Eat, Drink, Firms and Government: An Investigation of ...Eat, Drink, Firms and Government: An...
Transcript of Eat, Drink, Firms and Government: An Investigation of ...Eat, Drink, Firms and Government: An...
-
Eat, Drink, Firms and Government:
An Investigation of Corruption from Entertainment Expenditures of
Chinese Firms ∗
Hongbin Cai† Hanming Fang¶ Lixin Colin Xu§
Abstract
Entertainment and Travel Costs (ETC), an expenditure item in standard accounting books of firms in China, amount to about 20% of total wage bills in a sample of 3470 Chinese firms. Using a detailed dataset of these firms, we analyze the composition of ETC and effects of ETC on firm performance. We develop a simple model of managerial decisions on the amount of entertainment expenditures to spend on strengthening relational capital with suppliers and clients, bribing government officials, and private consumption. This model allows us to identify components of ETC by examining how they should respond to different environmental variables. We find strong evidence that firms’ ETC compromise a mix that includes expenditures on government officials both as “grease money” and “protection money,” expenditures to build relational capital with suppliers and clients, and managerial private consumption. Overall, ETC have significantly negative effects on firm performance, but their negative effects can be much less pronounced when their marginal returns are higher, particularly, under severe government expropriation and when the quality of government service is very poor.
JEL: L2, O1, H2 Key Words: Corruption, Firm performance, Government Expropriation, Corporate Governance.
∗ We gratefully acknowledge financial support from the United Kingdom’s Department for International Development (DFID), World Bank Research Committee and the Multi-donor Funded Knowledge for Change Program. The views expressed here do not represent those of the World Bank or its executive directors. We are grateful to Yang Jian for insights on accounting practices in Chinese firms. We thank George Clarke, Robert Cull, Philip Keefer, Naomi Lamoreaux, Enrico Moretti, Jean-Laurent Rosenthal, Dan Treisman and Christopher Udry for helpful discussions and comments. We are responsible for all remaining errors. † Department of Economics, University of California, Los Angeles, 405 Hilgard Ave, Los Angeles, CA 90024. Tel: (310) 794-6495. Fax: (310) 825-9528. Email: [email protected] ¶ Department of Economics, Yale University, P.O.Box208264, 37 Hillhouse Avenue, New Haven, CT 06520-8264 . Tel: (203) 432-3547. Fax: 203-432-6323. Email: [email protected] § World Bank and Guanghua School of Management, Peking University. MC 3-300, Development Research Group, World Bank, 1818 H Street, N.W., Washington, DC 20433. Tel: (202) 473-4664. Fax: (202) 522-1155. Email: [email protected].
-
- 1 -
1. INTRODUCTION
Corruption is a central issue in development, political economy and transition economics, yet
few studies provide systematic hard evidence of it. By its very nature, objective data on corruption
are difficult to come by. With few exceptions, researchers have resorted to indices of perceived
corruption across countries or micro-level surveys based on self-reporting by managers or
government officials.1 While useful for some purposes, such data can suffer from a number of biases
(Bertrand and Mullainathan, 2001).2
While finding reliable measures of corruption with solid data is important in and of itself, it is
perhaps more important to figure out the causes and effects of corruption. Corruption is a product of
bad institutions, broadly defined, but which kinds of weak institutions are more likely to breed
corruption? To use the dichotomy of Acemoglu and Johnson (2003), when “property rights
institutions” are weak and firms are subject to constant expropriation by government officials, firms
are forced to pay bribes in exchange for protection or less expropriation. In such cases corruption has
no productive role and only increases social waste. 3 On the other hand, when “contracting
institutions” are weak, in the sense that governments do not provide services at the level that
facilitates trade and commerce, such as slow license approvals or erratic contract enforcement, firms
are induced to bribe government officials to do what they are supposed to do. Some scholars argue
corruption such as this, which “greases the wheel,” can increase social welfare by “making things
work” in an otherwise rigid and stifling bureaucratic system (e.g., Lui, 1985). Others counter that
corruption makes things worse because firms and government officials cannot “trade efficiently”
(e.g., Shleifer and Vishny, 1993). This debate clearly cannot be settled without compelling empirical
evidence. Unfortunately, due to the limitations of data on corruption, the empirical work has shed
little light on the causes of corruption and on the theoretical debate about its effects.
Using a comprehensive dataset with a large sample of firms from China, we adopt a novel
approach to identify corruption and its causes. Specifically, we study an expenditure item in the
1 Rose-Ackerman pioneered studies of corruption in economics, see, e.g., Rose-Ackerman (1978). Bardhan (1997) summarizes the literature up to the date. For recent contributions that rely on perception data, see, for example, Mauro (1995), Ades and Di Tella (1999), and Treisman (2000). Studies that use subjective survey data of corruption and other related papers will be discussed later in the text. 2 Bertrand and Mullainathan (2001) discuss many potential problems such as perception biases, incentive issues in self-reporting, and other well-known biases, and reach the conclusion that their “findings cast serious doubts on attempts to use subjective data as dependent variables.” 3 The welfare implications of such protection bribery on ex ante investments can be ambiguous, depending on conditions such as whether official tax rates are set at socially optimal levels.
-
- 2 -
accounting books of Chinese firms called “Entertainment and Travel Costs” (ETC), which is
predominantly composed of expenditures on eating, drinking, Karaoke, sports club membership, and
travel. We develop a simple model of managerial decisions on the amount of entertainment
expenditures to spend on strengthening relational capital with suppliers and clients, bribing
government officials, and private consumption. This model allows us to identify components of ETC
by examining how they should respond to different environmental variables. We find compelling
evidence that corrupt payments to government officials account for a significant portion of
entertainment expenditures. Moreover, our findings suggest that firms pay bribes both to buy
protection against government expropriation and to grease the wheel in buying government services.
Correspondingly, the more severe the government expropriation and the worse the quality of
government services, the more ETC contributes to firm performance.
There are several reasons why we focus on entertainment expenditures. First, the magnitude
of ETC is large. For the 3,470 firms in our dataset, on average each firm spends 946,400 Chinese
Yuan (more than 110 thousand U.S. dollars) annually on ETC, which is about 20% of the average
firm’s total wage bill, about 2.6% of its total sales, and 4.0% of its total value added. Even though
we do not know of any comparable data on entertainment expenditures of firms in other economies,
these numbers suggest that ETC are an important phenomenon in Chinese firms and may be
excessive relative to normal business needs.4 A second reason to study ETC is that the accounting
category of “Entertainment and Travel Costs” in China is a very easy place to disguise corruption
due to extremely lax accounting regulations and enforcement. In addition, entertainment providers
are gladly willing to produce any kind of receipt clients would like. Entertainment expenditures are
widely believed to be corrupt. As a result, the central government has numerous times unsuccessfully
attempted to curb such practices.5 A third reason we focus on ETC to study corruption is that
entertainment expenditures are a hard accounting measure. These expenditures are legal and show up
4 The fact that firms spend large amounts of money on entertainment should not be surprising to observers of China. It is consistent with other available estimates. For example, Hu (2004) estimates that in 1997 Chinese firms and government agencies spent more than 200 billion Yuan (about 25 billions U.S. dollars) on eating and drinking, and another 283.5 billion Yuan (about 34 billion U.S. dollars) on tourism and oversea travel. Firms were responsible for the lion’s share of these expenditures (for their interactions, it is more likely that firms entertain the governments than vice versa). These entertainment expenditures of 484 billion Yuan account for about 6.5 percent of China’s GDP (about 7.4 trillion Yuan) and represent roughly 60 percent of the total national tax revenue in 1997. Recent news reports in fact suggest that such expenditures have only increased over time. 5 For example, the central government once announced a decree that there be no more than 4 dishes in any meal provided to government officials. As a result, many feasts for government officials featured 4 super sized plates, each with several courses of food. This decree is no longer enforced.
-
- 3 -
in the firms’ accounting books; managers must present receipts to get reimbursed. In contrast to the
problems inherent in surveys which quiz managers directly about bribery, there is both little incentive
and flexibility for managers to misreport on entertainment expenditures.6
However, entertainment expenditures likely contain both legitimate business expenses in
addition to illegitimate ones. How can one then identify corruption from those expenditures? Our
strategy is the following. We postulate that entertainment expenditures can fit into three different
categories: (i) normal business expenditures such as on those which build relationships with suppliers
and clients; (ii) manager own consumptions such as taking family and friends to restaurants, clubs,
and recreation; and (iii) corruption payments to government officials. With a simple theoretical
model, we derive predictions of how ETC respond to changes in firm characteristics and the business
environments (e.g., government expropriation and service quality), and what effects ETC have on
firm performance if a significant portion of ETC is used to serve each of the three functions.
Specifically, our model indicates that if firms are subject to severe government expropriation and are
forced to spend money entertaining government officials as a means of bribery, then ETC will be
higher for firms more prone to expropriation. If a city government is inefficient and refrains from
providing good services, firms in the city have to spend more on entertaining government officials to
“get things done.” In contrast, if entertainment expenditures are not spent as bribery to government
officials, then ETC should be decreasing in government expropriation or increasing in government
service quality.
Using our dataset of over 3,000 firms, we test these predictions to infer whether a significant
portion of ETC can be accounted for as corrupt payments to government officials.7 We find that ETC
is higher when government expropriation (proxied by the effective tax burden) is more severe, and
when the quality of government service is lower. Both effects are statistically significant. By the
identification results of our model, these findings show that corrupt payments to government officials
account for a significant portion of entertainment expenditures. Moreover, they also suggest that
firms pay bribes both to buy protection against government expropriation and to grease the wheel in 6 However, it is common for firms and government agencies in China to have off-balance-sheet accounts. Some entertainment expenditures, especially those that look more suspicious, are paid from those accounts. Thus, our measure of entertainment expenditures underestimates the actual entertainment expenditures of the firms in our sample. Actual managerial own consumption and corruption spending in entertainment expenditures are probably more significant than what we identify. 7 In this study we use a narrow definition of corruption that includes only entertainment type bribes firms pay to government officials in exchange for protection or services. Though excessive entertainment expenditures by managers on themselves and their family and friends are often called as “fraud” or “corruption” in China, especially when done by managers of state-owned firms, here they are classified as managerial perks.
-
- 4 -
buying government services. These findings are further reinforced in our analysis of the effects of
ETC on firm performance. Overall, ETC affect firm profits and productivity adversely. But the
effects of ETC are significantly less pronounced where the tax burden is higher, strongly suggesting
that ETC are spent to reduce government expropriation. Our estimates suggest that, once the
endogeneity of ETC is accounted for, for firms whose tax burdens are one standard deviation higher
than the average firm of the sample, the negative effect of ETC on profitability is reduced by 10%
(in absolute value) 8 and that on productivity reduced by 14%. Similarly, for firms that face
government officials offering services one standard deviation worse than the average, the negative
effect of ETC on profitability drops by 30 percent and that on productivity drops by 17 percent. In
summary, our empirical results show that a large portion of entertainment expenditures by Chinese
firms is corruption payments to government officials, both as protection money and grease money.
We also find evidence that a significant portion of entertainment expenditures is used
productively to conduct business and build relational capital with suppliers and clients. Specifically,
firms that sell their main products to other provinces spend more on entertainment than those who do
not, suggesting that entertainment expenditures increase in transportation costs and greater
difficulties in maintaining long-distance relationships.9 Moreover, the length of time that firms have
known their main suppliers and clients has a negative, though not statistically significant, effect on
ETC, and has a significantly negative effect on ETC’s contribution to firm performance. The latter
finding suggests that for firms without long-term relationships with their main suppliers and clients,
money spent on entertainment can be good investments. Together, these findings indicate that firms
spend on entertainment to build relationships with suppliers and clients.
Not all entertainment expenditures, however, are used by managers to advance the interests
of their firms. Our analysis also uncovers evidence supporting the view that a big chunk of
entertainment expenditures is spent to satisfy managers themselves. Overall, ETC are found to have
significantly negative effects on firm profits and productivity, suggesting that there is a strong
managerial own consumption component of entertainment expenditures. However, the negative
effects of ETC are much reduced for firms with greater private ownership and for firms that have
fired a manager in the previous 4 years. Once the endogeneity of ETC is accounted for, increasing
private ownership by one standard deviation reduces the negative effects of ETC on productivity by 9
8 We use the same convention in the rest of the Introduction. 9 Alternatively, some of this effect could arise because firms doing business in other provinces may have to pay additional bribes to government officials in those provinces. This will strengthen our main points on corruption.
-
- 5 -
percent, and on productivity by 17 percent. Similarly, increasing the likelihood of managerial
dismissal by one standard deviation reduces the negative effects of ETC on productivity by 31
percent, and on productivity by 15 percent. One interpretation of this finding is that private firms and
firms that have fired a manager in the previous four years have better corporate governance and place
greater emphasis on profit when evaluating managers. Following this logic, our regression results
suggest that as managers become more profit-oriented, they use entertainment expenditures more for
productive uses. Overall, our findings indicate that a significant portion of ETC goes toward
managers’ own consumption.
Our findings shed light on the relationship between economic growth, corporate governance
and institutions. In the case of China, both property rights institutions and contracting institutions, or
more specifically, non-expropriating and efficient city governments, are important for economic
growth. Across firms, improved corporate governance curbs managerial excess. Since ETC are a
significant expenditure item for firms and have significantly negative overall effects on firm
performance, more attention should be paid to monitoring them better and making them more
accountable. Possible measures include improving accounting regulations and enforcement, making
them more transparent, and auditing of entertainment expenditures by firms as well as government
agencies more intensively.10
Our paper belongs to a growing empirical literature on corruption that uses micro-level data.
Within this literature, Di Tella and Schargrosky (2003) is the only paper that uses hard measures.11
The authors find that hospital input prices in Buenos Aires are negatively correlated with wages and
auditing intensity, suggesting bribe-taking behavior of hospital procurement officers. Using
information on bribery reported by a sample of 176 Ugandan firms, Svensson (2003) finds firms’
choices to bribe or not depend on their regulatory environments, while their bribing payments depend
on their “ability to pay” and their bargaining positions relative to government officials. Using the
same dataset of Ugandan firms, Fisman and Svensson (2002) show that bribery has large negative
effects on firm growth. Clarke and Xu (2004) use enterprise-level data on bribes paid to utilities in 21
transitional countries in Eastern Europe and Central Asia, and find that corruption systematically
10 Of course, this is not to say that the central government is not corrupt and that just by imposing central regulation the problem can be solved. 11 Although not trying to measure corruption, Fisman (2001) provides a clever estimation of the value of relationships with governments to firms in Indonesia.
-
- 6 -
varies with characteristics of bribe takers and givers in economically intuitive ways.12 Kaufmann and
Wei (1998) analyze data from three worldwide firm surveys and find a positive correlation between
bribes and bureaucratic red tape, evidence they interpret as suggesting that corruption does not grease
the wheel of commerce. With survey data of private manufacturing firms in Eastern European
countries, Johnson et al. (2000) find bureaucratic corruption positively correlated with firms hiding
output. Finally, Hunt (2004) and Mocan (2004) use a same cross-country survey dataset on
individual bribery activities to analyze a host of interesting questions about corruption on the
individual level.
Our paper adds to this literature, and has several distinctive features. First, we use hard
measures of entertainment expenditures from over 3000 Chinese firms to identify corruption and its
causes. Since entertainment expenditures are quite significant in magnitude, corruption identified
from these expenditures should be important. Second, our methodology for identifying corruption is
new, and should be applicable in other contexts. Third, we use a rich dataset with a large sample of
firms to conduct a detailed analysis of effects of ETC on firm performance. The results provide
strong support for our basic findings from the standard regression analysis of the composition of
ETC. Fourth, we identify both protection money and grease money as causes of corruption.
In addition to considering corruption, our paper also sheds light on managerial perks. In this
sense it is closely related to Rajan and Wulf (2004), who examine managerial on-the-job
consumption. As they point out, there is little empirical work on managerial perks in the existing
literature, despite a great deal of attention from theory and from the public about managerial
incentives and compensations. With a sample of 300 large U.S. corporations, the authors find mixed
evidence for the view that perks are managers’ private benefits, but more compelling evidence that
perks are a means to enhance productivity. Consistent with Rajan and Wulf (2004), we find that a
significant portion of ETC spent in our sample is used to bribe government officials, a corrupt
activity that may be socially wasteful but beneficial to firms. On the other hand, we also find
evidence that a significant portion of ETC are spent toward the private benefit of managers. In fact,
overall, ETC have a negative effect on firm profit and productivity. The difference between our
findings and those of Rajan and Wulf (2004) likely reflects the different qualities of corporate
12 Glaeser (2003) also finds that ownership matters for controlling “grand” corruption. He suggests that the wave of utility nationalization in the early 20th century was intended to reduce corruption.
-
- 7 -
governance of firms in China and the U.S., and the different institutional environments within which
they operate.13
The remainder of the paper is structured as follows. In Section 2, we present a simple model
to illustrate our identification strategy. Section 3 describes our data and presents descriptive statistics.
In Section 4, we examine the determinants of total entertainment expenditures. Section 5 investigates
how entertainment expenditures affect firm performance, and Section 6 concludes.
2. THEORETICAL FRAMEWORK In this section, we propose a simple model of entertainment spending by a manager of a
representative Chinese firm. The manager decides the amount of firm funds to spend in three
categories: (1) normal business expenditures, in particular, spending on building relational capital
with suppliers and/or clients, rx ; (2) self-indulgent consumption or perks, which includes money that
goes directly into his own pocket, his own consumption, consumption by his family and friends, and
“gift exchange” with others, denoted by cx ; and (3) corruption payments to government officials,
denoted by bx . The total corruption spending in the last category can be further divided into two sub-
categories: bribes in exchange for government services (“grease money”), gx , and bribes in
exchange for government protection (“protection money”), px , where b g px x x= + .
By definition, the private consumption cx is simply a deduction from firm profit. The
bribes to government officials, gx and px , as well as the relational investment rx , however, have
a positive effect on the firm’s profit. We now detail the specific roles of gx , px and rx .
Image that the firm has a potential profit opportunity 0Q > .14 The probability that Q
can be realized, or, equivalently, how much of Q can be realized, is denoted by ( ),g rq K K . This probability depends on both cooperation from government officials and from the firm’s
13 The magnitudes of managerial perks differs greatly between firms in the U.S. and in China. According to Rajan and Wulf (2004), Hewitt Associates, a leading human resource consulting firm in the U.S., estimates the value of CEO perks at approximately 1 to 3 percent of total CEO compensation. In contrast, even if only 20 percent of entertainment expenditures are managerial perks in our sample, they would account for 4 percent of the firms’ total wage bills. Hence, total managerial perks, which include valuable items such as luxury cars and company condominiums, must be quite substantial relative to their monetary compensation. 14 In our empirical analysis, Q will be approximated by the firm’s basic characteristics, such as its industry, number of employees, age of the firm etc.
-
- 8 -
suppliers and clients, where gK is the firm’s actual relational capital with government, and rK is
the firm’s actual relational capital with its suppliers and clients. Naturally, we assume (, )q ⋅ ⋅ is a
strictly increasing and concave function of gK and rK ; moreover, we make the plausible
assumption that gK and rK are complements in ( , )g rq K K , that is, 2 0g rq K K∂ ∂ ∂ > .
The “grease money” bribes, gx , to government officials and the relational investment,
rx , with suppliers and clients improves the cooperation of the respective groups and makes it
easier for the firm to realize profit opportunities. We suppose that if the manager spends gx and
rx , then
; g g g g r r r rK k a x K k a x= + = +
where gk denotes the baseline government quality and is the cooperation that the firm can expect
from the government without additional “grease money” bribes. Similarly, rk is the baseline level of
the firm’s relational capital with its suppliers/clients and is the cooperation that the firm can expect
from them without any additional relational investment. The parameters ga and ra measure how the
firm’s entertainment expenditures of gx and rx improve its relationship with government and its
suppliers and clients, respectively.
The “protection money” bribe, px , affects the firm’s effective tax rate t . How much tax the
firm actually pays depends on three factors: (i) 0t , the official tax rate for the firm’s type, (ii) pk , the
baseline government expropriation that itself depends on both the general level of expropriation of
the city government and the bargaining power of the firm, and (iii) px , the firm’s “protection
money” bribes to government officials in exchange for lower expropriation. For simplicity, let the
actual level of government expropriation pK be p p p pK k a x= − , where pa measures how
effectively the firm’s spending, px , reduces government expropriation. The firm’s actual tax rate
( ; )p pt x k is assumed to be
( ) 0 0; ( ) ( )p p p p p pt x k t r K t r k a x= = − where ()r ⋅ is strictly increasing and concave.
-
- 9 -
Thus, if a manager chooses the vector ( ), , ,c r g px x x x , the firm’s net profit for the period,
( ), , ,c r g px x x xΠ , is given by
(1) ( ) ( ) ( ), , , [1 ( ; )] ,c r g p p p g r c r g px x x x t x k q K K Q x x x xΠ = − − + + + . Note that we make the realistic assumption that the tax base for Chinese firms is revenue, rather than
gross profit.
We also reasonably assume that the “owner” of the firm observes only the total net profit Π
and some noisy signal of the manager’s own consumption, cx , but cannot directly distinguish
corruption expenses from relationship building expenditures. 15 The firm owner decides in each
period whether to retain the manager based on the current period net profit and the manager’s own
consumption, cx .16 We assume that the probability that the manager is retained for the next period is
given by
(2) ( ) ( ), ; , 1 c ca a xc cx a a e ππρ τ − Π−Π = − ,
where (0,1)τ ∈ , and the positive parameters aπ and ca measure the corporate governance of the
firm. Higher aπ and ca imply that the manager is held responsible for the firm’s financial
performance to a greater degree and that the manager is better monitored.17 Note that the
function ( ), ; , ca aπρ ⋅ ⋅ nicely captures the idea that the manager is more likely to be retained if he
makes a higher profit and consumes less of the firm’s funds.
For simplicity, let 0V > be the manager’s (constant) continuation value from being retained
and let β be his discount factor. The manager in each period chooses ( , , , )c r g px x x x to maximize:
(3) ( ) ( ) ( )( )1 ; , , ; ,c p p g r c r g p c cU x t x k q K K Q x x x x x a a Vπβρ = + − − + + + . 15By firm “owner” we mean the party in the firm who decides on hiring and firing top managers. For corporations, this means the boards of directors; for state-owned firms, this means the government bureau supervising the firm; and for privately owned firms, the owner can be the manager or another private party. 16 For simplicity, we do not consider explicit incentive contracts here. It is easy to see that adding some profit-sharing, as long as it is not too large, in the manager’s payoff function will not change the qualitative results of the model, but can complicate the analysis substantially. 17 Implicitly we assume that firms face various constraints (such as verifiability, enforceability, legal constraints, etc.) in designing the incentive systems for their managers, and that and ca aπ represent the optimal solution under those constraints. In the ideal world, firms can make their managers maximize profit by committing to fire them for any own consumption or lower profit (i.e., by setting aπ and ca arbitrarily high).
-
- 10 -
The first order condition with respect to cx , assuming that its solution is interior is
(4) ( )[ ]* * lnc c ca a x a a Vπ πτβΠ − = + .
This can be rewritten as
(5) ( )[ ]** ln c
cc
a a a Vxa a
π π
π
ϕ τβ− +=+
where
(6) ( ) ( ) ( )* * * * * * *1 ; ,p p g g g r r r r g pt x k q k a x k a x Q x x xϕ = − + + − + + is the firm’s cash flow gross of the manager’s own consumption. This says that the manager
will consume more himself if the firm’s cash flow is greater. Not surprisingly, the manager’s
optimal self-consumption decreases in the monitoring intensity, ca , his discount factorβ and his
future value of employment, V .
The optimal levels of ( , , )r g px x x that maximize (3) should maximize the net profit function
(1). The following proposition summarizes the main testable implications of our model.
Proposition 1.
A. Suppose 0la = for some { }, ,l r p g∈ . Then * 0lx = , and the total entertainment
expenditure is increasing in lk when { , }l r g∈ , or decreasing in pk when l p= .
B. Suppose la is positive and sufficiently large for { }, ,l r p g∈ . Then *lx is strictly positive.
The total entertainment expenditure is decreasing in lk when { , }l r g∈ , or increasing in pk
when l p= .
Part A of Proposition 1 is easy to understand. When 0la = , any positive level of
expenditure on lx has no effect on revenue but decreases profit. Therefore the optimal expenditure
on lx is zero. To understand the rest of Part A, consider the case l g= . When 0ga = , we know
that * 0gx = , thus *g g g g gK k a x k= + = . Suppose and r pa a are positive and sufficiently large.
It is easy to see from Equation (1) that the profit function is supermodular ( and r px x are
complementary and have increasing differences in g gK k= ), and thus * * and r px x are increasing in
-
- 11 -
gk (Milgrom and Shannon, 1994). Moreover, the optimal level of cash flow *ϕ [see Equation (6)] is
increasing in gk regardless of whether ra and pa are zero or not, due to the general Envelope
Theorem (Milgrom and Segal, 2002). By Equation (5), *cx must always be increasing in gk .
Therefore, the total entertainment expenditure must be increasing in gk .
To understand Part B of Proposition 1, let us again consider the case l g= . When ga is
positive and sufficiently large, the optimal level of gx will be positive. Since 0gx > , we can use the
variable transformation ( )g g g gx K k a= − to rewrite the firm’s cash flow ϕ from Equation (1) as
(7) ( ) ( )[1 ( ; )] , g gp p g r r pg g
K kt x k q K K Q x x
a aϕ = − − + − +
With this variable transformation, we can think of the manager’s problem as choosing ,p rx x and gK
to maximize ϕ . Note that gk enters in this objective function only as a constant. Thus, the optimal
solution * *,p rx x and *gK must be independent of gk . This means that
* *( )g g g gx K k a= − decreases
linearly in gk . Let * * /g gB k aϕ = + , where
*B is the maximal value of
( ) ( )[1 ( ; )] , /p p g r r p g gt x k q K K Q x x K a− − + − that is independent of gk .18 By Equation (5),
(8) * * gcc g
kax Aa a a
π
π
= ++
where *A is a collection of terms that are independent of gk .19 Summing up the entertainment
expenditures of all categories, we have
* * * * * gcp r g cc g
kax x x x Da a aπ
+ + + = −+
where * * * * */p r g gD x x K a A= + + + is independent of gk . Clearly the total entertainment
expenditures are decreasing in gk . The analysis of the cases of l r= and l p= is analogous.
18 The explicit expression for *B is ( ) ( )* * * * * * *[1 ( ; )] ,p p g r r p g gB t x k q K K Q x x K a= − − + − . 19 The explicit expression for *A is
( )[ ]** ln cc
a B a a VA
a aπ π
π
τβ− +=
+.
-
- 12 -
Proposition 1 is the key result underlying our identification strategy. If, for example, the
“grease money,” gx , is not an important component of the firm’s ETC (i.e., 0ga = ), Part A of
Proposition 1 implies that the firm’s total ETC expenditure should be increasing in gk , the baseline
quality of government service. On the other hand, if gx is indeed an important component of the
firm’s ETC (i.e., ga is positive and sufficiently large), Part B of Proposition 1 implies that we should
expect the firm’s total ETC expenditure to be decreasing in gk . The distinct effects of rk and pk on
the firm’s total ETC expenditures are similar. To the extent that we can empirically observe the
firm’s total ETC and find reasonable proxies for rk , gk and pk , we can empirically verify whether la
is zero or sufficiently positive. This is exactly the empirical strategy we will follow.
Identification of the manager’s own consumption, cx , is somewhat different. In fact, our
specification of the manager’s payoff function in Equation (3) already assumes that the manager will
choose positive cx . Let us consider how managerial incentive intensity represented by the corporate
governance variable aπ and ca in the model affects total entertainment expenditures. By Equation
(2), if 0aπ = , then the manager will not use entertainment expenditures for other purposes except
for self-consumption, which implies that ETC would be independent of rk , gk and pk . Suppose both
aπ and ca are positive, then from the proof of Proposition 1, * * *, ,p r gx x x and the maximum cash flow
*ϕ are all independent of aπ and ca . From Equation (5), *cx is decreasing in ca . However,
*cx can
be decreasing in aπ (when aπ and *ϕ are relatively small), or increasing in aπ (when aπ and
*ϕ
are relatively large). The reason for the latter is that if profit is weighted heavily in managerial
evaluation, a manager will spend more on himself once the cash flow is high, since his job will
likely be secure. Thus, our model makes an ambiguous prediction about how the total
entertainment expenditure is affected by the firm’s corporate governance measures aπ and ca .
From the proof of Proposition 1, for any { }, ,l r p g∈ and for any la , the maximum cash
flow *ϕ is increasing in and r gk k and decreasing in pk . Subtracting *cx from
*ϕ , we have
Proposition 2. The optimal net profit *Π is increasing in and r gk k and decreasing in pk .
-
- 13 -
Proposition 2 is not a direct test of the importance of the expenditure categories of ETC, but
evidence consistent with it certainly lends support to our model as a whole.
Underlying our main results is the key assumption that the firm’s actual relational capital
stocks, rK , gK and pK , follow a linear accumulation technology. That is, the entertainment
expenditures rx , gx and px add to the baseline variables rk , gk and pk linearly. This type of
accumulation technology is commonly used, especially in the economic growth literature. More
importantly, qualitative results of the model should still hold when the actual accumulation
technologies are approximately but not exactly linear.
For simplicity, we have considered a static model of entertainment expenditures. When
parts of such expenditures are viewed as investments, particularly investments in long-term
relationships, dynamic considerations may be important. While we do formulate the manager’s
dynamic optimization problem in Equation (3), we really focus on the stationary state of the
dynamic problem. We make this modeling choice partly for simplicity, partly because of the
limitation of our data. Even if we solved a real dynamic model, we would not be able to test the
dynamic implications of the model since at this point we only have cross-section data.
We have so far illustrated the principles of our identification strategy by considering a
representative firm. In the empirical implementation of the identification strategy, we rely on
cross-firm variation along many dimensions. For example, profit opportunities, Q , among firms
vary by industry, size, age, and market coverage. A firm’s profit opportunities are also correlated
with its home city’s level of economic development, which varies greatly in our sample. Firms
vary greatly in their strength of private contracting relationship, rx , which we capture by the
number of years a firm has done business with its clients and suppliers. To measure a firm’s
relational capital with government officials, we use a variable called “government help”, which
is the likelihood that government officials tend to be helpful to the firm, to proxy for the baseline
quality of government service, gk . Firms in our sample also differ in ownership structure,
managerial dismissal, and tax burden.. Exploiting the ideas summarized in Proposition 1, we
infer the importance of the various components of the ETC by empirically examine how a firm’s
total ETC expenditures are affected by firm characteristics and environment variables.
-
- 14 -
3. DATA AND MEASUREMENTS Our data come from two firm-level surveys: the first covers 2,400 firms located in 18 cities
during 2000-02, and the second covers 1,070 firms located in 15 cities of Liaoning Province during
2001-03.20 Both surveys were conducted jointly by the World Bank and the Enterprise Survey
Organization of China, in order to investigate the investment climate of various cities in China. The
18 cities covered in the first survey were selected to achieve balanced representation across five
regions of China.21 For each city, either 100 or 150 firms were randomly sampled from an electronic
database of firms according to several criteria. First, firms in both manufacturing and service
industries were sampled. 22 Second, we pre-specify the firm sizes (measured by number of
employees) in selecting the firms to be surveyed.23 The second survey covered all major cities in
Liaoning Province. 80 firms were sampled in the cities of Shenyang and Dalian, the two largest cities
in the province, and 70 firms in all other 13 cities.24 Combining the two surveys yields a total of
3,470 firms located in 33 cities at very different stages of economic development. Within this sample
of cities, GDP per capita (in 2002 value) ranges from 3400 to 46,400 Chinese Yuan (about $400 to
more than $5,000 U.S. dollars).
The survey questionnaire has two parts. The first part, to be filled out by the firm’s senior
manager, asks for qualitative information on the firm. The second part covers financial and
quantitative information about the firm’s production and operation, and is obtained from interviews
with the firm’s accountant. Overall our dataset contains detailed information about firm
characteristics and performance. The average firm in our sample has 572 employees, and has been in
20 Two cities in Liaoning Province, Benxi and Dalian, appear in both surveys, so there are 31 cities in total in the two surveys. 21 Cities selected by region are. (1) The Northeast Region: Benxi, Dalian, Changchun, and Haerbin; (2) The Coastal Region: Hangzhou, Wenzhou, Shenzhen, and Jiangmen; (3) The Central Region: Nanchang, Zhenzhou, Wuhan, and Changsha; (4) The Southwest Region: Nanning, Guiyang, Chongqing, and Kunming; and (5) The Northwest Region: Xi’an and Langzhou. 22 The manufacturing industries include apparel and leather goods, electronic equipment, electronic components, consumer products, vehicles and vehicle parts. The service industries include accounting and related services, advertising and marketing, business logistics services, communication services, and information technology services. 23 The minimum number of employees for firms in the sample is 20 in manufacturing industries and 15 in service industries. We loosened the size criterion when there were not enough firms from a particular sector in a city. As a result, roughly 3% of firms in our sample have less than 15 employees. 24 The 13 other cities are Dalian Deveopment Zone, Anshang, Fushun, Benxi, Dandong, Jinzhou, Yingkou, Fuxin, Liaoyang, Tieling, Chaoyang, Pangjin, Huludao. We treat Dalian Development Zone as a separate city because it has relatively autonomous status and is significantly different from Dalian in general. The sample of firms in Dalian Development Zone is drawn independently from the firms in Dalian.
-
- 15 -
business for 15 years. Its annual total sales revenue is about 190 million Yuan, and makes a loss
equal to almost 9% of its total sales.25 Close to 60% of the firms in our sample sell their main
products to other provinces.
Obviously, the variable of central interest is Entertainment and Travel Costs (ETC), a readily
available category in the accounting book of every firm. Each reimbursement item in this category
has to have a receipt, so expenditures are measured with little error. The ETC are supposedly for
conducting business and building relations with suppliers and clients. However, accounting practice
in China is sufficiently lax so that managers may get reimbursed for almost any kind of entertainment
for any purpose.
ETC represent a significant portion of firms’ expenditures. The ratio of ETC to total wage
bill at the firm level has a median of 8.8%, and a mean of 19 percent (for the subsample excluding
the top 5 percent of the firms on this ratio). The ratio of ETC to value-added has a median of 1.8
percent and a mean of 4.0 percent.26 In our empirical analysis, we normalize Entertainment and
Travel Costs by total sales, and define the ratio as ETC/Sales. This variable has a median of 1
percent, and a mean of 2.6 percent in our data.27 ETC/Sales varies substantially across firms, with a
sample standard deviation of 6.2 percent. Across cities, the average ETC/Sales ratio ranges from 0.8
percent in Jiangmeng, a coastal and relatively affluent city, to 5 percent in Guiyang, an inland and
relatively backward city.
Table 1 presents descriptive statistics for all variables used in our analysis. Table 2 displays a
cross-city comparison of major variables. The cross-city comparisons of the ETC/Sales ratio show
some interesting patterns. First, the ETC/Sales ratio is on average lower in more developed cities.
Figure 1 shows a clear negative relationship between the mean ETC/Sales Ratio and the logarithm of
per capita GDP at the city level. Figure 2 plots the standard deviation of ETC/Sales ratio within the 25 There can be at least two reasons for such widespread losses among firms. First, Chinese firms hide a substantial amount of profits to evade taxes (Liu and Xiao, 2004). Second, profit variation is enormous across firms, and the standard deviation is almost 7 times the value of the mean. This is consistent with the fact that some state-owned firms are extremely good at destroying value. Due to these reasons, one should be careful in interpreting our regression results using profit data, if one believes that profit hiding is correlated with the variables used in our analysis. We have no way to deal with such data problems, nor do we know whether and how profit hiding is correlated with other variables we use. The other variable of firm performance, productivity, which uses value added or sales data, is more immune to these problems and hence more reliable. As will be shown later, regression results using profit and productivity are qualitatively similar, which suggests that potential problems of profit data do not ruin the analysis. 26 We exclude 44 firms with obvious data errors (with a ratio that is negative or greater than one) in this calculation. 27 We exclude firms reporting values of total ETC at the top 0.5 percentile in terms of ratio of ETC to sales (a total of 19 firms) from our sample. This ensures that our calculation is not driven by the outliers.
-
- 16 -
city against the logarithm of per capita GDP. It is interesting to note that there are larger variations
in the ETC/Sales ratio in less developed cities. Thus, less developed cities feature both higher levels
of the ETC/Sales ratio and higher variances. Importantly, GDP per capita differs greatly among the
cities. The ratio of the maximum per capita income in Shenzhen to the minimum in Chaoyang is
almost 14.
[Table 1 about Here]
[Table 2 about Here]
[Figure 1-2 About Here]
In our empirical analysis, we proxy the baseline government quality gk each firm faces by the
firm’s evaluation of the tendency of government officials to help firms, and we hereafter call this
government help. 28 Government help is a firm-specific measure of government helpfulness that
depends on the overall government quality and the relational capital each firm has with government
officials. Bounded between zero and one, Government help differs greatly across the cities, ranging
from 21% to 82%. Figure 3 graphs ETC/Sales against government help at the city level. Clearly, in
cities with better government help, firms tend to spend less on ETC.
[Figure 3 about here]
The baseline government expropriation pk that each firm faces is proxied for by the firm’s
total tax burden in the previous year, which we think represents the extent of government
expropriation if the manager does not spend additional entertainment money bribing government
officials.29 For several reasons, there is a substantial amount of variation in the tax burdens among
firms in China. First, central government tax rates differ across firm types (e.g., tax incentives to
28 Specifically, the variable is based on the answer to the following question: “Among the government officials that your firm regularly interacts with, what is the percentage that tends to help the firm develop?” 29 That is, consistent with our model, we assume that firms are in the stationary state of their dynamic optimization problem, where every period they respond to the stochastic environment shocks.
-
- 17 -
attract foreign investments) and across regions (e.g., negotiations between the central government
and provinces, tax reductions for special economic zones). Second, in the same city, the city
government often imposes different levels of local taxes. Third, tax law enforcement and collection
efforts differ greatly across cities and across firms within each city, so a firm’s actual tax burden
depends on the level of vigilance of local tax officials and the relationship its has with them. These
reasons explain why the actual tax burden, as measured by total taxes divided by sales, varies so
greatly across cities and across firms in the same city. Firms in our sample on average paid taxes
equal to 7.1 percent of their total sales, with a standard deviation of 9.7 percent. The lowest average
tax rate is found in Shenzhen at 3.8 percent, and the highest is Guiyang’s 11 percent. Figure 4 plots
ETC/Sales against the average tax burden at the city level. The graph shows a strong positive
relationship, suggesting that ETC may be used to cushion the blunt of government expropriations.
[Figure 4 about here]
To measure a firm’s baseline relational capital with its suppliers and clients, rk , we construct
a variable called Years of Relationship, which is the sum of the years that the firm has known its
most important supplier and the years that it has known its most important client. This variable shows
substantial variation across cities, with a low of 7 years in Lanzhou on average to a high of 9.1 years
on average in Dandong. The sample mean is 8.3 years, and the sample standard deviation is 2.1
years. In our regression, we use the logarithm of years of relationship. By Proposition 1, if managers
use entertainment expenditures to build private contracting relationships, ETC should be negatively
correlated with years of relationship; otherwise, ETC and years of relationship should be positively
correlated. Figure 5 below plots ETC/Sales against the logarithm of years of relationship with
contracting parties. There is a negative relationship, but it is not strong.
[Figure 5 about here]
Undoubtedly, managerial incentives are important to understand entertainment expenditures.
However, the incentive structure for managers in Chinese firms is not at all transparent, and no good
data are available on managerial incentives.30 For our purpose, we use two measures of corporate
30 There are several reasons for this. Most firms in our sample are not public firms, so they do not have to disclose compensations for mangers. Moreover, there are numerous accounting tricks which allow managers of Chinese firms to
-
- 18 -
governance to gauge how a manager’s incentives are aligned with those of the firm. The first variable
is “Private Ownership,” which measures the share of a firm owned by private parties, both domestic
and foreign. Motivated by profits, private owners are likely to have stronger incentives for
monitoring managers to reduce waste and increase efficiency. In our sample, the share of private
ownership varies across firms from 0 (purely state-owned) to 1 (purely private), and the intra-city
average varies from 82 percent in Wenzhou, a coastal city known for its entrepreneurship, to 17
percent in Fushun and Shenyang, two Northeastern cities. The average firm in our sample has 34.8%
of private ownership, with a standard deviation of 45.5%. Figure 6 plots ETC/Sales against private
ownership at the city level, suggesting a weak relationship between them. The second measure,
called “Fired Managers”, is a dummy variable indicating whether any senior manager (including
general managers and deputy managers) had been fired in the previous four years. Normally one
expects that the incidence of senior manager dismissal indicates strong monitoring of top managers
and hence greater pressure on current managers to improve firm performance.31 In our sample, 19
percent of firms fired a senior manager in the four years prior to the interview. There is substantial
variation in this variable across cities. In the city of Yingkou, for example, 33 percent of the firms
fired a manager in the previous four years, while only 7 percent of the firms in the city of Lanzhou
did so. Figure 7 does not show a clear relationship between managerial dismissal and ETC/Sales.
[Figure 6 about here]
[Figure 7 about here]
4. WHAT FIRMS SPEND MORE ETC? In this section, we empirically examine the determinants of ETC, using regression
specifications suggested by the simple model in Section 2. The dependent variable is the ETC/Sales
ratio, and the list of included explanatory variables varies by specification. For each regression get paid without being noticed by outsiders. Even when explicit incentive schemes are observed, they are typically subject to ex post renegotiations so the actual incentives managers face are still murky (Cai, Li and Zhou, 2003). 31 The opposite interpretation is also possible if firing senior managers indicates that things were very bad in a firm. Our empirical results do not seem to support this interpretation. Another alternative effect is that in a firm that fired a top manager before, the current manager may find it safer knowing that he is just hired. One thing to note, however, is that when we say a firm fired a top manager before, it does not mean that it fired the CEO of the firm. Thus, the current CEO may or may not be a replacement of some top manager fired before.
-
- 19 -
specification we only use the sample of firms for which there are no missing values for any of the
explanatory variables. For each firm, we only use the data from the last year in the survey because
ETC is observed only for the last year. The specification we estimate is
(9) 0 1 2 3/ ( _ )+ijc ijc j c i i ijcETC Sales IND Log GDP PC Z Xα α α α ε= + + + ,
where the subscript i stands for the firm, j for the industry, and c for the city. The variable jIND
is firm i ’s industry; and _ cGDP PC is the per capita GDP of the city where firm i is located.
Industry dummies are included to filter out industry-specific needs for normal business expenditures
in entertainment and travel costs. GDP per capita at the city level controls for various aspects related
to economic development. Alternatively, we also directly control for economic development with
city dummies. The variable iZ is a vector of basic firm characteristics, including size (log of the
number of employees) and age (log of firm age). The variable iX is a vector of firm-specific
variables that include
(1) Business conditions and relational capital: a dummy variable for whether the firm sells to
other provinces, the log of the number of years the firm has known its largest supplier and its
largest client;
(2) Service quality of the government: government help, defined as the likelihood of government
officials helping the firm in their interactions.
(3) Corporate governance: share of private ownership, a dummy variable for whether the firm
fired any senior manager in the last four years.
(4) Government expropriation: lagged tax burden. The tax burden variable is once lagged to
avoid contemporaneous bias.
In order to see if the results are sensitive to multicollinearity, we add variables step by step
starting with basic firm characteristics and municipal GDP per capita. Moreover, in the last two
columns, we include city dummies to more flexibly control for city characteristics. Since per capita
GDP at the city level is embedded in the city dummies, we cannot include this variable in the last two
columns when city dummies are included. We report the estimation results from 8 specifications in
Table 3. Since the estimates from different specifications differ little, we shall focus on the complete
specification in column 5.
-
- 20 -
[Table 3 about Here]
Our results show that larger firms, as measured by the number of employees, apparently have
lower ETC/Sales ratios. A one-standard deviation increase in log of number of employees would
reduce ETC by around 0.9 percent, or a reduction in ETC by 35 percent when evaluated at the mean
of 2.6 percentage point. This suggests that entertainment expenditures exhibit quite strong
economies of scale. There does not appear to be a robust relationship between ETC/Sales and firm
age. Consistent with Figure 1, firms in cities with higher per capita GDP tend to have lower
ETC/Sales ratio. Our estimates suggest that a one-standard-deviation increase in log per capita GDP
would reduce ETC by 0.2 percentage points, or 8% of the mean ETC level of 0.026.
Consistent with Figure 3, government help is negatively correlated with ETC/Sales. A one
standard deviation increase in this variable would decrease ETC/Sales by 0.2 percentage points. By
the identification results of Proposition 1, this finding suggests that firms entertain government
officials in exchange for better services, providing support for the grease money view of corruption.
The once lagged tax burden, which we use to measure government expropriation, has a
strong and positive effect on ETC/Sales. In column 6, tax burden is significant at the 5 percent level,
and the coefficient implies a 37% increase of ETC/Sales if the tax burden increases by one standard
deviation. Thus firms facing greater government expropriation have more entertainment
expenditures, consistent with the negative relationship at the city level seen in Figure 4. By
Proposition 1, this indicates that some ETC are protection money in exchange for lower
expropriation.
To allow for the possibility that the once lagged tax burden may be endogenous, we
experimented with the specification in column 6 that excludes the once lagged tax burden from the
regression. A comparison of columns 5 and 6 reveals that the other coefficient estimates barely
change, suggesting that there is no need to worry about the potential correlation of the once-lagged
tax burden with the firm-specific error term. This is reassuring, and is consistent with our implicit
stationarity assumption.
For the corporate governance variables, private ownership does not have a robust correlation
with ETC/Sales. The indicator of previous managerial dismissal, however, has strong and positive
effects on ETC/Sales. The coefficient is statistically significant across various specifications, and the
magnitude is large. Other things being equal, the ratio of ETC/Sales in firms that fired a senior
manager before is 0.8% higher than that in firms that did not, a 31% increase. If a firm that had fired
-
- 21 -
a senior manager before indeed has better corporate governance, this finding is consistent with our
model for the case of large aπ (the weight on profit in the managerial evaluation system) and large
ϕ (expected cash flow). It is also possible that managers at these firing firms tend to spend
entertainment on productive uses much more than their peers at firms with poor corporate
governance. This interpretation is reinforced by our results, to be presented later, that ETC/Sales has
a significantly more positive effect on firm performance when the firm has fired a senior manager.
Thus, these findings suggest that a significant portion of entertainment expenditures are spent on the
manager’s own consumption.
Finally, firms which serve a larger market, as captured by the selling-to-other-provinces
dummy variable, tend to have higher ETC/Sales. Other things being equal, the ratio of ETC/Sales in
firms that sell to other provinces is 0.4%-05% higher than that in firms that do not. Though
statistically insignificant, a firm’s strength of private contracting relationship has a negative effect on
ETC/Sales. By Proposition 1, these findings are consistent with the view that managers use
entertainment expenditures to conduct business and to build relationships with suppliers and clients.
The results that control for city dummies are reported in column 7. Overall, the estimates of
the other coefficients remain remarkably stable. The only noticeable difference is that the coefficient
of government help becomes slightly smaller (from –0.006 to –0.005) and no longer statistically
significant, with a t-stat of 1.33. This is not surprising since we expect that government help has a
significant city dimension.32 The coefficients on the city and industry dummies are also interesting.
Cities in less advanced regions, such as Tieling, Nanning, Haerbin, Shenyang, Yingkou and Fuxin,
in general tend to have higher ETC/Sales. On the other hand, lower ETC/Sales are observed for cities
in more developed regions such as Dalian, Dalian Development Zone, Jiangmeng, Shenzhen, Wuhan,
as well as some other cities in less advanced regions such as Chaoyang, Panjin and Liaoyang,. Thus,
there are some deviations from the pattern of income level predictions. High ETC/Sales ratios are
found in accounting, auditing, and non-bank financial services, pharmaceutical and biological
products, and information technology. Low ETC/Sales ratios are found in a variety of manufacturing
industries: vehicle and vehicle part manufacturing, various types of machine manufacturing,
petrochemical product manufacturing, and consumer product manufacturing.
One may raise two concerns about the government help variable. First, it is a subjective
measure based on managers’ self reports. Second, it may be a function of how much a firm spends 32 When we regress government help on city dummies, the R squared is 0.173. Therefore, there are clearly both firm-specific and city-specific dimensions to these variables.
-
- 22 -
entertaining government officials. To address these concerns, we alternatively use a city’s economic
development level, measured by per capita GDP of the city, as a proxy for municipal government
quality. Though a crude proxy itself, the city’s per capita GDP does avoid these two problems of the
government help variable. With this alternative proxy for government quality, all the main results
(not reported here), including those on firm performance effects, are qualitatively the same.
Alternatively, one may wonder if ETC is entirely a form of implicit CEO pay for tax evasion
or incentive purposes.33 To investigate this possibility, we run the final specification in column 8
which includes two measures of CEO relative compensation. We take the ratio of CEO wages to the
firm’s average wage, and the average wage of the firm’s mid-level managers. The ratios have means
of 4.06 and 2.35, respectively. When we include these ratios in log form, ETC/Sales is not
statistically correlated with either relative wage measure. The evidence therefore does not support
the view that ETC is entirely a form of CEO pay. If it were true, we would expect a statistically
significant negative relationship between CEO wages to ETC expenditures.
To summarize, following the identification strategy developed in our model, the evidence
strongly supports the view that entertainment expenditures serve three functions: normal business
expenditures, managerial own consumption, and corruption payments to government officials.
Moreover, bribes in the form of entertainment are used both for protection against excessive
government expropriation and for grease money in exchange for better government services. In
contrast, the evidence does not support that discretional ETC expenditures are used as a substitute
mechanism for regulated CEO pay. While some of our proxies can allow other possible stories and
one could come up with alternative explanations for ETC, we believe the results present a convincing
case for our theory. This is further supported by the detailed analysis of the effects of ETC on firm
performance in the next section.
5. ETC AND FIRM PERFORMANCE In this section, we shed further light on the nature of ETC by examining ETC’s impact on a
firm’s performance. If managers use ETC mainly for normal business needs and to build
relationships with clients and suppliers, or to build relationships with government officials, the
marginal return of ETC should respond to these variables as well.
33 This explanation is not completely orthogonal to our theory, since it can be consistent with the manager’s own consumption component of ETC in our theory.
-
- 23 -
To proxy for the performance of the firm, we use the ratio of profit to sales and the firm’s
total factor productivity. Specifically, we estimate the following two equations:
(10) ( )0 1 2 3 4_ /ijc j c i i i i ijcY IND Log GDP PC Z X ETC salesβ β β β β ε= + + + + + ;
(11) ( )0 1 2 3 4
5
_ /
/ ,ijc j c i i i i
i i i ijc
Y IND Log GDP PC Z X ETC sales
X ETC sales
β β β β β
β ε
= + + + +
+ × +
where the dependent variable ijcY is either profit/sales ratio (POS) or the logarithm of total value
added, and iZ and iX are the set of firm characteristics and variables from the previous section.
When the dependent variable is log total value added, we include industry dummies interacted with
log of the capital stock and log of the number of employees in the regressions.34 This allows for
sector-specific production functions, which is necessary since our sample spans a diverse set of
sectors. In the regressions with the total value added as the dependent variable and with sector-
specific production functions, we can interpret the regression coefficient on ETC/Sales as the effect
of ETC/Sales on total factor productivity (TFP). The specification of Equation (10) estimates the
effect of ETC on firm performance after controlling for firm characteristics and the business
environment. By including interaction terms of ETC and firm specific variables iX , the specification
of Equation (11) allows for estimation of the effect of iX on the returns to ETC, because in that
specification, 4 5ijc c iY ETC Xβ β∂ ∂ = + .
For the productivity estimates, we use both the value added and the output (sales) as the
dependent variables. Using the Cobb-Douglas production function, the estimation results are
qualitatively the same with the two dependent variables. We rely mainly on the value added
estimation since it is more parsimonious and less subject to the multicollinearity problem. We allow
for sector-specific coefficients for production factors, and adding one variable would amount to
adding 14 coefficients for the Cobb-Douglas specification, and 56 coefficients for the translog
specification. We shall also report the translog specification results for the value added estimation.
34 Capital stock is represented by the original value of fixed assets. Another measure of capital stock, the net value of fixed assets, has a correlation of 0.99 with our measure. Naturally, the results do not differ depending on this measure. All nominal figures are deflated with consumer price index.
-
- 24 -
It is likely that a firm’s ETC are endogenous. For example, there may be omitted factors that
both affect a firm’s ETC and its performance. To instrument for ETC/Sales, we use the city-industry
average of ETC/Sales and the district-industry average of ETC/Sales,35 excluding the firm itself in
computing the average.36 We use both averages since firms interact with both city- and district-level
governments, and therefore both instruments contain useful information on firm-level ETC/Sales.
Practically, we use the city-industry average ETC/Sales and its square, and the district-industry
average ETC/Sales as instruments for ETC.37 This choice of instrumental variables is justified by our
theoretical model. In Section 2, we showed that a firm’s “grease money” bribery to government
officials, *gx , is a function of government service quality gk . To the extent that firms within the
same city (or district) and industry face similar qualities of government service, their expenditures *gx should be correlated. Because *gx is a component of a firm’s ETC, we thus expect that firms’
total ETC will also be correlated within the same group. This instrumental variable argument is
similar to that of Nevo (2001), which argued that regional average prices (excluding the city being
instrumented) can be used as an instrument for the city-level price because both prices respond to the
product’s common marginal cost. We find that the averages are clearly correlated with a firm’s ETC
--- in the first stage regression, the three instrumental variables are highly statistically significant, and
we will later present evidence that they appear to be strong instruments in the sense of Bound, Jager
and Baker (1995). Yet there is no strong reason why they should be correlated with other firm-
specific performance, controlling for the city development (and in some specifications, city
dummies), industry types, firm size and age, and the extent of the product market. We have also
experimented with using only the city-industry average of ETC and its square as instruments, and the
qualitative conclusions remain similar.
As can be seen for columns using the GMM specifications (columns 3 to 7 for Tables 4, 5, 6
and A1), our instruments pass a variety of specification tests. The p values for the over-identifying
restrictions test—Hansen’s J test—are all greater than 0.10. The instruments thus appear to be
orthogonal to the error term of the firm performance equations. Moreover, the instruments are
35 The district is one administrative level below the city, and the typical number of districts in a city is around 10. District government officials also have substantial discretion. 36 When we use district-industry as the unit to compute average, there are some singleton cells. In those cases we replace those cells with the city-industry average. 37 The quadratic term for the district-industry level average is not included because it does not have predictive power in the first stage regression. The remaining three terms are all statistically significant.
-
- 25 -
reasonably strong in the sense of Bound, Jager and Baker (1995). The F-test statistics for the null
that the excluded instruments are jointly equal to zero are almost always greater than 10, and the
marginal R squares are also quite reasonable relative to those in the specifications of Bound, Jager
and Baker (1995) that yield reasonable IV estimates.
The sample used consists of firm-years with no missing variables. 38 Our time-varying
variables include firm performance, firm size, firm age and tax burden. Other variables are sampled
only once so they are time-invariant. To take advantage of both types of variation, we used two
years of data for each firm. To take into account potential within-firm correlation of the error term
and to avoid overstating the precision of the estimations, we report the White-corrected standard
errors. In some specifications we also rely only on one year of observations to see if this impacts the
results. The results for Profits/Sales (POS) are reported in Table 5, and those for the Cobb-Douglas
l;production function (using log value added as the dependent variable) are in Table 6. In the case of
productivity estimation, the coefficients on the interaction terms of industry dummies with the two
production factors (log capital stock and log labor) are not reported in order to conserve space, and
their coefficients appear sensible.39 In both tables we also use industry dummies. For each
dependent variable, we present two specifications corresponding to Equations (9) and (10). In
column 1 we also present a specification that leaves out ETC/Sales with the goal of evaluating
whether estimates of other variables are significantly affected by the potentially endogenous ETC.
[Tables 4-5 About Here]
4
For the most part, the control variables iZ and iX have the expected effects on firm
performance. The regression results are reported in Table 4 on profitability and Table 5 on
productivity. Firms located in more advanced regions, those characterized by higher GDP per capita,
have higher average POS (Table 4) and much higher productivity (Table 5). Larger firms tend to
have higher POS, and older firms lower POS. Firms that sell to other provinces have higher, though
largely statistically insignificant, POS and significantly higher productivity levels. Consistent with
38 In addition, we also cleaned the data slightly. For instance, we deleted 61 observations with implausible values of Profits/Sales (POS). For the value added regressions, we dropped 1% of observations with negative or zero value added; this was necessary to take the logarithm of the value added. 39 The industry-specific coefficients of production factors tend to be significant, and most sectors exhibit constant or decreasing returns to scale.
-
- 26 -
Proposition 2 of our model, a firm’s relationships with its suppliers and clients have a strong effect
on both POS and productivity. Increasing a firm’s strength of private contracting relationship by one
standard deviation (0.33) increases POS by roughly 2 to 3 percentage points, and productivity by
0.13 to 0.14. The estimates, based on the OLS and the GMM specifications, appear to be quite
stable. This finding suggests that private contracting relationships are very valuable to firms in
China, which is likely attributable to the nation’s weak contracting enforcement.
The two corporate governance variables have some effect on firm performance, but the direct
effects appear to be at most modest. Private ownership has a strong positive effect on POS.
Increasing private ownership by one standard deviation (0.46) increases POS by 4 to 5 percentage
points, which is a fairly large effect. However, private ownership does not appear to directly affect
productivity, as the coefficient is negative but statistically insignificant. The managerial dismissal
dummy variable has a statistically insignificant effect on POS, but a robust positive effect on
productivity. Firms with a previous managerial dismissal have a productivity premium of 9 to 15
percentage points.
Interestingly, though the tax burden does not appear to have a robust relationship by itself
with POS, it has a robust and negative relationship with productivity. The tax burden is statistically
insignificant for POS (Table 4), whether or not ETC is treated as endogenous, and regardless of
whether ETC is included in the regression. For the productivity equation (Table 6), the tax burden
affects productivity negatively. Even when ETC is treated as endogenous, the tax burden remains
statistically significant, though the magnitude decreases. By the GMM estimate in column 3,
increasing the tax burden by one standard deviation (0.097) would reduce productivity by a modest
12 percentage points.
In contrast with the tax burden, ETC has robust and negative effects on both POS and
productivity, with statistically significant coefficients in three of the four specifications. For POS,
increasing ETC by one standard deviation (0.062) reduces POS by 8 percentage points for the OLS
specification. Once the endogeneity of ETC is controlled for, the corresponding effect is 11
percentage points. Similarly, increasing ETC by one standard deviation (0.062) reduces productivity
by 0.22 (0.45) based on the OLS (GMM) estimate. The negative and strong effect of ETC on firm
performance—an effect that is much more pronounced than that of the tax burden—indicates that
ETC on average contain more managerial own consumption than expenditures on building
contracting relationships with suppliers and clients or bribes to government officials. It also suggests
that in addition to the direct costs of entertainment expenditures, there are likely substantial indirect
-
- 27 -
costs. These costs likely include time and attention of top managers on wining, dining, golfing, and
the like.
Estimations of Equation (11) provide more insight on the effect of ETC on firm performance.
In columns 4 of Tables 4 and 5, we allow the effect of ETC on firm performance to depend on
corporate governance and the strength of private contracting relationship. In columns 5 of Tables 4
and 5 we further allow the effect of ETC to depend on government expropriation, and government
help. The results clearly demonstrate that the effect of ETC on firm performance hinges on these
variables from our model.40
Using the results of the GMM estimation from columns 5 in Tables 4 and 5, we can calculate
the marginal effects of ETC on POS and productivity, evaluated at the mean vector of the
explanatory variables, as follows:
4 5 1 1 5 2 2 5 3 3 5 4 4 5 4 4ˆ ˆ ˆ ˆ ˆ ˆY ETC X X X X Xβ β β β β β− − − − −∂ ∂ = + + + + +
Here Y is either POS or TFP, 4β̂ is the coefficient estimate of ETC, 5 1 5 5ˆ ˆto β β− − are the coefficient
estimates of “ETC × % private”, “ETC × fired managers”, “ETC × tax burden”, “ETC × government
help”, and “ETC × log(years know supplier/client)”, and 1 5to X X are the sample means of the
corresponding variables. Thus of the marginal effects of ETC are
4.57 21.30POS ETC TFP ETC
Table 6 calculates the marginal effects of ETC/Sales on firm performance from results in
columns 5 of Tables 4 and 5. As shown in Table 6, both of these estimates are statistically
significant. Again, there is strong evidence of a negative effect of ETC on firm performance.
Now we examine how individual environmental variables affect the relationship between
ETC and firm performance. First consider the extent of government expropriation and government
quality. The interaction term of ETC with tax burden is positive for both performance measures (see
columns 5 in Tables 4 and 5). From Table 6, the magnitudes appear to be economically significant
based on the GMM estimates. Increasing the tax burden by one standard deviation (0.097) would
40 In empirical implementation, we do not include ETC times the dummy of selling to other provinces because it is never statistically significant in any specification.
-
- 28 -
increase POS ETC∂ ∂ from –4.57 (2.37) to –4.13 (2.15), 41 or by 10 percent, and increase
TFP ETC∂ ∂ from –21.30 (5.62) to –18.30 (5.28), or by 14 percent. Thus, when government
expropriation becomes sufficiently high, ETC becomes a more attractive investment.
Furthermore, there is some evidence that the marginal effects of ETC/Sales decreases with
government quality. The coefficients from the interaction of government help with ETC/Sales are
negative and close to being statistically significant for the GMM estimation of both the POS and the
productivity equations (t=1.61 and 1.52 respectively). Reducing government help by one standard
deviation (0.33) would increase POS ETC∂ ∂ from –4.57 (2.37) to –3.20 (1.99), or by 30 percent, and
increase TFP ETC∂ ∂ from –21.30 (5.62) to –17.80 (5.31), or by 17 percent. Bad government services
make ETC less harmful to firm performance. We have also used log GDP per capita as an alternative
proxy for government service---that is, using ETC×log city GDP per capita in place of
ETC×government help---and the qualitative results remain the same. These findings are consistent
with our theoretical predictions in Section 2 and provide strong support for our findings about the
composition of ETC from the preceding section. Because the marginal returns to ETC are higher
when government expropriation is higher and when government quality is lower, managers increase
entertainment expenditures precisely in environments with higher government expropriation and
lower government service quality.
The effects of ETC on firm performance clearly depend on corporate governance. From
columns 5 in Tables 4 and 5, the interaction terms of ETC with private ownership and managerial
dismissal are all positive and are frequently statistically significant for both performance measures.
Based on the GMM estimates, the magnitudes appear to be large. If private ownership increases by
0.46 (i.e., a one standard deviation increase), POS ETC∂ ∂ would increase from –4.57 (2.37) to –4.16
(1.97), or by 9 percent; TFP ETC∂ ∂ would increase from –21.30 (5.62) to –17.68 (4.87), or by 17
percent. Similarly for a one standard deviation change in the likelihood of managerial dismissal
(0.39), POS ETC∂ ∂ would increase from –4.57 (2.37) to –3.14 (1.91), or by 31 percent;
TFP ETC∂ ∂ would increase from –21.30 (5.62) to –18.15 (4.78), or by 15 percent. These results also
lend strong support to our findings about the composition of ETC. When better corporate governance
more closely aligns managerial and firm incentives, the marginal return to ETC is higher, and
managers use entertainment expenditures more productively. This increase in productive ETC
41 Standard errors in parentheses.
-
- 29 -
apparently more than offsets the reduction in managerial consumption of ETC, and leads to an
overall increase in ETC.
When the private contracting relationship is strong, the effect of ETC on firm performance
appears to be more negative. The coefficient on the interaction of ETC with the private contracting
variable is statistically significant in both performance equations. Reducing the private contracting
variable by one standard deviation (0.333) would increase POS ETC∂ ∂ from –4.57 (2.37) to –3.63
(2.06), or by 21 percent; and increase TFP ETC∂ ∂ from –21.30 (5.62) to –17.82 (4.84), or by 16
percent. Thus, when firms already have stronger private contracting relationships, the marginal
return to ETC is smaller, leading to lower investment in ETC. This is consistent with our findings in
Section 4 and the theoretical predictions from Section 2.
Sensitivity Checks
As we mentioned earlier, the main specifications use two years of data per firm. Since we
only have one year data on our key variable, ETC, we check whether the results are sensitive to
using only the last year of the data. In Columns 6 of Tables 4 and 5, we use only the last year of data
from each firm to re-estimate specification the specification in Column 5. As we can see, the results
in the two columns are very similar, both in terms of magnitude and precision.
Another concern is that the results may be driven by city-level heterogeneity. We have
controlled for the level of development of the firm’s home city with municipal log GDP per capita.
However, there may be other city-wide factors that matter for firm performance which may cause
omitted variable bias. Column 7 addresses this concern by including city dummies in our control
variables, and the qualitative results remain the same, though the estimation errors increase
somewhat for the POS results. This is not surprising since the X variables have important city-level
variation. These results show that the effects of ETC and of interactions between ETC and X hold
even if we focus on variation within city-industry cells.
Our estimates of the productivity equation are robust with respect to the functional form of
the production function. Table 7 assumes the translog functional form—that is, we allow for
quadratic terms of production factors and their interactions, and sector-specific production
functions—and we obtain estimates related to ETC and their interaction terms with X very similar to
those in the Cobb-Douglas case. In addition, using capital, labor and intermediate inputs to explain
out