DOES STATE CONTROL AFFECT MANAGERIAL INCENTIVES? EVIDENCE FROM CHINA’S PUBLICLY LISTED FIRMS*

Using data for 1203 publicly listed fi rms in China during 1999–2002, this paper empirically investigates whether and to what extent state control affects managerial incentives, including managerial compensation and CEO turnover. The paper fi nds that CEO turnover is negatively related to both current and lagged fi rm performance as measured by ROA and RPE (Relative Performance Evaluation) for non-state-controlled fi rms, while insensitive to performance measures for statecontrolled fi rms. In addition, CEO compensation is positively related to fi rm performance, but state ownership and control weaken this positive relation. Moreover, state control reduces the effectiveness of internal governance mechanisms such as the board of directors and supervisory committee. Overall, empirical results in the paper indicate that state ownership and control weaken managerial incentives and internal monitoring among publicly listed fi rms in China.


Introduction
In recent decades, worldwide privatization has drawn considerable research interest and generated a large number of theoretical and empirical studies concerning ownership, incentives and fi rm performance 1 . Megginson and Netter (2001) and Djankov and Murrell (2002) provide comprehensive surveys of over 200 empirical studies on privatization in both developed and emerging market economies, most of which focus on the differential performance between state-owned enterprises (SOEs) and privately-owned fi rms. Overall, existing evidence supports the proposition that private ownership is associated with better fi rm performance than state ownership is (e.g., Boubakri and Cosset 1998;D'Souza and Megginson 1999;Sun and Tong 2003;Tvaronavičienė and Kalašinskaitė 2005). A number of theories on privatization attribute the ineffi ciency of state ownership to weak incentives. For instance, Al-chian (1965) and Shleifer (1998) argue that dispersed owners of state fi rms (the citizens) make it diffi cult to write complete contracts linking manager's incentives to the returns from their decisions. In their survey of corporate governance, Shleifer and Vishny (1997) argue that SOEs are actually controlled by bureaucrats who have extremely concentrated control rights but no signifi cant cash fl ow rights, since the latter is widely dispersed amongst the country's taxpayers. Bureaucrats' main concern is to achieve their political objectives and economic benefi ts, which are often quite different from the objective of maximizing SOE profi ts (Shleifer and Vishny 1994;Boycko et al. 1996). Therefore, state owners are thought to be unwilling to adopt incentives that are tied to performance; thus, bureaucrats are free to use fi rms to address their own goals (Cragg and Dyck 2003). Even bureaucrats who have shareholders' interest at heart have very weak incentives to invest the time and effort required to design complete incentive contracts and monitor the performance of SOE managers, because the cost of doing so is much greater than the political/electoral payoff of modestly improving SOE performance (Megginson 2005). Furthermore, to safeguard economic rents, principals may use their power to protect SOEs from competition, bankruptcy and takeover through political mechanisms, such as soft budget constraints (Vickers and Yarrow 1991;Kornai 1998;Lin et al. 1998). As a result, the shield of state ownership weakens managerial incentives. Alas (2003) argues that countries with a socialist past have to deal with the satisfaction of needs at a lower level than traditional capitalist countries and this consequently infl uences managerial and workers' incentives. Tvaronavičienė (2004) provides a theoretical framework for formulating effi cient state policy in incentive design for transition economies. However, Grossman and Hart (1986) and Cragg and Dyck (2003) argue that the exact form of ownership may not matter because managerial incentives within private enterprises can be imitated using appropriate contracts under state ownership. The tools that align incentives to performance are available to both private fi rms and SOEs. For instance, SOE Managers could be motivated to improve fi rm performance through incentive compensation based on achievement of certain targets, with poor-performing managers being punished through demotion or dismissals (Shirley and Xu 1998). In fact, Laffont and Tirole (1993) point out that theory alone is unlikely to be conclusive and econometric analyses are badly needed in this area.
Despite the importance of the issue, little is known about the relation between state control and managerial incentives due to limited information available on factors such as ownership structure, executive compensation and managerial turnover 2 . Our study builds on and extends existing literature by providing a detailed examination of the relation between state control and managerial incentives using a large sample of partially privatized listed fi rms in China. China was chosen as the context for this study for three reasons. First, China is the largest emerging economy in the world. In terms of GDP (adjusted using Purchasing Power Parity), China is the second largest economy behind only the U.S. China's domestic stock exchanges rank eleventh among the largest stock exchanges in the world (Allen et al. 2005). Second, China's share issue privatization (SIP) program is a nation-wide ongoing program that has raised more than 386 billion RMB in private investments. In China the state still retains about onethird of the shares in the listed fi rms, and is the largest shareholder in more than 20% of them. Thus, studying Chinese fi rms can provide evidence about ownership, control and incentives that are diffi cult to detect in the U.S. or U.K. data. Third, the ongoing wage reforms in China transformed an egalitarian pay structure into an incentive-based compensation scheme through the adoption of "annual salary system", which provides an excellent context to examine the relation between state control and managerial incentives. Tirole (2001) argues that managerial incentives are tied to fi nancial performance in two ways. First, incentives derive from expected changes in compensation associated with fi nancial performance (explicit incentives). Second, incentives could be a result of managers' fear of losing their job or autonomy (implicit incentives). Prendergast (1999) argues that dismissal threat is the most important form of nonlinear incentive contracts in which wages vary little with performance but poor performance is punished by dismissal. This may be especially the case in China, where salary and bonus are not the only way to reward managers. In a rent-seeking society, on-the-job perks, such as better housing, the use of cars, entertainment, restaurant meals, travel, diversion of assets and business opportunities, can be substantial (Chang and Wong 2004). Therefore, implicit incentives may matter more to managers of Chinese fi rms.
Following previous studies on managerial incentives (e.g. Kaplan 1994; Hadlock and Lumer 1997;Brickley and Van Horn 2002), we examine the impact of state control on the relation between CEO turnover and fi rm performance, as well as the relation between CEO compensation and fi rm performance. Our fi ndings are as follows: First, the probability of CEO turnover is negatively related to both current and lagged fi rm performance as measured by ROA, indicating that China's publicly listed fi rms provide clear profi t incentives to CEOs so that poor-performing CEOs are more likely to be replaced. This result is robust to industry-adjusted relative performance evaluation (RPE) measures, suggesting that peer group benchmark performance is considered in the evaluation of management and the design of incentive schemes. However, when the sample is divided into two sub-samples that consist of state-controlled versus non-state-controlled fi rms, CEO turnover remains negatively related to current and lagged fi rm performance among non-state-controlled fi rms. In contrast, the likelihood of CEO turnover is not related to fi rm performance among state-controlled fi rms, indicating that state control weakens managerial incentives. This result is robust to model specifi cations (logit or probit models), performance measures (ROA or RPE) and turnover classifi cation (overall turnover or forced turnover).
Second, CEO compensation is positively related to current and lagged fi rm performance. However, coef-fi cient estimates of the interactive term between lagged performance and state control are all signifi cantly negative, providing strong evidence that state control weakens pay-performance sensitivities.
Third, there is some evidence that state control weakens the effectiveness of board and supervisory committee. Specifically, among state-controlled firms, independent directors and outside supervisors do not have any signifi cant impact on CEO turnover. In contrast, the number of independent directors and outside supervisors are positively related to the probability of CEO turnover among non-state-controlled fi rms. Furthermore, CEO duality reduces the probability of CEO turnover by a much larger magnitude among state-controlled fi rms than it does among non-state-controlled fi rms, suggesting that state-controlled fi rms are subject to more severe management entrenchment problems.
As Gupta (2005) points out, state is the controlling shareholder (percentage of shares greater than 20%) in more than 18% of the publicly listed fi rms in 27 developed economies and in many transition economies. Thus, understanding the impact of state ownership on managerial incentives and fi rm performance is crucial to evaluate the success or failure of partial privatization. Our study adds to the literature on ownership, control and managerial incentives by providing robust evidence that state ownership and control weaken managerial incentives. In addition, our study complements several studies of corporate governance in China. For example, Xu and Wang (1999), Sun and Tong (2003) and Wei et al. (2005) all fi nd that state ownership in the publicly listed fi rms in China has a negative impact on fi rm performance and market valuation. Our study, by focusing on the impact of state control on managerial incentives, provides empirical evidence that helps explain inefficiency of state ownership documented in these previous studies. As a result, our study also contributes to the literature on international corporate governance.
The rest of the paper is organized as follows. Section 2 describes the institutional context of this study. Section 3 provides information about our data and presents relevant summary statistics. Section 4 presents results and robustness tests. Section 5 concludes.

SOE reforms, partial privatization and wage reforms in China
Before the initiation of SOE reform in 1979, the Chinese state government owned all enterprises and operated them like production units of a single giant fi rm. Similar to other centrally planned economies, fi rms were not autonomous decision-making units. They were not fi nancially independent and did not have responsibility for sales or pricing (Estrin 2002). The central government set production objectives and product prices. Enterprise managers were evaluated and compensated for following orders from the national or regional planning hierarchy and for subservience to political dogma (Mengistae and Xu 2004). A competitive managerial labor market, where managers are rewarded or punished in response to market forces and fi rm performance, almost did not exist (Groves et al. 1995). During this period, the central government collected all profi ts and distributed wages to workers at nationally determined pay rates. Managerial compensation was determined by factors such as the seniority, occupation, rank within the civil service bureaucrats and size of enterprises. Compensation was not related to either fi rm performance or individual contribution. Differences in pay were minimal and more symbolic than substantive, refl ecting an egalitarian ideology. Since 1979, SOE reform in China has experienced three stages, which are briefl y described below.
The fi rst stage (1979)(1980)(1981)(1982)(1983) focused on administrative decentralization and profi t retention. To provide SOE managers with incentives and improve fi rm effi ciency, SOEs were allowed to retain 3% of their profi ts. The second stage of reform (1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992) focused on the separation of government ownership from the control of SOE operations. SOE managers were given more autonomy to be in charge of enterprise operations after they promised a certain amount of tax (maximum tax rate was 55%) to the government. In 1985, the Ministry of Labor (MOL) mandated that the budget allocated for wages be tied to the economic performance of SOEs. In 1987, the central government further increased managerial autonomy by allowing managers to retain some of their profi ts through the establishment of a Management Responsibility Contract System (MRCS). Under the MRCS, SOE managers were given a wide range of control rights in production, investment, sales, profi ts, personnel management, and distribution of bonuses and fringe benefi ts via contracts (Su 2005). Groves et al. (1995) fi nd that MRCS not only enhanced managers' incentives and productivity, but also increased their compensation. An internal managerial labor market, where managers are promoted and compensated on the basis of their human capital and economic performance, had emerged at SOEs during this period. In 1992, the State Council issued a circular to start a pilot implementation of the "annual salary system" at 100 largest SOEs. This move was intended to further cope with ineffi ciency problems arising from insuffi cient managerial incentives. The new salary system allowed SOEs to set their internal wage structure within the overall budget guidelines established by the government.
The third stage of the SOE reform started in 1993 following the third plenary session of the 14 th Party Congress. The goal of the reform was to transform and further improve the performance and effi ciency of the SOEs through corporatization, adoption of profi t making objectives and partial privatization. In the 15 th Party congress held in 1997, enterprise reform continued to be the main theme and the government called for continuous improvement by establishing a modern corporate governance system. Through SIPs, China fi rst listed 10 former SOEs in the two stock exchanges (SHSE and SZSE) in 1990. Since then, the number of publicly listed fi rms has increased steadily. By the end of 2006, there were 1434 fi rms listed in the SHSE and SZSE, with a total market capitalization of 8.94 trillion Renminbi Yuan (about 1.15 trillion U.S. Dollars), representing approximately 42% of China's GDP. 3 There are two distinct features of the enterprise reform in China in the 1990s. First, most listed fi rms have a dominant shareholder. In our sample, on average, the largest shareholder owns 44.5% (median is 43.8%) of the total shares while the second largest shareholder owns about 8.4% (median is 5.1%). The second feature is that state retains a signifi cant amount of ownership in many listed fi rms, often as the largest shareholder. State, as represented by the State Asset Management Bureau (SAMB) and their local branches, owned about one-third of total shares and is the largest shareholder in about one-fourth of the listed fi rms. Firth et al. (2006) point out that, while the state gave substantial autonomy to managers of the listed fi rms, it was unwilling to give up control. Therefore, the dilemma of giving full managerial autonomy while maintaining state control of voting rights persisted till the end of our sample period. As a result, the unique ownership structure among China's listed fi rms provides a good laboratory to test the impact of state control on managerial incentives during partial privatization.
Parallel to the enterprise reform, wage reform was also pressed ahead by the central government. In 1994, the guidelines for publicly listed fi rms to set their own wage structure were further reduced to two criteria. One was that the growth rate of total wages must be lower than that of the net operating profi t after tax. Another was that the growth rate of per capita wages must be lower than that of labor productivity. In recent years, the "annual salary system" has become increasingly popular because the central government has gradually recognized that compensation structure is one of the key elements in enhancing enterprise productivity and effi ciency. In November 2003, the SAMB issued a circular to formally implement incentive-based an-nual salary systems among 189 SOEs directly owned by the central government. According to the circular, managerial compensation in the "annual salary system" consists of two major parts: fi xed base salary and performance salary. This is very similar to the "salary plus bonus" compensation package offered by fi rms in developed market economies. In addition to the largest SOEs, the annual salary system has also been widely adopted by the publicly listed fi rms, most of which are partially privatized former SOEs. Fleisher and Wang (2003), Firth et al. (2006) document the adoption of incentive pay in different types of Chinese enterprises.

Data and summary statistics
The sample used in this study consists of 1203 fi rms listed on the SHSE or SZSE in China during 1999China during -2002. Most of the fi rms are partially privatized former SOEs, and 821 fi rms had gone public prior to 1999. Our dataset is an unbalanced panel that consists of 4343 fi rm-year observations. All data are obtained from the Chinese Stock Market and Accounting Research (CS-MAR) database, commercially available from Shenzhen GTA Information Technology Company and the University of Hong Kong. Table 1 provides summary statistics of key variables described below.

Ownership structure
The ownership structure of China's listed fi rms can be classifi ed into three main categories: state shares, legal entity share and publicly tradable shares. State shares are retained by the SAMB of the central or local government branches and are not allowed to be publicly traded, although reforms have been initiated to free up these shares since May 2005 4 . Legal entity shares are held by domestic institutional investors including banks, securities companies, insurance companies, mutual funds, industrial enterprises, transportation and power companies, and some private enterprises. Similar to state shares, legal entity shares are also nontradable. However, there exist important differences between state ownership and legal entity ownership, which lead to different implications for managerial incentives. Sun and Tong (2003) point out that unlike SAMB, many legal entities have close business connections with listed fi rms in which they have ownership. Because legal entity shareholders benefi t directly from listed fi rms' good performance and are hurt by their poor performance, they have incentives to be active in designing appropriate incentive contracts and engaging in active monitoring. Compared with either SAMB or individual shareholders, legal entities have more expert knowledge of fi rms in which they have shareholding interests and are better equipped to monitor managers through their infl uence on the board of directors. In contrast, public shares are held by the investment public and tradable on the two securities exchanges. Public shares are widely dispersed among millions of individual investors.
In our study, the sample is split into state-controlled fi rms (the controlling shareholder is the state, State = 1) and non-state-controlled firms (the controlling shareholder is not the State, State = 0). Among nonstate-controlled fi rms, the largest shareholder is generally various types of legal entities. In our sample, 23% of the listed fi rms have state as their controlling shareholder (see Table 1). Table 2 presents summary statistics of key variables for state-controlled and non-state-controlled fi rms, including ROA, CEO characteristics, board composition and the structure of the supervisory committee. We fi nd that, on average, CEO compensation in state-controlled fi rms is lower than that in non-state-controlled fi rms. The difference is statistically signifi cant at the 1% level. The average fi rm performance of non-statecontrolled fi rms (ROA = 0.025) is signifi cantly higher than that of the state-controlled fi rms (ROA = 0.013). On average, CEOs for state-controlled fi rms are significantly older than those for non-state-controlled fi rms. The frequency for a CEO to serve as the chairman of the board (DUALITY) is signifi cantly smaller for non-state-controlled fi rms. On average, state-controlled fi rms have larger supervisory committees while less outside committee members, indicating that supervisory committee is less independent among statecontrolled fi rms. The table reports the number of observations, mean, standard deviation, minimum value and maximum value for the key variables used in the analysis. Statistics are for a sample of 1203 fi rms listed on the Shanghai or Shenzhen stock exchanges in China during 1999China during to 2002. CEO compensation is the annual total compensation (salary and bonus) received by CEO of the fi rm in specifi c year. CEO turnover is a dummy variable, which equals 1 if the employment of the CEO was terminated during the year, and zero otherwise. ROA is return on asset, the ratio of net income to total asset. Lag ROA is the one-year lagged ROA. Largest investor is the percentage of the shares held by the largest investor. State equals 1 if the largest investor is state, represented by the Bureau of National Asset Management, and zero otherwise. CEO age is the age (in years) of the CEO. Duality equals 1 when the CEO holds the board chair position and zero otherwise. CEO tenure is the number of the years the CEO has held the position till now. Board size is the number of directors on the board. Independent director is the number of independent directors on the board. No. of Supervisors is the number of supervisors on the supervisory committee. Outside supervisor is the number of outside supervisors on the supervisory committee. Firm size is the natural logarithm of the total assets. Marketization is the NERI (National Economics Research Institute, China) marketization index that measures the overall market development in each province in China, including market competition, government regulation, and the legal environment.
Prior studies (e.g. Huson et al. 2004) emphasize the importance of distinguishing between forced and voluntary turnovers because only forced turnovers refl ect the quality of monitoring. Clearly, CEO turnovers in our sample are not all forced. However, it is extremely diffi cult to separate forced and voluntary turnovers. State-controlled fi rms are fi rms of which the state is the largest shareholder, while non-state-controlled fi rms are fi rms of which the state is not the largest shareholder. In the non-state-controlled fi rms, the largest shareholder is generally various types of legal entities including industrial banks, securities companies, insurance companies, mutual funds, industrial enterprises, transportation and power companies, research institutes and other private business. CEO compensation is the annual total compensation (salary and bonus) received by CEO of the fi rm in specifi c year. CEO turnover is a dummy variable, which equals 1 if the employment of the CEO was terminated during the year, and zero otherwise. ROA is return on asset, the ratio of net income to total asset. Lag ROA is the one-year lagged ROA. Largest Investor is the percentage of the shares held by the largest investor. State equals 1 if the largest investor is state, represented by the Bureau of National Asset Management, and zero otherwise. CEO age is the age (in years) of the CEO. Duality equals 1 when the CEO holds the board chair position and zero otherwise. CEO tenure is the number of the years the CEO has held the position till now. Board size is the number of directors on the board. Independent Director is the number of independent directors on the board. No. of Supervisors is the number of supervisors on the supervisory committee. Outside Supervisor is the number of outside supervisors on the supervisor committee. Firm size is the natural logarithm of the total assets. Marketization is the NERI (National Economics Research Institute, China) marketization index that measures the overall market development in each province in China, including market competition, government regulation, and the legal environment. Firms usually are reluctant to reveal genuine reasons of CEO turnover. In particular, Chinese culture emphasizes harmony and preserving face in interpersonal and social relationships. Outright dismissal is perceived as a huge embarrassment and causes people to lose face. Thus, the reported reasons do not always reveal the true cause of CEO turnover. Huson et al. (2004) use whether an outgoing CEO is under the age of 60 years as a criterion to identify forced turnover. This classifi cation is reasonable in the U.S. context, because serving as the CEO of a large fi rm is usually the last stage of one's career and almost all CEOs choose not to retire until they are at least 60 years old. However, this classifi cation may not be applicable in the Chinese context. In fact, about 94% of CEOs in our sample left the position before reaching the age of 60 years, which is also the ordinary retirement age. It is diffi cult to argue that all these CEOs were forced out.
Given the special context of China, our study analyzes the turnover for all CEOs and tests the robustness of results by tracking the post-turnover employment of outgoing CEOs in our sample 6 . We fi nd that in 654 out of 1130 cases, outgoing CEO remains an executive of the company (board chair, vice board chair, director, executive president and supervisory committee chair, etc.). We exclude these cases as forced turnover. For the remaining turnovers, we exclude turnovers with reported reasons as retirement (5 cases), health (13 cases), change of controlling shareholder (27 cases) and legal disputes (3 cases). We then exclude turnovers with outgoing CEOs more than 60 years old (14 cases). Eventually, we identify 372 turnovers as forced turnovers in our sample. Using this classifi cation, we conduct robustness tests in section 4.3.

Executive compensation
We use total annual compensation, including base salary and bonus, to calculate CEO pay 7 . We do not consider stock options or stock ownership because they are rare among Chinese executives. In fact, none of the listed fi rms had used stock options as part of the executive compensation package by the end of 2002. Executive stock ownership stems from the so-called employee shares and only represents less than 0.01% of total shares outstanding, which is much less than the fractional managerial shareholding in developed market economies 8 . More importantly, shares held by executives are not tradable and hence not tied to managerial performance. Even if a manager performs well and gets promoted, his stock ownership usually does not change, at least in our sample. Furthermore, CSRC issued a circular to cease the issuance of employee shares in 1998.

Performance measures
Firm performance is measured by ROA (return on asset) and RPE (relative performance evaluation) 9 . ROA is a widely used measure of accounting profi tability in corporate governance studies (e.g. Huson et al. 2001;Brickley and Van Horn 2002). Parrino (1997) points out that RPE, calculated as fi rm-specifi c ROA minus the median ROA of an industry, can fi lter out common factors within such as shocks that are beyond managers' control and thus provide more precise information about fi rm performance. We use both ROA and RPE as performance measures.

Board characteristics
A well-functioning board of directors is widely regarded as an important internal corporate governance mechanism for CEO compensation and turnover. How-ever, whether or not boards are effective has always been a controversial issue. Su (2005) argues that board directors in China's publicly listed fi rms are largely selected through political and administrative processes rather than endogenously chosen in competitive managerial labor markets. Therefore, the impact of board structure on managerial incentives in China's listed fi rms is an empirical question worthy of exploration. We use the following three variables to characterize board of directors. The fi rst variable is board size 10 . Jensen (1993) and Yermack (1996) argue that larger boards are less effective in monitoring management and more susceptible to the infl uence of CEOs. The second variable is the number of independent or outside directors, defi ned as directors who are not members of the management team 11 . Fama and Jensen (1983) argue that independent directors generally care about their reputations and social status, thus have incentives to monitor management. The third variable is CEO duality, which equals 1 if a CEO is also the board chair and 0 otherwise. Dalton and Kesner (1997) argue that holding board chair enables CEOs to exert more control over board agenda and decisions, which weakens the governance function of the board.

The structure of the supervisory committee
According to China's Company Law, publicly listed fi rms must establish a supervisory committee consisting of shareholder representatives and employee representatives in appropriate proportions. The duties of supervisors are to scrutinize decisions made by managers, directors and other senior personnel, review and audit reports provided by directors, safeguard fi rm's assets, and resolve disputes between shareholders and directors. In practice, supervisory committees are headed by Communist Party leaders of a fi rm and do not have fi nance or audit sub-committee. More importantly, it is only equipped with the right of supervision, without the right to select managers and directors and to veto the decision of the board or management. Hence, the power of supervisory committees is quite limited. We use the size of the supervisory committee and the number of outside supervisors to characterize supervisory committees.

Marketization index
One special feature of Chinese economy is the imbalanced regional development. The differential level of economic development, market competitiveness and legal environment can exert signifi cant impact on the quality of corporate governance. To address this issue, we use the NERI (National Economics Research Institute of China) provincial marketization index compiled by Fan and Wang (2004) and a set of provincial dummies in our empirical analysis. The marketization index characterizes the progress of transition towards market economy for 31 provinces and special administrative regions and encompasses various indicators such as the extent of government intervention, the degree of market competition, the development of product and factor markets and the strength of legal environment. Fan and Wang (2004) transform the value of the above indicators into 0-10, with higher values being indicative of higher degree of marketization, and then use principal component analysis to determine the weight for each indicator. A higher marketization index is associated with less government intervention and more regional economic freedom.

Other control variables
We use the age of CEOs, CEO tenure (the number of the years the CEO has held the position till to date), fi rm size (the natural logarithm of the total assets), and industry and year dummies as other control variables throughout our empirical analysis.

Basic models
We employ Probit models to examine the relation between state ownership and CEO turnover ( Table 5). The probability/likelihood function of CEO turnover can be expressed as: Pr(turnover) = f (performance, largest shareholder shareholding, CEO age, CEO duality, CEO tenure, board structure, supervisory committee structure, fi rm size, (1) marketization, industry dummies, province dummies, year dummies), where f (·)is the standard normal cumulative distribution (cdf) in the Probit model, which can be expressed as: where ( ) φ ⋅ is the standard normal density.
(3) The interactive effect can be expressed as: Ai and Norton (2003) summarize three implications for Probit models based on equation (3) and (4). First, the interaction effect could be zero, even if 12 0 β ≠ . Second, the statistical signifi cance of the interaction effect cannot be tested using simple t-test on the coeffi cient 12 β . Third, the interactive effect is conditional on independent variables and may have different signs when independent variables take different values. The aforementioned three implications are usually overlooked in empirical studies 12 . To apply the approach proposed by Ai and Norton (2003), we follow Powers (2005) and estimate a Probit regression using the full sample. Then we split the sample into state-controlled and nonstate-controlled fi rms, and estimate Probit regressions for each sub-sample. Finally, we compute and contrast marginal effects of performance on turnover for two types of fi rms. Table 5 contains empirical results for Probit models 13 . As shown in the table, CEO turnover is negatively related to fi rm performance. The coeffi cients of current and lagged performance are statistically signifi cantly negative at the 1% level, suggesting that high (low) performance decreases (increases) the probability of CEO turnover. Specifi cally, on average, a decrease in ROA or lagged ROA by 10 percentage points increases the probability of CEO turnover by 0.041 and 0.056, respectively. This fi nding indicates that China's listed fi rms provide clear profi t incentives to CEOs so that poor-performing CEOs are more likely to be replaced.
Ownership concentration, as measured by the largest shareholder shareholdings, reduces the probability of CEO turnover. As we have argued in Section 2, there is usually one overwhelmingly large shareholder with substantial controlling power in listed fi rms. One way for large shareholders to pursue their private interests is to exert control over the selection of managers and make sure that top managers represent their interests 14 . When fi rms are dominated by large shareholders who seek their private interests, the risk of CEO turnover will likely be decoupled from fi rm performance and infl uenced by how well CEOs serve the private interests of large shareholders 15 . A higher level of ownership is associated with more power of control, and increases the probability that CEOs will cater to the private interests of the largest shareholders. As a result, CEO turnover is negatively related to the share ownership of the largest shareholders.
Regarding the board structure, the coeffi cients of duality are negative and statistically signifi cant at the 1% level, supporting the notion that holding the board chair position by the CEO weakens monitoring and incentive effects. Specifi cally, CEO duality reduces the probability of CEO turnover by 0.217. Consistent to Table 5. Probit regression analysis of CEO turnover at the publicly listed fi rms in China The models are estimated using fi rm-level data of the fi rms listed on the Shanghai or Shenzhen stock exchanges in China during the period 1999-2002. State-controlled fi rms are fi rms of which the state is the largest shareholder, while non-statecontrolled fi rms are fi rms of which the state is not the largest shareholder. The dependent variable, CEO turnover, equals 1 if the employment of the CEO was terminated during the year, and zero otherwise. Performance is measured with ROA. Lag(performance) is the one-year lagged performance (ROA). Largest Investor is the percentage of the shares held by the largest investor. CEO age is the age (in years) of the CEO. Duality equals 1 when the CEO holds the board chair position, and zero otherwise. CEO tenure is the number of the years the CEO has held the position till now. Board size is the number of directors on the board. Independent Director is the number of independent directors on the board. No. of Supervisors is the number of supervisors on the supervisory committee. Outside supervisor is the number of outside supervisors on the supervisory committee. Firm size is the natural logarithm of the total assets. Marketization is the NERI (National Economics Research Institute, China) marketization index that measures the overall market development in each province in China, including market competition, government regulation, and the legal environment. Industry, province and year dummies are included in the models, but their coeffi cients are not reported. P-values for two-tailed tests that the coeffi cient estimates equal zero are reported. The P-values are estimated using robust stand errors that account for potential heteroskedasticity and time series autocorrelation within each company (clustering by company). The coeffi cient estimates are transformed to represent the marginal effects evaluated at the means of the independent variables from the Probit regression. The marginal effect of a dummy variable is calculated as the discrete change in the expected value of the dependent variable as the dummy variable changes from 0 to 1. our expectation, CEO turnover is negatively related to board size but positively related to the number of independent directors. One extra independent director on the board increases the probability of CEO turnover by 0.034, after controlling for fi rm performance and other variables. The results provide some evidence that independent directors play a positive role in monitoring CEOs. Regarding the supervisory committee, the size of the supervisory committee does not exert signifi cant impact on CEO turnover. Similar to the independent director, the number of outside supervisors is positively related to the probability of CEO turnover, although the magnitude is relatively small. This result indicates that outside supervisors only exert modest impacts on improving internal incentive and monitoring schemes. Furthermore, we fi nd that the coeffi cient for fi rm size is negative and statistically signifi cant at the 5% level, suggesting that the likelihood of CEO turnover at larger fi rms is lower than at smaller fi rms. Finally, CEO turnover is positively related to CEO age but negatively related to CEO tenure, both are statistically signifi cant at the 1% level.
When the sample was split into state-controlled and non-state-controlled fi rms, CEO turnover is not related to either current or lagged fi rm performance for state-controlled fi rms. In contrast, CEO turnover is signifi cantly negatively related to current and lagged ROA at the 1% level for non-state-controlled fi rms. In particular, a decrease of ROA by 10 percentage points increases the probability of CEO turnover by 0.0513 and a decrease in lagged ROA by 10 percentage points increases the probability of CEO turnover by 0.0674.
Furthermore, the effect of board composition and supervisory committee on CEO turnover differs signifi cantly between state-controlled and non-state-controlled fi rms. For example, CEO duality reduces the probability of CEO turnover in both types of fi rms, but the magnitude of the impact is much stronger for statecontrolled fi rms. The coeffi cients for the number of independent directors and outside supervisors are positively related to CEO turnover for non-state-controlled fi rms, but are insignifi cant for state-controlled fi rms. Overall, the evidence indicates that state-controlled fi rms have more severe management entrenchment problems, weaker internal governance mechanisms and less effective managerial incentive schemes.

Relative performance evaluation and CEO turnover
To test the robustness of the fi ndings in Section 4.1, we estimate Probit regressions using industry-adjusted ROA (calculated as the difference between fi rm ROA and the industry median ROA) as an alternative meas-ure of fi rm performance. Estimation results are illustrated in Table 6.
As shown in Table 6, empirical results using the RPE variable are similar to those using ROA. For the full sample, the coeffi cients for RPE are negative and statistically signifi cant at the 1% level. On average, a decrease in RPE and lagged RPE by 10 percentage points increases the probability of CEO turnover by 0.04 and 0.057, respectively. In addition, a 10-percentage-point increase in ownership concentration reduces the probability of CEO turnover by 0.019. CEO turnover is positively related to CEO age while negatively related to CEO tenure. CEO duality reduces the probability of CEO turnover by 0.22. Controlling for the fi rm performance and other variables, one additional board member decreases the probability of CEO turnover by 0.009 while one additional independent director increases the probability of CEO turnover by 0.034. The supervisory committee size does not have signifi cant impact on CEO turnover. However, one additional outside supervisory committee member increases the probability of CEO turnover by 0.019. Moreover, fi rm size is negatively related to the probability of CEO turnover.
A comparison of estimation results for state-controlled versus non-state-controlled fi rms reinforces our previous fi ndings. The likelihood of CEO turnover is not related to either current or lagged RPE for state-controlled fi rms, but is signifi cantly negatively related to current and lagged RPE for non-state-controlled fi rms. A decrease in RPE or lagged RPE by 10 percentage points increases the probability of CEO turnover by 0.049 and 0.065, respectively. In addition, board and supervisory committee do not appear to affect managerial turnover among state-controlled fi rms as the coeffi cients for the number of independent directors and outside supervisors are insignifi cant. In contrast, the number of independent directors and outside supervisors are signifi cantly and positively related to CEO turnover for non-state-controlled fi rms. Furthermore, CEO duality reduces the probability of CEO turnover by a much larger magnitude for state-controlled fi rms than it does for non-state-controlled fi rms, suggesting that state-controlled fi rms have more severe managerial entrenchment problems. Overall, the above results bolster our previous fi ndings that state control weakens managerial incentives and internal monitoring.

Forced versus voluntary turnover
Our empirical results have shown that fi rm performance is negatively related to the likelihood of CEO turnover, and that state control weakens managerial incentives and internal monitoring. However, as we have discussed in Section 3.2, many CEO turnovers may not be performance-related. Following Dahya et al. (2002) and Fich and Shivdasani (2006), we conduct robustness tests using forced turnovers identifi ed in Section 3.2.  *, **, and *** indicate statistical signifi cance at the 10%, 5%, and 1% levels for two-tailed tests, respectively. a decrease in ROA and lagged ROA by 10 percentage points increases the probability of forced turnover by 0.017 and 0.026, respectively. For state-controlled fi rms, the probability of forced CEO turnover is not related to either current or lagged fi rm performance. In contrast, the probability of forced turnover is significantly related to both current and lagged ROA for nonstate-controlled fi rms. A decrease in ROA and lagged ROA by 10 percentage points increases the probability of CEO turnover by 0.021 and 0.032, respectively.
In addition, CEO duality reduces the probability of forced CEO turnover by about 0.1 among non-statecontrolled fi rms, but does not affect the probability of forced CEO turnover for state-controlled fi rms. Supervisory committee does not have any impact on the probability of forced CEO turnover for state-controlled fi rms, but the number of outside supervisors is positively related to the probability of forced turnover. The models are estimated using fi rm-level data of the fi rms listed on the Shanghai or Shenzhen stock exchanges in China during the period 1999-2002. State-controlled fi rms are fi rms of which the state is the largest shareholder, while non-statecontrolled fi rms are fi rms of which the state is not the largest shareholder. The dependent variable is forced CEO turnover. Performance is measured with ROA. Lag(performance) is the one-year lagged performance (ROA). Largest Investor is the percentage of the shares held by the largest investor. CEO age is the age (in years) of the CEO. Duality equals 1 when the CEO holds the board chair position, and zero otherwise. CEO tenure is the number of the years the CEO has held the position till now. Board size is the number of directors on the board. Independent Director is the number of independent directors on the board. No. of Supervisors is the number of supervisors on the supervisory committee. Outside Supervisor is the number of outside supervisors on the supervisory committee. Firm size is the natural logarithm of the total assets. Marketization is the NERI (National Economics Research Institute, China) marketization index that measures the overall market development in each province in China, including market competition, government regulation, and the legal environment. Industry, province and year dummies are included in the models, but their coeffi cients are not reported. P-values for two-tailed tests that the coeffi cient estimates equal zero are reported. The P-values are estimated using robust stand errors that account for potential heteroskedasticity and time series autocorrelation within each company (clustering by company). The coeffi cient estimates are transformed to represent the marginal effects evaluated at the means of the independent variables from the Probit regression. The marginal effect of a dummy variable is calculated as the discrete change in the expected value of the dependent variable as the dummy variable changes from 0 to 1. Overall, estimation results using forced turnover strengthen our previous fi ndings that state control weakens managerial incentives and internal monitoring schemes.

Compensation and performance -basic models and results
We now examine the impact of state control on managerial pay-performance sensitivity. Brickley and Van Horn (2002) provide evidence that managerial compensation is related to current or lagged fi rm performance. In China, although some forms of incentive pay such as monthly and quarterly bonuses are distributed within the same calendar year, year-end bonuses are typically distributed at the end of the lunar calendar year, which is often two or three months subsequent to the New Year's day. Therefore, annual managerial compensation might be related to current and lagged fi rm performance.
To investigate whether and to what extent state control affects the relation between executive compensation and fi rm performance, we control for fi rm-and individualspecifi c factors such as fi rm age, size, CEO age and CEO tenure. We also take into account internal governance mechanisms such as board composition and supervisory committee structure, and external governance mechanisms as proxied by provincial marketization index. We then estimate the following panel data regression using fi xed effect and random effect methods: where W it is the logarithm of annual CEO compensation for fi rm i at time t, 0i β is the fi rm fi xed effect that captures fi rm-specifi c unobserved characteristics potentially correlated to regressors 16 . Table 8 presents fi xed effect and random effect estimation results. As shown in the table, CEO compensation is positively related to both current and lagged fi rm performance. A 1 percentage point increase in ROA and lagged ROA will increase the CEO compensation by 0.836-0.964% and 0.724-0.955%, indicating that fi rms provide clear incentives to tie fi rm performance with executive pay. In addition, there is some evidence that state ownership reduces CEO compensation, as the coeffi cient for STATE is signifi cantly negative in the random effect model but insignifi cant in the fi xed effect model. The coeffi cient estimates for the interaction term between lagged performance and state control is negative and statistically signifi cant at the 5% level in both fi xed effect and random effect models while the coeffi cient estimates for the interaction term between current performance and state control is not statistically signifi cant. The results indicate that state control weakens pay-performance sensitivity, but for lagged performance only. In unreported empirical analyses, we split the full sample into state-controlled and nonstate-controlled fi rms and estimate the regressions using each sub-sample. We fi nd that CEO compensation is signifi cantly positively related to fi rm performance for non-state-controlled fi rms, while insignifi cant for state-controlled fi rms. Overall, our results indicate that state control weakens managerial incentive schemes.
Moreover, CEO compensation is negatively related to ownership concentration (the percentage of shares held by the largest shareholder) but positively related to CEO age. Controlling for fi rm performance, the number of outside supervisors is negatively related to CEO pay. In particular, an additional outside supervisor reduces CEO compensation by 4-5%. Firm size is signifi cantly positively related to CEO compensation in the random effect model. A 10% increase in fi rm size increases CEO compensation by 2.2%. Finally, provincial marketization is signifi cantly and positively related to CEO compensation. An increase of marketization index by 1 increases the CEO compensation by 9-18%, indicating that CEO compensation is higher in more developed regions in China. Table 9 contains estimation results for regressions of CEO compensation using RPE as an alternative measure for fi rm performance.

Robustness tests -CEO compensation and RPE
As shown in the table, CEO compensation is significantly positively related to current and lagged RPE. A 1 percentage point increase in RPE and lagged RPE increases CEO compensation by 0.77-0.96% and 0.61-0.92%, respectively. In addition, state control has marginally negative impact on CEO compensation in random effect model with interaction term. The coeffi cient estimates of the interaction term between lagged performance and state control is negative and statistically significant in both the fixed effect and random effect models, while the coefficient estimates of the interaction term between current performance and state control are insignifi cant. This table reports results from fi rm-level random effects and fi xed effects (within) regressions that estimate the impact of fi rm performance on CEO compensation in different types of companies using fi rm-level unbalanced panel of the fi rms listed on the Shanghai or Shenzhen stock exchanges in China during the period 2000-2002. State-controlled fi rms are fi rms of which the state is the largest shareholder, while non-state-controlled fi rms are fi rms of which the state is not the largest shareholder. The dependent variable is the natural log of total annual compensation of the CEO. Performance is measured with ROA. Largest Investor is the percentage of the shares held by the largest investor. CEO age is the age (in years) of the CEO. Duality equals 1 when the CEO holds the board chair position, and zero otherwise. CEO tenure is the number of the years the CEO has held the position till now. Board size is the number of directors on the board. Independent Director is the number of independent directors on the board. No. of Supervisors is the number of supervisors on the supervisory committee. Outside Supervisor is the number of outside supervisors on the supervisory committee. Firm size is the natural logarithm of the total assets. Marketization is the NERI (National Economics Research Institute, China) marketization index that measures the overall market development in each province in China, including market competition, government regulation, and the legal environment. P-values for two-tailed tests that the coeffi cient estimates equal zero are reported.  This provides some evidence that state control weakens managerial incentives by reducing the pay-performance sensitivity. Furthermore, CEO compensation is negatively related to ownership concentration but positively related to CEO age. Controlling for fi rm performance, the number of outside supervisors is negatively related to fi rm performance. Firm size is signifi cantly positively related to CEO compensation, but in random effect models only. As before, the degree of provincial marketization is signifi cantly positively related to CEO compensation. An increase of marketization index by 1 increases CEO compensation by 9-18%.

Conclusion
Numerous empirical studies on privatization fi nd evidence that private ownership is associated with better fi rm performance than state ownership is. However, there is a long-standing theoretical debate on links between ownership and incentives. This study adds to the debate by providing empirical evidence on whether and to what extent state ownership and control affects managerial incentives among partially privatized fi rms in China.
Using data from a large sample of publicly listed fi rms in China during 1999China during -2002, we fi nd that both CEO turnover and CEO compensation are related to fi rm performance as measured by ROA and RPE, suggesting that fi rms provide clear performance-based incentives to motivate CEOs. Furthermore, the likelihood of CEO turnover is not related to either current or lagged fi rm performance for state-controlled fi rms while it is signifi cantly negatively related to both current and lagged fi rm performance for non-state-controlled fi rms. The results are robust to performance measures and turnover classifi cations. Moreover, state ownership is associated with lower pay-performance sensitivities in both fi xed effect and random effect models of managerial compensation. Overall, empirical results based on Chinese data render support to the theory that state ownership and control weaken managerial incentives, and can partially explain the ineffi ciency of state ownership found in previous studies, e.g. Xu and Wang (1999), Sun and Tong (2003) and Wei et al. (2005).
We also explore the relation between state control and the effectiveness of internal governance mechanisms such as board composition and supervisory committee. We fi nd that the number of independent directors and outside supervisors do not have signifi cant impact on CEO turnover among state-controlled fi rms. In contrast, the number of independent directors and outside supervisors are signifi cantly and positively related to the probability of CEO turnover among non-state-controlled fi rms. In addition, CEO duality reduces the probability of CEO turnover by a much larger magnitude among state-controlled fi rms, indicating that statecontrolled fi rms are subject to more severe management entrenchment problems. Furthermore, our results show that state ownership and control also weaken the effectiveness of board and supervisory committee.
Endnotes 1 According to Megginson (2005), the cumulative value raised through privatization in more than 100 countries now probably exceeds 1.5 trillion. In the transition economies, more than 150,000 large enterprises have experienced revolutionary changes in ownership and economic environment over the past decade (Djankov and Murrell 2002). 2 Tirole (2001) defi ned corporate governance as the design of institutions that induce or force management to internalize the welfare of stakeholders. The control structure and provision of managerial incentives are regarded as the most important factors to induce or force the internalization. Cragg and Dyck (2003) examine the impact of privatization on pay-performance sensitivity in 112 privatized fi rms in the U.K. They fi nd that there is no relationship between compensation and fi rm performance in SOEs but strong pay-performance sensitivity in privatized fi rms both before and after corporate governance reforms. 3 China Securities Regulatory Commission (CSRC) uses fi nancial performance as an important criterion in determining an SOE's eligibility for the stock market. An SOE must report three consecutive years of profi ts before applying for an IPO; and listed companies that report three consecutive years of losses will be delisted from the stock exchange. 4 As of February 2006, shares of over 100 publicly listed fi rms have moved from "partially tradable" to "fully tradable", or have become the so-called "G shares". During this process, the holders of non-tradable shares have agreed to provide the holders of tradable shares with free shares, cash, warrants or some other means of compensation in exchange for their shares to become tradable. 5 In China many SOEs carve out their most profi table assets and businesses into a joint stock company for the IPO in order to raise capital in the stock market. As a result, many listed fi rms in China have their parent companies as controlling shareholders. CEOs of these fi rms can be assigned by parent companies, and their turnover can be due to a change in their work assignment. The large proportion of CEO turnover reported as change in work assignment represents a unique feature of the Chinese context. 6 Our search is based on three data resources: the annual reports of the fi rms, CSMAR database and the information of China's listed fi rms available at http://www.jrj.com.cn. 7 Firms usually report compensations for the top three executives, including total annual compensation for CEOs and two other highest-paid executives (often vice CEOs). We use the average compensation for the top three executives to proxy for CEO compensation in this study. We compile compensation data from annual reports of listed fi rms during the period years 2000-2002. The data for the year 1999 are not included due to numerous missing observations. 8 With approval from the CSRC, a company can issue employee shares to managers and employees, typically at a signifi cant discount, at the time of Initial Public Offering (IPO). Employee shares are not tradable in the stock exchanges, although some companies have established Shareholding Association or Workers' Union to buy back shares in case that an employee retires, resigns, gets fi red or dies. In such cases, employee shares are generally priced on the basis of their net asset value and determined by managers. 9 Stock return is another frequently used performance measure. However, for the following reasons, it is not used in this study. First, the majority of shares in China's listed fi rms is held by the state and legal entities and not tradable. Hence, fi rm profi tability would be more important than stock return in evaluating managerial performance. Second, changes in stock prices are poor indicators of changes in fi rm fundamental values in China (Allen et al. 2005). Morck et al. (2000) fi nd that more than 80% of stocks listed on the two exchanges in China move in the same direction in a given week, which suggests that stock returns are not informative of fi rm performance, because stock returns tend to refl ect market-wide information rather than fi rm-specifi c information. Despite the above arguments, we use stock returns as a performance measure in robustness analysis and fi nd that it has no impact on CEO turnover and compensation. 10 According to the Company Law, a board of directors should consist of 5 to 19 directors. The average board size in our sample is 9.6. 11 According to the "Guidelines for Establishing Independent Director System in Listed Firms", prerequisites for independent director include: (a) neither the individual not his relatives work for the listed fi rms or its subsidiaries; (b) the individual does not directly or indirectly own more than 1% of shares of the listed fi rm; (c) neither the individual nor his close relatives (including parents, spouses and children) work for one of the largest 5 shareholders or a shareholder that owns more than 5% of shares of the listed fi rm. 12 Ai and Norton (2003) reviewed 73 published papers on JSTOR between 1980 and 1999 and found that none of the studies interpreted the coeffi cient on the interaction term correctly. Similarly, Powers (2005) surveyed 38 papers in the fi nance fi eld, only 5 of them interpret the results appropriately. 13 The reported p-values are estimated using robust stand errors that account for potential heteroskedasticity and time series autocorrelation within each company (clustering by company). The coeffi cient estimates are transformed to simplify their economic interpretation. In Table 5, the reported coeffi cients represent marginal effects evaluated at the means of the independent variables from the Probit regression. The marginal effect of a dummy variable is calculated as the discrete change in the expected value of the dependent variable when the dummy variable changes from 0 to 1. 14 According to the "Code of Corporate Governance for Listed Firms in China" issued by the CSRC, a controlling shareholder makes recommendations to the board of directors regarding the appointment of directors. The shareholders' meeting votes on whether to accept the recommendations. Thus the key player in the appointment of top managers is the controlling shareholder (Firth et al. 2006). 15 Shleifer and Vishny (1997) point out that, "as ownership gets beyond a certain point, the large owners gain nearly full control and are wealthy enough to prefer using fi rms to generate private benefi ts of control that are not shared by minority shareholders". 16 The rationale for presenting the random effect results together with the fi xed effect results is that controlling for fi rm fi xed effect may reduce the signifi cance of the coeffi cients because the coeffi cients are derived based on the within-fi rm changes of variable such as ownership structure, governance structure and fi rm size, which may be relatively constant over a short time period. In fact, for most panel data, particularly short panels such as this (3 years in our case), most of the variation of the data is in the cross-sectional dimension. Applying a fi xed effect model not only eliminates the invisible fi rm specifi c characteristics but also wipes out useful inter-fi rm variation, which may account for most of the total variation (Zhou 2001).