Wald test. This test tests for the presence of heteroskedasticity in the panel. In three out of the five regressions heteroskedasticity is present. To deal with this, heteroskedasticity-robust standard errors are used.
Panel E reports the results of the Woolridge test for serial autocorrelation. This test is rejected for all regressions, implying that serial autocorrelation is present. To account for this, the standard errors are clustered at the district-level for hypothesis 1 and at the state-level for hypothesis 2 and 3.
At last, panel F tests the validity of the instrument variable. The F-statistics of all the regressions are all larger than 10, supporting the validity of the instrument.17 The high F-statistics can be explained by the fact that in the sample period studied, early elections are not common with only 2 out of the 56 election being held early. Therefore, the instrument variable is strongly correlated with the actual occurrence of election years (the independent variable).
17
Table B.1 Diagnostic tests
Table B.1 provides the results of six diagnostic tests that are used in this paper to help specify the models employed. Total bank credit relates to the model used for testing hypothesis 1, Government-owned banks and Private banks to the model used for testing hypothesis 2 and Priority sector and Non-priority sector to the model used for testing hypothesis 3.
Panel A: Modified Chow test for validity of pooled OLS
Total bank credit Government-owned banks
Private banks Priority sector Non-priority sector
F-statistic 89.21 69.92 22.52 46.01 75.33
P-value 0.000 0.000 0.000 0.000 0.000
Pooled OLS valid No No No No No
Panel B: Modified Hausman test fixed effects versus random effects
Total bank credit Government-owned banks
Private banks Priority sector Non-priority sector
Sargan-Hansen statistic 0.505 0.330 0.309 0.583 0.081
P-value 0.477 0.565 0.578 0.445 0.776
Difference in coefficients is systematic No No No No No
Panel C: Hausman test for time fixed effects
Total bank credit Government-owned banks
Private banks Priority sector Non-priority sector
Chi2 429.38 687.44 714.71 1,894.31 380.04
P-value 0.000 0.000 0.000 0.000 0.000
Time-fixed effects needed Yes Yes Yes Yes Yes
Panel D: Modified Wald test for groupwise heteroskedasticity
Total bank credit Government-owned banks
Private banks Priority sector Non-priority sector
Chi2 9206.69 16.41 45.77 11.90 55.02
P-value 0.000 0.425 0.000 0.751 0.000
Heteroskedasticity present Yes No Yes No Yes
Panel E: Woolridge test for autocorrelation in panel data
Total bank credit Government-owned banks
Private banks Priority sector Non-priority sector
F-statistic 107.23 109.27 14.98 268.86 5.26
P-value 0.000 0.000 0.002 0.000 0.037
Serial autocorrelation present Yes Yes Yes Yes Yes
Panel F: Instrument relevance
Total bank credit Government-owned banks
Private banks Priority sector Non-priority sector Scheduled Election year (instrument) 1.000*** 1.000*** 1.000*** 1.000*** 1.000***
(0.000) (0.001) (0.001) (0.001) (0.001) F-statistic 8.84e+07 2,606,728 2,606,728 2,606,728 2,606,728 Instrument is relevant (F-statistic > 10) Yes Yes Yes Yes Yes
Observations 8.781 270 270 270 270
Appendix C
State election distribution
Table C.1
Election years per state and political affiliation between the state and the central government
Table C.1 provides an overview of the state elections held over the sample period (2002 to 2019) for the 16 states covered in this study. To match the election year data with the bank credit data, the election years are adjusted to the bank fiscal year calender. An election year is defined when the election is held after September 30 of the fiscal year and before October 1 of the next fiscal year. The X denotes whether a state election occurred in a specific state for a given year. Additionally, this table presents the distribution of political affiliation between the state and the central government during election years. Political affiliation in election years is illustrated by X*.
