Promotion Incentive: Corruption and Its Implications

on Political Budget Cycles in China

Liutang Gong1, Jie Xiao

2, and Qinghua Zhang

3

Guanghua School of Management, Peking University, 100871, China

Abstract

In addition to economic performance, this paper explores another incentive factor of local

political leaders in China---being clean (or staying away from corruption). Our data suggests that

both factors are crucial to provincial leaders’ promotion. Given these incentives, this paper

establishes a model to describe the decision-making process of political leaders regarding fiscal

expenditures, and conducts an empirical test of political budget cycles using Chinese

provincial-level data from 1990 to 2006. The findings show that promotion incentives drive

cyclical fluctuations in various types of fiscal expenditures that are synchronized with the timing

of the National Congress of the Communist Party (NCCP). Specifically, the growth rate of

infrastructure expenditure significantly reduces, while that of administration expenditure increases

during the years in which the NCCP takes place. This paper also tests another possible channel of

political budget cycles; that is, the time inconsistency effect caused by the turnover of provincial

leaders. This effect turns out to be insignificant for all types of expenditures.

Keywords: Corruption; Incentive Role; Political Budget Cycles; Fiscal Expenditure; Time

Inconsistency

JEL Classification: H72; D90; H30

1 Guanghua School of Management, Peking University, China, E-mail address: ltgong@gsm.pku.edu.cn

2 Guanghua School of Management, Peking University, China, E-mail address: xiaojiepku@pku.edu.cn

3 Guanghua School of Management, Peking University, China, E-mail address: zhangq@gsm.pku.edu.cn

1. Introduction

This paper studies the role of corruption in promotion of provincial leaders in

China and explores its implications on political budget cycles. In China’s M-structure

political system, the central government controls the promotion or renewal of the

tenure of local government leaders. The evaluation criteria used by the central

government thus greatly influence the decision-making processes of local government

leaders. Since the reform, economic performance has been the dominating factor in the

evaluation of provincial leaders (Li and Zhou, 2005; Maskin, Qian, and Xu 2000;

Blanchard and Schleifer, 2000; Qian, Weingast and Montinola, 1995). However,

accompanying the country’s rapid economic growth, corruption has become an

increasingly serious problem that is causing wide social discontent and eroding the

ruling power of the Chinese Communist Party (CCP). With the increasing concern over

corruption, central government has carried out sharp measures to corruption activities

in recent years. The nationwide anti-corruption event is a sign that the veto power of

corruption scandals is gaining more weight in the central government’s personnel

control over provincial leaders. A provincial leader who becomes involved in a

corruption scandal will find her chance of promotion to be greatly reduced, even if she

neither committed the corruption herself nor was directly responsible for the incident.

Moreover, such officials may be removed from office and may even face legal

investigation and indictment.

Using data on the turnover of provincial secretaries between 1990 and 2010, we

first examine the role of various incentive factors in determining provincial leaders’

turnovers. We find that in addition to the positive effect of economic performance,

which is well-documented in the literature, corruption scandals significantly lower the

probability of getting promotion. Another interesting finding is that increased

administration expenditure improves the chance of promotion. This may be achieved

by spending more money on forming relationship networks (in Chinese, Guanxi wang)

with central government officials or beautifying public images. In China, the

designation of provincial leaders is determined by the central government. However,

there is no institutionalized rule to follow regarding the decision process. As a result,

maintaining good Guanxi wang is essential to local leaders.

We then explore the implications of promotion incentives on provincial leaders'

fiscal decisions. We first develop a simple model to illustrate the decision-making

process of government leaders regarding fiscal expenditures. Our empirical analyze

shows that political budget cycles exist in China and they are synchronized with the

timing of the National Congress of the Communist Party (NCCP) which is the most

important event in the reshuffling of political power in China. More importantly,

different types of fiscal expenditures have different cyclical patterns, consistent with

the promotion incentive story.

Specifically, we find that provincial leaders increase their infrastructure

expenditure prior to an NCCP year because it has a long-lasting and lagged effect on

the local GDP growth. However, they significantly reduce infrastructure expenditure

during the NCCP year. One explanation would be that pork-barrel public

infrastructure projects t end to foster corruption, which the leaders would concern

more about during NCCP. A Chinese idiom best captures this idea: “bu qiu you gong,

dan qiu wu guo,” which means “no deeds, no mistakes.” Although the focus of this

paper is not to tackle the causality of infrastructure spending and corruption, there is

considerable evidence (Kenny, 2006 and Tanzi and Davoodi, 1998) of widespread

petty corruption in the area of infrastructure connections as well as large-scale of

corruption to gain construction contracts and licenses. Figure 1 reports the national

average infrastructure expenditure growth and the frequency of corruption scandals

by year and we can find a roughly synchronization between infrastructure

expenditure growth and corruption scandals. Figure 2 shows a positive correlation

between average infrastructure expenditure and the frequency of corruption scandals

across provinces. Moreover, we find that provincial leaders increase their

administration expenditure significantly during the NCCP year to build Guagnxi

wang or beautify public images, while such expenditure returns to its normal level in

other years. Our findings indicate that incentives for promotion drive cycles of

provincial fiscal expenditures in China.

Our paper is closely related to “the political tournament theory” proposed by Li

and Zhou (2005), who have employed a panel of leadership observations during

earlier reform years (1978-95) to show that economic performance was correlated

with leadership turnover until the mid-1990s. Other papers shedding light on cadre

evaluation in CCP institutions include Chen, Li and Zhou (2005), Shih, Adolph and

Mingxing Liu (2012) and Maskin, Qian, and Xu (2000). The above literature finds

that economic performance and political connections are important factors that

influence promotion. Edin (2003) analyzes the political control among township

leaders in China. Our paper contributes to the literature by incorporating another

important factor, corruption, into political leaders' promotion system.

This paper is also related to the literature on political budget cycles. The political

budget cycles theory studies the real effects of elections on policy makers’ instruments.

There is much literature documenting political budget cycles in democratic regimes.

For example, Rogoff (1990) designs a signaling model in which a political budget

cycle arises due to information asymmetries about the incumbent’s competence in

administering the production of public goods. Drazen (2001) incorporates both

monetary and fiscal policy in a rational opportunistic framework with separate

monetary and fiscal authorities. Empirical evidence from Canada (Kneebone and

McKenzie, 2001) and Mexico (Gonzales, 2002) indicates that the incumbent

government tends to increase spending in areas such as infrastructure and current

transfers to earn votes. Using data on 42 developing countries from 1975 to 2001,

Vergne (2009) finds that during election years, public spending shifts toward more

visible current expenditure, in particular wages and subsidies, and away from capital

expenditure. Katsimi and Sarantides (2012) also find that elections in OECD

countries shift public spending towards current expenditures at the cost of public

investment. Scholars have also uncovered economic policy cycles outside of Western

democracies. Bunce (1980) finds that in the Soviet bloc, first secretaries seem to pump

up public consumption in the period immediately following succession, and then

move toward less popular policies once the succession crisis has been resolved. Guo

(2009) applies China’s county-level data from 1997 to 2002 and shows that the growth

in fiscal spending per capita is fastest during a leader’s third or fourth year in office.

However, few studies have examined the channels that drive the political budget

cycles in China. Tsai (2013) studies the cyclical patterns of both aggregate and

disaggregate fiscal expenditures using Chinese provincial level data from 1980-2006.

Tsai's paper proposes a theory that illustrates how economic performance as

promotion incentive, drives fiscal cycles that synchronize with NCCP. Its empirical

analysis looks at capital expenditures and social expenditures separately and finds

capital expenditures increases two years before NCCP while returns to normal

during NCCP.

Our paper aims to understand the driving channels of political budget cycles.

This paper differs from Tsai's work in the following three aspects: First, using

political turnover data at the provincial level, we examine the effects of various

incentive factors, based on which we develop our model of fiscal cycles that

emphasizes both the roles of corruption and economic performance. We then conduct

our empirical analysis of provincial fiscal expenditures at both the aggregate and the

disaggregate level to test the implications of our model. Second, this paper and Tsai’s

work focus on different issues. While Tsai’s paper mainly focuses on the cycles of

aggregate and disaggregate expenditures, ours explore the driving forces of China’s

political budget cycles. We investigate a specific type of fiscal expenditure,

infrastructure, which is a more refined category than "capital" in Tsai's paper. On

the one hand, infrastructure expenditures may boost economic growth (Demurger,

2001; Czernich, Nina, et al., 2011). On the other hand, it may foster corruption and

hence is more risky. We find infrastructure spending increases in the year before

NCCP but declines during the NCCP. By looking at the distinct growth trend of

infrastructure expenditure over a political cycle, we may convince the roles of

different incentive factors in driving such cycles.

Thirdly, this paper explores another potential channel that may contribute to fiscal

fluctuations in China that receives little attention in the aforementioned works.

Specifically, it tests whether time inconsistency influences the fiscal cycles in China.

Time inconsistency plays an important role in a government’s decision-making process

when there are political turnovers in democratic countries (see Perrson and Svensson,

1989; Alesina and Tabellini,1990; De Haan and Sturm,1994, 1997; Grilli, Masciandaro

et al., 1991; Pettersson Lidbom, 2001; Crain and Tollison, 1993). Although China has

a single ruling party, political turnovers at the provincial level are not infrequent

compared to those in democratic countries. According to Table 1, between 1990 and

2010, the total number of turnovers was 123 party secretaries and 137 governors,

making an annual average of 5.9 and 6.5. Different local officials may have distinct

political preferences or belong to different political factions. Therefore, time

inconsistency might influence the cyclical patterns of local fiscal policies. Following

Pettersson Lidbom (2001) and Crain and Tollison (1993), we investigate whether the

frequency of provincial official turnovers affects the fluctuations of provincial fiscal

expenditures. We find that political turnover has little effect on the volatilities of

provincial fiscal expenditures at both the aggregate and the disaggregate level, except

that more frequent provincial party secretary turnovers are associated with more

fluctuations in the local government’s administration spending. Our findings indicate

that China’s centralized personnel control of provincial leaders, together with the

current incentive structure, seems to minimize the time inconsistency of local fiscal

policies, which is not common in democratic countries.

The main contributions of this paper are as follows. First, it examines the

promotion incentives of provincial leaders in China, incorporating the changes in the

CCP’s cadre evaluation system since 1990, which give more weight to the veto power

of corruption scandals, in addition to the economic performance that is well

documented in the literature. Second, it explores the implications of various incentive

factors on provincial governments' fiscal cycles. Finally, this paper tests the time

inconsistency effect of political turnovers in China, which is another potential channel

that may influence the pattern of fiscal cycles.

The rest of the paper is organized as follows. In Section 2 we provide a brief

institutional background of the political structure in China. In Section 3 we describe

the data. In Section 4 we outline the empirical identification strategy of officials’

turnover incentives and the key results. In Section 5 we establish a life-cycle model of

local government leaders to illustrate how the incentive role affects their decision

making regarding various types of fiscal expenditures and the implications on political

budget cycles. In Section 6 we tests the political budget cycles theory and report the

main results. Section 7 analyzes the time-inconsistency effect of officials’ turnover on

volatility of fiscal expenditures. Section 8 concludes.

2. China's Political Structure and Fiscal Incentive

China’s political system has five layers of administration: the central, provinces,

prefectures, counties, and townships. China’s Communist Party (hereinafter CCP) is

the sole ruling party in China. The Central Committee of the Communist Party has the

ultimate power over personnel within the system. The top political position in each

province is the provincial party secretary, who shapes the direction of policies,

followed by the provincial governor, who is in charge of implementing policies and the

day-to-day management of government. Therefore we only include provincial party

secretaries in our analysis of promotion incentives.

Under China’s highly centralized political structure, what matters most to a

provincial leader’s political career is how the CCP evaluates him or her, which

essentially determines whether the provincial leader will get promotion, remaining in

office or be removed from office. In China, government officials typically have few

outside career options, thus they have very strong motivation to hang on to their

political careers (see Li and Zhou, 2005). The evaluation criteria used by the CCP

provide important incentive factors that influence the decision making of provincial

leaders who have ultimate authority in the allocation of economic resources within

provinces. Promotion incentive thus becomes a distinctive channel through which

political turnovers may influence the cyclical fluctuations of local fiscal policies.

Accompanying the far-reaching economic reforms that began in 1978, China also

launched reforms in its personnel control system. In 1980, the lifetime tenure of party

and government officials was abolished. Replacing loyalty and obedience to the CCP,

economic performance became the most important factor in the evaluation of local

leaders, in addition to other competence-related indicators such as young age and

education. As a result, local officials raced to increase GDP growth in their efforts to

stand out among their peers and impress central government.

However, along with the rapid economic growth, corruption boomed and has

become an increasingly serious problem. According to Wang’s (2013) estimates, based

on the urban household survey conducted by China’s National Bureau of Statistics, the

total grey income amounted to 2.44 trillion, 4.65 trillion, and 6.24 trillion yuan in

2005, 2008, and 2011, accounting for 12.2%, 13.6%, and 12.2% of total annual GDP,

respectively. Grey income is defined as household income from unspecified sources

and is interpreted by the author as income related to corruption. The above numbers

indicate that corruption has largely infiltrated the Chinese economy. Moreover, Wang

finds that higher-income households have a higher percentage of grey income in their

total income, which suggests that corruption is worsening along with the widening of

income inequality that is afflicting Chinese society.

Aware of the reality that corruption is causing severe social discontent and

harming social stability, the Party has started to take a firm anti-corruption stand, and

efforts have been made to crack down on corrupt government officials in recent years.

The Central Commission for Discipline and Inspection of the Communist Party (CCDI)

and the Ministry of Supervision perform the supervision and inspection f unct ions

over cadres and government officials. People can anonymously report any corruption

or violations of the law by government officials to the CCDI. If the CCDI believes

there is enough evidence, it can Shuang Gui the officials involved. Shuang Gui means

putting an official into confinement at a certain location and time specified by the

CCDI, during which the official is cut off from office power and all outside

connections and must confess the truth to the Party. If any criminal conduct is found,

the case will be taken over by the police. Once Shuang Gui-ed, the official’s political

career is over. It turns out to be an effective way for the CCDI to investigate big

corruption cases.

According to an article on Xin Min Zhou Kan by Liang-fei Chen, published on

September 27, 2013, during the first decade after the beginning of the reform in 1978,

only two government officials at the provincial or ministry level were removed from

office because of corruption. During the second decade, this number increased to 15.

In the most recent decade, between 2003 and 2012, the number jumped to more than 80.

Meanwhile, the amount of money involved in corruption per case has skyrocketed,

from tens of thousands of yuan in the 1980s to tens of millions in more recent cases,

with a maximum of 200 million. After the 18th

NCCP, which was held in 2012, 50

government officials at the provincial or ministry level have been investigated or

arrested because of corruption scandals.

Against such a background, the CCP’s evaluation criteria for provincial leaders

have gradually changed, with the emphasis now on harmonic social development

rather than GDP. The veto power of corruption scandals has gained more weight. If a

provincial official becomes involved in a corruption scandal, her chance of promotion

will be greatly reduced, even if the official did not commit the corruption herself and

was not directly responsible for the incident. Moreover, the official may be removed

from office and may even face legal investigation and indictment.

This situation leaves provincial leaders with a dilemma. On the one hand, GDP

growth is still important, which means t h a t infrastructure investment is essential.

Note that 48% of the growth in Chinese GDP over the past three decades has come

from capital investment (Wang and Yao, 2003). On the other hand, in China,

government infrastructure investment is often associated with procurements of big

pork-barrel government projects and land development, which tend to foster corruption.

Figure 2 shows a positive correlation between government infrastructure expenditures

and the frequency of corruption scandals.

In China, the most important political re-shuffling event is the National Congress

of the Communist Party, which has been held every five years since the late 1970s.

During NCCP, new members are elected into Political Bureau of Central Committee

from provincial party secretaries and other government officials at similar levels. In

other words, this is the time when provincial party secretaries have the chance to get

promotion. Table 1 shows the frequency of turnovers of both provincial party

secretaries and governors by year. Almost 50% of the provincial leaders’ turnovers

during our study period occur in an NCCP year or one year after it. As there is no

institutionalized process regarding the appointment of provincial leaders in China, we

treat the regular occurrence of the NCCP as an exogenous and anticipated political

cycle, and examine its association with the cyclical fluctuations of local fiscal policies.

Figure 3 shows the average growth rates of various fiscal expenditures across

provinces over the 1990-2006 periods. The red bars indicate NCCP years. It is clear

that there are ups and downs coinciding with the NCCP. Preliminary findings are: total

fiscal revenue and fiscal expenditure growth rate decreases during NCCP and so does

the growth rate of infrastructure expenditure.

During the NCCP year, the CCDI and the Ministry of Supervision conduct stricter

supervision and inspection of government officials. Meanwhile, the competition from

peers becomes tenser. A corruption incident occurring in that year may trigger a CCDI

investigation into previous years, which may wake a sleeping volcano because nobody

can ensure complete cleanliness in the past. As a result, the provincial leaders switch

their infrastructure investment to the years prior to the NCCP. In the NCCP year, they

adopt the philosophy of “no deeds, no mistake,” and slow down by reducing fiscal

expenditures on infrastructure. At the same time, they increase their administration

expenditures in the NCCP year to strengthen their relationship networks with central

officials and to beautify their public images, which are visible to their superiors. Hence,

the cycles of fiscal expenditures are in synchrony with the NCCP.

Finally, the tenure system of government officials in China requires that Party

Cadres, governments of all levels should rotate on and off positions. On August 6,

2006, the central authorities issued “The Rotation Provisions of Party and

Government Leading Cadre”, which dictate that all government leaders at or above

county level must change positions after 10 consecutive years in office. These

requirements lead to more official turnovers in addition to those in NCCP years.

According to Table 1, 50% of turnovers take place in non-NCCP years, and these are

rarely synchronized across provinces. We explore this variation to test how

time-inconsistency influences fluctuations in fiscal policies in China.

3. Data

Our data include data on provincial leaders’ personal characteristics and turnover

information, corruption scandals and accidents data, provincial economic data and

provincial fiscal data. The details of the above four data sets are outlined in the

following sections. Table 2 shows the summary statistics of the data.

3.1. Provincial Leaders’ Turnover Personal Information

We collected data on the turnover of provincial leaders from 1990 to 2010.

Provincial leader refers to the provincial party secretary, who is the top political figure

in each province4. For each year, we record whether the current leader of each

province stays in office, moves to a lateral position, gets promoted or steps out of

office (either due to retirement or termination). Then we classify the turnovers into

three categories: promotion, termination and remain at the same level. We define

promotion as becoming a member of Political Bureau of Central Committee or other

government positions at vice-national level, such as the president of the Supreme

People's court. We define termination as retirement, demotion or death. Remaining at

the same level includes both shifting to a lateral position and stays in the same

position.

Table 5 shows the frequencies of provincial officials’ turnovers. There are four

NCCPs from 1990 to 2010, in 1992, 1997, 2002 and 2007 respectively. However,

there are four or more provincial secretary turnovers in about 75% of provinces and

four or more provincial governor turnovers in around 82% provinces. “The Rotation

Provisions of Party and Government Leading Cadre” may explain part of the frequent

turnovers, while another reason is that leaders may be permanently removed from

office as a result of political scandals related to either corruption or serious safety

accidents.

We also collect personal information on each provincial party secretary. From

1990 to 2010, 143 officials served as provincial party secretary in 28 provinces. Their

resumes are posted to the public online from which we can obtain their age, education

4 We include only provincial party secretary in our sample because: (1) provincial party secretary has ultimate

power in each province in China; (2) In practice provincial party secretary has higher level than governor and thus

are more qualified to be elected into Political Bureau of Central Committee. (There are exceptions such as Beijing)

and tenure, which is the length of time they have served in their current position.

Other features such as central connections (Li and Zhou, 2005; Shih et al, 2012) and

faction ties (Dittmer, 1995; Nathan and Tsai, 1995; Pye, 1992) also have impact on

the officials turnover and are therefore included in the data. A provincial leader has

central connection if he/she has worked in central government earlier in his/her life.

Also we divide official faction affiliations into four categories: Shanghai, Tuan, Taizi

and None. Shanghai faction represents followers of Jiang Zemin. Tuan faction

represents cadres and government officials who originated from the Communist Youth

League. These officials are also regarded as followers of Hu Jintao. Taizi faction is

similar to Princelings, which represents the descendants of prominent and influential

senior communist officials in the country. All the other officials are defined as none

faction.

One additional concern is that central government may first place some

promising officials as leaders of some provinces so that they can be promoted to the

central later. We use a dummy variable “Strategic Shift” to define this situation. It

equals 1 if the provincial party secretary comes from other provinces or institutions.

We also include the interaction of central connection and Strategic Shift to test

whether provincial leaders from central government are taking more advantage in

political competition.

We have 21 promotions and 58 terminations in our sample. In figure 5 and figure

6 we graph both the promotion and termination distributions in provinces and factions.

Data shows that officials from provinces such as Guangdong, Beijing, Shanghai,

Shandong and Zhejiang are more likely to get promotion. In addition, officials from

Shanghai faction take more advantage in the promotion process.

3.2. Corruption Scandals and Accidents Data

We argue that corruption scandals or accidents, once exposed to the public, are

highly likely to lead to the termination of a provincial leader’s political life. Therefore,

we search the news reports by year from the Xinhua multimedia database5, which

provides historical news and information and is operated by Xinhua News Agency, the

official press agency of China. For corruption exposures, we use the keywords

“province,” “tanwu” (pinyin for embezzlement), and “fubai” (pinyin for corruption) in

our search. The results include all of the corruption exposures in some provinces, from

provincial official to local bank teller, and from Party cadre to entrepreneur. However,

we only count the corruption exposures related to government officials at the municipal

level or above. We count each corruption scandal only once. As there is a rigid

hierarchy in China’s administration system, the corruption exposures at lower levels of

government or outside the political field have little effect on the turnover odds for

provincial leaders. To identify news about major accidents, we use the keywords

5 http://info.xinhua.org/cn/index.jsp

“province,” “zhongda” (pinyin for major or severe) and “shigu” (pinyin for accidents)

in our research. We then count the number of news stories on major accidents related to

environmental pollution or site safety.

The frequencies of corruption scandals and accidents are reported in Tables 6 and 7,

by year and by province, respectively. There is quite a wide variation across provinces

and over years. According to the summary statistics in Panel 2 of Table 2, the mean

number of exposed corruption scandals is 0.379 per year, while the standard deviation

is 0.905. The mean number of exposed major accidents is 0.413 per year, while the

standard deviation is 0.907.

A minor concern about the data is that Xinhua News Agency is closely related to

the central administration. To double check the reliability of the corruption scandals

data, we look up for more objective data source. We collected data about corruption

scandals from United Daily News in Taiwan using the same keywords. United Daily

News is founded in 1951 and is one of the three biggest newspapers in Taiwan. We

believe United Daily News is a better source because Taiwan is very close to China

but relatively independent from China mainland. In addition the report of Taiwan

media is more efficient and faster due to less regulation. However, data comparison

from Xinhua multimedia database and that from United Daily News (Table 8) shows

that in most cases, the scandal count from United Daily News is smaller than that

from Xinhua News Media, which brings the problem of small data variance. When we

look up into each piece of news, we found that United Daily News only reports

important or big corruption scandals (relating to government officials at the provincial

or ministry level), but ignores scandals at municipal levels. In our empirical work, we

use the data from Xinhua Media and make a thorough discussion about the corruption

issue.

3.3. Provincial economic Data

Our provincial economic data are taken from the China Statistics Yearbook

(1990-2010). Economic variables include population, GDP, per capita GDP, urban

income and the share of secondary industry.

In addition, to separating the aggregate shocks at the national level from the

cycles of the National Congress of the Communist Party, which is our main interest,

we also control for the national GDP over time.

3.4. Provincial Fiscal Data

In section 6 we use fiscal data to identify political budget cycles in China. Our

fiscal data covers 28 provinces in China from 1990 to 2006 (excluding Sichuan,

Chongqing, Tibet, Hong Kong, and Macao)6, including Beijing, Tianjin and Shanghai.

6 Sichuan and Chongqing are omitted because Chongqing was separated from Sichuan in 1997, which is in the

The data were sourced from the China Fiscal Statistics Yearbooks (1990-2006), China

Statistics Yearbooks (1990-2006) and New China 60 Years Statistics Data.

We include fiscal expenditure at both the aggregate and disaggregate levels.

Aggregate expenditure is the provincial total expenditure. Disaggregate expenditure

includes infrastructure expenditure, agricultural expenditure, administration

expenditure and education, science, medical and culture expenditure (hereafter

referred to as ESMC expenditure). The fiscal data ranges from 1990 to 2006 because

of the Government Revenue and Expenditure Classification Reform in 2007. We do

not include social security expenditure in our study due to limited data availability.

Note that all of the expenditures in our analysis are budgetary expenditures. We

do not include data on any funds outside of the government budget. All variables use

real values deflated by the RPI (Retail Price Index).

Within our sample period, agricultural expenditure accounts for about 7.2% of

total expenditure on average, ESMC for around 25% and administration and

infrastructure for 11% and 9.7%, respectively (see Table 3). Because China

experienced rapid economic growth during our sample period, we use the growth rate

of each expenditure category to construct the regression panel. A unit root test shows

that the growth rate panel is stationary (see Table 4).

Our data also includes total provincial budgetary revenues and the total

provincial fiscal support population, which is the population whose wages and

subsidies come from the fiscal budget. The data on the fiscal support population are

taken from Fiscal Statistics of Cities and Counties in China (1993-2009)7.

In the next section, we present our empirical analysis of promotion incentives.

We first study the turnovers of provincial secretaries to identify incentive roles that

affect officials’ promotion odds. Second in Section 5 we present a life-cycle model of

official based on different incentives. Then in section 6 we discuss the implication of

such incentives on political budget cycles, incorporating the timing of NCCP. Notice

that the NCCP is a reshuffle of personnel among top leaders, when provincial leaders

are most likely to get promoted. Finally, we examine whether time inconsistency

exists in provincial fiscal expenditures due to turnovers of provincial leaders.

4. Turnovers of Provincial Leaders and the Incentive Role

4.1. Basic Specification: Ordered Probit Model

In this subsection, we conduct an empirical analysis to study the turnovers of

provincial leaders, aiming to shed light on the incentive role that drives such fiscal

cycles.

middle of our sample period. Thus the reported statistics over years may be inconsistent. 7 We use the fiscal support population in cities and counties in each province because such data are not available at

the provincial level until 1998, which is too short for our analysis.

Following Li and Zhou (2005), we use an ordered probit model to examine

whether provincial leaders are promoted, remain at the same level, or their positions

are terminated. Suppose there is a continuous latent variable corresponding to the

probability of promotion for each provincial leader, *y . Although we cannot observe

this variable, we can observe the discrete outcome variable y , which equals 0 for

termination, 1 for remaining at the same level (including lateral moves and staying in

the same position), and 2 for promotion. Assume that the latent variable *y is a

linear function of our explanatory variables x , *y x , where is a vector of

coefficients and is the error term. Denote 1 and 2 as two cut off points of *y . A

provincial leader is terminated if *1y , remains at the same level if *

1 2y

and is promoted if *2y . The ordered probit model is expressed as:

1Prob( 0 | ) ( - )iy x x ,

2 1Prob( 1| ) ( - )- ( - )iy x x x ,

and

2Prob( 2 | ) 1- ( - )iy x x ,

where is the cumulative standard normal distribution function.

The explanatory variables that influence the promotion probability include

various performance measures that the Central Committee of the Party uses to evaluate

its leaders. As discussed, the Central Committee focuses on two issues. One is

economic performance, for which we include the provincial GDP growth in the

regression. We also include 1-year lag and 2-year lag GDP growth to capture the

effect of economic performance in earlier years. The other is whether the provincial

leader is clean, for which we include the number of news exposures related to

corruption in the regression, denoted as corruption. Our main hypothesis is that

provincial GDP growth has a positive effect on the probability of promotion for

provincial leaders, while corruption has a negative effect. We also include the number

of news exposures related to major environmental accidents or major work-site safety

accidents in our estimation, denoted as accidents, because such accidents may

stigmatize the provincial government’s public image and cast doubt on the leader’s

capability. We have checked the correlation between corruption and accidents; see

Table 9. They are positively correlated, but not so closely.

There are two major concerns about this identification strategy. The first is the

data source bias. Xinhua News Agency seems to be closely related to the central

administration. It might happen that there is a higher count of corruption/accidents in

some provinces just because the central administration wants to spread bad news and

damage the reputation of unwanted provincial leaders. In order to address the data

source bias, we collected data from United Daily News, a Taiwan media, and compare

the corruption scandal counts from both sources. For example, in Table 8, we report

corruption scandal counts in four representative provinces: Anhui, Guangdong, Jilin

and Shaanxi. They differ in both location and economic development levels. In most

years, the scandal counts from United Daily News are smaller than that from Xinhua

News Media. When we look up into each piece of news, we found that United Daily

News only reports important or big corruption scandals, often relating to government

officials at the provincial or ministry level, but ignores scandals at municipal levels.

Maybe that is because small scandals cannot easily draw people’s attention. But this

leads to the limitation of the corruption data and the data variation is quite small.

Therefore, we use data from Xinhua News Media in our empirical work.

The other problem is the endogeneity problem of corruption scandals. The

central administration may put more effort on investigations if they don’t want to

promote certain officials, which also will be reflected in the news count. We argue

that our corruption data does not relate to particular provincial leaders. It actually

includes all corruption scandals in the provincial leader’s jurisdiction, relating to

officials both at municipal level and provincial level. Therefore, it is less likely that

the central administration would rock the whole province off if they do not want the

provincial leader.

Other controls in our estimation include the provincial leader’s personal

characteristics such as age, education, tenure and ethnicity. Other political features

include central connection, strategic shift and faction. Because the level of economic

development of a province could affect the career prospects of its leaders, we also

control for the lagged provincial GDP per capita. Finally, we include the province fixed

effect and year dummies in our estimation.

4.2. Results

The results of the ordered probit estimation are reported in Table 10. In the first

column of Table 10, only current year corruption is included in the estimation. The

results show that coefficient of corruption is around -0.2 at the 5% significance level,

implying that corruption has a negative effect on the probability of promotion. For

provinces with one more corruption scandal, the probability of provincial leader

getting to the next higher level (termination to same level, or same level to promotion)

is reduced by 0.2. In Column 2, we add 1-year lagged corruption and 2-year lagged

corruption in the estimation. Similarly, the significance level of the negative effect of

current year corruption is 5%. Both coefficients of 1-year lagged corruption and

2 - year lagged corruption are insignificant, implying that there is little lagged effect.

Admittedly, corruption scandal count does not provide a completely untainted

assessment of the underlying corruption. Actually it reflects both underlying

corruption degree and the scrutiny from the central government. However, we cannot

distinguish these two effects and finding a feasible instrument variable for corruption

is quite difficult at the moment. But, think of it in another way, if the central

administration does put more efforts in digging out scandals to ruin the reputation of

unwanted officials, it is side evidence that corruption scandals have significant

negative effect on promotion chance. Therefore, in spite of possible endogeneity

problem it may have, the above analyze at least provides a possible way that how

corruption may affect the promotion odds.

Next, we turn to the coefficient of provincial GDP growth. We include the current

year GDP growth, lagged 1 - year growth and lagged 2 - year growth in column (3) and

(4). They are almost insignificant, but the coefficient of lagged 2 - year growth is

significant at 10% level in column (3), implying that the Central Committee evaluates

the provincial leader based on economic performance in earlier years (Li and Zhou,

2005). Since infrastructure is highly correlated with economic performance, this

result lends support to rapid economic growth in the middle period of official’s tenure

(Guo, 2009). Compared with Li and Zhou’s findings (2005), our results show that

earlier economic performance matters more in the evaluation of provincial leaders,

but the importance of economic growth in CCP’s evaluation has somehow weakened

during recent years.

We then explore the effect of accidents on the probability of provincial secretaries’

promotion. We add the current year accidents, 1-year lagged accidents and 2 - year

lagged accidents to the specification of Column (3) i n Table 10. The results are shown

in Column (4). The coefficient of 1-year lagged accidents, but not current year

accidents, is negative and significant at the 5% level, implying that major

environmental accidents or work-site safety accidents significantly reduce the

promotion odds for provincial secretaries. Accidents lagged by one year may be more

important than current year accidents because it usually takes some time to investigate

an accident and find out whether it is due to purely technical reasons or due to a human

mistake and, in the case of a human mistake, whether there is any political factor

involved.

In all of the regressions in Table 10, the coefficient of provincial secretary’s age is

negative and significant, indicating that older officials are less likely to get promoted.

Faction matters in the promotion odds too. The coefficient of faction_shanghai are

positive and significant at 10% level in column (4), while that of faction_tuan is

negative and significant at 10% level in column (1), suggesting that officials from

Shanghai faction take more advantages in political promotion.

Li and Zhou (2005) and Guo (2009) suggest that the political mobility of Chinese

cadres may be more institutionalized and based on their actual performance than is

commonly perceived. Our empirical studies confirm that the Central Committee makes

promotion decisions based on the evaluation of the actual performance of provincial

leaders. Moving the current literature one step forward, we show that in addition to

economic performance, the veto power of corruption scandals is gaining more weight

in the Party Central Committee’s personnel control over cadres. This gives provincial

leaders a strong incentive to manipulate fiscal expenditures over their political career,

which may generate fiscal cycles that coordinate with the NCCP, at which time the

Central Committee evaluates cadres and re-adjusts the assignment of government

officials. In Section 6 we show that during the NCCP years, provincial leaders are

likely to implement a “No deeds, no mistakes” strategy and slow down the

infrastructure construction to minimize the risks to their careers.

4.3. Robustness Check

We conduct several robustness checks, which we discuss one by one.

Counting corruption scandals and accidents by month

In the above analyses, we count the media exposures of corruption scandals and

accidents by year. However, it is entirely possible that the provincial secretary takes

office in January, and thus the corruption and accident exposures counted that year

actually happen after, rather than before, the secretary’s turnover. To deal with the

timing problem more cautiously, we conduct the following robustness check.

We find out the exact months of secretaries’ turnovers, and count the number of

corruption and accident exposures within 12 months before the turnover and between

13 and 24 months before the turnover. If there is no turnover within a specific year, we

count the current year exposures and the exposures one year before. We then use the

corruption and accident data thus constructed to replace the yearly corruption and

accident data and rerun the regressions in Table 10. The results are reported in Table

11.

Table 11 shows similar results to those in Table 10. The coefficient of 11

month-corruption is slightly smaller than before (in absolute value), but is still

significant. Overall, the adjusted data robustness check confirms our finding that

corruption scandals and accidents have negative effects on the probability of promotion,

while GDP performance has a slightly positive effect.

Administration spending and promotion

Aside from actual performance, there might be some idiosyncratic factors that also

affect the turnovers of provincial leaders. Maintaining good personal relationships with

the central government officials and peer officials is helpful. According to the official

classification of administration expenditure, it includes spending both on personnel and

on maintaining daily government functioning. Hence, it is possible that provincial

leaders increase administration spending around the NCCP to strengthen Guanxi wang

with central officials and peer officials. We then test whether administration spending

has a positive effect on the probability of promotion or renewal of tenure. As the scale

of fiscal expenditure varies widely across provinces, we use the log of administration

in our estimation.

A complication is the endogeneity issue. The turnover of a leader may cause

changes in administration spending. To deal with this problem, we use the fiscal

support population in corresponding years as an instrumental variable for

administration spending. The fiscal support population is defined as the population

whose wages, subsidies and fringe benefits come from government administration

spending. Thus, the fiscal support population is correlated with the personnel

expenditure part of the administration spending. However, the fiscal support

population is unrelated to the turnover of provincial officials. During turnovers, the

government might hire more temporary workers. However, those temporary workers

are not included formally in the fiscal support population.

Table 12 shows the results. Column (1) reports the ordinary ordered probit

regression and Column (2) reports the IV regression. Column (1) shows that

administration spending have positive and insignificant effect on the turnovers of

provincial leaders. In the IV regression, in Column (2), the sign of the coefficient of

log administration spending is positive and significant at the 10% level, indicating that

administration spending has some positive effect on the probability of provincial

leaders’ promotion or renewal of tenure. Note the coefficients of corruption and

accidents remain fairly similar to those i n Column (4) in Table 10.

In the next section, we set up a life-cycle model to describe provincial leader’s

fiscal policy decisions, based on the incentives of corruption scandals and economic

performance.

5. Theoretical Framework

In Section 4, we provide supporting empirical evidence regarding the incentive

factors of government leaders. In this section, we develop a simple life-cycle model to

describe a local government leader’s political career. Our model demonstrates how a

local leader makes fiscal decisions over time given promotion incentives.

5.1. The Model

Suppose that there are two types of public expenditure: safe expenditure 1G and

risky expenditure 2G ,

1 2G G G . Safe expenditures refer to expenditures that

have little direct influence on economic growth but help to increase the chance of

promotion or renewal of tenure.

Administration spending can be regarded as safe expenditure. Administration

expenditure is used to maintain the government’s daily functions. It may also be used

to strengthen social networks and develop good interpersonal relationships with other

government officials, especially those in the central government. Other examples of

safe expenditure include spending on public education, government-sponsored

science funds8, medicine, culture, and media. Such social expenditures may improve

social welfare through re-distribution and improve the public image of the

government.

Risky expenditures refer to expenditures that may increase economic growth, but

at the same time may bring career risk to the officials. Infrastructure spending has the

characteristics of risky expenditure. Generally, the higher the infrastructure

investment, the higher the GDP growth. However, it is known that public

infrastructure projects are easy to foster corruption. Tanzi and Davoodi (1998) find

that corruption is prevalent in infrastructure projects, especially large civil engineering

projects. Corruption and embezzlement, once exposed to the public, is very likely to

lead to the termination of an official’s political career, especially in recent years in

China.

Assumption 1. A provincial leader in office faces a probability of career failure

2( )p G in each period, 20 ( ) 1p G and 2'( ) 0p G , where 2G is the risky

expenditure during the period. When a failure occurs, the leader’s political career is

terminated. In addition, he has to bear punishment, which means for the rest of his life,

his utility, denoted as M , is lower than the reservation level of ordinary citizens.

Without career failure, the political leader's utility function is ( )U X , where

X is the consumption level which depends on his income. The leader’s income

consists of two parts. The first part is 1( )G , which we assume is positively

associated with the government’s safe expenditure 1G . The second part is 2( )G ,

which we assume is positively associated with the government’s risky expenditure

2G . The specific functional forms may be different. The idea is that bigger

governments bring more rents to leaders, although rents from different sources may

be different. Note that here the rents are all legal income. We assume the above two

8 In China, government sponsored science funds are usually allocated to public universities or public research

institutes, and the outcomes of such funding are apparent over the long term. This is quite different from firms’ R&D

expenses, whose outcomes can be more directly applied to production in the short term.

income functions are twice differentiable, concave, and monotonically increasing,

namely

1 1'( ) 0, ''( ) 0G G , and 2 2'( ) 0, ''( ) 0G G .

Next, we turn to the dynamics of a leader’s political career. We consider four

periods, 1,2,3,4t of a representative local leader’s political life. The first term in

office consists of periods 1 and 2. The National Congress takes place at the end of

period 2 and the Central Committee of the Party decides whether the local leader can

stay in office for one more term. For simplicity, we treat the renewal of tenure,

promotion and same-level relocation all as “remaining in office”. In such cases, the

local leader will continue to take charge in periods 3 and 4 and retire afterwards. If the

local leader does not get a second term, he will become a normal citizen starting from

periods 3 and 4. He chooses public expenditure 1

tG and 2

tG in each period during his

term(s) to maximize the discounted sum of utility over his whole political life cycle.

The following shows the time line of events.

At the end of period 2, the Central Committee of the Party evaluates the

performance of the leader in periods 1 and 2 and decides whether the leader can

remain in office for the next 2 periods. If he is out of office, it is assumed that his

utility level will be the same as the reservation utility level of ordinary citizens which

is typically much lower than that of a leader in office. Let the reservation utility be

0U , which in our model is an exogenously given constant. Thus every leader has

strong incentive to remain in office longer.

Two things are essential to the evaluation of the leader by the Central Committee

of the Party: economic growth and whether the leader is non-corruptive. For

simplicity, we assume that conditional on no career failure (which is related to

corruption scandal or accidents) in the first two periods, the leader faces a certain

probability of remaining in office for a second term. This probability function has two

arguments. The first is risky expenditure in period 1, 2

1G (the subscript refers to the

period and the superscript to the type of expenditure), because risky expenditure such

as that on infrastructure, can boost economic growth. However, this effect usually has

a one-period lag before realization. The second argument is the safe expenditure in

period 2, 1

2G . Increasing safe expenditure may increase the chance of remaining in

office. Such an effect is contemporaneous. Formally:

Assumption 2. Conditional on no career failure, at the end of period 2, the probability

of a provincial leader remaining in office for a second term is 2 1

1 2( , )f G G , with

2 1

1 1 2( , ) 0f G G and 2 1

2 1 2( , ) 0f G G .

We solve the leader’s dynamic decision problem by using the backward

induction method. First, let us look at the second term (periods 3 and 4) if the local

leader indeed remains in office for a second term. In this case, the optimization

problem in period 4 is

1 2

4 4 4

2 2

4 4 4, ,

max [1 ( )] ( ) ( )X G G

p G U X p G M

subject to

1 2

4 4 4 4( ) ( ) /G G X T N . (1)

The leader’s utility function is the same as that of ordinary citizens. Let 4X be

the leader’s consumption in period 4. M is the utility when there is a career failure

and punishment is considered. The official is also a taxpayer and has to pay the same

tax as ordinary citizens. Assume that the government balancing the budget every

period. Therefore, the total tax T to be collected in each period is

1 2 1 2( ) ( )T G G G G . In Period 4, each person has to pay

1 2 1 2

4 4 4 4 4/ [ ( ) ( )] /T N G G G G N .

Solving the above problem, we have the first-order conditions

1

4'( ) 1/ ( 1)G N (2)

and

2

2 4 44 2

4 4

'( )[ ( ) ]( 1) '( ) 1

[1 ( )] '( )

Np G U X MN G

p G U X

. (3)

Let the optimal public expenditures be 1*

4G and 2*

4G , and let the optimal

consumption level be *

4X . Then the maximized utility in period 4 is

* 2* * 2*

4 4 4 4[1 ( )] ( ) ( )V p G U X p G M

In period 3, the leader’s optimization problem is

1 23 3 3

2 * 2

3 3 4 3, ,

max [1 ( )] ( )+ ( )X G G

p G U X V p G M

subject to

1 2

3 3 3 3( ) ( ) /G G X T N (4)

where 3X denotes the consumption of the leader in period 3 and 0 1 is the

discount factor.

We can obtain the first-order conditions as follows:

1

3'( ) 1/ ( 1)G N (5)

and

2 *

2 3 3 43 2

3 3

'( )[ ( ) ]( 1) '( ) 1

[1 ( )] '( )

Np G U X V MN G

p G U X

. (6)

Let the optimal public expenditures be 1*

3G and 2*

3G , and let the optimal

consumption level be *

3X . Then the maximized utility from period 3 is

* 2* * * 2*

3 3 3 4 3[1 ( )][ ( )+ ] ( )V p G U X V p G M .

If the leader does not remain in office in periods 3 and 4 and if there is no career

failure, he simply receives a reservation utility 0U as for ordinary citizens, for

periods 3 and 4.

Now we turn to the first term (periods 1 and 2). The leader’s optimization

problem in period 2 can be written as

1 22 2 2

2 2 1 * 2 1 2

2 2 1 2 3 1 2 0 0, ,

2

2

max [1 ( )] ( )+ ( , ) +[1 ( , )][ ]

( )

X G G

p G U X f G G V f G G U U

p G M

subject to

1 2

2 2 2 2( ) ( ) /G G X T N . (7)

The optimal conditions can be derived as

2 1 *

1 2 1 2 3 02

2

( , ) [ ( 1)( )]( 1) '( ) 1

'( )

Nf G G V UN G

U X

(8)

and

2 2 1 *

2 2 0 1 2 3 02

2 2

2 2

'( ) ( ) [( 1) ( , )( ( 1) )]( 1) '( ) 1

[1 ( )] '( )

Np G U X M U f G G V UN G

p G U X

(9)

Assumption 3. *

3 0( 1)V U .

Assumption 3 requires that the maximized sum of discounted utility derived

from remaining in office in periods 3 and 4 is larger than the reservation utility level

of an ordinary citizen. This assumption is realistic in China: most government leaders

have more political power, which can bring them more benefits, and most of them

have few outside career options (Li and Zhou, 2005). Under this assumption, the right

hand of equation (8) is negative, and thus we have 1

2'( ) 1/( 1)G N . Therefore,

from equations (8) and (2), we have 1* 1*

2 4G G .

In period 1, the leader’s optimization problem can be written as:

1 2

1 1 1

2 * 2 2

1 1 2 1 1, ,

max [1 ( )]{ ( )+ ( )} ( )X G G

p G U X V G p G M

Subject to

1 2

1 1 1 1( ) ( ) /G G X T N . (10)

Note that from equations (8) and (9), the optimal solutions in period 2 ( 1*

2G , 2*

2G ,

and *

2X ) depend on 2

1G , as does the maximized utility in period 2, *

2V , because the

risky expenditure in period 1 influences the economic performance in period 2 and in

turn affects the probability of remaining in office. The optimal condition for safe

expenditure 1

1G in period 1 is

1

1'( ) 1/( 1)G N . (11)

From equations (2), (5), (8), and (11), we have 1* 1* 1* 1*

2 4 3 1G G G G . Thus we

have the following proposition.

Proposition 1. The safe expenditure is higher during period 2 when the Central

Committee of the Party evaluates provincial leaders and makes decisions on their

turnovers, compared to the level in the other three periods.

The intuition of the above proposition is clear. Increasing safe expenditure in the

evaluation period can increase the chance of remaining in office for a second term.

Now we turn to risky expenditure. The optimal condition for risky expenditure 2

1G , in

period 1 is

* 22 * 2 2 2 1

1 1 2 1 1 22 1

1 2

1 1

( )'( )[ ( ) ( )] [1 ( )]

( 1) '( ) 1[1 ( )] '( )

V GNp G U X M V G N p G

GN G

p G U X

. (12)

We cannot compare the risky expenditure in different periods without assuming

specific functional forms. Next, we use a numerical example to illustrate how the

risky expenditure varies in each period.

5.2. A Numerical Example of the Model

In this subsection, we construct a numerical example to illustrate the dynamic

pattern of safe and risky expenditures over a provincial leader’s life cycle. We specify

the leader’s utility function as:

( ) logU X X

The income functions of the leader is specified as: 1 1( ) logG G and

2 2( ) logG G . The probability of career failure in each period is a function of the

risky expenditure in this period 2

2

2( )

1

GP G

G

. At the end of the evaluation period,

i.e., at the end of period 2, conditional on no career failure, the probability of a leader

remaining in office for a second term is 2 1

2 1 1 21 2 2 1

1 2

( , )1

aG bGf G G

aG bG

, where 0a

and 0b are constants. One can verify that these specific functional forms all

satisfy the requirements in the basic setup of our model.

Next, we conduct numerical simulations to demonstrate the dynamic pattern of

fiscal expenditures over a provincial leader’s life cycle. We try different sets of

parameters and they all show similar patterns. In Figure 6, we show both the safe

expenditure and the risky expenditure over different periods for the following set of

parameters: 90N , 0M , 0.9 , 1a , 1b , and 0 =0.1U .

From Figure 6, we can see that safe expenditure is higher during the evaluation

period (NCCP) than other periods (the non-NCCP periods), as Proposition 1 predicted.

Risky expenditure shows the opposite pattern: It is lower during the evaluation period

than other periods.

The intuition of the model is clear. Safe expenditure during the NCCP period

may increase a provincial leader’s probability of remaining in office. Therefore, the

leader tends to spend more on safe expenditure during the NCCP, even though it does

not boost the economy. Risky expenditure has a lagged positive effect on the local

economy while increasing the career risk contemporaneously. Therefore, leaders tend

to reduce risky expenditure during the NCCP period. Risky expenditure is higher in

period 1 than in the NCCP period and one period afterwards because investing in

infrastructure earlier will boost economic growth later, thus increasing the chance of

being given renewal of tenure. Risky expenditure is also higher in the last period of

the life cycle because the opportunity cost of a career failure in the last period is lower

than in previous periods.

Our model and the empirical evidence in Section 4 provide a channel through

which the incentive roles affect the dynamic pattern of provincial fiscal expenditures.

In the following empirical analysis, we examine whether political budget cycles exist

at the provincial level in China and check whether the findings are consistent with our

model’s predictions.

6. Political Budget Cycles

6.1. Baseline Model

The baseline model we use to analyze how provincial public expenditures

synchronize with the NCCP is as follows:

0 1 2 3

4 5 1

1 + 1

+ 2

it t t t

t it it t i it

g NCCPpre NCCP NCCPpost

NCCPpost g X Z v u

(13)

The dependent variables itg is the growth rate of fiscal expenditure (including total

fiscal expenditure, administration spending, infrastructure expenditure, agricultural

expenditure, and ESMC expenditure), in province i in period t . Due to the

persistent nature of fiscal expenditures, we include the lagged variable 1itg as a

control, and four dummy variables, 1tNCCPpre , tNCCP , 1tNCCPpost , and

2tNCCPpost are used to capture the cyclical effect of the NCCP, which takes place

every five years. 1tNCCPpre equals 1 if one years before the NCCP and 0 otherwise;

tNCCP equals 1 if the NCCP takes place this year and 0 otherwise; 1tNCCPpost

equals 1 if one year after NCCP and 0 otherwise; 2tNCCPpost equals 1 if two years

after NCCP and 0 otherwise.

itX is a vector of control variables that affect the growth rate of public

expenditures and itu is the error term. itX includes both the provincial economic

variables and provincial leaders’ characteristics. The economic variables are the

provincial fiscal revenue growth rate, provincial fiscal expenditure growth rate

(controlled in regressions of disaggregate expenditures), population growth rate, GDP

per capita growth rate, urban income growth rate and the share of secondary industry.

The provincial leaders’ characteristics include age and educational level for both

provincial party secretary (number 1 leader) and provincial governor (number 2

leader). Educational level is a dummy variable in which 1 indicates college degree or

above and 0 otherwise.

The estimated cycle might be confounded by common shocks - which by

coincidence - may coincide with the NCCP cycle. Therefore we also include a vector

2, ,NGDP

t tZ g T T to control for aggregate shocks at the national level and the time

trend. Finally, the fixed effect of province is controlled for in all of the regressions.

Before running the regressions, we first test whether our panel data series are

stationary. As some of our variables are not balanced panel data, we use both the

Im-Pesaran-Shin and Fisher-ADF test to check the existence of a unit root. We include

both time trend and panel means. Table 4 reports the results, showing that all of the

variable series strongly reject the null hypothesis of a unit root.

Since the Government Revenue and Expenditure Classification Reform was

carried out in 2007, after which disaggregate expenditures are calibrated quite

differently, we use data from 1990 to 2006 to identify political budget cycles in fiscal

expenditures.

6.2. Results: The Effect of NCCP

Table 13 shows the effect of NCCP on the growth rates of both aggregate and

disaggregate expenditures. The coefficients of interest are those for the

1tNCCPpre , tNCCP , 1tNCCPpost , and 2tNCCPpost dummies, which capture

changes in the growth rate of public expenditures over the NCCP cycle. Column (1) in

Table 13 shows that on average the total expenditure growth decreases by more than

5% during the NCCP year, but increases by more than 4% one year before the NCCP.

To better understand what drives down the total expenditure growth, we next

investigate different expenditure categories (Column (2)-(5) in Table 13). The results

show that the growth rate of administration spending increases by 3.69% during the

NCCP year, significant at the 5% level, and also increases by 3.44% one year before

the NCCP, significant at 10% level The growth rate of ESMC expenditure shows a

similar pattern to that of administration expenditure, increasing by 1.85% during the

NCCP year at the significance level of 5%.

In contrast, the growth rate of infrastructure expenditure decreases by 7.31%

during the NCCP at the 5% significance level. In addition, one year before the NCCP

the growth rate of infrastructure expenditure decreases by 8.95% at the 1%

significance level. Since these coefficients are compared with the base year—two

years before NCCP, the results actually suggest that the growth rate of infrastructure

expenditure tends to increase two years before the NCCP. The growth rate of

agricultural expenditure increase significantly two years after the NCCP year.

To summarize, the empirical results of the baseline model indicates the existence

of political budget cycles in China that coincide with the timing of the NCCP.

Specifically, the growth rate of risky expenditures such as infrastructure declines

during the NCCP, while the growth rate of safe expenditures (such as administration

and ESMC increases during NCCP years, consistent with our model’s predictions.

Robustness Checks

GMM approach

The baseline model may suffer from potential bias due to the correlation between

lagged dependent variables and the error term. To assess robustness, we apply system

GMM approach9 and the results are reported in Table 14. We use lag two of the

dependent variable as the instrumental variable in the GMM estimation10

. Note that

all the regressions in Table 15 include a time trend and provincial fixed effect and we

specify the same control variables as the baseline model11

.

The results in Table 15 are fairly similar to those of the baseline model. The

coefficient of tNCCP for total expenditure growth rate is negative at significant at

level 10%. Coefficients of tNCCP for administration spending and infrastructure

expenditure growth rate remain the same sign. The coefficient of tNCCP for

agriculture becomes positive with a significance level of 1%, and that for ESMC

remains positive but becomes insignificant. This results support the earlier prediction

that public expenditure switch from infrastructure to other types of expenditure during

NCCP.

Some tests are reported in table 14 too. Sargan test reports the over-identifying

restrictions validity. The statistics show that we cannot reject the null hypothesis of

validity of the instruments. The second-order serial correlation test statistics show that

9 System GMM is used because it uses additional moment condition (level equations).

10 Using longer lags of the dependent variable as the instruments does not change the results.

11 We do not include the time-square trend in Table 9 because it causes collinearity in the GMM estimation.

there is no serial correlation in error terms. We also perform the Wald test of the null

hypothesis that all coefficients together are equal to zero. Wald test results show the

joint significance of all of the coefficients.

First Difference Approach

Tsai’s(2013) study uses first difference method to identify the effect of NCCP

from 1980 to 2006. Although we use shorter panel due to the limitation of corruption

data12

, it is important to check whether our results are robust to different approached

applied. We use first difference for all the variables and apply the same specifications

in Tsai’s paper: control lagged revenue in the regression of total expenditure and

control total revenue and total expenditure in the regressions of disaggregate

expenditures. Other control variables include: population, GDP per capita, urban

income, secondary industry ratio and officials’ characteristics such as age and

education. We still include national GDP growth rate to control for common shocks in

macro economy. To address the correlation between lagged dependent variable on

RHS, we apply system GMM approach as in table 14.

The results and test statistics are reported in table 15. Coefficients of NCCP

dummies remain the same signs, although with slightly different significance level,

implying that what we find earlier is robust when we use different measurements.

7. Time-Inconsistency

7.1. Fluctuations of Public Expenditures

In this section, we ask a slightly different question: in China, does a higher

frequency of turnovers of provincial leaders generate more fluctuations in fiscal

expenditures? The recent literature suggests that in democratic countries, more

frequent turnovers of partisan political control lead to more fluctuations in fiscal

decisions due to different preferences across parties. This is called time-inconsistency.

China is a single-party country where there is no turnover of ruling parties. However,

the turnovers of political leaders at the provincial level are not infrequent compared to

democratic countries. We observe turnovers not only during the NCCP year or one

year after; according to Table 1, around 50% percent of the provincial leaders’ turnovers

between 1990 and 2010 occurred in non-NCCP years. Because different provincial

leaders may have different political preferences or belong to different political factions,

time inconsistency may influence the fluctuations of provincial fiscal policies.

Following Crain and Tollison (1993), we examine whether the time-inconsistency

theory applies in China. If there is any strategic use of fiscal tools when a leader is

12

We try to keep the panel in section 4 and section 6 as consistent as possible.

likely to leave his position, we should observe a positive correlation between the

fluctuations in expenditure growth and the frequency of turnovers of leaders. We use

cross- sectional (province) data to conduct our econometric analysis using the

specification

0 1 2vari ps gov i ifreq freq W , (14)

where vari represents the variance in fiscal expenditure growth over time for

province i. We conduct the exercises for different expenditure categories, including

total expenditure, administration spending, infrastructure expenditure, agriculture

expenditure and ESMC expenditure. psfreq and govfreq represent the frequency of

provincial secretary turnovers and the frequency of provincial governor turnovers,

respectively, over 1995-200613

. iW is a vector of control variables, including the

variance of provincial population growth rate and the variance of provincial GDP

growth rate. The coefficients of interest are 1 and

2 , which capture the effects of

provincial secretary turnovers and provincial governor turnovers on the fluctuations of

public expenditures.

7.2. Results

Table 16 reports the estimation results of Equation (14). The coefficients of the

frequencies of provincial secretary and governor turnovers are all insignificant,

suggesting that the frequency of provincial leaders’ turnovers has little effect on the

volatility of any fiscal expenditure. This result lends support to the notion that even

though different provincial leaders may have different preferences or belong to different

political factions, the personnel control system of the Central Committee of the

Communist Party has adopted some unified evaluation criteria that give provincial

leaders incentives to implement the ruling Party’s ideas.

The coefficient of the variance of provincial GDP growth in Column (1) of Table

16 is positive and significant at the 5% level as expected, indicating that the total

expenditure growth experiences more volatility when the fluctuation of GDP growth

increases.

7.3. Robustness Check

The empirical model in Section 7.1 only takes the frequency of provincial leaders’

turnovers into account. However, frequency alone may not be enough to describe the

characteristics of turnovers. For example, suppose that two provinces experience one

provincial secretary turnover during our 10 years. In one province, the tenure of the

first secretary is 5 years and the tenure of the second is 5 years. In the other province,

13

We use data after 1995 due to the Tax Sharing Reform in 1994.

the tenure of the first secretary is 1 year and the tenure of the second is 9 years. Such a

variation in tenure length might cause variation in the fiscal policies across the two

provinces. Therefore, we add the standard deviation of provincial leaders’ tenure length

(for both the secretary and the governor) in our regression. The results are reported in

Table 17. The coefficient of secretary turnover frequency in total expenditure becomes

significant at 10% level, but the coefficient of governor turnover frequency is still

insignificant for all expenditure categories. Although the results are not very robust,

they are implying that provinces with more frequent secretary turnovers experience

more volatile total fiscal expenditures. However, the coefficient of the standard

deviation of provincial governors’ tenure are positive and significant at the 10% level

for total fiscal expenditure, which suggests that at the same turnover frequency, those

provinces with more irregular governors’ tenures tend to experience larger fluctuations

in total expenditure growth.

8. Conclusions

This paper first finds that in addition to good economic performance, being clean

(or staying away from corruption) is an important factor affecting provincial leaders’

chances of promotion. Given these promotion incentives, this paper develops a life

cycle model of provincial leaders to demonstrate the decision-making process

regarding various fiscal expenditures given promotion incentives. In the model, risky

expenditure, such as infrastructure spending, can boost the economy, while at the

same time increasing the career risk of a government leader because such projects

easily foster corruption. Meanwhile, safe expenditure such as administration spending

can improve the chance of promotion.

Before the evaluation and promotion year, provincial leaders have great

incentives to expand infrastructure expenditure to boost economic growth and impress

their superiors. In the year of evaluation and promotion, however, provincial leaders

adopt a more conservative strategy of “no deeds, no mistakes”; specifically, they

reduce risky expenditures such as infrastructure and increase safe expenditures such as

administration spending.

The paper then tests the model predictions using Chinese provincial level data.

The results show that political budget cycles exist in China. More specifically, the

growth of administration spending increases significantly during the NCCP year, as

does education expenditure. In contrast, the growth rate of infrastructure expenditure

decreases during the NCCP year.

This paper also tests another possible channel of political budget cycles; that is,

the time inconsistency effect caused by provincial leaders’ turnovers. It turns out to be

insignificant for almost all types of expenditures. This finding indicates that China’s

centralized personnel control of provincial leaders, accompanied by the current

incentive structure, seems to minimize the time inconsistency of local fiscal policies,

which is not uncommon in democratic countries.

Reference:

Alesina, A. and G. Tabellini, 1990, “A Positive Theory of Fiscal Deficits and

Government Debt”, The Review of Economic Studies 57(3), 403-414.

Benhabib, J. and A. Rustichini, 1997, “Optimal Taxes without Commitment”,

Journal of Economic Theory 77(2), 231-259.

Blanchard, O. and Shleifer, A., 2000, “Federalism with and without political

centralization: China versus Russia” (No. w7616), National Bureau of Economic

Research.

Bunce, V. J., 1980, “The Succession Connection: Policy Cycles and Political

Change in the Soviet Union and Eastern Europe”, The American Political Science

Review 74(4), 966-977.

Chari, V. V. and P. J. Kehoe, 1990, “Sustainable Plans”, Journal of Political

Economy 98(4), 783-802.

Chen, Ye, Hongbin Li, and Li-An Zhou, 2005, “Relative Performance

Evaluation and the Turnover of Provincial Leaders in China”, Economic Letters 88

(3): 421–25.

Crain, W. M. and R. D. Tollison, 1993, “Time Inconsistency and Fiscal Policy:

Empirical Analysis of US States, 1969-89”, Journal of Public Economics 51(2),

153-159.

Czernich, Nina, et al., 2011, “Broadband infrastructure and economic

growth”, The Economic Journal , 121 (552), 505-532.

De Haan, J. and J. E. Sturm, 1997, “Political and Economic Determinants of

OECD Budget Deficits and Government Expenditures: A Reinvestigation”, European

Journal of Political Economy 13(4), 739-750.

De Haan, J. and J. E. Sturm, 1994, “Political and Institutional Determinants of

Fiscal Policy in the European Community”, Public Choice 80(1), 157-172.

Demurger, Sylvie, 2001, “Infrastructure development and economic growth: an

explanation for regional disparities in China?”, Journal of Comparative Economics 29

(1), 95-117.

Dittmer, L. ,1995, “Chinese informal politics”, The China Journal, 34, 1-34.

Drazen, A., 2001, “The Political Business Cycle after 25 Years”, NBER

Macroeconomics Annual 15, 75-138.

Edin, Maria, 2003, “State Capacity and Local Agent Control in China: CCP

Cadre Management from a Township Perspective”, China Quarterly 173, 35–52.

Fischer, S., 1980, “Dynamic Inconsistency, Cooperation and the Benevolent

Dissembling Government”, Journal of Economic Dynamics and Control 2(1), 93-107.

Gonzalez, M. d. l. A., 2002, “Do Changes in Democracy Affect the Political

Budget Cycle? Evidence from Mexico”, Review of Development Economics 6(2),

204-224.

Grilli, V., D. Masciandaro, G. Tabellini, E. Malinvaud, and M. Pagano, 1991,

“Political and Monetary Institutions and Public Financial Policies in the Industrial

Countries”, Economic Policy 342-392.

Guo, G., 2009, “China’s Local Political Budget Cycles”, American Journal of

Political Science 53(3), 621-632.

Katsimi, M. and V. Sarantides, 2012, “Do Elections Affect the Composition of

Fiscal Policy in Developed, Established Democracies?” Public Choice 151(1-2),

325-362.

Kenny, C., 2006, Measuring and Reducing the Impact of Corruption in

Infrastructure, World Bank.

Kneebone, R. D. and K. J. McKenzie, 2001, “Electoral and Partisan Cycles in

Fiscal Policy: An Examination of Canadian Provinces”, International Tax and Public

Finance 8(5-6), 753-774.

Kydland, F. E. and E. C. Prescott, 1977, “Rules Rather than Discretion: The

Inconsistency of Optimal Plans”, Journal of Political Economy 85(3), 473-491.

Li, H. and L.-A. Zhou, 2005, “Political Turnover and Economic Performance:

the Incentive Role of Personnel Control in China”, Journal of Public Economics

89(9-10), 1743-1762.

Lucas, R. E. and N. L. Stokey, 1983, “Optimal Fiscal and Monetary Policy in an

Economy without Capital”, Journal of Monetary Economics 12(1), 55-93.

Maskin, Eric, Yingyi Qian, and Chenggang Xu, 2000, “Incentives, Information,

and Organizational Form”, Review of Economic Studies 67 (2), 359-78.

Montinola, G., Qian, Y. and Weingast, B. R., 1995, “Federalism, Chinese style:

the political basis for economic success in China”, World Politics, 48(01), 50-81.

Nathan, Andrew J., and Kellee S. Tsai, 1995, “Factionalism: a new

institutionalist restatement”, The China Journal, 34, 157-192.

Persson, T. and L. E. O. Svensson, 1989, “Why a Stubborn Conservative would

Run a Deficit: Policy with Time-inconsistent Preferences”, The Quarterly Journal of

Economics 104(2), 325-345.

Pettersson Lidbom, P., 2001, “An Empirical Investigation of the Strategic Use of

Debt”, Journal of Political Economy 109(3), 570-583.

Pye, L. W., 1992, The spirit of Chinese politics, Harvard University Press.

Rogoff, K., 1990, “Equilibrium Political Budget Cycles”, American Economic

Review 80(1), 21-36.

Schuknecht, L., 2000, “Fiscal Policy Cycles and Public Expenditure in

Developing Countries”, Public Choice 102(1), 113-128.

Shih, Victor, Christopher Adolph and Mingxing Liu, 2012, “Getting Ahead in

the Communist Party: Explaining the Advancement of Central Committee Members

in China”, American Political Science Review 106, 166-186.

Tanzi, V. and H. Davoodi, 1998, “Roads to Nowhere: How Corruption in Public

Investment Hurts Growth”, Economic Issues. International Monetary Fund.

Tsai, Pi-Han Christine, 2013, “Fiscal Incentives and Political Budget Cycles in

China”, working paper.

Vergne, C., 2009, “Democracy, Elections and Allocation of Public Expenditures

in Developing Countries”, European Journal of Political Economy 25(1), 63-77.

Zhong, Yang. 2003. Local Government and Politics in China: Challenges from

Below. Armonk, NY: M. E. Sharpe.

Wu, J., Y. Deng, J. Huang, R. Morck, and B. Yeung, 2013, “Incentives and

Outcomes: China’s Environmental Policy”, National Bureau of Economic Research

No. 18754.

Figure 1 Corruption Scandals and Infrastructure Expenditure Growth

Note: Figure 1 shows the frequency of corruption scandals and average infrastructure expenditure

growth (across provinces) each year. Data for corruption scandals is from 1990 to 2010. Data for

infrastructure expenditure growth is from 1991 to 2006. Vertical red lines represent the timing of

NCCP.

Figure 2. Correlation of Corruption, Accidents, and Infrastructure Expenditures

Note: Corruption, accidents, and infrastructure expenditures are average value across provinces each

year.

Figure 3. Growth Rate of Different Categories and NCCP

Note: Vertical lines refer to the year of NCCP. Dot represents the average growth rate among provinces

and capped spikes refer to the standard deviation of growth rate.

Figure 4 Promotion and Termination Distributions over Provinces

Figure 5 Promotion and Termination Distributions over factions

Figure 6. Simulations of Safe Expenditures and Risky Expenditures

Note: We set 90N , 0M , 0.9 , 0 =0.1U , 1a , and 1b in the above graphs. We have

also tried different values of a and b , but the variation trend doesn’t change much.

Table 1. Frequency of Provincial Official Turnovers

year Secretary Turnover Frequency Governor Turnover Frequency

1990 4 8

1991 3 2

1992 2 6

1993 10 9

1994 6 7

1995 5 5

1996 1 5

1997 11 3

1998 9 14

1999 4 4

2000 4 5

2001 8 7

2002 10 8

2003 4 13

2004 4 6

2005 4 1

2006 5 8

2007 14 12

2008 2 4

2009 5 2

2010 8 8

Note: NCCP years are 1992, 1997, 2002 and 2007.

Table 2. Descriptive Statistics of Variables

Variables Observations Mean Std.Dev. Min Max

Official Variables

Promotion 588 0.0357 0.186 0 1

Termination 588 0.0986 0.298 0 1

Secretary Age 588 60.04 4.069 47 70

Secretary Education 588 0.718 0.451 0 1

Secretary Tenure 588 3.969 2.547 0 16

Minority 588 0.0357 0.186 0 1

Central Experience 588 0.267 0.443 0 1

Strategic Shift 588 0.565 0.496 0 1

Faction 588 0.539 0.856 0 3

Corruption and Accidents

Corruption 588 0.379 0.905 0 7

Accidents 588 0.413 0.907 0 6

Economic Variables

Growth_Population 560 0.0106 0.0158 -0.0555 0.190

Growth_ProvincialGDP 560 0.128 0.0495 -0.0459 0.387

Growth_NationalGDP 560 0.105 0.0206 0.0762 0.142

Growth_ProvincialUrbanIncome 560 0.0936 0.0416 -0.0188 0.281

Secondary Industry Share 560 0.443 0.0794 0.197 0.617

Provincial per capita GDP (yuan) 560 6865 6985 849.7 41882

Fiscal Variables

Growth_Total Revenue 448 0.119 0.181 -0.677 1.233

Growth_Total Expenditure 448 0.140 0.109 -0.168 0.538

Growth_Administration 448 0.144 0.132 -0.506 1.221

Growth_Infrastructure 448 0.183 0.332 -0.394 2.066

Growth_Agriculture 448 0.135 0.310 -0.831 3.562

Growth_Education 448 0.126 0.0899 -0.216 0.487

Fiscal Support Population 392 11.95 6.491 1.407 29.59

Note: For official variables, Promotion is an indicator variable that equals one if a provincial leader is

promoted and zero otherwise. Termination equals 1 if termination occurs and 0 otherwise. Minority

equals 1 if the provincial leader is any ethnicity other than Han. Central Experience equals 1 if the

provincial leader has work experience in central government. Strategic Shift equals 1 if the provincial

leader comes from other provinces or institutions. Faction equals 1 if the leader belongs to Shanghai

faction, 2 for Tuan faction, 3 for Taizi faction, 0 for none faction. The fiscal data ranges from 1990 to

2006 because of the Government Revenue and Expenditure Classification Reform in 2007.

Table 3. Disaggregate Expenditure Ratios (1990-2006) (%)

Expenditure Category Mean Std.Dev. Min Max

Administration Expenditure Ratio 10.98 4.239 1.999 30.89

Infrastructure Expenditure Ratio 9.705 4.621 3.136 29.51

Agriculture Expenditure Ratio 7.242 2.958 1.196 15.48

ESMC Expenditure Ratio 24.69 3.681 13.89 36.36

Note: ESMC Expenditure includes education, science, medical and culture expenditures.

Table 4. Unit Root Test

Im-Pesaran-Shin test Fisher-ADF test

Variables p-value Time

Trend

Panel

Means p-value

Time

Trend

Panel

Means

Total Expenditure

Growth 0.0000 Yes Yes 0.0000 Yes Yes

Administration

Expenditure Growth 0.0000 Yes Yes 0.0000 Yes Yes

Infrastructure

Expenditure Growth 0.0000 Yes Yes 0.0000 Yes Yes

Agricultural Expenditure

Growth 0.0000 Yes Yes 0.0000 Yes Yes

ESMC Expenditure

Growth 0.0000 Yes Yes 0.0000 Yes Yes

Note: Im-Pesaran-Shin test and Fisher-ADF test are used to test unit root in panel data, including time

trend and panel means. We reject the null hypothesis of containing unit root and conclude that the

series is stationary.

Table 5. Frequency of Provincial Officials’ Turnover from 1990 to 2010

Secretary

Turnover

Frequency

Number

of

Provinces

Percent Cumulative

Frequency

Governor

Turnover

Frequency

Number

of

Provinces

Percent Cumulative

Frequency

2 1 3.570 3.570 2 1 3.570 3.570

3 6 21.43 25 3 4 14.29 17.86

4 7 25 50 4 5 17.86 35.71

5 11 39.29 89.29 5 7 25 60.71

6 1 3.570 92.86 6 9 32.14 92.86

7 2 7.140 100 7 2 7.140 100

Total 28 100 Total 28 100

Note: There are 123 Secretary Turnovers and 137 Governor Turnovers in total from 1990 to 2010.

Table 6. Corruption and Accidents Summary by Year

Summary by year Corruption Accidents

1990 1 3

1991 0 1

1992 2 2

1993 2 4

1994 1 1

1995 2 1

1996 0 1

1997 1 1

1998 1 5

1999 1 1

2000 3 8

2001 3 10

2002 2 10

2003 2 11

2004 18 28

2005 53 34

2006 32 28

2007 19 24

2008 23 23

2009 34 21

2010 23 26

Total 223 243

Table 7. Corruption and Accidents Summary by Province

Summary by province Corruption Accidents

Anhui 22 8

Beijing 18 2

Fujian 5 4

Gansu 7 3

Guangdong 6 6

Guangxi 12 5

Guizhou 4 8

Hainan 13 2

Hebei 7 22

Heilongjiang 5 10

Henan 17 20

Hubei 5 3

Hunan 14 18

Jiangsu 2 6

Jiangxi 9 10

Jilin 11 5

Liaoning 6 9

Neimenggu 3 10

Ningxia 4 4

Qinghai 3 3

Shandong 7 10

Shanghai 7 3

Shanxi 13 33

Shaanxi 7 15

Tianjin 2 1

Xinjiang 3 8

Yunnan 7 8

Zhejiang 4 7

Total 223 243

Table 8 Xinhua multimedia database VS United Daily News data

province year Corruption scandal count from

Xinhua multimedia database

Corruption scandal count

from United Daily News

Anhui 2005 7 3

Anhui 2006 6 4

Anhui 2008 2 1

Guangdong 2005 1 0

Guangdong 2007 1 0

Guangdong 2008 1 0

Guangdong 2009 0 1

Guangdong 2010 1 0

Jilin 2004 1 0

Jilin 2005 1 0

Jilin 2006 2 0

Jilin 2007 1 0

Jilin 2008 3 1

Jilin 2009 2 1

Jilin 2010 1 0

Shaanxi 1993 1 0

Shaanxi 1997 1 0

Shaanxi 2000 0 1

Shaanxi 2001 0 1

Shaanxi 2004 1 0

Shaanxi 2005 1 0

Shaanxi 2007 2 1

Shaanxi 2010 1 0

Note: Due to space constraints, the above table shows scandal data comparison in four representative

provinces: Anhui, Guangdong, Jilin and Shaanxi. They differ in both location and economic

development level. We only report the province-year data when the two news media have different

counts. We use the same keywords in searching for corruption scandals in the two media.

Table 9. Correlation of Corruption and Accidents

corruption Lagged

corruption

Lagged2

corruption

accidents Lagged

accidents

Lagged2

accidents

corruption 1

L1corruption 0.401 1

L2corruption 0.304 0.444 1

accidents 0.285 0.285 0.142 1

L1 accidents 0.318 0.273 0.278 0.439 1

L2 accidents 0.305 0.322 0.274 0.383 0.459 1

Table 10. Effects of Corruption and Accidents on Officials’ Turnover

(Ordered Probit Model)

Dependent variable: turnover (1) (2) (3) (4)

(0=termination,1=same level, 2=promotion)

Corruption -0.206** -0.225** -0.224** -0.206**

(0.0869) (0.0883) (0.0922) (0.0842)

Lagged corruption 0.0733 0.0708 0.0557

(0.0872) (0.0888) (0.0837)

Lagged2 corruption 0.00966 0.0149 0.0431

(0.102) (0.105) (0.107)

Accidents 0.122

(0.0927)

Lagged accidents -0.146**

(0.0687)

Lagged2 accidents -0.144

(0.0907)

Growth_Provincial GDP 2.094 2.013

(3.320) (3.459)

Lagged Growth_Provincial GDP -5.046 -4.613

(4.336) (4.437)

Lagged2 Growth_Provincial GDP 5.279* 5.089

(3.171) (3.356)

Lagged GDP per capita -1.48e-05 -5.57e-06 -5.23e-06 -1.48e-05

(3.81e-05) (3.57e-05) (3.74e-05) (3.92e-05)

Secretary Age -0.129*** -0.127*** -0.127*** -0.120***

(0.0249) (0.0238) (0.0234) (0.0231)

Secretary Education -0.151 -0.0884 -0.0837 -0.0348

(0.376) (0.290) (0.288) (0.303)

Secretary Tenure 0.00955 0.0222 0.0194 0.0142

(0.0469) (0.0493) (0.0488) (0.0502)

Minority -0.600 -0.632 -0.635 -0.630

(0.424) (0.441) (0.446) (0.446)

Central Experience -0.00644 -0.233 -0.230 -0.212

(0.378) (0.329) (0.329) (0.346)

Strategic Shift 0.0807 -0.0354 -0.0199 -0.0324

(0.189) (0.192) (0.209) (0.215)

Central Experience& Strategic Shift 0.183 0.352 0.343 0.318

(0.394) (0.364) (0.379) (0.389)

faction_shanghai 0.176 0.325 0.331 0.376*

(0.227) (0.199) (0.205) (0.198)

faction_tuan -0.350* -0.186 -0.183 -0.165

(0.189) (0.179) (0.184) (0.190)

faction_taizi -0.0392 0.192 0.231 0.384

(0.420) (0.467) (0.472) (0.414)

cut1 -9.406*** -9.036*** -8.747*** -8.472***

(1.448) (1.361) (1.530) (1.530)

cut2 -5.352*** -4.892*** -4.573*** -4.232***

(1.282) (1.174) (1.356) (1.346)

Observations 560 532 532 532

Note: Minority, Central Experience, Strategic Shift are all dummy variables. Central Experience equals

1 if the provincial leader has work experience in central government. Strategic Shift equals 1 if the

provincial leader comes from other provinces or institutions. *, **, and *** indicate significance at the

10%, 5%, and 1% levels, respectively. Robust standard errors are in parentheses. Year and province

dummies are included.

Table 11. Effects of Corruption and Accidents on Officials’ Turnover

(Adjusted corruption and accidents data)

Dependent variable: turnover (1) (2) (3) (4)

(0=termination,1=same level, 2=promotion) 1.2r 1.2r

Corruption 1-12 months before -0.164** -0.174** -0.182** -0.159**

(0.0825) (0.0796) (0.0867) (0.0763)

Corruption 13-24 months before 0.0614 0.0720 0.0370

(0.0745) (0.0778) (0.0678)

Accidents 1-12 months before 0.138

(0.0903)

Accidents 13-24 months before -0.231***

(0.0615)

Growth_Provincial GDP 2.348 2.305

(3.289) (3.434)

Lagged Growth_Provincial GDP -5.200 -4.760

(4.350) (4.399)

Lagged2 Growth_Provincial GDP 5.173 4.894

(3.183) (3.346)

Lagged GDP per capita -1.46e-05 -1.41e-05 -6.03e-06 -1.41e-05

(3.84e-05) (3.86e-05) (3.82e-05) (3.97e-05)

Secretary Age -0.129*** -0.128*** -0.125*** -0.118***

(0.0247) (0.0243) (0.0230) (0.0223)

Secretary Education -0.150 -0.148 -0.0742 -0.0318

(0.373) (0.372) (0.284) (0.297)

Secretary Tenure 0.00842 0.0101 0.0173 0.0153

(0.0468) (0.0468) (0.0491) (0.0499)

Minority -0.601 -0.585 -0.640 -0.628

(0.427) (0.419) (0.445) (0.454)

Central Experience 0.0193 0.0163 -0.178 -0.192

(0.363) (0.362) (0.331) (0.357)

Strategic Shift 0.0786 0.0728 -0.0202 -0.0227

(0.189) (0.190) (0.208) (0.215)

Central Experience& Strategic Shift 0.158 0.151 0.293 0.288

(0.380) (0.379) (0.366) (0.387)

faction_shanghai 0.186 0.188 0.342* 0.395**

(0.226) (0.222) (0.200) (0.197)

faction_tuan -0.339* -0.333* -0.174 -0.182

(0.187) (0.186) (0.185) (0.183)

faction_taizi -0.0262 -0.00345 0.238 0.457

(0.418) (0.413) (0.471) (0.398)

cut1 -9.351*** -9.240*** -8.651*** -8.315***

(1.435) (1.373) (1.501) (1.477)

cut2 -5.309*** -5.197*** -4.496*** -4.088***

(1.268) (1.205) (1.328) (1.284)

Observations 532 560 532 532

Note: Minority, Central Experience, Strategic Shift are all dummy variables. Central Experience equals

1 if the provincial leader has work experience in central government. Strategic Shift equals 1 if the

provincial leader comes from other provinces or institutions. *, **, and *** indicate significance at the

10%, 5%, and 1% levels, respectively. Standard errors are in parentheses. Year and province dummies

are included.

Table 12. Effects of Administration Spending on Officials’ Turnover

Dependent variable: turnover (1) (2)

(0=termination,1=same level, 2=promotion) Ordered Probit IV

Log(Administration) 0.517 1.716***

(0.483) (0.649)

Corruption -0.221** -0.246***

(0.0878) (0.0920)

Lagged corruption 0.0542 0.0436

(0.0855) (0.0869)

Lagged2 corruption 0.0283 -0.00695

(0.0983) (0.0907)

Accidents 0.123 0.129

(0.0916) (0.0879)

Lagged accidents -0.150** -0.145**

(0.0701) (0.0695)

Lagged2 accidents -0.155* -0.186**

(0.0903) (0.0934)

Growth_Provincial GDP 1.599 0.738

(3.432) (3.403)

Lagged Growth_Provincial GDP -4.793 -5.205

(4.470) (4.382)

Lagged2 Growth_Provincial GDP 4.495 2.919

(3.123) (2.935)

Lagged GDP per capita -3.21e-05 -7.10e-05

(4.34e-05) (5.42e-05)

Secretary Age -0.120*** -0.116***

(0.0226) (0.0225)

Secretary Education 0.0212 0.138

(0.306) (0.356)

Secretary Tenure 0.0145 0.0128

(0.0497) (0.0471)

Minority -0.587 -0.405

(0.458) (0.453)

Central Experience -0.223 -0.242

(0.354) (0.377)

Strategic Shift -0.0309 0.00402

(0.220) (0.228)

Central Experience& Strategic Shift 0.306 0.247

(0.395) (0.404)

faction_shanghai 0.306 0.144

(0.222) (0.253)

faction_tuan -0.175 -0.161

(0.188) (0.174)

faction_taizi 0.387 0.396

(0.425) (0.471)

cut1 -5.979** -2.035***

(3.033) (0.114)

cut2 -1.717 -0.272***

(2.907) (0.0974)

Observations 532 532

Note: Minority, Central Experience, Strategic Shift are all dummy variables. Central Experience equals

1 if the provincial leader has work experience in central government. Strategic Shift equals 1 if the

provincial leader comes from other provinces or institutions. *, **, and *** indicate significance at the

10%, 5%, and 1% levels, respectively. Robust standard errors are in parentheses. Year and province

dummies are included. We use the growth rate of fiscal population as instrument variable in column

(2).

Table 13. The Effect of NCCP on Growth Rate of Total Expenditure and Compositions

(The Baseline model)

(1) (2) (3) (4) (5)

Variables (real growth rate of ) Total

Expenditure

Administration Infrastructure Agriculture ESMC

NCCPpre1 0.0443*** 0.0344* -0.0895*** 0.0365 0.0199

(0.0104) (0.0188) (0.0306) (0.0467) (0.0124)

NCCP -0.0526*** 0.0369** -0.0731** 0.0661 0.0185**

(0.0119) (0.0169) (0.0350) (0.0482) (0.00817)

NCCPpost1 -0.0180* 0.00521 -0.0156 -0.0371 -0.00696

(0.00936) (0.0196) (0.0346) (0.0636) (0.00754)

NCCPpost2 -0.0283* 0.0103 -0.0493 0.168** 0.000889

(0.0140) (0.0181) (0.0508) (0.0616) (0.00808)

Secretary Age 0.00125 0.00177 0.00142 -0.00106 0.000655

(0.00123) (0.00272) (0.00281) (0.00344) (0.00117)

Secretary Education -0.00183 -0.00306 -0.0492 -0.0113 0.00183

(0.0116) (0.0135) (0.0409) (0.0465) (0.00919)

Governor Age -0.00204* 0.000474 0.00609* 0.00586 -0.000511

(0.00116) (0.00168) (0.00339) (0.00377) (0.000700)

Governor Education -0.0194 -0.00873 0.0331 -0.0460 0.00938

(0.0136) (0.0263) (0.0465) (0.0907) (0.00849)

Growth_ProvExpenditure 0.370*** 2.373*** 0.653** 0.368***

(0.0752) (0.240) (0.296) (0.0424)

Growth_ProvRevenue -0.00939 0.0604* 0.229** -0.0473 0.111***

(0.0360) (0.0353) (0.0866) (0.0677) (0.0175)

Growth_National GDP -2.911*** -1.121 -13.98*** 0.260 -0.360

(0.950) (1.007) (2.567) (3.502) (0.504)

Growth_ProvPopulation 0.0429 1.236*** -0.677 -0.614 0.243

(0.434) (0.390) (0.891) (1.851) (0.265)

Growth_GDPpercapita 0.00479*** -0.000208 -0.00551*** 0.00738 -0.000119

(0.000899) (0.000906) (0.00179) (0.00458) (0.000528)

Growth_ProvUrbanIncome 0.356*** 0.680*** 0.221 -0.213 0.473***

(0.108) (0.110) (0.250) (0.337) (0.0695)

Provincial Second Industry

Ratio

0.271* 0.333 -0.0429 0.391 0.138

(0.138) (0.258) (0.381) (0.517) (0.139)

Lagged Dependent Variables 0.0346 -0.0504 -0.0782 -0.193** -0.0638

(0.0454) (0.0565) (0.0510) (0.0749) (0.0556)

Fixed Effect Yes Yes Yes Yes Yes

Time Trend Yes Yes Yes Yes Yes

Time Trend Square Yes Yes Yes Yes Yes

Observations 420 420 420 420 420

R-squared 0.584 0.331 0.560 0.207 0.576

Number of Provinces 28 28 28 28 28

Note：*, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Robust standard

errors are in parentheses. The baseline model includes time trend and fixed effect.

Table 14. The Effect of NCCP on Growth Rate of Total Expenditure and Compositions (GMM)

(1) (2) (3) (4) (5)

Variables (real growth rate

of )

Total

Expenditure

Administration Infrastructure Agriculture ESMC

NCCPpre1 0.0719*** 0.0549*** -0.0680* 0.0902*** 0.0176

(0.0209) (0.0129) (0.0401) (0.0291) (0.0141)

NCCP -0.0454* 0.0456*** -0.0798** 0.0638*** 0.00125

(0.0253) (0.0104) (0.0383) (0.0202) (0.0177)

NCCPpost1 -0.0347* 0.0161 -0.0523 -0.00130 -0.0136

(0.0200) (0.0114) (0.0454) (0.0261) (0.00971)

NCCPpost2 -0.00384 0.0159 -0.0544 0.164*** -0.0325***

(0.0205) (0.0111) (0.0508) (0.0400) (0.00746)

Secretary Age -0.00210 8.32e-05 -0.00559 -0.00297 -0.00694

(0.00734) (0.00150) (0.00645) (0.00411) (0.00691)

Secretary Education 0.0208 -0.00324 -0.0220 -0.0498 -0.174**

(0.141) (0.0125) (0.0481) (0.0370) (0.0837)

Governor Age 0.00769 -0.00105 0.0113** 0.0125*** -0.00229

(0.0129) (0.00114) (0.00486) (0.00463) (0.00729)

Governor Education -0.225 -0.00875 0.178*** -0.0246 -0.000502

(0.202) (0.0161) (0.0419) (0.0581) (0.0597)

Growth_ProvExpenditure 0.329*** 2.472*** 0.584*** 0.184*

(0.0568) (0.159) (0.0915) (0.0964)

Growth_ProvRevenue -0.126 0.0281 0.195** -0.0482 0.149***

(0.122) (0.0191) (0.0890) (0.0504) (0.0253)

Growth_National GDP -1.338 1.056*** -2.731** 1.294* -0.948***

(1.064) (0.308) (1.107) (0.667) (0.327)

Growth_ProvPopulation -2.750 1.556*** -0.337 3.258 -0.704

(2.193) (0.282) (11.35) (2.618) (0.932)

Growth_ProvGDP 0.00630** 2.09e-05 -0.00269 0.00505*** -0.00178

(0.00246) (0.000537) (0.00184) (0.00156) (0.00116)

Growth_ProvUrbanIncome 0.502 0.561*** 0.207 -0.261 0.681*

(0.595) (0.0948) (0.335) (0.281) (0.375)

Provincial Second Industry

Ratio

0.121 -0.116 -0.754 -0.655 0.119

(0.536) (0.309) (0.882) (0.421) (0.287)

Lagged Dependent

Variables

0.125 -0.0214 -0.0636* -0.137*** -0.144***

(0.260) (0.0521) (0.0353) (0.0196) (0.0538)

Fixed Effect Yes Yes Yes Yes Yes

Time Trend Yes Yes Yes Yes Yes

Observations 420 420 420 420 420

Number of provinces 28 28 28 28 28

Sargan test 14.73 16.59 14.85 12.38 10.80

(0.26) (0.94) (0.97) (0.99) (0.46)

AR(2) test -0.66

(0.51)

0.91

(0.36)

-0.93

(0.35)

0.61

(0.54)

0.34

(0.73)

Wald Test 1231.45

(0.00)

1511.80

(0.00)

3466.80

(0.03)

1240.66

(0.00)

1020.62

(0.00)

Note: System GMM is applied, with two lag dependent variable as instruments. Time trend and

province fixed effect are included. Control variables are the same as those in baseline model. Sargan

test shows that the instruments are valid. AR(2) test shows that there is no auto correlation of error

terms. Wald Test results show the joint significance of all of the coefficients. *, **, and *** indicate

significance at the 10%, 5%, and 1% levels, respectively. Robust standard errors are in parentheses.

Table 15. The Effect of NCCP on Total Expenditure and Compositions

(First Difference, GMM Approach )

(1) (2) (3) (4) (5)

Difference in Variables Expenditure Administration Infrastructure Agriculture ESMC

NCCPpre1 5.972** 0.915*** -1.928*** 1.080** 1.314***

(2.417) (0.294) (0.736) (0.541) (0.255)

NCCP -2.946 0.119 -1.313* 1.535*** 0.361

(2.971) (0.276) (0.773) (0.452) (0.299)

NCCPpost1 -5.180** 1.467*** -0.541 -0.652 0.216

(2.109) (0.481) (0.968) (0.669) (0.243)

NCCPpost2 -1.175 0.478*** -0.953* 1.774*** 5.54e-05

(1.698) (0.174) (0.522) (0.492) (0.376)

Secretary Age -0.0557 0.00550 0.00854 -0.0846*** 0.0463*

(0.303) (0.0271) (0.0383) (0.0234) (0.0261)

Secretary Education -6.044 0.293 -0.960* -0.629** 0.472

(10.65) (0.312) (0.504) (0.304) (0.369)

Governor Age -0.542 0.0237 0.0580 0.109** 0.0558*

(0.453) (0.0242) (0.0923) (0.0473) (0.0331)

Governor Education -23.42 0.330** -0.773 -1.815 0.703

(16.98) (0.150) (5.781) (1.981) (0.441)

Dif_Total Exp 0.0778*** 0.213*** 0.00892 0.119***

(0.00657) (0.0188) (0.0120) (0.0120)

Dif_Total Rev -0.00964* -0.0423*** 0.0261*** -0.0315***

(0.00565) (0.00975) (0.00695) (0.00496)

Dif_ National GDP -124.8 55.52*** 22.48 -28.84 31.38**

(79.86) (16.44) (38.34) (23.01) (12.58)

Dif_population -0.0307*** 0.00914*** -0.00973*** 0.00329*** 0.00414***

(0.00418) (0.000932) (0.00288) (0.000879) (0.00117)

Dif_GDP per capita 0.0118** 0.00180*** -0.00116** 0.000917 0.00116***

(0.00550) (0.000632) (0.000546) (0.000686) (0.000406)

Growth_ProvUrbanIncome 0.0122** 0.00109 4.50e-05 -0.000641 0.00397***

(0.00508) (0.000704) (0.00111) (0.00105) (0.000879)

Provincial Second Industry

Ratio

131.4 -103.0*** -41.42 16.15 -53.78*

(200.2) (33.27) (60.83) (37.32) (30.10)

Lagged dependent

variables

0.623*** 0.326*** -0.0544 -0.402*** 0.322***

(0.0896) (0.0445) (0.0548) (0.0349) (0.0298)

Dif_Lagged Total Rev -0.194***

(0.0344)

Fixed Effect Yes Yes Yes Yes Yes

Time Trend Yes Yes Yes Yes Yes

Observations 420 420 420 420 420

Number of provinces 28 28 28 28 28

Sargan test 17.03 9.14 18.86 13.40 13.95

(0.93) (0.99) (0.88) (0.99) (0.98)

AR(2) test -1.04

(0.30)

1.29

(0.20)

-1.50

(0.13)

0.59

(0.55)

0.80

(0.42)

Wald Test 30695.19

(0.00)

14969.53

(0.00)

4185.46

(0.00)

2677.84

(0.00)

84746.53

(0.00)

Note: This table uses first difference instead of growth rate to check whether the results are robust to

outliers. We use the same specifications as those in Tsai(2013): controls lagged revenue for the

regression of total expenditure, controls revenue and expenditure for the regressions of disaggregate

expenditures. System GMM is applied, with two lag dependent variable as instruments. Sargan test

shows that the instruments are valid. AR(2) test shows that there is no auto correlation of error terms.

Wald Test results show the joint significance of all of the coefficients. Time trend and province fixed

effect are included. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

Robust standard errors are in parentheses.

Table 16. Frequency of Turnovers and Fluctuation of Expenditure Categories (Baseline)

Variables (variance in real

annual growth of )

(1) (2) (3) (4) (5)

Total

Expenditure

Administration Infrastructure Agriculture ESMC

Secretary Turnover

Frequency

0.00106 0.00119 -0.00599 -0.00101 -0.00175

(0.00115) (0.0346) (0.0316) (0.183) (0.00384)

Governor Turnover

Frequency

-0.000577 -0.0127 0.0177 0.0332 -0.000749

(0.000980) (0.0292) (0.0266) (0.154) (0.00323)

Total Expenditure Variance 1.421 17.15*** 34.95 1.675**

(6.292) (5.743) (33.17) (0.697)

Total Revenue Variance 0.00767 0.636 -0.861 -6.308 0.204

(0.0872) (2.574) (2.350) (13.57) (0.285)

Population Growth

Variance

-0.999 -64.27 -21.01 74.52 -5.319

(2.917) (86.33) (78.80) (455.1) (9.564)

GDP Growth Variance 2.325* -17.37 -31.67 -144.7 -4.393

(1.205) (38.46) (35.10) (202.8) (4.261)

Observations 28 28 28 28 28

R-squared 0.220 0.039 0.333 0.076 0.242

Note: Dependent variables are the variance of different types of expenditure growth from 1990 to 2006.

*, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Standard errors are in

parentheses.

Table 17. Frequency of Turnovers and Fluctuation of Expenditure Categories

Variables (variance in real

annual growth of )

(1) (2) (3) (4) (5)

Total

Expenditure

Administration Infrastructure Agriculture ESMC

Secretary Turnover

Frequency

0.00225* -0.0180 -0.00252 -0.105 -0.00443

(0.00115) (0.0412) (0.0390) (0.220) (0.00458)

Secretary Tenure_sd 0.000861 -0.0440 0.00335 -0.146 -0.00192

(0.00136) (0.0450) (0.0426) (0.240) (0.00501)

Governor Turnover

Frequency

0.00118 -0.0291 0.0213 -0.0687 -0.00368

(0.00115) (0.0389) (0.0368) (0.207) (0.00433)

Governor Tenure_sd 0.00374* -0.0268 0.00801 -0.207 -0.00668

(0.00181) (0.0655) (0.0620) (0.349) (0.00729)

Total Expenditure Variance 4.899 16.56** 53.08 2.133**

(7.346) (6.951) (39.16) (0.817)

Total Revenue Variance 0.0203 0.315 -0.821 -7.676 0.176

(0.0797) (2.622) (2.481) (13.98) (0.292)

Population Growth

Variance

1.654 -79.69 -15.93 -54.15 -9.603

(2.952) (97.73) (92.47) (521.1) (10.87)

GDP Growth Variance 2.819** -17.92 -29.67 -185.4 -6.230

(1.265) (46.43) (43.93) (247.5) (5.165)

Observations 28 28 28 28 28

R-squared 0.410 0.085 0.134 0.027 0.014

Note: Dependent variables are the variance of different types of expenditure growth from 1990 to 2006.

*, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Standard errors are in

parentheses.