How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China

ARTICLE IN PRESS

JID: YJCEC

[m3Gsc;October 20, 2016;22:17]

Journal of Comparative Economics 0 0 0 (2016) 1–23

Contents lists available at ScienceDirect

Journal of Comparative Economics journal homepage: www.elsevier.com/locate/jce

How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China Gang Xu∗, Go Yano Graduate School of Economics, Kyoto University Yoshida-honmachi, Sakyo-ku, Kyoto, 606-8501 Japan

a r t i c l e

i n f o

Article history: Received 20 June 2016 Revised 5 September 2016 Accepted 15 October 2016 Available online xxx JEL classification: D73 O38 Keywords: Anti-corruption Corruption Innovation Corporate R&D China

a b s t r a c t Xu, Gang, and Yano, Go—How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China This paper investigates the effect of anti-corruption on the financing and investing in innovation by using a detailed dataset of Chinese listed companies from 2009 to 2015. We find that stronger anti-corruption efforts make firms more likely to acquire external funds, mainly the long-term debt. Moreover, we show that firms located in provinces with stronger anti-corruption efforts invest significantly more of their newly acquired funds in R&D and generate more patents. Further empirical tests suggest this positive and statistically significant effect almost comes entirely from the current massive anticorruption campaign launched by President Xi Jinping since 2013. We further test two mechanisms regarding the corruption-innovation nexus: the expropriation hypothesis and the rent-seeking hypothesis. The results show that only firms without political connections, non-state owned enterprises (non-SOEs), firms operating in non-regulated industries and younger firms benefit from the stronger anti-corruption efforts, all supportive of the former mechanism. Journal of Comparative Economics 0 0 0 (2016) 1–23. Graduate School of Economics, Kyoto University Yoshida-honmachi, Sakyo-ku, Kyoto, 606-8501 Japan. © 2016 Association for Comparative Economic Studies. Published by Elsevier Inc. All rights reserved.

1. Introduction Corruption has become one of the central social and political issues in China and has increased to an epidemic level since the advent of the “reform and opening policy” in 1978. Moreover, corruption in China is becoming more intensified and institutionalized in the sense that high-level, big-stakes corruption increased more rapidly than ordinary ones (Wedeman, 2004). Even though the Chinese leaders never stop their efforts in fighting corruption, it is only since President Xi Jinping came to power after 18th Congress of the Communist Party of China (CCPC) in late 2012 that a large-scale systematic anticorruption campaign has been put in place. However, as the anti-corruption campaign continues and deepens, some critics begin to cast doubt on the real impact of the anti-corruption campaign, contending that anti-corruption is detrimental to the country’s economy. How does anti-corruption affect the economy? Is it beneficial or detrimental to economic development? However, to our knowledge no previous studies have ever tackled this important topic empirically. In this paper, we try to address this question by shedding light on the causal effect of anti-corruption on innovation, one of the most important driving forces of economic growth. Thus, our study both tries to fill this gap in literature and aims to provide policy ∗

Corresponding author. E-mail addresses: [email protected] (G. Xu), [email protected] (G. Yano).

http://dx.doi.org/10.1016/j.jce.2016.10.001 0147-5967/© 2016 Association for Comparative Economic Studies. Published by Elsevier Inc. All rights reserved.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

JID: YJCEC 2

ARTICLE IN PRESS

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

implications with respect to whether the current anti-corruption campaign should continue based on systematic empirical analyses. With respect to the link between corruption and innovation, it has received very limited attention relative to the large strand of literature focusing on the effects of corruption on economic performance indicators.1 Murphy et al. (1991) argue that innovators are particularly vulnerable to expropriation by government officials due to their inelastic need for government services such as permits and licenses, resulting in high risks, uncertainty and vulnerability. Blackburn and ForguesPuccio (2009), however, show that the effects of corruption depend on the extent to which bureaucrats coordinate their rent-seeking behavior. Several empirical studies using firm-level data generally come to the conclusion that corruption is detrimental to innovation and put innovative firms at a disadvantage (e.g., Waldemar, 2012; Paunov, 2016; Habiyaremye and Raymond, 2013), and Anokhin and Schulze (2009) show similar results by analyzing cross-country macro data. This paper tries to provide insights into this important topic from a very different perspective by exploring the causal impact of anti-corruption on innovation. We focus on China’s anti-corruption mainly for two reasons. First, China has long been plagued by corruption since the foundation of PRC. Corruption has managed to flourish and even become more rampant over time, despite the introduction of some anti-corruption campaigns. However, shortly after Xi Jinping came to power in late 2012, he launched an unprecedentedly massive anti-graft campaign in the history of Communist rule in China. This creates a precious opportunity to understand and study anti-corruption, an issue closely related to corruption, yet left unexplored in literature. Second, as the largest developing economy with unbalanced development across its regions, anti-corruption effort (also corruption) within China varies greatly across provinces and time (especially between pre-2013 and post-2013). The geographical diversity along with its evolution over time provides wide variations needed for identifying the effect of anti-corruption while making us still able to control the relatively homogenous underlying institutional and cultural factors within the country. We adopt two measures of anti-corruption in this paper. The first one is the standardized provincial number of officials of vice county-division rank and above investigated in the registered corruption cases each year, reported by the Procuratorial Yearbook of China. In the case of China, this proxy typically reflects the efforts devoted in anti-corruption by individual provinces rather than corruption itself. The second one measures the determinations and commitments of the provincial leaders along with its Commission for Discipline Inspection (CDI) to fight corruption in coordination with the Party Central Committee. As the provincial official newspaper of Communist Party of China (CPC) mainly serves a crucial function of promoting policy directives and shaping public opinions, the number of articles published in provincial CPC newspaper that advocate anti-corruption or denounce corruption is expected to well capture the provincial anti-corruption efforts. We employ the data of Chinese listed companies from CSMAR database, which allows us to construct a rich panel spanning a relatively long time period with the latest data also available. We propose that anti-corruption campaigns can be conducive to the financing and investing in corporate R&D mainly by mitigating the expropriation problem. When corruption is pervasive, firms are reluctant to innovate due to the high probability that their rents generated from risky innovative activity will be expropriated by corrupt-prone bureaucrats (the expropriation hypothesis). This significantly aggregates the risks and uncertainty associated with innovative pursuits. In this sense, firms located in provinces with stronger anti-corruption intensity are expected to both get access to and invest more funds for R&D, which may lead to more patents as a result, due to a higher expected return to innovation. Our concerns for the causal effect of anti-corruption include the existence of unobserved provincial factors that may correlate with both anti-corruption efforts and the financing & investing behavior of the firms, as well as the potential reverse causality, generated by the mechanism where a higher level of financing and investing in innovation may result in a greater demand for anti-corruption. We address these concerns by adopting system GMM developed by Arellano and Bover (1995) and Blundell and Bond (1998). This approach can not only control for the firm-level unobservable heterogeneity, but can also deal with the potential endogeneity of all the regressors by using their own lags as instruments. To increase the identification power of GMM, we further use two historical variables measuring the extent to which individual province was once exposed to the Anglo-American influence and a dummy indicating whether the province’s Secretary of Committee for Discipline Inspection (SCDI) is “airborne” as three external IVs for anti-corruption. These variables are expected to be correlated with the provincial anti-corruption efforts while basically unlikely to directly affect the financing and investing behavior in corporate R&D. We find that stronger anti-corruption efforts can make firms more likely to have access to external funds, mainly longterm debt. Moreover, the empirical results suggest that firms located in provinces with stronger anti-corruption efforts invest a significantly larger proportion of their newly acquired funds in R&D and generate more patents. Further empirical tests suggest this significantly positive effect comes entirely from the current massive anti-corruption campaign launched by President Xi Jinping since he came to power in late 2012. Thus facilitating financing and investing in R&D is one of the potential channels through which anti-corruption efforts can positively affect economic growth. However, besides the expropriation hypothesis, another mechanism may also account for the negative impact of corruption on firms’ innovation. As implied by Baumol (1990) and Murphy et al. (1991), when the relative payoff of corruption is 1 A large number of early studies using cross-country data generally find that a higher level of corruption is associated with lower economic growth (e.g., Mauro, 1995; Mo, 2001), less foreign direct investment (e.g., Wei, 20 0 0) and more social unrest (Manion, 2004). Among the empirical literature that examines the effect of corruption on firm performance by using firm-level survey data, the conclusions are mixed (e.g., Fisman and Svensson, 2007; Cai et al., 2011; Hellman et al., 2003).

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

JID: YJCEC

ARTICLE IN PRESS G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

[m3Gsc;October 20, 2016;22:17] 3

higher, firms are incentivized to choose rent seeking by cultivating relations with bureaucrats over productive activities, like innovation (the rent-seeking hypothesis). Stronger anti-corruption efforts may force those firms to substitute innovation for rent seeking due to the rising cost of corruption and the depreciated value of relational capital, particularly with those officials that collapsed in the campaign. To test the two hypotheses, we separate firms into two groups–one more susceptible to expropriation and one more likely to have access to rent. The results suggest only firms without political connections, non-SOEs, firms operating in non-regulated industries and younger firms are responsible for the significantly positive effect of anti-corruption on innovation, consistent with the expropriation hypothesis. This paper contributes to the literature in the following ways. First of all, as far as we know, this is the first study that systematically investigates the effect of anti-corruption on innovation. This paper adds to the limited studies relating to the link between corruption and innovation from a novel perspective. Almost all of the previous firm-level empirical studies regarding this topic use cross-sectional survey data, making it impossible to take into account the firm fixed effects and the dynamic effect of corruption on innovation. Cross country studies generally use perception-based subjective corruption indexes, which are potentially susceptible to spurious regressions (Aidt, 2009), and neither allow us to analyze the variation of corruption within a country nor to examine individual heterogeneity. This paper overcomes the above drawbacks by combining a rich firm-level panel dataset with the objective provincial-level anti-corruption data. Second, this paper also tries to test and disentangle two mechanisms that propose the detrimental effect of corruption on firms’ innovation–the expropriation hypothesis and rent-seeking hypothesis, by examining the heterogeneous effect of anti-corruption. When it comes to the impact of corruption on innovation, previous studies have exclusively focused on the expropriation mechanism, while overlooking alternative explanations that may also have important implications for the link between corruption and innovation. We reassess the validity of the expropriation hypothesis and explicitly take into account the possibility that firms that are not severely subject to expropriation is also likely to positively react to anti-corruption efforts. This leads us to clear and distinct policy implications. Moreover, this paper also adds to the strand of literature related to the financing of innovation. Previous studies generally suggest that debt is poorly suited for financing innovation due to the moral-hazard problem (Himmelberg and Peterson, 1994), lack of collateral (Hall, 2002) and the rising marginal costs of financial distresses (Opler and Titman, 1994), particularly associated with innovative firms.2 Given that the previous studies generally ignored the crucial role of institutions in making debt an available source for funding innovation, we complement this line of studies by showing that as the institutional quality improves (as proxied by the stronger anti-corruption efforts), firms tend to both obtain and invest more external debt for funding their innovations. This highlights the positive role of institutions in affecting the financing patterns of innovative activity. Finally, our study is also related to the very limited amount of literature that emphasizes the role of individual leaders on the economy (Jones and Olken, 2005).3 The pronounced effect of the large-scale anti-corruption campaign under President Xi suggests that even in a non-democratic country, anti-corruption campaigns led by a strong and determined leader can still make a huge difference. The rest of the paper is organized as follows. The next section discusses the anti-corruption situation in China, economic theories regarding the link between corruption and innovation as well as the expected effect of anti-corruption on innovation. Section 3 describes the sample and variables. Section 4 discusses our empirical approach and the main findings from the empirical results. Section 5 conducts several robustness checks and Section 6 concludes the paper. 2. Institutional background, theoretical framework and hypothesis development 2.1. The anti-corruption situation in China In China, the Commission for Discipline Inspection (CDI) is the most crucial organ responsible for anti-corruption. According to the constitution of CPC, the tasks of CDI at all levels mainly include upholding the Constitution and other statutes of the Party, checking up on the implementation of the line, principles, policies and resolutions of the Party and assisting the respective Party committees in organizing and coordinating the work against corruption.4 The process of CDI’s work routine in carrying out anti-corruption mainly includes the following steps (Nie and Wang, 2016). The first is to discover and investigate the officials that committed malfeasance and then to implement formal investigative procedures like Shuanggui (the practice of detaining individual party members for investigations). After this, the CDI determines the penalties according to the disciplinary rules of CPC and if necessary, forwards the evidence gathered to the procuratorial organs, which proceed to charge the accused with criminal wrongdoing and move the case to court. At last, the sentence is pronounced by the court. Anti-corruption have always been on the agenda of Chinese leaders who have never stopped their efforts in combating corruption, though the efforts devoted and the effectiveness of anti-corruption have varied. The anti-corruption efforts made

2 The only exceptions we know are Benfratello et al. (2008) and David et al. (2008). David et al. (2008) argue that different from transactional debt, relational debt (bank loans) can better overcome the above problems and thus provide more appropriate governance for R&D investment, which are corroborated by their empirical tests using a sample of Japanese listed companies. 3 Our findings closely echo those in Jones and Olken (2005), which show leaders matter for economic growth and the effects of individual leaders are strongest in autocratic settings where there are fewer constrains on a leader’s power. 4 http://www.china.org.cn/chinese/18da/2012-11/19/content_27156212_10.htm

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

JID: YJCEC 4

ARTICLE IN PRESS

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

by Xi Jingping’s predecessors were relatively weak and far from systematic or massive.5 According to statistics on the website of the Central Commission for Discipline Inspection (CCDI), the average number of officials convicted of corruption with the rank of deputy prefecture-department (Tingjuji) and above was only around 30 from 2003 to 2012, while this number soared to 186 and 473 in 2013 and 2014 respectively. Shortly after Xi assumed office since the 18th CCPC in late 2012, he vowed to crack down on “tigers and flies”, that is, high-ranking officials and petty civil servants alike.6 The coordination of anti-corruption in provinces and SOEs has been implemented by the Central Inspection Team (CIT), also led by Wang Qishan, who is the Secretary of CCDI. These teams are generally “stationed” at the organizations they are to oversee for a few months, and are in charge of thorough audits into the practices of officials and organizations. The first batch of CITs was dispatched in August 2013 to a variety of provinces and SOEs, e.g., Jiangxi, Guizhou, Inner Mongolia, and the Sinograin. Up to now, in each year since 2013, at least two batches of CITs have been dispatched to different provinces and SOEs. The CITs have uncovered a multitude of corruption cases and brought down thousands of officials of different ranks, thus functioning as an important organ for detecting and investigating local corruption in this campaign. The massive anti-corruption campaign has resulted in fruitful achievements. According to the data of Chinafile, as of June 2016, the campaign has “netted” 174 “big tigers” since the 18th CCPC, including five national leaders, several top executives of SOEs and about a dozen high-ranking military officers with the total value of funds and assets involved in these sentenced officials amounting to 2 billion yuan.7 The most striking case in the campaign is definitely the one that implicates Zhou Yongkang, the former member of the Politburo Standing Committee (PSC). Zhou was formally expelled from the party in December 2014 and then sentenced to life imprisonment in June 2015, making him the first member of PSC that collapsed due to corruption in the history of CPC. This broke the unspoken rule of “PSC criminal immunity”, which has been the norm for decades. This reflected the strong determination of President Xi in anti-corruption and sent out a clear message that anyone, no matter what his/her background and position is, cannot get away with corruption. Geographically, as of the beginning of 2016, there have been at least one provincial-level officials investigated for corruption in each of the 31 provinces. Among the provinces, Shanxi, Guangdong, Sichuan and Henan have seen the highest number of corruption cases. 2.2. The potential impact of corruption on corporate innovation The literature generally suggests a detrimental impact of corruption on firms’ innovative activity. Corruption necessarily increases the level of uncertainty and transaction cost associated with holdup or expropriation because the innovators’ value chain must involve related officials and authority at some stage of their innovation process. Corruption can make the cost of permits, licenses and other government services more expensive, for which the innovators tend to have very high and inelastic demand (Murphy et al., 1991). For example, corrupt officials in national granting bodies may ask for bribes to grant patents or quality certificates (Paunov, 2016). To make things worse, those competitors who imitate the innovators’ ideas can even get away with infringement if the corrupt-prone court takes bribes from the infringers. The expropriation from corrupt officials will necessarily lower the expected returns from innovative activity, making an otherwise promising innovative project difficult to commercialize lucratively. It is very difficult or costly for firms to take measures to guard against corruption due to its arbitrary and ex-post opportunistic nature (Luo, 2004). This significantly aggregates the risks associated with inherently risky innovative pursuits. We term this view the expropriation hypothesis. Moreover, according to Baumol (1990) and Murphy et al. (1991), in countries where the relative payoff of rent seeking is higher, talented entrepreneurs with potential innovative capabilities have incentives to direct their energy to rent seeking away from innovation. This highly fits the case of China, particularly before Xi’s campaign, where corruption can generate higher revenue at lower cost. This theory can provide a possible explanation for the low innovation incentives for firms without the fear of expropriation by government, e.g., politically connected firms and SOEs.8 We term this conjecture the rent-seeking hypothesis. 2.3. The expected effect of anti-corruption on financing and investing in innovation Stronger anti-corruption efforts can better alleviate the expropriation problem since the probability of being exposed becomes higher and the punishment associated with corruption becomes more severe, which significantly reduces the returns of expropriation for corrupt bureaucrats. This in turn reduces the uncertainty and transaction costs associated with innovation. Anti-corruption efforts thus contribute to fostering the institutional trust that proves to be necessary for the development of innovation and entrepreneurial activities. In terms of firms, lower expropriation risk through enhanced anti-corruption increases the probability of succeeding in their innovation projects with higher expected returns, thus encouraging them to invest more and use more debt. Due to the same reason, creditors also have more confidence in getting repaid, thus willing to expand the supply of finance. Moreover, stronger anti-corruption efforts also make creditors face a 5

For a comprehensive overview and analysis of the anti-corruption campaigns before Xi, please refer to Wedeman (2005). Generally, the “tigers” mainly refer to party and governmental officials of Sub-Provincial/Ministerial level and above & military officers holding a rank of Major General and above. 7 https://www.chinafile.com/infographics/visualizing-chinas-anti-corruption-campaign 8 Bellettini et al. (2013) and Kim (2015) provide empirical evidence that political connections hinder firms’ incentive to innovate through increased rent seeking opportunities. 6

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

[m3Gsc;October 20, 2016;22:17] 5

lower risk that the collateral underlying their secured loans will be expropriated by the predatory bureaucrats (Berkowitz et al., 2015). Therefore, firms are predicted to receive more debt financing in provinces with more intense anti-corruption efforts. However, we do not expect the same impact of anti-corruption on facilitating firms’ access to new equity. Most importantly, equity proves to be a superior and more efficient way for financing innovation relative to debt, as the former can, to a greater extent, overcome the moral hazard problem (Stiglitz and Weiss, 1981; Himmelberg and Peterson, 1994) and at the same time, have lesser requirements on collateral and the financial status of innovative firms (Hall, 2002; Opler and Titman, 1994). Thus these features of equity financing are expected to compensate for the unsatisfactory underlying business environment. In addition, as our samples firms used in this paper are publicly listed, they can easily get access to external equity as they want. In this sense we expect access to equity financing, unlike debt financing, is relatively insensitive to the improvement of institutional quality, as proxied by the anti-corruption intensity. More importantly, we expect anti-corruption is conductive to corporate innovation. Stronger anti-corruption efforts are expected to enable entrepreneurs to capture a larger proportion of revenues generated from innovation, increase the reliabilities of their cash flows and thus motivate higher levels of innovative activity, particularly for firms susceptible to government expropriation. On the other hand, if anti-corruption effort is truly effective in the sense that stronger anti-corruption campaign raises the cost of rent seeking via cultivating guanxi with government and diminishes the value of connections with bureaucrats, especially those who have fallen under the anti-graft dragnet, former politically favored firms are also expected to devote more resources and efforts to innovative activity accordingly as future developing strategy. In this sense, both theories imply that firms in provinces with more intense anti-corruption efforts invest more funds in R&D, which is also likely to generate more patents as a result. 3. Data and variables 3.1. Database of Chinese listed companies Most of the firm-level data used in this paper is obtained from the China Stock Market and Accounting Research (CSMAR) database at GTA. The CSMAR database provides various accounting and financial information of companies listed on Shanghai and Shenzhen Stock Exchange since their IPO. Besides, data regarding firm’s ultimate controller and political backgrounds of (vice) chairmen and (vice) CEOs is derived from two sub-databases of CSMAR, the Chinese Listed Firm Shareholder Research Database and Corporate Governance Structure Research Database. With respect to political connection information, we also exploit other available sources, such as firms’ annual reports, Baidu and Sina Finance to make sure we do not miss key information to identify their connections with government. Finally, corporate R&D expenditure and information regarding patent applications come from Wind financial database and the State Intellectual Property Office of China (SIPO) respectively. We focus on Chinese companies listed on the A-share Stock Exchanges. We drop firms in the financial industry as their financing and investment behaviors differ systematically from other firms. Observations from within 2 years before Special Treatment (ST) to the year when ST is removed are also excluded. We then eliminate any firm without at least a three-year complete record of basic financial information and positive R&D expenditures during the sample period,9 i.e., from 2009 to 2015. Most of the excluded firms do not report any information on R&D expenses (rather than reported as zero). 96% of our sample firms are from manufacturing and service sector, which account for most of the R&D in the economy. At last, we trim all the firm-level variables at upper and lower 2.5th percentiles to avoid outliers driving our results.10 This leaves a total of 1895 firms and 8025 firm-year observations. Table 1 shows the sample distribution with respect to their industries and years. 3.2. Measures of anti-corruption efforts In this paper we measure the provincial anti-corruption efforts in two ways. Similar to Cole et al. (2009) and Nie and Wang (2016), the first proxy is measured as the provincial number of officials of vice county-division rank (xianchuji) and above investigated in the registered cases on corruption per 100 thousand population, which stems from the Procuratorial Yearbook of China.11 The registered cases mainly include the ones charged with misappropriation of public property, extortion and acceptance of bribes, abuse of power and dereliction of duty. We believe this is an appropriate proxy that can capture the anti-corruption efforts devoted by provinces. First of all, according to Cole et al. (2009), assuming inherent corruption levels are equal across provinces, which would be the case

9 The availability of R&D data, to the largest extent, determines our sample size used in the regressions. For example, among the firm-year observations with non-missing total asset in 2009, only 27% of them report non-missing R&D data, which sharply contrasts with 79% for 2014 and 81% for 2015. Additional tests suggest that firms that report R&D do not differ systematically from firms with missing R&D. 10 All the firm-level variables are subject to extreme outliers and for most of them, trimming at 1st percentile level is not enough to avoid outliers driving our results. For example, the 99th percentile for New Equity and Tobin’s Q is 2.65 and 15.4 respectively, which is still unrealistically large. However, our regressions results remain unaffected even if we trim or winsorize variables at 1st percentile levels (Tables OA3 and OA4 in online appendix). 11 However, this data is not reported for some province-year observations, totaling 229 from 2008 to 2015.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

JID: YJCEC 6

ARTICLE IN PRESS

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23 Table 1 Sample distribution. Freq.

%

A. Industry distribution of the sample firms Agriculture, forestry, livestock farming and fishery (A) Mining (B) Food and beverage (C0) Textile, clothes and fur (C1) Timber and furniture (C2) Paper making and printing (C3) Petroleum, chemistry, rubber and plastic (C4) Electronic (C5) Metal and non-metal (C6) Machinery, equipment and instrument (C7) Medical and biological products (C8) Other manufacturing (C9) Electric power, heat, gas and water production (D) Construction (E) Wholesale and retail (F) Transportation and storage (G) Information technology and software (I) Real estate (K) Leasing and commercial service (L) Scientific research and technological service (M) Communication and culture (R) Miscellaneous (S) Total

121 207 322 302 51 175 996 1715 764 1526 670 68 76 171 143 58 452 33 24 59 37 55 8025

1.51 2.58 4.01 3.76 0.64 2.18 12.4 21.4 9.52 19.0 8.35 0.85 0.95 2.13 1.78 0.72 5.63 0.41 0.30 0.74 0.46 0.69 100

B. Year distribution of the sample firms 2009 2010 2011 2012 2013 2014 2015 Total

436 691 991 1390 1515 1460 1542 8025

5.43 8.61 12.4 17.3 18.9 18.2 19.2 100

if individuals are equally susceptible to temptation, this number directly reflects the outcomes of anti-corruption efforts of each province. Furthermore, a little different from that study, we use the number of officials above a certain rank that are investigated in the registered corruption cases rather than all of the officials investigated in those cases. This can better reflect Chinese leaders’ determination to fight corruption as dealing with corruption cases involving officials of higher rank is much trickier and requires more courage.12 Moreover, even though some US-related studies adopt similar proxies to measure corruption (e.g., Fisman and Gatti, 2002; Glaeser and Saks, 2006), we believe in the case of China, it indicates more of anticorruption rather than corruption. For the US, institutions that prevent bureaucratic corruption, such as checks and balances, are well in place, which makes corruption less of a concern. Besides, the sound legal system can impartially and effectively uncover the corrupt cases and bring down the implicated bureaucrats. Nevertheless, in China, a non-democratic country with missing and dysfunctional institutions, corruption is more pervasive and those exposed to the public may be merely the tip of iceberg. Particularly for corruption cases that involve officials of higher ranks, the investigations and further legal sanctions on those targets must follow the instructions of upper-level authorities. Fig. 1 plots the time trend of this measure across 8 different regions in China. It shows an abrupt increase since 2013 and an even steeper rising trend over 2014 and 2015, highly consistent with the trajectory of the large-scale anti-corruption campaign launched by President Xi since he assumed office. As the underlying determinants of corruption are continuous and unlikely to change abruptly, the sharp increase of this proxy captures exactly the intensity of anti-corruption. In this sense, we believe this variable is a reliable indicator of anti-corruption rather than corruption in the context of China. The second measure of anti-corruption is the number of articles in the provincial CPC newspaper that advocate anticorruption or denounce corruption in each year (divided by 100). The main function of the provincial CPC newspaper, published by each province’s circulated daily, is to promote party’s platform, guidelines and policies as well as to shape public opinions.13 This newspaper is subscribed to by all the government agencies of different levels in the province. Given the crucial role of this newspaper as the party’s mouthpiece, a stronger advocacy of anti-corruption in the provincial party’s official newspaper reflects a firmer determination of provincial leaders and its CDI to combat corruption in response to

12 We also use the number of all the officials investigated in the registered corruption cases per 100 thousand populations as a measure for anti-corruption and the regression results basically remain unaffected. 13 Ang et al. (2014) also adopt similar methods to measure the intellectual property rights protection at provincial level in China.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

JID: YJCEC

ARTICLE IN PRESS G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

[m3Gsc;October 20, 2016;22:17] 7

Fig. 1. Time trend of Anticor1 of different regions in China.

the anti-corruption guidelines and instructions of the Party Central Committee. We use the Duxiu database, an integrated database that contains rich information in books, journals, newspapers and academic papers, to track relevant newspaper articles with the following key words in the title: anti-corruption (fanfu), corruption (fubai), and honest and clean government (dangfenglianzheng). We then read through the identified articles to ensure the contents satisfy our requirements.14 ,15 3.3. Other variables To analyze the effect of anti-corruption efforts on access to external financing, two variables are constructed. A firm is counted as having raised new long-term debt in year t if the net increase in long-term debt for firm i in year t exceeds 5% of its total assets at the beginning of year t. Similarly, a firm is counted as having raised new external equity in year t if the net increase in external equity for firm i in year t exceeds 5% of its total assets, where the net increase in external equity is defined as the net increase in book equity minus the net increase in retained earnings (Baker et al., 2003; Ang et al., 2014). Following Hovakimian et al. (2001) and De Haan and Hinloopen (2003), we set the 5% threshold to ensure that our analysis is confined to relatively substantial financing activities.16 To ensure the estimated effect of anti-corruption on the financing and investing in R&D is not driven by some omitted variables that may correlate with our anti-corruption measures, we further add two provincial-level control variables in the regression. The first one is the real GDP growth rate to account for the differences in economic development across provinces. More importantly, as better protection for intellectual property rights (IPR) may contribute to the availability of external financing and investment in R&D, to capture this effect, we use a second control variable to measure the extent of provincial IPR protection. This variable is constructed as the natural logarithm of the sum of two sub-indices in Fan et al. (2011) that measure the extent of provincial IPR protection, i.e., the number of three kinds of patent applications accepted

14 We show in an unreported validity test that our anticorruption measures are positively correlated with the extent of marketization and negatively correlated with government interventions imposed on firms (Fan et al., 2011) 15 To further deal with the potential endogeneity problem, we obtain the residuals from regressing the original anti-corruption measures on a set of provincial-level variables, including GDP per capita, log (population), fiscal expenditure, education level, GDP growth, natural resources, media exposure, IPR protection and relative wage of public servant as well as year dummies using fixed effect model, to extract the relative random part of anti-corruption. We use the residuals as the new anti-corruption measures and the results remain unchanged (Table OA5). 16 The conclusions still hold if we use continuous dependent variables, i.e., the amount of new debt and new equity (Table OA2).

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC 8

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23 Table 2 Summary statistics and correlations. Variable

Mean

Sd

Min

RD intensity LnPatent New Equity New Debt Cash Flow Age Tobin’s Q Sales Access to NewDebt Access to NewEquity Asset ROA Sale Growth Leverage Fixed Asset Intangible Asset

A. Firm-level Summary Statistics 0.022 0.017 0 1.679 1.100 0 0.048 0.141 −0.069 0.010 0.046 −0.110 0.059 −0.060 0.102 1.694 0.930 0 2.190 1.508 0.296 0.736 0.409 0.113 0.139 0.346 0 0.214 0.410 0 21.63 1.047 19.29 0.050 0.043 -0.108 0.244 0.520 -0.535 0.400 0.196 0.063 0.228 0.138 0.005 0.044 0.035 0

Anticor1 Anticor2 IP Protection GDP Growth Airborne UKsettle Churchuniv

B. Provincial-level Summary Statistics 0.260 0.153 0.036 0.411 0.248 0.040 8.168 1.037 5.631 0.111 0.025 0.049 0.641 0.481 0 0.258 0.438 0 0.452 0.757 0

1.Anticor1 2.Anticor2 3.IP Protection 4.GDP Growth 5.Airborne 6.UKsettle 7.Churchuniv

C. Correlation Matrix 1 2 1 1 0.784∗ ∗ ∗ −0.105 −0.059 ∗∗∗ –0.218∗ ∗ ∗ –0.170 0.155∗ ∗ 0.244∗ ∗ ∗ –0.201∗ ∗ ∗ –0.242∗ ∗ ∗ –0.129∗ ∗ –0.168∗ ∗

Median

Max

N

0.093 1.792 1.813 0.648 0 0 21.49 0.044 0.115 0.391 0.203 0.036

0.090 4.970 1.600 0.238 0.353 3.219 9.710 2.515 1 1 24.81 0.185 4.462 0.914 0.663 0.210

8025 6106 8025 8025 8025 8025 8025 8025 6774 6774 6774 6774 6774 6774 6774 6774

0.215 0.345 8.157 0.110 1 0 0

0.932 1.270 10.97 0.178 1 1 3

229 248 248 248 248 248 248

3

4

5

6

7

1 –0.210∗ ∗ ∗ 0.315∗ ∗ ∗ 0.518∗ ∗ ∗ 0.591∗ ∗ ∗

1 –0.141∗ ∗ –0.002 –0.163∗ ∗ ∗

1 0.401∗ ∗ ∗ 0.392∗ ∗ ∗

1 0.624∗ ∗ ∗

1

0.019 1.609 0.007 0

Note: Panel A includes all the firm-level variables and Panel B includes all the provincial-level variables. Panel C lists the correlation among the provincial-level variables. ∗ ∗ ∗ ,∗ ∗ and ∗ : significant at 1%, 5% and 10%, respectively

and granted per 10,0 0 0 science and technology personnel.17 We lag both controls by one year in the regression. Other relatively time-invariant provincial characteristics are expected to be captured by the 31 province dummies.

3.4. Summary statistics According to Table 2 Panel B, on average, there are 0.26 officials of vice county-division rank and above investigated in the registered corruption cases in every 10 0,0 0 0 population with a standard deviation of 0.153, ranging from 0.036 to 0.932. The number of articles in the provincial CPC newspapers that advocate anti-corruption or denounce corruption ranges from 4 to 127, averaged at 41 with a standard deviation of 24.8. Fig. 2 maps the distribution of anti-corruption measured by the first proxy from 2008 to 2012 and from 2013 to 2015, respectively. The figures in Fig. 2 imply that the anti-corruption efforts vary greatly across provinces and become more intensified dramatically after 2012. The extent of protection for IPR varies from 1.04 to 10.97, with a mean of 8.17, and on average provinces experience an 11% real GDP growth. The correlations among the provincial-level variables are shown in Table 2, Panel C. Table 2 panel A reports the summary statistics for firm-level variables. R&D intensity is measured as the ratio of R&D expenditures to the start-of-year book value of total assets. On average, companies in our sample invest 2.2% of their assets in R&D and file 9.4 patent applications each year. Moreover, 21.4% of our sample firms on average have raised new equity while only 13.9% of them have raised new debt. Given that they are all listed companies, it’s natural for them to depend more on equity. The detailed definition of the key variables can be found in Appendix Table A1.

17 Note that Fan and his team have stopped updating this NERI index since 2009. Therefore we construct similar indexes based on their methods mentioned in Fan et al. (2011). The relevant data to calculate this variable is obtained from the Statistical Yearbook of China.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

9

Fig. 2. Anti-corruption efforts across provinces in China.

4. Empirical results and analyses We report the empirical results in the following 4 subsections. We first examine the effect of anti-corruption on firms’ access to new debt and new equity. Then we investigate the impact of anti-corruption on R&D input and output. Furthermore, we particularly focus on the effect of the current anti-corruption campaign under Xi since 2013. Finally we explore the heterogeneous effects of an-corruption on R&D investment for different types of firms, aiming to test two theories regarding the corruption-innovation nexus, i.e., the expropriation and the rent-seeking hypothesis. 4.1. The effect of anti-corruption efforts on debt and equity financing In this part, we examine the impact of anti-corruption efforts on firms’ ability to obtain external debt and equity financing. We model the probability of whether a firm obtains new funds in a given year as the function of provincial anticorruption efforts, various firm and provincial characteristics and a series of dummies controlling for provinces, industries and years. The model is specified as Eq. (1):

P rob(Yi,t = 1 ) = ∅(α0 + α1 Anticori,t + α2 F irm Characi,t−1 + α3 P rovincial Characi,t−1 +

α4 Province Dummiesi,t + α5 Industry Dummiesi,t + α6Year Dummiesi,t + i,t )

(1)

The dependent variable Yi, t is equal to 1 if the net increase in long-term debt/equity of firm i exceeds 5% of its start-ofyear book value of total asset in year t. ∅ is the cumulative function of normal distribution. The firm-level control variables include the natural logarithm of total assets, ROA, sales growth rate, natural logarithm of age, leverage, R&D intensity, the ratio of fixed assets to total assets and the ratio of intangible assets to total assets. Provincial characteristics include the real GDP growth rate and extent of IPR protection. All the above control variables are lagged by one year. We also control for 31 province dummies, 22 industry dummies according to the “Guidelines on the Industry Classification of Listed Companies” issued by China Securities Regulatory Committee (CSRC) and 7 year dummies. Standard errors are heteroskedasticity-robust and clustered at the provincial level. Columns 1 and 2 in Table 3 report the effect of Anticor1 (the normalized provincial number of officials of vice-county division level and above investigated in registered corruption cases) on the probability that a firm has access to new longterm debt using the Probit model, without and with provincial controls and dummies. We find that as expected, anticorruption efforts have a significantly positive effect on the probability that a firm has access to new long-term debt. Besides, we find that firms with larger size, better profitability, younger age, higher leverage and more fixed assets are more likely to access new debt. Columns 5 and 6 in Table 3 show the results using Anticor2 (the number of articles published in provincial Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC 10

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23 Table 3 The effect of anti-corruption on access to new debt.

AntiCor1

Probit

Probit

GMM

(1)

(2)

(3)

0.077∗ ∗ ∗ (0.028)

0.076∗ (0.042)

0.341∗ ∗ (0.143)

GMM with external IVs (4)

ROA Sale Growth Age Leverage RD Intensity Fixed Asset Intangable Asset

0.011∗ ∗ ∗ (0.004) 0.315∗ ∗ ∗ (0.099) 0.013 (0.010) –0.020∗ ∗ ∗ (0.006) 0.217∗ ∗ ∗ (0.029) –0.353 (0.237) 0.128∗ ∗ ∗ (0.025) 0.176 (0.109)

0.012∗ ∗ ∗ (0.004) 0.334∗ ∗ ∗ (0.106) 0.012 (0.010) –0.021∗ ∗ ∗ (0.007) 0.215∗ ∗ ∗ (0.035) –0.326 (0.244) 0.131∗ ∗ ∗ (0.024) 0.177 (0.110) 0.830 (0.511) –0.044∗ ∗ (0.021)

0.039 6558

0.048 6550

GDP Growth IPR Protection Acess to NewDebt t–1 Pseudo R-square Observations Groups AR(2) (p-value) AR(3) (p-value) J (p-value)

Probit

GMM

(5)

(6)

(7)

GMM with external IVs (8)

0.044∗ ∗ (0.017) 0.013∗ ∗ ∗ (0.004) 0.302∗ ∗ ∗ (0.099) 0.014 (0.010) –0.021∗ ∗ ∗ (0.006) 0.221∗ ∗ ∗ (0.029) –0.408 (0.254) 0.127∗ ∗ ∗ (0.027) 0.112 (0.125)

0.049∗ (0.029) 0.014∗ ∗ ∗ (0.004) 0.317∗ ∗ ∗ (0.104) 0.013 (0.010) –0.022∗ ∗ ∗ (0.007) 0.220∗ ∗ ∗ (0.034) –0.410 (0.262) 0.130∗ ∗ ∗ (0.024) 0.101 (0.125) 0.782 (0.510) –0.037 (0.023)

0.209∗ ∗ (0.089) –0.156 (0.149) –0.028 (2.823) 0.193 (0.214) 0.078 (0.072) –0.115 (0.657) –2.641 (2.675) 0.704 (0.559) 0.494 (1.659) 2.518∗ (1.483) –0.040 (0.049) –0.005 (0.028)

0.363∗ ∗ (0.161) 0.485∗ ∗ (0.247) –3.011 (3.165) 0.350 (0.332) 0.132 (0.169) –0.738 (1.077) 0.444 (12.702) –0.230 (0.970) –3.565 (3.455) 5.792∗ ∗ (2.640) –0.046 (0.064) –0.087 (0.200)

0.041 6774

0.049 6763

6774 1772 0.873 0.800 0.225

6774 1772 0.269 0.520 0.194

0.500∗ ∗ ∗ (0.181)

AntiCor2 Asset

Probit

0.100 (0.169) –2.321 (2.083) 0.025 (0.161) 0.192∗ (0.103) –0.907∗ (0.546) 3.526 (3.915) 0.588 (0.689) 0.407 (1.550) 1.919 (1.273) –0.057 (0.047) –0.020 (0.036)

–0.301 (0.288) 4.451∗ ∗ (2.230) 0.258 (0.234) 0.125 (0.173) 0.186 (0.754) –13.697 (8.568) 1.728∗ ∗ (0.809) 0.142 (3.687) –0.532 (1.680) 0.003 (0.042) –0.031 (0.049)

6558 1766 0.501 0.592 0.314

6558 1766 0.367 0.240 0.334

Note: The dependent variable is a dummy that equals 1 if the net increase in long-term debt of firm i in year t exceeds 5% of its total asset at the beginning of the year. For columns (1), (2), (5) and (6), Probit model is applied to estimate the regressions, and the estimated effects of marginal change in the corresponding regressors on the probability of getting access to new debt, computed at the sample means of the regressors, are reported. For columns (3), (4), (7) and (8), system GMM is employed to further deal with the potential endogeneity of the model. In columns (4) and (8), besides using lagged independent variables as their own IVs, we also use three external IVs, UKsettle, Churchuniv and Airborne for the anticorruption proxies. We use three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year dummies and industry dummies. Except columns (1) and (5) that do not control for province dummies, all the regressions control for industry dummies, province dummies and year dummies which are not reported for brevity. Standard errors in parentheses are heteroskedasticity-robust and clustered at the provincial level. ∗ ∗ ∗ ,∗ ∗ and ∗ :significant at 1%, 5% and 10%, respectively

CPC newspapers that advocate anti-corruption) as our anti-corruption proxy. The results are highly consistent with the ones using Anticor1, again showing a statistically significant and positive relationship between anti-corruption efforts and firms’ ability to access new long-term debt. However, our concern for above results is that the estimated effect of anti-corruption on firms’ access to new debt may be driven by omitted variables (both province- and firm-level variables) and reverse causality, generated by the mechanism where a higher level of financing and investing in innovation may result in a greater demand for anti-corruption. We address these concerns by adopting system GMM (Arellano and Bover, 1995; Blundell and Bond, 1998). This approach enables us to take into account the dynamic effect of anti-corruption on firms’ debt financing by including past debt financing, and at the same time to control for the firm-level unobservable heterogeneity by including firm fixed effects. It can easily deal with the potential endogeneity of all the independent variables, by simultaneously estimating a regression of Eq. (1) in levels and in differences, using lagged differences as instruments for regression in levels and lagged levels as instruments for regression in differences. To strengthen the identification power of GMM, we further employ three external IVs. La Porta et al. (1999) argue that the common law system originating in England has developed to some extent as a defense of Parliament and property owners against expropriation from the sovereign, while the civil law has developed more as an instrument for building state and controlling economy used by the sovereign. Therefore the common law system in England and its former colonies attaches more importance to protecting property rights, thus leading to higher government efficiency and lower corruption than countries of the civil law system. We thus use two historical variables to capture the extent of exposure to the influence of Great Britain and its former colonies. The first one is the number of church universities founded by Christian missionaries (mainly including American, British and Canadian) in each province by 1920 Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

JID: YJCEC

ARTICLE IN PRESS

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

11

(Stauffer et al., 1922) and the second is a dummy indicating whether there was once a concession or leased territory established by the Great Britain in the province during the late Qing dynasty (Yang and Ye, 1993). After China was defeated by foreign colonial powers in a sequence of wars of aggression in the early 1990s, the powers carved spheres of influence in China through a series of unequal treaties. Headed by the Britain, they reconstructed the local administrative and legal system, infrastructure and education within the place they occupied based on their own values and practices (Dong and Torgler, 2013). The persistent influences in these dimensions may be still responsible for the regional differences in corruption nowadays. We therefore expect that these time-invariant historical variables can predict the cross-sectional variations in anti-corruption efforts and that regions once exposed to the influence of the Great Britain and its former colonies may have lower level of anti-corruption, due to relatively less severe underlying corruption. Besides, we also choose as a third IV, Airborne, a dummy variable indicating whether the province’s Secretary of the Commission for Discipline Inspection (SCDI) is transferred directly from other provinces or from central government/departments rather than locally promoted from the same province. Compared with locally promoted SCDIs, as the “airborne” SCDIs have neither close guanxi with the provincial leaders nor complex personal network in the locale, they are expected to be less susceptible to interventions from the same level party committee and to act more independently in fighting corruption (Nie and Wang, 2016), thus leading to stronger anti-corruption effort. The significantly positive correlation coefficient between Airborne and Anticor shown in Table 2 confirms the above argument. While the appointment of provincial SCDI may reflect the will of central government and is not randomly assigned, to the extent that SCDIs’ terms of reference and related influence are predominantly confined to discipline inspection and anti-corruption on government officials, whether the SCDI is “airborne” should only indirectly affect individual firm’s R&D decision through the channel of anti-corruption rather than directly.18 In addition, even if SCDIs can exert certain direct influence on the economy, their influence will be exceedingly limited and it should not be correlated with error term after controlling for provincial GDP growth and firm-level controls. We estimate the equations using one-step system GMM. To evaluate whether the models are correctly specified and instruments are valid, we conduct two tests to check whether our instruments are uncorrelated with the error term of the regression. The first test is the Sargan J-test with the null that the over-identifying restrictions are valid. The second is the m(n) test, based on the serial correlation in the differenced residuals. In the presence of serial correlation of order n in the differenced residuals, only lags n + 1 and deeper should be used as instruments. These instruments are valid if serial correlation of order n + 1 is not detected in the differenced residuals.19 Columns 3 and 7 report the GMM estimates using only internal instruments while columns 4 and 8 report the GMM estimates that further use the three external IVs. The results of these specifications again corroborate a positive, statistically significant and quantitatively important link between anti-corruption efforts and firms’ access to new debt. The estimated effect in column 4 suggest that holding all the other regressors constant at their means, moving from the province with the lowest Anticor1 to the one with the highest Anticor1 increases the probability that a firm will obtain new debt by 45%, much larger than the estimate by Probit model in column 2. In Table 4 we investigate the effect of anti-corruption on equity financing. While the coefficients of anti-corruption are significantly positive in columns 1 and 4 using the simplest specification estimated by Probit model, we fail to find robust and significant results when adding provincial controls and apply system GMM. Thus, as expected, equity financing is relatively insensitive to anti-corruption given that our sample firms are all listed and that equity is better suited for financing innovation. 4.2. The effect of anti-corruption efforts on R&D input and output In this part, we investigate the effect of anti-corruption on R&D input and output. Based on the standard model specifications in this line of studies (Borisova and Brown, 2014; Ang et al., 2014), the empirical model takes the following form:

RD Intensit yi,t =

β0 + β1 Sources o f F undsi,t + β2 Anticori,t × Sources o f F unds i,t + β3 Anticori,t + β4 Agei,t + β5 Salesi,t + β6 Qi,t−1 + β7 P rovincial Characi,t−1 + β8 P rovince Dummiesi,t + β9 Industry Dummiesi,t + β10Year Dummiesi,t + εi,t

(2)

The R&D intensity is modeled as a function of newly raised funds, anti-corruption efforts and the interaction terms between newly raised funds and anti-corruption efforts in addition to some firm and provincial controls as well as dummies of province, industry and year. Newly acquired funds include newly acquired long-term debt, newly acquired equity and cash flows, all scaled by total assets at the beginning of the year. The firm-level control variables include sales (scaled by total assets) and lagged Tobin’s Q to control for firm demand or investment opportunities, as well as firm age (in natural logarithm), as is the standard practice in the literature. The provincial control variables are the same as the previous section. We also estimate the impact of anti-corruption on firms’ R&D output, measured by the number of patents applied for during a given year, as this can better reflect the actual time of innovation efforts than the number of patents granted. To

18

As introduced in the beginning of Section 2, the SCDI’s terms of reference do not include any work related to economic development. Initially, we used two lags of the regressors as instruments for the equation in first-differences. However, in all our specifications, we found evidence of serial correlation of order two in the differenced residuals, so we turn to using three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year and industry dummies. 19

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC 12

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23 Table 4 The effect of anti-corruption on access to new equity.

AntiCor1

Probit

Probit

GMM

(1)

(2)

(3)

GMM with external IVs (4)

0.087∗ ∗ (0.041)

–0.053 (0.082)

–0.069 (0.169)

–0.006 (0.132)

AntiCor2 Asset ROA Sale Growth Age Leverage RD Intensity Fixed Asset Intangable Asset

–0.053∗ ∗ ∗ (0.007) 1.316∗ ∗ ∗ (0.158) –0.003 (0.008) 0.006 (0.009) 0.400∗ ∗ ∗ (0.062) 0.344 (0.342) 0.131∗ ∗ (0.054) 0.249 (0.164)

–0.054∗ ∗ ∗ (0.007) 1.308∗ ∗ ∗ (0.143) –0.006 (0.008) 0.007 (0.010) 0.398∗ ∗ ∗ (0.064) 0.401 (0.327) 0.126∗ ∗ (0.054) 0.189 (0.177) 0.762 (0.532) 0.005 (0.044)

0.057 6558

0.064 6550

GDP Growth IPR Protection Acess to NewEquity t–1 Pseudo R-square Observations Groups AR(2) (p-value) AR(3) (p-value) J (p-value)

–0.574∗ ∗ ∗ (0.170) –0.724 (2.605) 0.304 (0.208) 0.213 (0.141) 2.204∗ ∗ ∗ (0.740) –0.629 (8.173) –0.994 (1.011) –1.857 (3.135) –1.137 (2.028) 0.015 (0.039) 0.063 (0.059)

–0.377 (0.223) –1.774 (2.569) 0.351∗ ∗ (0.170) 0.077 (0.100) 1.355∗ (0.697) –0.183 (5.920) 0.017 (0.638) 0.509 (2.654) –0.461 (1.721) –0.043 (0.049) 0.052 (0.059)

6558 1766 0.179 0.147 0.111

6558 1766 0.244 0.144 0.102

Probit

Probit

GMM

(5)

(6)

(7)

GMM with external IVs (8)

0.056∗ ∗ (0.025) –0.051∗ ∗ ∗ (0.007) 1.289∗ ∗ ∗ (0.152) –0.004 (0.008) 0.003 (0.010) 0.400∗ ∗ ∗ (0.062) 0.353 (0.339) 0.131∗ ∗ (0.051) 0.220 (0.170)

–0.053 (0.055) –0.053∗ ∗ ∗ (0.007) 1.278∗ ∗ ∗ (0.139) –0.006 (0.008) 0.004 (0.010) 0.400∗ ∗ ∗ (0.064) 0.380 (0.326) 0.129∗ ∗ (0.051) 0.160 (0.179) 0.595 (0.481) 0.012 (0.044)

0.070 (0.119) –0.215 (0.186) 0.024 (2.032) 0.105 (0.131) 0.221∗ (0.127) 1.243∗ (0.657) 3.018 (5.494) –0.859 (0.688) –1.743 (2.145) –2.005 (1.766) –0.009 (0.039) 0.021 (0.042)

0.093 (0.073) –0.105∗ ∗ (0.043) 2.590∗ ∗ (1.221) 0.200∗ ∗ (0.084) 0.029 (0.038) 0.507∗ ∗ (0.207) –0.893 (1.666) 0.342∗ ∗ (0.154) 0.775 (0.927) –0.399 (1.312) –0.036 (0.039) 0.021 (0.024)

0.058 6774

0.065 6763

6774 1772 0.393 0.150 0.129

6774 1772 0.266 0.155 0.109

Note: The dependent variable is a dummy that equals 1 if the net increase in external equity of firm i in year t exceeds 5% of its total asset at the beginning of the year. For columns (1), (2), (5) and (6), Probit model is applied to estimate the regressions, and the estimated effects of marginal change in the corresponding regressors on the probability of getting access to new equity, computed at the sample means of the regressors, are reported. For columns (3), (4), (7) and (8), system GMM is employed to further deal with the potential endogeneity of the model. In columns (4) and (8), besides using lagged independent variables as their own IVs, we also use three external IVs, UKsettle, Churchuniv and Airborne for the anticorruption proxies. We use three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year dummies and industry dummies. Except columns (1) and (5) that do not control for province dummies, all the regressions control for industry dummies, province dummies and year dummies which are not reported for brevity. Standard errors in parentheses are heteroskedasticity-robust and clustered at the provincial level. ∗ ∗ ∗ ,∗ ∗ and ∗ :significant at 1%, 5% and 10%, respectively

also take into account the quality of patents, our patent measure only includes the invention patent which is the most original among the three types of patents in China.20 Following Tan et al. (2015), we use the natural logarithm of one plus the number of patent applications as the dependent variable. We lag the anti-corruption measures by two years because there exists a delay between firms’ decision to innovate in response to the anti-corruption situation and the actual generation of observable outputs.21 This also reduces our sample size. Therefore, for the purpose of our study, R&D input is a better proxy as it can immediately respond to the anti-corruption situation and adequately reflect the short-term effect anti-corruption engenders. Due to these reasons, in this paper we mainly focus on R&D input as our innovation measure and only report the results of baseline regression using patent as the dependent variable. However, conclusions of this study are not affected by the choice of innovation variables. The estimated results of Eq. (2) are reported in Table 5. The interaction terms between newly raised funds from different sources and anti-corruption are the key variables of interest. Columns 1 and 2 show the results using Anticor1 as the anti-corruption measure estimated by a within-firm regression controlling for firm fixed effects, with column 2 further controlling for the lagged R&D expenditures. The coefficients of the interaction term between new debt and anti-corruption are significant and positive in both specifications, indicating that in provinces with stronger anti-corruption intensity, firms

20 The three types of patents in China are invention, utility model and design. Our results are robust if we measure patent using both invention and utility model as Tan et al. (2015). 21 Our results are robust if we lag anti-corruption by three or four years.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

13

Table 5 Baseline regression: the effect of anti-corruption on R&D input.

New Equity New Debt Cash Flow New Equity

∗

AntiCor1

New Debt

∗

AntiCor1

Cash Flow

∗

AntiCor1

AntiCor1 New Equity

∗

FE

FE

GMM

FE

FE

GMM

(3)

GMM with external IVs (4)

(5)

(6)

(7)

GMM with external IVs (8)

(1)

(2)

0.001 (0.002) –0.004 (0.004) 0.081∗ ∗ ∗ (0.006) –0.002 (0.007) 0.034∗ ∗ (0.015) 0.018 (0.018) –0.001 (0.004)

0.0 0 0 (0.002) –0.006 (0.005) 0.078∗ ∗ ∗ (0.006) 0.001 (0.006) 0.043∗ ∗ (0.018) 0.003 (0.018) –0.001 (0.004)

0.008 (0.016) –0.101 (0.065) 0.091∗ (0.052) –0.026 (0.044) 0.420∗ ∗ (0.181) 0.101 (0.127) –0.008 (0.011)

–0.001 (0.011) –0.074 (0.061) 0.115∗ ∗ (0.052) 0.031 (0.028) 0.379∗ ∗ (0.188) 0.106 (0.113) –0.010 (0.010)

0.002 (0.002) –0.003 (0.004) 0.081∗ ∗ ∗ (0.006)

0.001 (0.002) –0.003 (0.005) 0.079∗ ∗ ∗ (0.006)

0.004 (0.020) –0.074 (0.068) –0.001 (0.045)

–0.014 (0.011) –0.090 (0.054) 0.087∗ (0.043)

–0.002 (0.004) 0.018∗ (0.009) 0.009 (0.012) –0.0 0 0 (0.003) –0.002∗ ∗ ∗ (0.001) 0.004∗ ∗ (0.001) 0.0 0 0 (0.0 0 0) 0.0 0 0 (0.001) 0.003 (0.010)

0.0 0 0 (0.004) 0.022∗ ∗ (0.010) –0.001 (0.012) –0.0 0 0 (0.003) 0.0 0 0 (0.001) 0.004∗ ∗ ∗ (0.001) 0.0 0 0∗ (0.0 0 0) 0.0 0 0 (0.001) 0.003 (0.012) 0.125∗ ∗ ∗ (0.015)

0.008 (0.034) 0.234∗ ∗ (0.111) 0.130∗ ∗ (0.062) –0.011 (0.007) –0.005∗ ∗ (0.002) 0.010∗ (0.005) 0.004∗ ∗ (0.002) –0.003∗ (0.002) 0.069 (0.052) 0.233 (0.169)

0.034∗ ∗ (0.015) 0.214∗ ∗ (0.099) 0.059 (0.042) –0.006 (0.004) –0.001 (0.001) 0.007 (0.008) 0.0 0 0 (0.003) –0.001 (0.001) –0.029 (0.035) 0.190∗ ∗ ∗ (0.055)

0.272 8025 1895

0.274 7009 1848

7009 1848 0.578 0.080 0.348 0.595

7009 1848 0.545 0.113

AntiCor2

New Debt

∗

AntiCor2

Cash Flow

∗

AntiCor2

AntiCor2 Age Sales Tobin’s Q IPR Protection GDP Growth

–0.002∗ ∗ ∗ (0.001) 0.004∗ ∗ (0.001) 0.0 0 0 (0.0 0 0) 0.0 0 0 (0.001) 0.002 (0.011)

0.001 (0.001) 0.004∗ ∗ ∗ (0.001) 0.0 0 0 (0.0 0 0) 0.0 0 0 (0.001) 0.002 (0.012) 0.139∗ ∗ ∗ (0.014)

0.272 7724 1894

0.278 6768 1845

RD Intensity t–1 Adjusted R2 Observations Groups AR(2) (p-value) AR(3) (p-value) AR(4) (p-value) J (p-value)

–0.003 (0.002) 0.008 (0.008) –0.001 (0.002) –0.002 (0.002) 0.003 (0.049) 0.270∗ ∗ ∗ (0.096)

–0.001 (0.001) 0.008 (0.007) 0.0 0 0 (0.002) 0.001 (0.001) –0.019 (0.034) 0.217∗ ∗ ∗ (0.039)

6768 1845 0.973 0.272 0.919 0.214

6768 1845 0.552 0.124 0.111

0.188

Note: For columns (1), (2), (5) and (6), firm specific fixed effects are controlled for in the regressions. For columns (3), (4), (7) and (8), system GMM is employed to further deal with the potential endogeneity of the model. In columns (4) and (8), besides using lagged independent variables as their own IVs, we also use three external IVs, UKsettle, Churchuniv and Airborne for the anticorruption proxies. We use three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year dummies and industry dummies. All the regressions control for industry dummies, province dummies and year dummies which are not reported for brevity. Standard errors in parentheses are heteroskedasticityrobust and clustered at the provincial level. ∗ ∗ ∗ ,∗ ∗ and ∗ :significant at 1%, 5% and 10%, respectively.

invest a larger proportion of their newly acquired debt in funding R&D. However, the assumptions of fixed effect model require that all the independent variables are strictly exogenous to the idiosyncratic error (Woodridge, 2002), which is almost unlikely to hold in our specification.22 We thus turn to system GMM to deal with the potential endogeneity. The results of columns 3 and 4 are highly in line with the previous ones. According to the result in column 4, if a firm moves from the province with the lowest Anticor1 to the one with the highest Anticor1, the proportion of new debt invested in R&D increases from almost 0% to 27.9%. The results still hold and are highly robust across column 5 to column 8, where we use Anticor2 as the measure of anti-corruption efforts. Moreover, the coefficient on the interaction between new equity and anti-corruption also becomes significant at 5% level in column 8. The coefficients on age and sales are significant in most of the columns with the expected signs, while the coefficients of Tobin’s Q, GDP growth rate and IPR protection basically

22 Note that including the lagged R&D intensity in a regression with firm fixed effects also introduces the dynamic panel bias in a standard within-firm estimator (Nickell, 1981)

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC 14

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23 Table 6 The effect of anti-corruption on R&D output–patents.

AntiCor1

FE

FE

GMM

(1)

(2)

(3)

GMM with external IVs (4)

0.465∗ ∗ (0.186)

0.513∗ ∗∗ (0.183)

1.696∗∗∗ (0.502)

1.388∗∗∗ (0.461)

AntiCor2 Asset Age Leverage ROA Fixed Asset Intangable Asset IPR Protection GDP Growth

0.160∗∗∗ (0.057) 0.055 (0.075) 0.119 (0.181) 1.485∗∗∗ (0.522) 0.105 (0.228) –0.310 (0.911) 0.128 (0.126) 2.866∗∗ (1.366)

0.133∗∗ (0.053) 0.072 (0.072) 0.138 (0.224) 1.680∗∗∗ (0.413) 0.091 (0.264) –0.250 (0.999) 0.145 (0.129) 3.122∗∗ (1.400) 0.069∗∗∗ (0.023)

0.044 5778 1532

0.053 5243 1458

LnPatent t–1 Adjusted R2 Observations Groups AR(2) (p-value) AR(3) (p-value) J (p-value)

–0.864 (0.531) 0.105 (0.217) 1.042 (1.365) 0.310 (3.125) 0.022 (1.316) 2.391 (4.754) 0.188 (0.130) 3.428 (2.720) 0.224∗∗∗ (0.041)

–0.846 (0.530) 0.169 (0.207) 0.564 (1.375) –0.663 (2.809) –1.408 (1.004) 4.853 (4.265) 0.176 (0.134) 5.228∗∗ (2.617) 0.222∗∗∗ (0.050)

5243 1458 0.271 0.189 0.866

5243 1458 0.079 0.294 0.530

FE

FE

GMM

(5)

(6)

(7)

GMM with external IVs (8)

0.239∗∗ (0.103) 0.166∗∗∗ (0.056) 0.044 (0.074) 0.071 (0.177) 1.409∗∗ (0.513) 0.166 (0.218) –0.022 (0.919) 0.105 (0.124) 2.525∗ (1.319)

0.256∗∗ (0.104) 0.135∗∗ (0.052) 0.061 (0.071) 0.102 (0.217) 1.595∗∗∗ (0.415) 0.127 (0.247) 0.128 (1.005) 0.121 (0.128) 2.810∗∗ (1.350) 0.060∗∗∗ (0.021)

1.481∗∗ (0.689) –0.167 (0.906) 0.136 (0.278) 0.418 (2.097) 0.220 (5.727) 0.350 (1.731) 0.939 (4.397) 0.297 (0.264) 9.726∗∗∗ (3.699) 0.416∗∗ (0.192)

1.155∗∗ (0.546) –1.304 (0.783) 0.074 (0.263) 2.263 (1.916) 0.812 (5.707) –1.060 (1.696) 0.875 (4.978) 0.121 (0.212) 10.601∗∗ (4.112) 0.702∗∗∗ (0.163)

0.043 6106 1546

0.051 5537 1472

5537 1472 0.066 0.398 0.151

5537 1472 0.042 0.686 0.114

Note: For columns (1), (2), (5) and (6), firm specific fixed effects are controlled for in the regressions. For columns (3), (4), (7) and (8), system GMM is employed to further deal with the potential endogeneity of the model. In columns (4) and (8), besides using lagged independent variables as their own IVs, we also use three external IVs, UKsettle, Churchuniv and Airborne for the anticorruption proxies. We use three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year dummies and industry dummies. All the regressions control for industry dummies, province dummies and year dummies which are not reported for brevity. Standard errors in parentheses are heteroskedasticity-robust and clustered at the provincial level. ∗ ∗ ∗ ,∗ ∗ and ∗ :significant at 1%, 5% and 10%, respectively

remain statistically insignificant. This confirms our hypothesis that stronger anti-corruption intensity can better contribute to boosting firms’ investment in R&D. Table 6 reports the estimated effect of anti-corruption on firms’ innovation output measured by the number of patent applications. From columns 1 to 8, regardless of the model specifications, estimation techniques and anti-corruption measures, the coefficients on anti-corruption are always positive, statistically significant and quantitatively large. This evidence again supports the hypothesis that anti-corruption is conducive to corporate innovation. 4.3. The effect of the current massive anti-corruption campaign on R&D input As introduced in the Section 2, China has been undergoing a nationwide massive campaign against corruption launched by President Xi since he came to power in the late 2012. Thousands of officials, including more than 100 officials with very high ranks (the “big tigers”) have been swept out of offices because of corruption. The corrupt case involving Zhou Yongkang broke the unspoken rule of “PSC criminal immunity”, and sent out a clear message that anyone, no matter what his/her background and position is, cannot get away with corruption free of punishment. Thus this campaign is believed to generate a very strong deterring effect on officials, making them “unable and unwilling to be corrupt”, as Xi put it. Therefore we expect the effect of the current massive anti-corruption campaign should be larger than that of the previous ones. In this part, we try to separate the effects estimated in the last part and particularly focus on the effect of the current campaign under President Xi. Table 7 shows the estimated effect of the current large-scale anti-graft campaign on corporate R&D. We interact a XiCamp dummy, which indicates the period of the massive anti-corruption campaign launched by President Xi, with all the main independent variables in Eq. (2). We mainly focus on the triple interaction terms between funding sources, anti-corruption efforts and the XiCamp dummy. In columns 1 and 3, the XiCamp dummy is defined as years after 2013 (including 2013). This is natural because almost immediately after Xi came to power in the late 2012, he began to devote himself to cracking down on corruption with unprecedented zeal and acumen. The coefficients on the interaction term between new debt, antiPlease cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

15

Table 7 The effect of the current large-scale anti-corruption campaign on R&D investment.

New Equity New Debt Cash Flow New Equity

∗

XiCamp

New Debt

∗

XiCamp

Cash Flow

∗

XiCamp

New Equity

∗

AntiCor1

New Debt

∗

AntiCor1

Cash Flow

∗

AntiCor1

New Equity

∗

∗

AntiCor2

Cash Flow

∗

AntiCor2 ∗

AntiCor1

New Debt

∗

AntiCor1

∗

Cash Flow

∗

AntiCor1

∗

New Equity

∗

XiCamp=after 2014 (2)

XiCamp=after 2013 (3)

XiCamp=after 2014 (4)

–0.013 (0.021) 0.169∗∗∗ (0.061) 0.152∗∗∗ (0.040) 0.019 (0.023) –0.222∗∗∗ (0.064) –0.045 (0.043) 0.085 (0.084) –0.379 (0.297) 0.027 (0.224)

–0.086 (0.053) 0.114∗ (0.057) 0.135∗∗∗ (0.041) 0.099∗ (0.054) –0.268∗∗∗ (0.095) 0.004 (0.055) 0.431 (0.257) –0.223 (0.267) 0.064 (0.185)

–0.016 (0.026) 0.163∗∗∗ (0.061) 0.087 (0.053) 0.017 (0.027) –0.216∗∗∗ (0.069) –0.066 (0.051)

–0.047 (0.028) 0.153∗ (0.081) 0.049 (0.054) 0.048 (0.035) –0.280∗∗ (0.107) 0.091 (0.060)

0.051 (0.066) –0.278 (0.177) 0.111 (0.150)

0.124 (0.077) –0.273 (0.227) 0.260 (0.182)

–0.056 (0.069) 0.393∗∗ (0.189) –0.101 (0.146)

–0.107 (0.081) 0.521∗∗ (0.242) –0.288 (0.183)

7009 1848 0.432 0.005 0.468 0.069

7009 1848 0.664 0.227

AntiCor2

New Debt

New Equity

XiCamp=after 2013 (1)

AntiCor2

∗

XiCamp

XiCamp XiCamp ∗

–0.429 (0.254) 0.736∗∗ (0.290) –0.074 (0.182)

XiCamp

New Debt

∗

AntiCor2

∗

XiCamp

Cash Flow

∗

AntiCor2

∗

XiCamp

Observations Groups AR(2) (p-value) AR(3) (p-value) AR(4) (p-value) J (p-value)

–0.080 (0.085) 0.607∗∗ (0.286) –0.018 (0.212)

6768 1845 0.298 0.190

6768 1845 0.780 0.318

0.368

0.584

0.173

Note: This table investigates the effects of the currently ongoing anticorruption campaign launched by President Xi Jinping since he came to power in late 2012 on corporate R&D investment. All the regressions in this table are estimated by system GMM to further deal with the potential endogeneity of the model. Besides using lagged independent variables as their own IVs, we also use three external IVs, UKsettle, Churchuniv and Airborne for the anticorruption proxies. We use three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year dummies and industry dummies. We actually interacted the XiCamp dummy with all the variables in the specification in Table 5 except for province and industry dummies and only report the relevant variables. All the regressions control for industry dummies and province dummies which are not reported for brevity. Standard errors in parentheses are heteroskedasticity-robust and clustered at the provincial level. ∗ ∗ ∗ ,∗ ∗ and ∗ :significant at 1%, 5% and 10%, respectively.

corruption efforts and XiCamp dummy are positive and statistically significant in both specifications, and their magnitudes are much larger than those in Table 5. Furthermore, we find none of the interactions between funding sources and anticorruption are significant anymore. However, considering that it may take some time both for the campaign to take effect and for firms to react to the anti-corruption situation, in columns 2 and 4 we also specify the XiCamp dummy to be 1 if it is after year 2014 (including 2014). Highly in line with the previous results, the coefficients of the triple interaction between new debt, anti-corruption and XiCamp are again positive, statistically significant and even quantitatively larger than those in columns 1 and 3. This indicates that as the anti-corruption campaign continues and deepens, its full effect has been brought out gradually.23 Our empirical results uniformly demonstrate an economically and statistically significant impact of Xi’s anti-corruption campaign on corporate innovation. In contrast, the former campaigns under Xi’s predecessor seem to 23 According to the results in column 2, during Xi’s anti-corruption campaign, firms located in the province with the highest anticor1 will on average invest 32.4% (significant at 5% level) of their new debt in R&D, 16% larger than the corresponding value estimated in the baseline regression, while this proportion is not statistically different from 0 before the campaign, regardless of the anti-corruption intensity.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

JID: YJCEC 16

ARTICLE IN PRESS

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

have no real effect.24 Considering that former anti-corruption efforts were basically sporadic and unsystematic with limited intensity, Xi’s massive anti-graft campaign with its strong deterring effect is expected to have more far-reaching impact on the economy.

4.4. The heterogeneous effects of anti-corruption on corporate R&D We have found that stronger anti-corruption intensity contributes to more corporate innovation. However, up to now we have not specifically touched on the mechanisms through which this effect takes place. Besides the expropriation hypothesis, another mechanism may also be responsible for this effect. According to the rent-seeking hypothesis (Baumol, 1990; Murphy et al., 1991), stronger anti-corruption efforts may force former politically favored firms to substitute innovation for rent seeking due to the rising cost of corruption and the depreciated value of relational capital, particularly with officials that collapsed in the campaign. Note that even though expropriation hypothesis and rent-seeking hypothesis differ significantly in their working mechanisms, they are not mutually exclusive in an economy in that bureaucrats may use regulations to both create and extract rents from business activities (McChesney, 1987; Dong et al., 2016). However, the two mechanisms should impact on different types of firms, with the former acting on potential expropriation victims and the latter on potential rent seekers. Therefore we rely on widely used ex ante criteria to sort firms into two subsamples, one more susceptible to expropriation and one more likely to have access to rents, to test the two mechanisms separately. If the former group of firms spends significantly more in R&D in response to stronger anti-corruption efforts, the expropriation hypothesis is empirically verified. So is the rent-seeking hypothesis if the latter group of firms experiences a significant increase in R&D accordingly. Literature documents that connections with politicians can not only shield firms from expropriations by government but can also provide them with access to political rents (De Soto, 1990; Hellman et al., 2003; Chen et al., 2011), e.g., preferential access to government subsidies and finance. Two cases need to be further shed light on for our identification strategy. Firstly, while it’s possible that politicians also expropriate connected firms (e.g., SOEs and firms with political connections) to attain political goals (Shleifer, 1998; Cull and Xu, 2005), expropriation is by no means the channel through which corruption depresses innovation for connected firms. In China, the managers of SOEs generally face a different reward system, i.e., the prospect of promotion rather than increased pay associated with higher firm performance. These managers typically attach greater importance to social and political objectives, such as maintaining employment and climbing the political ladder, with little focus on profit or firm-value maximization (Shleifer, 1998; Chen et al., 2016). In this sense, they do not care whether the SOEs under their control are manipulated by government simply as means of channeling benefits, and they even exploit resources to bribe upper-level officials as long as this can boost their chance of promotion (Cull et al., 2015). Besides, politically connected firms have formed profit-sharing partnerships and simply on the same boat with local bureaucrats who own discretional powers (Dong et al., 2016). Secondly, although unconnected firms may also try to pursue additional benefit through rent seeking, due to the illegal nature of corruption, in an effort to reduce the risk of being exposed and the opportunism of contracting parties of the corrupt deals, bureaucrats generally collect bribes only from a small number of firms they have trust in (Li and Wu, 2010). This means without solid guanxi with government, firms may not be eligible to compete for the political rents.25 , 26 Besides the direct measures like firms’ political connection and ownership type,27 we also employ another two indirect criteria to split our sample firms. Heavily regulated industries in China are mainly the strategically important ones, which are generally associated with concentrated state ownership and close connections with government.28 Firms operating in regulated industries, where monopoly rents abound, are more incentivized to concentrate on rent seeking to extract more resources and scramble for more rents. Finally, we expect mature firms more likely to have guanxi relative to young firms, thus less susceptible to expropriation and more likely to engage in rent seeking since establishing guanxi with officials generally needs time and experience (Nee and Opper, 2012). Fig. 3 demonstrates that connected firms in general have significantly lower R&D intensity relative to their unconnected counterparts during our sample period.29 Except the group based on age, this gap becomes even larger since the start of Xi’s anti-corruption campaign.

24 This is highly consistent with Wedeman (2005) who argues that the Chinese style anti-corruption campaigns before Xi characterized by its low detection rate are unlikely to deter corruption. This indicates anti-corruption campaigns in a non-democratic country may only be effective if their intensity exceeds certain threshold. 25 However, we do not mean that unconnected firms make no efforts to seek political rents. In order to be eligible to compete for those rents, they have to establish certain connections at first. Before that, expropriation risk is a dominant factor that depresses innovation rather than potential access to rents. 26 We are grateful to the anonymous referee for helping us disentangle the two mechanisms in a clearer way. 27 Political background of top corporate executives and ownership type are two commonly used criteria to define political connections (Francis et al., 2009). A firm is coded as politically connected if any of the (vice) CEOs and (vice) chairmen is or was a government official of vice prefecture-department rank (tingjuji) and above or is a member of National People’s Congress (NPC), Chinese People’s Political Consultative Conference (CPPCC) or Congress of the Communist Party of China (CCPC) of the region with the rank not lower than the vice prefecture level (Diji Shi). A firm is coded as a SOE if the firm’s ultimate controlling shareholder is a state-related entity. 28 Heavily regulated industries in China mainly include the following ones: natural resources, mining, public utilities, real estate, telecommunications, communication and culture. 29 This pattern is consistent with the finding of Lin et al (2010).

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

JID: YJCEC

ARTICLE IN PRESS G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

[m3Gsc;October 20, 2016;22:17] 17

Fig. 3. Time trend of R&D intensity: connected vs. unconnected.

Table 8 shows the heterogeneous effects of anti-corruption on R&D investment based on whether firms have political connections or are SOEs respectively.30 We find in provinces with stronger anti-corruption intensity, companies without political connections invest a significantly larger proportion of new debt in R&D (columns 1 and 3). In addition to new debt, non-SOEs also invest significantly more of their newly acquired equity in R&D (columns 5 and 7). In contrast, we fail to reach similar conclusions for politically connected firms nor SOEs (columns 2, 4, 6 and 8). Similarly, in Table 9 we find that stronger anti-corruption efforts boost more R&D investment only for firms operating in non-regulated industries and young firms (columns 1, 3, 5 and 7).31 All the above results are consistent with the expropriation mechanism, indicating that the expropriation from corrupt officials poses a serious threat to innovation and is particularly responsible for the low level of R&D investment for Chinese firms.32 This also implies that anti-corruption campaigns may benefit a very wide range of firms, as those with no background thus susceptible to expropriation by government make up the majority in the economy.33 Moreover, the insignificant results for latter group of firms also indicate that anti-corruption campaign alone is insufficient to boost innovation for these connected firms as they may still be able to enjoy some implicit favorable treatments that do not apparently involve corruption. In this sense, further initiatives to reform the government-business relationship should be put into agenda.34

30

The results using number of patents as the dependent variable are reported in Table A2. A firm is coded as young if its age since firm’s establishment is less than the median, and old otherwise. 32 The mean (median) of R&D intensity for our Chinese listed firms is 0.022 (0.019) while the corresponding value for European and US listed firms in Brown et al. (2012) and Borisova and Brown (2013) is 0.085 (0.041) and 0.105 (0.057) respectively. 33 Our results may only capture the lower bound of the real effect of anti-corruption on innovation as the sample only consists of listed firms. Among non-listed firms, those having certain kinds of guanxi with government or bureaucrats should be much rarer than their listed counterpart. 34 One potential concern regarding our result is that anti-corruption campaigns, which can diminish the value of connections, may make previously connected firms more financially constrained, thus leading to lower investment in R&D even if they wish to engage in innovative activity. If this is really the case, the insignificant results for connected groups may be attributed to a combination of intensified financial constraints and the rent-seeking hypothesis. However, this is unlikely in our case due to two reasons. First of all, the coefficient of Cashflow as well as that of Cashflow∗ Anticor is always insignificant for connected groups, indicating that the connected firms are not financially constrained. Moreover, we regress a measure of firm-level financial constraints, the WW index (Whited and Wu, 2006), on Anticor, Anticor∗ PC (or Anticor∗ SOE), Anticor∗ PC∗ XiCamp (or Anticor∗ SOE∗ XiCamp) as well as other controls and dummies (Table OA1). We find none of these coefficients is significant. The results are similar for firms operating in regulated industries and mature firms and robust to alternative financial constraint measures, like KZ index. The evidence suggests that anti-corruption does not make connected firms so financially constrained that they are unable to afford sufficient R&D expenses. We thank an anonymous referee for pointing this out. 31

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC 18

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23 Table 8 The effect of anti-corruption on R&D investment based on political connection status and ownership type. Political Connection Status

New Equity New Debt Cash Flow New Equity

∗

AntiCor1

New Debt

∗

AntiCor1

Cash Flow

∗

AntiCor1

AntiCor1 New Equity

∗

(1)Non-PC

(2)PC

(3)Non-PC

(4)PC

(5)Non-SOE

(6)SOE

(7)Non-SOE

(8)SOE

–0.021 (0.022) –0.051 (0.039) 0.110∗∗ (0.043) 0.076 (0.064) 0.250∗∗∗ (0.094) 0.098 (0.119) –0.011 (0.012)

0.012 (0.013) –0.030 (0.049) 0.083 (0.098) –0.026 (0.027) 0.079 (0.122) 0.095 (0.137) –0.003 (0.011)

0.013 (0.013) –0.049 (0.030) 0.119∗∗ (0.055)

0.010 (0.011) 0.007 (0.041) 0.096 (0.058)

–0.014 (0.012) –0.088 (0.053) 0.147∗∗∗ (0.057) 0.049∗∗ (0.022) 0.390∗∗ (0.157) 0.011 (0.089) 0.003 (0.010)

0.005 (0.018) –0.025 (0.034) 0.049 (0.036) 0.037 (0.059) 0.144 (0.102) –0.012 (0.081) –0.005 (0.007)

–0.005 (0.011) –0.077 (0.047) 0.123∗∗∗ (0.045)

0.014 (0.013) –0.066 (0.042) 0.065 (0.060)

–0.011 (0.023) 0.173∗∗∗ (0.057) 0.053 (0.062) –0.006 (0.008)

–0.034 (0.021) –0.057 (0.072) 0.091 (0.071) –0.001 (0.006)

0.034∗∗ (0.014) 0.174∗∗ (0.084) –0.019 (0.058) 0.003 (0.007)

–0.018 (0.026) 0.132 (0.086) 0.039 (0.080) –0.005 (0.009)

3168 994 0.404 0.641

3841 1053 0.096 0.186 0.548 0.514

4638 1259 0.009 0.611 0.959 0.978

2371 623 0.077 0.105

AntiCor2

New Debt

∗

AntiCor2

Cash Flow

∗

AntiCor2

AntiCor2 Observations Groups AR(2) (p-value) AR(3) (p-value) AR(4) (p-value) J (p-value)

Ownership Type

3081 990 0.781 0.711 0.490

3687 1050 0.288 0.509 0.760 0.349

0.787

4511 1257 0.013 0.528

2257 622 0.094 0.152

0.231

0.183

0.317

Note: This table investigates the heterogeneous effects of anticorruption on corporate R&D investment across firms with different political connection status and ownership type. All the regressions in this table are estimated by system GMM to further deal with the potential endogeneity of the model. Besides using lagged independent variables as their own IVs, we also use three external IVs, UKsettle, Churchuniv and Airborne for the anticorruption proxies. We use three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year dummies and industry dummies. All the regressions control for industry dummies, province dummies and year dummies which are not reported for brevity. Standard errors in parentheses are heteroskedasticity-robust and clustered at the provincial level. ∗∗∗ ∗∗ , and ∗ :significant at 1%, 5% and 10%, respectively.

5. Robustness checks In this part, we show that the findings above remain robust to a number of alternative specifications in Table 10. First of all, to account for the possibility that it may take some time for firms to alter their R&D investment decisions in response to the anti-corruption situation, we use one-year lagged anti-corruption measures (columns 1 and 2). Secondly, following the model specification developed by Bond and Meghir (1994), we further control for lagged squared R&D to construct an Euler Equation (columns 3 and 4). Thirdly, we use an alternative measure for corporate innovation—intangible assets investment (columns 5 and 6). Since intangible capital mainly includes R&D capital and human capital as a result of training, etc, intangible capital formation can be thought as an investment for innovation in a broad sense and has been proved to have played very important roles in promoting economic growth (e.g., Fukao et al., 2009; Tomer, 2012). Corrado et al. (2013) also measure innovation based on intangible investment. In all of the specifications from columns 1 to 6, we can still find at least one coefficient of the interaction terms between funding sources and anti-corruption efforts that is positive and statistically significant. These again substantiate our hypothesis that anti-corruption can play a crucial role in spurring corporate R&D and innovation.35 6. Conclusions Now we can turn to answering the questions raised at the beginning of the paper: How do anti-corruption campaigns affect the economy? Is it beneficial or detrimental to economic growth? This paper examines this important topic by shedding light on the causal effect of anti-corruption efforts on innovation, which is a main driver for economic growth and is responsible for the accumulation of economic welfares. By examining the anti-corruption efforts in different provinces within China, we find that anti-corruption efforts do positively affect the financing and investing of corporate R&D as well 35

Our results are also robust to the inclusion of industry-by-year fixed effects (Tables OA3 and OA4).

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

19

Table 9 The effect of anti-corruption on R&D investment based on industry attributes and age. Industry Attributes

New Equity New Debt Cash Flow New Equity

∗

AntiCor1

New Debt

∗

AntiCor1

Cash Flow

∗

AntiCor1

AntiCor1 New Equity

∗

(1)Non-Reg

(2)Regulated

(3)Non-Reg

(4)Regulated

(5)Young

(6)Mature

(7)Young

(8)Mature

0.018 (0.014) –0.112 (0.070) 0.188∗∗∗ (0.045) 0.007 (0.030) 0.454∗∗ (0.194) 0.173∗ (0.089) –0.016 (0.012)

–0.001 (0.026) 0.100∗∗ (0.045) 0.037 (0.056) 0.013 (0.114) –0.312 (0.192) –0.038 (0.117) –0.002 (0.010)

0.010 (0.012) –0.072 (0.043) 0.109∗∗ (0.053)

–0.055 (0.033) 0.002 (0.057) –0.007 (0.064)

–0.024 (0.026) –0.127 (0.075) 0.171∗∗∗ (0.061) 0.123∗ (0.064) 0.401∗∗ (0.193) –0.075 (0.154) –0.012 (0.029)

0.015 (0.025) 0.052 (0.092) 0.011 (0.050) –0.022 (0.082) –0.092 (0.325) 0.285 (0.239) –0.059 (0.035)

–0.034 (0.022) –0.091 (0.056) 0.143∗∗∗ (0.048)

0.022∗ (0.013) –0.035 (0.042) 0.074 (0.050)

0.012 (0.020) 0.184∗∗∗ (0.056) 0.097 (0.075) –0.007 (0.007)

0.142 (0.086) 0.050 (0.153) –0.018 (0.093) 0.002 (0.009)

0.051∗ (0.030) 0.250∗∗∗ (0.096) 0.061 (0.096) –0.010 (0.012)

–0.036 (0.026) 0.091 (0.078) –0.008 (0.051) 0.008 (0.006)

5957 1570 0.228 0.660 0.507 0.411

1052 325 0.161 0.221

3418 1127 0.260 0.378

3350 1225 0.252 0.723

3563 1141 0.539 0.321

3446 1227 0.069 0.567

0.706

0.926

0.114

0.812

0.129

AntiCor2

New Debt

∗

AntiCor2

Cash Flow

∗

AntiCor2

AntiCor2 Observations Groups AR(2) (p-value) AR(3) (p-value) AR(4) (p-value) J (p-value)

Age

5762 1567 0.008 0.529

1006 324 0.304 0.117

0.132

0.122

Note: This table investigates the heterogeneous effect of anticorruption on corporate R&D investment across firms in different industries and with different ages. All the regressions in this table are estimated by system GMM to further deal with the potential endogeneity of the model. Besides using lagged independent variables as their own IVs, we also use three external IVs, UKsettle, Churchuniv and Airborne for the anticorruption proxies. We use three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year dummies and industry dummies. All the regressions control for industry dummies, province dummies and year dummies which are not reported for brevity. Standard errors in parentheses are heteroskedasticity-robust and clustered at the provincial level. ∗ ∗ ∗ ,∗ ∗ and ∗ :significant at 1%, 5% and 10%, respectively.

as its innovation output among Chinese listed firms. Further empirical tests suggest that this significant and positive effect is entirely driven by the massive anti-corruption campaign launched by President Xi since 2013. The empirical results also suggest that only firms without political background thus susceptible to expropriations invest significantly more in R&D in response to the stronger anti-corruption efforts, consistent with the expropriation hypothesis. Some critics and observers argue that the current anti-corruption campaign under President Xi may have taken its toll on the economy based on the fact that the economic growth of China has begun to slow down since 2014. However, we should not establish a causal relationship between the large-scale anti-corruption campaign and the economic downturn so easily. China, as the largest emerging economy in the world, is going through a process of economic structure adjustment and updating after experiencing an average 10% growth rate of so many years. It’s more rational to view this as a transition into a new stage of development—the New Normal as Xi put it— rather than signs of economic recessions. This transition is inevitable and can even be painful regardless of the anti-corruption campaign. Moreover, the superficial prosperity upheld by an underlying “corruption economy” is unsustainable and fundamentally harmful in the long run. Instead creating a corrupt-free and fair business environment has been proved by both theories and practices to be a sustainable and effective way to boost economic development. The findings in this paper suggest that public policy efforts to protect firms from expropriations particularly from corrupt bureaucrats, either through massive anti-corruption campaign or strictly enforcing property right protections, can have a pronounced effect on innovation and intangible investment. Moreover, it’s also imperative to take further initiatives to reform the government-business relationship in order to boost innovation for firms with certain connections to government. As demonstrated by this paper, facilitating financing and investing in R&D is one of the potential channels through which anti-corruption efforts can positively affect economic growth.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC 20

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23 Table 10 Robustness Checks.

New Equity New Debt Cash Flow New Equity

∗

AntiCor1

New Debt

∗

AntiCor1

Cash Flow

∗

AntiCor1

AntiCor1 New Equity

∗

One-year Lagged Anti-corruption

Euler Equation

(1)

(2)

(3)

(4)

(5)

(6)

0.002 (0.011) –0.053 (0.047) 0.099∗ (0.058) 0.016 (0.027) 0.313∗∗ (0.156) 0.074 (0.135) 0.001 (0.011)

0.007 (0.010) –0.043 (0.031) 0.173∗∗∗ (0.054)

0.005 (0.009) –0.106 (0.064) 0.007 (0.041) 0.010 (0.020) 0.432∗∗ (0.202) 0.194∗ (0.105) –0.017 (0.011)

–0.019 (0.013) –0.058 (0.039) 0.104∗∗∗ (0.036)

–0.013 (0.018) –0.093 (0.079) –0.111 (0.081) 0.103∗∗ (0.046) 0.463∗∗ (0.226) 0.296∗ (0.156) –0.026 (0.018)

0.003 (0.018) –0.060 (0.063) –0.126 (0.082)

AntiCor2

New Debt

∗

AntiCor2

Cash Flow

∗

AntiCor2

–0.008 (0.016) 0.165∗∗∗ (0.054) –0.012 (0.065) 0.005 (0.005)

AntiCor2

–8.531∗∗ (3.236)

RD Intensity Square t–1

0.037∗∗ (0.017) 0.180∗∗ (0.075) 0.077 (0.061) –0.005 (0.004) –4.604∗∗∗ (1.116)

Intangible Investment t–1 Observations Groups AR(2) (p-value) AR(3) (p-value) AR(4) (p-value) J (p-value)

6674 1843 0.054 0.791

7009 1848 0.127 0.149

0.159

0.404

6768 1845 0.397 0.143 0.561 0.861

Alternative Innovation Proxy

7009 1848 0.888 0.106 0.666

0.044∗∗ (0.022) 0.244∗∗ (0.095) 0.152∗ (0.087) –0.012 (0.009)

–0.432∗∗∗ (0.120)

–0.241∗ (0.136)

7492 1910 0.0 0 0 0.853 0.856 0.395

7816 1914 0.068 0.806 0.787 0.446

Note: In column (1) and (2), we use one-year lagged anticorruption measures to account for the possibility that it may take some time for firms to alter their R&D expenditures in response to the anti-corruption situation. In columns (3) and (4), we further control for lagged squared R&D investment to construct an Euler Equation. In columns (5) and (6), we use an alternative measure for corporate innovation—intangible assets investment. System GMM is applied in all the regressions to further deal with the potential endogeneity of the model. We also use three external IVs, UKsettle, Churchuniv and Airborne for the anticorruption proxies besides using lagged regressors as their own IVs. We use three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year dummies and industry dummies. All the regressions control for industry dummies, province dummies and year dummies which are not reported for brevity. Standard errors in parentheses are heteroskedasticity-robust and clustered at the provincial level. ∗ ∗ ∗ ,∗ ∗ and ∗ :significant at 1%, 5% and 10%, respectively

Acknowledgement We would like to thank Professor Daniel Berkowitz (the editor), an anonymous referee and Vixathep Souksavanh for their insightful comments and suggestions, which greatly improve the paper. We are also grateful to Dongyang Zhang for sharing the patent data of SIPO with us. This work was funded by Scientific Research (C) No. 26380296 from the Ministry of Education, Science, Sports, and Culture of Japan (MESSC). All the remaining errors are our own. Supplementary materials Supplementary data associated with this article can be found, in the online version, at 10.1016/j.jce.2016.10.001.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

21

Appendix

Table A1 Definition of key variables. Variable

Definition

RD intensity LnPatent New Equity New Debt Cash Flow Age Access to NewDebt Access to NewEquity Fixed Asset Intangible Asset PC

SOE Regulated

Anticor1 Anticor2 IPR Protection

GDP Growth Airborne

UKsettle Churchuniv

The R&D expenditures scaled by the book value of total assets at the beginning of the year The natural logarithm of one plus the number of patents applied for during a given year The net increase in book equity minus the net increase in retained earnings scaled by the book value of total assets at the beginning of the year The increase in long-term debt scaled by the book value of total assets at the beginning of the year The net profits plus depreciation and amortization plus R&D expenses scaled by the book value of total assets at the beginning of the year The natural logarithm of current year minus the year when the firm went public. A dummy that indicates whether the amount of newly acquired long-term debt in a given year exceeds 5% of the book value of its total assets at the beginning of the year A dummy that indicates whether the amount of newly acquired external equity in a given year exceeds 5% of the book value of its total assets at the beginning of the year The ratio of net fixed assets to the book value of total assets at the end of the year The ratio of net intangible assets to the book value of total assets at the end of the year A dummy indicating whether the firm has political connections. A firm is coded as politically connected if any of the (vice) CEOs and (vice) chairmen is or was a government official of vice prefecture-department rank (Tingju ji) and above or is a member of NPC, CPPCC, CCPC of the region with the rank not lower than vice prefecture level (Diji Shi). A dummy indicating whether the firm’s ultimate controlling shareholder is a state-related entity A dummy indicating whether the firm operates in a heavily regulated industry. Heavily regulated industries in China mainly include the following ones: natural resources, mining, public utilities, real estate, telecommunications, communication and culture. The provincial number of officials of vice county-division rank (Xianchu ji) and above investigated in the registered cases on corruption per 100 thousand population in a given year The number of articles in the provincial CPC newspaper that advocate anticorruption or denounce corruption (divided by 100) in a given year The natural logarithm of the sum of two sub-indices in Fan et al. (2011) that measure the extent of provincial intellectual property rights protection in a given year, i.e., the number of three kinds of patent applications accepted and granted per 10,0 0 0 science and technology personnel The yearly real GDP growth rate A dummy that indicates whether the province’s Secretary of the Commission for Discipline Inspection (SCDI) is transferred directly from other provinces or from the central government/departments rather than locally promoted from the same province A dummy that indicates whether the province was once a concession or leased territory of the Great Britain from the late Qing Dynasty to the early years of the Republic The number of colleges in the province that were established by western missionaries (including UK, US, Canada, South America and so on)

Table A2 The effect of anti-corruption on patents based on political connection status and ownership type. Political Connection Status

AntiCor1

(1)Non-PC

(2)PC

1.394∗ ∗ (0.666)

2.973 (2.166)

AntiCor2 Asset Age Leverage ROA

0.571 (0.530) –0.289 (0.236) 2.520∗ (1.362) 2.882 (2.496)

–2.718∗ (1.459) –0.051 (0.327) 3.979 (2.664) 1.562 (5.240)

Ownership Type (1)Non-PC

0.999∗ ∗ ∗ (0.385) 0.929∗ (0.478) –0.188 (0.160) 1.188 (0.857) –0.416 (2.802)

(2)PC

0.943 (0.868) –1.175 (0.904) 0.093 (0.202) 2.873 (1.748) –1.559 (5.085)

(1)Non-SOE

(2)SOE

1.952∗ ∗ ∗ (0.748)

0.585 (0.665)

–1.090∗ ∗ ∗ (0.367) –0.308 (0.205) 2.525∗ ∗ (1.068) –0.756 (3.022)

–0.274 (0.584) 0.078 (0.521) 0.920 (1.794) 1.081 (2.508)

(1)Non-SOE

(2)SOE

1.980∗ ∗ ∗ (0.702) –1.297 (0.882) –0.629 (0.521) 5.167∗ ∗ (2.620) –4.834 (8.371)

0.479 (0.416) –0.500 (0.582) 0.219 (0.375) 2.154 (1.674) 2.699 (2.671)

(continued on next page)

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

ARTICLE IN PRESS

JID: YJCEC 22

[m3Gsc;October 20, 2016;22:17]

G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23 Table A2 (continued) Political Connection Status

Fixed Asset Intangable Asset IPR Protection GDP Growth LnPatent t–1 Observations Groups AR(2) (p-value) AR(3) (p-value) J (p-value)

Ownership Type

(1)Non-PC

(2)PC

(1)Non-PC

(2)PC

(1)Non-SOE

(2)SOE

(1)Non-SOE

(2)SOE

–0.335 (1.194) 1.264 (4.528) –0.292 (0.182) –1.990 (3.874) 0.137∗ (0.075) 2301 772 0.361 0.560 0.525

–2.869 (2.063) 8.329 (9.337) 0.217 (0.199) 8.477 (5.964) 0.579∗ ∗ (0.241) 2942 847 0.940 0.133 0.116

–0.020 (1.159) 0.276 (3.738) –0.076 (0.199) 2.461 (2.703) –0.114 (0.149) 2407 781 0.384 0.748 0.299

–1.713 (1.510) 0.956 (5.039) 0.348∗ (0.202) 7.475∗ ∗ (3.735) 0.350∗ ∗ ∗ (0.102) 3130 857 0.688 0.958 0.133

–0.266 (1.469) 3.416 (2.964) –0.011 (0.103) 8.676∗ ∗ ∗ (2.941) 0.162∗ ∗ ∗ (0.053) 3524 1024 0.210 0.631 0.326

0.079 (1.175) –0.350 (5.732) 0.519∗ ∗ ∗ (0.165) –1.183 (2.792) 0.269∗ ∗ ∗ (0.099) 1719 464 0.548 0.129 0.258

1.815 (1.771) –0.474 (5.969) 0.320 (0.290) 9.859 (6.394) 0.127 (0.127) 3672 1031 0.643 0.785 0.497

0.017 (1.433) –1.333 (6.127) 0.504∗ ∗ ∗ (0.188) 1.052 (2.705) 0.312∗ ∗ ∗ (0.097) 1865 472 0.433 0.216 0.357

Note: This table investigates the heterogeneous effects of anti-corruption on patents across firms with different political connection status and ownership type. All the regressions in this table are estimated by system GMM to further deal with the potential endogeneity of the model. Besides using lagged independent variables as their own IVs, we also use three external IVs, UKsettle, Churchuniv and Airborne for the anticorruption proxies. We use three or four lags of our regressors as instruments and report the p-value of m2, m3 and m4 tests in the tables. We treat all the independent variables as potentially endogenous except for age, year dummies and industry dummies. All the regressions control for industry dummies, province dummies and year dummies which are not reported for brevity. Standard errors in parentheses are heteroskedasticity-robust and clustered at the provincial level. ∗ ∗ ∗ ,∗ ∗ and ∗ : significant at 1%, 5% and 10%, respectively.

References Aidt, T.S., 2009. Corruption, institutions, and economic development. Oxford Rev. Econ. Policy 25 (2), 271–291. Ang, J.S., Cheng, Y., Wu, C., 2014. Does enforcement of intellectual property rights matter in China? Evidence from financing and investment choices in the high-tech industry. Rev. Econ. Stat. 96 (2), 332–348. Anokhin, S., Schulze, W.S., 2009. Entrepreneurship, innovation, and corruption. J. Bus. Venturing 24 (5), 465–476. Arellano, M., Bover, O., 1995. Another look at the instrumental variable estimation of error-components models. J. Economet. 68 (1), 29–51. Baker, M., Stein, J.C., Wurgler, J., 2003. When does the market matter? Stock prices and the investment of equity-dependent firms. Q. J. Econ. 969–1005. Baumol, W.J., 1990. Entrepreneurship: productive, unproductive, and destructive. J. Polit. Economy 893–921. Bellettini, G., Berti Ceroni, C., Prarolo, G., 2013. Persistence of politicians and firms’ innovation. Econ. Inq. 51 (4), 2056–2070. Benfratello, L., Schiantarelli, F., Sembenelli, A., 2008. Banks and innovation: microeconometric evidence on Italian firms. J. Financ. Econ. 90 (2), 197–217. Berkowitz, D., Lin, C., Ma, Y., 2015. Do property rights matter? Evidence from a property law enactment. J. Financ. Econ. 116 (3), 583–593. Blackburn, K., Forgues-Puccio, G.F., 2009. Why is corruption less harmful in some countries than in others? J. Econ. Behav. Organ. 72 (3), 797–810. Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel data models. J. Economet. 87 (1), 115–143. Bond, S., Meghir, C., 1994. Dynamic investment models and the firm’s financial policy. Rev. Econ. Stud. 61 (2), 197–222. Borisova, G., Brown, J.R., 2013. R&D sensitivity to asset sale proceeds: new evidence on financing constraints and intangible investment. J. Banking Finance 37 (1), 159–173. Brown, J.R., Martinsson, G., Petersen, B.C., 2012. Do financing constraints matter for R&D? Eur. Econ. Rev. 56 (8), 1512–1529. Cai, H., Fang, H., Xu, L., 2011. Eat, drink, firms and governments: an investigation of corruption from entertainment expenditures in Chinese firms. J. Law Econ. 54, 55–78. Chen, C.J., Li, Z., Su, X., Sun, Z., 2011. Rent-seeking incentives, corporate political connections, and the control structure of private firms: Chinese evidence. J. Corp. Finance 17 (2), 229–243. Chen, D., Ljungqvist, A., Jiang, D., Lu, H., & Zhou, M., 2016. State Capitalism vs. Private Enterprise. Private Enterprise (March 23, 2016). Cole, M.A., Elliott, R.J., Zhang, J., 2009. Corruption, governance and FDI location in China: a province-level analysis. J. Dev. Stud. 45 (9), 1494–1512. Corrado, C., Haskel, J., Jona-Lasinio, C., Iommi, M., 2013. Innovation and intangible investment in Europe, Japan, and the United States. Oxford Rev. Econ. Policy 29 (2), 261–286. Cull, R., Xu, L.C., 2005. Institutions, ownership, and finance: the determinants of profit reinvestment among Chinese firms. J. Financ. Econ. 77 (1), 117–146. Cull, R., Li, W., Sun, B., Xu, L.C., 2015. Government connections and financial constraints: evidence from a large representative sample of Chinese firms. J. Corp. Finance 32, 271–294. David, P., O’Brien, J.P., Yoshikawa, T., 2008. The implications of debt heterogeneity for R&D investment and firm performance. Acad. Manage. J. 51 (1), 165–181. De Haan, L., Hinloopen, J., 2003. Preference hierarchies for internal finance, bank loans, bond, and share issues: evidence for Dutch firms. J. Empirical Finance 10 (5), 661–681. De Soto, H., 1990. The Other Path. Harper and Row, New York. Dong, B., Torgler, B., 2013. Causes of corruption: evidence from China. China Econ. Rev. 26, 152–169. Dong, Z., Wei, X., Zhang, Y., 2016. The allocation of entrepreneurial efforts in a rent-seeking society: evidence from China. J. Comp. Econ. 44 (2), 353–371. Fan, G., Wang, X.L. & Zhu, H.P., 2011. NERI Index of Marketization of China’s Provinces, 2011 Report (Beijing: Economic Science Press, 2011) Fisman, R., Gatti, R., 2002. Decentralization and corruption: evidence across countries. J. Public Econ. 83 (3), 325–345. Fisman, R., Svensson, J., 2007. Are corruption and taxation really harmful to growth? Firm level evidence. J. Dev. Econ. 83, 63–75. Francis, B.B., Hasan, I., Sun, X., 2009. Political connections and the process of going public: evidence from China. J. Int. Money Finance 28 (4), 696–719. Fukao, K., Miyagawa, T., Mukai, K., Shinoda, Y., Tonogi, K., 2009. Intangible investment in Japan: measurement and contribution to economic growth. Rev. Income Wealth 55 (3), 717–736. Glaeser, E.L., Saks, R.E., 2006. Corruption in America. J. Public Econ. 90 (6), 1053–1072. Habiyaremye, A., & Raymond, W., 2013. Transnational corruption Corruption and innovation Innovation in transition Transition economiesEconomies, (unpublished working papers). Hall, B.H., 2002. The financing of research and development. Oxford Rev. Econ. Policy 18 (1), 35–51.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001

JID: YJCEC

ARTICLE IN PRESS G. Xu, G. Yano / Journal of Comparative Economics 000 (2016) 1–23

[m3Gsc;October 20, 2016;22:17] 23

Hellman, J.S., Jones, G., Kaufmann, D., 2003. Seize the state, seize the day: state capture and influence in transition economies. J. Comp. Econ. 31 (4), 751–773. Himmelberg, C.P., Petersen, B.C., 1994. R&D and internal finance: a panel study of small firms in high-tech industries. Rev. Econ. Stat. 38–51. Hovakimian, A., Opler, T., Titman, S., 2001. The debt-equity choice. J. Financ. Quant. Anal. 36 (01), 1–24. Jones, B.F., Olken, B.A., 2005. Do leaders matter? National leadership and growth since world war II. Q. J. Econ. 120 (3), 835–864. Kim, T., 2015. Does a Firm’s Political Capital Affect Its Investment and Innovation? University of Illinois working paper. La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R., 1999. The quality of government. J. Law Econ. Organ. 15 (1), 222–279. Li, S., Wu, J., 2010. Why some countries thrive despite corruption: the role of trust in the corruption–efficiency relationship. Rev. Int. Polit. Economy 17 (1), 129–154. Lin, C., Lin, P., Song, F., 2010. Property rights protection and corporate R&D: evidence from China. J. Dev. Econ. 93 (1), 49–62. Mauro, P., 1995. Corruption and growth. Q. J. Econ. 110, 681–712. McChesney, F.S., 1987. Rent extraction and rent creation in the economic theory of regulation. J. Legal Stud. 16 (1), 101–118. Mo, P., 2001. Corruption and economic growth. J. Comp. Econ. 29, 66–97. Murphy, K.M., Shleifer, A., Vishny, R.W., 1991. The Allocation of talent: implications for growth. Q. J. Econ. 106 (2), 503–530. Nee, V., Opper, S., 2012. Capitalism from Below. Harvard University Press. Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica 1417–1426. Nie, H., Wang, M., 2016. Are foreign monks better at chanting? The effect of ‘airborne’ SDICs on anti-corruption. Econ. Polit. Stud. 4 (1), 19–38. Opler, T.C., Titman, S., 1994. Financial distress and corporate performance. J. Finance 49 (3), 1015–1040. Paunov, C., 2016. Corruption’s asymmetric impacts on firm innovation. J. Dev. Econ. 118, 216–231. Shleifer, A., 1998. State versus private ownership. J. Econ. Perspect. 12 (4), 133–150. Stauffer, M.T., Wong, T.C., Tewksbury, M.G., 1922. The Christian Occupation of China: A General Survey of the Numerical Strength and Geographical Distribution of the Christian Forces in China. China Continuation Committee Press, Shanghai 1922. Stiglitz, J.E., Weiss, A., 1981. Credit rationing in markets with imperfect information. Am. Economic Rev. 71 (3), 393–410. Tan, Y., Tian, X., Zhang, X., & Zhao, H., 2015. The real Real effects Effects of privatizationPrivatization: Evidence from China’s split Split share Share structure Structure reformReform. Kelley School of Business Research Paper, (2014-33). Tomer, J.F., 2012. Intangible capital and economic growth. Int. J. Behav. Healthcare Res. 3 (3-4), 178–197. Waldemar, F.S., 2012. New products and corruption: evidence from Indian firms. Developing Econ. 50 (3), 268–284. Wederman, A., 2004. The intensification of corruption in China. China Q. 180, 895–921. Wedeman, A., 2005. Anticorruption campaigns and the intensification of corruption in China. J. Contemp. China 14 (42), 93–116. Wei, S.J., 20 0 0. How taxing is corruption on international investors? Rev. Econ. Stat. 82 (1), 1–11. Whited, T.M., Wu, G., 2006. Financial constraints risk. Rev. Financ. Stud. 19 (2), 531–559. Woodridge, J.M., 2002. Econometric Analysis of Cross Sectional Data and Panel Data. MIT Press, Cambridge. Yang, Z., Ye, F., 1993. Studies on Semi-Colonization of Qing Dynasty. Higher Education Press, Beijing.

Please cite this article as: G. Xu, G. Yano, How does anti-corruption affect corporate innovation? Evidence from recent anti-corruption efforts in China, Journal of Comparative Economics (2016), http://dx.doi.org/10.1016/j.jce.2016.10.001