From Government to Enterprises: How Tax Sharing Interacts with Tax Rates

LyuBingyang(吕冰洋)etal.

Abstract: Tax sharing embodies central-local government fiscal relations and tax rates reflect government-market relations. Research on the interactions between tax sharing and tax rates helps uncover the effects of central-local fiscal relations on government-market relations. According to our study, China’s flexible tax sharing and differential tax rates facing firms are two important typical facts; theoretical analysis discovered that effective corporate tax rates are influenced by local government preferences and tax sharing ratio; empirical analysis found that increasing CIT and VAT sharing ratios for governments at city and county levels led to the reduction of tax evasion and increase of effective tax rates. The above conclusions have revealed the unique mechanism of how government-market relations are influenced by fiscal system, explains the sources of differential tax rates facing Chinese firms, and provides reference for next-step fiscal reform. Keywords: flexible sharing, differential tax rates, central- local fiscal relations, government and market

JEL Classification: H77, H71, H26

1. Introduction and Literature Review

In a large market economy, the government must address two economic relations: first, central- local fiscal relations and second, government-market relations. While the former is defined by the sharing of tax revenues and administrative responsibilities, the latter is defined from a fiscal perspective by the determination of tax rate and the provision of public goods. Will these two relations interact with each other? Intuitively, there must be some kind of correlation: for instance, central government may reduce the ratio of tax revenue attributable to local governments or increase their administrative responsibilities, thus putting local governments at greater fiscal pressure and

affecting their tax collection efforts. The question is the entry point for the research on central-local fiscal relations and government-market relations.

This paper believes that tax sharing and effective corporate tax rate are the key issues of China’s central- local fiscal relations and government-market relations, given the following reasons:

First, tax sharing is a typical feature of

China’s central- local fiscal relations. China’s central- local fiscal relations after reform and opening up in 1978 can be divided into two chapters of “fiscal contract” and “tax sharing” systems with the boundary in 1994. Under the fiscal contract system, local governments were required to submit tax revenue to central government according to a certain method and base amount; under the tax sharing system, the respective scopes of tax collection and ratios of tax sharing between central and local governments are predetermined. Obviously, central-local fiscal relations are underpinned by tax distribution.

When i t comes t o ce n t r a l - l o c a l tax distribution, China is different from many other countries in the following ways: first, tax sharing is the primary instrument for regulating central- local tax distribution. For instance, the central-local tax sharing ratios for value-added tax ( VAT) and corporate income tax ( CIT) as two major types of tax in China are 75:25 and 60: 40. Second, the adoption of a flexible tax sharing system: central government regulates the tax sharing ratios only between central and provincial- level governments. Within provinces, however, tax revenues collected at city and county levels are shared with provincial governments at varying ratios. From a contract perspective, Lyu and Nie ( 2014) defined the nature of overall tax sharing contract system as a “flexible sharing contract” system. Section 2 of this paper will employ data for depiction of this phenomenon. Such flexibility is inevitably embodied in regional differences, which serve as an important entry point for our study.

Furthermore, effective corporate tax rate is closely related to government tax collection. Considering the role of enterprises as the backbone of market economy, governmentmarket relations are embodied in governmententerprise relations. No matter for direct tax or indirect tax, inconsistency between effective and nominal corporate tax rates is not uncommon and has been discussed extensively in literature. As for the most important two types of tax paid by Chinese enterprises, i. e. ( corporate income tax) CIT and ( value- added tax) VAT, it has been found in quite a few literature studies that effective CIT and VAT rates are substantially below nominal tax rates and their distribution among enterprises varies greatly as well (Li and Xu, 2010; Chen, 2013) . The discrepancy derives from not only the tax evasion of enterprises (Liu and Ye, 2014.) but different practices of tax collection on the part of government as well. As a result of fierce competition for tax revenues among local governments below prefecturelevel city, effective tax rates vary greatly among adjacent districts and counties. Although China’s local governments have no legislative powers for taxation, local governments influence effective tax rates by changing their tax collection efforts and offering tax incentives in disguised forms1.

Hence, flexible tax sharing and differentiated tax rates can be seen as the two typical features of China’s central- local fiscal relations and government- market relations. This paper will investigate the correlation between these two features based on the following rationality: effective corporate tax rates affects government tax revenue and corporate development potentials; thus, when central government changes tax sharing ratios, such change will affect the fiscal interests of local governments and their enthusiasm to develop local economy, resulting in the change of local government tax collection behaviors and effective corporate tax rates.

This paper investigates how the interaction of central-local government strategies affects tax

rates. Despite a large body of literature on this topic over the past two decades, most studies have focused on horizontal and vertical centrallocal government tax competition. Horizontal tax competition will affect effective corporate tax rates because local governments competitively attract mobile tax base by adjusting effective tax rates to maximize household welfare in their respective jurisdictions ( Wildasin, 1989; Hoyt, 2001; Marceau et al., 2010). Yet another view holds that under information asymmetry, when the government of one jurisdiction cuts tax rate, politicians of an adjacent jurisdiction will follow suit under the pressures of local public opinion, i. e. “benchmark competition” of taxation ( Besley and Case, 1995; Dincer et al., 2010). Studies on vertical tax competition focus on the effects of different hierarchies of government independently exercising their respective tax rights on the same shared tax base ( Thierry et al., 2004; Alejandro et al., 2012). Yet this type of literature emphasizes that the occurrence of vertical tax competition derives from the interaction of concurrent taxes imposed by different levels of government on the same tax base rather than shared tax between different levels of government (Keen, 1998).

While relevant literature on horizontal or vertical tax competition also examined the effects of central- local government strategies on tax rates, little attention has been paid to how centrallocal tax sharing induces change in tax collection behaviors. The reasons may be twofold: first, not many countries have resorted to extensive tax sharing to adjust central- local fiscal relations. For instance, in countries like the United States, Canada and Russia, although federal and state governments impose tax on corporate income, each level of government has the power to set tax rates, i.e. concurrent tax rather than shared tax. Second, even if tax revenue is shared, the ratio of tax sharing is normally unified and stable. Obviously, data without much change cannot be used for empirical analysis. It is likely that due to the above two reasons, researchers did not spend much effort to examine the effects of tax sharing on tax rates.

In fact, incentive theory already offered relevant analysis on the mechanism of action. In their classic paper on principal-agent model, Holmstrom and Milgrom (1987) assume that the agent takes action and obtains the benefits and the principal determines the ratio for the sharing of benefits. In this manner, the ratio for the sharing of benefits will influence the behaviors of the agent. In our case, if the principal is central government, the agent is local government and the benefits are the efforts of local government in developing the economy and the resultant tax revenue, then the ratio for the sharing of tax revenue will apparently influence the tax collection behaviors of local governments.

We believe that the effect of tax sharing on tax rates offers a unique perspective for observing China’s central-local and governmentmarket relations. In On Ten Major Relations ( Mao, 1956) 2, Mao said that an important principle is to “increase the initiative of central and local governments.” If tax sharing is a leverage of central government in regulating the initiative of local governments, tax rate is an outcome in promoting such initiative. Hence, by examining the relationship between tax sharing and tax rates, this paper uncovers the root causes of problems in government-market relations and sheds light on how the initiative of central and local governments can be better engaged.

By innovatively capturing the typical facts of flexible tax sharing and differential tax rates, this paper elaborates the effects of tax sharing on tax rates from theoretical and empirical

perspectives. The rest of this paper is structured as follows: Section 2 employs data to describe the flexible tax sharing between China’s central and local governments and the differential effective corporate tax rates as two typical facts; Section 3 creates a theoretical model to illustrate the mechanism through which tax sharing affects tax rates; Section 4 and Section 5 offers empirical tests of how tax sharing affects corporate effective tax rates; Section 6 provides conclusions.

2. Two Typical Facts: Flexible Tax Sharing and Differential Tax Rates

This paper is based on the following assumptions: local government has a flexible attitude for the ratio of tax sharing and firms face differential effective tax rates. The question is: Whether are these assumptions consistent with the reality? In answering this question, we need to carry out estimation and analysis. Judging by China’s current tax system, VAT and CIT are the most important taxes in terms of tax income and happen to be subject to the central-local tax ratios of 75:25 and 60:40 respectively. Therefore, we focus our research on VAT and CIT.

2.1 Flexible Tax Sharing: Estimation and Description

As mentioned before, tax sharing ratios are differentiated among various cities and counties. Here, city and county governments include prefecture-level and county-level (including citydistrict and county-level city) governments. The ratio of tax shared by city/county governments is best measured by the amount of tax revenue received by city/ county governments divided by total local tax revenue. Regretfully, the government did not reveal the overall data of taxes collected by state tax bureaus and local tax bureaus of city/county governments (except for VAT), so the tax sharing ratios for each and every city/county cannot be calculated3. To overcome this dilemma, this paper adopts an alternative: all cities/counties of a province are regarded as a whole and the sum of CIT or VAT revenues obtained by city/ county governments of each province is divided by the tax revenue collected by tax authorities of the province to denote the ratio of tax revenue shared by city/ county governments. The equation is:

The reason for this approach is that judging by China’s political reality, the ratio of tax sharing is a key issue that affects superior/ subordinate government relations. Determination of tax sharing ratio for city/county governments is usually cascaded by provincial government4. As a result, the difference of tax sharing ratios between city/ county government and the subordinate governments within the same province would be limited. We may use VAT as an example to justify this approach. We use VAT as an example because Compendium of Fiscal Statistics for All Prefectures, Cities, and Counties provide the VAT collected by county tax authorities and the VAT retained by county governments, allowing us to conveniently calculate the ratio of VAT sharing for various counties. If the difference of VAT sharing ratios among various counties in a province is very small, the implication is that tax sharing below provincial level is mainly determined by the provincial government. First, we calculated the standard deviation of VAT sharing ratios among various county governments within each province in 2007, arriving at the mean value of 0.013 for various provinces. Among them, the standard deviation of VAT sharing ratios among county governments in the same province is smaller than 0.01 for 19 provinces.

Meanwhile, we estimated the standard deviation of VAT sharing ratios among all county governments across China in 2007 to be 0.053. Thus, intra-provincial differences may all explain 25% of differences in VAT sharing ratios among

all county governments nationwide. That is to say, if all the cities and counties of a province are regarded as a sample group, there will be 30 sample groups for China’s 31 provinces, autonomous regions and municipalities ( not including Hong Kong, Macao and Taiwan) and the differences of tax sharing ratios are relatively small within each sample group and much more significant between sample groups. Hence, the tax sharing ratios of city/ county governments may largely reflect the effective tax sharing ratios below provincial level. Of course, this approach is followed as an alternative due to limited data.

Data source: total CIT revenue and VAT revenue collected by tax authorities of various provinces ( including state and local tax authorities) are from China Tax Statistical Yearbook and the CIT and VAT revenues of city/ county governments across various provinces are from Compendium of Fiscal Statistics for All Prefectures, Cities, and Counties ( hereinafter “Compendium”). Due to the absence of CIT revenue of city/ county governments in the Compendium before 2001, the coverage of CIT sharing ratios is between 2001 and 2007, while the coverage of VAT sharing ratios is between 1998 and 2007. Considering the significant differences between tax collection and management system of the four municipalities directly under the State Council ( Beijing, Shanghai, Tianjin and Chongqing) and provinces, we deleted the data of the four municipalities in our samples5. Moreover, we also deleted Tibet due to missing data.

Figure 1 reports the provincial average of CIT sharing ratio for city/county governments. The enactment of Income Tax Sharing Scheme in 2002 slashed CIT sharing ratio at city/county level. Although the ratio of income tax sharing was 100% prior to the reform of income tax sharing, the ratio of income tax sharing for city/county governments varied greatly among provinces, ranging from the lowest 21.7% for Heilongjiang Province to the highest 72.1% for Sichuan Province. While the ratio of income tax sharing for local governments dropped to 40% after 2003, great differences still existed in the ratios of income tax sharing across provinces, ranging from the lowest 10.2% for Heilongjiang Province to the highest 31.0% for Xinjiang Uyghur Autonomous Region.

Figure 2 shows the average of VAT sharing ratio at city/ county level during 1998- 2007. Overall, the ratio has been on the decline. During this timeframe, the VAT sharing ratio remained constant for Beijing, Heilongjiang, Anhui, Fujian, Jiangxi, Henan, Hunan, Chongqing, Guizhou, Yunnan, Ningxia and Xinjiang, while

the ratio was significantly lowered in 10 other provinces, by around 10 percentage points, after 2002, including Hebei, Shanxi, Inner Mongolia, Liaoning, Hubei, Guangxi, Sichuan, Shaanxi, Gansu and Qinghai. Different VAT arrangements across provinces provide us with a great opportunity to identify how VAT sharing ratio of city/county governments affect effective corporate tax rate.

2.2 Differentiated Tax Rate: Estimation and Description

This paper employs the annual survey data of the National Bureau of Statistics (NBS) for industrial enterprises to measure effective corporate tax rate. The survey was conducted on SOEs and large non- SOEs (“large” means revenue above 5 million yuan). This paper employs the data of manufacturing sector, where subsectors are more comparable, and deleted the data of mining sector. We also deleted enterprises governed by central government (“central SOEs”) because their income tax had been collected by the central government before 2002 and the local part of their income tax has been transferred to provincial governments for mot provinces since 2002.

In this database, each enterprise is assigned a unique code together with the code of province. Using similar methods of Brandt et al. (2011), we verified company codes with reference to company name, address, postcode, telephone number and legal representative and re-matched those enterprises that remained in the market despite the change of codes and thus created the new panel data.

We also deleted those observations with the values of major indicators ( sales turnover, assets, liabilities, exports and profits, etc.) that are missing or smaller than zero. To prevent the interference of extreme values, we further deleted the observations with key variables greater than their 99th percentile and smaller than 1st percentile. To ensure comparability of values across years, we also utilized the ex-factory price index of industrial goods to convert all variables into values measured by constant price in 1998.

(1) Estimation and description of effective

corporate income tax rate

The simplest measure of effective corporate tax rate is the ratio between income tax payment and pre-tax profit. However, two problems are associated with the use of this indicator: instead of lowering applicable tax rates, tax authorities allow firms to understate pre-tax profits (Cai & Liu, 2009). If this is the case, the ratio between tax payment and pre- tax profits will seriously overestimate corporate effective tax rate. Another problem is that an outsize share of enterprises report negative profits in the China Industrial Enterprises Database; 25% of enterprises in our samples reported negative pre- tax profits. For these enterprises, the ratio between tax payment and pre- tax profits cannot be used to measure effective corporate tax burden6.

To overcome this problem, we tried to find an alternative indicator. Given the fixed statutory tax rates, enterprises had to understate their profits in order to pay less income tax. For corporate income tax, however, the complex calculation of tax base and profits creates more opportunities for firms to avoid tax. Under the assumption that corporate effective income tax rate is formed through the tax collection efforts of tax authorities, we may reference the method of Cai & Liu (2009) to measure the level of tax evasion of firms in order to reflect the effective tax rates of firms. When the tax collection efforts are insufficient, the level of corporate tax evasion will increase ( profit understatement becomes more serious) and corporate effective tax rates will fall.

Yet the amount of CIT evasion is not directly observable. Despite our knowledge of profitability reported by firms, we have no idea about their real profits. Existing literature studies often employed book income and tax income to measure the level of corporate tax evasion (e.g. Desai, 2005; Desai & Dharmapala, 2006). However, this method is applicable only to listed companies as the book income of non- listed companies is usually unavailable. For China’s non-listed companies, the level of corporate tax evasion may be estimated based on the difference between financial accounting principles and national income principles as two methods of profit calculation.

Referencing the method of Cai & Liu (2009), this paper subtracts all intermediate costs from corporate gross output based on national income accounting principles to arrive at imputed profit:

Where, Y is corporate gross output; MED is industrial intermediate input; FC is financial cost; WAGE is wage spending; DEP is current depreciation; VAT is payment of VAT. Financial accounting principles are different from national income accounting principles in many ways. In calculating profits according to accounting principles, for instance, not all current output

will go into corporate operating income. The treatment of asset depreciation is even more complex. Overhead expenses are allowed to be extracted from profits, etc. All these differences will lead to differences between the above imputed profit and real profit πit. Imputed profit does not equal real profit. Differences between imputed profit ( PRO) and reported profit ( RPRO)

are not entirely what firms have deliberately concealed.

Some differences are completely in line with regulations while others are deliberate concealments. For instance, firms may understate their business revenue and avoid stating current-year output as sales volume; firms may also overstate overhead expenses in order to understate profits; firms may also manipulate the provisioning of depreciation to understate profits. We divide the pre-tax profit of each enterprise in the industrial enterprises database by total corporate assets to arrive at RPRO. Similarly, we divided corporate customers profit by total corporate assets to arrive at PRO7.

Figure 3 shows the distribution of PRO

and RPRO in our samples. The mean value of reported profitability ( RPRO) is 0.071 and median value is 0.033. The mean value of imputed profitability ( PRO) is 0.240 and median value is 0.110. On average, reported profit is far below imputed profit. The median value of difference between imputed profit and reported profit (GAP= PRO-RPRO) is 0.055 and standard deviation is 0.171, which shows significant differences of GAP among different firms. This partially reflects significant differences of effective CIT tax rate facing enterprises. Of course, imputed profit is not completely the real profit of enterprises and it is inaccurate to use GAP as hidden corporate profit. In Section 4 of this paper, we will further reference Cai & Liu’s (2009) method to reflect the degree of corporate profit concealment by estimating the sensitivity of PRO and RPRO.

In addition, we also extended Cai & Liu’s (2009) method by simultaneously encompassing the two channels for enterprises to enjoy lower effective tax rate, i.e applicable preferential tax rate and profit understatement. Specifically, we calculated imputed CIT (IMCIT), which is the product between PRO and standard statutory tax rate ( 33%), and the real payment of CIT (RCIT) in corporate financial statements, both of which are divided by total corporate assets for standardization. The deviation between RCIT and IMCIT measures the deviation between effective tax rate and imputed tax rate to reflect the level of effective CIT tax rate. For our sample enterprises, the mean value of IMCIT is 0.079 and the mean value of RCIT is 0.011, i. e. the former is significantly greater than the latter. The above estimates will be used in the empirical analysis in the following section of this paper.

(2) Estimation and description of effective VAT tax rates

According to the practice in literature (Chen, 2013), the effective VAT tax rate ETR is defined as the ratio between corporate payment of VAT and the amount of value added. Despite the nominal VAT tax rate of 17% in China, the mean value of ETR for sample enterprises is 12.1% and standard deviation is 11.6%. Obviously, ETR is far below nominal tax rate and the difference of effective tax rates is quite significant among enterprises. According to a study of Chen (2013), 90% of the difference of ETR is caused by various enterprises within the same industry and only 10% of that difference derives from different industries.

This section reveals that flexible tax sharing and differential tax rates are indeed two typical facts that exist in China. Now, in the following sections, a theoretical and empirical analysis will be conducted to uncover their correlation.

3. Theoretical Analysis

In the classic paper of Chamley ( 1986),

interactions between government and private sector (market) are a type of Stackelberg game, i. e. the government will set the optimal tax rate based on its projection of private sector’s response. Following Chamley’s approach, this section offers a theoretical analysis under the framework of corporate cash flow maximization created by Jorgenson (1963).

3.1 Basic Model

We assume corporate production function to be as follows:

Where, K and L respectively denote capital and labor employed by an enterprise. Assuming that the government levies a tax on corporate output, nominal tax rate is t, the tax collection effort of tax authorities is , where , and the effective tax rate for the enterprise is . Such specification of tax effort seems to explain the characteristic of marginal diminishing tax revenue to tax collection efforts, e.g. with more repetitive tax audits, fewer outstanding tax obligations may still be found and collected.

Thus, the tax payable of an enterprise is as follows:

Where, the production inputs of enterprise are labor input and capital input. Assuming labor compensation is W and labor supply is L, labor input will be WL. Capital input ( investment) is I and investment with the inclusion of cost adjustment is C( I). Following the general assumptions of literature, we have .

Since corporate objective is to achieve maximal cash flow, we have the objective function:

To obtain explicit solution, in light of the feature of investment function, we make:

Assuming that capital depreciation rate is , the capital accumulation function of enterprise is as follows:

The above equation explains that corporate capital stock is affected by government efforts of tax collection.

Here, we specify two objectives of local governments: corporate output Y maximization (equivalent to emphasis on economic aggregate) and tax revenue maximization TL ( equivalent to emphasis on tax interest). Assuming that the weight assigned by local governments to the former is and the weight assigned to the latter is , local governments may determine the tax collection efforts of tax authorities, so the objective function of local governments is as

Economic implications of equation ( 17): first, when government attaches more importance to economic aggregate in its jurisdictions, it is inclined to reduce tax collection efforts; second, when government attaches more important to tax revenue, it will be inclined to increase tax collection efforts; third, when the ratio of tax sharing declines, government is inclined to reduce tax collection efforts, thus reducing the effective tax rate for enterprises. As far as the third implication is concerned, the rationale is that if the ratio of tax sharing drops, local governments will be less incentivized to grow tax revenue and collect taxes, thus resulting in the decrease of effective tax rate. As revealed by contract economics, without risk sharing and joint input, the smaller ratio of benefits the agent shares under the contract, the less effort the agent will exert.

3.2 Model Extension: Local Government Preference Is a Function of Tax Sharing

In the above analysis, local government preference is strictly exogenous of tax sharing. Yet according to experience, this assumption could be too strict because in reality, local government preference could be endogenous of tax sharing. In history, central government adjusted tax sharing ratio for many times to incentivize local governments to develop the economy8, which explains that local government preference could be affected by tax sharing. Hence, it is reasonable to assume that and are the function of s. To be cautious, we do not preset the directions of the symbols of and

.

Thus, equation (16) can be written as:

, which respectively denote the s elasticity of and , i. e. elasticity of economic preference and elasticity of fiscal preference . They reflect the degree to which local government preference changes with tax sharing ratio.

The following can be derived from equation (18): The above equation indicates that when

, we have , and vice versa. If local government preference is affected by tax sharing ratio, the effect of change in tax sharing ratio of government tax collection efforts is indefinite. If the elasticity of economic preference is smaller than the elasticity of fiscal preference, it is more likely for corporate effective tax rate to increase. The reason is simple: more tax collection helps increase fiscal revenue but will hamper corporate development, and vice versa. This tradeoff will be carefully weighed by local governments. That is to say, the effect of tax sharing on tax rate could be positive or negative.

We may reach reliable conclusions on the real impact of tax sharing on tax rate only by conducting an empirical analysis.

4. Specification of Empirical Model

In this section, our strategy for conducting empirical test is as follows: the ratio of CIT and VAT sharing for city/county governments is the key explanatory variable, the effective tax rates of CIT and VAT are explained variables and the impact of tax sharing on tax rates is revealed through regression.

4.1 Specification of Econometric Model for the Impact of CIT Sharing on Tax Rate

Given that imputed profit is not real corporate profit, it is imprecise to use the difference between imputed profit and reported profit as the hidden profit of enterprises. Hence, Cai & Liu ( 2009) estimated the sensitivity of the two as a reflection of the degree of corporate tax evasion. Considering the positive correlation between imputed profit PRO and real profit πit, we may assume that they have the following correlation:

Where, ηit is an unknown parameter, which reflects the intrinsic difference of profit accounting between national income accounting principles and financial accounting principles. ηit can be greater than zero or smaller than zero and is a general assumption. θit is the random disturbance term of expected value of 0.

We make RPROit denote reported profit, which is also positively correlated with effective profit πit, and we assume that the two are in the following relationship:

Where, dit reflects the degree of corporate tax evasion. We assume that 0< dit< 1 and smaller denotes more serious profit understatement. ζit is random disturbance term with mean value of

0. eit is intercept term, which denotes reported profit RPRO when effective corporate profit is 0, and we assume eit< 0. Given that real profit πit is unknown, we cannot directly estimate equation ( 20). But by substituting equation ( 20) into equation (21), we have:

The value of dit reflects the sensitivity between imputed profit PRO and reported profit RPRO. The smaller the sensitivity dit is, the more serious tax evasion of enterprises9. εit=ditθit+ζit is error term.

As shown in equation ( 22), based on the positive correlation between PRO and RPRO, if we wish to investigate the effect of city/ county government tax sharing ratio on effective corporate tax rate, we must examine whether PRO will come closer to RPRO when the tax sharing ratio of city/ county government increases, i.e. whether the two’s sensitivity will increase. If the sensitivity dit between PRO and RPRO increases, the implication is that the tax evasion of enterprises will fall and effective tax rate will increase. Of course, in addition to the impact of city/ county government tax sharing ratio, dit is also subject to many factors such as its own characteristics and the characteristics of its province. We assume that dit is influenced by the following factors:

Where, CITSharept is the CIT sharing ratio for city/county governments in province p; Zpt denotes provincial-level control variable; and Xit is the characteristic variable of enterprise itself. Dprovince, Dyear and Dindustry respectively denote the dummy variables of province, year and two-digit industry and βp, βγ and βj are all the coefficients of these dummy variables respectively. By substituting equation ( 23) into equation ( 22), we may arrive at the regression equation that identifies the impact of CIT sharing ratio on tax evasion. Specifically,

As the value and symbol of ηit for each firm is unknown, we cannot directly estimate the value of ηit for each enterprise in regression. But they are all included in such variables as α0 and α1. Among them, α0 is intercept term, which encompasses factors like eit, β0 and ηit, while α1 equals , i.e. the product between β1 and mean value of ηit. Error term μit includes such factors as ζit, θit and ωit. β1, β2 and β3 reflect the impact of CIT sharing ratio, province characteristic and enterprise characteristics on the sensitivity between imputed profit and reported profit. Here, we are primarily concerned with the coefficient of β1: if β1 is positive, the CIT sharing ratio for city/ county governments will increase, dit will increase and the sensitivity between reported profit and imputed profit will increase, so that there will be less tax evasion and higher effective CIT. Because CITSharept facing firms of the same province is the same, if the correlation between random disturbance term of firms in the same province is not taken into account, the standard error of coefficient could be seriously underestimated (Angrist & Pischke, 2009). Hence, we cluster standard errors at the level of province.

As for firm ch a r a c t e r i s t i c , we have controlled such variables as firm size, age, access to loans, share of expert, ownership type and the ratio between current-year sales volume and total output value. Impact of firm size on tax evasion may have two opposite directions: first, tax authorities may focus more attention on major taxpayers, so that larger firms are more likely to be caught for tax evasion, i. e. higher opportunity cost of tax evasion. On the other hand, given the economies of scale for tax evasion, it is less costly for larger firms to evade tax. In addition, larger firms are more capable to

bribe local governments and tax officials, thus more likely to evade tax. We use the logarithmic value of the number of employees as the proxy variable of firm size. Firm age is also likely to affect tax evasion. Newly established firms are more difficult to supervise and are more likely to escape the attention of tax authorities. Access to loans has a negative impact on tax evasion. Firms with easy access to bank financing face a higher opportunity cost of tax evasion because once caught for tax evasion, they may lose access to loans. Moreover, banks have very high requirements on the standardization of enterprise financial accounts, making it more difficult to evade tax (Gordon & Li, 2009). We use the ratio between interest expenses and total assets to reflect access to loans.

We used to dummy variables to control the type of firm ownership. Dsoe denotes SOEs. When the ratio of state shares exceeds 50%, the value of this variable is 1. Dfrn denotes a overseas- invested enterprise ( including enterprises with investment from Hong Kong, Macao and Taiwan). In addition, local governments may relax tax law enforcement for exporting firms in order to encourage exports. Thus, we also controlled for the share of export in the output value of a firm. We also controlled for the current-year sales revenue of a firm in its current-year total output. Because the timing for the entry of income is different under financial accounting principles and national income accounting principles, this ratio will affect the difference between imputed profit PRO and reported profit RPRO. The higher this indicator, the closer PRO is to RPRO.

Control variables at provincial level include per capita GDP, fiscal autonomy and marketization index. Among them, fiscal autonomy is the ratio between local fiscal revenue and local fiscal spending in a province. Marketization index is given by Fan et al. (2009) and other provincial-level data are from China Statistical Yearbook.

4.2 Model Specification for the Impact of VAT Sharing on Tax Rate

We use the following fixed effect regression model to examine the impact of VAT sharing ratio for city/ county governments on effective VAT tax rate:

Where, dependent variable ETRipt is effective VAT tax rate, i.e. the ratio between VAT paid by firms and value-added, and subscripts i, p and t respectively denote firm, province and year. Key explanatory variable VATSharept is the ratio of VAT sharing for city/ county government in province p. Xipt is control variable at firm level, including firm size (logarithmic value of valueadded), value-added ratio (ratio between valueadded and total output) 10, sales profitability ( ratio between profit and sales volume), age ( logarithmic value), access to loans, share of export in total output value11, whether the firm is SOE, and whether the firm is overseas-invested. Zpt is provincial-level control variable and also includes per capita GDP, marketization index and fiscal autonomy. ρi and αt are fixed effect of firm and fixed effect of year respectively. By controlling for the fixed effect of firm, we can largely control the difference of effective VAT tax rates caused by different industries and types of product. ζict is random disturbance term. We cluster standard errors at the province level.

4.3 Statistical Description of Variables

Statistical description of regression samples for CIT and VAT are shown in Tables 1 and 2, respectively.

5. Empirical Results and Analysis

5.1 Impact of CIT Sharing on Effective CIT Tax Rate

Table 3 presents our regression results of equation (24). In the interest of length, we did not report the dummy variables of province, year, industry and the estimation coefficient of their interaction term with imputed profits ( PRO).

Column 1 of Table 3 presents imputed profit ( PRO), CIT sharing ratio for city/county governments ( CITShare) and their interaction term, dummy variables of province, industry, year and their interaction term with PRO. As

can be seen from the table, the coefficient of interaction term between CITShare and PRO is significantly positive at 1% level, which indicates that in provinces with higher CIT sharing ratio for city/county governments, higher sensitivity between imputed profit and reported profit means less tax evasion of firms or higher effective tax rate.

Column 2 introduces the control variable at firm level and its interaction term with PRO. The coefficient of interaction term between CITShare and PRO is almost unchanged and still significantly negative at 1% level. Columns 3 further introduces control variables at province level and their interaction terms with PRO, and the coefficient of interaction term between

CITShare and PRO slightly increased and is still significant at 1% level. Regression coefficient 0.264 suggests that at the mean value of imputed profit 0.240, an increase of CIT sharing ratio for city/county governments by 10 percentage points will lead to an increase of reported profit ( RPRO) by 0.63 percentage points. The implication is that for firms with reported profit ( RPRO) at the mean value ( 0.071), it will report 8.9% more profits, i.e. payment of income tax will increase by 8.9%. Column 5 further controls for the fixed effect of firm and obviously, despite a slightly decreased coefficient of PRO×CITShare, it is still statistically significant at 5% level.

The coefficients of other control variables are also consistent with our assumptions. The interaction term between firm access to loans and PRO is significantly positive, which means that the easier it is for firms to obtain bank loans, the less likely they will evade tax. The coefficient of interaction term between firm age and PRO is significantly positive, i. e. newly established firms are more likely to evade tax and vice versa. The interaction term between firm size and PRO is positive: a like explanation is that the larger a firm is, the more likely it is to be subjected to tighter regulation, thus less tax evasion. The interaction term between the ratio of sales revenue to total output value and PRO is also positive with significance level of 15% in Columns 2-3 and 1% in Column 4, which verify our prior assumption. Compared with private firms, profit understatement is much more serious for SOEs. A possible reason is that for SOEs with more powerful political ties, tax inspections are not as strong as they should be. The survival of SOEs, which assume more policy burdens, is the responsibility of local governments. Supervision on SOE tax evasion is thus less severe. There is not much difference in the profit understatement between overseas-funded firms and private firms. For export-dependent firms, however, their tax evasion is more serious, possibly due to the policy preferences granted by local governments to encourage export.

As for control variables at province level, the level of economic development and marketbased operation in a province has a limited impact on the tax evasion of firms. In provinces with higher fiscal autonomy, tax evasion of firms is less severe. A possible explanation is that if the fiscal spending of a province is more dependent on the transfer payment of central government (i.e., less fiscally dependent) rather than its own

fiscal revenue, the tax collection effort of local government will reduce.

The above empirical analysis adopts the sensitivity between reported profit and imputed profit as the measurement of effective corporate tax rate. In the robustness test, we also extended the method of Cai & Liu ( 2009), using the sensitivity between real payment of CIT (RCIT) and imputed CIT (IMCIT) as the measurement of effective tax rate. We replaced the imputed profit ( PRO) in regression equation ( 24) with imputed CIT ( IMCIT), which equals the product between the imputed profit and standard statutory tax rate (33%). We replaced reported profit ( RPRO) with real payment of CIT (RCIT) for a similar regression with other variables held constant. The deviation of RCIT from IMCIT not only takes into account the degree of profit understatement but the level of tax preference granted by tax authorities and thus offers a more complete reflection of the deviation of real tax rate from nominal tax burden. This regression result is reported in Table 4. The coefficient of interaction term between IMCIT and CITShare is significantly positive. The implication is that with rising tax sharing ratio for city/ county governments, the sensitivity between nominal tax burden and effective tax rate will increase, i.e. higher effective tax rate12.

Then, we in v e s t i g a t e d whether ta x sharing has a heterogeneous effect on different types of firms. First, we examine whether an increase in the tax sharing ratio for city/county governments has a heterogeneous effect on the tax evasion of firms of different size. For this purpose, we constructed a dummy variable and if the number of firm employees is greater than the median value of samples, the value of this dummy variable is 1. We put Large, and Large*CITShare and Large*CITShare*PRO into the regression equation, with results reported in the first column of Table 5. It can be seen that the coefficient of Large* CITShare* PRO is insignificant. This implies that when the tax sharing ratio of city/ county governments increases, the impact on large firms and SMEs is equal.

Then, we also examine firms of which type of ownership will reduce their tax evasion the most when CIT sharing ratio for city/ county governments increases. For this purpose, we put the dummy variable of SOE, SOE* CITShare and SOE* CITShare* PRO into the regression equation with results reported in Column 2 of Table 5; put dummy variable of overseas- funded firms FOE, FOE* CITShare and FOE* CITShare* PRO into the regression equation, with results shown in the Column

3 of Table 5. The regression result shows that the coefficient of SOE* CITShare* PRO isinsignificant but the coefficient of FOE*CITShare*PRO is significantly negative. This means that when CIT sharing ratio for city/ county governments increases, the tax evasion of private firms, SOEs and overseas-funded firms will significantly reduce. The magnitude of such reduction is not much different between private firms and SOEs but much more significant for overseas-funded firms. A possible reason is that overseas capital is an important target of local government competition. With increasing tax sharing ratio for city/county governments, city/ county governments will lose motivation to engage in tax rate competition.

Lastly, local tax bureaus are managed by local governments, while state tax bureaus are under vertical management by central government. It is thus believed that local governments will exert a greater degree of tax law enforcement on local tax bureaus. As discovered by Fan and Tian ( 2013), the CIT of firms established after 2002 is collected by state tax bureaus and the degree of tax evasion for firms established after 2002 is smaller than those established before 2002. Because local governments may not intervene in the tax law enforcement of state tax bureaus, effective tax rate of CIT collected by state tax bureaus is not as sensitive as the CIT burden of firms under local tax bureaus is to change in tax sharing ratio for city/county governments.

In order to test such a possibility, this paper has created a dummy variable New. For firms established before 2002, this dummy variable is defined to be 1. Then, we put virtual variable New, New*CITShare and New*CITShare*PRO into the regression equation ( 24) and define samples to be Chinese-funded firms. We deleted overseas- funded firms because overseasfunded firms established around 2002 are subject to tax collection by state tax bureaus. Regression results are shown in Column 4 of Table 5. Obviously, the coefficient of New* CITShare* PRO is not significant. This implies that for the two groups of firms established before and after 2002, raising CIT sharing ratio for their local city/ county governments would significantly reduce the level of tax evasion for both groups of firms. It also implies that even for firms under state tax bureaus, when the CIT sharing ratio for city/ county governments increases, the state tax bureaus will also enhance tax collection and management13.

5.2 Impact of VAT Sharing Ratio on Effective VAT Tax Rate

Table 6 reports the impact of city/ county VAT sharing ratio on VAT tax burden. Column 1 only controls for the fixed effect of firm and year; Column 2 additionally controls for the control variable at firm level; Column 3 additionally controls for the control variable at province level. As can be seen from the table, the coefficient of VAT sharing ratio for city/ county governments is statistically significantly positive. Take the coefficient of Column 3 for instance, when city/ county VAT sharing ratio increases by 10 percentage points, effective VAT tax rate will rise by 1.74 percentage points.

In respect of control variables, determinants of effective VAT tax rates are not entirely consistent with the determinants of CIT evasion. Firms with greater industrial value- added and value-added ratio are subject to lower effective VAT tax rates. This implies the tax preferences are given to large firms and firms that create more value- added. More profitable firms are subject to higher effective VAT tax rates, i. e. profitable firms not only paid more CIT but

paid more VAT as well. Older firms also need to pay more VAT. Compared with private firms, overseas- funded firms are subject to lower effective VAT tax rates but SOEs are subject to higher effective tax rates. Firms with a high percentage of export have also enjoyed relatively low effective VAT tax rates. Levels of economic development, fiscal autonomy and marketization at province level have no significant impact on corporate VAT tax rates14.

The above results indicate that for China as a whole, increasing VAT revenue sharing for city/ county governments will significantly increase their efforts to collect VAT from firms. However, are there any differences for various types of firms? On the basis of regression equation ( 25), we have introduced the interaction terms between VAT sharing ratio for city/ county governments VATSharept and firm size and ownership type respectively, with results shown in Table 7. As can be seen from Column 1, there is no significant heterogeneity between large and small firms. As shown in Columns 2-3, with rising VAT sharing ratio for city/ county governments, the effective VAT tax rates for private firms, SOEs and overseasfunded firms all increase significantly. There is not much difference in the degree of increase between SOEs and private firms but the effective tax rate for overseas-funded firms will increases by a larger margin. This discovery coincides with the foregoing discovery on CIT.

6. Concluding Remarks

By investigating the relationship between tax sharing, a key issue of central- local fiscal relations, and tax rate, a key issue of government-market relations, this paper observes how China’s central- local government fiscal relations affect government-market relations and has arrived at the following conclusions:

First, flexible tax sharing and differentiated tax rate in China are two important typical facts. Despite stable ratios for the sharing of central tax, local tax and shared tax, tax distribution of city/county government is subject to provincial government intervention and is thus different from the sharing ratios for important types of tax. Despite the nationally unified nominal tax rates, local governments cannot adjust tax rates but may adjust the level of their tax collection efforts, which means that effective tax rates facing firms vary greatly.

Second, theoretical analysis has discovered that tax sharing will affect tax rate. With constant preference of local government for economic growth and the scale of tax revenue, when tax sharing ratio declines, the government will reduce tax collection effort, thus causing effective tax rate to reduce. If local government preference changes with the ratio of tax sharing, the effect of tax sharing on tax rate is uncertain and subject to the tax sharing elasticity of various preferences.

Third, our empirical analysis has discovered that the increase of tax sharing ratio will induce tax rate to rise. When CIT sharing ratio for city/ county income tax increases, the tax evasion of SOEs, private firms and overseas-funded firms will reduce significantly and such reduction is even greater for overseas-funded firms. As far as VAT is concerned, with rising city/ county VAT sharing ratio, effective VAT tax rates will significantly increase for SOEs, private firms and overseas- funded firms and the increase is particularly significant for overseas- funded firms.

This paper explains the sources of differential tax rates from the perspective of tax sharing. We know that differential effective tax rates will impede fair competition among firms and is unfavorable to unified tax collection and the formation of a unified market. China’s current rapidly advancing “business tax to VAT” reform will tighten local government finance. Without effective improvement of tax system, central government may adopt more extensive tax sharing as the last resort, resulting in more significant differences in tax sharing ratios facing city/county