Controlling Shareholders Liquidity Constraints and Corporate Payout Policies ABSTRACT

Transcription

1 Controlling Shareholders Liquidity Constraints and Corporate Payout Policies ABSTRACT We show that firms that are partially affiliated with a business group reduce dividend payouts when their controlling shareholder experiences liquidity constraints. Our identification strategy relies on a series of reforms in China that mandate central state owned enterprises (CSOEs) to return to the government a proportion of consolidated net income of their controlled business groups. Empirical analyses reveal that listed firms controlled by these CSOEs experience a significant reduction in dividend payouts upon the inception of these reforms. The dividend reductions are concurrent with an increase in within-group borrowings and related party transactions, i.e. two primary channels of reallocating resources within the group. The dividend cut around CSOE reforms reflects CSOE managers incentives to seek better career outcome, and is more pronounced when managers face a higher likelihood of demotion in the near future. Finally, the reduction in listed firms dividend payout hurt minority shareholders interests, manifested in lower level of valuation after the reform. We conclude that controlling shareholders liquidity constraints exert a causal and first-order impact on listed firms payout policies. JEL classifications: Dividend payout, Controlling shareholder, Expropriation, Business Group Keywords: G30, G28, G35, P1 1

2 Controlling Shareholders Liquidity Constraints and Corporate Payout Policies 1. INTRODUCTION Traditional models of dividend payout rely on the Fisher Separation Theorem, i.e., all investors prefer managers to adopt value-maximizing payout policies (DeAngelo, DeAngelo and Skinner, 2008). However, as DeAngelo et al. (2008) note, controlling shareholders may prefer non-value-maximizing payout policies due to their idiosyncratic utility functions. In this study, we examine whether controlling shareholders liquidity constraints affect payout policies. There are two opposing views relating to this research question. The first view contends that, with liquidity constraint, controlling shareholders have incentives to maintain resources of firms under their control. Inter-corporate borrowings and related party transactions within a business group provide latitudes for controlling shareholders to reallocate retained resources across firms within the group, particularly so when minority shareholders are not well protected (Faccio, Lang and Young, 2001; Jiang, Lee and Yue, 2010; Johnson, La Porta, Lopez-de-Silanes and Shleifer 2000; La Porta Lopez-de-Silanes, Shleifer and Vishny, 2000). On the contrary, the second view holds that controlling shareholders can increase corporate payout and receive cash injections directly through dividends. For instance, Gopalan, Nanda and Seru (2014) show that cash-rich firms within a business group payout dividends to be used as other group peers investments. To disentangle the two views, we employ an identification strategy that relies on a series of staggering reforms in China introducing shocks to controlling shareholders liquidity. In 2007, the Chinese government initiated the State Capital Operation Program which mandates central state-owned enterprises (CSOEs) to begin turning in a proportion of their profits to the state. The Chinese government justifies this regulatory reform with the argument that, as CSOEs are formed entirely through state investment, the state has the right to collect its investment returns (i.e. the economic profits of CSOEs). The initial mandate requires CSOEs in monopolistic industries and those in generally competitive industries to turn in 10% and 5% of their annual net income, respectively. Later in 2010, both proportions were further increased by 5%. At the same time, the state required more CSOEs to begin turning in 5% of their net income. A third revision was made in Jan, 2012 whereby additional CSOEs were required to begin turning in profits. The latest revision was made in April The proportions of returning-profit were increased by an additional 5%. Together, the series of government mandates generate plausibly exogenous variation, both cross-sectional and inter-temporal, in the extent of liquidity shocks applying to CSOEs. The mandated rules, in combination with the organizational structure of business groups in China, renders the reform ideal to examine how controlling shareholders liquidity constraints 2

3 affect corporate dividend payout. First, the historical wave of partial privatizations results in CSOEs becoming controlling shareholders of many listed firms. The pervasive partial privatizations starting in 1978 in China have formed business groups representing a significant portion of the economy (Sun and Tong, 2003). In an effort to transform a centrally controlled economy to a market economy, state owned assets spin off divisions while retaining significant stakes through pyramidal layers (Fan, Wong and Zhang, 2013). Such a reform process produces a significant number of pyramidal business groups in which central state-owned enterprises (CSOEs) are located at the top of the pyramid, controlling other state-owned enterprises (SOEs) through an ownership chain. 1 Second, the staggering adoption of the State Capital Operation Program facilitates a difference-in-differences analysis to estimate the causal effects of CSOEs liquidity shocks on their group members operational and financial policies. We begin our analyses by estimating the effect of CSOE reform on dividend payout of firms with CSOEs as controlling shareholders (CSOE-controlled firms). We document a significant reduction in cash dividends of CSOE-controlled firms upon the inception of the reform. An analysis of the dynamic effects around the reform reveals no significant change in controlled firms dividend payout prior to the reform, confirming the exogeneity of the reform. Further, the dividend reduction is persistent in time and large in magnitude. Incorporating the potential differences between firms controlled by CSOEs (treatment firms) and those controlled by other entities or individuals (control firms), we employ several matching procedures and find that our inferences continue to hold. Combined, these findings are consistent with the notion that CSOEs tend to retain resources within the business group after the government mandates a proportion of the group-level income. Our earlier discussions hinge on the presumption that retained resources through cutting dividends of listed firms can be reallocated within the business group and utilized by peer firms within the group. We provide related tests to examine whether intra-group resource reallocation intensifies after the passage of the CSOE reform. We consider two alternative mechanisms that the controlling entity can shift resources away from the listed firm: (1) inter-corporate loans with the listed firm as creditors (Jiang et al., 2010) (2) the listed firms purchase of commercial products from related parties within the same business group. The latter transaction type is the predominant form of Chinese firms related party transactions. Further, the direction of the trade involves the listed firm cash outflows. We find that, after the passage of the CSOE reform, the listed firm experiences a significant increase in the proportion of inter-corporate loans to total assets. Further, it also exhibits an increase in its purchases of commercial products from related parties and no change in its sale of commercial products to related parties. We conclude from this set of analyses that controlling shareholders (CSOEs) are able to reallocate the resources 1 Our empirical sample is thus representative of organizations in other emerging markets with historical partial privatizations, such as India, Russia and Brazil (Megginson and Netter, 2001; Gupta, 2005). 3

4 associated with the reduced dividends within the business group. The tendency to maintain resources within the group and the ensuing dividend reduction of CSOEs-controlled listed firms can be traced to CSOE managers political incentives. Unlike the private sector whereby managers behaviors are mostly affected by monetary incentives, the public sector is shaped to a greater extent by managers political incentives. CSOE managers who are also government officials compete in a closed pyramidal labor market to seek higher political ranks (Chen, Kim, Li and Liang, 2017; Deng, Morck, Wu and Yeung, 2015). They have strong incentives to seek political promotions, avoid demotions and retain their positions after the official retirement age (Yang, Wang and Nie, 2013). Their career outcome depends heavily on the SASAC s triennial evaluations which assign significant weight to economic indicators (e.g., asset value appreciation and economic-value-added) and harmonious indicators (e.g., avoiding layoff). Maintaining group-level income and profitability and avoiding employee layoffs therefore become critical for CSOE managers career concern. We perform two lines of analyses to seek related evidence. First, we find that a greater reduction of controlled firms dividend payout is associated with better career outcome for the CSOE manager in the future. The result is largely driven by a lower incidence of future demotions. Second, we model the effect of CSOE reform on controlled firms dividend payout as a function of CSOE manager s political incentives. We find that, the association is more pronounced when a CSOE manager received worse rating in her last evaluation, is serving her evaluation year, is elder or closer to official retirement age, i.e. when CSOE managers have a greater likelihood of demotion. These results combine to suggest that an incentive to avoid demotion plausibly explains CSOE managers decisions to reduce controlled firms dividends and maintain resources within the business group. Finally, we examine whether minority shareholders discount the treatment firms share valuation after seeing the increased level of expropriation activities. We argue that, controlling shareholders preference to maintain resources for the business group can bring harm to minority shareholders of their partially controlled listed firms. More specifically, we estimate the change in the valuations of expropriated firms around the treatment of CSOEs within the same pyramids. We find that Tobin s Q of expropriated firms significantly decreases. The pattern of the valuation effect in different years around the treatment suggests a lag in investors responses. The reduction in firm value began two years after the treatment year. Empirical analyses here suggest that outside shareholders discount a firm s shares upon the realization of their invested firms resources being reallocated to other group members receiving adverse cash flow shocks. Our study contributes to the payout literature by documenting how controlling shareholders idiosyncratic incentives induce non-value maximizing dividend policies, an area categorized in DeAngelo et al. (2008) as important and under-explored. The importance of this research 4

5 question is further elevated by the prevalence of concentrated ownership structure around the world. Outside the U.S, it is common for a firm to have a controlling shareholder, being either an individual (e.g., the founder of a family firm) or an institution (e.g., a financial institution or government entity). Further, even in the U.S., block ownership also prevails (Holderness, 2010; Edmans, 2014). Related to this field, Perez-Gonzalez (2003) shows that controlling shareholders tax profiles shape corporate payout policies. We note that, DeAngelo el al. (2008) outline two conditions under which the controlling shareholder is incentivized to seek non-value maximizing dividend policies. First, a value-maximizing payout level does not best suit the controlling party s consumption needs. Second, trading the firm s shares would weaken her control over the firm. Our testing ground utilizing the series of reforms imposed on Central State-owned enterprises are precisely compatible with both conditions. As the reforms mandate a significant proportion of group-level resources to be returned to the state, the CSOE has a heightened incentive to retain resources within the group to maintain pre-reform operations (i.e., consumptions at a group level). Further, offloading the listed firm s shares will result in the CSOE, and therefore the state, relinquishing their control over the listed firm. Utilizing this setting, we suggest that controlling shareholders incentives to maintain resources within their controlled business groups result in a significant reduction of cash dividends of their partially controlled listed firms. Second, and more broadly, our study contributes to the literature on the consequences of having controlling shareholders or blockholders (Edmans, 2014). Instead of maximizing firm value, controlling shareholders likely extract private benefits. These benefits, regardless of the source, are enjoyed exclusives by the owner in control (Dyck and Zingales, 2004). However, as Edmans (2014) highlights, examining the causal impact of block ownership on corporate policies is empirically challenging due to either the two-way relationship between them or omitted variables simultaneously affecting the two factors, or both. We address this issue by introducing exogenous variation in controlling shareholders incentives to retain resources and estimating the causal effects of these varying incentives on controlled firms dividend policies. The remainder of our article proceeds as follows. Section 2 introduces the organizational structure of business groups and describes the series of reforms implemented by the Chinese government to demand government-owned business groups to return profits. Section 3 describes our sample and identification strategies. Section 4 presents empirical findings and our inferences. Finally, Section 5 concludes. 2. INSTITUTIONAL BACKGROUND 2.1 Partial privatization and CSOE controlled business groups 5

6 Upon the inception of Shenzhen and Shanghai stock exchanges in the 1990s, state owned enterprises in China were instructed to assist the development of the capital market in general, and stock listing in particular, by spinning off a proportion of their assets. As a result, SOEs organized equity carve-outs to form subsidiaries eligible for listing on stock exchanges. However, as Fan, Morck and Yeung (2012) note, the Chinese government has been careful to uphold the principle of Market Socialism with Chinese Characteristics with the concern of a potential shift in political ideology from socialism to capitalism. Therefore, the state only floated minority interests in these listed subsidiaries while maintaining the controlling stakes. This partial privatization essentially results in SOEs controlling business groups where the listed firms inhabit. This organizational structure is similar with family-controlled business group commonly seen in other countries, e.g. business chaebols in South Korea (Bae, Kang and Kim, 2012). In Appendix 2, we provide an illustrational example using the business group controlled by China Resources National Corporation, itself a CSOE. In March 2003, the State Council of China formed the State-owned Assets Supervision and Administration Commission (SASAC) as the investor of previously state-owned assets on the government s behalf. The formation of SASAC lies under then-time government objective to reform SOEs to facilitate the clarification of property rights. As a unique institution, the SASAC controls each CSOE which itself is at the apex of a business group comprising listed firms and their peer firms. More importantly, the SASAC is empowered to formalize and release regulations that CSOEs under their control have to obey. 2.2 The state operation program On Sep 8, 2007, the State Council (Guo Wu Yuan) released the Opinions of The State Council on The Pilot Implementation of State-owned Capital Operating Budget (Guofa No. 26, 2007), mandating central stated-owned enterprises (CSOEs) to begin returning profits to the state. The requirement only applies to CSOEs over whom the government has direct and complete control (i.e. 100% share ownership). The Chinese government justifies this regulatory reform with the argument that, as CSOEs are formed entirely through state investment, the state has the right to collect its investment returns (i.e. the economic profits of CSOEs). Subsequently, on Dec 11, 2007, the Ministry of Finance (MOF) and the State-owned Assets Supervision and Administration Commission (SASAC) jointly announced a mandate requiring CSOEs to begin returning profits to the state. The initial mandate groups CSOEs into four categories: Category I: China Tobacco Corporation; Category II: Central State-Owned Enterprises in monopolistic industries, including Petroleum, Mining, Telecommunication, and Electrical power; Category III: Central State-Owned Enterprises in generally competitive industries, including Transportation, Steel, Trade and commerce, and Construction, etc; Category IV: Central State-Owned Enterprises in military related industries and those affiliated with scientific research institutions; 6

7 Category V: China Grain Reserves Corporation and China Cotton Reserve Corporation. Starting in Year 2007, CSOEs in Category I and Category II are mandated to return 10% of their net income. CSOEs in Category III are mandated to return 5% of their net income. CSOEs in Category IV were given a three-year exemption after which they should also begin to return profits to the state. Finally, CSOEs in Category V are not mandated to return any profit. On Dec 23, 2010, the MOF announced a second mandate introducing following revisions to the program. First, CSOEs in Categories I, II, III, and IV increase the proportions of returning-profits by 5% to 15%, 15%, 10%, and 5%, respectively. CSOEs in Category V remain exempted from the mandate. Further, the mandate extended beyond the existing CSOEs and included CSOEs affiliated with following government agencies: Ministry of Education, Ministry of Culture, Ministry of Agriculture, the State Administration of Radio, Film and Television, and so on. These newly added CSOEs began to return 5% of their net income to the state. On Jan 13, 2012, another mandate further included CSOEs affiliated with following government agencies: Ministry of Health, Ministry of Information and Technology, the State Administration of Sports, and so on. Again, these newly added CSOEs began to return 5% of their net income to the state starting in Fiscal Year In addition, the returning-profits proportion of CSOE in Category I (i.e. the China Tobacco Corporation) was increased by 5% to 20% in On April 17, 2014, the MOF announced that CSOEs, except those in Category V, increase the proportion of returning-profit by another 5%. In the table below, we present the percentages of CSOEs profits claimed by the state by categories and sub-periods. We summarize the series of reforms pertaining to CSOEs profit returning in Appendix 1. By mandating CSOEs to return a proportion of the consolidated net income of the business group, the series of reforms introduced exogenous and significant shocks to group-level resources. To maintain pre-reform level of operations and investing activities, CSOEs experience an increase in the incentive to maintain resources within the business group. Building upon this institutional design, we explore whether controlling shareholders increased incentives to retain resources within the group translates into lowering pro rata dividends distributed to outside shareholders. 3.1 Sample and Data Sources 3. Sample and Identification Strategy Our empirical analyses utilize data from different sources. We broadly categorize the data as follows Data on CSOEs Profit-returning Information We obtain filings of government mandates to identify CSOEs that are required to return profits to the state, along with the proportions of their profits to be claimed by the state. We manually code these data by reading each filing. The filings corresponding to the four mandates are: [Caiqi No. 309, 2007] - The Notice of Interim Procedures to Administrate Central State-owned 7

8 Capital Gains ; [Caiqi No. 392, 2010] - The Notice of Revising The Budget Plan of Central State-owned Capital issued by the Ministry of Finance; [Caiqi No. 3, 2012] - The Notice of Expanding The Budget Plan of Central State-owned Capital issued by the Ministry of Finance; [Caiqi No. 59, 2014] - The Notice of Further Increasing the Returning Proportion of Central State-owned Capital Gains Data on accounting and market variables We obtain data on firm fundamentals and stock prices from the China Stock Market and Accounting Research (CSMAR) database. Our sample period covers 2003 to We begin with Year 2003 because the SASAC who currently owns and supervises CSOEs was established in Year Our sample consists of two groups of firms: [1] firms whose controlling shareholders are central state-owned enterprises (CSOEs), and [2] non-state owned enterprises 2. To identify firms whose controlling shareholders are CSOEs, we manually read information of each public firm s ownership chain disclosed in its annual report. If a firm s controlling shareholder is a CSOE included in the State Capital Operation Program, then it is included into our first group of firms. The second group of firms remains unaffected by the CSOE reform. Such a sample consists of both treatment and benchmark firms, facilitating a difference-in-differences empirical design. We begin with 12,531 firm-year observations. We exclude 112 observations for firms in the financial industry. Further, we omit 1,817 observations with missing values for variables required in the empirical analyses. Our main empirical sample thus consists of 10,602 firm-year observations representing 1,645 unique firms Data on political incentive We manually collect information on CSOE managers employers, positions, political ranks, and demographic information by searching SASAC s website, CSOEs website, Baidu, etc. We obtain information on CSOE managers historical performance in political evaluations from SASAC s website. 3.2 Identification strategy Since the reform has the merit of a staggering-adoption program, we perform the following difference-in-differences analysis: y i,t = α i + α t + δpr i,t + γx i,t + ε i,t, (1) 2 Our empirical sample excludes another subgroup of publicly listed firms SOEs controlled by the local governments due primarily to the inadequate amount of information. Although local governments also began to initiate profit-returning programs for their controlled SOEs in 2007, regulatory filings lack information on the identities of SOEs selected into their mandates and the corresponding profit-returning proportions. 8

9 where PR indexes the post-mandate period during which a firm s group member is required to return a proportion of its profits to the state; α i and α t denote fixed effects for firm and year; X i,t is a set of covariates. The estimate of δ suggests the impact of a CSOE s liquidity shock on its group member s economic outcome examined. As firms also differ in their percentages of profits to be claimed by the state, we further augment Equation (1) and estimate the following regression: 4.1 Summary statistics y i,t = α i + α t + δ 1 PR_5% i,t + δ 2 PR_10% i,t + δ 3 PR_15% + i,t + γx i,t + ε i,t, (2) 4. Empirical Analyses In Table 1, we report summary statistics for key variables used in our empirical analyses. An average firm pays out 21.8% of its earnings as cash dividends and has a dividend yield at There are 16.8% post-reform firm-year observations out of the entire sample. Return on assets (ROA), measuring profitability, averages at 3.1%. The relatively lower level of profitability potentially explains the smaller magnitude of dividend payout, compared with firms from the US capital market. Related, free cash flow deflated by total assets (FCF) has a negative mean value, again suggesting a lack of resources to be distributed as cash dividends. In terms of ownership structure, the controlling shareholder on average owns 35.2% of our sample firms shares, revealing a typical concentrated ownership structure. [Insert Table 1 around Here] 4.2 The CSOE reform and controlled firms dividend payout Documenting the main effects Our primary hypothesis states that, after the reform requires Central State Owned Enterprises (CSOEs) to turn in a proportion of their profits to the state, CSOEs have an incentive to expropriate partially-owned listed firms to maintain a pre-reform level of resources for operations and investments. 3 To test this hypothesis, we employ the following regression model. D/E (D/M) i,t = α i + α t + δ 1 PR i,t + γx i,t + ε i,t, (3) In Equation (3), the dependent variable measures firm i s dividend payout in year t. α i and α t indicate fixed effects for firms and years, respectively. We employ two commonly used 3 Such an incentive is also mentioned by CSOE managers and government officials in the media. 9

10 constructs for a firm s dividend payout. D/E is computed as the dividend-to-earnings ratio and D/M is computed as the dividend yield. PR is an indicator that equals one if a firm s controlling shareholder is mandated to return non-zero proportion of group-level consolidated net income to the state, and zero otherwise. X indicates a vector of control variables that are shown in prior literature to affect dividend payout. We cluster standard errors by firm to address the concern over potential serial correlations of the residual term. We are interested in the coefficient δ 1 which captures the causal effect of the reform on treatment firms dividend payout. We report empirical results in Table 2. The coefficient on PR is (t = -3.40) in column (1) when the dependent variable is dividend-earnings ratio (D/E). Considering the average dividend-earnings ratio at 0.218, this effect is practically important, approximating one-fourth of an average firm s dividend payout. In Column (3) where the dependent variable is dividend yield (D/M), the coefficient on PR remains negative and significant (-0.003, t = -4.25). 4 A second appealing feature of the series of reforms lies in the fact that the proportion of profit to return for a CSOE increases in subsequent mandates. Related to our main hypothesis, the effect of the CSOE reform on listed firms dividend payout should become stronger when the proportion increases. To test this prediction, we decompose PR into PR_5%, PR_10% and PR_15% +. The coefficients on the three indicators are mostly negative, although less so when the required returning proportion is low. D/E (D/M) i,t = α i + α t + δ 1 PR_5% i,t + δ 2 PR_10% i,t + δ 3 PR_15% + i,t + γx i,t + ε i,t, (4) In Equation (4), PR_5%, PR_10% and PR_15% + are indicators that equal to one when a firm s controlling shareholder is mandated to return 5%, 10% or more than or equal to 15% of the group-level consolidated net income to the state, respectively. The patterns of coefficients δ 1, δ 2 and δ 3 are consistent with our predictions. More specifically, when we employ dividend-earnings ratio as the dependent variable, the magnitude of δ 3 is greater than that of δ 2, which further exceeds that of δ 1. Further, the coefficient on PR_5% is even statistically insignificant, suggesting a weak initial effect. When we employ dividend yield as the dependent variable, the magnitudes of δ 2 and δ 3 are greater than that of δ 1, again suggesting a more pronounced impact when the mandated returning proportion is higher. Overall, empirical findings here suggest that after the series of mandates requiring CSOEs to return a proportion of their profits, dividend payouts of their partially controlled listed firms 4 In unreported analyses, we cluster standard errors by industries and re-estimate Equation (3). This adjustment is to address the concern that firms in different industries receive treatment at different points in time, creating potential within-industry correlation of the residual term. We obtain qualitatively similar findings. 10

11 significantly decrease. [Insert Table 2 around Here] The validity of the identification strategy above requires treatment and control firms to have similar dividend payout before the treatment s occurrence. To assess this, we follow Jiang, Levine and Lin (2016) to perform the following regression: D/E (D/M) i,t = α i + α t + δ -3 PR -3 i,t + δ -2 PR -2 i,t + δ -1 PR -1 i,t + δ 0 PR 0 i,t + δ 1 PR 1 i,t + δ 2 PR 2 i,t + δ 3 PR 3+ i,t + ε i,t, (5) where PR -n i,t indicates the nth year prior to the treatment year, and PR n i,t indicates the nth year after the treatment year. PR 3+ i,t indicates the 3 rd year after the treatment year, along with years after the 3 rd year. α i and α t again are fixed effects for firms and years, respectively. In Equation (5), coefficient δ n captures the difference in abnormal dividend payout between treatment and control firms at different points in time around the treatment year. Figure 1 plots the estimated coefficients and the 5% confidence intervals. We have two major observations. First, pre-reform, average dividend payout of treatment firms and that of control firms move in parallel. Such a pattern suggests limited evidence of firms responding to the reform before its enactment. Second, treatment firms experience an immediate and significant decline of average dividend payout after the reform, compared with control firms. Combined, graphical evidence in Figure 1 suggests the exogeneity of the reform and its causal effect on treatment firms dividend payout. [Insert Figure 1 Here] Falsification test dynamic effects around the reform To ensure that the reform provides an unexpected and persistent shock to CSOEs, we perform a falsification test in the spirit of Bertrand and Mullainathan (2003). We decompose the variable PR into dummy variables indicating years surrounding the reform, namely BEFORE 1, AFTER 0, AFTER 1, AFTER 2, and AFTER 3+, and estimate the following equation: D/E (D/M) i,t = α i + α t + δ 1 BEFORE 1 i,t + δ 2 AFTER 0 i,t + δ 3 AFTER 1 i,t + δ 4 AFTER 2 i,t + δ 5 AFTER 3+ i,t + γx i,t + ε i,t, (6) Should the CSOE reform exert a causal impact on listed firms dividend payout as we conjecture, we have two predictions. First, the coefficient on BEFORE 1 should be statistically insignificant, suggesting no pre-reform dividend payout adjustment. Second, coefficients on dummy variables indicating post-reform years should be consistently negative and statistically significant, suggesting a consistent post-reform adjustment in CSOEs dividend payout. 11

12 Empirical results reported in Table 3 are in line with these predictions. In both specification where we employ D/E and D/M as the dependent variable, respectively, BEFORE 1 consistently load insignificantly. The coefficient on BEFORE 1 captures the difference in treatment firms dividend change from earlier years to one year before the treatment year and the control firms dividend change during the same period. An insignificant estimate suggests that the two firms exhibit parallel trends prior to the treatment. Further, coefficients on variables indicating post-reform periods (AFTER 0, AFTER 1, AFTER 2, AFTER 3+ ) are consistently negative and statistically significant, except the coefficient on AFTER 0 in Column (1). This set of coefficient estimates combine to reveal a significant and persistent impact of the CSOE reform on listed firms payout policies. Overall, empirical evidence here further supports the exogeneity of the reform and the validation of our main findings Robustness matching analyses [Insert Table 3 around Here] Our primary sample comprises both treatment firms and control firms. Treatment firms are listed firms partially controlled by CSOEs. Control firms are listed firms controlled (either partially or fully) by private entities or individuals, i.e. these firms inhabit in non-state owned business groups. A natural concern arises as to the comparability between these two groups of firms. We attempt to mitigate this concern through matched sample analyses. More specifically, we implement a propensity score matching (PSM) approach. In the first stage, we estimate the probability of a firm receiving treatment in the subsequent year, i.e., its controlling shareholder being mandated to return a non-zero proportion of profits. For treatment firms, we only keep their firm-year observations till one year before the treatment year. This is because once a CSOE is included into the State Capital Operation Program, it will continue returning profits to the state. For control firms not experiencing treatment during the entire sample period, we keep all of their firm-years observations. To estimate the logit model, we control for all variables in Equation (1). Further, because Sun and Tong (2003) argue that stock exchanges vary in the proportion of state-owned enterprises, we further control for EXCHANGE. It is a dummy variable that equals one if the firm is listed in the Shenzhen Stock Exchange, and zero otherwise. We present first stage results in Table 4 Panel A. We then match each treatment firm with a control firm using the nearest-neighbor criteria. Specifically, for each treatment firm, we pair with it a control firm with the closest propensity 12

13 score that is estimated from the first stage logit regression (with replacement). To assess the matching quality, we compare firm characteristics of treatment and control firms both before and after the matching in Table 4 Panel B. Two findings emerge. Prior to the matching, treatment and control firms exhibit significant differences in most variables except return on assets (ROA), market to book ratio (MB) and free cash flow (FCF). After the matching, the differences in all variables are statistically insignificant between the treatment and control groups of firms. Therefore, our matching procedure is effective to eliminate fundamental differences between treatment and control firms. Utilizing this matched sample, we re-perform our analyses on the effect of the CSOE reform on listed firms dividend payout. We again estimate Equation (3) and Equation (4) and report results in Table 4 Panel C. We find that our main results continue to hold. The coefficient on PR is negative and significant in columns (1) and (3) with D/E and D/M as the dependent variable, respectively. Further, when we decompose PR into indicators for different proportions of returning-profits (PR_5%, PR_10% and PR_15% + ), the coefficients on these indicators are mostly negative and significant, with the economic magnitudes becoming greater when the returning proportion is higher. [Insert Table 4 around Here] To ensure robustness of our results, we also employ two parsimonious matching procedures and present results in Table IA1 in our Internet Appendix. In Panel A, we match with each treatment firm a control firm that is in the same industry and that has the closest total assets at the end of the pre-reform fiscal year. In Panel B, we match with each treatment firm a control firm that has the closest level of dividend payout in the pre-reform fiscal year. We estimate our main regressions on both matched samples and find results that are consistent with earlier findings Alternative measures of dividend payout Our main analyses employ two common measures of dividend payout: (1) dividend-to-earnings ratio (D/E) and (2) dividend yield (D/M). In sensitivity analyses, we construct two alternative measures of dividend payout. D/S measures a firm s dividend to sales ratio, defined as annual cash dividends divided by sales during the current year. D/A measures a firm s dividends to asset ratio, defined as annual cash dividends divided by total assets at the end of the year. Despite not being conventional measures of dividend payout, the two measures have their own advantages. Using sales or total assets as the deflator does not subject to the concern of negative earnings. Further, compared with market capitalization used in D/M, sales and total assets are less likely to be affected by non-fundamental factors such as investor sentiment. We replace the dependent variable with D/S and D/A and re-estimate Model (1). Our empirical results in Table IA2 in our Internet Appendix suggest that our main inferences are unchanged. 13

14 4.2.5 An alternative treatment measure Earlier, we rely on a PR indicator and three indicators representing 5%, 10% and more than or equal to 15% of returning percentage of profits to identify the effect of the CSOE reform on dividend payout. Such an empirical design resembles difference-in-differences strategy and facilitates results interpretation. In this section, we perform a robustness analysis by constructing a continuous variable that captures the full set of variation information in the proportion of profits mandated to be returned to the state, PR_RATIO. It is computed as the percentage of a CSOE s profits mandated to be returned to the state in the current year. We then estimate Equation (7) below and present results in Table 3. y i,t = α i + α t + δpr_ratio i,t + γx i,t + ε i,t, (7) We estimate Equation (7) and present results in Table IA3 in our Internet Appendix. Using the two alternative proxies for dividend payout (D/E, and D/M), we obtain negative and significant coefficients on PR_RATIO. 4.3 Mechanism analyses intra-group resource reallocation Dividend reductions of listed firms do not necessarily translate into resources that can be used by other firms within the group. For this to occur, certain forms of transactions/activities must exist to reallocate these reduced dividends from one firm to another. In this section, we provide related evidence. We consider two primary forms of intra-group business transactions that indicate resource reallocation from the listed firm to other group members. First, we identify a listed firm s inter-corporate loans provided to the controlling shareholder or its affiliates. Jiang et al. (2010) show that these loans likely reflect the controlling shareholder s expropriation activities over the listed firm, and are associated with lower future profitability of the listed firm. 5 We expect listed firms to experience an increase in these loans upon the inception of the CSOE reform. Results in Table 5 are confirmatory. The coefficient on PR is positive and statistically significant (0.010, t = 2.86), suggesting an increase in inter-corporate loans borrowed by the controlling shareholder and its affiliates. Further, such an effect becomes stronger when the proportion of profits to return becomes greater. The coefficients on PR_5%, PR_10% and PR_15% are increasing in 5 For our sample period from 2007 to 2014, we repeat the analyses in Jiang et al. (2010) that examine the association between current level of other receivables (ORECTA) and the return-on-assets (ROA) in the subsequent year. Consistent with their findings, we find a negative and statistically significant coefficient on ORECTA (-0.177, t = -3.08), suggesting that current other receivables deteriorate future ROA. We control for current year s ROA, firm size, leverage and an indicator for negative value of book equity along with fixed effects for firm and year in this analysis. Results appear in Table IA4 in our Internet Appendix. 14

15 magnitudes. [Insert Table 5 around Here] Second, we consider related party transactions between the listed firm and its parent firm or its affiliates (Jian and Wong, 2010; Liao, Liu and Wang, 2014). To seek related evidence, we utilize granular data on directional commercial trades within a business group. We choose this specific type of related party transactions to analyze for following reasons: (1) it is the dominating form of related party transactions, accounting for 33.1% in proceeds and 37.1% in frequency; (2) the identities of buyers and sellers of the trade clearly shows the direction of money flow; and (3) earlier literature such as Jian and Wong (2010) also rely on commercial trades to seek more precise estimation of the role of related party transactions. Categorizing trades into two groups based on whether our sample firm is a buyer or a seller, we are able to obtain related evidence over whether saved cash dividends are re-distributed within the business group through related party transactions. If so, it will mostly apply to cases when the sample firm is a buyer in a commercial trade. Based on this split, we re-estimate the effect of CSOE reform on CSOE-controlled firms related party transactions. We present results in Table 6. When the listed firm is a buyer, the coefficient on PR, PR_5%, PR_10% and PR_15% + are consistently positive and significant. Differently, when the listed firm is a seller, the coefficients are statistically insignificant. On aggregate, the reform increases the level of related party transactions of the form of commercial trades, and mostly resulting from transactions when the listed firm is a buyer. This set of empirical findings are in line with our earlier argument that saved cash dividends of CSOE-controlled firms are re-distributed within the business group through related party transactions. In brief, we conclude that, after the inception of the CSOE reforms, there is an increase in the intensity of transactions that shift resources away from the listed firm to peer firms within the business group. This evidence, in combination with the significant reduction of dividend payout of listed firms around the reform, suggest that CSOEs intend to maintain the pre-reform level of resources through resource reallocation within their controlled business groups. 4.4 Political incentives of CSOE managers Dividend cut and future career outcome [Insert Table 6 around Here] CSOE managers are government bureaucrats employed by the SASAC to supervise the operations of CSOEs. Unlike managers in the private sector, CSOE managers compete for 15

16 political ranks in a closed pyramidal labor market (Chen et al., 2017; Deng et al., 2015). Within this internal managerial labor market, CSOE managers strive to seek higher political ranks which bring both financial and reputational capital in the future. The mechanism of this labor market therefore resembles the tournament game of local government officials in China, as modeled in Zhou (2007). Competing in the pyramidal labor market, CSOE managers have strong incentives to seek promotions, avoid demotions and retain their positions when exceeding the official retirement age. 6 The political outcome depends heavily on the evaluation of CSOE managers economic performance (e.g., asset value appreciation and economic-value-added) and harmonious development (e.g, avoiding layoffs). The SASAC conducts annual and triennial evaluations of CSOE managers, and the outcome of the latter evaluation significantly affect a CSOE manager s future turnover, i.e. promotion, demotion or remaining status quo (Deng et al., 2015). We argue that, the incentive to retain resources and maintain pre-reform level of operations and investments is a function of CSOE managers political incentives. Those with greater incentives to seek promotions and/or avoid demotions in the near future are more likely to alter policies of firms within the CSOE-controlled business group, upon the inception of the series of CSOE reforms. To understand the CSOE managers motives to cut listed firms dividends, we analyze whether such behavior is associated with their future career outcome, measured by political promotions, demotions and aggregated turnover outcome 7. We construct the variable TURNOVER to measure CSOE managers career outcome. TURNOVER is an ordinal variable that equals two if a CSOE manager is promoted to a higher political rank or enters the government system with an equal political rank 8 ; equals one if a CSOE manager stays in the same position or changes job at equal political rank; or equals zero if a CSOE manager is demoted to a lower rank or has his/her job 6 The Chinese government sets official retirement age which varies with bureaucrats political ranks. For instance, CSOE managers in principle should retire at the age of 60 while governors of provinces are mandated to retire at the age of 65. Most importantly, official retirements are never strictly enforced. Retaining the current position after the official retirement age serves as a reward to officials (Yang et al., 2013). 7 In the test of CSOE manager s career outcome, we focus on the highest level manager in each CSOE. For CSOEs with board of directors, the highest level manager is the board chair. For CSOEs without board of directors, the highest level manager is the chief executive officer (also termed general manager in China). 8 Entering the government system with an equal political rank is a career improvement for CSOE managers for following reasons. First, there is limited chance of seeking further promotions if, as the highest level individual, the CSOE manager stays within a CSOE (Yang, Wang and Nie, 2013). Second, abundant anecdotal evidence reveals that former CSOE managers will be promoted to a higher political rank soon after they enter the government system. For instance, Shulin Su, the former Chairman of China Petrochemical Corporation, was appointed in March, 2011 as the Deputy Secretary of Fujian Province with the same political rank (Fu Bu Ji, 副部级 ). In July 2011, Su was promoted to a higher political rank (Bu Ji, 部级 ) as the Governor of Fujian Province. In our study, there are six CSOE managers who enter the government system with an equal political rank. In addition, our findings are qualitatively unchanged if we set TURNOVER to one for those six observations. 16

17 terminated. To be more precise, we present in Table 7 Panel A the frequencies of all CSOE managers reasons for job turnover. Out of our initial sample comprising 367 observations, 9 managers are promoted (TURNOVER = 2). There are 320 managers retaining their original position and an additional 17 managers changing job at equal political rank (TURNOVER = 1). 9 Although there is only one observation whereby a manager receives demotion, retirement and job-leave observations represent an alternative form of demotion, accounting for 15 and 5 managers respectively (TURNOVER = 0). We perform ordered probit regression when the dependent variable is TURNOVER. In addition, we construct two alternative measures of managerial turnover. PROMOTION is a dummy variable that equals one if TURNOVER equals two, and zero otherwise; and DEMOTION is a dummy variable that equals one if TURNOVER equals zero, and zero otherwise. We perform probit regression when the dependent variable is either PROMOTION or DEMOTION. In Table 7 Panel B, we present results on the effect of dividend reductions on CSOE managers career outcome. We measure dividend cut (DIV_CUT) as the change in dividends of listed firms in the same CSOE business group, deflated by total earnings (or total market capitalizations) of the business group s listed firms at last fiscal year end, multiplied by minus one. The change in dividends for each firm is calculated as the difference between listed firms average cash dividends for three years before the firm s controlling shareholder is mandated to return profits to the state and firm s cash dividends in current year. A higher value of DIV_CUT therefore implies more dividend cut for listed firms within this business group. 10 We find that, on average, group-level dividend cut improves a CSOE manager s career outcome. When the dependent variable is TURNOVER, the coefficients on DIV_CUT are reliably positive and statistically significant. Subsequently, we separately consider whether the results are caused by increased chance of promotion or reduced likelihood of demotion, or both. We find that, when the dependent variable is PROMOTION, coefficients on DIV_CUT are statistically insignificant. Differently, when the dependent variable is DEMOTION, coefficients on DIV_CUT are consistently negative and statistically significant. The effect of dividend cut on reducing demotion likelihood is also economically sizable. Using results in Column (3) for illustration (i.e., DIV_CUT measured using D/E), a dividend cut of one standard deviation of D/E is associated with 7.5% (= 0.616*0.122) lower probability of job demotion. 11,12 9 We exclude two observations with sudden death of the manager and one observation whereby a manager has job turnover because of frauds. None of which is related to economic performance of the CSOE. 10 The majority of CSOE business groups cut dividends during our sample period. Using dividend cuts measured by D/E as an illustration, 54.6% of the observations have negative DIV_CUT. 27.6% of the observations have positive values of DIV_CUT, i.e. increasing dividends. The remaining 17.8% have zero value of DIV_CUT because these CSOE groups did not issue cash dividends in the first place. 11 When computing the reduced likelihood of demotion, is the standard deviation of DIV_CUT measured using D/E ratio (untabulated), and equals the marginal effect on DIV_CUT in Table 7 Panel B (Column 3), multiplied by minus. 17

18 Combined, the difference in the results of PROMOTION and DEMOTION suggests that CSOE managers cut listed firms dividends to maintain the current level of operations in an effort to avoid demotions. [Insert Table 7 Here] Conditional analyses on CSOE managers political incentives Related to this political incentive explanation, we form additional predictions and provide corroborative analyses. We argue that, since the incentive to obtain political capital is greater when (1) a CSOE manager is serving his/her last evaluation year (the third year of a triennial evaluation period), and (2) a CSOE manager was rated lower in his/her last political evaluation, the effect of the reform on listed firms dividends should be more pronounced in these cases. Further, we condition our estimation of the main effect on CSOE managers age. If CSOE managers try to avoid demotions as we document above, the dividend cut should be more pronounced for elder managers. We seek to empirical findings to better understand above issues. In Table 8, we present the results of related analyses. In the first two columns of Table 8, we decompose PR into two indicators. PR_EVAL is coded one when the observations are in the third years of the CSOEs term-evaluation cycle, and zero otherwise. PR_NOT_EVAL is coded one when the observations are in the first two years of the CSOEs term-evaluation cycle, and zero otherwise. We find that the effect of the reform on listed firms dividend payout is greater when a CSOE manager serves his/her evaluation year. The coefficient on PR_EVAL is greater in magnitude than those on PR_NOT_EVAL. In columns (3) and (4), we decompose PR into following two indicators. PR_A is coded one when the CSOE manager received an A (good performance) in her last political evaluation, and zero otherwise. PR_NOT_A is coded one when the CSOE manager did not receive an A (bad performance) in her last political evaluation, and zero otherwise. We obtain information on CSOE managers ratings in each political evaluation from SASAC s website. We find that the effect of the reform on listed firms dividend payout is greater when a CSOE manager did not receive an A in her last political evaluation. The coefficient on PR_NOT_A is greater in magnitude than those on PR_A. 12 In unreported analyses, we employ multinomial logistic regression analyses to estimate the effect of business group dividend cut on the CSOE manager s promotion and demotion likelihood. There are three classes of CSOE managers career outcome: (1) Promotion, (2) Job-retaining or turnover at an equal rank (termed Equal-rank ) and (3) Demotion. Regression results suggest that the likelihood of moving from Class (2) to Class (3) is significantly lower when there is more group-level dividend cut. Differently, the likelihood of moving from Class (2) to Class (1) is insignificantly related to the group-level dividend cut. Therefore, multinomial logistic regressions provide consistent empirical findings. 18