Is the Chinese Corporate Anti-Corruption Campaign Genuine?

JOHN GRIFFIN

CLARK LIU

TAO SHU*

February, 2018

* We are very grateful for helpful comments from Vikas Agarwal, Lin Chen, Mark Chen, Yongheng Deng, Gerald Gay, Jie (Jack) He, Sara Holland, Omesh Kini, Jun Hoon Lee, David Lesmond, Harold Mulherin, Jeffry Netter, Xuhui Pan, Bradley Paye, Annette Poulsen, Clemens Sialm, Baolian Wang, Yongxiang Wang, Bernard Yeung, Xiaoyun Yu, and participants at 2017 AFE Meeting, 2017 ABFER Meeting, the Fifth Symposium on Emerging Financial Markets: China and Beyond, 2016 China International Conference of Finance, 2016 MIT Asian Conference in Accounting, 2016 China Financial Research Conference, Tulane University, University of Georgia, and Georgia State University. We thank many research assistants including Haorui Bai, Hunter Bezner, Tianyuan Liu, Kush Patel, and Wenqing Zhao for their research support. Griffin is at the McCombs School of Business, University of Texas, Austin. Email: [email protected]. Liu is at the PBC School of Finance, Tsinghua University. Email: [email protected]. Shu is at the Terry College of Business, University of Georgia. Email: [email protected].

Is the Chinese Corporate Anti-Corruption Campaign Genuine?

February, 2018

Although one of the largest anti-corruption initiatives in modern history, little is known as to whether the Chinese anti-corruption campaign is ensaring corrupt firms, contains political bias, and is effective in reducing corporate corruption. Consistent with the anti-corruption campaign investigating corrupt firms, Chinese firms with characteristics commonly associated with measures of poor governance and corporate self-dealing are more likely to have their executives investigated by the campaign. However, specific affiliations with the investigated former leaders increases investigation likelihood, while connections with the current leaders decrease investigation likelihood, indicating possible political favoritism. Chinese firms in general exhibit little overall decrease in measures of potential corporate corruption, with the exception of heavily scrutinized entertainment expenditures.

Corporate managers use political connections to funnel resources to their firms, but these distortions often lead to inefficient decisions and underperformance. 1 Nevertheless, China has been rather

anomalous in being the world’s second largest economy and growth engine, while having a corruption

level on par with most developing markets.2 Amidst stated concerns about lower future growth, the new top leadership in China embarked on an anti-corruption campaign to widely reduce political, military, and business corruption in December 2012. The campaign has already caught more than 1.3 million people, including many executives in the corporate world.3 We examine three important

questions related to the corporate portion of the campaign with emphasis on the first two: Is the

campaign actually exposing executives from more corrupt firms? Does the prosecution of executives

contain signs of political favoritism? Is the campaign effective at reducing corporate corruption?

As perhaps the most important event in China today, the anti-corruption campaign has the

potential to bring fundamental changes to China’s corporate world. Although the campaign was

mainly started to target political corruption, it has spread to the corporate world and has launched investigations on a large group of executives from a variety of companies. The campaign provides a unique laboratory to evaluate an effort to reduce corruption because the campaign is both widespread and in a corporate setting with substantial room for improvement. In particular, a large number of studies document pervasive problematic behavior by Chinese firms such as financial fraud [Chen,

Firth, Gao, and Rui (2006)], tunneling [Jiang, Lee, and Yue (2010)], and excessive workplace fatalities due to avoidance of compliance measures [Fisman and Wang (2015)]. Political connections have also been shown to have broad influence on Chinese firms in areas such as bank financing [Li, Meng,

1 See research by Fisman (2001) and Faccio (2006) which document the magnitude of these connections, and the negative externalities are examined by Shleifer and Vishny (1994), Fan, Wong, and Zhang (2007), Aggarwal, Meschke, Wang (2012), Duchin and Sosyura (2012), among others. 2 China’s Corruption Perception Index score of 40 in 2016 (and rank of 79 of 176) was half of the score of most Northern European countries (from 75 to 90). Allen, Qian, and Qian (2005) provide a comprehensive analysis of the Chinese business environment and show it lacks a strong legal system and effective corporate governance mechanisms. 3 Announced by the Communist Party of China’s Central Commission for Discipline Inspection (CCDI) in October 2017.

1 Wang, and Zhou (2008), Chen, Liu, and Su (2013)], earnings management [Fan, Guan, Li, and Yang

(2014)], expropriation of minority shareholders [Cheung, Rau, and Stouraitis (2010), Chen, Li, Su, and

Sun (2011)], and the influence of political connections can be seen in the firms’ stock returns

[Calomiris, Fisman, and Wang (2010) and Liu, Shu, and Wei (2017)].

Given the close connection between political and corporate actors, and that it has been shown that political connection is linked to poor corporate decision-making and self-dealing, a campaign focused on political corruption should prosecute executives from more corrupt firms. There is substantial controversy surrounding the genuineness of the campaign and if it is simply a ruse to remove political opponents [Economist (2014)]. Others argue that the campaign is not a short-term political one as had been used in the past, but instead genuinely focuses on those actually engaged in corruption [Li and McElveen(2014), Magnus (2015)]. While it is difficult to disentangle these possibilities, research is needed to establish working facts regarding the nature of this campaign that is of academic and practical importance to the world’s second-largest economy and many developing countries which look to emulate Chinese policies.

We focus on the corporate aspect of the campaign because of the availability of firm-level data that will allow us to examine characteristics of firms related to corruption as well as information on executives that allow us to examine connections. By carefully searching a broad set of sources including databases of managerial turnovers of Chinese listed firms, disclosures by the Chinese

Communist Party, corporate announcements, and news articles from over 300 Chinese financial newspapers, we construct a sample of 150 Chinese-listed firms where the CEOs or other top executives were investigated during the anti-corruption campaign (from December 2012 to December

2015). 130 (or 87%) are state-owned-enterprises (SOEs). The sample firms account for 5.6 percent of

China’s listed firms in terms of number, and 18.1 percent in terms of market capitalization (RMB 5.73 trillion) as of December 2014. The major listed charges against the executives are receiving bribes (82),

2 illegally benefiting family members (29), embezzling funds (26), bribing others (21), and unspecified

offenses (31).4

We first examine whether the campaign is capturing executives from firms with signs of

distorted managerial incentives, potential self-dealing, and other dubious behavior. Twelve measures

of potential corruption are grouped into five categories: corporate governance and managerial incentives, related-party transactions indicating illegal or unethical behavior, regulation breaches and entertainment expenditures, inefficiencies in operation and investment, and a measure of corruption- related postings on a popular Chinese online investor forum. Sample firms with prosecuted executives generally have higher measures of corruption than their matched firms that have no prosecutions. In probit regressions, we find that (poor) monitoring by minority shareholders, CEO compensation, a near-retirement dummy, related-party sales, related-party loans, operational inefficiency, investment

inefficiency, and corruption postings are all significantly related to the probability of an executive being

investigated for corruption. Interestingly, business entertainment expenditures is not related to

investigation probability, perhaps indicating that the campaign is more complex than the perceived

notion of targeting firms with excessive consumption expenditures.5

These results not only reveal that the campaign captures corrupt firms, but also have important

implications for China’s corporate world. For example, the finding on the near-retirement dummy

confirms the wide-spread anecdote of the “59 Years Old Phenomenon” where China’s strict

retirement policy (at the age of 60) can trigger distorted incentives of managers. The finding on

corruption postings indicates the importance and informativeness of on-line outlets of grass-root

Chinese investors given the dominance of government media in China.

We then turn to an examination of political connections. We first study general political

4 Note that these corrupt behaviors are not mutually exclusive. 5 The Eight-Point Regulation by CPC, which marked the start of the anti-corruption campaign, states that “(Officials) must practice thrift and strictly follow relevant regulations on accommodation and cars.”

3 connections and find an increased investigation probability for firms whose executives have previous

government working experience. We then examine specific connections between firm executives and

politicians who attended the same universities and birthplace connections where companies are

located in the politicians’ home provinces. Using samples of politicians at different levels, we find

evidence that managers’ university or oil-industry affiliations with Zhou, Yongkang, the highest ranked

investigated leader from the previous administration, predict a higher probability of corporate

investigation.6 This result is consistent with a “spillover effect” or political cleansing in political investigations reaching to corporate investigations. Another spillover effect is that managers with connections to local politicians are more likely to be investigated, but not for managers with central government connections. Consistent with a “protection effect” where political connections shelter firms from scrutiny, university connections with the top seven leaders of the 18th Politburo Standing

Committee (PSC) through Peking University and Tsinghua University predict a lower probability of

investigation.7 It is possible that executives from top universities such as Peking and Tsinghua simply

have better job prospects and self-descipline and hence are less likely to cheat, however, we see no

reduced probability of investigation for other top Chinese universities which are not associated with

the PSC members.

Additionally, the relations between the probability of corporate investigation and most of our

corruption measures remain even after controlling for detailed measures of political connectedness.

The corruption measures are also important after excluding the corporate investigations associated

with political investigations that are identified in the news. These results suggest that the influences of

corruption and political factors both exist in the corporate investigations.

Nevertheless, like with most analyses, one cannot rule out the possibility that the campaign is

6 It is widely recognized that oil system is one of Zhou’s power bases (more details in Section 4.2). 7 The Politburo Standing Committee (PSC) of CPC’s Central Committee is the most powerful decision-making body in China. The members of the 18th PSC were elected in November 2012 and their terms ended in October 2017.

4 entirely politically motivated and that the corruption measures correlate highly with the wrong political

camp. This possibility is substantialy reduced since even after controlling for an extensive set of

political measures, both corruption measures and political connections explain similar portions of the

variation in the probability of investigation. For example, in the Probit regressions of corruption

investigations, there is a 14 percentage-point increase in pseudo R2 when adding the corruption

measures and a 19 percentage-point increase for the political variables. The campaign seems to both

have a strong political component, while also capturing corrupt firms that cannot be identified through

their political connections. Our tests are not causal; and like with most empirical exercises, there are

omitted variables which can affect both corruption measures and investigations.

To the extent that the campaign removes bad corporate managers and causes other managers

to stop engaging in dubious activities for fear of prosecution, the campaign could lead to

improvements in governance, efficiency, and transparency in Chinese firms. In contrast, if managers

believe the campaign and in turn investigation is mostly politically motivated, then the campaign may

have little effect on managerial behaviors and measures of self-dealing. We examine whether the

reform affects the culture of the Chinese corporate world by examining the same corruption measures

for all Chinese listed firms over time. Out of our firm-level measures of corruption, the only one that

shows a large and reliable improvement in 2013-2015 compared to 2011 is business entertainment

expenditures, which decreases.8 It is possible that the other measures are poor proxies for questionable firm behavior, but these measures have strong intuitive underpinnings, empirical support from prior literature, and are useful for predicting corruption investigations. In addition to the lack of improvements in corruption measures, we also find little evidence that the anti-corruption campaign improves the disclosure and information environment of Chinese financial markets, or encourages

8 Although practicing thrift is a stated objective of the campaign, our earlier empirical analysis did not show a relation between this measure and investigations perhaps because entertainment expenses are so widespread. Nevertheless, companies seem to perceive that the campaign monitors the conspicuous entertainment expenditures.

5 foreign direct investment in China. In consideration of the possibility that that our measures are

affected by macroeconomic conditions, and given that there is no anti-corruption campaign in Hong

Kong, we match our sample of Chinese firms to a sample of Hong Kong firms and conduct a

difference-in-difference analysis, and again find little improvement for Chinese firms.

Overall, other than a drop in entertainment expenditures, we find little evidence that there has

been a broader decrease in corporate corruption indicators. Our findings provide an important

observation as to the debate about the relative effectiveness of fighting corruption through centralized

government initiatives [Olken (2007)] or broader and more micro-level changes such as corporate

governance, legal reform [Wei (2001), Svensson (2005), Magnus (2015)], and news-based and

grassroots campaigns [Reinikka and Svensson (2004), Bjorkman and Svennson (2009)]. In bank

lending, Beck, Demirguc-Kunt, and Levine (2006), Barth, Lin, Lin, and Song (2009), and Houston,

Lin, and Ma (2011) find that official supervisory agencies do not reduce corruption in bank lending,

but more micro-level approaches such as information sharing and transparency, increasing bank

competition, and private media can more effectively promote lending integrity.

It is both interesting and puzzling that a campaign ensaring many executives from firms with

weaker governance, more self-dealing, and more inefficiencies is not leading to any broader

improvements in these exact same measures. This contrast could be due to the perception of Chinese

executives about the anti-corruption campaign. Self-dealing is widespread and lucrative in China’s

corporate world, and if the corrupt executives believe that the campaign has a strong political

component or the campaign will be short-term like the previous ones, then they might focus more on

the shoring up political connections and continue in rent-extraction in all but the most conspicuous

ways.9 We hope to see research further examine these issues.

9 This interpretation might be consistent with luxury liquor prices. Moutai, China’s national liquor served at luxury banquets, price plummeted after the start of campaign, but gradually recovered, especially in 2017 with the expectation that the price will rise when the campaign ends. (http://www.scmp.com/news/china/policies- politics/article/2095446/targeted-crackdown-moutai-liquor-once-again-toast-town)

6

1. The Anti-Corruption Campaign Background

Corruption is generally thought to be a problem in the political, military, and business spheres in

China.10 Pei (2007) estimates the direct costs of corruption at three percent of Chinese GDP per year.

Yet, due to secrecy, excessive focus on rent-seeking, and distortionary incentives, the indirect costs are likely much greater than the direct costs [Shleifer and Vishny (1993)]. Corruption is a larger impediment for small and medium-sized firms [Beck, Demirguc-Kunt, and Maksimovic (2005)], and

also stifles foreign direct investment [Wei (2000)]. Laws in the U.S. and U.K. seek to curtail bribery abroad, but may disproportionately disadvantage global firms.11 Pointing to successful cases of anti-

corruption reform in Hong Kong and Singapore, developing economies often launch anti-corruption

campaigns.

Soon after Xi Jinping took China’s leadership on November 15, 2012, he emphasized the

determination to crack down on corruption, targeting both “tigers and flies”. On December 4, 2012,

the Communist Party of China (CPC) announced the “Eight-Point Regulation” which provides clear guidance for the party and government officials to eliminate corruption.12 Xi’s leadership and the

issuance of the “Eight-Point Regulation” are generally regarded as the start of the anti-corruption

campaign in China.

Although previous Chinese leaders had repeatedly criticized the severe corruption problems

and made anti-corruption efforts, top government officials were rarely touched. In contrast, Xi, the

10 The knabbing of top U.S. firms for bribery in China by U.S. regulators also provides glimpses into the extent of the bribery problem using U.S. standards. For example, Avon, the cosmetics company, admitted guilt and paid $135 million in 2014 to settle U.S. Department of Justice charges for bribing Chinese government officials (http://fortune.com/2014/12/17/avon-bribery-probe-settlement/). J.P. Morgan paid $264 million in 2016 to settle a probe by U.S. officials into its practice of improperly influencing top Chinese officials by hiring their children. (https://www.ft.com/content/fc32b64e-ac87-11e6-ba7d-76378e4fef24). 11 For example, Karpoff, Lee, and Martin (2017) document that bribe-paying firms caught under the U.S. Foreign Corrupt Practices Act of 1977 experience large loss. Zeume (2016) finds that the passage of the U.K. Bribery Act decreases U.K. firms’ value and investment. 12 The content of the Eight-Point Regulation is presented in Section A.1 of the Internet Appendix.

7 son of a top communist veteran, put anti-corruption at the center of his platform. The intensive and

extensive anti-corruption campaign has been “more prolonged and far-reaching than anyone

anticipated” [The Guardian (2015)]. The campaign has punished over 1.3 million people, including six

national leaders and hundreds of high-ranking government officials and military officers. The

campaign has also investigated corrupt managers in China’s corporate world, such as Lin Song, former

Chairman of the state-owned enterprise (SOE) China Resources and one of Fortune’s “50 Most

Influential Business Leaders”.

Nevertheless, corruption campaigns are typically conducted by less transparent legal and

financial institutions, and are often thought to mask the cleansing of political oppositions [Bardhan

(1997), Svensson (2005)]. 13 Additionally, one wonders if the anti-corruption campaign is driving

positive changes in the Chinese corporate world and corporate culture. Academic research is needed

to further understand these important questions.

The Chinese anti-corruption campaign has also been examined from other angles. Lin, Morck,

Yeung, and Zhao (2017) find that the overall market reaction is quite positive to the announcement

of the campaign, especially for non-SOEs located in developed provinces or with lower business and

entertainment expenditures. Qian and Wen (2015) find that jewelry imports decreased dramatically after the anti-corruption campaign, which is consistent with our firm-level findings of decreases in entertainment expenditures. 14 Ang, Bai, and Zhou (2016) find that corruption as measured by

investigations of government officials in a province is reflected in municipal bond spreads.

13 For example, Bardhan (1997) states that “…as happened many times in the recent history of Africa or China, anti- corruption campaigns are usually ad hoc, and targeted at political enemies or at best at small fry, exempting the big fish, or the important cronies and accomplices of the political rulers.” 14 Qian and Wen (2015) also document that, in contrast to the dramatic drop in import of the luxury goods that are easily observed by the public, there is no effect on goods that can be consumed away from public view. Ke, Liu, and Tang (2017) document that the anti-corruption campaign reduced the excessive consumption of luxury goods but had little impact on the growth or value of the luxury goods and services consuming firms.

8 2. Data, Summary Statistics, and Measures of Potential Corruption

2.1 Sample selection

There are two parts of analysis and sample construction. First, we examine a sample of firms

with corrupt managers investigated during the anti-corruption campaign (henceforth “sample firms”

or “event firms”). Second, we examine all Chinese listed firms to study the impact of the anti-

corruption campaign on the Chinese corporate world.

As of December 2015, there are 2,909 listed stocks in Shanghai and Shenzhen stock exchange,

with a total free-float market capitalization of RMB 41.8 trillion (USD 6.4 trillion).15 The sample firms

include the listed firms in China whose top managers were investigated during the anti-corruption

campaign for corrupt behavior. We require a sample firm to be listed on either the Shanghai Stock

Exchange or the Shenzhen Stock Exchange, and for either the CEO or other top executives to have

been investigated for corrupt behavior during the campaign from December 4, 2012 to December 31,

2015.16

We identify investigated firms using three approaches. First, we obtain information of CEOs

of listed companies from the China Stock Market & Accounting Research (CSMAR) database, and

identify a total of 2,862 CEO turnovers during December 4, 2012 to December 31, 2015. We manually

search these CEOs’ internet biographies and the news for the reasons of each turnover, and identify

the events involving corrupt behavior.

Next, to collect corruption cases of non-CEO top executives, we obtain the list publicized by

the Commission of Discipline Inspection of CPC’s Central Committee which discloses investigated

high-level party members, including executives of large SOEs.17 We manually read through the list and

15 In December 2005, there were 1,464 listed companies in Shanghai and Shenzhen stock exchange, with a total free-float market capitalization of RMB 1.1 trillion (USD 131.7 billion). 16 Eligible titles include CEO, Chairman of the board, director on the board, firm controller, Vice President, and CEO/Chairman or Vice President of the parent company. See Table 1 for distribution of manager types. The anti- corruption campaign is still continuing, and December 31, 2015 is when our sample construction ends. 17 The disclosures can be found at http://www.ccdi.gov.cn/jlsc/.

9 identify the investigations of managers of listed firms.

Third, we conduct key word searches on two large bodies of publications: a) more than

800,000 corporate announcements for all listed companies in our sample period from the CNINFO

dataset; and b) news articles from Genius Finance, a widely-used database covering news articles from

over 300 Chinese financial newspapers. We manually read through the corporate announcements and

news articles about corruption cases from the first two sources (CEO turnovers and CPC disclosure),

and compose a list of 34 corruption-related keywords that are commonly used to describe corrupt

behavior and investigations (provided in Section A.2 of the Internet Appendix).18 Due to the large

number of news articles, we first obtain the list of 35,353 director turnovers during our sample period

from CSMAR, and narrow down the sample to the 40,000 news articles that mention names of at least

one of these directors. We then search corporate announcements and these news articles and identify

1,049 corporate announcements and 2,236 news articles containing the key words. We manually read them through to identify an additional sample of corruption cases.

We further investigate the details of managers’ corrupt behavior from the above three sources, and use a conservative approach to exclude a small number of events where: 1) the manager’s corrupt behavior took place before joining the company; 2) the manager’s corrupt behavior is unrelated to the firm.19 3) the manager is found clean after the investigation (one event); and 4) the firms experience reverse merger or major asset restructuring within one year prior to the corruption investigation, in which case the top manager might not have full control of the firm (two events).

When a firm experiences several investigations involving multiple executives, we keep only one event per firm by choosing the most important manager or the earliest event if the investigated managers are of similar importance. For each event, we carefully go through announcements and news

18 Examples of keywords (in Chinese) include “discipline violation”, “under corruption investigation”, “suspicion of bribery”. 19 For example, a vice president of a listed firm represented a blockholder and his corrupt behavior occurred in the blockholder firm instead of the listed firm.

10 articles, and identify the event date as the earliest day when the news of investigation becomes

available. We also include in our sample the cases where the parent company’s top managers engage

in corrupt behavior, as Chinese parent companies have very tight control of their subsidiaries, either

by directly managing them or through influencing their major decisions.

Our final sample includes 150 listed companies whose managers were investigated and

dismissed during the anti-corruption campaign for corrupt behavior. The size of this sample as of

December 31, 2014 is 5.73 trillion RMB (USD 936 billion) in market capitalization. They account for

5.6% of China’s listed firms, and 18.1% of market capitalization.

2.2 Summary statistics

Panel A of Figure 1 plots the distribution of sample firms by year, where firms are divided into

SOEs and non-SOEs. A firm is classified as SOE if its controlling shareholder is affiliated with the

Chinese government or its largest shareholder is affiliated with the Chinese government and holds at

least 25% of the firm’s outstanding shares. The data on SOE status is directly obtained from the

CSMAR database, and we manually check and correct misclassifications. Panel A of Table 1 shows

that the anti-corruption campaign has accelerated since its start in December 2012, as the number of

firms investigated increased from just one in 2012 (December) to 28 in 2013, 50 in 2014, and 71 in

2015. Additionally, 86.7% of the investigated firms are SOEs, which is consistent with SOE managers having greater conflicts of interest and resources compared to non-SOEs and reflects the fact that

SOEs are direct targets of the campaign.20 Panel B of Figure 1 plots the positions of corrupt managers

by year, and Panel A of Table 1 reports the numbers. Out of the 150 sample firms, 67 have corrupt

CEOs, 25 have corrupt non-CEO executives, and 58 have corrupt top managers (CEOs or Vice

Chairmans) from the parent company.

20 For example, the Eight-Point Regulations applies to SOEs but not non-SOEs. Our sample composition is in line with Ke, Liu, and Tang (2017) who construct a sample of 62 investigated firms during the campaign and find that 92% of investigated firms are SOEs.

11 Panel C of Figure 1 plots the distribution of corrupt behavior (not mutually exclusive) for

SOEs and non-SOEs and Panel B of Table 1 presents the numbers. There is a stark contrast between

SOEs and non-SOEs in terms of corrupt behavior. Specifically, out of the 130 SOEs, the most common corrupt behaviors are receiving bribes (82), embezzling company funds (25), and illegally

benefiting family members or relatives (29). Managers of four SOEs, all financial firms, bribe

government officials or other parties to obtain licenses or compete for underwriting business. The

remaining 29 SOEs do not have detailed information about corrupt behavior by managers.21 For the

20 non-SOEs in our sample, we are able to identify specific corrupt behaviors for 18 of them, where

17 non-SOE managers bribe other parties, and the other firm’s manager is investigated for embezzling company funds. These results shed light on the vastly different incentives and forms of corruption across ownership structures.

2.3 Measures of potential corruption

Motivated by the existing literature, we examine a broad set of measures that capture five aspects of corruption: corporate governance and managerial incentives, related-party transactions, regulation breaches and entertainment expenditures, inefficiencies, and corruption-related postings from a popular online investor forum. The definitions of our corruption proxies are briefly listed below with more details in Section A.3 of the Internet Appendix.

2.3.1 Corporate governance and managerial incentives

Poor corporate governance or distorted managerial incentives can lead to agency problems

[Jensen and Meckling (1976)], and this relation has been documented in China [as reviewed Jiang and

Kim (2015), Fan, Wei, and Xu (2011)]. We examine four measures.

Our first measure is monitoring by minority shareholders. This is proposed by Chen, Chen,

21 For these firms the sources, such as corporate announcements or news articles, simply mention “involvement in financial issues”, etc. All of these managers are non-CEO executives where media coverage is relatively sparse compared to CEOs.

12 Schipper, Xu, and Xue (2012) who argue that because China is characterized by weak shareholder

protection and controlling shareholders that exploit other shareholders, high and concentrated

ownerships of minority shareholders is an important monitoring mechanism. We follow their

approach and construct the measure as the total ownership of the second to fifth largest shareholders,

multiplied by the Herfindahl index of their ownerships.22

Our second measure is CEO compensation. Previous studies, such as Borokhovich, Brunarski,

and Parrino (1997) and Coles, Daniel, and Naveen (2014), have shown that higher CEO compensation

indicates poor corporate governance and entrenched CEOs. High CEO compensation could be the outcome of extraction of money from the firm by the CEO, and the Chinese government seems to believe this is the case as they started capping the salary of SOE managers in 2015. We construct the measure as the sum of salary, bonus, value of granted restricted stocks, options, and appreciation rights, scaled by total assets.

Our third measure is CEO pay-for-performance sensitivity, a commonly used measure of managerial incentive. We follow the literature [Core and Guay (1999), Bergstresser and Philippon

(2006)] to construct this measure using CEO’s holdings of the company’s stocks and options. Our fourth measure is a near-retirement dummy which equals one if CEO’s age is greater than or equal to

59, and zero otherwise. China has a strict retirement policy that require government officials, including employees of SOEs, to retire at the age of 60.23 There have been many news articles about distorted

incentives and corrupt behavior for government officials and SOE managers approaching retirement,

a phenomenon generally referred to as the “59 Years Old Phenomenon”.24

22 Our findings hold when we do not multiply share ownership by the Herfindahl index. 23 Retirement age varies among the top government officials, for example, 65 for ministry-level government officials, and generally 70 for the top national leaders. 24 For example, the 2014 China Legal Development Yearbook issued by the Chinese Academy of Social Science shows that the “59 Years Old Phenomenon” is pronounced in China’s corruption cases (http://www.chinanews.com/fz/2014/02-24/5873850.shtml). We also repeat the analysis using 57 or 58 to construct the dummy variable and the results are similar. If the investigated manager for an event firm is not the CEO but another top manager, we use the investigated manager’s age rather than CEO age to construct the dummy.

13 2.3.2 Related party transactions

The existing literature suggests that related-party transactions can be associated with unethical

or illegal behavior by Chinese firms. We therefore examine three measures based on the existing

literature on related party transactions: 1) Related-party sales (scaled by revenue), as Jian and Wong

(2010) find that Chinese firms use related party sales to prop up earnings to meet the exchanges’ listing

requirements for financial performance. 2) Related-party loans (scaled by total assets), as Jiang, Lee,

and Yue (2010) reveal “tunneling” behavior in Chinese firms, where controlling shareholders take

advantage of the firm and other shareholders through large amounts of borrowing from the company

at very low or no cost. 3) Other receivables from parent company (scaled by total assets), as Jiang,

Lee, and Yue (2010) suggest that this also reflects tunneling behavior through trade credit.

2.3.3 Regulation breaches and business entertainment expenditures

The third group of measures includes regulation breaches and entertainment expenditures,

both potentially associated with the degree of unethical or illegal behaviors in Chinese companies. We

obtain the number of regulation breaches of all Chinese listed firms from CSMAR’s Enforcement

Actions Research Database, and aggregate by firm-year. In counting the number of breaches, we

exclude those described as “non-material accounting errors”, because they are associated with

common accounting mistakes that are unlikely to be associated with corruption (a complete list of

regulation breach categories provided in Section A.3 of the Internet Appendix).25 The second measure, business entertainment expenditures (BEE), is widely considered by news media and academic studies to be associated with corruption because it is often used by firms as perks to employees and especially executives, or to establish relations with other parties to gain business [e.g., Cai, Fang, and Xu (2011)].

2.3.4 Inefficiencies

25 To avoid any look-ahead bias, we define the year of a regulation breach as the year when the breach is disclosed. On average there is a time lapse of 0.7 year between disclosure and the last incident of a regulation breach.

14 Corrupt behavior by top managers, especially embezzling funds and receiving bribes, can

increase the firm’s costs or misallocate the firm’s resources. We therefore examine two measures of

inefficiency. Our first measure is operational inefficiency, calculated as the difference between sales growth and net income growth. News articles about investigated firms often mention that corruption

caused these firms to have a much slower growth of profit than their growth of revenue. A higher

value of operational inefficiency indicates increased costs for the firm and more potential corruption.

Our second measure is investment inefficiency proposed by Biddle, Hilary, and Verdi (2009).

While several measures of investment efficiency have been proposed at the country or industry level,

this measure is at the firm level and based on the neo-classical framework where the marginal Q ratio should be the only determinant of capital investment. The departure from this optimal level is

considered abnormal, possibly indicative of friction-induced inefficient investment. News articles

about our event firms also mention abnormal investment behavior associated with instances of

corruption. We follow Biddle, Hilary, and Verdi (2009) and estimate cross-sectional firm-level

regressions of corporate investment on sales growth within industry-years, and calculate the measure

as the absolute value of the residual.

2.3.5 Corruption-related postings

We collect corruption-related postings from “GuBa” (“StockBar” in English, http://guba.eastmoney.com/), one of the most popular online investment forums in China.

According to Alexa Internet, a subsidiary of Amazon, the number of new posts per day on Guba is as

high as six million. We download all the posts discussing listed companies in GuBar and identify a

post as corruption-related if its title contains a corruption-related keyword (as described in Section

A.3 of the Internet Appendix). We then measure corruption postings for a firm-year as the number

of corruption-related posts scaled by the total number of posts for the firm-year.

15 3. Is the Corporate Anti-Corruption Campaign Capturing More Corrupt Firms?

We first examine if the campaign is capturing more corrupt firm. If the campaign is a sincere

effort to reduce corruption, then we expect that event firms will have a greater degree of corruption

than peer firms.

3.1 Univariate analysis

For each event firm, we identify a matched firm by first selecting a subsample of firms that

are in the same industry, have the same SOE status as the event firm, and have market capitalization within the range of 50% and 150% of the event firm. We then choose from this subgroup a matched firm that has the closest book-to-market ratio to the event firm.26

Figure 2 plots firm-level corruption measures for event firms and matched firms in years t-2,

t-1, and t, where t is the year of event (announcement of corruption investigation). Since we examine

drivers of investigations, we focus on the corruption measures in years t-1 and t-2 but also report

those in year t which is partially before investigation. The numbers and statistical tests behind the

differences of even and matched firms in Figure 2 is shown in Table 2. The relatively small sample

size allows for the hand-collection of many of the political measures, but may make it difficult to

observe statistical significance due to a potential lack of power. Despite these limitations, many of the

measures are significant. Eleven of the twelve corruption measures indicate greater degree of

corruption in event firms than in matched firms in year t-1, and seven of them are statistically

significant. Specifically, Table 2 shows that in years t-1, when compared to matched firms, event firms

have statistically significantly less monitoring by minority shareholders, higher CEO compensation,

lower pay-for-performance sensitivity, more near-retirement managers, higher related-party sales, and

26 Firms are divided into 19 industries using the first-digit of China Securities Regulatory Commission’s 2012 industry classification codes. For five event firms there are no other firms in the same industry with close enough market capitalization. We therefore identify their matched firms as those having the closest market capitalization among firms in the same industry and with the same SOE status. We repeat the event firm analyses by excluding these five event firms, and the results are unchanged (reported in Section A.4 of the Internet Appendix).

16 more operational inefficiency.27 Additionally, corruption postings are significantly higher for event

firms, indicating that investors more often discuss corruption issues about event firms than peer firms.

Among the remaining measures in year t-1, related party loans, other receivables, and investment

inefficiency are in the correct direction but marginally insignificant.

Regulatory breaches and business entertainment expenditures show no noticeable differences

between investigated and matched firms in year t-1. The fact that business entertainment expenditures

is no higher for event firms is puzzling because the common perception is that entertainment

expenditures is a main focus of the campaign. It is possible that entertainment expenditures is so

pervasive in the Chinese corporate world that it is a noisy signal of differential corruption across firms.

This result, together with our finding that other measures predict corruption investigation, could

indicate that the campaign is targeting various aspects of corruption beyond entertainment

expenditure. Event firms’ regulation breaches is higher in year t than matched firms but the difference

is marginally insignificant. Overall, the results in Figure 2 and Table 2 show that event firms are more

corrupt than peer firms along many of the corruption measures.

3.2 Probit regressions of investigation on corruption measures

We estimate probit regressions to jointly test for a relationship between our various corruption

measures and the occurrence of an investigation after controlling for various firm-level characteristics.

The sample includes event firms and their matched firms. The dependent variable is a binary variable

that equals one if the firm is investigated for corruption, and zero otherwise. The independent

variables are corruption measures of the year prior to corruption investigation (year t-1). We further

control for firm size, SOE status, year fixed effects, industry fixed effects, and region fixed effects.28

27 CEO compensation is significant at the 0.10 level and all other variables at the 0.05 percent level. 28 China is divided into seven regions including North, Northeast, East, Central, South, Southwest, and Northwest. For robustness, we repeat the regression analyses using alternative fixed effects such as year fixed effects and industry-region fixed effects, or industry fixed effects and year-region fixed effects. These results are in Section A.5 of the Internet Appendix.

17 Models (1) to (5) in Table 3 present regressions on the five groups of corruption measures,

respectively. Consistent with the univariate analysis, the signs of coefficients of most corruption

measures indicate that the degree of corruption positively predicts corruption investigation. The only

exception is business entertainment expenditures (BEE) in model (3), where the coefficient is negative

rather than positive, but the t-statistic is only -0.21.29 Despite the small sample, the coefficients of

monitoring by minority shareholders, CEO compensation, CEO near-retirement dummy, related-

party sales, related-party loans, operational inefficiency, investment inefficiency, and corruption

postings are statistically significant.30

Model (6) includes all corruption measures, and the sample size is reduced by over one-third

because of missing BEE. The results are similar as previous models, except the CEO compensation,

investment inefficiency, and corruption postings become insignificant with the signs of coefficients

unchanged. However, other receivables from parents becomes significant. To address the concern of

reduced sample size, we repeat the regression in Model (7) without including BEE, and these measures

become significant with the exception of investment inefficiency which is now marginally insignificant

(t-stat 1.63). Regarding economic significance, other receivables from parents has the largest impact, where one standard deviation increase is associated with a 11.0 percentage points increase in the probability of investigation. The marginal effects for a one standard-deviation increase in the other measures are also large, such as the near-retirement dummy (9.3 percentage points), monitoring by minority shareholders (8.8 percentage points), corruption postings (7.7 percentage points), operational inefficiency (7.4 percentage points), related-party loans (6.8 percentage points), related party sales (6.4

29 Chinese firms disclose business entertainment expenditures on a voluntary basis, which could introduce noise as high BEE firms might choose not to disclose. For robustness, we repeat the regression analysis using an alternative measure of BEE that includes travel expenses. The coefficient on this alternative BEE measure turns positive but remains insignificant, with t-stat equal 0.66 in the full model. For brevity this result is presented in Section A.6 of the Internet Appendix (Table IA.10). 30 For CEO compensation, we also examine the alternative measures of CEO compensation scaled by operating profit, CEO compensation of matched non-SOEs, or employee compensation. Table IA.11 of the Internet Appendix report Probit regressions of investigation on these alternative measures, and their coefficients remain significantly positive.

18 percentage points), and CEO compensation (6.2 percentage points).

Since the anti-corruption campaign directly targets corrupt behaviors of members of the

Chinese Communist Party and most SOE managers are Party members, we conduct the regression

analysis only for SOE firms. Table 4 presents regression results corresponding to Table 3 but using

only SOE firms. The results are similar to those in Table 3 in terms of both magnitude and statistical

significance. Interestingly, entertainment expenditures is again insignificant like in the full sample.

Overall, our results indicate that investigated firms are generally more corrupt along many

dimensions than their peers. We will further discuss possible interpretations of this finding below after

examining the political dimension.

4. Are political factors related to corporate investigations?

Political factors could influence corporate investigations through two main channels.

Connections with non-investigated politicians can serve as protection to the firms and decrease the

probability of investigation (a “protection effect”). However, political connections can increase the

probability of corporate investigation (a “spillover effect”) where a political connection draws a firm

into an investigation. The investigations that occur due to political connections could be be because

the firm is caught in the middle of a campaign to weaken or eliminate a political enemy, or because

the campaign has caught a corrupt politician associated with a firm. We examine 14 measures of political connectedness that provide a thorough examination of how political factors impact the corporate campaign, but nevertheless will not capture all political influences. We discuss possible interpretations of our findings at the end of this section.

4.1 General government connection

We first examine a measure of general government connection proposed by the existing

19 literature [e.g., Fan, Wong, and Zhang (2007), Fisman and Wang (2015)]. The measure is a dummy

variable equal to one if a C-Suite executive of the company previously served as a high-ranking official

within the government.31 We find that 43.3 percent of the event firms have general government

connections, over twice that of matched firms (18.7 percent). More formally, we estimate similar

probit regressions as in Table 3 but with government connection measures. In Model (1) of Table 5,

the coefficient of government connection is significantly positive, suggesting that, consistent with the

spillover effect, firms with closer relations to government are more likely to be investigated.

We further examine whether the managers’ government working experience was with the

central or local government. Central government connections could provide protection for firms

because the anti-corruption campaign was launched and has been carried out largely by the central

government (CPC central leadership). In Model (2) of Table 5, the coefficient of local government

connection remains significantly positive but that of central government connection is insignificant,

possibly indicating that the protection effect potentially counteracting the spillover effect at the

national level.

4.2 Specific connections with non-investigated and investigated officials

To more directly analyze the spillover and protection effects, we separately examine

connections with non-investigated officials and those with investigated officials, using samples of the

very top leaders, national leaders, and general government officials.

We first look at loyalties using school-ties [Cohen, Frazzini, and Malloy (2010)] with the

universities attended by the seven members of the 18th Politburo Standing Committee (PSC) of CPC’s

31 Chinese firms have different organization structure from U.S. firms, and we define “C-Suite executives” to be CEO, CFO, board chairman, and vice-chairman. If the managers investigated are top managers of parent company, we construct the political connection measures using the managers of the parent firm. The Chinese political system has five ranks: 1) Nation (“Guo Ji”, e.g., President, Vice President, Premier, Vice Premier); 2) Province/Ministry (“Sheng/Bu Ji”, e.g., Provincial Governor, Deputy Provincial Governor, Minister, Deputy Minister); 3) Prefecture/City (“Ting/Ju Ji”, e.g., Mayor and Deputy Mayor of prefecture-level cities); 4) County (“Xian/Chu Ji” e.g., county’s chiefs); 5) Township (“Xiang/Zhen Ji”, e.g., Town Mayor). We follow Fisman and Wang (2015) and identify executives that held a Prefecture/City or higher position before joining the company.

20 Central Committee, the most powerful decision-making body in China. The members of the 18th PSC

were elected in November 2012 and their terms ended in October 2017. We identify cases where a C-

Suite executive of the company graduated from the same university as a PSC member. For both event

firms and matched firms, we manually read their managers’ biographies (most from CSMAR and some

from online searches) and collect the university data.32 The universities attended by the top seven

leaders include the two most prestigious universities in China, Tsinghua University (school of Xi,

Jinping) and Peking University (PKU, school of the Prime Minister Li, Keqiang), as well as Xiamen

University.

For a comparison, we also look at school ties with Zhou, Yongkang, the highest-ranked

politician investigated during the campaign. Zhou, Yongkang was once the country’s third most powerful leader and a member of the 17th PSC. According to Chinese officials, Zhou plotted to usurp

the party’s leadership and seize state power.33 In 2013 the Party began an investigation into Zhou and

arrested a number of Zhou’s high-ranked former suboridnates. After a corruption probe Zhou

received lifetime incarceration. He attended China Petroleum University and had strong ties with the

oil industry.

Figure 3 plots the school ties for investigated firms and matched firms. Each line connecting

firms with a university represents a C-suite executive from one of the firms that graduated from the

university. Red (blue) lines are for investigated (non-investigated) firms.34 Interestingly, all the firms

32 We also considered a school-tie measure capturing whether leaders and managers overlap at a university. However, possibly because of mandatory retirement requirements, there is little intersection between firm managers and national leaders at universities. Managers of SOEs are required to retire at 60, whereas the average age for PSC members is 67. 33 See the news article from South China Morning Post (http://www.scmp.com/news/china/policies- politics/article/2116176/coup-plotters-foiled-xi-jinping-fended-threat-save). The report also says that Wang Qishan, China’s top graft-buster, said that the central authorities managed to punish Zhou so as to eradicate a number of conspirators and ambitious schemers within the party. For another example, see the news article from Los Angeles Times (http://articles.latimes.com/2013/dec/16/world/la-fg-wn-china-corruption-zhou-yongkang-20131216). 34 The figure does not include three universities attended by the top seven leaders but not corporate managers (Kim Il- sung University, Harbin Engineering University, and Northwest University). The university affiliation analysis excludes the Party School of the Central Committee of CPC, attended by a current PSC member. Rather than a normal school, the Party School is a unit of the Central Committee of CPC training (high-rank) officials. For robustness, we repeat the analysis of university affiliation including the Party School and the results remain similar (Table IA.12 of the Internet Appendix).

21 with executives ties to China Petroleum University are investigated. Firms with executive ties to

Tsinghua or Peking University appear less likely to be investigated.

To examine this relation more formally, Model (3) of Table 5 presents the probit regression of investigation on university affiliation, which is measured for a company as the number of connections where a C-Suite executive of the company graduated from the same university as a PSC member. We also examine birthplace connections with the top seven leaders using a dummy variable that equals 1 if the company’s headquarter is located in the home province of a PSC member, and 0 otherwise. The coefficient of university affiliation is significantly negative (t-stat -4.14) and that of birthplace connection is small and insignificant. These results are consistent with Figure 3 that university affiliation with the existing leadership is negatively related to corporate investigation, consistent with the protection effect.

While the result on university affiliation is consistent with protection effect, an alternative explanation is that managers who graduated from the schools, especially Tsinghua and Peking could be less corrupt than their peers. Tsinghua and Peking are the top two universities in China, so their graduates could be more self-disciplined or operate with greater career potentials in mind. To examine this alternative explanation, we repeat the regression analysis using university affiliations with two alternative pairs of universities: The top 3rd and 4th universities (University of Science and Technology of China and Fudan University), and the top 5th and 6th universities (Shanghai Jiao Tong University and Zhejiang University).35 While these universities are also prestigious in China, they are not attended by any of the top 7 leaders like Tsinghua and Peking. The results in Table IA.13 shows that in the regressions of corporate investigations, the coefficients of both alternative university affiliation measures are insignificant while the coefficient of affiliation with Tsinghua and Peking is significantly positive. These results suggest that our finding of university affiliation is driven by protection effect

35 We use the U.S. News & World Report ranking of Best Global Universities in China.

22 rather than self-discipline.

Model (4) of Table 5 presents the regression including a dummy variable which equals one if

the firm is connected to the investigated former leader Zhou, Yongkang. We classify firms located in

Sichuan province or in the oil industry as connected to Zhou because it is widely recognized that

Zhou’s power bases are “Sichuan Gang” and “Oil Gang”.36 The result shows that a connection with

Zhou is significantly positively related to the probability of investigation, which is consistent with the

spillover effect.

Next, we extend the analysis to 112 national leaders identified using the official website of the

government of China (described in Section A.7 of the Internet Appendix), among them six are

investigated during the campaign and the other 106 are not investigated. Model (5) of Table 5 presents

a regression using the measures of university affiliation and birthplace connection with both

investigated and non-investigated national leaders.37 University affiliation with investigated leaders is

significantly positively related to the probability of investigation (t-stat 2.07), while that with non-

investigated leaders is significantly negatively related to investigation (t-stat -2.64). Birthplace

connection with investigated leaders is insignificantly positive (t-stat 1.14) while that with non-

investigated leaders is significantly negative (t-stat -2.01). Furthermore, we obtain a list of 1,021 general

government officials investigated from December 2012 to December 2015 published by the CPC’s

Central Discipline Committee. We focus on birthplace connection because university information is

missing for the majority of these politicians. Regression results using this measure are shown in Model

(6), and the coefficient of the birthplace connection with these politicians is insignificant (t-stat -1.04).

Overall, the results on connections with investigated and non-investigated politicians provide

evidence of both spillover and protection effects on corporate investigations, which we will further

36 The figure https://en.wikipedia.org/wiki/Zhou_Yongkang#/media/File:Relative_map_of_Zhou_Yongkang.png illustrates the four circles of Zhou, including family members, public security service, Sichuan circle, and oil system. 37 When we construct birthplace connections for national leaders and later the 1,021 investigated officials, we construct the measure as the number of politicians having the same home province as the company’s headquarters.

23 discuss at the end of this section.

4.3 Provincial political investigations

We further examine whether political investigations in a province spills over to local

corporations. We construct two measures of political investigation at the province level following Ang,

Bai, and Zhou (2016). “Graft-Tigers” is the average rank of officials investigated in a province, which

captures intensity of political investigation. “Graft-Flies” is the total number of officials investigated

in a province, which captures the scope of political investigation. Ang, Bai, and Zhou find that yields

of China’s municipal bonds increase as a result of local political investigation as captured by the

“Graft-Tigers” measure. For our analysis, we construct the measures “Province Graft-Tigers” and

“Province Graft-Flies” similarly as Ang, Bai, and Zhou but instead use investigated provincial

politicians in the six months prior to a corporate investigation. Because provinces with more fractures

and counties tend to have a larger official base, we scale “Province Graft-Flies” by the number of

counties in the province. These are included in Model (7) of Table 5, where the coefficient for

“Province Graft-Tigers” is positive and significant at the 0.10 level (t-stat 1.80), and that of “Province

Graft-Flies” is insignificant (t-stat -0.86). This result provides some mixed evidence that intensity of

local political investigations positively predicts corporate investigation.

We also construct a variable to measure the timing of local political investigations. The

measure is a dummy equal to one if an investigation team from CPC’s central leading group for

inspection work carried out the investigation in the investigated firm’s province in the six-month

period prior to the investigation.38 This measure is included in Model (8), where the coefficient of the

investigation team dummy is significantly positive (t-stat 1.99), consistent with local political

investigations having a positive impact on corporate investigations.

38 The central leading group for inspection work is headed by Wang Qishan, the top seven leader in charge of the anti- corruption campaign. Each year during the campaign, the Group conducts two to three rounds of inspection. Each round takes about three months, and about ten teams are sent out to provinces, major SOEs, and government units.

24 We include all the political measures in Model (9) of Table 5, where five of them remain

significant. The sample size is smaller because the two provincial measures require the previous six-

month data and exclude investigation events in early 2013. Model (10) repeats the regression without

including the two provincial measures, where the results remain similar but t-statistics of the

coefficients generally increase. In particular, local government connections and affiliations with Zhou,

Yongkang increase the probability of investigations, whereas university affiliations with the top seven

leaders and birthplace connections with non-investigated national leaders decrease the probability of

investigation. Regarding economic significance, university affiliation with top seven leaders has the

largest impact, where one standard deviation increase is associated with a 19.9 percentage points

decrease in the probability of investigation. The marginal effects for a one standard deviation increase

in the other measures are also large, such as the association with Zhou Yongkang (14.8 percentage points), local government connection (9.4 percentage points), birthplace connection with non-

investigated national leaders (-9.6 percentage points) or with investigated national leaders (6.0

percentage points), province graft-tigers (6.0 percentage points), and investigation team in the

province (4.5 percentage points).

For robustness, we also repeat the regression analysis for SOEs. The results in Table 6 are

similar to those using the full sample in Table 5. The findings suggest that political factors seem to

strongly influence the corporate anti-corruption campaign and evidence supports both a protection

and a spillover effect.

4.4 Are corporate investigations entirely driven by political investigations?

Because we find evidence that political factors impact corporate investigations, a natural

question is whether the corporate anti-corruption campaign is entirely driven by political factors. Table

3 includes different sets of political measures in regressions, but surprisingly, the coefficients of the

corruption measures remain similar both in magnitude and in statistical significance for all model

25 variations. For example, in Model (10) of Table 5 where all political measures are included, most of the eight corruption measures which are significant in the other models of Table 5 remain significant with the only exception being related-party sales.

We further examine pseudo R2s of the probit models to compare the explanatory power of

our corruption measures and political factors. The R2 is 3.8% with only control variables, 17.8% with

corruption measures and controls, 22.8% with political measures and controls, and 39.2% with

corruption measures, political measures, and controls. There is a 14 percentage-point increase in R2 for the corruption measures beyond the control variables, and a 19 percentage-point increase for the political variables. These results indicate that both corruption measures and political factors explain a sizeable portion of the cross-sectional variation in occurrence investigations and that each capture aspects of corruption and political influence.

To further investigate possible political motives of corporate investigations, we classify the news articles surrounding the investigation of event firms to identify a potential association with political investigations. We find direct evidence of an association with political investigation for 41 sample firms investigated (27.3% of sample), indirect evidence for 13 sample firms investigated (8.7% of sample), and no evidence for 96 sample firms investigated (64.0% of sample). Indirect evidence refers to cases where a corporate manager has a personal relationship with an investigated politician

(e.g., attending the same private club), but there is no direct evidence that the manager’s investigation is the consequence of the politician’s investigation.

The news coverage about corporate investigation may be incomplete and could fail to capture some connection to a political investigation. To validate the news analysis, it is useful to recall that the anti-corruption campaign (e.g., Eight-Point Regulation) targets SOEs but not Non-SOEs. Therefore, we expect investigations of non-SOEs to be mainly due to their involvement in ongoing political investigations. This is verified by the news analysis. Out of the 20 non-SOEs investigated, 15 have

26 direct evidence of a political connection, three have indirect evidence, and only two cases (10%) have

no evidence. This is a sharp contrast to investigated SOEs, where 26 out of 130 have direct evidence,

10 have indirect evidence, and 94 (72%) show no evidence of a connection. Thus, SOEs seem to be

directly examined for corrupt behavior, rather than just being targeted because of connection to an

ongoing political investigation.

Next, we repeat the regressions on corruption measures after excluding the event firms with

direct or indirect evidence of association with a political investigation. Table IA.14 of the Internet

Appendix shows that the coefficients for the corruption measures remain similar for this smaller sample, indicating that the effects shown by our corruption measures are not driven solely by measured political investigations.39

4.5 Discussion and Interpretations

There are strong linkages between corporate manager investigations and political

investigations. How to interpret these findings? The most natural interpretation is that the campaign

has a strong political component which ensares corrupt firms. However, an alternative narrative is that

a campaign solely focused on corruption would capture corrupt politicians and those corrupt

politicians would then be connected to corrupt firms. So, could the campaign be purely focused on

corruption? The analysis of Zhou, the previous leader is instructive. Since he was clearly part of a

failed party struggle it is widely believed that the strong reaction against him was at least partly

motivated by political cleansing. The fact that firms connected to him both through the University

and oil industry experienced a large increase in investigation probabililty indicates that the purging was

broad. Additionally, the fact that connections with the current PSC is associated with a strong decrease

in investigation probability indicates either a political protection effect, or that graduates of the very

top Chinese Universities have better job prospects are less likely to engage in corruption. Inconsistent

39 We also repeat this this analysis by excluding firms with direct evidence of association or connectedness with Zhou Yongkang, and the coefficients for the corruption measures are similar (Tables IA.15 and IA.16 of the Internet Appendix).

27 with the second interpretation, we do not observe decreases in the investigation probability for

managers affiliated with other top universities. These finding suggests that investigators may avoid

targeting managers from firms with a Tsinghua or Peking affiliation as doing so could be considered

an affront to the current leadership. Thus, the most evidence points to a strong political component.

Is it possible that the campaign is mostly politically motivated and the corruption measures

are highly significant because corrupt firms have unfavorable political ties that we do not observe?

This possibility can not be completely ruled out because there are likely other political connections

that we are unable to capture, like that through friends or relatives. However, the 14 extensive

measures of political connection minimize this possibility, as does the fact that magnitude of variation

explained by corruption measures is large and similar to that explained through political measures.

Overall, we caution that our empirical measures are not causal and are difficult to pin to precise

motives behind the measures. The findings indicate that the campaign is prosecuting corrupt firms,

but appears to contain a political spillover (cleansing) and protection effect dimension.

5. Is the Anti-Corruption Campaign Effective at Reducing Corporate Corruption in China?

A key question about the anti-corruption campaign is whether or not it accomplishes its stated purpose

of reducing corruption. To answer this question, we examine corruption measures for all Chinese

listed firms.40

5.1 Corruption measures for all firms: Before and after the start of anti-corruption campaign

Figure 4 plots the annual averages of corruption measures for all Chinese listed companies

from 2005 to 2015, and we focus on the changes around 2012, the start of the anti-corruption

campaign. Panels A to D show that the improvement associated with the anti-corruption campaign

40 There are two types of shares in China’s stock markets. A-shares are denominated in RMB and traded by only Chinese citizens. B-shares are denominated in either US dollars or Hong Kong dollars, and traded by foreign investors or domestic residents using foreign currency. We follow the literature and exclude the firms that issue only B-shares but not A-shares.

28 seems to concentrate in business entertainment expenditures (BEE). Specifically, Panel C shows a

dramatic decline in BEE, which provides formal support for the intense media coverage of China’s

efforts to eliminate luxury gifts and social events with the campaign. Consistent with our finding, a

contemporaneous study by Ke, Liu, and Tang (2017) documents that the anti-corruption campaign

has reduced the consumption of luxury goods and services, but this reduction does not increase values

or growth of the underlying firms.41 The other corruption measures either do not significantly improve

after the campaign or improve due to the general trend started long before the campaign. Specifically,

Panel A shows that CEO compensation declines slightly after the campaign. While CEO pay-for-

performance sensitivity increases after the campaign, the increase seems to be part of the general

upward trend starting in 2005.42 Therefore, while the anti-corruption campaign successfully reduced

business entertainment expenditures for Chinese firms, it has not significantly improved along other

dimensions of good governance.

Corresponding to Figure 4, Panel A of Table 7 presents annual corruption measures during

2005-2015 and statistical tests of the differences between 2013~15 and 2011. There is significant

decrease in business entertainment expenditures after the start of the anti-corruption campaign. CEO

compensation decreases slightly but Figure 4 shows that it began before the campaign. Pay-for-

performance sensitivity increases significantly after the campaign, but at a slower rate than before the

campaign. The other groups of measures either deteriorate or display mixed evidence. We also conduct

robustness test of SOEs in Panel B of Table 8 because SOEs are direct targets of the campaign, and

the results are similar to those of the full sample.43

41 Panel C also plots regulation breaches over the years. Since there is on average a time lapse of 0.7 year between disclosure and the last incident of a regulation breach, we focus on 2014 and 2015 which do not experience a decline in regulation breaches compared to earlier years. 42 Investment inefficiency decreases after 2012 but the decrease is largely due to the outlier year of 2011 and is concentrated in 2014 (there are increases in 2013 and 2015). 43 Table IA.17 of Internet Appendix further repeat the analysis for non-SOEs, and the results are similar to those of full sample and SOEs. To examine the differential impact of the campaign between SOEs and non-SOEs, we further report the differences of corruption measures between SOEs and and Non-SOEs, and their changes [2013~2015-2011] in Table

29 5.2 How does the campaign impact the investigated firms?

We also analyze changes in corruption measures for event firms after an investigation is finished. For each event firm, we identify a matched firm using the propensity score matching approach (PSM) based on pre-event corruption measures and firm characteristics (details in Section

A.10 of the Internet Appendix).44 CEO compensation significantly decreases (improves) but related- party loans significantly increases (deteriorate), and other measures remain little changed. Overall, there do not seem to be dramatic improvements in the corruption measures for event firms relative to matched firms, which is consistent with an overall lack of effectiveness.45

We further examine how the investigations affect the event firms’ stock prices and corruption measures. To study market reaction to corruption investigations, we calculated cumulative abnormal returns (CARs) of event firms in the [-1,+1] and the [-1,+15] windows, where day 0 is the announcement day of investigation and report the results in Table 8.46

Panel A of Table 8 shows a CAR [-1,+1] of around -1.5% (t-stats between -2.63 and -2.77), revealing a negative market response to investigations. CAR [-1,+15] remains negative but becomes insignificant (t-stat between -0.83 and -1.05). Since long-term returns can be noisy as they include various events during the long-term window, we focus on the short-term CARs that provide us with cleaner test. We nevertheless examine the long-term windows [-15,+125] and [-15,+250] and the

IA.18 of the Internet Appendix, where the only significant results are that relative to non-SOEs, SOEs experience a greater decline in entertainment expenditure (t-stat 2.19) and a smaller decline in investment inefficiency (t-stat 1.72). 44 This analysis restricts the sample to 69 firms as investigations in 2015 are dropped because of missing data for year t+1. Table IA.19 in the Internet Appendix reports corruption measures in years t-1 and t+1 for event firms and matched firms, where t is the year of corruption investigation. Because the focus is corrupt behaivor, we exclude monitoring by minority shareholders and CEO near-retirement as they are factors that affect corruption rather corrupt behavior. 45 We also examine the effect of investigations on financing and investment of event firms and their connected firms and find no evidence of significant changes in financing or investment after investigation (result reported in Tables IA.20 and IA.21 of the Internet Appendix). 46 Because event windows of some investigations overlap, we use the calendar-time approach to calculate abnormal returns. Specifically, we first calculate the daily abnormal return of a firm using one of three approaches: 1) daily return in excess of market return; 2) size-adjusted return by subtracting value-weighted return of the firm’s size decile portfolio; and 3) DGTW-adjusted return by subtracting the value-weighted return of the firm’s benchmark portfolio of size, book-to- market ratio, and past return. Then, for each trading day, we calculate the average daily abnormal returns of event firms whose event windows include this day, and average these average daily abnormal returns across days in the sample period.

30 CARs are insignificant, which provides suggestive evidence that the negative impact could be largely

temporary.

Panel B further reports CARs for non-event firms that are related to event firms through

customer-supply relation, locating in the same province, and connections among managers and

directors.47 Customers of event firms exhibit significantly negative CARs in the [-1,+1] window,

whereas other related firms experience insignificant price responses.

5.3 The effect of anti-corruption campaign on corporate disclosure, financial markets, and

foreign investments

We first examine if the anti-corruption campaign positively impacts China’s corporate culture from the perspective of accounting manipulation. Accounting manipulation is also directly associated

with corruption and Fan, Guan, Li, and Yang (2014) find that when political networks are broken earnings management decreases.48

Our first measure of accounting manipulation is discretionary accruals, which has previously

been used to examine Chinese firms [e.g., Liu and Lu (2007)]. A higher value of absolute discretionary accruals indicates greater earnings manipulation. Our second measure is discontinuity in earnings distribution around zero [Hayn (1995), Burgstahler and Dichev (1997)]. A much higher number of firms with small positive earnings than that of firms with small negative earnings indicates earnings manipulation.49 The construction of these two measures are detailed in Section A.3 of the Internet

Appendix.

47 Customer-supplier relation is identified using voluntary disclosure by Chinese firms on the names of top five customers and suppliers. The firm related to an event firm through manager and director connection is the non-event firm whose C- suite managers and directors have the highest total number of university affiliations and birthplace connections with those of the event firm. For supplier firms, we exclude an outlier which has a return of 77% due to a restructuring event. 48 Dass, Nanda, and Xiao (2014) and Huang (2016) document a greater tendency to manage earnings for U.S. firms that are more corrupt. 49 Chinese regulation adds to the incentive for Chinese firms to avoid negative earnings, as a Chinese listed firm with two years’ losses in a row will be assigned a label “ST” (special treatment) prior to its ticker, which is a negative signal to the market.

31 Figure 5 plots earnings distributions for all Chinese listed firms before and after the start of

the campaign. Panel A shows that there is a strong earnings discontinuity around zero for Chinese

firms overall during 2005-2011, indicating massive earnings manipulation in China’s corporate world

before the campaign. In the meantime, Panel B shows that earnings discontinuity remains similar in

the 2013-2015 period. For statistical analyses, we present in Table 9 the annual statistics from 2005 to

2015 as well as the differences [2013~2015 – 2011]. The change in absolute discretionary accruals is

significantly negative but the drop started in 2012 which is largely before the anti-corruption campaign.

The difference in earnings discontinuity is insignificantly positive. Overall, these results show that the

anti-corruption campaign does not reduce accounting manipulation for Chinese firms.

We further examine whether the anti-corruption campaign reduces leakage of inside

information and inside trading, both of which provide insight on the financial market environment.

Our test design follows Griffin, Hirschey, and Kelly (2011) and examines abnormal stock return

volatility on earnings-announcement days (Details of the construction described in Section A.3 of the

Internet Appendix). Lower stock return volatility indicates fewer price movements upon

announcement and therefore more information leakage or insider trading. Table 9 reports the volatility

measures from 2005 to 2015, which shows that stock return volatility on announcement days decreases

rather than increases after the anti-corruption campaign, and changes are statistically significant. This

result suggests that the campaign may not improve the information environment of China’s financial

markets.50

Previous studies such as Wei (2000), Habib and Zurawicki (2002), and Cuero-Cazurra (2006)

document that foreign direct investment (FDI) is affected by the host country’s corruption. We

therefore examine if the anti-corruption campaign positively affects FDI in China. Because FDI is

driven by many factors other than corruption, we develop a clean test design by examining FDIs

50 For robustness we repeat the analyese in Table 9 for SOEs and non-SOEs separately, and the results (Tables IA.22 and IA.23 of the Internet Appendix) also show little improvement in these measures for the two subsamples.

32 across provinces with different levels of marketization. Specifically, we use the provincial-level

marketization index developed by the China’s National Economic Research Institute (Fan, Wang, and

Zhu, 2011) based on official statistics and enterprise and household surveys. A higher marketization

index indicates smaller roles of government in economy, greater share of non-state sectors, freer price

setting and less trade barriers, better factors market development, and better legal environment.51 If the anti-corruption campaign increases FDI by promoting competition and reducing frictions, then we expect the effect on FDI to be stronger in provinces with lower levels of marketization.

We estimate a regression where the dependent variable is annual province-level FDIs, either in dollar value or scaled by GDP of the province. Consistent with our approach of calculating the difference [2013~2015 – 2011] for other measures in Tables 8 and 9, we include the years 2011 and

2013-2015 for this analysis and construct a post-campaign dummy which equals one for the years

2013-2015. The main independent variable is the interaction of the province-level marketization index

and the post-campaign dummy, and we further include their first-degree terms.52 In Table IA.24 of the Internet Appendix, the coefficient of the marketization index is significantly positive, indicating higher FDI in the provinces of greater marketization. More importantly, the interaction of marketization index and post-campaign dummy is insignificant. This result therefore provide little evidence that the anti-corruption campaign has positively affected foreign direct investments in China.

5.4 Benchmarking corruption measures of Chinese firms to Hong Kong firms

The changes in corruption measures of all Chinese listed firms, especially those based on financial variables, can be affected by varying macroeconomic condition. We attempt to address this

51 This index is a widely recognized measure of market reform and used by a number of financial studies (e.g., Wang, Wong, and Xia, 2008; Lin, Morck, Yeung, and Zhao, 2017). The values of marketization index range between 0 and 10 in the base year of 2001, and can be lower than 0 or higher than 10 in the following years to reflect changes in marketization over time. 52 We use the marketization index of 2009, which is the the most recent year from the repor of Fan, Wang, and Zhu (2011). The index was published every two year but the updating stopped from 2011 and resumed in 2016. Since FDI is one of the 23 components of the index, we reconstruct the index by purging the FDI component.

33 issue using Hong Kong firms as the benchmark. Because the economies of Mainland China and Hong

Kong are closely related, Hong Kong firms share similar economic conditions with Chinese firms but there is no anti-corruption campaign in Hong Kong during the same period.

We first screen Hong Kong listed firms to identify Hong Kong local firms instead of foreign or Mainland Chinese firms listed in Hong Kong, and require these firms to have December fiscal year ends like Chinese firms. We then use a propensity score matching procedure (PSM) to match Chinese listed firms with Hong Kong firms using available corruption measures as well as firm size, year and industry dummies. Due to data availability of Hong Kong firms, we are able to examine six of the measures for the benchmark analyses: 1) regulation breaches, 2) operational inefficiency, 3) investment inefficiency, 4) earnings discontinuity, 5) absolute value of discretionary accruals, and 6) return volatility around earnings announcement. For each measure, we first calculate the difference between

Chinese firms and matched Hong Kong firms, and then calculate the difference-in-difference measure before and after the anti-corruption campaign [2013~2015 - 2011]. More details of sample selection and the PSM approach are described in Section A.11 of the Internet Appendix. The results in Table

10 are consistent with our previous findings of no obvious improvement after the anti-corruption campaign. Specifically, within both pairs of earnings discontinuity and return volatility measures, one sees improvement while the other sees deterioration, and the changes are significant only at the 10% level. The other four measures all experienced significant deterioration after 2012.53 Nevertheless, benchmarking to Hong Kong firms alleviates rather than fully controls for economic conditions because even Hong Kong’s economy differs from that of Mainland China in many ways.

Overall, the results across measures of corruption, accounting manipulation, information

53 We test the parallel trends assumption and examine the difference [2009~2010 – 2011]. This analysis excludes 2008 when China launched a huge stimulus of 8 trillion Yuan in response to the global financial crisis, which can cause changes in accounting variables of Chinese firms relative to Hong Kong firms in 2008. Table IA.25 of the Internet Appendix shows that only three out of the eight variables exhibit violation of parallel trends assumption (small profit, operational inefficiency, and investment inefficiency). The five variables that satisfy the parallel trends assumption do not experience significant improvement after the campaign with the exception of differenced volatility (t-stat 1.66).

34 environment, and foreign investment show little signs of decreasing corruption with the exception of business and entertainment expenditures.

6. Conclusion

The Chinese anti-corruption campaign is the largest anti-corruption campaign in recent history and provides an interesting laboratory to examine the motivation for and effectiveness of anti- corruption efforts. Further understanding this initiative has practical ramifications as many developing countries follow China’s approach. Consistent with the campaign ensaring corrupt firms, investigated firms have lower minority shareholder ownership, higher CEO compensation, more near-retirement

CEOs, more related-party sales and related-party loans, more operational inefficiency, and more corruption postings than their matched-firm counterparts.

We further collect and investigate a wide variety of ties to government and current and former political leaders. Firms with connections to investigated political leaders such as Zhou, Yongkang exhibit considerably higher investigation probabilities. However, firms with ties to the top seven leaders exhibit a lower investigation probability. The results highlight how political connections can potentially provide protection but also be costly after a change in leadership. Importantly, in probit regressions that include both political and corporate connections, both sets of variables explain a sizeable share of the variation in investigation probability and are of similar economic importance.

Although our tests cannot disentangle precise motives behind the campaign, the findings suggest that the campaign contains a strong political dimension but also identifies executives at some of the more corrupt firms.

We ask if the campaign improves the environment for corporations by examining if there is broad improvement in corporate governance and managerial incentives, self-dealing, earnings manipulation, informational environment of financial markets, or foreign direct investment. For

35 Chinese firms as a whole, while there is a large decrease in highly visible business entertainment

expenditures in 2013-2015, less conspicuous but important indicators of self-dealing exhibit no

improvement. This lack of improvement is also present for earnings manipulation, the information

environment of financial markets, and foreign direct investment, and after benchmarking with Hong

Kong firms. Overall, our findings suggest that any measureable decreases in corruption indicators

appear to be largely limited to conspicuous expenditures.

While more time may be necessary to assess the this expanding campaign and its full impact,

our findings suggest that the reform may not be squarely focused on its stated anti-corruption goal,

and relatedly, may not improve corporate culture and reducing corporate corruption. The findings are

generally consistent with a literature skeptical of centralized initivies without broader reforms.

Summaries by Bardhan (1997), Wei (2001), Svensson (2005) suggest that any anti-corruption reform may have minimal effectiveness without fundamentally improving the legal, regulatory, and

informational landscapes.

36 References Aggarwal, R. K., F. Meschke, and T. Y. Wang. 2012. Corporate political donations: investment or agency? Business and Politics 14: 1-40. Allen, F., J. Qian, and M. Qian. 2005. Law, finance, and economic growth in China. Journal of Financial Economics 77: 57-116. Ang, A., J. Bai, and H. Zhou. 2016. The great wall of debt: The cross section of Chinese local government credit spreads. Working paper. Banyan. 2014. Tiger in the net. The Economist. Bardhan, P. 1997. Corruption and development: A review of issues. Journal of Economic Literature 35: 1320-1346. Barth, J. R., C. Lin, P. Lin, and F. M. Song. 2009. Corruption in bank lending to firms: Cross-country micro evidence on the beneficial role of competition and information sharing. Journal of Financial Economics 91: 361-388. Beck, T., A. Demirguc-Kunt, and V. Maksimovic. 2005. Financial and legal constraints to growth: does firm size matter? The Journal of Finance 60: 137-177. Beck, T., A. Demirguc-Kunt, and R. Levine. 2006. Bank supervision and corruption in lending. Journal of Monetary Economics 53: 2131-2163. Bergstresser, D., and T. Philippon. 2006. CEO incentives and earnings management. Journal of Financial Economics 80: 511–529. Biddle, G. C., G. Hilary, and R. S. Verdi. 2009. How does financial reporting quality relate to investment efficiency? Journal of Accounting and Economics 48: 112-131. Bjorkman, M., and J. Svensson. 2009. Power to the people: Evidence from a randomized field experiment on community-based monitoring in Uganda. The Quarterly Journal of Economics 124: 735–769. Borokhovich, K., K. Brunarski, and R. Parrino. 1997. CEO contracting and antitakeover amendments. Journal of Finance 52: 1495–517. Branigan, T. 2015. Politburo, army, casinos: China’s corruption crackdown spreads. The Guardian. Burgstahler, D., and I. Dichev. 1997. Earnings management to avoid earnings decreases and losses. Journal of Accounting and Economics 24: 99-126. Cai, H., H. Fang, and L. C. Wu. 2011. Eat, drink, firms, government: An investigation of corruption from the entertainment and travel costs of chinese firms. Journal of Law and Economics 54: 55– 78. Calomiris, C. W., R. Fisman, and Y. Wang. 2010. Profiting from government stakes in a command economy: Evidence from Chinese asset sales. Journal of Financial Economics 96: 399–412. Chen, C. J.P., Z. Li, X. Su, and Z. Sun. 2011. Rent-seeking incentives, corporate political connections, and the control structure of private firms: Chinese evidence. Journal of Corporate Finance 17: 229-243. Chen, Q., X. Chen, K. Schipper, Y. Xu, and J. Xue. 2012. The sensitivity of corporate cash holdings to corporate governance. Review of Financial Studies 25: 3610-3644. Chen, G., M. Firth, D. N. Gao, and O. M. Rui. 2006. Ownership structure, corporate governance, and

37 fraud: Evidence from China. Journal of Corporate Finance 12: 424-448. Chen, Y., M. Liu, and J. Su. 2013. Greasing the wheels of bank lending: Evidence from private firms in China. Journal of Banking and Finance 37: 2533-2545. Cheung, Y. L., P. R. Rau, and A. Stouraitis. 2010. Helping hand or grabbing hand? Central vs. local government shareholders in Chinese listed firms. Review of Finance 14: 669-694. Cohen, L., A. Frazzini, and C. Malloy. 2010. Sell-side school ties. The Journal of Finance 65: 1409-1437. Coles, J. L., N. D. Daniel, and L. Naveen. 2014. Co-opted boards. The Review of Financial Studies 27: 1751-1796. Core, J., and W. Guay. 1999. The use of equity grants to manage optimal equity incentive levels. Journal of Accounting and Economics 28: 151–184. Cuervo-Cazurra, A. 2006. Who cares about corruption? Journal of International Business Studies 37: 807-822. Dass, N., V. Nanda, and C. S. Xiao. 2014. Firms in corrupt environments and the value of corporate governance. Working Paper, Georgia Tech. Duchin, R., and D. Sosyura. 2012. The politics of government investment. Journal of Financial Economics 106: 24-48. Faccio, M. 2006. Politically connected firms. The American Economic Review 96: 369-386. Fan, J. P.H., F. Guan, Z. Li, and Y. G. Yang. 2014. Relationship networks and earnings informativeness: Evidence from corruption cases. Journal of Business Finance and Accounting 41: 831-866. Fan, J. P.H., K.C. J. Wei, and X. Xu. 2011. Corporate finance and governance in emerging markets: A selective review and an agenda for future research. Journal of Corporate Finance 17: 207-214. Fan, J. P.H., T. J. Wong, and T. Zhang. 2007. Politically connected CEOs, corporate governance and Post-IPO performance of China’s newly partially privatized firms. Journal of Financial Economics 84: 330–357. Fisman, R. 2001. Estimating the value of political connections. The American Economic Review 91: 1095- 1102. Fisman, R., and Y. Wang. 2015. The mortality cost of political connections. Review of Economic Studies 82: 1346-1382. Griffin, J. M., N. H. Hirschey, and P. J. Kelly. 2011. How important is the financial media in global markets? Review of Financial Studies 24: 3941-3992. Hayn, C. 1995. The information content of losses. Journal of Accounting and Economics 20: 125-153. Huang, M. 2016. Financial reporting quality in corrupt firms: Evidence from the foreign corrupt practices act. Working Paper, University of Rochester. Habib, M., and L. Zurawicki. 2002. Corruption and foreign direct investment. Journal of International Business Studies 33: 291-307. Hong, H., and I. Liskovich. 2017. Crime, punishment and the value of corporate social responsibility. Working Paper. Houston, J. F., C. Lin, and Y. Ma. 2011. Media ownership, concentration and corruption in bank

38 lending. Journal of Financial Economics 100: 326-350. Jensen, M. C., and W. H. Meckling. 1976. Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics 3: 305-360. Jian, M., and T. J. Wong. 2010. Propping through related party transactions. Review of Accounting Studies 15: 70-105. Jiang, G., C. M. C. Lee, and H. Yue. 2010. Tunneling through intercorporate loans: The China experience. Journal of Financial Economics 98: 1–20. Karpoff, J. M., D. S. Lee, and G. S. Martin. 2017. Foreign bribery: incentives and enforcement. Working Paper. Ke, B., N. Liu, and S. Tang. 2017. The effect of anti-corruption campaign on shareholder value in a weak institutional environment: Evidence from China. Working Paper. Li, C., and R. McElveen. 2014. Debunking misconceptions about Xi Jinping’s anti-corruption campaign. Brookings. Lin, C., R. Morck, B. Yeung, and X. Zhao. 2017. Anti-corruption reforms and shareholder valuations: Event study evidence from China. Working Paper. Li, H., L. Meng, Q. Wang, and L.A. Zhou. 2008. Political connections, financing and firm performance: Evidence from Chinese private firms. Journal of Developing Economics 87: 283-299. Liu, Q., and Z. J. Lu. 2007. Corporate governance and earnings management in the Chinese listed companies: A tunneling perspective. Journal of Corporate Finance 13: 881-906. Liu, L. X., H. Shu, and K.C. J. Wei. 2017. The impacts of political uncertainty on asset prices: Evidence from the Bo scandal in China. Journal of Financial Economics 125: 286-310. Magnus, G. 2015. Xi’s anti-corruption campaign key to China’s prospects. Nguyen, T. T., and M. A. van Dijk. 2012. Corruption, growth, and governance: Private vs. state-owned firms in Vietnam? Journal of Banking & Finance 36: 2935-2948. Olken B. A. 2007. Monitoring corruption: Evidence from a field experiment in Indonesia. Journal of Political Economy 115: 200-249. Pei, M. 2007. Corruption threatens China’s future. Policy Brief 55. Qian, N., and J. Wen. 2015. The impact of Xi Jinping’s anti-corruption campaign on luxury imports in China. Working Paper. Reinikka, R., and J. Svensson. 2004. Local capture: Evidence from a central government transfer program in Uganda. Quarterly Journal of Economics 119: 679-705. Shleifer, A., and R. W. Vishny. 1993. Corruption. Quarterly Journal of Economics 108: 599-617. Shleifer, A., and R. W. Vishny. 1994. Politicians and firms. Quarterly Journal of Economics 109: 995-1025. Svensson, J. 2005. Eight questions about corruption. Journal of Economic Perspectives 19: 19-42. Wei, S. 2000. How taxing is corruption on international investors. Review of Economics and Statistics 82: 1-11. Wei, S. 2001. Corruption in Economic Transition and Development: Grease or Sand? Working Paper. Zeume, S. 2016. Bribes and firm value. Review of financial studies 30: 1457-1489.

39 Figure 1 Description of Sample Firms This figure plots the distribution of 150 sample firms with corrupt managers investigated during China’s anti-corruption campaign. The sample period starts from the beginning of the anti-corruption campaign on December 4, 2012 to December 31, 2015. Panel A plots the distributions of state-owned enterprises (SOEs) and Non-SOEs by year. Panel B plots the distribution of firms with different positions of corrupt managers by year. The corrupt managers in the sample are CEOs, other top managers who also serve as internal directors, and top managers of parent company. Panel C plots the distributions of firms involved in different specific corrupt behaviors for SOEs and Non-SOEs. These corrupt behaviors are the most common ones among sample firms, and they are not mutually exclusive. Panel A: State-Owned Enterprises (SOEs) and Non-SOEs across Years

80 70 60 50 40 30

Number of of Number Firms 20 10 0 2012 2013 2014 2015 Year SOE Non-SOE

Panel B: Distribution of Firms with Different Manager Positions across Years

35 30 25 20 15 10

Number of of Number Firms 5 0 2012 2013 2014 2015 Year CEOs Other Managers Parent Managers

40 Panel C: Distribution of Firms with Different Corrupt Behaviors for SOEs and Non-SOEs

90 80 70 60 50 40 30 Number of of Number Firms 20 10 0 Receiving Embezzling Benefiting Bribing Others Unspecified Bribes Funds Family Members Corrupt Behaviors SOE Non-SOE

41 Figure 2 Firm-Level Corruption Measures before Corruption Investigation Events for Event Firms and Matched Firms This figure presents twelve corruption measures for event firms and matched firms. The sample includes 150 Chinese listed firms with corrupt managers investigated since China’s anti-corruption campaign from December 4, 2012 to December 31, 2015. For each event firm, we identify a matched firm by first selecting a subsample of firms satisfying the following conditions: 1) In the same industry as the event firm; 2) Have the same SOE status as the event firm; and 3) Market capitalization is within the range of 50% and 150% of the event firm. We then choose from this subgroup a matched firm that has the closest book-to-market ratio to the event firm. The figure plots the corruption measures in the years t-2, t-1 and t where t is the year of corruption investigation. All Chinese firms’ fiscal years end in December so the fiscal year coincides with the calendar year. The firm-level corruption measures include: 1) Monitoring by minority shareholders, calculated using ownerships of the 2nd to the 5th largest shareholders; 2) CEO compensation, scaled by total assets; 3) CEO pay-for-performance sensitivity; 4) CEO near-retirement dummy, which equals one if CEO’s age is greater than or equal to 59; 5) Related-party sales, scaled by revenue; 6) Related-party loans, scaled by total assets; 7) Other receivables from parent firm, scaled by total assets; 8) Number of regulation breaches in a year; 9) Business entertainment expenditures, scaled by total assets; 10) Operational inefficiency, calculated as growth of sales minus growth of net income; 11) Investment inefficiency, calculated as the absolute value of residue from regression of investment on sales growth within each industry-year; and 12) Corruption postings, measured as percentage of posts that discussed corruption in the total posts for a firm on “Guba” (“Stock Bar” in English), a popular online investor-forum. Operational Inefficiency is Winsorized at 5% and 95% for each year because of the large number of outliers. All the other firm-level corruption measures, except CEO near-retirement dummy and number of regulation breaches, are Winsorized at 1% and 99% for each year. We exclude seven financial firms for the following measures: related-party sales, related-party loans, other receivables from parent firm, operational inefficiency, and investment inefficiency. To ease reading, CEO compensation, other receivables from parent firm, and business entertainment expenditures are expressed in percentage.

42 Monitoring by Minority Shareholders CEO Compensation CEO Pay for Performance Sensitivity

1.0 0.020 0.08 0.07 0.8 0.015 0.06 0.6 0.05 0.010 0.04

0.4 (%) Monitor 0.03 0.005 0.02 0.2 0.01

0.0 0.000 PerformanceCEO Pay for 0.00 CEO Compensation Compensation CEO / Asset t-2 t-1 t t-2 t-1 t t-2 t-1 t Event Matched Event Matched Event Matched

CEO Near-Retirement Dummy Related-Party Sales Related-Party Loans 0.40 0.15 0.010 0.35 0.008 0.30 0.10 0.25 0.006 Retirement

- 0.20

party sales / 0.004 -

0.15 Assets

Revenue 0.05 party loansparty / Total 0.10 - 0.002 CEO Near 0.05 Related

0.00 0.00 Related 0.000 t-2 t-1 t t-2 t-1 t t-2 t-1 t Event Matched Event Matched Event Matched

43 Other Receivables from Parent Firm Number of Regulation Breaches Business Entertainment Expenditures 0.012 0.25 0.30

0.010 0.20 0.25 0.008 0.20 0.15 0.006 0.15

breaches 0.10 0.004 0.10 0.002 0.05 Number of Number of regulation 0.05 / Total Assets (%) Assets Total / Other Receivables from Parent/ Assets Total (%) 0.000 0.00 0.00 t-2 t-1 t t-2 t-1 t Expenditures Entertainment t-2 t-1 t Event Matched Event Matched Event Matched

Operational Inefficiency Investment Inefficiency Corruption Postings 0.8 0.3 0.15

0.2 0.6 0.12 0.2 0.09 0.4 0.1 0.06 0.2 0.1 0.03 Investment Inefficiency Operational Inefficiency Operational Inefficiency Percentage of Percentage of corruption related posts StockBar related in 0.0 0.0 0.00 t-2 t-1 t t-2 t-1 t t-2 t-1 t Event Matched Event Matched Event Matched

44 Figure 3 Corporate Investigations and Political Connections This figure presents the university affiliations of the investigated firms and their matched firms with the current and investigated top leaders. Investigated firms include the 150 sample firms investigated for corruption during the anti-corruption campaign. For each event firm, we identify a non- investigated matched firm using the approach described in the heading of Figure 2. The universities include three (Tsinghua University, Peking University, and Xiamen University) attended by the seven members of the Politburo Standing Committee (PSC) of CPC’s Central Committee, as well as one (China Petroleum University) attended by Yongkang Zhou, the past member of the PSC investigated during the campaign. Each red solid (blue dotted) line connecting firms with a university represents a C-Suite executive of an investigated (non-investigated) firm graduated from the university. The figure does not include three other universities attended by the current top seven leaders but not corporate managers (Kim Il-sung University, Harbin Engineering University, and Northwest University).

45 Figure 4 Corruption Measures for All Firms: 2005-2015 This figure plots the annual averages of eleven corruption measures for Chinese listed companies from 2005 to 2015. The sample includes all firms listed on Shanghai and Shenzhen stock exchanges (A shares). All Chinese firms’ fiscal years end in December so the fiscal year coincides with the calendar year. The corruption measures are grouped into five categories and plotted in five figures. Two measures of CEO incentives include: 1) CEO compensation, scaled by total assets; and 2) CEO pay-for-performance sensitivity. The three measures of related-party transactions include: 1) Related-party sales, scaled by revenue; 2) Related-party loans, scaled by total assets; and 3) Other receivables from parent firm, scaled by total assets. The two measures of illegal/unethical behaviors include: 1) Number of regulation breaches; and 2) Business entertainment expenditures, scaled by total assets. The two measures of efficiency include: 1) Operational inefficiency, calculated as growth of sales minus growth of net income; and 2) Investment inefficiency, calculated as the absolute value of residue from regression of investment on sales growth within each industry-year. Operational inefficiency is Winsorized at 5% and 95% for each year because of the large number of outliers. All the other firm-level corruption measures, except number of regulation breaches, are Winsorized at 1% and 99% for each year. We exclude financial companies for five measures: related-party sales, related-party loans, other receivables from parent firm, operational inefficiency, and investment inefficiency. To ease reading, CEO compensation, other receivables from parent firm, and business entertainment expenditures are expressed in percentage.

46 Panel A: CEO Incentives and Compensation Panel B: Related-Party Transactions 0.033 0.30 0.07 1.0 0.031 0.9 0.25 0.06 0.029 0.8 0.05 0.7 0.027 0.20 0.6 0.025 0.04 0.15 0.5 0.023 0.03 Party Loan/Asset 0.4 -

0.021 0.10 0.3 Parent/Asset(%)

Party Sales/Revenue 0.02 -

0.019 0.2 Other Receivables from 0.05 0.01

CEO Compensation (%) 0.017 0.1 CEO Pay for PerformanceCEO Pay for

Related 0.00 0.0 0.015 0.00 and Related 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Year Year Related-Party Sales Related-Party Loans CEO Compensation CEO Pay for Performance Other Rec. from Parent

Panel C: Regulation Breaches & Entertainment Expenditures Panel D: Operating Inefficiency and Investment Ineffiency

1.00 0.20 0.25 0.30 0.29 0.80 0.18 0.20 0.28 0.27 0.60 0.16 0.15 0.26 0.25 0.40 0.14 0.10 0.24 0.20 0.12 0.23 0.05 0.22 Investment Inefficiency Operational Inefficiency 0.00 0.10 0.21 0.00 0.20

Entertainment Expenditures(%) Entertainment -0.20 0.08 Number of Number of BreachesRegulation 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 Year Year Regulation Breaches Entertainment Exp. Operational Inefficiency Investment Inefficiency

47 Figure 5 Distribution of Earnings for All Firms before and after Anti-Corruption Campaign Started This figure plots the distribution of earnings for all Chinese listed firms before and after the start of the anti-corruption campaign on December 4, 2012. The sample includes all firms listed on Shanghai and Shenzhen stock exchanges (A shares), and earnings is defined as net income (NI , ) scaled by market value of equity in the previous year end (ME ). All Chinese firms’ fiscal years end in , 𝑖𝑖 𝑡𝑡 December so the fiscal year coincides with the calendar year. Panel A presents the distribution of 𝑖𝑖 𝑡𝑡−1 earnings for all firms from 2005 to 2011, and Panel B presents the distribution of earnings for all firms from 2013 to 2015. We exclude earnings of 2012 because its indication is not clear: Most of 2012 is before the start of the anti-corruption campaign, but the 2012 earnings figures were composed and announced after the start of the anti-corruption campaign. Panel A: Distribution of Earnings for All Listed Firms: 2005-2011 16% 14% 12% 10% 8% 6% 4% Percentage ofPercentage Firms 2% 0% -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 Earnings

Panel B: Distribution of Earnings for All Listed Firms: 2013-2015 16% 14% 12% 10% 8% 6% 4%

Percentage ofPercentage Firms 2% 0% -0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 Earnings

48 Table 1 Distribution of Corruption Investigation Events This table presents the distribution of 150 sample firms with corrupt managers investigated during China’s anti-corruption campaign. The sample period starts from the beginning of the anti-corruption campaign on December 4, 2012 to December 31, 2015. Panel A presents the distribution of sample firms across year, and distribution of positions of corrupt managers. The corrupt managers in the sample are CEOs, other top managers who also serve as internal directors, and top managers of parent company. Panel B presents the distribution of specific corrupt behaviors for state-owned enterprises (SOEs) and non-SOEs separately. These corrupt behaviors are the most common ones among sample firms, and they are not mutually exclusive. Panel A: Characteristics of Event Firms Categories of Event Firms # Firms # SOEs # Non-SOEs Year of Events 2012 1 1 0 2013 28 23 5 2014 50 42 8 2015 71 64 7 Positions of Corrupt Managers CEO/Chairman 67 49 18 Other Top Managers 25 23 2 Managers of Parent Firms 58 58 0 Total #Firms 150 130 20 Panel B: Distribution of Specific Corrupt Behaviors Main Corrupt Behaviors # Firms # SOEs # Non-SOEs Receiving Bribes 82 82 0 Embezzling Company Funds 26 25 1 Illegally Benefiting Family Members 29 29 0 Bribing Others 21 4 17 Unspecified 31 29 2 Total #Firms 150 130 20

49 Table 2 Corruption Measures for Event Firms before Corruption Investigations This table presents corruption measures for event firms in the years before corruption investigations. The sample includes 150 Chinese listed firms with corrupt managers investigated from the beginning of the anti-corruption campaign on December 4, 2012 to December 31, 2015. All Chinese firms’ fiscal years end in December so their fiscal year coincides with the calendar year. For each event firm, we identify a matched firm by first selecting a subsample of firms satisfying the following conditions: 1) In the same industry as the event firm; 2) Have the same SOE status as the event firm; and 3) Market cap is within the range of 50% and 150% of the event firm. We then choose from this subgroup a matched firm that has the closest book-to-market ratio to the event firm. The table presents the corruption measures in the years t-1 and t, where t is the year of corruption investigation. The firm- level corruption indicators include: 1) Monitoring (Monitoring by minority shareholders), calculated using ownerships of the 2nd to the 5th largest shareholders; 2) Compensation (CEO Compensation), scaled by total assets; 3) Pay for performance (CEO pay-for-performance sensitivity), which is the change in dollar value of CEO’s stock and option holdings in response to one percent change in stock price, scaled by the sum of the dollar value change, CEO salary, and CEO bonus; 4) Near-retirement (CEO near-retirement dummy), which equals one if CEO’s age is greater than or equal to 59; 5) Related-party sales, scaled by revenue; 6) Related-party loans, scaled by total assets; 7) Other receivables (Other receivables from parent firm), scaled by total assets; 8) Regulation breaches (Number of regulation breaches in a year); 9) Entertain. exp. (Business entertainment expenditures, scaled by total assets); 10) Operational inefficiency, calculated as growth of sales minus growth of net income; 11) Investment inefficiency, calculated as the absolute value of residue from regression of investment on sales growth within each industry- year; and 12) Corruption postings, measured as percentage of posts that discussed corruption in the total posts for a firm on “Guba” (“Stock Bar” in English), a popular online investor-forum. Operational inefficiency is Winsorized at 5% and 95% for each year because of the large number of outliers. All the other firm-level corruption measures, except CEO near-retirement dummy, number of regulation breaches, are Winsorized at 1% and 99% for each year. We exclude seven financial firms for the following measures: related-party sales, related-party loans, other receivables from parent firm, operational inefficiency, and investment inefficiency. To ease reading, CEO compensation, other receivables from parent firm, and business entertainment expenditures are expressed in percentage. Bold is used for numbers that are statistically significant at 0.10 level. Year t-1 Year t Event Match Event Match Firms Firms Diff t-stat Firms Firms Diff t-stat Monitoring 0.436 0.713 -0.277 (-2.18) 0.421 0.730 -0.309 (-2.48) Compensation (%) 0.016 0.012 0.005 (1.75) 0.013 0.014 0.000 (-0.13) Pay for Performance 0.018 0.061 -0.042 (-2.30) 0.029 0.071 -0.043 (-2.12) Near-Retirement 0.280 0.153 0.127 (2.77) 0.336 0.161 0.175 (3.36) Related-Party Sales 0.100 0.065 0.035 (1.99) 0.110 0.062 0.048 (2.65) Related-Party Loans 0.006 0.002 0.003 (1.58) 0.008 0.005 0.003 (1.33) Other Receivables (%) 0.009 0.003 0.006 (1.45) 0.002 0.002 0.000 (0.18) Regulation Breaches 0.153 0.153 0.000 (0.00) 0.213 0.120 0.093 (1.61) Entertain. Exp. (%) 0.211 0.189 0.022 (0.44) 0.127 0.169 -0.041 (-1.29) Operational Inefficiency 0.501 0.111 0.390 (2.50) 0.880 0.609 0.271 (1.14) Invest. Inefficiency 0.142 0.129 0.012 (1.44) 0.149 0.137 0.012 (1.10) Corruption Postings 0.087 0.058 0.030 (3.12) 0.140 0.072 0.068 (5.84)

50 Table 3 Probit Regressions of Corruption Investigation on Corruption Measures This table presents probit regressions of corruption investigation on corruption measures. The sample includes Chinese listed firms with corrupt managers investigated from the beginning of the anti-corruption campaign on December 4, 2012 to December 31, 2015, as well as their matched firms. For each event firm, we identify a matched firm by first selecting a subsample of firms satisfying the following conditions: 1) In the same industry as the event firm; 2) Have the same SOE status as the event firm; and 3) Market cap is within the range of 50% and 150% of the event firm. We then choose from this subgroup a matched firm that has the closest book-to-market ratio to the event firm. The dependent variable is a dummy variable that equals one if the firm was investigated (event firm), and zero if the firm was not investigated (matched firm). The major independent variables are firm-level corruption measures of the year prior to corruption investigation (year t-1). The firm-level corruption indicators include: 1) Monitoring (Monitoring by minority shareholders), calculated using ownerships of the 2nd to the 5th largest shareholders; 2) CEO compensation, scaled by total assets; 3) CEO pay for performance (CEO pay-for-performance sensitivity), which is the change in dollar value of CEO’s stock and option holdings in response to one percent change in stock price, scaled by the sum of the dollar value change, CEO salary, and CEO bonus; 4) CEO near-retirement dummy, which equals one if CEO’s age is greater than or equal to 59; 5) Related-party sales, scaled by revenue; 6) Related-party loans, scaled by total assets; 7) Other receivable from parent (Other receivables from parent firm, scaled by total assets); 8) # Regulation breaches (Number of regulation breaches in a year); 9) Entertainment Expenditures (Business entertainment expenditures, scaled by total assets); 10) Operational inefficiency, calculated as growth of sales minus growth of net income; 11) Investment inefficiency, calculated as the absolute value of residue from regression of investment on sales growth within each industry-year; and 12) Corruption postings, measured as percentage of posts that discussed corruption in the total posts for a firm on “Guba” (“Stock Bar” in English), a popular online investor-forum. Operational inefficiency is Winsorized at 5% and 95% for each year because of the large number of outliers. All the other firm-level corruption measures, except CEO near-retirement dummy, number of regulation breaches, are Winsorized at 1% and 99% for each year. The regressions also control for firm characteristics including natural log of market capitalization, and a dummy variable for state-owned enterprises (SOE). All Chinese firms’ fiscal years end in December so the fiscal year coincides with the calendar year. We exclude seven event firms in the finance industry and their matched firms for the models using five measures: related-party sales, related-party loans, other receivables from parent firm, operational inefficiency, and investment inefficiency. All models include year fixed effects, industry fixed effects, and region fixed effects. T-statistics associated with coefficients are reported in the parentheses. The coefficients on CEO compensation, other receivables and business entertainment expenditures are divided by 1,000 to ease reading. ***, **, and * represent statistical significance at the 0.01, 0.05, and 0.10 levels , respectively. Bold is used for numbers that are statistically significant at 0.10 level.

51 Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) Monitoring -0.187** -0.213* -0.241*** (-2.53) (-1.81) (-2.68) CEO Compensation 0.564* 0.483 0.700** (1.69) (1.32) (1.98) CEO Pay for Performance -0.812 -0.843 -0.880 (-1.42) (-1.34) (-1.45) CEO Near-Retirement Dummy 0.537*** 0.686*** 0.692*** (2.76) (2.63) (3.24) Related-Party Sales 1.029* 1.225* 1.288** (1.94) (1.80) (2.28) Related-Party Loans 8.702* 10.537* 12.394** (1.71) (1.79) (2.24) Other Receivables from Parent 0.719 1.356* 1.031* (1.45) (1.73) (1.79) # Regulation Breaches 0.015 0.077 0.117 (0.07) (0.34) (0.57) Entertainment Expenditures -0.008 -0.015 (-0.21) (-0.36) Operational Inefficiency 0.165** 0.161* 0.180** (2.50) (1.84) (2.43) Investment Inefficiency 2.101* 1.293 1.945 (1.90) (0.94) (1.63) Corruption Postings 1.922** 1.737 2.474** (2.15) (1.22) (2.42) Ln(ME) 0.152* 0.155** 0.175* 0.155** 0.107 0.187 0.127 (1.92) (2.03) (1.68) (2.00) (1.42) (1.52) (1.40) SOE Dummy -0.164 -0.226 -0.250 -0.208 -0.176 -0.680** -0.439 (-0.63) (-0.93) (-0.96) (-0.83) (-0.75) (-2.06) (-1.51) Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes # Obs 293 286 213 281 300 197 274

52 Table 4 Probit Regressions of Corruption Investigation on Corruption Measures: SOEs Only This table presents probit regressions of corruption investigation on corruption measures for the subsample of state-owned enterprises. The regression setting and definitions of variables are the same as those in Table 3. T-statistics associated with coefficients are reported in the parentheses. ***, **, and * represent statistical significance at the 0.01, 0.05, and 0.10 levels, respectively. Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) Monitoring -0.225*** -0.335** -0.319*** (-2.77) (-2.42) (-3.16) CEO Compensation 1.261* 0.746 1.724** (1.79) (0.77) (2.23) CEO Pay for Performance -1.582 -2.259* -2.102* (-1.59) (-1.66) (-1.72) CEO Near-Retirement Dummy 0.472** 0.727** 0.618*** (2.28) (2.45) (2.69) Related-Party Sales 1.526*** 1.841** 1.993*** (2.62) (2.49) (3.13) Related-Party Loans 7.722 10.637 11.630** (1.48) (1.62) (2.00) Other Receivables from Parent 0.649 1.273 1.018* (1.30) (1.51) (1.72) # Regulation Breaches -0.020 0.038 0.097 (-0.09) (0.15) (0.44) Entertainment Expenditures 0.016 0.018 (0.36) (0.38) Operational Inefficiency 0.168** 0.160* 0.189** (2.43) (1.67) (2.40) Investment Inefficiency 2.636** 1.574 2.186* (2.25) (1.03) (1.70) Corruption Postings 1.765* 0.964 2.138** (1.94) (0.62) (1.99) Ln(ME) 0.193** 0.142* 0.152 0.136 0.091 0.222 0.202* (2.06) (1.71) (1.28) (1.63) (1.11) (1.40) (1.83) Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes # Obs 253 248 178 245 260 164 238

53 Table 5 Probit Regressions of Corruption Investigation on Political Connection Measures This table presents probit regressions of corruption investigation on political connection measures. The regressions are similar to those in Table 3 but include the political connection measures: 1) Government connection, a dummy variable that equals one if a C-Suite executive was previously a high-ranked government official; 2) Central and local government connection, two dummy variables constructed according to whether a C-Suite manager was previously a high-ranked government official for the central or local government; 3) University affiliation: Top 7 leaders, number of connections where a C-Suite executive of the company graduated from the same university as a top 7 leader; 4) Birthplace connection: Top 7 leaders, a dummy variable that equals 1 if the company’s headquarter is located in the home province of a top 7 leader; 5) Association with Yongkang Zhou, a dummy variable that equals one if the firm is located in the Sichuan province or in the oil industry, the power base of Zhou, Yongkang, the highest ranked leader investigated; 6) University affiliations with investigated (non-investigated) national leaders, constructed similarly as in 3) but using the investigated or non-investigated national leaders; 7) Birthplace connections with investigated or non-investigated national leaders, constructed similarly as in 4) but using the number of investigated or non-investigated national leaders whose home province is the same as the company’s headquarters province; 8) Birthplace connections with the 1,021 investigated government officials, constructed similarly constructed as in 7) but using the 1,021 investigated officials publicized on the CCDI website; 9) Province graft-tigers and graft-flies. The former is the average rank of investigated provincial officials in the six months prior to the investigation month, and the latter is the total number of investigated provincial official in the prior six months, scaled by the number of counties in the province; 10) Investigation team in province, a dummy variable that equals one if an investigation team was in the province in the prior six months. For the university and birthplace connection variables, we take natural log of the sum of raw value and one. ***, **, and * represent significance at the 0.01, 0.05, and 0.10 levels, respectively. Bold is used for numbers statistically significant at 0.10 level. Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Government Connection 0.725*** (3.45) Local Government Connection 0.927*** 1.068*** 0.831*** (3.98) (3.72) (3.34) Central Government Connection -0.051 -0.089 0.239 (-0.13) (-0.18) (0.53) University Affiliation: Top 7 Leaders -1.636*** -2.003*** -2.405*** (-4.14) (-2.98) (-3.71) Birthplace Conn.: Top 7 Leaders 0.037 -0.006 0.068 (0.10) (-0.01) (0.15) Association with Zhou, Yongkang 0.997** 1.851*** 1.972*** (2.20) (2.58) (2.94) Univ. Conn: Invest. National Leaders 0.871** 0.064 0.364 (2.07) (0.12) (0.74) Birth Conn: Invest. National Leaders 0.457 0.484 0.806* (1.14) (0.87) (1.76)

54 Dependent Variable: Dummy of Corruption Investigation Univ. Conn: Non-Invest. National Leaders -0.348*** 0.266 0.299 (-2.64) (1.11) (1.32) Birth Conn: Non-Invest. National Leaders -0.377** -0.800*** -0.566** (-2.01) (-2.90) (-2.38) Birth Conn: 1,021 Invest. Politicians -0.142 0.021 -0.152 (-1.04) (0.10) (-0.82) Province Graft-Tigers 0.365* 0.334 (1.80) (1.27) Province Graft-Flies -1.119 -1.356 (-0.86) (-0.91) Investigation Team in Province 0.472** 0.633** 0.421 (1.99) (2.09) (1.57) Monitoring -0.222** -0.215** -0.189* -0.266*** -0.248*** -0.233** -0.200** -0.235*** -0.173 -0.220** (-2.45) (-2.36) (-1.94) (-2.86) (-2.61) (-2.57) (-2.03) (-2.61) (-1.37) (-2.02) CEO Compensation 0.692** 0.720** 0.784** 0.719** 0.724** 0.648* 0.253 0.601* 0.338 0.636* (1.99) (2.06) (2.15) (2.03) (2.02) (1.84) (0.60) (1.71) (0.80) (1.77) CEO Pay for Performance -0.861 -0.880 -0.706 -0.801 -0.792 -0.867 -1.131* -0.806 -0.903 -0.543 (-1.42) (-1.43) (-1.06) (-1.32) (-1.22) (-1.43) (-1.72) (-1.34) (-1.19) (-0.79) CEO Near-Retirement Dummy 0.488** 0.565** 0.664*** 0.704*** 0.682*** 0.711*** 0.735*** 0.689*** 0.694** 0.548** (2.18) (2.46) (3.03) (3.27) (3.13) (3.32) (3.15) (3.20) (2.49) (2.20) Related-Party Sales 1.095* 1.149** 1.046* 1.254** 1.461** 1.238** 1.492** 1.300** 0.682 0.483 (1.91) (1.96) (1.78) (2.15) (2.41) (2.18) (2.50) (2.30) (0.88) (0.67) Related-Party Loans 11.135* 10.961* 16.911** 12.536** 13.806** 12.942** 13.388** 12.368** 18.345** 17.536** (1.92) (1.89) (2.57) (2.25) (2.42) (2.31) (2.24) (2.24) (2.26) (2.33) Other Receivables from Parent 0.898 0.830 1.611** 1.092* 1.521** 1.015* 0.948 0.922 1.441* 1.515** (1.55) (1.46) (2.34) (1.92) (2.32) (1.74) (1.57) (1.61) (1.92) (2.02) # Regulation Breaches 0.100 0.122 0.064 0.107 0.083 0.124 0.225 0.109 0.212 0.100 (0.48) (0.58) (0.30) (0.52) (0.39) (0.60) (0.98) (0.53) (0.86) (0.46) Operational Inefficiency 0.179** 0.171** 0.195** 0.179** 0.199*** 0.177** 0.127 0.187** 0.102 0.164* (2.38) (2.25) (2.51) (2.38) (2.64) (2.38) (1.53) (2.47) (1.09) (1.93) Investment Inefficiency 1.713 1.758 1.992 1.670 1.953 1.986* 2.268* 2.012* 2.440 1.729 (1.40) (1.42) (1.62) (1.39) (1.61) (1.65) (1.73) (1.67) (1.60) (1.30) Corruption Postings 2.375** 2.620** 3.101*** 2.024* 2.641** 2.562** 2.435** 2.379** 2.432* 2.507** (2.33) (2.51) (2.90) (1.92) (2.45) (2.49) (2.28) (2.31) (1.94) (2.09) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year, Industry, Region Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes # Obs 274 274 274 274 274 274 235 274 235 274

55 Table 6 Probit Regressions of Corruption Investigation on Political Connection Measures: SOEs Only This table presents probit regressions of corporate investigation on political connection measures for the subsample of state-owned enterprises. The regression setting and definitions of variables are the same as those in Table 5. ***, **, and * represent significance at the 0.01, 0.05, and 0.10 levels, respectively. Bold is used for numbers statistically significant at 0.10 level. Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Government Connection 0.722*** (3.26) Local Government Connection 0.934*** 0.965*** 0.804*** (3.78) (3.11) (2.98) Central Government Connection -0.024 -0.326 0.195 (-0.06) (-0.62) (0.41) University Affiliation: Top 7 Leaders -1.763*** -2.358*** -2.648*** (-3.96) (-3.12) (-3.63) Birth Conn: Top 7 Leaders 0.061 0.195 0.174 (0.15) (0.34) (0.35) Association with Zhou Yongkang 1.410** 2.753*** 2.599*** (2.51) (2.92) (3.14) Univ. Conn: Invest. National Leaders 1.077** 0.248 0.589 (2.19) (0.38) (1.00) Birth Conn: Invest. National Leaders 0.503 0.289 0.965* (1.15) (0.46) (1.82) Univ. Conn: Non-Invest. National Leaders -0.344** 0.369 0.358 (-2.30) (1.36) (1.45) Birth Conn: Non-Invest. National Leaders -0.399* -0.986*** -0.664** (-1.85) (-2.81) (-2.37) Birth Conn: 1,021 Invest. Politicians -0.199 0.059 -0.293 (-1.32) (0.24) (-1.43) Province Graft-Tigers 0.522** 0.645** (2.23) (1.99) Province Graft-Flies -1.356 -0.182 (-0.93) (-0.10) Investigation Team in the Province 0.459* 0.548 0.254 (1.71) (1.48) (0.80) 56

Dependent Variable: Dummy of Corruption Investigation Monitor by Minority Shareholders -0.306*** -0.302*** -0.259** -0.357*** -0.323*** -0.316*** -0.241** -0.311*** -0.231 -0.322** (-3.01) (-2.96) (-2.34) (-3.36) (-3.04) (-3.12) (-2.16) (-3.08) (-1.53) (-2.57) CEO Compensation ** ** ** ** ** ** * ** 1.571 1.557 2.042 1.764 1.772 1.564 1.696 1.604 1.802 1.307 (2.06) (2.02) (2.52) (2.25) (2.23) (2.00) (1.81) (2.05) (1.61) (1.49) CEO Pay for Performance * * * * ** -2.169 -2.330 -1.478 -2.142 -1.714 -2.008 -2.693 -2.027 -2.360 -1.685 (-1.72) (-1.75) (-0.96) (-1.75) (-1.20) (-1.66) (-2.08) (-1.64) (-1.29) (-0.94) CEO Near-Retirement Dummy * ** ** *** ** *** ** *** ** * 0.411 0.499 0.580 0.664 0.604 0.639 0.657 0.613 0.756 0.538 (1.71) (2.01) (2.43) (2.84) (2.57) (2.77) (2.55) (2.65) (2.36) (1.94) Related-Party Sales 1.771*** 1.859*** 1.773*** 1.978*** 2.267*** 1.928*** 2.499*** 1.963*** 2.124** 1.604* (2.76) (2.83) (2.64) (2.95) (3.24) (3.00) (3.58) (3.09) (2.26) (1.91) Related-Party Loans 10.184* 9.936 17.242** 11.727** 13.002** 12.280** 12.411** 11.591** 16.934* 16.990* (1.66) (1.62) (2.40) (1.98) (2.16) (2.07) (2.00) (1.99) (1.78) (1.93) Other Receivables from Parent 0.826 0.747 1.675** 1.047* 1.542** 0.958 0.899 0.922 1.340* 1.325* (1.41) (1.31) (2.36) (1.81) (2.28) (1.60) (1.50) (1.56) (1.89) (1.85) # Regulation Breaches 0.070 0.088 0.065 0.092 0.071 0.103 0.287 0.092 0.299 0.101 (0.32) (0.39) (0.29) (0.42) (0.32) (0.47) (1.14) (0.42) (1.09) (0.42) Operational Inefficiency 0.188** 0.178** 0.215*** 0.180** 0.204** 0.185** 0.120 0.198** 0.110 0.163* (2.34) (2.18) (2.60) (2.25) (2.53) (2.35) (1.34) (2.46) (1.01) (1.72) Investment Inefficiency 1.802 1.799 2.314* 1.770 2.228* 2.245* 3.040** 2.229* 3.128* 1.857 (1.37) (1.35) (1.74) (1.36) (1.71) (1.73) (2.10) (1.72) (1.80) (1.27) Corruption Postings 2.033* 2.257** 2.904** 1.526 2.213* 2.252** 1.823 2.027* 1.696 2.036 (1.90) (2.07) (2.57) (1.37) (1.93) (2.09) (1.62) (1.87) (1.25) (1.57) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes # Obs 238 238 238 238 238 238 204 238 204 238

57 Table 7 Corruption Measures for All Firms: 2005-2015 This table presents annual averages of corruption measures for all Chinese listed firms from 2005 to 2015. Panel A presents the sample of all firms listed on Shanghai and Shenzhen stock exchanges (A shares). Panel B present the subsample of SOEs. All Chinese firms’ fiscal years end in December so the fiscal year coincides with the calendar year. The firm-level corruption indicators include: 1) CEO comp. (CEO compensation), scaled by total assets; 2) CEO pay-perform. (CEO pay-for-performance sensitivity); 3) Related sales (Related-party sales, scaled by revenue); 4) Related loans (Related-party loans, scaled by total assets); 5) Other receiv. (Other receivables from parent firm, scaled by total assets); 6) Reg. breaches (Number of regulation breaches in a year); 7) Entertain. Exp. (Business entertainment expenditures, scaled by total assets); 8) Operational Inefficiency, calculated as growth of sales minus growth of net income; and 9) Inv. inefficiency (Investment inefficiency), calculated as the absolute value of residue from regression of investment on sales growth within each industry-year. Operational inefficiency is Winsorized at 5% and 95% for each year because of the large number of outliers. All the other firm-level corruption measures, except number of regulation breaches, are Winsorized at 1% and 99% for each year. We exclude financial firms for these measures: related-party sales, related-party loans, other receivables from parent firm, operational inefficiency, and investment inefficiency. We also report differences between 2011 and the average of 2013-2015 (after the start of anti-corruption campaign), as well as the associated t-statistics calculated by clustering at the firm level. %Diff. refers to the percentage change from before the anti-corruption campaign (2011). To ease reading, CEO compensation, other receivables from parent firm, and business entertainment expenditures are expressed in percentage. Bold is used for numbers that are statistically significant at 0.10 level. Panel A: All Chinese Listed Firms Corruption Measures Year CEO CEO Pay- Related Related Other Reg. Entertain. Operational Inv. Comp. (%) Perform. Sales Loans Receiv. (%) Breaches Exp. (%) Inefficiency Inefficiency 2005 0.016 0.022 0.067 0.022 0.946 0.058 0.284 0.947 0.086 2006 0.018 0.029 0.060 0.015 0.604 0.052 0.278 0.293 0.095 2007 0.023 0.056 0.060 0.013 0.135 0.057 0.280 -0.087 0.121 2008 0.026 0.060 0.061 0.011 0.039 0.061 0.271 0.924 0.153 2009 0.028 0.094 0.051 0.008 0.018 0.103 0.288 0.301 0.125 2010 0.031 0.114 0.044 0.004 0.007 0.095 0.294 0.196 0.109 2011 0.029 0.167 0.038 0.001 0.010 0.121 0.293 0.300 0.193 2012 0.028 0.201 0.037 0.001 0.010 0.198 0.290 0.428 0.138 2013 0.028 0.228 0.039 0.002 0.008 0.225 0.284 0.415 0.178 2014 0.028 0.229 0.039 0.004 0.003 0.202 0.232 0.345 0.097 2015 0.028 0.250 0.040 0.003 0.000 0.208 0.211 0.595 0.121 2013~2015 - 2011 Diff. -0.001 0.069 0.001 0.001 -0.006 0.090 -0.051 0.154 -0.062 t-stat (-2.03) (10.94) (0.52) (6.33) (-4.34) (9.15) (-8.48) (5.52) (-11.43) % Diff. -4.92% 40.98% 2.26% 104.14% -64.05% 74.78% -17.37% 51.52% -32.14%

58 Panel B: SOEs Corruption Measures Year CEO CEO Pay- Related Related Other Reg. Entertain Operational Inv. Comp. (%) Perform. Sales Loans Receiv. (%) Breaches Exp. (%) Inefficiency Inefficiency 2005 0.013 0.008 0.077 0.020 0.957 0.037 0.263 0.843 0.085 2006 0.015 0.014 0.073 0.014 0.609 0.033 0.262 0.226 0.090 2007 0.018 0.026 0.075 0.014 0.121 0.042 0.233 -0.124 0.117 2008 0.018 0.013 0.079 0.011 0.043 0.048 0.251 0.936 0.150 2009 0.018 0.031 0.071 0.009 0.021 0.081 0.239 0.340 0.125 2010 0.020 0.025 0.069 0.005 0.010 0.074 0.238 0.183 0.106 2011 0.018 0.024 0.065 0.002 0.015 0.095 0.261 0.311 0.206 2012 0.017 0.027 0.063 0.002 0.015 0.170 0.249 0.461 0.139 2013 0.015 0.031 0.068 0.002 0.012 0.187 0.213 0.435 0.180 2014 0.015 0.033 0.068 0.005 0.006 0.176 0.156 0.409 0.101 2015 0.014 0.041 0.074 0.005 0.000 0.186 0.127 0.762 0.139 2013~2015 - 2011 Diff. -0.003 0.011 0.005 0.002 -0.009 0.088 -0.094 0.225 -0.066 t-stat (-3.82) (3.03) (1.51) (5.25) (-3.37) (6.38) (-11.76) (4.92) (-8.78) % Diff. -15.86% 48.00% 7.33% 119.07% -60.67% 91.74% -36.20% 72.37% -32.10%

59 Table 8 Stock Returns of Event Firms and Related Firms on and after Corruption Investigation Events This table presents event firms’ short-term returns around corruption investigation events and long- term returns after events. The event firms include 150 Chinese listed firms with corrupt managers investigated from the beginning of the anti-corruption campaign on December 4, 2012 to December 31, 2015. We calculate cumulative abnormal returns (CARs) for all firms in the [-1,+1] and the [-1,+15] windows, where day 0 is the date of investigation announcement. We first calculate daily abnormal return using one of the three approaches: 1) Daily return in excess of market return; 2) Size-adjusted return by subtracting return of the firm’s size decile portfolio; 3) DGTW-adjusted return by subtracting the value-weighted return of the firm’s benchmark portfolio of size, book-to-market ratio, and past return. For a specific window, we calculate daily abnormal return using the calendar time approach, and then multiply the daily abnormal return by the number of days in the window to calculate cumulative abnormal return. T-statistics associated with returns are also reported. Panel A reports abnormal returns for all event firms and non-event firms that are related to event firms, including customers of event firms, suppliers of event firms, firms in the same province as even firms, and firms with the largest overlap of C-suite managers with event firms. Panel B reports the abnormal returns for firms in the top or the bottom quintile of the propensity score of investigations. Bold is used for numbers that are statistically significant at 0.10 level. Panel A: Abnormal Returns of Event Firms and Connected Firms Market-Adj. Ret. Size-Adj. Ret. DGTW-Adj. Ret. Return t-stat Return t-stat Return t-stat Event Firms CAR [-1,+1] -1.55% (-2.63) -1.41% (-2.64) -1.42% (-2.77) CAR [-1, +15] -1.28% (-0.83) -1.33% (-0.92) -1.39% (-1.05) CAR [-15, +125] -6.91% (-1.49) -3.67% (-0.90) -2.26% (-0.68) CAR [-15,+250] -0.23% (-0.03) 4.68% (0.58) 2.77% (0.39) Panel B: Abnormal Returns of Subsamples of Firms Customers of Event Firms CAR [-1,+1] -1.46% (-2.13) -1.39% (-2.18) -1.01% (-1.60) CAR [-1, +15] -1.72% (-1.13) -1.51% (-1.06) -1.24% (-0.90) Suppliers of Event Firms CAR [-1,+1] 1.07% (0.68) 1.09% (0.67) 1.81% (1.28) CAR [-1, +15] -2.55% (-0.81) -0.77% (-0.25) -0.58% (-0.23) Firms in the Same Province CAR [-1,+1] 0.05% (0.36) 0.07% (0.67) 0.06% (0.65) CAR [-1, +15] 0.77% (1.78) 0.53% (1.64) 0.54% (1.97) Firms with Managerial Connections CAR [-1,+1] -0.35% (-0.83) -0.56% (-1.36) -0.34% (-0.93) CAR [-1, +15] 2.24% (1.59) 2.09% (1.48) 2.16% (1.64) Top Quintile of Propensity Scores of Investigation CAR [-1,+1] 0.01% (0.04) -0.05% (-0.49) -0.06% (-0.99) CAR [-1, +15] -0.75% (-2.04) -0.12% (-0.35) -0.22% (-1.16)

60

Panel B: Abnormal Returns of Subsamples of Firms Bottom Quintile of Propensity Scores of Investigation CAR [-1,+1] -0.19% (-0.70) -0.05% (-0.32) 0.00% (0.03) CAR [-1, +15] 2.23% (2.50) 1.12% (2.18) 0.91% (2.63)

61 Table 9 Accounting Manipulation and Financial Market Environment: 2005-2015 This table presents annual averages of variables on accounting manipulation and financial market environment for all Chinese listed firms from 2005 to 2015. The sample includes all firms listed on Shanghai and Shenzhen stock exchanges (A shares). All Chinese firms’ fiscal years end in December so the fiscal year coincides with the calendar year. The variables of interest include: 1) Abs.(DACC) (Absolute value of discretionary accruals, scaled by total assets); 2) Standardized difference of small profit and standardized difference of small loss, which measure earnings discontinuity; and 3) Normalized volatility and differenced volatility around earnings announcement. All the variables are Winsorized at 1% and 99% for each year. We exclude financial firms for the measure of absolute value of discretionary accruals. We also report differences between 2011 (before anti-corruption campaign) and the average of 2013- 2015 (after the start of anti-corruption campaign), as well as the associated t-statistics calculated by clustering at the firm level. Bold is used for numbers that are statistically significant at 0.10 level. Earnings Return Volatility around Discontinuity Earnings Announcement Abs. Small Small Normalized Differenced Year (DACC) Profit Loss Volatility Volatility 2005 0.063 0.069 -0.055 0.153 0.261 2006 0.058 0.035 -0.037 0.065 0.116 2007 0.065 0.033 -0.034 0.256 0.493 2008 0.067 0.138 -0.123 0.180 0.328 2009 0.066 0.047 -0.039 0.119 0.177 2010 0.056 0.059 -0.070 0.272 0.361 2011 0.054 0.064 -0.077 0.331 0.429 2012 0.048 0.068 -0.062 0.107 0.123 2013 0.047 0.057 -0.053 0.263 0.345 2014 0.046 0.079 -0.068 0.050 0.114 2015 0.046 0.086 -0.075 0.100 0.137 2013~2015 - 2011 Diff. -0.008 0.011 0.012 -0.194 -0.232 t-stat (-6.34) (0.85) (1.44) (-11.76) (-9.44)

62 Table 10 Abnormal Corruption Measures for All Chinese Firms in 2005-2015: Benchmarked to Hong Kong Firms This table presents annual averages of abnormal corruption measures for all Chinese listed firms from 2005 to 2015. The sample includes all firms listed on Shanghai and Shenzhen stock exchanges (A shares). For each Chinese firm, we identify a matched firm using the propensity score matching based on the corruption measures. Abnormal corruption measures are calculated as the differences between the Chinese firms and matched Hong Kong firms. The firm-level corruption measures include: 1) Standardized difference of small profit and standardized difference of small loss, which measure earnings discontinuity; 2) Abs.(DACC) (Absolute value of discretionary accruals, scaled by total assets); 3) # Regulation breaches (Number of regulation breaches in a year); 4) Operational Inefficiency, calculated as growth of sales minus growth of net income; 5) Investment inefficiency, calculated as the absolute value of residue from regression of investment on sales growth within each industry-year; 6) Normalized volatility and differenced volatility around earnings announcement. Absolute value of discretionary accruals, operational inefficiency, and profit margin are Winsorized at 5% and 95% for each year because of the large number of outliers. All the other firm-level corruption measures, except number of regulation breaches, are Winsorized at the 1% and 99% levels for each year. We exclude financial firms for four measures: absolute value of discretionary accruals, operational inefficiency, profit margin, and investment inefficiency. We also report differences between 2011 (before anti-corruption campaign) and the average of 2013-2015 (after the start of anti-corruption campaign), as well as the associated t-statistics. Bold is used for numbers that are statistically significant at 0.10 level. Earnings Return Volatility around Discontinuity Earnings Announcement Small Small Abs. # Regulation Operational Investment Normalized Differenced Year Profit Loss (DACC) Breaches Inefficiency Inefficiency Volatility Volatility 2005 0.059 -0.045 -0.034 0.049 0.863 -0.059 -0.212 -0.005 2006 0.035 -0.032 -0.014 0.043 0.325 -0.038 -0.018 0.000 2007 0.053 -0.041 -0.024 0.045 -0.080 -0.027 0.033 0.000 2008 0.124 -0.113 -0.032 0.050 -0.083 0.047 -0.175 -0.004 2009 0.027 -0.031 -0.021 0.107 0.685 0.037 -0.218 -0.004 2010 0.065 -0.064 -0.021 0.106 0.242 -0.062 -0.098 -0.002 2011 0.100 -0.066 -0.020 0.130 -0.066 0.043 -0.119 -0.004 2012 0.092 -0.064 -0.023 0.206 0.259 0.073 -0.247 -0.005 2013 0.072 -0.051 -0.014 0.223 0.065 0.107 -0.099 -0.003 2014 0.069 -0.052 -0.017 0.217 0.203 0.028 -0.158 -0.002 2015 0.100 -0.102 -0.013 0.226 0.262 0.058 -0.255 -0.004 2013~2015 - 2011 Diff. -0.029 0.015 0.005 0.092 0.248 0.020 -0.046 0.001 t-stat (-1.90) (1.32) (2.11) (5.81) (4.17) (3.63) (-1.66) (1.63)

63

Internet Appendix

Is the Anti-Corruption Campaign Effective at Reducing Corporate Corruption in China?

February 2018

1 Section A.1 Content of Eight-point Regulation The content of Eight-point Regulation includes the following [China Daily (2012)]. 1. Leaders must maintain close contact with the grassroots. They must understand the real situation facing society through in-depth visits at the grassroots level. Greater attention should be focused on places where social problems are more acute, and inspection tours must be carried out more thoroughly. Inspection tours which are a mere formality should be strictly prohibited. Leaders should work and listen to the public and lower level officials; the most practical problems facing ordinary people must be tackled. For official visits, there should be no welcome banner, no red carpet, no floral arrangement or grand receptions for officials. 2. Meetings and major events should be strictly regulated, and their efficiency improved. Politburo members are not allowed to attend ribbon-cutting or cornerstone-laying ceremonies, or celebrations and seminars, unless they get approval from the Central Committee. Official meetings should be shortened, be specific and to-the-point, and be free of empty-talk and blather. 3. The issuing of official documents should be reduced. 4. Officials’ visits to foreign countries should only be arranged when absolutely necessary, with fewer accompanying members; on most occasions, there is no need to mobilize a reception by Chinese expatriates, institutions and students at the airport. 5. There should be fewer traffic controls when leaders travel by car to avoid unnecessary inconvenience to the public. 6. The media should seek to reduce the number of news reports related to members of the Politburo, their work and their activities. The media should also seek to reduce the amount of time spent on these news pieces and minimize their scope. Such stories should only be reported depending on work needs, news value, and potential social impact. 7. Leaders should not publish any works by themselves or issue any congratulatory letters in their own name unless an arrangement has been made with the central authorities. Official documents without much meaningful content and without much actual importance should be withheld. Publications dedicated to senior officials' work and activities are also restricted. 8. Leaders must practice thrift and strictly follow relevant regulations on accommodation and cars.

2 Section A.2 List of 34 Key Words for News Search

Below is the list of 34 key words used for news searches described in Section 2.1 of the paper. English translation is provided next to the original Chinese key words. Note that the structure of Chinese language differs from English, so some key words in English may sound redundant but are not so in Chinese. For example, “investigated” (key word #4) is a substring of “investigated by party organizations” (key word #1), but the corresponding Chinese key word 接受调查 is not a substring of 接受组织调查 .

1 接受组织调查 Investigated by party organizations 2 接受检察部门调查 Investigated by prosecuting department 3 接受有关部门调查 Investigation by corresponding department 4 接受调查 Investigated 5 公安机关调查 Investigated by police department 6 纪检机关调查 Investigated by discipline inspection department 7 涉嫌受贿 Suspected of receiving bribes 8 涉嫌严重违纪 Suspected of severe disciplinary violations 9 涉嫌个人违纪问题 Suspected of personal disciplinary violations 10 涉嫌经济问题 Suspected of monetary issues 11 涉嫌违纪 Suspected of disciplinary violations 被检察机关批准执 12 Arrested with the approval of prosecuting department 行逮捕 被检查机关带走接 13 Taken away by investigation department to be investigated 受调查 被检察机关带走接 14 Taken away by prosecuting department to be investigated 受调查 15 被警方控制 Under police control 16 被立案侦查 Being filed a case for investigation 17 被带走 Taken away 18 被拘 Detained 采取认定为不适当 19 Considered an inappropriate candidate 人选措施 20 刑事拘留 Criminal detention 21 拘留审查 Detained for examination 22 拘留调查 Detained for investigation “Shuanggui” (a disciplinary measure taken by CPC that requires a party member 23 双规 to be investigated at a given time and a given place) 24 两规 Another name for “Shuanggui” 25 两指 “Liangzhi” (similar to “Shuanggui” but applied to non-Party members) 26 逮捕 Arrested 27 批捕 Approval of arrest 28 失去人身自由 Lose personal freedom 29 投案自首 Surrender oneself 30 未能取得联系 Unable to contact 3 31 逃至国外 Escape abroad 32 跑路 Escape 33 执行监视居住 Under residential surveillance 34 关于媒体报道 Regarding the media report

4 Section A.3 Additional Description of Corruption Measures 1. Monitoring by minority shareholders We follow Chen, Chen, Schipper, Xu and Xue (2012) and construct the measure as the total share ownership of the 2nd to 5th largest shareholders (scaled by total shares outstanding) multiplied by the Herfindahl index for the concentration of ownership among these shareholders. The share ownership data are obtained from CSMAR’s corporate governance database. 2. CEO compensation We calculate CEO compensation as the sum of salary, bonus, value of granted restricted stocks, options and appreciation rights. The valuation of granted options, restricted stocks and appreciation rights generally follow the procedures in the U.S. litereature [e.g. Core and Guay (2002)] except that instead of counting total option value towards compensation of grant year, we track how many units are vested or forfeited in the each of following years, calculate the value of vested amount in the vesting year, and count it to the compensation of vesting year. We adopt this method for our sample of Chinse firms because: 1) Usually, Chinese companies make one-time option grant to executives as a lump-sum compensation for the next three to ten years(the length of vesting period). Therefore counting the total value of option grants towards only the grant year would overestimate the compensation in the grant year and underestimate the compensation in the following non-grant years. 2) Chinese option grants almost all have performance-vesting provisions, which are applied to each vesting period (usually, year-by-year) in addition to time-based vesting provision. So the value the manager can collect depends on firm performance over time. Calculating the option compensation using the total grant amount in the grant year would ignore the performance-vesting provisions and very likely inflate the compensation because of the forfeit provision. The salary, bonus and option grant data are obtained from CSMAR’s corporate governance database. 3. CEO pay-for-performance sensitivity Following Bergstresser and Philippon (2006), we first calculate the dollar change in the value of a CEO’s stock and option holdings in response to a one percentage increase in the company stock price using the Black-Scholes formula and following the procedures in Core and Guay (2002) (referred to as “OnePct”), and then normalize it with respect to the sum of OnePct, CEO salary and bonus, i.e. Pay-for-Performance Sensitiviey = OnePct / (OnePct + Salary +

5 Bonus). CEO’s stock and option holdings data are obtained from CSMAR’s corporate governance database and WIND database. 4. Related-party sales Related-party sales for a firm are obtained from CSMAR’s related party transaction database at the transaction level in the annual report, and then aggregated and scaled by revenue (REV). 5. Related-party loans We obtain related-party loans from WIND database (available from annual report) and scale by total assets (AT). 6. Other receivables from parent We obtain data on other receivables from the parent firm WIND database (available from annual report) and scale by total assets (AT). 7. Regulation Breaches Below is a complete list of the categories of regulation breaches, with English translation. We exclude the type of “non-material accounting errors” (P2515) because they are associated with common accounting mistakes which are unlikely to be asscicated with corruption. P2501=虚构利润 Fake profit P2502=虚列资产 Fake assets P2503=虚假记载(误导性陈述) Fake record (misleading description) P2504=推迟披露 Delayed disclosure P2505=重大遗漏 Important missing items P2506=披露不实(其它) False disclosure (other) P2507=欺诈上市 Cheating for IPO P2508=出资违规 Illegal fund investment P2509=擅自改变资金用途 Change in uses of funds without permission. P2510=占用公司资产 Embezzle corporate assets. P2511=内幕交易 Trading on inside information. P2512=违规买卖股票 Illegal stock trading. P2513=操纵股价 Manipulating stock prices. P2514=违规担保 Illegal guarantee P2515=一般会计处理不当 Non-material accounting errors P2599=其他 Other 8. Business Entertainment Expenditures We collect the data of business entertainment expenditures from the footnotes of firms’ financial statements using a Python program. The item could be reported under three sections: “management expenses” and “sales expenses” in the income statement, and “other cash payments for the expenses related to operating activities” in the cash flow statement. We follow the literature [Ou-Yang, Shu, and Wong (2015)] and construct the BEE measure as 6 follows. First, if BEE is disclosed under both sections of “management expenses” and “sales expenses” in the income statement, we take their sum as BEE. Second, if BEE is only disclosed in either one of expenses accounts or “other cash payments” account, we take the reported BEE as the total BEE. Third, if BEE is disclosed only in the “other cash payments” section in the cash flow statement, and one of the expense accounts in the income statement, we take the larger amount as BEE. 9. Operational Inefficiency We calculate operational inefficiency as the sales growth rate minus the net income growth

, , rate: , = . 𝑅𝑅𝑅𝑅𝑅𝑅,𝑖𝑖 𝑡𝑡 𝑁𝑁𝑁𝑁,𝑖𝑖 𝑡𝑡 10. Investment𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 Inefficiency𝑖𝑖 𝑡𝑡 𝑅𝑅𝑅𝑅𝑉𝑉𝑖𝑖 𝑡𝑡− 1 − 𝑁𝑁𝐼𝐼𝑖𝑖 𝑡𝑡−1 We follow Biddle, Hilary, and Verdi (2009) and measure investment inefficiency as the absolute value of residue from a regression of firm’s investment on sales growth within each industry-year. A firm’s investment is calculated as change in gross property, plant and equipment plus change in inventories, scaled by lagged total assets. 11. Corruption Postings We manually read a sample of the “Guba” posts and find that posters normally use simple and casual language to discuss corruption rather than the formal language in the list of 34 key words in Section A.2. Therefore, we construct a list of 7 keywords based on our manual reading of the subsample of “Guba” posts that discuss corruption. we identify a “Guba” post as discussing corruption if its title contains one of seven keywords below.

1 腐败 Corrupt 2 腐化 Corrupt 3 贪污 Embezzlement 4 反腐 Anti-Corruption 5 受贿 Receiving Bribes 6 行贿 Bribing Others 7 中纪委 Central Commission for Discipline Inspection of the CPC 12. Discretionary accruals We follow the literature and construct discretionary accruals using annual accounting variables. Specifically, we first define total accrual as the difference between net income (NI) and cash flows from operating activities (CFO), divided by total assets (AT). Next, we use the modified Jones’ (1991) model for each industry-year.

, 1 , , = + + + , , , , , 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖 𝑡𝑡 𝛥𝛥𝛥𝛥𝛥𝛥𝑣𝑣𝑖𝑖 𝑡𝑡 𝑃𝑃𝑃𝑃𝐸𝐸𝑖𝑖 𝑡𝑡 𝑎𝑎1 𝑎𝑎2 𝑎𝑎3 𝜀𝜀𝑖𝑖 𝑡𝑡 𝐴𝐴𝑇𝑇𝑖𝑖 𝑡𝑡 𝐴𝐴𝑇𝑇𝑖𝑖 𝑡𝑡 7 𝐴𝐴𝑇𝑇𝑖𝑖 𝑡𝑡 𝐴𝐴𝑇𝑇𝑖𝑖 𝑡𝑡 where ∆REV is change in revenue, and PPE is the gross property, plant, and equipment. Discretionary accruals (DACC) is residual from the regression. Since discretionary accruals reverse over time, we follow the literature and use the absolute value of discretionary accruals as a measure of accounting manipulation. 13. Standardized difference of small profit and small loss Standardized difference of small profit (small loss) equals the difference between the actual and expected number of firms in the small profit interval (small loss interval), divided by the difference’s estimated standard deviation. We follow the literature [Beaver, McNichols and Nelson (2007)] and calculate the expected number of firms in an interval as the average of the two immediately adjacent intervals, and variance as (1 ) + ( + )(2 1 + ) where is the sum of the number of firms 𝑁𝑁and𝑝𝑝𝑖𝑖 − is𝑝𝑝 𝑖𝑖the probability�4� 𝑁𝑁 𝑝𝑝𝑖𝑖−1 that𝑝𝑝𝑖𝑖+ 1a firm−

𝑝𝑝falls𝑖𝑖−1 in 𝑝𝑝interval𝑖𝑖+1 . 𝑁𝑁 𝑝𝑝𝑖𝑖 14. Abnormal stock𝑖𝑖 return volatility around earnings announcement We consider annual earnings annoucements for all Chinese listed firms and construct two measures of volatility around earnings announcement. We first define stock return volatility in a window as the mean of absolute daily abnormal return (in excess of market return) in this window, and calculate normalized volatility as the return volatility during the 4-day window [- 1, +2] divided by the return volatility during the [-56, -2] window (55 days before the announcement window) and the [+3, +57] window (55 days after the announcement window). Day 0 refers to the earnings announcement day. We further calculate differenced volatility as the return volatility during the 4-day window [-1, +2] minus the return volatility during the [- 56, -2] window (55 days before the announcement window) and the [+3, +57] window (55 days after the announcement window). We require at least four days of consecutive trading around the announcement to calculate the volatility measures.

8 Section A.4. Additional Results on the Corruption Measures of Sample of Event Firms Prior to Investigations

As discussed in Section 3.1, our approach of identifying matched firms is unable to find matched firms for five event firms because no other firms in the same industry have close enough market capitalization. As a result, we relax the size requirement and identify matched firms as those having the closest market capitalization among firms in the same industry and with the same SOE status. We conduct robustness tests by excluding these five event firms, and Tables IA.1 to IA.3 correspond to all the event analyses in the paper.

9 Table IA.1 Corruption Measures of Event Firms before Corruption Investigations This table corresponds to Table 2 in the paper. Year t-1 Year t Event Match Event Match Firms Firms Diff t-stat Firms Firms Diff t-stat Monitoring 0.425 0.679 -0.254 (-1.98) 0.412 0.694 -0.282 (-2.25) Compensation (%) 0.015 0.012 0.003 (1.43) 0.014 0.014 0.000 (-0.13) Pay for Performance 0.014 0.062 -0.049 (-2.68) 0.026 0.074 -0.048 (-2.36) Near-Retirement 0.276 0.159 0.117 (2.53) 0.326 0.160 0.167 (3.14) Related-Party Sales 0.101 0.066 0.035 (1.88) 0.111 0.059 0.052 (2.85) Related-Party Loans 0.006 0.002 0.003 (1.59) 0.008 0.005 0.004 (1.32) Other Receivables (%) 0.009 0.002 0.007 (1.72) 0.003 0.001 0.001 (0.75) Regulation Breaches 0.159 0.152 0.007 (0.15) 0.214 0.124 0.090 (1.51) Entertain. Exp. (%) 0.211 0.189 0.022 (0.44) 0.127 0.169 -0.041 (-1.29) Operational Inefficiency 0.518 0.107 0.411 (2.54) 0.913 0.584 0.329 (1.34) Invest. Inefficiency 0.140 0.129 0.011 (1.27) 0.151 0.138 0.013 (1.16) Corruption Postings 0.082 0.058 0.025 (2.75) 0.133 0.071 0.062 (5.31)

10 Table IA.2 Probit Regressions of Investigation on Corruption Measures This table corresponds to Table 3 in the paper. Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) Monitoring -0.170** -0.250** -0.224** (-2.25) (-2.07) (-2.43) CEO Compensation 0.380 0.192 0.513 (1.05) (0.47) (1.33) CEO Pay for Performance -1.001* -1.164* -1.062* (-1.68) (-1.72) (-1.67) CEO Near-Retirement Dummy 0.508** 0.681*** 0.629*** (2.57) (2.60) (2.92) Related-Party Sales 0.935* 1.201* 1.163** (1.76) (1.78) (2.06) Related-Party Loans 8.945* 10.437* 12.134** (1.74) (1.75) (2.19) Other Receivables from Parent 1.468** 1.402* 1.590** (2.15) (1.76) (2.25) # Regulation Breaches 0.025 0.083 0.105 (0.12) (0.36) (0.50) Entertainment Expenditures -0.005 -0.005 (-0.14) (-0.13) Operational Inefficiency 0.173*** 0.165* 0.189** (2.59) (1.87) (2.51) Investment Inefficiency 2.023* 1.249 1.806 (1.82) (0.91) (1.51) Corruption Postings 1.657* 1.539 2.204** (1.78) (1.07) (2.06) Ln(ME) 0.139 0.175** 0.177 0.158* 0.119 0.174 0.151 (1.53) (1.96) (1.62) (1.76) (1.41) (1.28) (1.43) SOE Dummy -0.233 -0.264 -0.264 -0.233 -0.180 -0.817** -0.531* (-0.87) (-1.04) (-0.97) (-0.89) (-0.74) (-2.35) (-1.75) Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes # Obs 283 276 210 271 290 194 264

11 Table IA.3 Probit Regressions of Investigation on Political Connection Measures This table corresponds to Table 5 in the paper.

Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Government Connection 0.714*** (3.33) Local Government Connection 0.927*** 1.136*** 0.868*** (3.88) (3.75) (3.37) Central Government Connection -0.054 -0.025 0.314 (-0.14) (-0.05) (0.68) University Affiliation: Top 7 Leaders -1.626*** -2.286*** -2.619*** (-3.76) (-3.17) (-3.77) Birth Conn: Top 7 Leaders -0.015 -0.055 -0.035 (-0.04) (-0.11) (-0.08) Association with Zhou Yongkang 0.971** 1.927*** 1.950*** (2.10) (2.66) (2.90) Univ. Conn: Invest. National Leaders 0.810* 0.084 0.385 (1.89) (0.15) (0.75) Birth Conn: Invest. National Leaders 0.507 0.529 0.914* (1.21) (0.95) (1.90) Univ. Conn: Non-Invest. National Leaders -0.312** 0.333 0.349 (-2.24) (1.33) (1.49) Birth Conn: Non-Invest. National Leaders -0.369** -0.802*** -0.542** (-1.97) (-2.89) (-2.28) Birth Conn: 1,021 Invest. Politicians -0.152 0.014 -0.165 (-1.09) (0.07) (-0.88) Province Graft-Tigers 0.347* 0.384 (1.67) (1.41) Province Graft-Flies -0.848 -1.187 (-0.65) (-0.79) Investigation Team in Province 0.473** 0.651** 0.437 (1.97) (2.13) (1.61)

12 Dependent Variable: Dummy of Corruption Investigation Monitoring -0.214** -0.209** -0.186* -0.250*** -0.242** -0.218** -0.179* -0.221** -0.167 -0.240** (-2.30) (-2.23) (-1.88) (-2.62) (-2.50) (-2.34) (-1.78) (-2.39) (-1.30) (-2.13) CEO Compensation 0.500 0.529 0.622 0.539 0.568 0.446 0.257 0.392 0.308 0.474 (1.31) (1.38) (1.57) (1.40) (1.46) (1.16) (0.61) (1.02) (0.73) (1.22) CEO Pay for Performance * * * * -1.063 -1.091 -0.896 -0.985 -0.982 -1.063 -1.098 -1.014 -0.920 -0.758 (-1.67) (-1.69) (-1.28) (-1.55) (-1.45) (-1.67) (-1.68) (-1.60) (-1.21) (-1.05) CEO Near-Retirement Dummy * ** *** *** *** *** *** *** ** ** 0.437 0.511 0.642 0.648 0.644 0.649 0.682 0.631 0.666 0.510 (1.93) (2.21) (2.89) (2.99) (2.93) (2.99) (2.91) (2.91) (2.34) (2.02) Related-Party Sales 0.999* 1.044* 0.979* 1.144** 1.350** 1.109* 1.403** 1.178** 0.565 0.346 (1.74) (1.79) (1.67) (1.96) (2.23) (1.96) (2.35) (2.08) (0.73) (0.48) Related-Party Loans 10.955* 10.801* 16.646** 12.285** 13.473** 12.691** 13.051** 12.072** 18.594** 17.735** (1.89) (1.86) (2.51) (2.20) (2.36) (2.26) (2.20) (2.19) (2.24) (2.28) Other Receivables from Parent 1.371* 1.286* 1.879** 1.615** 1.894*** 1.585** 1.356* 1.451** 1.536* 1.652** (1.89) (1.79) (2.48) (2.26) (2.59) (2.22) (1.79) (2.04) (1.87) (1.98) # Regulation Breaches 0.111 0.141 0.075 0.099 0.081 0.115 0.245 0.097 0.325 0.164 (0.53) (0.66) (0.35) (0.47) (0.38) (0.55) (1.05) (0.46) (1.29) (0.73) Operational Inefficiency 0.184** 0.176** 0.196** 0.185** 0.201*** 0.185** 0.129 0.196** 0.094 0.158* (2.41) (2.27) (2.52) (2.43) (2.65) (2.46) (1.55) (2.54) (1.00) (1.83) Investment Inefficiency 1.603 1.653 1.920 1.531 1.851 1.847 2.202* 1.872 2.599* 1.747 (1.31) (1.33) (1.56) (1.27) (1.52) (1.53) (1.68) (1.55) (1.67) (1.30) Corruption Postings 2.055* 2.324** 2.584** 1.748 2.175* 2.316** 2.083* 2.104* 1.658 1.772 (1.93) (2.13) (2.34) (1.59) (1.93) (2.16) (1.87) (1.95) (1.25) (1.41) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes # Obs 264 264 264 264 264 264 228 264 228 264

13 Section A.5. Robustness Tests for the Regression Analyses: Alternative Fixed Effects

This subsection presents robustness tests for our main regression analyses in Tables 3 and 5 of the paper using alternative fixed effects, including year and industry-region fixed effects (Tables IA.4 and IA.5), industry and region-year fixed effects (Tables IA.6 and IA.7), and region and industry-year fixed effects (Tables IA.8 and IA.9).

14 Table IA.4 Probit Regressions of Corruption Investigation on Corruption Measures: Year and Industry-Region Fixed Effects This table is the same as Table 3 except that we control for year and industry-region fixed effects. Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) Monitoring -0.201*** -0.211* -0.262*** (-2.58) (-1.66) (-2.75) CEO Compensation 0.319 0.351 0.520 (0.84) (0.79) (1.29) CEO Pay for Performance -1.034 -1.165 -1.078 (-1.58) (-1.59) (-1.58) CEO Near-Retirement Dummy 0.641*** 0.858*** 0.825*** (3.04) (2.90) (3.57) Related-Party Sales 0.978* 0.994 1.176** (1.80) (1.40) (2.01) Related-Party Loans 8.053 14.862** 14.642** (1.45) (2.02) (2.34) Other Receivables from Parent 0.743 1.414* 1.052* (1.44) (1.68) (1.74) # Regulation Breaches 0.086 0.130 0.128 (0.37) (0.52) (0.57) Entertainment Expenditures -0.012 -0.021 (-0.31) (-0.51) Operational Inefficiency 0.177*** 0.165* 0.181** (2.60) (1.77) (2.37) Investment Inefficiency 1.778 1.821 1.999 (1.53) (1.21) (1.57) Corruption Postings 2.007** 1.874 2.426** (2.14) (1.19) (2.28) Ln(ME) 0.117 0.145* 0.104 0.139* 0.091 0.099 0.100 (1.36) (1.81) (0.94) (1.71) (1.15) (0.72) (1.02) SOE Dummy -0.178 -0.204 -0.155 -0.138 -0.148 -0.562 -0.371 (-0.65) (-0.80) (-0.57) (-0.53) (-0.60) (-1.60) (-1.23) Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Industry-Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes # Obs 273 267 193 263 279 182 257

15

Table IA.5 Probit Regressions of Corruption Investigation on Political Connection Measures: Year and Industry-Region Fixed Effects This table is the same as Table 5 except that we control for year and industry-region fixed effects.

Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Government Connection 0.789*** (3.55) Local Government Connection 1.033*** 1.210*** 0.922*** (4.07) (3.84) (3.36) Central Government Connection -0.019 -0.063 0.301 (-0.05) (-0.12) (0.62) University Affiliation: Top 7 Leaders -1.688*** -2.174*** -2.669*** (-4.09) (-3.03) (-3.87) Birth Conn: Top 7 Leaders 0.135 0.143 0.205 (0.34) (0.27) (0.44) Association with Zhou Yongkang 1.015** 1.591** 1.826*** (2.16) (2.21) (2.64) Univ. Conn: Invest. National Leaders 1.038** 0.209 0.487 (2.33) (0.38) (0.94) Birth Conn: Invest. National Leaders 0.515 0.566 0.823* (1.23) (0.96) (1.68) Univ. Conn: Non-Invest. National Leaders -0.354** 0.304 0.360 (-2.53) (1.18) (1.48) Birth Conn: Non-Invest. National Leaders -0.435** -0.877*** -0.653** (-2.14) (-2.93) (-2.50) Birth Conn: 1,021 Invest. Politicians -0.130 0.053 -0.104 (-0.92) (0.24) (-0.54) Province Graft-Tigers 0.356* 0.334 (1.71) (1.19) Province Graft-Flies -0.688 -0.594 (-0.51) (-0.39) Investigation Team in Province 0.494** 0.554* 0.411 (2.00) (1.73) (1.43) 16 Dependent Variable: Dummy of Corruption Investigation Monitoring -0.242** -0.232** -0.203* -0.291*** -0.283*** -0.257*** -0.219** -0.256*** -0.238* -0.235* (-2.48) (-2.36) (-1.92) (-2.93) (-2.76) (-2.67) (-2.12) (-2.68) (-1.76) (-1.86) CEO Compensation 0.497 0.531 0.682 0.565 0.582 0.456 0.266 0.394 0.330 0.591 (1.25) (1.33) (1.63) (1.40) (1.42) (1.13) (0.62) (0.99) (0.75) (1.41) CEO Pay for Performance * -1.093 -1.158 -0.920 -0.972 -1.002 -1.078 -1.070 -1.014 -1.155 -0.842 (-1.59) (-1.66) (-1.25) (-1.42) (-1.38) (-1.57) (-1.51) (-1.48) (-1.37) (-1.07) CEO Near-Retirement Dummy ** *** *** *** *** *** *** *** ** ** 0.617 0.663 0.772 0.855 0.806 0.842 0.833 0.812 0.719 0.608 (2.52) (2.68) (3.21) (3.63) (3.39) (3.63) (3.31) (3.48) (2.36) (2.20) Related-Party Sales 0.931 0.980 0.975 1.131* 1.379** 1.128* 1.373** 1.186** 0.309 0.099 (1.56) (1.62) (1.61) (1.87) (2.16) (1.92) (2.23) (2.02) (0.37) (0.12) Related-Party Loans 12.687* 11.625* 18.761** 15.075** 16.264** 15.009** 15.686** 14.410** 20.565** 17.881** (1.88) (1.73) (2.55) (2.38) (2.51) (2.39) (2.36) (2.31) (2.29) (2.05) Other Receivables from Parent 0.927 0.884 1.753** 1.126* 1.656** 1.038* 0.900 0.959 1.742* 1.921** (1.51) (1.45) (2.35) (1.87) (2.36) (1.70) (1.48) (1.59) (1.86) (2.07) # Regulation Breaches 0.076 0.071 0.065 0.126 0.129 0.135 0.282 0.121 0.255 0.096 (0.33) (0.31) (0.28) (0.56) (0.55) (0.60) (1.12) (0.54) (0.95) (0.40) Operational Inefficiency 0.181** 0.173** 0.178** 0.177** 0.190** 0.178** 0.145* 0.183** 0.101 0.147* (2.34) (2.19) (2.26) (2.30) (2.47) (2.33) (1.71) (2.36) (1.05) (1.66) Investment Inefficiency 1.856 1.855 1.976 1.712 1.927 1.998 2.519* 2.034 2.161 1.707 (1.43) (1.42) (1.50) (1.34) (1.48) (1.56) (1.76) (1.59) (1.30) (1.18) Corruption Postings 2.255** 2.510** 3.139*** 1.974* 2.715** 2.478** 2.199** 2.317** 2.406* 2.462* (2.13) (2.30) (2.83) (1.81) (2.37) (2.32) (1.98) (2.16) (1.82) (1.93) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry-Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes # Obs 257 257 257 257 257 257 224 257 224 257

17 Table IA.6 Probit Regressions of Corruption Investigation on Corruption Measures: Industry and Region-Year Fixed Effects This table is the same as Table 3 except that we control for industry and region-year fixed effects. Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) Monitoring -0.169** -0.207 -0.233** (-2.18) (-1.57) (-2.42) CEO Compensation 0.545 0.523 0.662* (1.62) (1.39) (1.86) CEO Pay for Performance -0.716 -0.901 -0.787 (-1.20) (-1.37) (-1.25) CEO Near-Retirement Dummy 0.564*** 0.713*** 0.750*** (2.78) (2.58) (3.33) Related-Party Sales 1.101** 1.418** 1.332** (2.05) (2.04) (2.32) Related-Party Loans 11.354** 14.208** 15.868*** (2.13) (2.06) (2.60) Other Receivables from Parent 0.543 1.135 0.882 (1.13) (1.37) (1.50) # Regulation Breaches 0.011 0.073 0.129 (0.05) (0.30) (0.60) Entertainment Expenditures -0.010 -0.036 (-0.24) (-0.75) Operational Inefficiency 0.158** 0.167* 0.164** (2.29) (1.76) (2.09) Investment Inefficiency 2.288** 1.487 2.262* (2.02) (1.03) (1.84) Corruption Postings 2.034** 1.557 2.590** (2.21) (1.04) (2.45) Ln(ME) 0.182** 0.194** 0.201* 0.180** 0.144* 0.177 0.144 (2.24) (2.41) (1.85) (2.23) (1.84) (1.36) (1.53) SOE Dummy -0.095 -0.164 -0.258 -0.165 -0.137 -0.672* -0.367 (-0.35) (-0.65) (-0.93) (-0.63) (-0.56) (-1.93) (-1.22) Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Year-Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes # Obs 290 283 212 275 297 194 268

18 Table IA.7 Probit Regressions of Corruption Investigation on Political Connection Measures: Industry and Region-Year Fixed Effects This table is the same as Table 5 except that we control for industry and region-year fixed effects.

Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Government Connection 0.718*** (3.25) Local Government Connection 0.935*** 1.106*** 0.827*** (3.83) (3.63) (3.15) Central Government Connection -0.110 -0.148 0.195 (-0.27) (-0.29) (0.42) University Affiliation: Top 7 Leaders -1.740*** -1.952*** -2.343*** (-4.21) (-2.79) (-3.50) Birth Conn: Top 7 Leaders 0.066 0.116 0.092 (0.17) (0.22) (0.20) Association with Zhou Yongkang 0.948** 1.837** 1.863*** (2.03) (2.42) (2.66) Univ. Conn: Invest. National Leaders 0.860* 0.030 0.372 (1.83) (0.05) (0.68) Birth Conn: Invest. National Leaders 0.445 0.475 0.738 (1.05) (0.82) (1.50) Univ. Conn: Non-Invest. National Leaders -0.397*** 0.240 0.272 (-2.74) (0.94) (1.13) Birth Conn: Non-Invest. National Leaders -0.406** -0.829*** -0.585** (-1.97) (-2.79) (-2.28) Birth Conn: 1,021 Invest. Politicians -0.076 0.048 -0.100 (-0.52) (0.22) (-0.51) Province Graft-Tigers 0.369* 0.400 (1.68) (1.42) Province Graft-Flies -0.464 -0.635 (-0.34) (-0.40) Investigation Team in Province 0.386 0.507 0.311 (1.54) (1.56) (1.07)

19 Dependent Variable: Dummy of Corruption Investigation Monitoring -0.230** -0.227** -0.171 -0.257*** -0.228** -0.232** -0.179* -0.234** -0.166 -0.231** (-2.40) (-2.34) (-1.64) (-2.58) (-2.24) (-2.40) (-1.64) (-2.43) (-1.17) (-2.00) CEO Compensation * * ** * * * * * 0.658 0.680 0.728 0.673 0.688 0.639 0.265 0.593 0.402 0.658 (1.87) (1.92) (1.96) (1.89) (1.89) (1.79) (0.62) (1.68) (0.92) (1.79) CEO Pay for Performance -0.831 -0.849 -0.556 -0.742 -0.670 -0.797 -1.037 -0.731 -0.834 -0.512 (-1.32) (-1.34) (-0.79) (-1.18) (-0.97) (-1.27) (-1.54) (-1.17) (-1.07) (-0.72) CEO Near-Retirement Dummy ** *** *** *** *** *** *** *** ** ** 0.551 0.640 0.715 0.767 0.714 0.756 0.741 0.740 0.695 0.576 (2.33) (2.63) (3.08) (3.38) (3.10) (3.35) (3.01) (3.27) (2.36) (2.19) Related-Party Sales 1.174** 1.233** 1.151* 1.303** 1.563** 1.307** 1.479** 1.352** 0.782 0.616 (2.02) (2.09) (1.91) (2.20) (2.53) (2.26) (2.43) (2.35) (0.99) (0.84) Related-Party Loans 14.309** 14.651** 21.233*** 15.674** 17.417*** 16.162*** 16.508*** 15.509** 20.993** 19.086** (2.20) (2.21) (2.93) (2.55) (2.72) (2.63) (2.60) (2.56) (2.45) (2.31) Other Receivables from Parent 0.755 0.700 1.483** 0.952 1.403** 0.875 0.828 0.799 1.251 1.404* (1.26) (1.18) (2.08) (1.64) (2.05) (1.48) (1.40) (1.38) (1.61) (1.80) # Regulation Breaches 0.106 0.128 0.070 0.114 0.072 0.136 0.266 0.132 0.245 0.079 (0.50) (0.59) (0.32) (0.53) (0.33) (0.63) (1.12) (0.61) (0.95) (0.35) Operational Inefficiency 0.161** 0.151* 0.171** 0.164** 0.178** 0.161** 0.106 0.169** 0.077 0.143 (2.03) (1.88) (2.07) (2.06) (2.23) (2.06) (1.21) (2.13) (0.76) (1.60) Investment Inefficiency 1.922 1.965 2.483* 1.994 2.321* 2.274* 2.707** 2.298* 2.877* 1.928 (1.54) (1.56) (1.94) (1.61) (1.85) (1.85) (2.01) (1.86) (1.82) (1.41) Corruption Postings 2.427** 2.729** 3.300*** 2.224** 2.808** 2.632** 2.396** 2.507** 2.374* 2.588** (2.31) (2.52) (2.95) (2.05) (2.50) (2.48) (2.19) (2.36) (1.84) (2.10) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year-Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes # Obs 268 268 268 268 268 268 232 268 232 268

20 Table IA.8 Probit Regressions of Corruption Investigation on Corruption Measures: Region and Industry-Year Fixed Effects This table is the same as Table 3 except that we control for region and industry-year fixed effects. Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) Monitoring -0.193** -0.289** -0.248*** (-2.56) (-2.24) (-2.69) CEO Compensation 0.487 0.352 0.642* (1.41) (0.88) (1.75) CEO Pay for Performance -0.861 -0.903 -0.818 (-1.47) (-1.34) (-1.31) CEO Near-Retirement Dummy 0.523*** 0.763*** 0.678*** (2.63) (2.77) (3.12) Related-Party Sales 1.023* 1.318* 1.308** (1.92) (1.89) (2.30) Related-Party Loans 8.744* 12.644* 12.969** (1.69) (1.93) (2.27) Other Receivables from Parent 0.682 1.119 0.958* (1.39) (1.49) (1.67) # Regulation Breaches -0.036 0.051 0.121 (-0.16) (0.21) (0.57) Entertainment Expenditures -0.019 -0.018 (-0.49) (-0.42) Operational Inefficiency 0.163** 0.138 0.176** (2.41) (1.54) (2.34) Investment Inefficiency 2.929** 1.738 2.610* (2.28) (1.07) (1.86) Corruption Postings 1.869** 1.517 2.430** (2.07) (1.05) (2.38) Ln(ME) 0.155* 0.156** 0.249** 0.156** 0.109 0.236* 0.129 (1.87) (1.99) (2.19) (1.96) (1.40) (1.78) (1.38) SOE Dummy -0.249 -0.283 -0.286 -0.240 -0.228 -0.712** -0.458 (-0.92) (-1.14) (-1.04) (-0.94) (-0.95) (-1.98) (-1.52) Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes Industry-Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes # Obs 292 285 210 280 299 196 273

21 Table IA.9 Probit Regressions of Corruption Investigation on Political Connection Measures: : Region and Industry-Year Fixed Effects This table is the same as Table 5 except that we control for region and industry-year fixed effects.

Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Government Connection 0.786*** (3.59) Local Government Connection 0.983*** 1.118*** 0.869*** (4.08) (3.81) (3.40) Central Government Connection -0.002 -0.061 0.224 (-0.00) (-0.12) (0.48) University Affiliation: Top 7 Leaders -1.611*** -1.881*** -2.397*** (-4.04) (-2.69) (-3.60) Birth Conn: Top 7 Leaders 0.051 -0.046 0.076 (0.13) (-0.09) (0.17) Association with Zhou Yongkang 0.977** 1.721** 1.942*** (2.14) (2.38) (2.86) Univ. Conn: Invest. National Leaders 0.886** 0.088 0.365 (2.09) (0.17) (0.74) Birth Conn: Invest. National Leaders 0.462 0.480 0.820* (1.10) (0.86) (1.73) Univ. Conn: Non-Invest. National Leaders -0.346*** 0.234 0.299 (-2.61) (0.95) (1.29) Birth Conn: Non-Invest. National Leaders -0.376** -0.821*** -0.576** (-1.99) (-2.96) (-2.40) Birth Conn: 1,021 Invest. Politicians -0.154 0.068 -0.172 (-1.09) (0.32) (-0.92) Province Graft-Tigers 0.413** 0.360 (1.97) (1.34) Province Graft-Flies -1.043 -1.373 (-0.79) (-0.91) Investigation Team in Province 0.490** 0.635** 0.425 (2.04) (2.07) (1.57)

22 Dependent Variable: Dummy of Corruption Investigation Monitoring -0.230** -0.219** -0.190* -0.275*** -0.252*** -0.240*** -0.203** -0.243*** -0.177 -0.212* (-2.47) (-2.35) (-1.92) (-2.87) (-2.59) (-2.58) (-1.97) (-2.63) (-1.36) (-1.91) CEO Compensation * * * 0.581 0.602 0.735 0.664 0.667 0.571 0.291 0.550 0.345 0.518 (1.59) (1.64) (1.96) (1.82) (1.79) (1.54) (0.70) (1.52) (0.81) (1.37) CEO Pay for Performance -0.858 -0.894 -0.726 -0.735 -0.755 -0.819 -0.905 -0.746 -0.921 -0.664 (-1.35) (-1.39) (-1.06) (-1.17) (-1.12) (-1.31) (-1.33) (-1.20) (-1.17) (-0.93) CEO Near-Retirement Dummy * ** *** *** *** *** *** *** ** ** 0.438 0.521 0.661 0.690 0.665 0.707 0.685 0.676 0.653 0.548 (1.89) (2.19) (2.96) (3.14) (3.00) (3.22) (2.87) (3.08) (2.24) (2.11) Related-Party Sales 1.109* 1.153** 1.050* 1.289** 1.483** 1.260** 1.533** 1.318** 0.747 0.450 (1.92) (1.96) (1.77) (2.19) (2.43) (2.20) (2.56) (2.31) (0.96) (0.61) Related-Party Loans 11.709* 11.504* 16.998** 13.192** 14.129** 13.603** 14.431** 12.892** 19.015** 17.586** (1.92) (1.89) (2.54) (2.28) (2.41) (2.34) (2.28) (2.26) (2.29) (2.32) Other Receivables from Parent 0.798 0.738 1.560** 1.020* 1.442** 0.937 0.816 0.850 1.400* 1.438* (1.40) (1.31) (2.26) (1.81) (2.20) (1.62) (1.41) (1.49) (1.87) (1.92) # Regulation Breaches 0.093 0.118 0.057 0.109 0.094 0.140 0.245 0.114 0.215 0.097 (0.44) (0.55) (0.26) (0.52) (0.43) (0.66) (1.04) (0.54) (0.84) (0.43) Operational Inefficiency 0.177** 0.169** 0.193** 0.175** 0.193** 0.171** 0.118 0.184** 0.094 0.161* (2.33) (2.19) (2.45) (2.30) (2.52) (2.27) (1.40) (2.37) (1.00) (1.87) Investment Inefficiency 1.986 2.076 2.349 2.297 2.617* 2.593* 3.403** 2.760* 2.719 1.753 (1.39) (1.44) (1.62) (1.63) (1.82) (1.83) (2.16) (1.96) (1.47) (1.10) Corruption Postings 2.280** 2.542** 3.123*** 1.999* 2.608** 2.532** 2.338** 2.352** 2.454* 2.576** (2.24) (2.43) (2.91) (1.90) (2.41) (2.46) (2.19) (2.28) (1.93) (2.12) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry-Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes # Obs 273 273 273 273 273 273 234 273 234 273

23 Section A.6 Robustness Tests for the Regression Analysis: Alterantive Measures or Samples

This subsection presents robustness tests for the regression analyses using alternative measures or samples. Table IA.10 repeats the regression analysis in Table 3 but using the alternative definition of business entertainment expenditures that includes travel expenses. Table IA.11 repeats the regression analysis in Table 3 but using the alternative definitions of CEO compensation. Table IA.12 repeats the regression analysis in Table 5 but using the measures of university affiliation that includes the Party School of CPC. Table IA.13 repeats the regression analysis in Table 5 but using the alternative pairs of university affiliation using the 3rd & 4th, and the 5th & 6th top Chinese universities.

Table IA.10 Probit Regressions of Corruption Investigation: Alternative Definition of Business Entertainment Expenditures that Includes Travel Expenses Dep. Var.: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) Monitoring -0.238** (-2.26) CEO Compensation 0.523 (1.44) CEO Pay for Performance -0.856 (-1.37) CEO Near-Retirement Dummy 0.737*** (2.95) Related-Party Sales 1.141* (1.89) Related-Party Loans 11.901** (2.11) Other Receivables from Parent 1.247* (1.66) # Regulation Breaches 0.048 0.106 (0.23) (0.48) Entertainment Expenditures 0.014 0.009 (1.08) (0.66) Operational Inefficiency 0.152* (1.87) Investment Inefficiency 1.658 (1.27) Corruption Postings 2.477** (2.06) Ln(ME) 0.201** 0.188 (2.16) (1.64) SOE Dummy -0.172 -0.536* (-0.69) (-1.78) Year Fixed Effect Yes Yes Industry Fixed Effect Yes Yes Region Fixed Effect Yes Yes # Obs 242 223

24 Table IA.11 Probit Regressions of Corruption Investigation on Corruption Measures: Alternative Measures of CEO Compensation This table is the same as Table 3 except that we use alternative measures of CEO compensation. For these alternative measures, numerators are all the same as the original measure and the difference only lies in denominators. Specifically, the first alternative measure is scaled the CEO compensation by the firm’s operating profits. The second alternative measure is scaled it by the median CEO compensation of Non-SOEs in the same industry. The third alternative measure is scaled it by the average compensation of non-executive employees in the firm. Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) Monitoring -0.243** -0.266*** -0.269*** (-2.55) (-2.92) (-2.81) CEO Compensation: Scaled by Operating Profits 11.213* (1.83) CEO Compensation: Scaled by Non-SOEs 0.154** (2.04) CEO Compensation: Scaled by Employee Compensation 0.021 (1.52) CEO Pay for Performance -0.701 -0.862 -1.275* (-1.17) (-1.44) (-1.65) CEO Near-Retirement Dummy 0.573** 0.695*** 0.636*** (2.50) (3.27) (2.77) Related-Party Sales 1.289** 1.290** 1.558** (2.12) (2.28) (2.54) Related-Party Loans 10.744 12.525** 13.583** (1.62) (2.26) (2.41) Other Receivables from Parent 0.627 1.099* 1.017* (1.05) (1.86) (1.72) # Regulation Breaches 0.197 0.134 -0.082 (0.87) (0.65) (-0.34) Operational Inefficiency 0.273** 0.190** 0.173** (2.50) (2.53) (2.12) Investment Inefficiency 2.593** 1.891 2.381* (1.98) (1.58) (1.87) Corruption Postings 2.504** 2.672*** 2.920*** (2.25) (2.61) (2.69) Ln(ME) 0.160 0.013 0.052 (1.58) (0.13) (0.54) SOE Dummy -0.401 -0.479* -0.582** (-1.34) (-1.68) (-1.97) Year Fixed Effect Yes Yes Yes Industry Fixed Effect Yes Yes Yes Region Fixed Effect Yes Yes Yes # Obs 236 274 246

25 Table IA.12 Probit Regressions of Investigation on University Affiliation: Including the Party School of CPC This table repeats the regressions of corruption investigation on the university affiliation measures (Table 5), except that the university affiliation measures include the Party School of the Central Committee of CPC.

Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) Local Government Connection 1.135*** 0.897*** (3.98) (3.59) Central Government Connection 0.037 0.145 (0.08) (0.32) University Affiliation: Top 7 Leaders -0.803*** -2.146*** -2.599*** (-2.76) (-3.10) (-3.99) Birth Conn: Top 7 Leaders 0.030 -0.089 0.017 (0.08) (-0.18) (0.04) Association with Zhou Yongkang 1.975*** 2.151*** (2.60) (3.07) Univ. Conn: Investigated National Leaders 1.388*** 0.698 0.900 (2.99) (1.19) (1.64) Birth Conn: Investigated National Leaders 0.529 0.542 0.846* (1.31) (0.98) (1.83) Univ. Conn: Non-Invest. National Leaders -0.366*** 0.261 0.350 (-2.58) (1.06) (1.52) Birth Conn: Non-Invest. National Leaders -0.371** -0.793*** -0.563** (-1.98) (-2.86) (-2.36) Birth Conn: 1,021 Investigated Politicians 0.039 -0.131 (0.18) (-0.71) Province Graft-Tigers 0.319 (1.22) Province Graft-Flies -1.385 (-0.93) Investigation Team in Province 0.554* 0.388 (1.83) (1.45) Monitoring -0.210** -0.262*** -0.187 -0.243** (-2.24) (-2.76) (-1.47) (-2.19) CEO Compensation ** * * 0.730 0.703 0.336 0.630 (2.03) (1.96) (0.79) (1.75) CEO Pay for Performance -0.824 -0.769 -0.915 -0.569 (-1.30) (-1.18) (-1.19) (-0.83) CEO Near-Retirement Dummy *** *** ** ** 0.730 0.682 0.640 0.575 (3.37) (3.11) (2.30) (2.31) Related-Party Sales 1.156** 1.338** 0.482 0.298 (1.99) (2.21) (0.61) (0.41) Related-Party Loans 16.177*** 12.904** 17.832** 17.585** (2.74) (2.23) (2.23) (2.39)

26 Dependent Variable: Dummy of Corruption Investigation Other Receivables from Parent 1.249** 1.686** 1.603** 1.631** (1.98) (2.50) (2.02) (2.10) # Regulation Breaches 0.067 0.093 0.181 0.083 (0.32) (0.44) (0.74) (0.38) Operational Inefficiency 0.203*** 0.193** 0.111 0.183** (2.70) (2.52) (1.20) (2.17) Investment Inefficiency 1.948 1.954 2.139 1.622 (1.61) (1.60) (1.40) (1.20) Corruption Postings 2.721*** 2.743** 2.412* 2.640** (2.61) (2.50) (1.88) (2.14) Controls Yes Yes Yes Yes Year Fixed Effect Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Region Fixed Effect Yes Yes Yes Yes # Obs 274 274 235 274

27 Table IA.13 Probit Regressions of Corruption Investigation on Political Connection Measures: Alternative Pairs of Universities This table is similar to Table 5 in the paper except that we consider university affiliations with two alternative pairs of universities: The top 3rd and 4th universities (University of Science and Technology of China and Fudan University), and the top 5th and 6th universities (Shanghai Jiao Tong University and Zhejiang University).

Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) (8) University Affiliation: Top 2 (Tsinghua & PKU) -1.523*** -1.577*** -2.027*** -2.013*** (-3.80) (-3.82) (-3.10) (-2.97) University Affiliation: Top 3 & 4 0.819 0.341 0.598 -0.784 (0.89) (0.25) (0.61) (-0.53) University Affiliation: Top 5 & 6 1.520* 1.496 1.670 1.643 (1.71) (1.46) (1.53) (1.31) Local Government Connection 0.828*** 0.880*** 0.834*** 0.795*** (3.35) (3.59) (3.36) (3.12) Central Government Connection -0.094 0.025 0.040 -0.071 (-0.22) (0.06) (0.09) (-0.17) Birth Conn: Top 7 Leaders -0.083 0.062 0.065 -0.083 (-0.19) (0.15) (0.15) (-0.19) Association with Zhou Yongkang 1.844*** 1.598** 1.631*** 1.870*** (2.82) (2.53) (2.59) (2.84) Univ. Conn: Invest. National Leaders 0.478 0.633 0.768 0.597 (0.98) (1.32) (1.53) (1.14) Birth Conn: Invest. National Leaders 0.812* 0.714 0.750* 0.847* (1.79) (1.60) (1.68) (1.85) Univ. Conn: Non-Invest. National Leaders 0.174 -0.343** -0.392** 0.118 (0.79) (-2.24) (-2.48) (0.50) Birth Conn: Non-Invest. National Leaders -0.532** -0.630*** -0.656*** -0.552** (-2.26) (-2.70) (-2.79) (-2.33) Birth Conn: 1,021 Invest. Politicians -0.139 -0.159 -0.170 -0.156 (-0.76) (-0.89) (-0.95) (-0.84) Investigation Team in Province 0.389 0.483* 0.476* 0.395 (1.47) (1.88) (1.84) (1.48)

28

Dependent Variable: Dummy of Corruption Investigation Monitoring -0.219** -0.237*** -0.237*** -0.209** -0.257** -0.243** -0.235** -0.253** (-2.29) (-2.64) (-2.59) (-2.13) (-2.38) (-2.39) (-2.30) (-2.31) CEO Compensation ** ** * ** * * * * 0.780 0.698 0.673 0.754 0.656 0.641 0.629 0.635 (2.16) (1.98) (1.92) (2.11) (1.84) (1.81) (1.79) (1.78) CEO Pay for Performance -0.706 -0.860 -0.826 -0.645 -0.575 -0.587 -0.533 -0.524 (-1.08) (-1.42) (-1.37) (-0.99) (-0.85) (-0.88) (-0.80) (-0.77) CEO Near-Retirement Dummy *** *** *** *** ** *** *** ** 0.625 0.703 0.735 0.677 0.530 0.631 0.665 0.557 (2.88) (3.29) (3.42) (3.10) (2.14) (2.61) (2.73) (2.22) Related-Party Sales 1.131* 1.249** 1.106* 0.962 0.723 1.418** 1.419** 0.747 (1.96) (2.20) (1.89) (1.61) (1.00) (2.09) (2.07) (1.03) Related-Party Loans 16.926*** 12.483** 12.631** 17.341*** 17.123** 13.807** 14.199** 17.602** (2.59) (2.25) (2.27) (2.60) (2.35) (2.20) (2.26) (2.39) Other Receivables from Parent 1.637** 1.060* 1.051* 1.680** 1.556** 1.188* 1.236* 1.588** (2.37) (1.83) (1.81) (2.39) (2.04) (1.75) (1.78) (2.03) # Regulation Breaches 0.066 0.109 0.133 0.077 0.102 0.072 0.096 0.130 (0.31) (0.53) (0.64) (0.37) (0.46) (0.33) (0.44) (0.59) Operational Inefficiency 0.191** 0.176** 0.172** 0.181** 0.160* 0.176** 0.169** 0.155* (2.48) (2.35) (2.28) (2.32) (1.91) (2.18) (2.07) (1.83) Investment Inefficiency 2.063* 1.959 1.974* 2.068* 1.811 1.679 1.647 1.794 (1.69) (1.64) (1.65) (1.68) (1.37) (1.31) (1.28) (1.36) Corruption Postings 2.945*** 2.480** 2.260** 2.681** 2.309* 2.182* 1.981* 2.067* (2.80) (2.41) (2.16) (2.49) (1.94) (1.87) (1.67) (1.71) Controls Yes Yes Yes Yes Yes Yes Yes Yes Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Industry-Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes # Obs 257 257 257 257 224 257 224 257

29 Section A.7 List of Positions for National Leaders

According to the official website of the government of China, leaders holding the following positions are recognized as national leaders. Note that a few national leaders assume multiple positions. For example, besides being the President, Xi, Jinping also holds the positions of General Secretary of the CPC Central Committee, the Chairman of the Military Commission of the CPC Central Committee, and the Chairman of the Central Military Commission of the People’s Republic of China. The list of positions for national leaders can also be found at the Baidu Baike (China’s Wikipedia): http://baike.baidu.com/link?url=lLljb-Ts74IJJ6gJzL-S_9WjVVXxGVmFrqpBakfXMri- usc8Tmb8mj42NZS3iAiDLy3Rn9avCUxmhKs8z2sBu3zxPpjxaTl1gG2XmHMAsVSMZyW03ZC9 kcoAJdPZg4Tyj7-CelJl8VroWy8coaHZqtUKOU5uYJLyBBmDSm5RVNS#reference-[1]-1381052- wrap.

1) The Communist Party of China • General Secretary of the CPC Central Committee. • Members of the Political Bureau of the CPC Central Committee (over 20 members, including the 7 members of the Politburo Standing Committee). • Members of the Secretariat of the CPC Central Committee. • Chairman and Vice-Chairman of the Military Commission of the CPC Central Committee. • Secretary of the CPC Central Commission for Discipline Inspection.

2) The Central People’s Government of China • President. • Vice Presidents. • Premier. • Chairman and Vice-Chairman of the Central Military Commission of the People's Republic of China.

3) Other Branches • The National People’s Congress: Chairman and Vice-Chairmans of the Standing Committee. • The Supreme People’s Court: President. • The Supreme People’s Procuratorate: Procurators-General. • The People’s Political Consultative Conference: Chairman and Vice-Chairmans of The National Committee.

30 Section A.8 Robustness Tests of Corruption Measures: Excluding Politically Associated Investigations. Table IA.14 Probit Regressions of Corruption Investigations: Exclude the Corporate Investigations with Direct or Indirect Evidence of Political Association This table is similar to Table 3 in the paper but excluding the events where there is direct or indirect evidence that the corporate investigation is a consequence of political investigation (defined in Section 3.2.4). Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) Monitoring by Minority Shareholders -0.233** -0.334** -0.329*** (-2.36) (-2.03) (-2.64) CEO Compensation 0.941* 0.635 1.181 (1.23) (0.60) (1.37) CEO Pay for Performance -0.701 -0.949 -0.684 (-0.78) (-0.91) (-0.69) CEO Near-Retirement Dummy 0.680** 0.882** 0.950*** (2.49) (2.19) (2.98) Related-Party Sales 0.923 1.429* 1.283* (1.38) (1.80) (1.79) Related-Party Loans 8.876 9.971 12.882** (1.55) (1.42) (1.97) Other Receivables from Parent 2.618* 3.849 3.483** (1.90) (1.63) (2.28) # Regulation Breaches 0.075 0.016 0.129 (0.32) (0.06) (0.54) Bus. Ent. Expenditures 0.004 -0.018 (0.10) (-0.36) Operational Inefficiency 0.131 0.161 0.149 (1.58) (1.37) (1.53) Investment Inefficiency 2.338* 2.096 2.395 (1.79) (1.20) (1.61) Corruption Postings 2.052* 2.184 2.785** (1.94) (1.29) (2.24) Controls Yes Yes Yes Yes Yes Yes Yes Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes # Obs 186 186 140 183 192 130 177 31 Table IA.15 Probit Regressions of Corruption Investigations: Exclude Investigations with Direct Evidence of Political Association This table is similar to Table 3 in the paper but excluding the events where there is direct evidence that the corporate investigation is a consequence of political investigation (defined in Section 3.2.4). Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) Monitoring -0.283*** -0.326** -0.363*** (-3.08) (-2.04) (-3.05) CEO Compensation 0.265 -0.019 0.201 (0.63) (-0.04) (0.43) CEO Pay for Performance -0.623 -1.125 -0.942 (-0.79) (-1.19) (-1.06) CEO Near-Retirement Dummy 0.642** 0.797** 0.884*** (2.57) (2.18) (3.04) Related-Party Sales 1.222* 1.459* 1.523** (1.87) (1.86) (2.19) Related-Party Loans 9.858* 14.689** 15.042** (1.81) (2.04) (2.35) Other Receivables from Parent 1.442* 2.647* 2.240** (1.88) (1.65) (2.31) # Regulation Breaches 0.083 0.071 0.147 (0.35) (0.27) (0.62) Entertainment Expenditures -0.011 -0.014 (-0.26) (-0.30) Operational Inefficiency 0.165** 0.194* 0.166* (2.12) (1.72) (1.81) Investment Inefficiency 1.984 1.465 2.190 (1.59) (0.90) (1.59) Corruption Postings 2.405** 2.201 2.935** (2.34) (1.32) (2.45) Ln(ME) 0.177* 0.180* 0.139 0.159 0.093 0.242 0.164 (1.79) (1.86) (1.08) (1.64) (0.97) (1.47) (1.37) SOE Dummy -0.255 -0.383 -0.494 -0.426 -0.227 -1.406* -0.902 (-0.44) (-0.83) (-0.97) (-0.90) (-0.50) (-1.84) (-1.31) Year, Industry, Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes # Obs 211 208 153 204 218 141 197

32 Table IA.16 Probit Regressions of Corruption Investigation: Excluding Investigated Firms Associated with Zhou, Yongkang This table is similar to Table 3 in the paper but excluding the investigated firms associated with Zhou, Yongkang (defined in Section 3.2.3). Dependent Variable: Dummy of Corruption Investigation Independent Variables (t-1) (1) (2) (3) (4) (5) (6) (7) Monitoring -0.236*** -0.215* -0.264*** (-2.93) (-1.78) (-2.70) CEO Compensation 0.629* 0.419 0.693* (1.82) (1.13) (1.94) CEO Pay for Performance -0.965 -1.143* -1.026 (-1.63) (-1.74) (-1.64) CEO Near-Retirement Dummy 0.643*** 0.808*** 0.782*** (3.14) (2.93) (3.52) Related-Party Sales 0.866 1.235* 1.116* (1.51) (1.75) (1.83) Related-Party Loans 8.304 11.543* 12.241** (1.60) (1.92) (2.14) Other Receivables from Parent 0.690 0.994 0.988* (1.42) (1.35) (1.75) # Regulation Breaches 0.075 0.175 0.160 (0.34) (0.74) (0.74) Entertainment Expenditures 0.001 0.005 (0.03) (0.11) Operational Inefficiency 0.131* 0.091 0.137* (1.89) (0.99) (1.75) Investment Inefficiency 1.930* 1.232 1.661 (1.71) (0.87) (1.35) Corruption Postings 1.712* 1.341 2.077* (1.72) (0.88) (1.79) Ln(ME) 0.142* 0.142* 0.169 0.118 0.098 0.150 0.105 (1.65) (1.70) (1.56) (1.40) (1.19) (1.16) (1.06) SOE Dummy -0.245 -0.208 -0.391 -0.225 -0.187 -0.891** -0.484 (-0.83) (-0.77) (-1.35) (-0.81) (-0.72) (-2.39) (-1.47) Year Fixed Effect Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Region Fixed Effect Yes Yes Yes Yes Yes Yes Yes # Obs 266 259 196 255 273 181 248

33 Section A.9 Additional Analysis for Change in Corruption Measures of All Chinese Firms or Event Firms

Table IA.17 Corruption Measures for Non-SOEs: 2005-2015 This table corresponds to Table 8 except that we include only Non-SOEs. Corruption Measures Year CEO CEO Pay- Related Related Other Reg. Entertain. Operational Inv. Comp. (%) Perform. Sales Loans Receiv. (%) Breaches Exp. (%) Inefficiency Inefficiency 2005 0.021 0.051 0.045 0.027 0.923 0.105 0.329 1.175 0.087 2006 0.024 0.056 0.037 0.015 0.595 0.087 0.313 0.421 0.105 2007 0.032 0.105 0.037 0.012 0.159 0.082 0.365 -0.022 0.128 2008 0.037 0.133 0.034 0.010 0.034 0.080 0.306 0.907 0.159 2009 0.040 0.177 0.027 0.007 0.014 0.130 0.359 0.245 0.126 2010 0.042 0.202 0.021 0.003 0.003 0.116 0.347 0.211 0.112 2011 0.038 0.278 0.018 0.001 0.006 0.140 0.317 0.290 0.178 2012 0.037 0.324 0.019 0.001 0.006 0.218 0.318 0.403 0.138 2013 0.037 0.364 0.018 0.001 0.005 0.251 0.328 0.401 0.177 2014 0.036 0.354 0.021 0.003 0.002 0.219 0.275 0.303 0.094 2015 0.037 0.369 0.020 0.002 0.000 0.221 0.256 0.490 0.109 2013~2015 - 2011 Diff. -0.002 0.084 0.002 0.001 -0.004 0.089 -0.031 0.111 -0.053 t-stat (-1.90) (8.23) (1.16) (4.03) (-2.52) (6.37) (-3.75) (3.09) (-6.85) % Diff. -5.50% 30.25% 10.92% 100.00% -65.19% 63.28% -9.87% 38.23% -29.63%

34 Table IA.18 Abnormal Corruption Measures for SOEs in 2005-2015: Benchmarked to Non-SOEs This table presents annual averages of abnormal corruption measures for SOEs relative to non-SOEs from 2005 to 2015. For each SOE firm, we identify a matched non-SOE using the propensity score matching based on the all the corruption measures in our Table 3. Abnormal corruption measures are calculated as the differences between the SOEs and matched Non-SOEs. We also report differences between 2011 (before anti-corruption campaign) and the average of 2013-2015 (after the start of anti-corruption campaign), as well as the associated t-statistics. Bold is used for numbers that are statistically significant at 0.10 level.

Corruption Measures: SOE – Non-SOE Year CEO CEO Pay- Related Related Other Reg. Entertain. Operational Inv. Comp. (%) Perform. Sales Loans Receiv. (%) Breaches Exp. (%) Inefficiency Inefficiency 2007 -0.011 -0.058 0.030 -0.002 -0.079 -0.033 -0.033 -0.315 -0.011 2008 -0.017 -0.051 0.035 0.000 -0.009 -0.032 0.004 -0.034 -0.017 2009 -0.015 -0.094 0.042 -0.001 0.009 -0.055 -0.009 0.168 -0.003 2010 -0.021 -0.157 0.038 -0.001 0.005 -0.092 -0.088 -0.019 -0.005 2011 -0.019 -0.168 0.043 0.000 0.004 -0.075 -0.042 -0.012 0.025 2012 -0.018 -0.214 0.038 0.001 0.005 -0.032 -0.107 0.016 -0.006 2013 -0.018 -0.217 0.045 0.001 0.004 -0.066 -0.091 -0.015 0.002 2014 -0.018 -0.147 0.041 0.002 0.003 -0.077 -0.111 0.068 0.007 2015 -0.020 -0.138 0.045 0.001 0.000 -0.098 -0.083 0.151 0.016 2013~2015 - 2011 Diff. 0.001 0.000 0.000 0.001 -0.001 -0.005 -0.054 0.080 -0.016 t-stat (0.26) (0.01) (0.02) (1.27) (-0.25) (-0.17) (-2.19) (0.76) (-1.72)

35 Table IA.19 Change in Corruption Measures for Event Firms after Corruption Investigations This table presents changes in corruption measures for event firms after corruption investigations. The initial sample includes 150 Chinese listed firms with corrupt managers investigated from the beginning of the anti-corruption campaign on December 4, 2012 to December 31, 2015. We further require firms to have corresponding measures available for both years t-1 and t+1, where year t is the year of corruption investigation. The firm-level corruption indicators include: 1) CEO compensation, scaled by total assets; 2) CEO pay-for-performance sensitivity; 3) Related-party sales, scaled by revenue; 4) Related-party loans, scaled by total assets; 5) Other receivables from parent firm, scaled by total assets; 6) Number of regulation breaches in a year; 7) Business entertainment expenditures, scaled by total assets; 8) Operational inefficiency, calculated as growth of sales minus growth of net income; 9) Investment inefficiency, calculated as the absolute value of residue from regression of investment on sales growth within each industry-year; and 10) Corruption postings, measured as percentage of posts that discussed corruption in the total posts for a firm on “Guba” (“Stock Bar” in English), a popular online investor-forum. All Chinese firms’ fiscal years end in December so the fiscal year coincides with the calendar year. For each event firm, we identify a matched firm using the propensity score matching based on the corruption measures. We report corruption measures for both event firms and matched firms for years t-1 and t, and then the difference-in- difference. CEO compensation, other receivables, and business entertainment expenditures are expressed in percentages to ease reading. T- statistics associated with difference-in-difference are also reported. Investigated Firms Matched Firms Diff-in- Measures t-1 t+1 Diff. t-stat t-1 t+1 Diff. t-stat Diff t-stat #Obs CEO Compensation (%) 0.019 0.020 0.001 0.25 0.017 0.016 -0.001 -1.13 0.003 0.49 69 CEO Pay for Performance 0.014 0.018 0.004 0.99 0.031 0.042 0.011 2.26 -0.007 -1.27 59 Related-Party Sales 0.070 0.077 0.007 0.58 0.096 0.099 0.003 0.26 0.005 0.38 73 Related-Party Loans 0.003 0.009 0.007 2.18 0.005 0.004 0.000 -0.20 0.007 1.87 73 Other Receivables (%) 0.014 0.003 -0.011 -1.47 0.015 0.003 -0.012 -1.55 0.001 0.11 73 Regulation Breaches 0.178 0.219 0.041 0.53 0.151 0.123 -0.027 -0.42 0.068 0.67 73 Entertainment Exp. (%) 0.189 0.137 -0.052 -3.41 0.277 0.213 -0.064 -4.04 0.013 0.72 27 Operational Inefficiency 0.549 0.950 0.400 1.36 0.307 0.913 0.606 2.79 -0.206 -0.55 73 Investment Inefficiency 0.167 0.147 -0.020 -0.65 0.178 0.136 -0.042 -1.84 0.022 0.56 73 Corruption Postings 0.066 0.113 0.047 2.94 0.047 0.069 0.021 2.24 0.026 1.48 73

36 Table IA.20 Change in Financing and Investment for Event Firms after Corruption Investigations This table is the same as Table IA.19 except that we examine the measures of financing and investment of event firms. Investment is calculated as the change in gross property, plant and equipment plus change in inventories. Market or book leverage ratios are defined as total debt divided by the sum of market or book value of equity and total debt. Short-term debt is measured as total current liabilities. Long-term debt is measured as total long-term liabilities. Short-term (long-term) loans are loans borrowed from banks and other financial institutions that are due in less than one year (longer than one year). Long-term bonds are bonds payable due more than one year. Panel A: Year t-1 versus Year t Investigated Firms Matched Firms Diff-in- Measures t-1 t Diff. t-stat t-1 t Diff. t-stat Diff t-stat #Obs Investment/Asset 0.039 0.039 0.000 0.03 0.064 0.053 -0.010 -0.68 0.011 0.53 139 Market Leverage 0.370 0.343 -0.027 -4.26 0.361 0.327 -0.034 -4.61 0.007 0.80 139 Book Leverage 0.525 0.522 -0.003 -0.41 0.517 0.507 -0.010 -1.25 0.007 0.78 137 Short-term Debt/Asset 0.411 0.407 -0.004 -0.62 0.381 0.376 -0.004 -0.62 0.000 0.04 138 Long-term Debt/Asset 0.097 0.097 0.000 0.01 0.118 0.112 -0.006 -1.27 0.006 0.99 139 Short-term Loans/Asset 0.122 0.123 0.000 0.08 0.099 0.097 -0.002 -0.47 0.002 0.38 139 Long-term Loans/Asset 0.063 0.059 -0.004 -1.15 0.080 0.074 -0.007 -1.57 0.003 0.54 139 Long-term Bonds/Asset 0.021 0.024 0.003 1.09 0.030 0.029 -0.001 -0.50 0.004 1.18 139 Panel B: Year t-1 versus Year t+1 Investigated Firms Matched Firms Diff-in- Measures t-1 t+1 Diff. t-stat t-1 t+1 Diff. t-stat Diff t-stat #Obs Investment/Asset 0.039 0.026 -0.013 -0.63 0.066 0.045 -0.021 -0.75 0.008 0.24 73 Market Leverage 0.389 0.320 -0.069 -5.31 0.385 0.305 -0.079 -5.88 0.010 0.58 73 Book Leverage 0.534 0.547 0.013 1.05 0.514 0.503 -0.010 -0.92 0.023 1.36 72 Short-term Debt/Asset 0.413 0.415 0.003 0.27 0.378 0.372 -0.006 -0.60 0.008 0.60 72 Long-term Debt/Asset 0.099 0.100 0.000 0.03 0.120 0.110 -0.010 -1.16 0.010 0.91 73 Short-term Loans/Asset 0.139 0.124 -0.016 -1.77 0.098 0.086 -0.012 -1.37 -0.004 -0.33 73 Long-term Loans/Asset 0.061 0.069 0.008 1.44 0.078 0.075 -0.003 -0.38 0.011 1.21 73 Long-term Bonds/Asset 0.025 0.018 -0.007 -1.31 0.036 0.029 -0.007 -1.45 0.000 0.03 73

37 Table IA.21 Change in Financing and Investment after Corruption Investigations: Connected Firms This table is the same as Table IA.19 except that we examine the measures of financing and investment of connected firms (defined in Section 3.3). Panel A: Customers of Event Firms t-1 vs t t-1 vs t+1 Measures t-1 t Diff. t-stat t-1 t Diff. t-stat Investment/Asset 0.053 0.027 -0.026 -1.51 0.047 0.034 -0.013 -0.50 Market Leverage 0.426 0.411 -0.014 -1.42 0.385 0.334 -0.051 -3.21 Book Leverage 0.533 0.544 0.012 1.64 0.507 0.528 0.021 2.18 Short-term Debt/Asset 0.403 0.406 0.002 0.37 0.378 0.380 0.002 0.25 Long-term Debt/Asset 0.073 0.083 0.009 1.73 0.061 0.077 0.016 2.24 Short-term Loans/Asset 0.117 0.110 -0.007 -1.28 0.108 0.098 -0.010 -1.41 Long-term Loans/Asset 0.046 0.049 0.003 1.01 0.036 0.041 0.005 0.91 Long-term Bonds/Asset 0.019 0.020 0.001 0.37 0.018 0.024 0.006 1.70 Panel B: Suppliers of Event Firms t-1 vs t t-1 vs t+1 Measures t-1 t Diff. t-stat t-1 t Diff. t-stat Investment/Asset 0.037 0.056 0.019 0.39 0.041 0.001 -0.040 -0.49 Market Leverage 0.485 0.492 0.007 0.31 0.636 0.547 -0.089 -1.88 Book Leverage 0.549 0.557 0.007 0.31 0.647 0.649 0.001 0.08 Short-term Debt/Asset 0.422 0.429 0.007 0.29 0.491 0.497 0.006 0.21 Long-term Debt/Asset 0.115 0.114 -0.001 -0.20 0.153 0.142 -0.011 -0.68 Short-term Loans/Asset 0.126 0.108 -0.018 -0.85 0.187 0.170 -0.017 -0.51 Long-term Loans/Asset 0.073 0.074 0.001 0.13 0.092 0.087 -0.004 -0.37 Long-term Bonds/Asset 0.038 0.036 -0.002 -0.52 0.055 0.046 -0.010 -1.65 Panel C: Firms with Managerial Connections with Event Firms t-1 vs t t-1 vs t+1 Measures t-1 t Diff. t-stat t-1 t Diff. t-stat Investment/Asset 0.062 0.042 -0.020 -1.09 0.089 0.051 -0.038 -1.08 Market Leverage 0.303 0.277 -0.026 -3.92 0.354 0.310 -0.044 -3.67 Book Leverage 0.497 0.511 0.014 2.33 0.523 0.524 0.001 0.12 Short-term Debt/Asset 0.341 0.344 0.003 0.49 0.336 0.307 -0.029 -3.38 Long-term Debt/Asset 0.076 0.084 0.008 1.93 0.083 0.105 0.021 3.16 Short-term Loans/Asset 0.073 0.080 0.006 1.15 0.073 0.074 0.001 0.16 Long-term Loans/Asset 0.044 0.040 -0.004 -1.58 0.054 0.059 0.005 0.95 Long-term Bonds/Asset 0.027 0.036 0.009 3.12 0.023 0.036 0.013 2.97

38 Table IA.22 Accounting Manipulation and Financial Market Environment 2005-2015: SOEs Only This table corresponds to Table 9 except that we include only SOEs. Earnings Return Volatility around Discontinuity Earnings Announcement Abs. Small Small Normalized Differenced Year (DACC) Profit Loss Volatility Volatility 2005 0.059 0.071 -0.055 0.165 0.280 2006 0.053 0.038 -0.042 0.056 0.096 2007 0.058 0.033 -0.038 0.256 0.499 2008 0.061 0.131 -0.119 0.139 0.241 2009 0.061 0.035 -0.035 0.088 0.109 2010 0.050 0.074 -0.070 0.232 0.297 2011 0.052 0.074 -0.066 0.276 0.339 2012 0.047 0.055 -0.051 0.132 0.152 2013 0.046 0.073 -0.051 0.288 0.336 2014 0.044 0.064 -0.050 0.019 0.026 2015 0.046 0.094 -0.073 0.068 0.060 2013~2015 - 2011 Diff. -0.007 0.002 0.008 -0.150 -0.197 t-stat (-3.86) (0.15) (0.68) (-6.22) (-5.74)

39 Table IA.23 Accounting Manipulation and Financial Market Environment 2005-2015: Non-SOEs Only This table corresponds to Table 9 except that we include only Non-SOEs. Earnings Return Volatility around Discontinuity Earnings Announcement Abs. Small Small Normalized Differenced Year (DACC) Profit Loss Volatility Volatility 2005 0.071 0.065 -0.056 0.127 0.217 2006 0.067 0.028 -0.026 0.083 0.153 2007 0.078 0.034 -0.029 0.256 0.483 2008 0.076 0.151 -0.131 0.243 0.464 2009 0.073 0.063 -0.046 0.159 0.265 2010 0.064 0.039 -0.070 0.309 0.420 2011 0.056 0.053 -0.088 0.373 0.498 2012 0.049 0.077 -0.071 0.090 0.103 2013 0.048 0.047 -0.054 0.246 0.352 2014 0.048 0.090 -0.080 0.070 0.171 2015 0.047 0.080 -0.076 0.120 0.186 2013~2015 - 2011 Diff. -0.009 0.019 0.018 -0.229 -0.264 t-stat (-5.13) (1.06) (1.53) (-10.14) (-7.63)

40 Table IA.24 Regressions of Provincial Foreign Direct Investment This table presents the regression of provincial foreign direct investment (FDI) on marketization index. In model (1), the dependent variable is annual inflation-adjusted provincial FDI, in trillion USD. In model (2), the dependent variable is the annual provincial FDI scaled by annual provincial GDP. Our sample includes the years 2011 and 2013-2015. The independent variables include provincial marketization index, a post-2013 dummy which equlas one for the years from 2013 (after anti- corruption campaign), and their interaction term. The provincial marketization index is constructed in Fan, Wang and Zhu (2011) except that we purge the one component index on foreign investment.

Dependent Variable: Foreign Direct Investment in a Province in Year t Raw FDI FDI/Provincial GDP (1) (2) Marketization × Post-2013 Dummy 0.199 -0.000 (0.46) (-0.20) Marketization 2.237*** 0.006*** (5.97) (3.48) Post-2013 Dummy -0.339 0.002 (-0.10) (0.13) Intercept -10.166*** -0.015 (-3.47) (-1.21) # Obs 124 124

41 Section A.10 Propensity Score Matching (PSM) Approach

For each event firm, we identify a matched firm using the propensity score matching approach (PSM) based on the pre-event corruption measures and firm characteristics. The matching variables include monitoring by minority shareholders, CEO compensation, CEO pay-for-performance, related-party sales, related-party loans, other receivables, regulation breaches, operational inefficiency, investment inefficiency, corruption postings, firm size, year, SOE status, industry, and region fixed effects. Specifically, we first estimate a logit regression of the dummy of investigation on these corruption measures and firm characteristics using all firm-years from 2012-2015. We then find for each investigated firm in our sample a matched control firm using the nearest neighbor matching technique without replacement and setting the caliper to 0.25. This procedure yields a final sample of 139 pairs of treatment and control firms with valid data.

42 Section A.11 Sample Selection and Propensity Score Matching: Matched Hong Kong Firms

Among the 1,869 Hong Kong listed firms, we first exclude Mainland Chinese firms, namely, H shares, Red Chips, and Chinese private enterprises, which in total account for 947 Hong Kong listed firms. Next, we exclude 49 foreign firms using three sources: Wolrd Federation of Exchanges Database, constituents of Hang Seng Foreign Companies Composite Index, and a manual reading of company tickers and names. In the end, we are left with 873 Hong Kong local firms. Additionally, Chinese firms have December fiscal year ends but Hong Kong firms’ fiscal year ends may vary. To avoid misalignment of accounting periods, we further require Hong Kong firms to have fiscal year ends between Septermber and March, which left us with 741 Hong Kong local firms. Then we use the propensity score matching procedure (PSM) to match Chinese listed firms with Hong Kong firms. This approach uses the matching variables that are also available for Hong Kong firms, including monitoring by minority shareholders, operational inefficiency, investment inefficiency, firm size, year dummies, and industry dummies.

43 Section A.12 Parallel Trend Analysis for Table 6 in the Paper.

Table IA.25 Abnormal Corruption Measures for All Chinese Firms in 2005-2015: Benchmarked to Hong Kong Firms: Parallel Trend Analysis We follow the literature [Bertrand and Mullainathan (2003), Heider and Ljungqvist (2015)] and test the parallel trend assumption by conducting the following regression (Yevent,i,t – Ymatch,i,t) = α + β × Post + γ × Pre + ε, where Y takes each of corruption measures below. Post is a dummy variable that equals one for years 2013-2015 and zero for years 2009-2011. Pre is a dummy variable that equals one for year 2009- 2010 and zero for years 2011 and 2013-2015. The table below reports the coefficient estimates of γ and associated t-stats. Note that year 2012 is skipped in this test because the anti-corruption campaign started in Dec-4th, 2012. Earnings Return Volatility around Discontinuity Earnings Announcement Small Small Abs. Regulation Operational Investment Normalized Differenced Year Profit Loss (DACC) Breaches Inefficiency Inefficiency Volatility Volatility 2009~2010 - 2011 Diff. -0.053 0.018 -0.001 -0.024 0.493 -0.058 -0.036 0.001 t-stat (-3.30) (1.58) (-0.39) (-1.45) (7.04) (-9.67) (-1.15) (1.37)

44 References

Beaver, W. H., M. F. McNichols, and K. K. Nelson. 2007. An alternative interpretation of the discontinuity in earnings distributions. Review of Accounting Studies 12: 524–556. Bertrand, M., and S. Mullainathan. 2003. Enjoying the quiet life? Corporate governance and managerial preferences. Journal of Political Economy 111: 1043-1075. Core, J., and W. Guay. 2002. Estimating the value of employee stock option portfolios and their sensitivities to price and volatility. Journal of Accounting Research 40: 613–630. Fan, G., X. Wang, and H. Zhu. 2011. NERI index of marketization of China’s provinces: 2011 report (in Chinese). Economic Science Press, Beijing. Heider F., and A. Ljungqvist. 2015. As certain as debt and taxes: estimating the tax sensitivity of leverage from state tax changes. Journal of Financial Economics 118: 684–712. Jones, J. J. 1991. Earnings management during import relief investigations. Journal of Accounting Research 29: 193–228. Ou-Yang, H., H. Shu, and S. M. Wong. 2015. Does it pay for entertaining your stakeholders. Working Paper.

45