Government Credit, a Double-Edged Sword: Evidence from the China Development Bank

Transcription

1 THE JOURNAL OF FINANCE VOL. LXXIII, NO. 1 FEBRUARY 2018 Government Credit, a Double-Edged Sword: Evidence from the China Development Bank HONG RU ABSTRACT Using proprietary data from the China Development Bank (CDB), this paper examines the effects of government credit on firm activities. Tracing the effects of government credit across different levels of the supply chain, I find that CDB industrial loans to state-owned enterprises (SOEs) crowd out private firms in the same industry but crowd in private firms in downstream industries. On average, a $1 increase in CDB SOE loans leads to a $0.20 decrease in private firms assets. Moreover, CDB infrastructure loans crowd in private firms. I use exogenous timing of municipal politicians turnover as an instrument for CDB credit flows. GOVERNMENT-DIRECTED LENDING PROGRAMS are pervasive around the world and are often justified as a way to support economic development. 1 A central question in the debate on government credit is whether it crowds out or crowds in private-sector activities. The theoretical literature suggests that government credit can have many countervailing effects. On the one hand, government credit that supports high social return projects such as infrastructure can have positive spillover effects (e.g., Stiglitz (1993)). On the other hand, government credit might crowd out more productive private-sector investments (e.g., King and Levine (1993a, 1993b)). Due mainly to data limitations, previous empirical Hong Ru is with Nanyang Technological University. I am indebted to my advisors Nittai Bergman, Andrey Malenko, Robert Townsend, and especially Antoinette Schoar for their invaluable guidance and encouragement. I thank Jean-Noel Barrot, Stephen Dimmock, Wei Jiang, Mark Kritzman, Chen Lin, Deborah Lucas, Eric Maskin, Stewart Myers, Tran Ngoc-Khanh, Michael Roberts (the Editor), Stephen Ross, Zheng Song, Richard Thakor, Wei Wu, an associate editor, and two anonymous referees. This paper benefited greatly from seminar participants at Cornell, HBS, MIT, NTU, NUS, Olin, Rotman, SMU, Texas A&M, UIUC, and Wharton for insightful comments. I thank the discussants and participants at the CFRC, SFS Cavalcade, SEFM, and TCGC conferences. I am also grateful to Gao Jian. I thank Yue Wu and Endong Yang for their excellent research assistance. I thank all anonymous local government officials in China for long and engaging discussions. I thank Haoyu Gao for help on the CBRC data, Nanyang Technological University for financial support, and the China Development Bank access to its data. To my knowledge, there is no financial or other conflict of interest relevant to the subject of this paper. All errors are my own. 1 Lucas (2014) states that the total amount of credit supported by OECD (Organisation for Economic Co-operation and Development) governments was recently estimated at several tens of trillions euros, and Elliott (2011) states that in 2010 the U.S. government s outstanding commitments for loans and guarantees totaled approximately $2.3 trillion, which was roughly one-third the size of the loans of all U.S. banks combined. DOI: /jofi

2 276 The Journal of Finance R studies have only been able to explore the net effects of these opposing forces and revealed mixed evidence. Using detailed data from China Development Bank (CDB) on different types of government credit, in this paper, I aim to separate these countervailing channels of government credit by tracing its effects across different levels of the supply chain. The CDB is the largest policy bank in China and lends mainly to state-owned enterprises (SOEs) in strategic industries (e.g., energy and mining) and to local governments for infrastructure development. The CDB loan data contain outstanding loan amounts and issuance amounts at the province-industry level between 1998 and I document two main findings. First, CDB loans to SOEs crowd out private firms in the same industry as indicated by decreases in asset investment, employment, and sales, but they crowd in private firms in downstream industries. More efficient private firms in downstream industries can benefit significantly more from CDB credit to upstream SOEs. Second, CDB loans to local government infrastructure projects have positive (crowding-in) effects on private firms activities. By disentangling the different forces of government credit in China, this analysis sheds light on the mixed results of previous studies regarding the net effects of aggregate government credit. To establish the causal effects of CDB credit on firm activities, I exploit exogenous variation in CDB credit flows using predetermined political turnover cycles of municipalities in China, which occur every five years on average. This allows me to alleviate the concern that the CDB endogenously targets areas with specific economic needs for credit. For example, the CDB may maximize spillover effects by strategically lending to industries in which downstream private firms have high growth potential. In particular, I use the predicted timing of municipal government turnover as an instrument for CDB loans. In China, city secretaries are appointed and typically serve a five-year term. 3 Moreover, cities in China have their own five-year turnover cycles. This allows me to exploit variation in different five-year cycles across cities. Instead of using actual turnover cycles, I take the first year of the secretary in the previous term and add five years to calculate the predicted first year of the current city secretary. 4 These predetermined predicted municipal turnover cycles depend solely on past information and hence are not affected by current economic factors (e.g., local GDP, employment, and income). I begin the analysis by regressing the amounts that cities borrow from the CDB on the predicted turnover cycles. I find a zig-zag borrowing pattern in most cities whereby city secretaries borrow significantly more from the CDB during their predicted first year and decrease borrowing monotonically as their 2 I categorize CDB loans into two groups: industrial loans to firms and loans to infrastructure projects. Among CDB industrial loans to manufacturing firms, approximately 95% go to SOEs. I consider CDB industrial loans as SOE loans in this paper. See the more detailed discussion in Section II.A. 3 In China, the political leader of a municipal government is called the Secretary of the Municipal Committee of the Communist Party of China (CPC) (equivalent to a mayor in the United States). 4 I follow the strategy in Shue and Townsend (2014) of using predicted cycles as instruments.

3 Government Credit, a Double-Edged Sword: 277 terms progress. Borrowing rises again in the predicted first year of the next city secretary. On average, city secretaries reduce CDB outstanding loan amounts by 15.4% each year during their tenure. This zig-zag borrowing cycle is driven primarily by career concerns of city secretaries. In China, the promotion of local politicians depends largely on their GDP performance (e.g., Li and Zhou (2005)). To increase local GDP quickly over the short term, city secretaries tend to borrow from the CDB and in turn invest as much and as early as possible during their terms. I next investigate the heterogeneous effects of CDB loans on the private sector across different levels of the supply chain. First, I examine the effects of CDB industrial SOE loans on private-sector firms in the same industry. I use city-level turnover timing as an exogenous shock to CDB credit at the province-industry level. In particular, I identify each city s largest SOE industry (i.e., focal industry), which does not change much over time. 5 I then interact dummies of the predetermined focal industry in each city with its predicted turnover cycle. Using these interactions as instruments for CDB provinceindustry loan amounts, I perform two-stage least squares (2SLS) analysis. If the city secretary is in an early year of the term, I consider this a shock to the province-level CDB loans in the city s focal industry. In the first-stage regression, I find that the province-level CDB loan amount in an industry is 41.3% higher when the corresponding city secretary is in the first two years of the term. In the second-stage regressions, I find that, consistent with the OLS regression results, a 100% increase in CDB SOE loan amount to the focal industry leads to a decrease in the assets, employment, and sales of private firms in the same focal industry and province of 2%, 1.7%, and 4.1%, respectively. By contrast, increasing CDB SOE loans leads to increases in SOEs activities. Second, I examine the effects of CDB SOE loans in upstream and downstream industries. Using an input-output matrix for China, for each focal industry I identify its downstream industries that source the majority of their inputs from it. On average, each focal industry has 2.3 downstream industries. I find that a 100% increase in CDB loan amount to the focal industry leads to an increase in the assets, sales, and sales per worker of downstream private firms in the same province of 3.4%, 2.6%, and 2.6%, respectively. The evidence also suggests that more efficient private firms capture significantly more benefits from these CDB upstream industrial loans. In sum, although CDB industrial SOE loans crowd out private firms in the same industry (i.e., the focal industry), they crowd in private firms in downstream industries. These opposing effects could explain the mixed empirical findings in previous studies that use aggregate data on government credit. For the exclusion condition of the instruments, I find that other channels through which the city secretary may influence a city s business (e.g., borrowing from other banks, selling more land, requesting fiscal transfers, and better enforcing tax treaties) are not correlated with the turnover cycles. In particular, 5 The distribution of SOE industries across cities is predetermined mainly for historical reasons and remains stable over time. See a detailed discussion in Section III.C.

4 278 The Journal of Finance R the joint F-tests of predicted turnover cycle dummies are not significant in regressions of these other channels. I further find that, for cities with potentially better access to CDB credit, the effects of the predicted turnover cycles are significantly more pronounced. These findings, together with CDB SOE loans opposing effects on private firms and SOEs in focal industries, alleviate the concern that political turnover cycles are associated with other unobservable factors that might affect firm activities (e.g., changes in a city s investment opportunities over its turnover cycle). Moreover, political turnover cycles affect firms not only through CDB SOE loans but also through CDB infrastructure loans. However, the effects of CDB SOE loans across different levels of the supply chain are not likely to be confounded with the effects of CDB infrastructure loans. First, infrastructure projects would impact most firms in the area, whereas the crowding-out and crowding-in effects of CDB SOE loans depend on the industrial levels of the supply chain. Second, the distributions of CDB infrastructure versus SOE loans differ across cities. For example, in 2002, for half of the cities, more than 90% of their CDB loans were issued to SOEs, while the other cities borrowed mainly for infrastructure. I define the dummy SOECity by setting it equal to one if the city s SOE assets ratio is above the median in Using the interactions between the predetermined SOECity dummies and the predicted municipal turnover cycles as instruments, I break down various borrowing patterns of CDB industrial SOE loans and infrastructure loans in the first stage. In the second stage, I again find that increasing CDB industrial SOE loans leads to decreases in private firms assets, sales, and sales per worker. By contrast, increasing CDB infrastructure loans leads to increases in private firms assets, employment, debt, and total sales. These opposing effects provide further evidence that the instruments satisfy the exclusion condition. Finally, based on the regression coefficients, I perform a back-of-the-envelope calculation to estimate the overall effects of CDB loans on individual firms. I multiply the growth of different types of CDB loans (i.e., infrastructure loans, industrial SOE loans, and upstream loans) by the estimated coefficients. On average, a $1 increase in CDB SOE loans leads to a $0.20 decrease in private firm assets. This indicates that the crowding-out effects are larger than the crowding-in effects. Moreover, a $1 increase in CDB infrastructure loan amounts leads to a $0.47 increase in private firms assets. This paper adds to the literature that examines whether government credit and spending crowds the private sector in or out. As noted above, findings on this question have been inconclusive in both theoretical and empirical studies. On the one hand, the social view (e.g., Stiglitz (1993)) argues that governments should allocate funds to high social return projects with positive externalities when returns from such loans are difficult for private banks to capture. 6 On the other hand, many other studies argue that government credit will crowd 6 The social view of government intervention is suggested by Atkinson and Stiglitz (1980), who argue that SOEs can be justified under market failures (e.g., Stiglitz and Weiss (1981), Greenwald and Stiglitz (1986)).

5 Government Credit, a Double-Edged Sword: 279 out private-sector investment, particularly when the subsidized credit is given to firms with distorted incentives, such as SOEs (e.g., King and Levine (1993a, 1993b), Demirguc-Kunt and Maksimovic (1998), Rajan and Zingales (1998), La Porta, Lopez-de-Silanes, and Shleifer (2002)). Previous empirical studies use aggregate data to estimate the overall net effect of government credit and show either crowding-in or crowding-out effects. 7 The major contribution of this paper is to disentangle these different forces of government credit. The results suggest that it is important to consider the heterogeneous effects of government credit across different levels of the supply chain and different types of credit. Such analysis could help researchers and policy makers fully understand the economic consequences of government credit. 8 This paper also relates to another strand of the literature on political business cycle. Starting with Nordhaus (1975) and followed by many other studies, this literature argues that politicians capture firms or use economic policy to increase their chances of election. 9 I show suggestive evidence that politicians career concerns play a role in credit allocation in China, as they do in some other countries. This leads to the zig-zag borrowing pattern over political turnover cycles that helps me tease out the exogenous variation in CDB credit flows to establish the causal effects of government credit, which is another challenge in the literature. 10 The remainder of the paper is organized as follows. Section I provides background on the CDB and local government debt in China. Section II describes the data. Section III summarizes empirical analysis and presents the results. Section IV concludes. I. Background: The CDB and Local Government Financing in China A. History of the CDB The CDB was established in 1994 out of six State Planning Commission (SPC) Investment Corporations. 11 Among policy banks, the CDB is the largest with US$1.83 trillion total assets in The CDB s two main shareholders 7 See, for example, Gale (1991), Schwarz (1992), Ramey and Shapiro (1998), Burnside, Eichenbaum, and Fisher (2004), Craig, Jackson, and Thomson (2007), Shaffer and Collender (2009), Cohen, Coval, and Malloy (2011), Banerjee, Duflo, and Qian (2012). 8 The CDB focuses on strategic industries at the top of the supply chain and infrastructure development. The positive spillover effects of CDB infrastructure credit and credit to upstream industries are consistent with the social view of government credit. 9 See, for example, MacRae (1977), Kornai (1979), Alesina and Sachs (1988), and Shleifer and Vishny (1994). 10 Many empirical studies use political turnover cycle as a source of exogenous variation to identify causal effects (e.g., Sapienza (2004), Dinc (2005), Khwaja and Mian (2005), Bertrand et al. (2007), Cole (2009), Dinc and Gupta (2011), Carvalho (2014)). 11 The SPC is under the Chinese State Council and has broad administrative and planning control over the Chinese economy. These six Investment Corporations were policy institutions established in the late 1980s as long-term investment instruments on behalf of the government.

6 280 The Journal of Finance R are the Ministry of Finance and the China Investment Corporation. 12 The CDB can thus be viewed as an extension of the government s fiscal function. The CDB is fully state-owned, which is similar to state-owned commercial banks such as the ICBC, CCB, BOC, and ABC, but the CDB s lending strategy distinguishes it from commercial banks. 13 The CDB typically covers infrastructure sectors and uncontested markets in which commercial banks have little interest for three possible reasons. First, the CDB s mandate locates the bank in policy-related areas. In particular, the main role of the CDB is to grant subsidized credit to infrastructure projects in undeveloped and underdeveloped areas in China (such as the provinces in western China) and to SOEs in strategic industries, whereas commercial banks weigh their businesses more heavily in the wealthier areas of China (such as the eastern coastal provinces). Second, as a policy bank, profit maximization is not on the CDB s agenda; in contrast, although commercial banks in China are also state-owned, profit is one of their primary performance targets. Third, the CDB finances its loans by issuing long-term bonds with sovereign ratings, whereas state-owned commercial banks rely primarily on short-term deposits. Therefore, the CDB engages in long-term lending that not only caters to infrastructure-sector requirements but also matches the durations of its assets and liabilities. 14 B. Relationship between the CDB and Local Governments The CDB generally has closer relationships with local governments than commercial banks do. Since 1989, budgetary law has prohibited local governments in China from incurring debt. However, under the tax-sharing system, local governments retain only approximately 30% of tax revenue. As a result, while local governments are responsible for infrastructure development, they do not have the money to do so. To solve this dilemma, in 1998 the CDB began to work with local governments to help them create 100% state-owned companies as their borrowing platforms. Local governments are thus able to use these companies to borrow from banks off the balance sheet. In November 2008, commercial banks began to lend to local governments aggressively, as part of a four-trillion RMB stimulus plan. At the conclusion of this stimulus program in 2010, many commercial banks pulled back, but the CDB continued to increase its lending to local governments. Between 2006 and 2013, on average the CDB contributed approximately 50% to 60% of the outstanding loans of local governments in China. In contrast, on average each commercial bank in China accounts for less than 3% of total local government 12 The China Investment Corporation is a sovereign wealth fund responsible for managing part of China s foreign exchange reserves. 13 ICBC stands for Industrial and Commercial Bank of China; CCB for China Construction Bank; BOC for Bank of China; and ABC for Agricultural Bank of China. 14 The CDB s long-term loan rates have remained lower than those of state-owned commercial banks and much lower than those of private or shareholding commercial banks. The subsidized loans from the CDB can thus be viewed as government spending (Lucas (2012)).

7 Government Credit, a Double-Edged Sword: 281 debt. Even the big five commercial banks contribute approximately 6% each to total local government debt (Gao, Ru, and Tang (2016)). Compared with commercial banks, the CDB is a long-term, stable financial resource for local governments. This paper focuses on the period from 1998 to 2009, which overlaps the stimulus plan by only one year. 15 During the sample period, the CDB played an important role in local government borrowing, and local politicians strongly affected CDB credit allocation. Commercial banks also lend massive amounts to SOEs and typically have close connections with SOE executives who do not follow five-year turnover cycles, which helps explain why CDB loans and not commercial bank loans are sensitive to local government turnover cycles. II. Data Description A. CDB Loan Data The proprietary loan data in this paper, obtained directly from the CDB, contain both city- and province-level loan data. At the province level, the CDB data contain monthly aggregate outstanding loan amounts and loan issuances in 95 industries for each of the 31 provinces in mainland China from 1998 to The industries include infrastructure sectors (such as road, air, rail transportation, water supply, and public facilities) and industry sectors (such as agriculture, tobacco, software, oil refining, and textiles). At the city level, the CDB data contain annual aggregate outstanding loan amounts and loan issuances to both infrastructure projects and industrial firms across 310 cities in China from 1998 to I obtain city- and province-level economic variables (such as GDP, income per capita, total employment, and fiscal income) from the China Statistical Yearbook. Table I presents summary statistics and Table AI in the Appendix provides detailed variable definitions. For CDB city-level loan data, reported in Table I, Panel A, the average total outstanding loan amount is 3.2 billion RMB per city-year. Among these loans, the average outstanding loan amount for infrastructure is 1.1 billion RMB and the average outstanding loan amount for industrial firms is 2.1 billion RMB. Panel B of Table I shows that, for CDB province-industry-level loan data, the average outstanding loan amount is 0.8 billion RMB per province, industry, and year. 17 Figure 1 plots the ratio of city-level infrastructure loans and industrial loans to total loan amount. For infrastructure loans, the ratio was almost zero from 1998 to 2000, after which it began to increase and the gap between the top 15 I repeat the main analysis in the paper by restricting the sample period to 1998 to 2008; the results remain the same. Please see Table IAIII in Internet Appendix for details. The Internet Appendix may be found in the online version of this article. 16 The set of cities does not include Beijing, Shanghai, Tianjin, or Chongqing, which are classified as provinces. 17 Figure A1 in the Appendix plots the time trend of CDB province- and city-level outstanding loan amounts.

8 282 The Journal of Finance R Table I Summary Statistics Panel A provides summary statistics at the city year level for CDB city-level loan data, municipal local economic data, and municipal politician profile data. The data cover 310 cities between 1998 and Panel B provides summary statistics at the province industry year level for the CDB province-level loan data, which cover 31 provinces and 95 industries between 1998 and Panel C provides summary statistics at the firm year level for the Chinese Industrial Census data between 1998 and See Table AI for detailed variable definitions. Panel A: City Data Variable N Mean SD Min Max Loan_City 3, , Issuance_City 3, Loan_INF_City 3, Issuance_INF_City 3, Loan_IND_City 3, Issuance_IND_City 3, ActualTurnover 3, PredictedTurnover 3, GDP 3, , AvgIncome 3,568 10, , , FiscalIncome 3, , Employment 3,579 1, , , Age 3, Gender 3, Relation 3, Promotion 3, Panel B: Province Data Loan_PI 44, , Issuance_PI 44, , Loan_Road Loan_Rail Loan_Water Loan_Tel Panel C: Firm Data LogAssets 2,949, LogWorkers 2,944, LogDebt 2,930, ROA 2,949, Log(Sales/W) 2,918, LogSales 2,931, Tax_Corp 1,520, Tax_VAT 1,356, and bottom quartiles of the infrastructure loan ratio widened from 1999 to 2003 but closed a bit after The bottom panel of Figure 1 reveals similar patterns for the industrial SOE loan ratio, which indicates that different cities have different combinations of infrastructure and industrial SOE loans.

9 Government Credit, a Double-Edged Sword: 283 Figure 1. Distribution of CDB infrastructure loans versus industrial SOE loans. This figure plots the distribution of CDB infrastructure loans to total city-level loans (top panel) and the distribution of CDB industrial loans to total city-level loans (bottom panel) across 310 cities. Infrastructure includes transportation (e.g., road, railway, airport, bridge, and tunnel), water supply, energy supply (e.g., gas, electric), telecommunications, and public service (e.g., sewage discharge). City-level loans do not include province-level projects even if part of such projects may be located in the city, such as a highway. The solid lines represent the median ratios across 310 cities and the dashed lines are the top and bottom quartiles of the ratios across 310 cities each year. (Color figure can be viewed at wileyonlinelibrary.com) In particular, CDB industrial loan amounts are significantly higher in cities with more SOEs. These cities with large state-owned sectors borrow relatively less for infrastructure. The distribution of SOEs across cities is predetermined by historical factors. For example, Baotou has large rare-earth mining SOEs because it has rich rare earth resources. I construct the dummy SOECity by

10 284 The Journal of Finance R setting it equal to one if the weight of SOE assets in a city was greater than the median level across all 310 cities in I then interact this predetermined dummy with predicted municipal turnover cycles to capture differences in borrowing patterns between infrastructure and industrial SOE loans across cities. In other words, using this dummy, I explore the exogenous variation in how a city chooses to allocate CDB credit between infrastructure and industrial SOEs. The CDB collaborates primarily with local governments to grant credit. CDB infrastructure loans are lent directly to local governments. For CDB industrial loans to firms, local governments also play a key role, as most of CDB s industrial loans go to local SOEs. The China Banking Regulatory Commission (CBRC) loan-level data, which cover the 2007 to 2012 period, allow me to examine the allocation of CDB industrial loans between SOEs and private firms. 18 Based on the CBRC loan-level data and Chinese Industry Census (CIC) data, I find that, among manufacturing firms, approximately 95% of CDB industrial loans go to SOEs. Only 5% of CDB industrial loans go to private firms. Private firms with CDB loans are typically large and have close connections with the government, and some of these private firms with CDB loans have been privatized from SOEs. For example, Huawei, the largest telecommunications equipment manufacturer in China, was state-owned in 1998 and has been borrowing from the CDB. When I exclude these privatized firms from the sample, only 2% of CDB industrial loans go to the private sector. In sum, the CDB rarely lends to small entrepreneurs in China, and thus in this paper I consider CDB industrial loans as SOE loans. Another caveat in the analysis using aggregate CDB loan data is that only a few SOEs or small SOEs borrow from the CDB, which might lead to biased results because aggregate CDB credit to SOEs does not well represent the majority of SOEs in a certain industry and province. To mitigate this concern, I use CBRC loan-level data to calculate the coverage of CDB credit at the industry-province level. In particular, I first merge the CBRC loan data with CIC manufacturing firm data to get each firm s outstanding loan amount from the CDB. I then aggregate the assets of SOEs with outstanding loans from the CDB and divide by the total assets of all SOEs at the province-industry level. This ratio is approximately 86%, which means that the majority of the large SOEs have been borrowing from the CDB. The aggregate outstanding CDB loan data in this paper provide a good representation of SOEs in China. This mitigates the concern that the crowding-out and crowding-in effects of CDB SOE loans come from only a few small SOEs. B. Politician Profile Data I manually collect local politicians profiles from the Zechen Database and the Baidu Encyclopedia. The data contain the names of all city mayors and 18 The CBRC loan-level data record all loan issuances from the 19 largest Chinese banks including the CDB between 2007 and This data set has been used in other studies, for example, Ai et al. (2016) and Gao, Ru, and Tang (2016).

11 Government Credit, a Double-Edged Sword: 285 secretaries at the city-month level across all 334 cities in China between 1949 and The data also contain politician demographics such as gender, age, and birthplace. In total, the data contain information for 1,227 city secretaries. I cross-check these data with other sources to ensure their quality. When I merge the CDB city-level data with politicians profiles, I obtain a sample of 310 cities in total (the remaining 24 cities did not receive CDB loans). In China, the political leader of a municipal government is called the Secretary of the Municipal Committee of the CPC. City secretaries terms are generally five years. The national political turnover cycle is also five years, and occurs around the National Congress of the CPC. Many cities in China were built during the 1990s due to the urbanization process. For a new city, the secretary begins the five-year tenure in the year the city was incorporated, which does not coincide with the national cycle. From 1998 to 2010, 67% of the 310 sample cities are off the national turnover cycle, which allows me to explore variation in different five-year cycles across cities in China. During municipal political turnover, since there are no elections in China, promotion decisions involving local politicians are frequently made by higher level Communist Party officials (such as province governors). At the end of these terms, approximately 38% of city secretaries were promoted, 14% of city secretaries were transferred to other cities for another term as city secretary, and the remainder were no longer city secretaries for various reasons (e.g., retiring, under arrest, or serving as another type of government official). I construct the dummy Promotion by setting it equal to one if a city secretary was appointed to a higher position in the political hierarchy at the end of her term. In China, political turnover itself is endogenous as politicians are assigned by the CPC instead of being elected by voters. For example, those politicians with good connections can be assigned to better cities and can borrow more from the CDB. Notably, the exogenous variation in CDB credit flows I exploit come from predetermined five-year political turnover cycles rather than from turnover itself. C. CIC Data The firm-level data come from the CIC data collected by the Chinese National Bureau of Statistics (NBS). The CIC data, which are widely used in many academic studies (e.g., Hsieh and Klenow (2009) and Song, Storesletten, and Zilibotti (2011)), cover all manufacturing firms in China with annual sales over five million RMB (about US$700,000) from 1998 to The CIC features detailed annual accounting data and firm characteristics, such as number of workers, industry categories, locations, registration types, political hierarchies, government subsidies, and wages. In total, the CIC contains information on 711,892 firms. The CIC appears to be the most detailed database on Chinese manufacturing firms, and the content and quality of the database are both sufficient. Using the CIC data, I classify firms as SOEs based on their annual registration types. In particular, I classify two registration types as SOEs: firms owned by government department and collective-owned enterprises (COEs). In

12 286 The Journal of Finance R China, the first type comprises firms in which the majority of shares are owned by government departments (e.g., State-Owned Assets Supervision and Administration Commission, Ministry of Finance, Ministry of Transport), while COEs comprise firms in which assets are owned collectively by all residents in a community (e.g., cities, counties, and villages). These COEs are mainly owned and controlled by local governments. Officially, these two types of firms are considered public. 19 Approximately 9% of private firms in the CIC data were privatized from SOEs. Consistent with other studies that use the CIC data, I consider these firms as private firms. In a robustness test, the results of this paper continue to hold when I reclassify these privatized firms as SOEs. 20 Table I, Panel C reports the summary statistics for the CIC data. III. Empirical Analysis and Results A. Career Concerns of Local Politicians Promotion is one of the most important career aspirations of politicians in China (Maskin, Qian, and Xu (2000)). When China began its economic reforms in 1978 under Deng Xiaoping, local governments began to play an increasingly important role in developing the local economy and local government officials became increasingly accountable for local economic growth. On December 9, 2013, the Organization Department of the Central Committee of the CPC announced that, from that point forward, local political promotion would depend on performance on various factors such as environmental protection. Prior to this change, however, GDP served as the primary measure of local politicians performance. Thus, during the sample period of this paper (i.e., 1998 to 2009), promotions of city secretaries and mayors in China depend largely on local GDP growth. Many empirical studies find evidence of GDP s importance for promotion decisions prior to the 2013 change. For example, Li and Zhou (2005) find that the likelihood of promotion of Chinese provincial leaders increased with economic performance (i.e., local GDP growth over politicians tenure) between 1979 and Under this promotion system, local politicians in China are incentivized to invest in an effort to boost local GDP during their five-year terms. As promotion decisions for city secretaries are made in the last or second-to-last year of their term, local GDP growth is typically more heavily weighted in the first three or four years of their terms because GDP data generally take a few months to be released. City secretaries therefore have strong incentives to increase local GDP as much and as soon as possible. Growth in GDP is fueled mainly by borrowing from the CDB. To verify this hypothesis, I regress the promotion 19 The NBS in China officially categorizes firms into eight types where government-owned and collective-owned enterprises are considered public firms. 20 In particular, I repeat the analysis in Table V after excluding privatized firms from the private sector. The results remain the same. Please see Table IAV in Internet Appendix for more details.

13 Government Credit, a Double-Edged Sword: 287 probability on the increase in CDB loans using the following Probit model: Promotion i, j = α + β 1 Loan Increase t,i, j + β 2 Relation i, j + β 3 Age i, j + β 4 Gender i + ε i, (1) where Promotion i, j is a dummy variable indicating whether city secretary i in city j is promoted during the turnover year, and Loan Increase t,i, j is the logarithm of the increase in CDB loan outstanding from secretary i s first year in city j in year t. Isett = 1, 2, 3, 4, 5 to examine the effects of increases in the CDB loan amount at various points in a city secretary s term. Relation i, j is a dummy indicating whether city secretary i in city j is from the same hometown as the provincial governor, Age i, j is the age of secretary i in city j during the turnover year, and Gender i is a dummy indicating whether city secretary i is female. Standard errors are clustered at the city level. Note that this estimation is informative about the correlation between CDB credit and local politician promotions; it does not provide evidence on a causal relation. Table AII shows that CDB loan increases are positively associated with promotion probabilities, and that this effect is driven primarily by loan increases during the first two years of a secretary s term. In columns (1) and (2) of Table AII, the coefficients on loan increases are and 0.087, respectively, both of which are significant. When I include later years in a secretary s term (columns (3) to (5)), the coefficients are lower and less significant. This suggests that CDB loans in particular taken in the early years of local politicians terms may help their careers. During the most recent 15 years in China, borrowing from the CDB has been the primary mechanism used by city secretaries to boost local GDP. Because the loans take time to affect the economy, city secretaries typically borrow from the CDB as early as possible. This is consistent with the findings in Li and Zhou (2005) that local GDP growth is an important determinant of politicians promotions in China. B. CDB Credit Flows and the Timing of Political Turnover In this subsection, I investigate the effects of the timing of municipal politician turnover on borrowing from the CDB. Instead of actual turnover cycles, I employ predicted turnover cycles using past turnover in a city to predict future city secretary terms. This approach mitigates the concern that the actual timing of city secretary turnover may be affected by other factors. For example, new provincial governors tend to replace city secretaries with their own people. Since a provincial governor with greater political power might be assigned to a province with better investment opportunities and might have greater access to CDB credit, the turnover timing of city secretaries in this province may be correlated with local investment opportunities. To predict the first years of city secretaries terms, I use the following simple prediction algorithm. For each city secretary s term, let y be the first year of the previous city secretary s term, which is also the turnover year of the previous cycle. I predict that year y + 5 will be the first year of the current city secretary.

14 288 The Journal of Finance R Figure 2. Predicted political turnover of city secretaries in China. This figure plots the distribution of predicted city secretaries term length in China. The data cover 334 cities and 1,227 city secretaries from 1997 to Term length is calculated at the city-politician level from the predicted turnover cycles of city secretaries. Approximately 46% of the city secretaries end their terms in the fifth year. If there was no previous cycle, I use the actual first year of the city secretary as the predicted year. For example, to predict the first year of city secretary A, I begin with the term of secretary A s predecessor, secretary B. If secretary B began her term in 2000, I mark 2005 as the predicted first year of secretary A. Overall, I correctly predict approximately 60% of the cycles actual first years. Approximately 28% of the predictions occur within a year of the actual first year. Approximately 12% of the predicted first years differ from the actual first years by more than one year. These incorrect predictions result from actual cycles with lengths that are less than five years. Figure 2 plots the length of predicted city secretary terms. Approximately 46% of the predicted terms are five years. The predicted turnover cycles depend solely on past information and tease out exogenous variation from the actual turnover cycles that might be correlated with concurrent economic conditions. I use the Cox proportional hazard model to test whether predicted political turnover timing is affected by factors such as current local economic conditions and politicians demographics. Table II shows that one-year-lagged local economic conditions (such as GDP, household income, fiscal income, employment, and the CDB s outstanding loans) are uncorrelated with predicted turnover timing. The NationalCycle dummy is positively related to predicted turnover because 33% of city turnovers occur during national turnover years. The secretary s age positively affects predicted turnover timing because city secretaries are more likely to retire as they get older. In columns (2) and (3), I separate the sample into those politicians who are promoted and those who are not (e.g., lateral transfers and demotions). The results remain the same in these two subsamples. Next, I explore borrowing patterns over various periods of a city secretary s term. The CDB s primary lending method involves coordinating with local

15 Government Credit, a Double-Edged Sword: 289 Table II City Secretary Turnover Timing and Local Conditions This table presents results of Cox proportional hazard regressions following Wooldridge (2002). Data are restricted to 1,106 city secretaries across 310 cites between 1998 and The origin and failure events are the starts and ends of each city secretary s term. Estimated coefficients are reported in columns (1) to (3). GDP, AvgIncome, FiscalIncome, and Employment are city-levelgdp, income per capita, fiscal income, and total number of employees, respectively. Log(Loan City) is the logarithm of CDB city-level total loans outstanding. Age is the city secretary s age. Gender is a dummy for whether the city secretary is female. NationalCycle is a dummy for whether the year is a national turnover year (e.g., National Congress Party in 1998, 2003, and 2008). Column (1) covers the full sample. Column (2) focuses on city secretaries who were not promoted at their turnover. Column (3) focuses on city secretaries who were promoted at their turnover. Standard errors are clustered at the city secretary level. T-statistics of the coefficient estimates are reported in parentheses. ***, **, * indicate statistical significance at the 1%, 5%, and 10% level, respectively. (1) (2) (3) Coefficient Coefficient Coefficient GDP t (0.090) (0.113) (0.128) Urban_Income t (0.004) (0.006) (0.011) Fiscal_Income t (1.019) (1.801) (1.519) Labor t (0.269) (3.677) (9.390) Log(Loan_City t 1 ) (0.020) (0.027) (0.032) NationalCycle 0.343*** 0.378*** 0.339*** (0.067) (0.093) (0.110) Age 0.029*** 0.044*** 0.025* (0.008) (0.013) (0.014) Gender (0.176) (0.364) (0.199) Observations 2,812 1, χ governments to support infrastructure projects and SOEs. The city secretary is the top-ranking politician in the city and typically plays a large role during the lending process. The regression that I employ is as follows: Log Loan j,t = α + β 1 Year 1 i, j,t + β 2 Year 2 i, j,t + β 3 Year 3 i, j,t + β 4 Year 4 i, j,t + β 5 Year 5 i, j,t + β 6 Year 6 i, j,t + X Control j,t 1 + Fixed Effects + ε j,t. (2) The dependent variable in equation (2), LogLoan j,t, is the logarithm of the CDB loans outstanding (for industrial SOE loans, infrastructure loans, or total loans) in city j in year t. The variables Year 1 i, j,t,..., Year 6 i, j,t are dummies indicating those years in which secretary i was in city j in year t. For example, Year 1 i, j,t

16 290 The Journal of Finance R equals one for the first year secretary i was in city j and zero otherwise. Control j,t 1 is a matrix of variables that control for economic conditions, such as GDP, urban income per capita, fiscal income, and the size of the working population; X is the vector of coefficients on these control variables. The fixed effects include year fixed effects, which mitigates the concern that national turnover cycles drive the borrowing patterns; city fixed effects, because cities are often characterized by unique situations; and secretaries personal fixed effects, because secretaries have their own investment styles. Standard errors are clustered at the city level. Table III, Panel A, presents the regression results. Columns (1) to (3) employ actual turnover cycles. In column (1), the dependent variable is the logarithm of total CDB city outstanding loans. The dummy Year 1 is the missing category. The coefficient on Year 2 is 0.386, with a significance level of 1%, which implies that on average city secretaries borrow 38.6% less during their second year than during their first year. The coefficients on Year 3, Year 4, Year 5, and Year 6 are 0.749, 1.071, 1.429, and 1.900, respectively. Thus, the amount borrowed from the CDB decreases monotonically in the tenure of a city secretary, that is, city secretaries borrow more heavily as soon as they take office and then decrease their borrowing monotonically over the duration of their term. In Panel B, I employ Turnover, which equals the number of years that the secretary was in the city, as the independent variable. In line with Panel A, I find that city secretaries borrowing from the CDB decreases over their tenure: on average, if a city secretary stays one more year, borrowing from the CDB decreases by 36.4%. In columns (2) and (3), I separate loans into infrastructure and industrial SOE loans, respectively. Both measures have patterns similar to that in column (1), with the patterns stronger for industrial SOE loans. Columns (4) to (6) in Table III present predicted turnover cycles. The patterns are similar to actual turnover cycles but are a bit weaker. For example, on average, if a city secretary stays one more year, borrowing from the CDB decreases by 15.4%, which is smaller than the effect for actual turnover cycles because some predicted turnover cycles are incorrect (i.e., different from the actual cycles). These prediction errors eliminate potential endogeneity in turnover timing while lowering the instruments power. I use these predicted turnover cycles in all of the empirical exercises below. Figure 3 plots the average logarithm of total CDB loans made to cities after eliminating year, city, and politician fixed effects. There are three national turnover cycles during my sample period: 1998 to 2002, 2003 to 2007, and 2008 to I cluster cities using these three cycles and calculate the average logarithm of total CDB city loan amounts for each year in a predicted five-year cycle. Figure 3 displays a zig-zag pattern during these three cycles. On average, city secretaries borrow significantly more during their predicted first year in office, with this amount decreasing monotonically over time. When a new city secretary comes in, borrowing spikes again, in line with the results in Table III. I also examine borrowing patterns for each individual

17 Government Credit, a Double-Edged Sword: 291 Table III City Secretary Turnover and Borrowing from the CDB This table presents results for regressions of CDB loan amounts on city secretary turnover cycles estimated at the city year level for 310 cities from 1998 to Panel A presents results for the OLS regression in equation (2). Year 2 is a dummy for the second year in a city secretary s term. Year 3toYear 6 are defined similarly. The dummy for year 1 is the missing category. In Panel B, Turnover is the number of years a secretary has served the city. Columns (1) to (3) give the effects of the actual turnover cycles on city-level CDB total loans outstanding, infrastructure loans, and industrial SOE loans, respectively. Columns (4) to (6) give the corresponding effects of the predicted turnover cycles. LogLoan = log(loan City), which is the logarithm of city-level CDB total loans outstanding. LogLoanINF = log(loan INF City), which is the logarithm of infrastructure loans. LogLoanIND = log(loan IND City), which is the logarithm of industrial SOE loans. Control variables include city-level GDP, income per capita, fiscal income, and total number of employees in year t 1. All columns also control for year, city, and politician fixed effects. Standard errors are clustered at the city level. T-statistics of the coefficient estimates are reported in parentheses. ***, **, * indicate statistical significance at the 1%, 5%, and 10% level, respectively. Actual Turnover Predicted Turnover Dependent (1) (2) (3) (4) (5) (6) Variable LogLoan LogLoanINF LogLoanIND LogLoan LogLoanINF LogLoanIND Panel A: Turnover Dummies Year_ *** 0.109*** 0.330*** 0.182*** 0.117*** 0.149*** (0.031) (0.028) (0.044) (0.030) (0.032) (0.040) Year_ *** 0.240*** 0.703*** 0.303*** 0.176*** 0.336*** (0.042) (0.035) (0.064) (0.034) (0.039) (0.051) Year_ *** 0.351*** 1.001*** 0.473*** 0.231*** 0.495*** (0.063) (0.041) (0.086) (0.037) (0.044) (0.052) Year_ *** 0.467*** 1.341*** 0.635*** 0.343*** 0.645*** (0.108) (0.081) (0.151) (0.071) (0.067) (0.095) Year_ *** 0.691*** 1.668*** 0.729*** 0.451*** 0.680*** (0.144) (0.110) (0.204) (0.116) (0.096) (0.159) Control&FE Yes Yes Yes Yes Yes Yes Observations 3,127 2,449 2,727 3,161 2,429 2,759 R Panel B: Turnover Years Turnover 0.364*** 0.122*** 0.338*** 0.154*** 0.080*** 0.160*** (0.020) (0.013) (0.028) (0.012) (0.012) (0.015) Control&FE Yes Yes Yes Yes Yes Yes Observations 3,127 2,449 2,727 3,161 2,429 2,759 R city and find that most cities follow this zig-zag pattern. This result alleviates the concern that certain cities with extreme values drive the results in Table III. 21 As discussed in Section III.A, these patterns are consistent with the promotion incentives of city secretaries. 21 In Table AIII in the Appendix, I employ the off-national-cycle cities. I find that this subsample of cities exhibits similar borrowing patterns.

18 292 The Journal of Finance R Figure 3. Local government borrowing pattern. This figure plots the logarithm of CDB total city loan amounts over city secretaries predicted turnover cycles. The right vertical axis is the logarithm of CDB total city loans after removing the year, city, and politician fixed effects from the regressions. The horizontal axis is the predicted turnover cycle of city secretaries. There are three national five-year turnover cycles between 1998 and 2012: 1998 to 2002, 2003 to 2007, and 2008 to The left vertical axis is the number of years that a city secretary serves the city (predicted), which is from year 1 to year 5. For example, for the first cycle (1 to 5 on the horizontal axis) from 1998 to 2002, I cluster cities by Years in Office (1 to 5) from 1998 to 2002 and plot the average CDB city loan amounts for each Years in Office bin. I repeat this procedure for the second cycle (6 to 10 on the horizontal axis) from year 2003 to The third cycle (11 to 15 on the horizontal axis) has only three years from 2008 to 2010, as the CDB city-level loan data are between 1998 and (Color figure can be viewed at wileyonlinelibrary.com) C. CDB SOE Loans Effects on Firms in the Same Industry To explore the effects of CDB loans on private firms, I begin by tracing the heterogeneous effects of CDB industrial loans across different levels of the supply chain. I use province-industry-level CDB loan data covering 31 provinces and 95 industries in China. These data offer the advantage of allowing me to analyze the effects of government credit on various industries. Previous studies are based mainly on aggregate government credit or spending and hence explore only the net effects. However, as I discuss in Section II.A, more than 95% of CDB industrial loans are extended to SOEs. Only 5% of CDB industrial loans flow to private firms, which typically have government backgrounds. Moreover, at the province-industry level, as weighted by total assets, approximately 86% of SOEs have loans outstanding from the CDB. Using these aggregate CDB province-industry loan data, I capture the effects of CDB credit to major SOEs on nearby private firms. In Table IV, I merge CDB province-industry loans outstanding each year with firm-level CIC data by firms locations and industry codes. In OLS regressions, I explore the correlations between CDB loan amounts and a firm s asset investment, employment, borrowing, sales, ROA, and sales per worker. I