Competition and Corporate Tax Avoidance: Evidence from Chinese Industrial Firms



Similar documents
FOREIGN TAXES AND THE GROWING SHARE OF U.S. MULTINATIONAL COMPANY INCOME ABROAD: PROFITS, NOT SALES, ARE BEING GLOBALIZED.

Do R&D or Capital Expenditures Impact Wage Inequality? Evidence from the IT Industry in Taiwan ROC

Yao Zheng University of New Orleans. Eric Osmer University of New Orleans

Social Security Eligibility and the Labor Supply of Elderly Immigrants. George J. Borjas Harvard University and National Bureau of Economic Research

Investment and financing constraints in China: does working capital management make a difference?

The Determinants and the Value of Cash Holdings: Evidence. from French firms

Journal of Comparative Economics

Abnormal Audit Fees and Audit Opinion Further Evidence from China s Capital Market

The (mis)allocation of capital

FDI as a source of finance in imperfect capital markets Firm-Level Evidence from Argentina

DOES A STRONG DOLLAR INCREASE DEMAND FOR BOTH DOMESTIC AND IMPORTED GOODS? John J. Heim, Rensselaer Polytechnic Institute, Troy, New York, USA

Investment and financing constraints in China: does working capital management make a difference?

ANNUITY LAPSE RATE MODELING: TOBIT OR NOT TOBIT? 1. INTRODUCTION

On the Growth Effect of Stock Market Liberalizations

An Empirical Analysis of Insider Rates vs. Outsider Rates in Bank Lending

TRADE CREDIT AND CREDIT CRUNCHES: EVIDENCE FOR SPANISH FIRMS FROM THE GLOBAL BANKING CRISIS

Export Pricing and Credit Constraints: Theory and Evidence from Greek Firms. Online Data Appendix (not intended for publication) Elias Dinopoulos

A.2 The Prevalence of Transfer Pricing in International Trade

Understanding the Risk of China s Local Government Debts and Its Linkage with Property Markets

The Elasticity of Taxable Income: A Non-Technical Summary

Gao Peiyong* * Gao Peiyong, Professor, Renmin University, Beijing, China. gaopy@263.net.

Internal finance and growth: microeconometric evidence on Chinese firms

Do Direct Stock Market Investments Outperform Mutual Funds? A Study of Finnish Retail Investors and Mutual Funds 1

Institutional Trading, Brokerage Commissions, and Information Production around Stock Splits

Markups and Firm-Level Export Status: Appendix

ONLINE APPENDIX TO Bridging the gap: the design of bank loan contracts and distance

An Analysis of the Telecommunications Business in China by Linear Regression

Moral Hazard. Itay Goldstein. Wharton School, University of Pennsylvania

Are Chinese Growth and Inflation Too Smooth? Evidence from Engel Curves

Short sales constraints and stock price behavior: evidence from the Taiwan Stock Exchange

Efficiency Analysis of Life Insurance Companies in Thailand

Measuring BDC s impact on its clients

The Impact of Interest Rate Shocks on the Performance of the Banking Sector

Autoria: Eduardo Kazuo Kayo, Douglas Dias Bastos

CAPITAL STRUCTURE AND DEBT MATURITY CHOICES FOR SOUTH AFRICAN FIRMS: EVIDENCE FROM A HIGHLY VARIABLE ECONOMIC ENVIRONMENT

How To Model Corporate Investment In China

Determinants of Capital Structure in Developing Countries

FORECASTING DEPOSIT GROWTH: Forecasting BIF and SAIF Assessable and Insured Deposits

Panel Data. Tomohito Hata. In the present paper, I will examine the determinants of grant-loan allocation in Japanese ODA,

DIVIDEND POLICY, TRADING CHARACTERISTICS AND SHARE PRICES: EMPIRICAL EVIDENCE FROM EGYPTIAN FIRMS

LAW & PROJECT FINANCE

TAX EVASION AND SELF-EMPLOYMENT OVER TIME: EVIDENCE FROM SWEDEN

Insurance Markets in China 1

TERMS OF LENDING FOR SMALL BUSINESS LINES OF CREDIT: THE ROLE OF LOAN GUARANTEES

The National Accounts and the Public Sector by Casey B. Mulligan Fall 2010

China s Economic Reforms and Growth Prospects. Nicholas Lardy. Anthony M Solomon Senior Fellow. Peterson Institute for International Economics

Small Bank Comparative Advantages in Alleviating Financial Constraints and Providing Liquidity Insurance over Time

This PDF is a selection from a published volume from the National Bureau of Economic Research

DETERMINANTS OF INSURANCE COMPANIES PROFITABILITY: AN ANALYSIS OF INSURANCE SECTOR OF PAKISTAN

A Review of Cross Sectional Regression for Financial Data You should already know this material from previous study

China Tax Monthly. 1. Recent anti-avoidance cases in China. Beijing/Hong Kong/Shanghai. a. The Shanxi Permanent Establishment ( PE ) Case

Capital Structure and Taxes: What Happens When You (Also) Subsidize Equity?

Working Capital, Financing Constraints and Firm Financial Performance in GCC Countries

WORKING DRAFT. Chapter 5 - Transfer Pricing Methods (Transactional Profit Methods) 1. Introduction

Kauffman Dissertation Executive Summary

Inflation. Chapter Money Supply and Demand

NCSS Statistical Software Principal Components Regression. In ordinary least squares, the regression coefficients are estimated using the formula ( )

Paul Brockman Xiumin Martin Emre Unlu

Case Study of Unemployment Insurance Reform in North Carolina

The Bank of Canada s Senior Loan Officer Survey

Gains from Trade: The Role of Composition

Earnings Announcement and Abnormal Return of S&P 500 Companies. Luke Qiu Washington University in St. Louis Economics Department Honors Thesis

A GENERALIZATION OF THE PFÄHLER-LAMBERT DECOMPOSITION. Jorge Onrubia Universidad Complutense. Fidel Picos REDE-Universidade de Vigo

Trade, Finance, Specialization and Synchronization: Sensitivity Analysis

The Impact of Bank Consolidation on Small Business Credit Availability

Introduction to time series analysis

CHAPTER 11: The Problem of Global Inequality

ACHIEVABLE CORPORATE TAX REFORM 2013 PETERSON INSTITUTE FOR INTERNATIONAL ECONOMICS DECEMBER 12, 2012

ECON20310 LECTURE SYNOPSIS REAL BUSINESS CYCLE

On the Dual Effect of Bankruptcy

Determinants of Non Performing Loans: Case of US Banking Sector

ESTIMATING AN ECONOMIC MODEL OF CRIME USING PANEL DATA FROM NORTH CAROLINA BADI H. BALTAGI*

Asian Economic and Financial Review THE CAPITAL INVESTMENT INCREASES AND STOCK RETURNS

This PDF is a selection from an out-of-print volume from the National Bureau of Economic Research

DETERMINANTS OF THE CAPITAL STRUCTURE: EMPIRICAL STUDY FROM THE KOREAN MARKET

Economic Analysis on Development of Marine Insurance in Shanghai. 1 Introduction

How Do Small Businesses Finance their Growth Opportunities? The Case of Recovery from the Lost Decade in Japan

Factors affecting the inbound tourism sector. - the impact and implications of the Australian dollar

Firm and Product Life Cycles and Firm Survival

Chapter 5: Analysis of The National Education Longitudinal Study (NELS:88)

Stock market booms and real economic activity: Is this time different?

Constructing a Basefile for Simulating Kunming s Medical Insurance Scheme of Urban Employees

Introduction to Regression and Data Analysis

How To Calculate Export Earnings

Corporate Income Taxation

Response to Critiques of Mortgage Discrimination and FHA Loan Performance

THE IMPACT OF MACROECONOMIC FACTORS ON NON-PERFORMING LOANS IN THE REPUBLIC OF MOLDOVA

Taxation of Shareholder Income and the Cost of Capital in a Small Open Economy

INTERNAL DEBT AND MULTINATIONALS' PROFIT SHIFTING EMPIRICAL EVIDENCE FROM FIRM-LEVEL PANEL DATA. Thiess Buettner and Georg Wamser

Do broker/analyst conflicts matter? Detecting evidence from internet trading platforms

1 Determinants of small business default

Causes of Inflation in the Iranian Economy

Multinational Firms, FDI Flows and Imperfect Capital Markets

National Small Business Network

U.S. DEPARTMENT OF THE TREASURY

Why a Floating Exchange Rate Regime Makes Sense for Canada

MACROECONOMIC ANALYSIS OF VARIOUS PROPOSALS TO PROVIDE $500 BILLION IN TAX RELIEF

READING 11: TAXES AND PRIVATE WEALTH MANAGEMENT IN A GLOBAL CONTEXT

Integrating Financial Statement Modeling and Sales Forecasting

Transcription:

Competition and Corporate Tax Avoidance: Evidence from Chinese Industrial Firms Hongbin Cai Qiao Liu Abstract This article investigates whether market competition enhances the incentives of Chinese industrial firms to avoid corporate income tax. We estimate the effects of competition on the relationship between firms reported accounting profits and their imputed profits based on the national income account. To cope with measurement errors and potential endogeneity, we use instrumental variables, exogenous policy shocks and other robustness analysis. We find robust and consistent evidence that firms in more competitive environments engage in more tax avoidance activities. Moreover, all else equal, firms in relatively disadvantageous positions demonstrate stronger incentives to avoid corporate income tax. JEL Classification: L10, D21, H26, G30 Keywords: Competition, firm behavior, tax avoidance and evasion, Chinese economy Guanghua School of Management and IEPR, Peking University, Beijing, China 100871. Contact: (86)10-6276-5132, hbcai@gsm.pku.edu.cn School of Economics and Finance, Faculty of Business and Economics, University of Hong Kong, Pokfulam, Hong Kong. Contact: (852)2859-1059, qliu@hku.hk. 1

Recently tax avoidance and evasion has received increasing attentions both practically and in academic research. 1 The literature has come to view tax avoidance and evasion as an important corporate strategy and has examined various determinants affecting tax avoidance activities (see, e.g., Slemrod, 2004; Desai and Dharmapala, 2006). However, how industrial characteristics such as competitive environment affect firms incentives to engage in tax avoidance activities has not been analyzed. Moreover, almost all empirical works in this literature focus on the U.S. experiences (for a notable exception, see Sivadasan and Slemrod, 2006, on India). Using a large dataset of Chinese industrial firms, we examine empirically how product market competition affects firms incentives to avoid corporate income tax. We study corporate tax saving behavior of Chinese firms, because it is an important economic phenomenon in an increasingly important economy in the world. 2 For example, the Chinese National Auditing Office uncovered 11.89 billion Chinese yuan or RMB (about $1.6 billion) in 2003 based on a four-month, nationwide investigation of 788 companies selected at random in 17 provinces and cities (The Asian Wall Street Journal, A2, September 20, 2004). Fisman and Wei (2004) also identify evidence of pervasive tariff evasion in China. It is safe to say that these cases of uncovered tax evasion represent only a tiny fraction of tax avoidance and evasion by Chinese firms. Intuitively, firms under greater competition pressure are more motivated to avoid tax so as to have more investment money to compete in the market place. Thus, firms in more competitive industries and firms in relatively disadvantageous positions within an industry should have stronger incentives to avoid tax. In an earlier working paper (Cai and Liu, 2007), we demonstrate these results in a simple theoretical model. In this paper we focus on testing these conjectures empirically. The dataset we use is maintained by the National Bureau of Statistics of China (NBS) and contains firm-level information based on the annual accounting briefing reports filed by all above scale industrial firms in China from 2000 to 2005. On average we have about 190,000 firms per year in our sample. However, a main challenge for our empirical analysis, which is common to the literature on tax avoidance and evasion, is that firms true accounting profits 1

are not observable. To overcome this difficulty, scholars often use book incomes as a proxy for true profits and use the book-tax gap as a measure of tax avoidance, e.g., Desai (2003; 2006) and Desai and Dharmapala (2006). 3 But this approach works only for public companies since book incomes for non-listed companies are usually unavailable. Using a similar approach, we calculate an imputed corporate profit based on the national income account that is, by deducting intermediate inputs from gross output. For many reasons, this imputed corporate profit can legitimately differ from a firm s true accounting profit based on the General Accepted Accounting Principles (GAAP). A main reason is the differences in the revenue and expense recognition rules of the two systems, for example, not all gross output in the current year necessarily converts into firm revenue in the same year. Asset depreciation rules can also be different. Tax credits (such as earning-reinvestment credits) and tax loss carryovers can be other sources of the gap between the imputed profit and true accounting profit. For these reasons, the imputed corporate profit PRO based on the national income account is probably not a good proxy for true accounting profit, and certainly not as good as book income. Thus, using the gap between the imputed profit and the reported accounting profit as a measure of tax avoidance is not appropriate in our context. However, for our purposes, we only need to assume that the imputed profit and the true accounting profit are positively correlated, which is likely to hold since both reflect a firm s economic fundamentals. The reason is that our theoretical predictions are mainly concerned with comparative statics results about the sensitivity of reported profits to true accounting profits. As long as the imputed profit and the true accounting profit are positively correlated, our comparative statics results will carry over to the sensitivity of the reported profit to the imputed profit. Therefore, our empirical strategy is to test hypotheses regarding how competition (and other variables of interest) affects the sensitivity of the reported profits to the imputed profits. Our theoretical predictions are all confirmed by our empirical results. Specifically, we find strong evidence indicating that competition in the product market enhances firms incentives to engage in tax avoidance activities. The estimated effect on profit under-reporting has the predicted sign and is statistically significant for several measures of competition (the 2

number of firms, concentration, or industry average profit margin) and alternative definitions of industries (3-digit or 4-digit industry codes) and markets (national or regional). Our main empirical results are robust to alternative specifications. Besides OLS regressions, we run 2SLS regressions instrumenting for both the imputed profit and competition. In particular, we use the number of permissions required for establishing a new firm in a four-digit industry as the instrument for competition, and the average imputed profit at the four-digit industry level (excluding the firm itself) as the instrument for the imputed profit. In all model specifications, we find supporting empirical evidence. 4 In addition, we investigate a natural experiment on competitive environment (lifting of restrictions on foreign investment) for two industries. All these investigations yield the same result that competition encourages under-reporting profits by firms. This means that even though endogeneity (such as competition) and measurement errors (such as the imputed profit) pose potentially serious econometric issues in our estimations, they are unlikely to be the driving forces of our main results. The competition effect is also economically significant. Take the 2SLS regression in column (4) of Table 4 as an example. When all independent variables take their mean values, the reported profit will increase by 0.504 if the imputed profit increases by 1. If the competition measure used in the regression (industry average profit margin in this case) increases from its mean by one standard deviation (i.e., competition is less intensive by one standard deviation), then the responsiveness of the reported profit to the imputed profit increases to 0.584, representing a 15.7% increase from its previous level. In other words, a representative firm in an industry that is one standard deviation less competitive than the average industry reports about 16% more profit for each unit of imputed profit than an identical firm in the industry with the average degree of competition. Our analysis also yields useful results regarding other factors that may affect firms incentives to engage in tax avoidance activities. After controlling for other characteristics, firms facing higher tax rates or tighter financial constraints and smaller firms report less profits for each unit of imputed profit. These are all consistent with our theoretical predictions. The estimated effects of these factors have the predicted signs and are statistically and economi- 3

cally significant. Based on the 2SLS estimation of the baseline model (e.g., see model (4) in Table 4), all else equal, for each unit of imputed profit, a one-standard-deviation increase in tax rate reduces a firm s reported profit by 5.6% from its mean level; a one-standard-deviation increase in our measure of accessibility to capita markets leads to a 2.1% increase in the firm s reported profits; and lastly, a one-standard-deviation increase in firm employment size causes the firm s reported profits to increase by 27.4%. Our paper builds on and contributes to the aforementioned growing literature on corporate tax avoidance and evasion. Our main contribution is to present systematic evidence from China that product market competition increases tax avoidance. Insofar as some of the tax avoidance and evasion is illegal or socially undesirable, our paper is also related to papers that point out the dark side of competition, e.g., Cummins and Nyman (2005) and Shleifer (2004). In particular, Shleifer (2004) argues that competition encourages the spread of a wide range of unethical behavior such as employment of child labor, corruption, excessive executive pay, and corporate earnings manipulation. The basic idea is simple: the effects of competition critically depend on the instruments firms use to compete in the product markets. If firms use socially unproductive means to gain competitive advantage, then competition may not lead to socially desirable outcomes. In this regard, ours is the first paper to our knowledge that provides systematic empirical evidence consistent with that claim. The rest of the paper proceeds as follows. Section 1 describes briefly the institutional background of corporate taxation in China. Then Section 2 develops theoretical conjectures and empirical hypotheses. We describe the dataset and our empirical strategy in Section 3, and then present the main empirical results in Section 4. Section 5 examines robustness of our empirical analysis. Concluding remarks are in Section 6. 1 Institutional Background In the central planning system before economic reforms started in 1978, Chinese industrial firms were mostly state-owned (except small collective firms). They were not independent accounting units and must send all surpluses to the government agencies controlling them. 4

There was no corporate income tax in the planning system (for a description of China s economic reforms, see Wu, 2003). Starting in 1979 and continuing in the 80 s, the Chinese government introduced a number of enterprise taxation reforms. Large state-owned enterprises (SOEs hereafter) were subject to 55% income tax, small SOEs were subject to a 10 55% tax schedule, and private firms s income tax rate was set at 35%. For large SOEs, there were also complicated and evolving sharing formula to divide the after tax profits between SOEs and government. SOE reforms focused on delegating decision power and giving incentives (fang quan yang li) in the 80 s, and then continued in the 90 s with modern corporation system emphasizing corporatization and governance. Many small and medium sized SOEs were quietly privatized, others were transformed into corporations. At the same time, non-state firms (private, foreign, Hong Kong/Taiwan) grew fast and became more and more important in the economy. As a part of the systematic fiscal and taxation reforms, in 1994 China enacted the Corporate Income Tax Code that overhauled corporate taxation. All domestic firms, independent of ownership types, pay 33% corporate income tax, except that small firms with taxable income less than 30,000 RMB pay 18% income tax. Since the reforms, SOEs have been allowed to keep the after-tax profits for re-investing and for covering reform costs (e.g., paying for layoff workers). 5 Most foreign invested industrial firms pay 15% corporate income tax. The Code allowed various exemptions, tax credits (e.g., recycling, environmental protection, retained earning reinvestment) and reductions (e.g., to high tech firms, new firms in poor areas), and foreign investment firms in general enjoy much more favorable treatments (Cui, 2005). The tax collection agencies were also reformed in 1994. Before 1994, provincial and local tax collectors collected all taxes, and then the central government and provincial governments divided the total tax revenue according to some previously negotiated sharing formula. In the 1994 reforms, taxes are classified into central and local taxes, and a National Taxation bureau (Guoshuiju) and provincial bureaus (Dishuiju) are responsible for collecting central taxes and local taxes separately. Both the National Taxation Bureau and provincial bureaus are under the supervision of the State Administration of Taxation. Right now, corporate income tax is classified as a central tax, thus is collected by the National Taxation Bureau and its 5

branches in all provinces. 6 Along with impressive economic growth in the last three decades, the economic landscape in China has changed dramatically over the reform years. Starting from a central planning economy, the Chinese economy has become a mixed economy in which the non-state sector plays an increasingly dominant role. Among all above scale industrial firms in our dataset, in 2005 the SOEs and collective firms account for only 10.2% and 15.1% respectively, mixedownership firms for 21.4%, and private Firms (domestic, foreign, and Hong Kong and Taiwan) account for 53.3%. Moreover, through many years of reforms, managers of the remaining SOEs and collective firms are now given considerable decision power and relatively strong incentives that tie their compensations with firm performance, giving rise to a common concern about insider control in these firms (Wu, 2003). The high economic growth and increasing tax collection efforts together produce impressive growth of corporate income tax revenue for the Chinese government since the reforms in 1994. In 2001, the total revenue from corporate income tax was 263 billion RMB, and in 2005 it increased to 534.4 billion RMB (China Statistical Yearbook, 2006). 7 However, enforcement and collection of corporate income tax are still considered rather weak. For example, according to one report by a city branch of the National Taxation Bureau, there are three main problems: (1) insufficient manpower in the collection agency to deal with the increasing number of firms; (2) lack of training and skills in the collection agency to collect corporate income tax (which is much more complicated than other taxes such as VAT); and (3) ineffective management system in the collection agency. 8 In each city, tax collection offices usually identify large firms as important tax targets (zhongdianhu) and each tax official is assigned to deal with a certain number of these firms. Except this, tax collection officials do not appear to have effective systematic ways of auditing and checking firms. For example, sampling tax evasion cases reported in the industry magazine China Taxation, one finds that a vast majority of the cases were reported by insiders to the tax authority. Given the weak enforcement of corporate income tax, one would expect tax avoidance and evasion to be quite pervasive. Even though there is no systematic study, that is quite evident from anecdotal evidence and numerous cases reported in the media. Among the 6

many avoidance and evasion methods, the following seem to be among the most common: (a) mis-recording sales revenue (e.g., under accounts receivable ); (b) abusing tax credits (e.g., claiming recycling materials); (c) transfer prices with affiliated firms; (d) earning management (e.g., to smooth profit and loss); and (e) fake receipts. 9 Moreover, there are also numerous books of taxation planning that teach how to minimize corporate tax legally (e.g., Cui, 2005). The large demand of such knowledge indicates that firms do pay close attention to tax efficiency strategies. 2 Theoretical Conjectures, Empirical Methodology and Testable Hypotheses Let π i,t be firm i s true accounting profit in year t. We postulate that it reports a profit of ˆπ i,t = d i,t π i,t + e i,t + ζ i,t (1) where ˆπ i,t is the profit firm i reports, d i,t < 1 and e i,t 0 are two parameters, and ζ i,t is a mean zero error term. This means that firms under-report their profits. Clearly, the amount of profit under-reported, π i,t ˆπ i,t, is decreasing in d i,t and e i,t. In other words, when d i,t and e i,t are greater, firm i reports more truthfully. In an earlier working paper (Cai and Liu, 2007), we present a simple theoretical model that yields a profit-reporting decision rule as in Equation (1). We also show that d i,t and e i,t depend on market competition and other firm characteristics. Intuitively, firms under greater competitive pressure are more motivated to avoid tax so as to have more investment money to compete in the market place. Thus, firms in more competitive industries and firms in relatively disadvantageous positions within an industry should have stronger incentives to avoid tax. Our theoretical predictions can be summarized into the following two conjectures (for details, see Cai and Liu, 2007): Conjecture 1 All else equal, profit under-reporting becomes more severe in more competitive environments. 7

Conjecture 2 All else equal, profit under-reporting becomes more severe when tax rates or marginal returns of capital are larger, or when the cost of tax avoidance is smaller. Intuitively, with higher tax rates, one yuan of un-reported profit saves more tax, hence profit under-reporting is more profitable. With higher marginal returns of capital, one yuan of saved tax will generate more future profit. In either case, firms will tend to report less profits. On the other hand, if the cost of tax avoidance is higher, firms will report more truthfully. Note that we focus on the effect of competition on tax avoidance and do not explicitly consider the possible agency problems in the tax avoidance decisions. Recently the effects of managerial incentives on firms tax avoidance have been emphasized by several scholars (Crocker and Slemrod, 2005; Chen and Chu, 2005; Desai and Dharamapala, 2006). These studies clearly show that agency issues can have important implications for firms tax avoidance behavior. In this paper, as long as managers partially care about firm value, our qualitative results (Conjectures 1 and 2) should still hold. Specifically, if managers payoff functions are increasing in firm value, competition pressure will force managers to engage in more tax avoidance activities. Thus, to focus on the competition effect and also due to data limitation, we abstract from agency issues in this paper. See Section 5.4 for more discussion on this. Since π i,t is not observable, we cannot directly estimate Equation (1). To overcome this difficulty, we adopt the following approach. Using the NBS database, which we will detail in Section 3, we compute firm i s corporate profit P RO i,t in year t according to the national income accounting system as follows: P RO i,t = Y i,t MED i,t F C i,t W AGE i,t CURRD i,t V AT i,t (2) where Y i,t is the firm s gross output; MED i,t measures its intermediate inputs excluding financial charges; F C i,t is its financial charges (mainly interest payments); W AGE i,t is the firm s total wage bill; CURRD i,t is the amount of current depreciation, and V AT i,t is the value added tax. Note that P RO i,t defined here, as an imputed profit, can legitimately differ from firm i s true accounting profit π i,t. A main reason is the differences in the revenue and expense 8

recognition rules of the two systems, for example, not all gross output in the current year necessarily converts into firm revenue in the same year. Asset depreciation rules can also be different. Tax credits (such as earning-reinvestment credits, recycling and environmental credit) and tax loss carryovers can be other sources of the gap between P RO i,t and π i,t. For these reasons, this imputed profit is at best a very noisy proxy for π i,t. In fact, a key challenge for our empirical strategy is to effectively address the measurement errors in P RO i,t, which we will detail in later sections. On the other hand, the imputed profit from the national income account system and the (unobservable) true profit should be positively correlated since they both reflect the firm s fundamentals. We suppose that the imputed profit and the true profit are related in the following way: π i,t = η i,t + P RO i,t + θ i,t (3) where η i,t is an unknown parameter, and θ i,t is a mean zero error term. The unobservable η i,t reflects the (firm-specific) differences in profit calculation between the accounting system and the national income account system. A priori, we do not know the sign of η i,t : it can be positive or negative. By substituting Equation (3) into Equation (1), we derive RP RO i,t = d i,t P RO i,t + E i,t + ɛ i,t, (4) where we use RP RO i,t to replace ˆπ i,t ; E i,t = d i,t η i,t + e i,t ; and ɛ i,t = d i,t θ i,t + ζ i,t. Furthermore, we propose the following econometric specification for d i,t (note that j denotes the industry of firm i): d i,t = β 0 + β 1 Compet j,t + β 2 T ax i,t + β 3 F inance i,t + β 4 F irm Size i,t +β 5 X i,t + ɛ i,t (5) where Compet j,t measures the level of competition in industry j; T ax i,t is firm i s tax rate in year t; F inance i,t is a measure of how easily firm i can access capital market; and X i,t is a 9

set of control variables that includes other firm characteristics, time fixed effect, and location fixed effects. From Conjectures 1 and 2, we have the following hypotheses: Hypothesis 1 β 1 < 0, i.e., a firm s incentives to engage in tax avoidance are positively correlated with the degree of product market competition. Hypothesis 2 β 2 < 0, i.e., a firm s incentives to engage in tax avoidance are positively correlated with its tax rate. Hypothesis 3 β 3 > 0, i.e., a firm s incentives to engage in tax avoidance are negatively correlated with its accessibility to capital market. The impact of firm size on firms tax avoidance is harder to determine. On the one hand, as mentioned earlier, tax collectors in China pay more attention to larger firms, thus making the expected cost of avoiding tax higher for larger firms. Moreover, firm size may also serve as a proxy for easier access to capital market, which again reduces larger firms incentives to avoid tax. On the other hand, one could argue that there are economies of scales in tax avoidance activities, thus larger firms have stronger incentives to hide profits. While it remains an empirical issue to find out which direction the net effect goes, our prior is that the former is likely to be more important and hence we state Hypothesis 4 as follows: Hypothesis 4 β 4 > 0, i.e., larger firms have weaker incentives to engage in tax avoidance. We have a specification for E i,t, which is similar to that for d i,t as in Equation (5). However, since we cannot determine the sign of η i,t in Equation (3) a priori, and since E i,t = d i,t η i,t + d i,t e i,t, our model has no prediction about how E i,t will be affected by competition or other variables. Thus, we do not have predictions about the signs of the coefficients in the estimation of E i,t. 10

3 Data and Variable Definitions 3.1 Dataset In order to study the impact of product market competition on corporate tax avoidance, we analyze a large dataset developed and maintained by the National Bureau of Statistics of China (NBS). The NBS data contains annual survey data of all above scale industrial firms in China (i.e., industrial firms with sales above a certain level). On average, close to 190,000 firms per year for the period from 2000 to 2005 are included in the dataset, spanning 37 twodigit manufacturing industries and 31 provinces or province-equivalent municipal cities. They account for most of China s industrial value added and have 22% of China s urban employment in 2005. The NBS collects the data to compute the Gross Domestic Product. For that purpose, every industrial firm in the dataset is required to file with the NBS an annual report of production activities and accounting and financial information. The information reported to the NBS should be quite reliable, because the NBS has implemented standard procedures in calculating the national income account since 1995 and has strict double checking procedures for above-scale firms. Moreover, firms do not have clear incentives to misreport their information to the NBS, because such information cannot be used against them by other government agencies such as the tax authorities. 10 Misreporting of statistical data was commonly suspected for some time in China, the most notorious was local GDP data provided by local governments. However, the national income accounting of above-scale firms is done by the central NBS, hence is much less subject to manipulation by local governments. The NBS designates every firm in the dataset a legal identification number and specifies its ownership type. Firms are classified into one of the following six primary categories: stateowned enterprises (SOEs), collective firms, private firms, mixed-ownership firms (mainly joint stock companies), foreign firms, and Hong Kong, Macao, and Taiwan firms. The NBS does not treat publicly listed companies in China separately, which are all grouped under the mixedownership category. By the end of 2005, there are about 1,300 publicly listed companies in China s two stock exchanges, and only slightly over 700 of them are manufacturing firms. 11

To obtain a clean sample and to rule out outliers, we delete the following kinds of observations from the original data set: observations whose information on critical parameters (such as total assets, the number of employees, gross value of industrial output, net value of fixed assets, or sales) is missing; misclassified observations whose operation scales are clearly smaller than the classification standard of above scale firms, specifically, observations for which one of the following is true: (i) the value of fixed assets is below RMB 10 million; (ii) the value of total sales is below RMB 10 million; and (iii) the number of employees is smaller than 30; observations that have one of the following variables at a negative value: (i) total assets minus liquid assets, (ii) total assets minus total fixed assets, (iii) total assets minus net value of fixed assets, or (iv) accumulated depreciation minus current depreciation; observations with extreme variable values (the values of key variables are either larger than the 99.5 percentile or smaller than the 0.5 percentile). From this procedure, we obtain a sample of 514,394 observations representing 194,635 unique firms. 11 All monetary terms are in terms of 2000 constant Renminbi (RMB). 3.2 Profit Measures The NBS dataset contains input and output information for every sample firm, which allows us to compute the imputed profit (P RO) defined under the national income account system as in Equation (2). The dataset also contains the pre-tax accounting profit reported by each firm (RP RO). We normalize both imputed profits and reported profits by firms total assets (TA). As shown in Table 2, the sample mean of the reported profit is 0.0515 and that of the imputed profit is 0.1431 during 2000 2005. We define a variable GAP to denote the difference between the two profit measures. The variable GAP has a sizable mean of 0.0916, but a much smaller median of 0.027. 12 12

As mentioned before, the positive difference between the two profit measures can simply reflect the exogenous differences between the accounting system and the national income account system, such as different expenses recognition rules, different asset depreciation rules, different tax credit and tax loss carryovers rules, etc. But tax avoidance activities by firms can also make additional contributions to the gap between the imputed profit and the reported profit. It is the purpose of this paper to empirically detect the latter source of the gap and analyze how market competition affects firms incentives to engage in tax avoidance activities. An example may illustrate the difference between the imputed profit and the reported profit in China. In 2005, Hudong Heavy-machines Co., a listed company in the Shanghai Stock Exchange, reported a pre-tax operating profit of 141 Million RMB. However, the pretax profit under the national income account system according to Equation (2) is 382 million RMB. The gap is mainly due to the differences in output vs. sales (RMB 1,744 million vs. RMB 1,423 million), and intermediate inputs vs. expenses (RMB 1,351 million vs. 1,282 million). As discussed early, the differences inherent in revenue and expense recognition rules between the two systems can lead to legitimate differences between outputs and sales and between inputs and expenses, thus cause a substantial difference between the reported profit and the imputed profit. The central question in our analysis is whether behind the legitimate differences in the two systems, firms manage their accounting and reporting practices to reduce their income taxes. And if they do, how do they do it differently across industries of different competition? 3.3 Competition Variables Following the standard practice in the Industrial Organization literature, we construct four variables to measure competition intensity. The first is simply the number of above-scale firms operating in a four-digit industry. We use its natural logarithm in our analysis. The second measure is the Herfindahl index of total sales in a four-digit industry, which is the sum of squares of the market shares (by sales) by all firms in industry j. As another way of measuring concentration, we compute the total market share accounted for by the four 13

largest firms in industry j (by sales). Both the Herfindahl index and the concentration ratio are negatively correlated with competition. The fourth measure of competition we use is the industry average profit margin, defined as the ratio of total pre-tax profit to total sales in a four-digit industry. As competition increases, one expects that firm profit on average will fall, thus the industry average profit margin should be negatively correlated with competition. All four competition variables are computed for each of the 503 four-digit manufacturing industries in China. For brevity, we present the competition variables for the thirty-seven two-digit manufacturing industries averaged over 2000-2005 in Table 1. All measures show substantial variations across industries. We calculate the bivariate correlations among the four competition measures of 4-digit industries. Not surprisingly, the Herfindahl index and the concentration ratio are highly correlated with a correlation coefficient of 0.864, since both measure concentration. The logarithm of firm number is negatively correlated with both the Herfindahl index and the concentration ratio, the coefficients are 0.564 and 0.745, respectively. Industrial average profit margin is negatively correlated with the logarithm of firm number (-0.0241), and positively correlated with both the Herfindahl index and the concentration ratio (the correlation coefficients are 0.013 and 0.011, respectively). Note that the correlation between profit margin and the other measures has the expected sign, but is very small. Overall, the four variables measure the degree of competition consistently, yet offer somewhat differentiated perspectives. We report the summary statistics of the competition variables based on both the two-digit and four-digit industry levels in Table 2. The two concentration measures, the Herfindahl index and the concentration ratio, are very sensitive to market definition. As one would expect, concentration increases as the market becomes smaller (from 2-digit to 4-digit industry). On the other hand, the industrial average profit margin is stable. 3.4 Other Variables From our data set, we construct the following variables of firm characteristics and report their summary statistics in Table 2. 14

We construct a variable T AX from the ratio of actual corporate income tax paid by a firm to its reported pre-tax profit, and set it to zero for loss-making firms. Although the standard corporate income tax rate is 33% in China, the Chinese government gives various preferential tax treatments (e.g., tax reduction for a certain period of time) to various kinds of firms (e.g., foreign firms, high-tech firms, joint ventures). Local governments also grant tax holidays and rebates to various types of companies in order to promote local economic development. Furthermore, tax collection and enforcement is quite discretionary, leaving large room for distortion and bribery in exchange of tax reduction. As a result, there are substantial variations in the effective tax rates across firms. From Table 2, the variable T AX has a sample mean of 25.01% with a standard deviation of 13.14%. Note that a small fraction of firms have very high effective tax rates (the sample max is 77.29%), which likely indicates that there are large tax carryovers from previous years. We use access to credit to measure firms marginal returns to capital. Firms that are more financially constrained should have higher marginal returns to capital. However, it is difficult to measure a firm s access to credit markets. In our context, we use the following strategy. In development economics it is well established that access to credit is crucial for firm performance and economic growth in developing countries (e.g., Rajan and Zingales, 1998; Banerjee and Duflo, 2004). In the case of China, as the Chinese economy has been growing at a very fast pace, Chinese firms demand for credit has grown rapidly. But the banking sector and stock market have not developed quickly enough to keep pace with this growing demand, thus Chinese firms are usually facing binding credit constraints. Thus, the actual amount of debt a firm has reflects mostly how much it manages to borrow, not its endogenously chosen optimal capital structure. Consequently, we expect firms access to credit and their debt to equity ratios to be positively correlated. Note that banks in China have little discretion over interest rates they charge borrowing firms (the corporate lending rates were regulated throughout our sample period). Also note that the corporate bond market is tiny due to strict regulations. Therefore, interest payments on loans reflect how much a firm is able to borrow. Therefore, we compute the ratio of total financial charges to total assets for each firm and use it as a proxy for the firm s access to credit markets. 13 From Table 2, this variable has a 15

sample mean of 1.56% and a standard deviation of 1.43%. We construct six dummy variables to represent a firm s ownership status: D SOE, D private, D foreign, D HK/T W, D mixed, and D collective. These binary variables take the value of one if a firm falls into a corresponding ownership category and zero otherwise. From the means of these dummy variables in Table 2, in our sample observations of (domestic) private firms, Hong Kong or Taiwan invested firms, and foreign firms account for 25.5%, 14.8% and 13.1%, respectively. This shows how far the Chinese economy has changed from the central planning system since reforms started. Roughly speaking, more than 50% of the above scale industrial firms are pure private firms, while SOEs and collective firms only account for 10.3% and 14.6% of our sample observations. The remainder of 21.7% consists of mixed ownership firms. We measure firm size by the logarithm of the number of employees. The mean of employment in our sample is 501, with the maximum and minimum being 161,654 and 30 respectively. In Table 2, one can see that on average, our sample firm has total assets of 163 million yuan. Using the logarithm of total assets as a measure of firm size yields similar results. We also include the ratio of sales to total output, as one control variable. To some extent, this variable controls for the difference in the timing of when revenues are recognized into income under the accounting system and the national income account system. Its sample mean is 0.994 and standard deviation is 0.244 (Table 2). 14 We create twenty-eight location dummies to capture the geographical effect. For the 31 provinces and province-equivalent municipal cities, we merge Tibet, Qianghai, and Ningxia into one location because each of the three adjacent provinces has too few observations. We also merge Chongqing and Sichuan because Chongqing was separated from Sichuan to become a province-equivalent municipal city only since 1997. Finally, we create six year dummies to capture time-varying effects. 4 Empirical Results To get a feeling of how competition may affect a firm s profit hiding incentive, we plot the difference between the imputed profit and reported profit against the logarithm of the number of 16

firms in an given two-digit industry in Figure 1. We observe a significantly positive correlation between the gap of the two profit measures and the level of competition. 15 There are no clear outliers in the scatter plot, implying that the relation between the gap between the imputed profit and reported profit and competition measures is unlikely due to outliers. Plotting the gap against other competition measures yields a similar pattern. Although this quick look at the data should be taken with caution, it does suggest that the difference between the imputed and reported profits may be related to the degree of market competition. Based on Equations (4) and (5), we estimate the following equation: RP RO i,t = (β 0 + β 1 Compet + β 2 T AX + β 3 F INANCE + β 4 LNLABOR + β 5 RSALE +β 6 D OW N + β i d year + β j d loc ) P RO i,t + α 1 Compet + α 2 T AX i=year j=loc +α 3 F INANCE + α 4 LNLABOR + α 5 RSALE + α 6 D OW N + α i d year + j=loc i=year α j d loc + ɛ i,t (6) where D OW N is a set of five ownership dummy variables with the SOE dummy taken as the benchmark; d year is a set of year dummies; and d loc is a set of location dummies. Including d year and d loc controls for time-specific, and location-specific effects on profit reporting behavior. In the above specifications, the signs and magnitudes of the coefficients of the interactive terms (i.e., βs) indicate whether and to what extent each of the variables affects the sensitivity of reported profits to imputed profits. Our main focus is on β 1. If it is negative and significant, then firms in more competitive industries tend to report less profit (Hypothesis 1). Our other hypotheses say that β 2 < 0, β 3 > 0 and β 4 > 0. Note that we include all control variables and dummy variables in the constant term in Equation (6) corresponding to E i,t in Equation (4), but as discussed before, we do not have predictions about the signs of the coefficient estimates of αs. 4.1 OLS Regressions We first use the OLS regression to estimate Equation (6), in which we use four alternative measures of competition at the four-digit industry level. The results are reported in Table 3. 17

In each column (except columns (1) and (2) where competition measures are excluded), the heading identifies the measure of competition used in the regression. To save space, only the estimates of interest are reported. We report robust standard errors in parentheses under the estimates. In column (1), we only include the constant and the imputed profit in the regression. The imputed profit explains about 20.2% of cross-sectional variations in the reported profit. The estimated coefficients for the imputed profit and intercept are 0.202 and 0.06 respectively, and are both significant. If (i) the imputed profit is a good proxy for the true accounting profit, (ii) there are no measurement errors in the imputed profit, and (iii) firms do not under-report profit, then the coefficient of the imputed profit should be one and that of the intercept should be zero. Not surprisingly, this is not the case. In column (2), we add all variables specified in Equation (6) except the competition variables and their interactions with the imputed profit. The model fitness improves as the adjusted R-squared increases from 20.2% to 24.2%, indicating that these covariates do help explain the divergence between the reported profit and the imputed profit. In columns (3) and (4), we only consider the impact of competition (measured by the Herfindahl index and the profit margin, respectively) on the relationship between the reported profit and the imputed profit. The interactions of the imputed profit with both competition variables are significantly positive, showing that competition reduces the sensitivity of the reported profit to the imputed profit. Compared to column 1, adding competition variables increases the adjusted R-squared of the OLS regression by 1 2 percentage points, indicating that competition variables are important in explaining firms profit reporting decisions. In columns (5)-(8), we estimate Equation (6) with OLS regressions using different competition measures. The estimated coefficient of the interaction of competition with the imputed profit, β 1 in Equation (6), is negative when the logarithm of the number of firms is used, and is positive in all other cases. In all the regressions, the estimated β 1 is precisely estimated with a standard error ranging from 0.005 to 0.0188, and is statistically very significant. The evidence here shows that firms tend to hide more profits in industries that are less concentrated, have more firms, or have lower average profit margin. 18

The results on other variables of interest are also largely consistent with our conjectures. Table 3 shows that in all regressions, the estimated coefficient of the interaction of effective tax rate with the imputed profit is negative and statistically significant. This is consistent with Hypothesis 2. All else equal, a firm s incentives to hide profits are negatively correlated with its access to external financing (Hypothesis 3). It is evident from Table 3 that this hypothesis is strongly supported by the OLS results. In all the regressions, the estimated coefficient of the interaction of access to credit with the imputed profit is positive, statistically significant, and quite stable across all regressions. Table 3 also shows that in all regressions, the estimate of the interaction of the logarithm of the number of employees (our proxy for firm size) with the imputed profit is positive and statistically significant. The estimates are in a close range from 0.011 and 0.012. These results suggest that on the balance larger firms are less motivated to avoid corporate tax (Hypothesis 4). Table 3 also reports the estimated coefficients of ownership dummies. From the sensitivity of the reported profit to the imputed profit, private firms and Hong Kong and Taiwan firms report less profits than SOEs, collective firms and foreign firms report more profits than SOEs, while mixed ownership firms are not significantly different from SOEs. 4.2 The Impact of Competition on Profit Hiding: IV Estimates While the OLS regression results support our hypotheses, they should be interpreted with caution. First of all, there might be omitted variables in our model specifications, which could cause an endogeneity problem if these omitted variables are related to observed righthand-side variables such as competition variables and the imputed profit (we defer a more detailed discussion of potential omitted variables to Section 5). Second, the imputed profit and the competition variables in our empirical analysis are likely measured with error, in which case OLS will yield biased and inconsistent estimates. Third, we mainly rely on industrylevel variations in competition to capture the effect of competition on firms tax avoidance behavior. It may lead to concerns because of the difficulty of comparing competition levels across industries. 19

These concerns justify the need to identify sources of variations in both the industry-level competition and the imputed profit that are likely to be independent of corporate choice variables. We use two approaches in this paper to address these concerns. Our primary approach is to use the method of instrumental variables. We instrument for both the imputed profit and the measure of competition the two variables of interest that arguably are subject to an endogeneity problem. We identify appropriate instrument variables and use the 2SLS regression to estimate the β 1 s in Equation (6). In particular, we conduct several tests to study the validity of instrumental variables used in our empirical analysis. We show that these instruments are indeed exogenous and relevant, and are not subject to the weak instrument problem (see e.g., Stock, Wright, and Yogo, 2002). For a second approach, we use variations introduced by a natural experiment in China in 2002, i.e., China s WTO entrance, which exposed several Chinese manufacturing industries to heightened product market competition. We examine how firms in those industries change their tax avoidance behavior in Section 4.3. We start by discussing the method of instrumental variables. We use the four-digit industry average imputed profit (excluding the firm itself) as the instrument for a firm s imputed profit. Furthermore, we instrument for the competition variables with the number of application procedures a firm has to go through in order to enter a four-digit industry (ST EP ). 16 This variable is a direct measure of entry barriers to a four-digit industry imposed by the government. It should be correlated with the intensity of competition, but is unlikely to be related to factors affecting firms profit-reporting once they have entered. 17 Using the above IVs, we perform the 2SLS estimations of Equation (6). To save space, we only report the results of using the profit margin as the measure of competition. We report the second stage regression results in Table 4. Column (1) repeats the OLS results from Table 3 for comparison. Columns (2) (4) present the IV results. In column (2), only the profit margin and its interaction with the imputed profit are instrumented with ST EP and the interaction of ST EP with the imputed profit. In column (3), we instrument for the imputed profit with its four-digit industry average. The interactions of the imputed profit with other covariates are instrumented with the interactions of industry average imputed profit with corresponding covariates. Column (4) presents the IV estimates in which the profit margin, imputed profit, 20

and their interactions with other covariates are all instrumented. Each regression includes ownership dummies, location dummies, year dummies and their interactions with the imputed profit. Table 5 presents the results of the first-stage regressions for models (2) (4) of Table 4. Because the implications drawn on the first stage regression results for different models are similar, we take the first-stage regression of model (2) from Table 4 as the example. Here, only the profit margin and its interaction with the imputed profit are instrumented. The instrumented variables are used as dependent variables in the first stage regressions. The independent variables include the two instruments ST EP and its interaction with the imputed profit and all exogenous variables used in the second stage regressions (we only report the coefficient on the two instruments for brevity). The estimates of ST EP in the first column and the interaction of ST EP with the imputed profit in the second column are all statistically significant. Following the suggestion of Bound, Jaeger and Baker (1995), Table 5 also presents F-statistics with the null hypothesis that instruments jointly equal zero. The F statistics in all of our first-stage regressions are significantly larger than the critical values suggested in the literature (e.g., Bound et al., 1995; Staiger and Stock, 1997). It should be emphasized that in case of multiple endogenous variables, the equation by equation F statistics might be misleading. The instruments can be weak although they are very significant in each first stage regression. As pointed out by Stock and Yogo (2005), the reason for this is that when the predicted endogenous explanatory variables are close to be collinear, it is difficult to separate their effects. Stock and Yogo (2005) tabulate critical values that enable the use of the Cragg-Donald (1993) statistic to test whether a set of instruments are weak in models with more than one endogenous variable. The bottom of Table 4 reports the results of the weak identification test (the Cragg-Donald statistics, see e.g., Stock and Yogo, 2005). We now discuss the IV estimation results in Table 4. In all model specifications (note that the profit margin is negatively correlated with the competition level), we find that β 1 from Equation (6) is always positive and statistically significant, indicating that a higher level of competition increases firms incentives to hide profits. We report the Cragg-Donald 21