THE POLITICAL TOURNAMENT IN CHINA: THE DYNAMIC NATURE

THE POLITICAL TOURNAMENT IN CHINA: THE DYNAMIC NATURE Zhenyu Zhao Guanghua School of Management, Peking University, CHINA, 100871 Email: patrickzhao33@qq.com ABSTRACT The local officials tournament competition in China is dynamic in nature. This paper documents a unique and robust U-shaped relationship between officials contemporaneous performance and their interim rank, which is different from all the existing studies: the forerunners and underdogs outperform their middleranked counterparts. I argue that the phenomenon originates from distinct incentives of officials in China: the pursuit of promotion and avoidance of punishment. The U-shaped curve gradually diminishes for officials who are in the final round of the tournament competition as their political careers are close to the end. I provide further evidence for the U-shaped curve by examining officials subsequent public goods/services provisions and risk taking behaviors after interim rank disclosure. I check the robustness of the U-shaped curve by ruling out alternative explanations such as competition intensity, term length, age and heterogeneous ability, and conducting falsification tests on fake rankings. The study extends the dynamic tournament theory to a qualitatively different setting in the real world, helps to have a deeper understanding of dynamic tournament and sheds light on officials appraisal system in China. Keywords: Dynamic Political Tournament; Contemporaneous Performance; Interim Rank. JEL classification: H11;H77; P26 1 INTRODUCTION What distinguishes the dynamic tournament from its static counterpart is its salient nature of interim information disclosure; the contestants progress can be informed at the intervals of the stages. A vast body of literature studies the impact of halfway evaluation on the final outcome, though far from reaching an agreement on the direction of the influence. According to conclusions in these studies, contemporaneous performance and interim rank can exhibit downward slope (see Gürtler, Münster, and Nieken (2013)), upward slope (see Ederer (2010)) or inverse U-shaped relationships (see Genakos and Pagliero (2012)). In sharp contrast, this paper documents a unique and robust U-shaped curve linking contemporaneous performance and interim rank in a qualitatively different dynamic tournament of the real world, the competition among the China local officials 1. The tournament competition theory is often employed to comprehend strong economy-promoting incentives of China local officials. According to the argument, local government officials are competing for economic growth in order to maximize the likelihood of promotion (see Chen, Li, and Zhou (2005); Li and Zhou (2005), Yu, Zhou, and Zhu (2013) and among the others). However, the competition is dynamic in nature since (1) the tournament competitions among the officials take place in multiple periods (on annual basis) rather than single one and (2) officials are informed of their interim rank of economic growth after the release of economic statistics (see Goltsman and Mukherjee (2011) for the clarification of differences between static and dynamic tournament). The unique finding, which is different from all the existing literature, can be explained Progressive Academic Publishing, UK Page 84 www.idpublications.org

from the perspectives of two distinct types of incentives induced by interim rank, rooted in the contextual differences of tournament competition among China local officials: China s top-down power structure and the well informed higher-level author- ity distinguish itself from the Western style yardstick competition as studied by Besley and Case (1995) (see Yu, Zhou, and Zhu (2013) for concrete interpretations). On the other hand, the official- s incentives and strategic behaviors after knowing their ranking positions in halfway during their incumbencies are important for their jurisdictions, as they have full discretion and can implement a wide variety of policy instruments to affect the economy (see Yu, Zhou, and Zhu (2013)): the correlation between officials incentives and outcome are highly correlated (see Maskin, Qian, and Xu (2000); Chen, Li, and Zhou (2005); Li and Zhou (2005); Xu (2011); Choi (2012) and among the others). In the current study, I apply the dynamic tournament theory to investigate how interim rank, i.e., the rank of last year GDP per capita growth within the province 2, affects the contempo- raneous economic performance in the dynamic tournament among China s local officials with a concrete panel data set from 1999 to 2010. I document a unique and robust U-shaped curve linking contemporaneous performance and interim rank, i.e., the forerunners and underdogs outperform their middle-ranked counterparts. I argue that the phenomenon stems from two distinct types of incentives owing to the qualitatively different setting of China officials: the one to get promoted for officials in leading positions and the other to avoid punishment for those who fall far behind. The similar U-shaped patterns derived from different samples of officials other than cities party chiefs back up the argument, and the pattern gradually diminishes for officials who are in the final round of the tournament competition as their political careers are close to the end. I provide fur- ther evidence by investigating officials subsequent public goods/services provisions and risk taking behaviors after the interim rank disclosure. Moreover, I address the concerns such as competition intensity, term length, age and heterogeneous ability through a wide range of robustness checks and falsification tests. The findings of this study on one hand extends dynamic tournament theory to a qualitatively different setting in the real world and unites the world s second largest economy, while on the other hand considers the dynamic nature of the tournament competition among China local officials. Moreover, the conclusions made here reflect a new problem of officials appraisal system in that interim rank might further distort officials incentives and aggravate the existing problems such as insufficient public goods/services provisions and deteriorating environment. The paper is closely related to two stands of literature. The first stand of literature studies how the interim information disclosure impacts the performance in the dynamic tournament. Ederer (2010) argue that the interim performance disclosure can reveal heterogeneous ability, and thus reduce players incentives since they might feel it impossible to overcome the ability gap; those who fall behind might slack off. Sabotage is another problem since the forerunners are more dangerous rivals and thus sabotaged more heavily, i.e., they incur an extra cost to take the leading position (see Gürtler, Münster, and Nieken (2013)), and therefore they might shirk. Evidence on both heterogeneous ability and sabotage effect are found in a lot of studies such as Genakos and Pagliero (2012). They discover an inverse U-shaped relationship between performance and interim rank, and interpret the phenomenon from contextual/psychological perspectives owing to the specific data of world weight-lifting tournament. In contrast, someone maintain that the effect i.e., the rank of last year GDP per capita growth within the province 2, affects the contemporaneous economic performance in the dynamic tournament among China s local officials with a concrete panel data set from 1999 to 2010. I document a unique and robust U-shaped curve linking contemporaneous performance and interim rank, i.e., the forerunners and underdogs outperform their middle-ranked counterparts. I argue that the phenomenon stems from two distinct types of incentives owing to the qualitatively different setting of China officials: 1 I refer (local) officials to party chiefs of city level government who are the political leaders of the jurisdiction, as did by Chen, Li, and Zhou (2005) and Li and Zhou (2005). The city level governments include all prefectural level cities, prefectures, autonomous prefectures and leagues which are direct subordinates of provinces and at the same government level/political rank. Progressive Academic Publishing, UK Page 85 www.idpublications.org

The one to get promoted for officials in leading positions and the other to avoid punishment for those who fall far behind. The similar U-shaped patterns derived from different samples of officials other than cities party chiefs back up the argument, and the pattern gradually diminishes for officials who are in the final round of the tournament competition as their political careers are close to the end. I provide fur- ther evidence by investigating officials subsequent public goods/services provisions and risk taking behaviors after the interim rank disclosure. Moreover, I address the concerns such as competition intensity, term length, age and heterogeneous ability through a wide range of robustness checks and falsification tests. The findings of this study on one hand extends dynamic tournament theory to a qualitatively different setting in the real world and unites the world s second largest economy, while on the other hand considers the dynamic nature of the tournament competition among China local officials. Moreover, the conclusions made here reflect a new problem of officials appraisal system in that interim rank might further distort officials incentives and aggravate the existing problems such as insufficient public goods/services provisions and deteriorating environment. The paper is closely related to two stands of literature. The first stand of literature studies how the interim information disclosure impacts the performance in the dynamic tournament. Ederer (2010) argue that the interim performance disclosure can reveal heterogeneous ability, and thus reduce players incentives since they might feel it impossible to overcome the ability gap; those who fall behind might slack off. Sabotage is another problem since the forerunners are more dangerous rivals and thus sabotaged more heavily, i.e., they incur an extra cost to take the leading position (see Gürtler, Münster, and Nieken (2013)), and therefore they might shirk. Evidence on both heterogeneous ability and sabotage effect are found in a lot of studies such as Genakos and Pagliero (2012). They discover an inverse U-shaped relationship between performance and interim rank, and interpret the phenomenon from contextual/psychological perspectives owing to the specific data of world weight-lifting tournament. In contrast, someone maintain that the effect is overall conducive. Eriksson, Poulsen, and Villeval (2009) hold this view by arguing that the underdogs almost never quit the competition and front runners do not slack off, and meanwhile provide some experimental evidence. Similar findings are obtained by Casas- Arce and Martínez- Jerez (2009) with data of salesmen contests organized by a commodity company. The third view distinguishes encouraging interim news from the discouraging one, and insists that the influence might be positive or negative, depending on its content delivered: Goltsman and Mukherjee (2011) propose that information should be disclosed only if all contestants perform poorly, and similar conclusions are made by Chen and Chiu (2013) who raise a solution to from the perspective of contract. The paper contributes to the first strand of literature in the following two aspects. As far as I know, this study for the first time extends the dynamic tournament theory to a qualitatively different setting, the China local officials to examine the relationship between contemporaneous performance and interim rank, and thus enrich the empirical evidence of dynamic tournament. Unlike the existing literature that employs experimental data (like Casas-Arce and Martínez-Jerez (2009) and Eriksson, Poulsen, and Villeval (2009)) or sports competition (like Genakos and Pagliero (2012)), the current application to this real world problem unites the dynamic tournament theory and the world s second largest economy. Besides, officials in China are vital for the jurisdictions they govern as their incentives can exert substantial influence on the economy and welfare of residents. The study also sheds light on the current officials appraisal system. Meanwhile, the setting is qualitatively different in that tournament takes place in several rounds at different levels of government/political ranks, and the level rising from city mayors, cities party chiefs to governors and provincial party chiefs 3. Therefore, the dynamic political tournament competition occurs in different rounds and each round has several stages, which resembles soccer league system. The discovered unique and robust U-shaped curve linking contemporaneous performance and interim rank combined with 2 Official compete with each other in a greater region they belong to, i.e., provincial officials competition takes place in the whole country, cities officials compete in the same provinces they belong to, counties officials rival each other in the same cities they belong to, etc. Progressive Academic Publishing, UK Page 86 www.idpublications.org

further evidence helps to have a deeper understandings of the dynamic tournament. The second strand of literature discusses the correlation between promotion of government officials and their economic performance, and most of the studies resort to tournament theory (see Chen, Li, and Zhou (2005); Li and Zhou (2005) and among the others). According to the theory, the superior official assesses the subordinates by their relative economic performance and decides promotion. Maskin, Qian, and Xu (2000) for the first time document a significant positive correlation between relative economic performance (measured by rank position) and political status with the sample of provincial officials from 1976 to 1986, and argue that officials are likely to engage in economic growth competition. Li and Zhou (2005) contribute fundamentally to the theory and present evidence that economic performance significantly affects officials likelihood of promotion and termination with sample of provincial party chiefs, while Chen, Li, and Zhou (2005) show further that the chance of promotion also depends on the immediate predecessor s economic performance. Choi (2012) verifies the positive effect of economic performance on the odds of promotion with data of provincial officials from 1989 to 2009, while Yu, Zhou, and Zhu (2013) provides further evidence on investment attraction. This paper is distinguished from existing ones by considering the dynamic nature of the tournament competition among the officials, since interim evaluations can be disclosed to them during the intervals of their competition stages; officials compete annually and feedback can be obtained easily from release of economic statistics of the jurisdictions they govern. I show evidence on how local officials respond strategically to the interim rank. The rest of the paper proceeds as follows. Section 2 introduces the data and specification. Section 3 presents stylized facts about the contemporaneous performance and interim rank. Sec- tion 4 provides further evidence on the relationship between them. Section 5 checks the robustness. Section 6 concludes. 2 The Data and Specification 2.1 The Data The data employed here incorporate 1999-2010 city level panel data from China Regional E- conomic Statistical Yearbook and 1998-2008 provincial level panel data from China Compendium of Statistics. Information on officials is manually collected from a wide range of data sources like governmental documents, searching engines, etc. Since a handful of observations (up to 50) contain errors, I correct them with alternative data sources such as China City Statistical Yearbooks and statistical yearbooks of different provinces in corresponding years (See Table 1). 3 In China, mayors/governors are less important than city/provincial party chiefs and they are often vice party chiefs, and thus transfer from position as city mayor/governor to city/provincial party chief is a kind of promotion, although mayor/governor and party chiefs share the same political rank. Progressive Academic Publishing, UK Page 87 www.idpublications.org

Table 1: Summary Statistics-City Level Sample Variable & Definition No.Obs. Mean Std. Dev. Min Max Panel A City Statistics GDP Per Capita Growth (%) 3657 12.803 7.962-7.717 39.334 Population Growth (%) 3657 0.562 2.579-10.134 10.407 Investment Rate (%) 3902 45.54 23.724 9.926 122.642 Education Expenditure 2941 2.169 0.932-4.605 4.917 Medical Expenditure 1330 1.844 0.846-2.996 4.130 Fiscal Expenditure 3658 3.682 0.999 0.495 7.144 Waste Water 1590 8.594 1.098 2.833 11.359 Sulphur Dioxide 1590 10.666 1.182 2.303 13.115 Smoke Dust 1584 9.789 1.168 3.526 14.422 GDP 3992 14.904 1.201 10.215 18.516 Waste Water Firms Investment 979 12.759 2.129 1.099 17.518 Sulphur Dioxide Firms Investment 951 11.929 2.181 0.693 16.437 Smoke Dust Firms Investment 969 11.933 2.091 2.996 16.423 Nitrogen Oxides Firms Investment 946 11.685 2.284 0 16.431 Dust Firms Investment 941 11.104 2.342 0 16.429 Total Firms Investment 985 13.145 2.128 1.099 17.538 Panel B Characteristics of City Party Chiefs Rank 3324 0.541 0.288 0.048 1 D: =1 Not the First Year 3996 0.69 0.463 0 1 Education 3887 0.918 0.275 0 1 Age 3885 50.842 4.335 31 63 Term Length 3885 2.395 1.703 0 10.083 Provincial Government Working Experience 3994 0.544 0.498 0 1 Number of Experiences as Party Chief/Mayor 2848 0.097 0.334 0 2 Panel C Characteristics of Mayors Rank 3324 0.541 0.288 0.048 1 D: =1 Not the First Year 3984 0.661 0.473 0 1 Education 3984 0.897 0.304 0 1 Age 3983 48.660 4.813 29 64 Term Length 3984 2.139 1.786 0 10.333 Provincial Government Working Experience 3983 0.457 0.498 0 1 Data are mainly from China Regional Economic Statistical Yearbook, Statistical Yearbooks of different provinces. Data on officials ar from governmental documents and search engines. All rank variables are lagged one period. Panel A in Table 1 tabulates the summary statistics of the city level sample. A problem asso- ciated with the sample is the outliers due to the adjustment of administrative divisions in China during these years. I winsorize the top and bottom 1% observations on GDP Per Capita Growth, Population Growth and Investment Rate, i.e., replace the original values with the corresponding 1% and 99% quantile values, while the results stay robust to different winsorizing quantiles and unwinsorized sample. One can see Progressive Academic Publishing, UK Page 88 www.idpublications.org

that the GDP per capita grows at an extraordinary rate of 12.803% per year during 2000-2010, while the growth of the population experiences only 0.562% annual increase due to the one-child policy. The maximum and minimum values of the economic performance and population growth suggest substantial difference across cities: the fastest growth can almost reach 40% since these cities are mainly underdeveloped ones and catch up with the oth- ers quickly while the slowest one incur a 7% loss due to the 2008 financial crisis, and the population growth can increase or decrease by almost 10%. Moreover, we can see from the high investment rate that China is still featured as an investment-driven economy, and the mean value is 45.54%. I also exploit the variations on fiscal variables to see the officials public goods/services provisions behav- ior, including education, medical and total fiscal expenditure in logarithm (Education Expenditure, Medical Expenditure and Fiscal Expenditure). Computing the ratios of the two expenditures over the total, we can see that the average values are 19.9% and 14.5%, suggesting that they are not an emphasis of officials. Furthermore, I employ variables to further test officials risk taking behav- iors, i.e., the pollution attractions, including the logarithm of industrial waste water release (Waste Water), sulphur dioxide (Sulphur Dioxide) and smoke dust (Smoke Dust) emissions from China Statistical Yearbooks on Environment, and control for logarithm of GDP (GDP). To exclude the co-trend of GDP, I (1) scale the three variables by GDP and (2) obtain governmental documents to get 2-digit code industries that release/emit the most pollutants in the five categorical industries respectively: waste water, sulphur dioxide, smoke dust, nitrogen oxides and dust. I calculate the sum of investment made by firms in these five categorical industries (denoted as Waste Water Firms Investment, Sulphur Dioxide Firms Investment, Smoke Dust Firms Investment, Nitrogen Oxides Firms Investment and Dust Firms Investment), while also control for the total investment made by firms within the entire cities (Total Firms Investment) utilizing data from China Industrial Survey. By calculating the ratios of investment, one can see that the ratio of waste water pollution firms investment to total firms investment is as high as 71.8%, while the figures are about 40% for sulphur dioxide and smoke dust pollution firms, 33.1% for nitrogen oxides pollution firms and 24.1% for dust pollution firms. Panel B in Table 1 demonstrates party chiefs characteristics. Unlike the standard tournaments in which number of players are the same (see, e.g., Genakos and Pagliero (2012)), the number of cities within each province vary substantially (Hainan Province has only 2 cities while Guangdong Province has 21 during the sample period), I normalize the original rank by the total city number within the province (Rank) to indicate the position. One has to take into consideration that the turnover of the officials since the panel is constructed in terms of city and year, namely, the panel identifier is city rather than individual officials. To distinguish the officials last year performance from their predecessors (see Chen, Li, and Zhou (2005)), I interact the rank with a dummy (D) to indicate whether it is the first year; if not, then the last year performance is indeed the current incumbent s. Meanwhile, the variable D measures the turnover rate of cities party chiefs (D=0 if it is a year when turnover happens), and it is about 0.31. Officials characteristics are also important in explaining the economic growth within their jurisdictions, and I select the corresponding variables of official s characteristics as Li and Zhou (2005). One can see that 91.8% of the officials earn an undergraduate degree or above (Education), indicating the emphasis on education attainment on officials by central government. The mean age of cities party chiefs is 50.842, in accordance with governmental documents stipulating that the maximum inaugurate age for city party chief is 50. The average length of term (Term Length) is only 2.395 years, signaling frequent turnover of local officials, and the high frequency enables me to better assess the tournament as argued by Denis and Denis (1995) in the example of firm managers. Over a half of the city officials have ever worked at provincial government (Provincial Government Working Experience), which might indicate the importance of connection and experience on promotion. As seen in quite a few literature, like Ederer (2010) who points out the importance of heterogeneous ability in dynamic tournament, I use the number of terms at position of city party chief or mayor (Number of Experiences as Party Chief/Mayor) to proxy, and it can be seen that most of the officials are new to the positions as party chief. (See Figure 1 ) Progressive Academic Publishing, UK Page 89 www.idpublications.org

Per Capita GDP Growth Rate in % Per Capita GDP Growth Rate in % Per Capita GDP Growth Rate in % European Journal of Business, Economics and Accountancy Vol. 5, No. 5, 2017 0 5 10 15 20 pcgdpgrmean Fitted values (a) Performance vs Original Interim Rank 1 2 3 4 5 pcgdpgrmean1 Fitted values (b) Performance vs Interim Rank in Five Quintile 0 2 4 6 8 10 Rank Decile of Last Year pcgdpgrmean2 Fitted values (c) Performance vs Interim Rank in Ten Decile Figure 1: Interim Rank and Performance in a Dynamic Political Tournament Figure 1 illustrates the contemporaneous performance and interim rank in this dynamic political tournament, a striking nature of the graphs tells us that there is a U-shaped curve linking them together, which is hardly seen in any existing literature on dynamic tournament. I present the average GDP per capita growth (at vertical axis) to each original rank (at horizontal axis) of economic performance of last year GDP per capita growth (Figure 1(a)), the rank (scaled by number of cities within each Progressive Academic Publishing, UK Page 90 www.idpublications.org

province) in quintile and decile (1(b) and 1(c), and the pattern stays virtually unchanged. I divide the rank also into 5, 6, 7, 8 or 9 groups, and the U-shape remains. To facilitate a more thorough understanding of the officials tournament competition, I exploit variations of mayors and provincial officials. Panel C in Table 1 presents the mayors characteristics. The turnover rate is 0.339, even higher than cities party chiefs as many of them are then promoted to position of city party chief as is seen in the data set. About 89.7% of mayors own undergraduate degree or above, reflecting again the emphasis on education for cadres. On average, mayors are about 2 years younger than party chiefs as it takes some years for a mayor to be promoted to the position of party chief. The length of term for mayors are 0.25 years shorter than party chiefs, consistent with their higher turnover rate, and about 10% less of them have provincial government working experience as compared with party chiefs. The data of provincial sample span from 1998 to 2008 so that comparison between provincial and city level sample becomes feasible. Moreover, since Hainan and Chongqing are newly established province/provincial city in 1988 and 1997 respectively, data start from 1998 can exclude alternative explanation for the U-shaped curve from the perspective of administrative division adjustments; before 1988 Hainan is part of Guangdong Province while Chongqing used to be a part of Sichuan Province before 1997 ( See Table 2 ) Table 2: Summary Statistics-Provincial Level Sample Variable & Definition No.Obs. Mean Std. Dev. Min Max Panel A Provincial Statistics GDP Per Capita Growth (%) 310 11.464 3.985 1.741 20.782 Population Growth (%) 341 0.601 0.359-0.324 1.59 Investment Rate 341 44.664 13.014 26.687 79.248 Panel B Characteristics of Provincial Party Chiefs Rank 279 0.493 0.281 0.032 1 D: =1 Not the First Year 341 0.762 0.426 0 1 Education 341 0.868 0.339 0 1 Age 341 58.32 3.982 46 67 Term Length 340 2.767 2.474 0 14.583 Central Government Working Experience 341 0.337 0.473 0 1 Panel C Characteristics of Governors Rank 279 0.516 0.289 0.032 1 D: =1 Not the First Year 341 0.692 0.462 0 1 Education 341 0.862 0.345 0 1 Age 341 57.161 4.333 42 65 Term Length 341 2.34 1.865 0 9.25 Central Government Working Experience 341 0.226 0.419 0 1 Data are mainly from China Regional Economic Statistical Yearbook, Statistical Yearbooks of different provinces. Data on officials ar from governmental documents and search engines. All rank variables are lagged one period. Panel A in Table 2 illustrates the descriptive statistics of provinces. We can still see the mirac- ulous economic development in the twenty-first century; the average GDP Per Capita Growth rate is 11.464%, while the population growth rate averages at 0.601%. Investment driven feature is still salient, and the mean value of investment rate is 44.664%. Panel B in Table 2 reports the characteristics of provincial party chiefs. The rank is still normalized by number of provinces due to the administrative Progressive Academic Publishing, UK Page 91 www.idpublications.org

division adjustments. I also utilize the dummy variable D to indicate whether the last year performance is from the same official, and it can be seen that the turnover rate of provincial party chiefs is 0.238, more stable than that of cities since provincial party chiefs are close to the end of their political careers. The high educational attainment of provincial party chiefs still reflects the stress on education, and 86.8% of them obtain undergraduate degree or above. The mean value of the provincial party chiefs age is 58.32, older than that of city level officials since it still requires several years to get promoted to provincial government positions. The length of term is 2.767 years, longer than city officials and consistent with the lower turnover rate. Only 33.7% of them have central government working experience compared to 54.4% at city level, since much fewer positions are available in central government than in the provincial ones. Panel C in Table 2 provides further the characteristics of governors. One can see that the turnover rate 0.308 is also lower than city officials, and 86.2% of the governors acquire undergraduate degree or above implying again the emphasis on education. Still, the mean age of governors is smaller and their average term length is shorter than provincial party chiefs, as it takes some time to get promoted to position of party chief. 2.2 The Specification I exploit the panel data regression model to test the relationship between contemporaneous performance and interim rank: yi,t=α+β1ranki,t 1+β2Rank 2 +β3ranki,t 1*D +β4rank 2 *D+ x l i,t γ+µi+λt+ei,t i,t 1 where yi,t is the dependent variable which can be GDP Per Capita Growth in baseline regressions, Education Expenditure (in logarithm or scaled by total fiscal expenditure) and Medical Expenditure (in logarithm or scaled by total fiscal expenditure) for the investigation of public goods/services provisions behavior, Waste Water (in logarithm or scaled by GDP), Sulphur Dioxide (in logarithm or scaled by GDP), Smoke Dust (in logarithm or scaled by GDP), Waste Water Firms Investment, Sulphur Dioxide Firms Investment, Smoke Dust Firms Investment, Nitrogen Oxides (Nitrogen Oxides) Firms Investment and Dust Firms Investment for the examination of risk taking behavior. Rank and Rank 2 are level and quadratic terms of interim rank normalized by number of cities within each province, to distinguish the genuine effect of officials interim performance from their immediate predecessors, I interact the Rank and Rank 2 with the aforementioned dummy D. x l i,t is a set of variables including Population Growth, Investment Rate, the officials characteristics (Term Length, Education, Age and Provincial/Central Government Working Experience). µi and λt are individual and time fixed effects respectively, the former represents the time-invariant unobservable specific city/provincial effect that might affect the economic growth and is correlated with rank, like culture and natural resource endowment, while the latter implies the common shock of specific year that is invariant to different individual cities/provinces like the 2008 financial crackdown. The parameters of interest are β3 and β4 in the above equation that capture the real effect of interim rank on contemporaneous performance. The model specification is different from that of Genakos and Pagliero (2012) in that I don t use dummies for ranks due to the substantial difference of city numbers within provinces in China. However, one can obtain similar results by employing original rank dummies, quintile and decile dummies after obtaining Rank, i.e., if 0.1 <Rank 0.2, then the first quintile dummy and the second decile dummy takes value 1 respectively. However, the rank dummies lack economic interpretations. For example, Ningxia Hui Autonomous Region has only 5 cities so that only the second decile rank dummy for the top ranked official takes value 1, yet the performance and behavior for the first decile ranked officials are missing. As a consequence, I use normalized rank variable and its quadratic term for the subsequent regressions. i,t 1 Progressive Academic Publishing, UK Page 92 www.idpublications.org

3 Interim Rank and Performance: Stylized Facts 3.1 Baseline Regressions (See Table 3 ) Table 3: Baseline Regressions (1) (2) (3) (4) GDP Per Capita Growth Rank -2.624* -2.624* -2.119-0.225 (1.582) (1.582) (2.266) (2.343) Rank 2 3.112* 3.021* 1.905 0.292 (1.628) (1.626) (2.344) (2.386) Rank*D -1.611* -3.138* Rank 2 *D (0.970) 1.540* (1.834) 3.591* (0.921) (2.111) Population Growth -1.301*** -1.305*** -1.299*** -1.304*** (0.049) (0.049) (0.049) (0.049) Investment Rate 0.039*** 0.041*** 0.039*** 0.042*** (0.009) (0.009) (0.009) (0.009) Term Length 0.223*** 0.282*** (0.077) (0.094) Education -1.495** -1.501** (0.594) (0.595) Age -0.009-0.008 (0.042) (0.042) Provincial Government 0.111 0.114 Working Experience (0.305) (0.305) Constant 8.551*** 9.647*** 8.573*** 9.392*** (0.619) (2.340) (0.620) (2.351) Observations 3,303 3,279 3,303 3,279 adjusted R 2 0.343 0.349 0.343 0.349 Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. Table 3 exhibits the empirical results from the baseline regressions, i.e., regressing GDP per capita growth on both level and quadratic terms of rank normalized by city number within each province as well as their interaction with dummy D. I stepwise test the models, and in the first step use only Rank and Rank 2 within or without the officials characteristics. From Column (1) and (2), it can be seen that there is a significant U-shaped curve linking contemporaneous performance and interim rank for local officials; the coefficients estimates on both level and quadratic terms of normalized rank variable are significantly negative and positive, and the lowest point is around the middle (0.434). To distinguish the true effect of interim rank on performance, I add in the Rank*D and Rank 2 *D to indicate official s own rank, and the results are reported in Column (3) and (4). The coefficients estimates on Rank*D and Rank 2 *D are significantly negative and positive, while those on Rank and Rank 2 are not. Again, the lowest point at this time is also around the middle (0.437). Thus according to Table 3, officials whose interim rank is in the Progressive Academic Publishing, UK Page 93 www.idpublications.org

middle (0.437) perform most poorly. Besides, the control variables are within the expectation. In accordance with the classical growth theory, the population growth rate negatively affects the GDP per capita growth while the investment rate influences economic growth in the opposite direction, and both variables coefficients estimates are statistically significant at 1% level: one percent rise in population growth lowers down the GDP per capita growth by 1.3 percent, while one percent increase in investment rate raises GDP per capita growth rate by 0.04 percent. The length of term is positively correlated with per capita GDP growth since the longer the official stays, the more he/she knows about the economic situation of the jurisdiction he/she governs; ceteris paribus, one year increase in term can accelerate the economic growth by 0.28 percent. Higher education attainment lowers down the economic growth, according to Li and Zhou (2005), perhaps promotion of officials with more advanced degree will be easier and thus their effort in boosting the economy is weaker; the coefficient estimates are significantly negative. Age and provincial government working experience do not significantly affect the economic growth. rate exhibits a U-shaped relationship with the last year rank of economic performance within the province, and I argue that this phenomenon stems from the two distinct incentives for front runners and underdogs respectively, i.e., the promotion incentive and punishment fear. 3.2 The Interpretation The previous studies argue that the information disclosed during the intervals of competition stages in dynamic tournament can impact the future outcome of players in different rank positions, but haven t reached an agreement. The negative impact can arise from the revealed heterogeneity or the sabotage effect (see Ederer (2010), Genakos and Pagliero (2012) and Gürtler, Münster, and Nieken (2013)), and then contemporaneous performance and interim rank shall exhibit upward slope, downward slope or inverse U-shaped curve rather than a U-shaped one. The positive influ- ence might come from the exit barrier for underdogs (see Casas-Arce and Martínez-Jerez (2009); Eriksson, Poulsen, and Villeval (2009)) while the undetermined effect is attributive to different contents in news delivered (Goltsman and Mukherjee (2011); Chen and Chiu (2013)), then officials in all ranks shall improve their performance or behave contingently according to the news, and it is hard to see some certain pattern between contemporaneous performance and interim rank. Whatever the effect the interim rank has on contemporaneous performance documented by the previous literature, it is not likely that a U-shaped relationship between them can emerge. I argue that the competition among the local officials is quite different, compared with the ones studied by the existing dynamic tournament literature. The discovered U-shaped relationship between contemporaneous performance and interim rank is unique but intuitive. The phenomenon originates from the two distinct kinds of incentives which are grounded in the qualitatively different settings of China local officials. From the results reported in Table 3, one can see that the contemporaneous GDP per capita growth In this dynamic political tournament, winners can be a few rather than only one. The institu- tional arrangements in China are comprised of four pillars, i.e., communist party (the leaders are called party chiefs at different government levels), government (the leaders are called governors, mayors, magistrates, etc.), national people s congress (the leaders are called directors at different government levels) and Chinese people s political consultative conference (the leaders are called directors at different government levels). All the four pillars at higher level government offer posi- tions to the winners in the tournament, and in this case they get promoted. Meanwhile, although party chief, mayor, director of national people s congress or Chinese people s political consultative conference share the sample political rank, the party chief leads the other three and is the most im- portant one. As a result, the other three unimportant Progressive Academic Publishing, UK Page 94 www.idpublications.org

positions other than party chief at city level can absorb the losers, the transfer from position as party chief to position of city mayor, director of national people s congress or Chinese people s political consultative conference at city level is regarded as demotion. In addition, officials can be demoted in political rank, or even terminated. Hence, the front runners are motivated to work even harder since only a limited number of them can win the game and get promoted to the higher level positions, and they have to keep the leading position by working harder. Simultaneously, the underdogs might be afraid of punishment and thus perform better. On the contrary, the middle ranked officials are neither motivated by promotion nor fear for punishment, and hence their performance lags behind the above two groups. Moreover, there are several rounds of the competition, from lowest to highest levels are mayors, cities party chiefs, governors and provincial party chiefs (See Xu (2011) for an excellent review of institutional arrangements in China). And in each round, there are several stages as officials have to stay several years before moving to next position. Combining the two distinguished features, the tournament competition among the local officials can be regarded as a soccer league system, where teams are competing for a champion and a few (3 to 4) qualifications to (1) participate in international sports-field such as UEFA Champions League for teams in highest level league or (2) be promoted to higher level league for those not in the highest level league, while a few bottom ranked teams are relegated to lower level league. For officials, a few of them can get promoted to provincial positions (like the few winners in soccer league) while the other few might encounter punishment (being moved to other three unimportant positions, demotion in political rank or even segregation, like the relegation in soccer league). Thus in this circumstance, forerunners still work hard to compete for the champion or qualifications while underdogs spare no effort to stay in the current league. The middle ranked teams might slack off as they are neither motivated to obtain the qualifications opportunity let alone the champion, nor deterred to be relegated to lower level league. The phenomenon is world-wide, and we can easily see that some teams in soccer leagues across the globe behave in this way during the last season. The argument can be found in a strand of literature on sports economics, like Noll (2002). As a result, the fundamental argument made in this study is that the U-shaped curve comes from two distinct kinds of incentives: the incentive to be promoted and the one to avoid the potential punishment. The officials in leading positions are motivated to perform even better as they are still keen on being promoted to the limited number of higher level positions, while those in the bottom rank are deterred to be punished and thus have strong incentive to work harder to improve the economic performance. On the contrary, the middle ranked officials are neither motivated by promotion nor afraid of being punished, and hence they slack off and underperform the other two groups of officials. 4 Interim Rank and Performance: Further Evidence 4.1 Officials Competition at Other Levels The incentive for officials at other levels shall be different from that of cities party chiefs. Mayors incentives are stronger than party chiefs since they are in lower level of tournament than the party chiefs in the same city, and transfer from position as mayor to party chief is regarded as promotion. Besides, the provincial officials incentives should be weaker than the cities party chiefs since it is extremely hard to get promoted to national level as the positions are even fewer. Moreover, governors are competing in the lower level of tournament than the party chiefs within the same province, and thus the incentives for the former should be stronger than the latter. As a consequence, the U-shaped pattern should diminish gradually from mayors to provincial party chiefs (See Table 4 ). Progressive Academic Publishing, UK Page 95 www.idpublications.org

Table 4: Competition of Officials Other than Cities Party Chiefs (1) (2) (3) GDP Per Capita Growth (1) (2) (3) Rank -2.082-2.517-2.471 (2.433) (3.374) (2.509) Rank 2 2.933 0.407 2.123 (2.439) (3.688) (3.881) Rank*D -2.545** -1.390* -4.193 (1.281) (0.768) (3.777) Rank 2 *D 2.471** 1.964* 10.467 (1.24) (1.123) (6.725) Population Growth -1.245*** -5.480*** -4.743** (0.039) (1.947) (1.926) Investment Rate 3.054*** 0.127*** 0.120*** (1.000) (0.028) (0.027) Term Length 0.056 0.060-0.202 (0.107) (0.125) (0.140) Education -0.427 1.055-1.008 (0.587) (0.792) (0.668) Age -0.072* 0.061 0.050 (0.041) (0.067) (0.059) Provincial Government -0.411 Working Experience (0.316) Central Government -0.163 0.329 Working Experience (0.482) (0.479) Constant 12.785*** 5.083 7.676* (2.227) (4.441) (4.016) Sample Mayor Governor Provincial Secretary Chief Observations 3,296 268 264 adjusted R 2 0.365 0.569 0.565 Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. Table 4 report the results, and we can see that the U-shaped relationship between contempo- raneous performance and interim rank gradually fade away from mayors to provincial party chiefs: most of the coefficients estimates on Rank*D and Rank 2 *D are significantly negative and positive respective, except for the ones derived from provincial party chiefs sample. For mayors, the U- shaped relationship between contemporaneous performance and interim rank is the most obvious with coefficients estimates significant at 5% level, and the lowest point is just in the middle (0.515). The pattern is even more salient than that of cities party chiefs since the mayors are at lower level tournament. Comparatively, the U-shaped pattern for governors is more ambiguous than mayors and cities party chiefs, although the coefficients estimates on Rank*D and Rank 2 *D are marginally significant, the lowest point approaches to Progressive Academic Publishing, UK Page 96 www.idpublications.org

0.354. Comparatively, the U-shaped curve for provincial party chiefs is the most vague one; the coefficient estimate on both Rank*D and Rank 2 *D are both insignificant although the pattern still exists with level term being negative and quadratic term being positive. Taking into consideration the estimated lowest point 0.200, chances are that the provincial party chiefs contemporaneous performance might decrease with interim rank with formerly top-ranked being the lowest the bottom-ranked being the highest, consistent with pre- vious literature arguing sabotage effect (see Gürtler, Münster, and Nieken (2013)). Estimates in Table 4 show that U-shaped relationship between contemporaneous performance and interim rank diminishes as tournament level escalates as it becomes much harder to get promoted to national level for provincial party chiefs/governors, and their political careers are approaching to the end. 4.2 Public Goods/Services Provisions Behaviors A possible consequence related to the interim rank should be the public goods/services provi- sions. If the U-shaped curve between contemporaneous performance and interim rank holds, then we shall see that the public goods/services provisions shall exhibit an inverse U-shaped relationship: top and bottom ranked officials might be keen on economic growth and slack off on providing public goods/services, while those rank in the middle might consider to provide more public goods/services to local residents under less promotion/demotion pressure (See Table 5 ) Table 5: Public Goods/Services Provisions Behavior (1) (2) (3) (4) VARIABLES Education Expenditure Medical Expenditure Functional Form Logarithm /Expenditure Logarithm /Expenditure Rank -0.068 0.000-0.225-0.007 (0.119) (0.004) (0.198) (0.005) Rank 2 0.123 0.003 0.234 0.009* (0.122) (0.004) (0.207) (0.005) Rank*D 0.143 0.003 0.006 0.008** (0.098) (0.003) (0.161) (0.004) Rank 2 *D -0.167-0.004-0.053-0.010** (0.116) (0.003) (0.194) (0.005) Term Length -0.004 0.000 0.006-0.000 (0.005) (0.000) (0.009) (0.000) Education -0.033-0.001 0.071 0.003 (0.034) (0.001) (0.087) (0.002) Age -0.001-0.000-0.002 0.000 (0.002) (0.000) (0.005) (0.000) Provincial Government -0.012-0.001-0.016 0.000 Working Experience (0.016) (0.000) (0.035) (0.001) Fiscal Expenditure 0.424*** 0.904*** (0.056) (0.092) Constant -4.236*** 0.039*** -2.399*** 0.006 (0.822) (0.004) (0.462) (0.006) Observations 2,640 2,640 1,330 1,330 adjusted R 2 0.905 0.868 0.879 0.764 Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. Progressive Academic Publishing, UK Page 97 www.idpublications.org

Table 5 demonstrates the results of officials public goods/services provisions behavior after obtaining the interim rank. I interchangeably use logarithm terms of two kinds of expenditures while controlling for logarithm of total fiscal expenditure and ratios of them over total fiscal expenditure. We can see in most of specifications the coefficients estimates on level and quadratic terms of rank are positive and negative respectively, indicating an inverse U-shaped relationship. However, only the coefficients estimates on Rank*D and Rank 2 *D in Column (4) are significantly positive and negative with lowest point around 0.4. Therefore, I obtain marginal evidence that public goods/services provisions exhibit U-shaped relationship with interim rank. The pattern still results from the two distinct incentives: the promotion incentive and punishment fear. As a result, officials rank in top and bottom are engaged in some aspects for the economic growth while slack off on the public goods/services provisions which cannot directly pump the economy, while those rank in the middle might focus on enhancing residents welfare and providing sufficient public goods/services. 4.3 Risk Taking Behavior Risk taking behavior is another latent consequence brought by the interim rank. After acquiring the interim evaluation, the top and bottom ranked officials might accelerate the economic growth at the sacrifice of environment, like attracting more pollution for better economic performance. The risk taking behavior resembles what Chevalier and Ellison (1999) argue in the example of fund managers and Genakos and Pagliero (2012) for weight lifting athletes. If it is the case, then the pollution should exhibit a U-shaped relationship with interim rank. I exploit the data from China Statistical Yearbook on Environment and China Industrial Survey to investigate risk taking behavior, i.e., pollution attractions (See Table 6 ) Table 6: Risk Taking Behavior: Pollution Amount (1) (2) (3) (4) (5) (6) VARIABLES waste water Sulphur Dioxide Smoke Dust Functional Form Logarithm /GDP Logarithm /GDP Logarithm /GDP Rank -0.232-6.135 0.045 47.363-0.156 9.087 (0.169) (3.947) (0.275) (66.406) (0.290) (79.583) Rank 2 0.268 9.072** -0.008-30.994 0.176-2.523 (0.170) (3.995) (0.278) (67.225) (0.294) (80.720) Rank*D -0.002 4.440* -0.206-4.746-0.105* -103.316* (0.135) (2.665) (0.220) (53.328) (0.060) (60.120) Rank 2 *D -0.039-6.226* 0.167-0.006 0.136* 116.765* (0.160) (3.761) (0.261) (63.355) (0.080) (67.323) Term Length 0.006 0.034-0.003 0.535 0.003 6.223* (0.007) (0.165) (0.011) (2.772) (0.012) (3.323) Education 0.064-0.244 0.024-10.701 0.068 26.412 (0.046) (1.083) (0.075) (18.096) (0.079) (21.777) Age 0.005-0.122 0.004 1.417 0.011* 2.226 (0.003) (0.079) (0.005) (1.319) (0.006) (1.584) Provincial Government -0.045* -1.050* -0.012 13.954 0.003 4.182 Working Experience (0.025) (0.572) (0.040) (9.633) (0.042) (11.545) Progressive Academic Publishing, UK Page 98 www.idpublications.org