Asymmetric Corporate Tax Compliance: Evidence from ... - SSRN papers

Asymmetric Corporate Tax Compliance: Evidence from a Tax Reform in China Hua Cheng, University of Texas at Austin Kishore Gawande, University of Texas at Austin Yuanyuan Ma, Zhongnan University of Economics and Law

We wish to thank Jiahua Che, Shuping Chen, Kimsau Chung, Michael Geruso, Lin Gui, Sukjin Han, Lixing Li, Yu Liu, Leigh Linden, Xiaobing Liu, Yongzheng Liu, Lilian Mills, Mathias Thoenig, Tianyang Xi, Jiacai Xiong, Colin Lixin Xu, Yang Yao, Wei Yu, Muyang Zhang, Xiaobao Zhang, Li’an Zhou, Guozhong Zhu, Shenghao Zhu, and participants in UT-Austin, Shanghai University of Finance and Economics, Southwestern University of Finance and Economics, Jiangxi University of Finance and Economics, 2017 CCER Summer Institute, and 2015 China Economics Annual Meeting, for helpful comments and discussions. All the possible errors are our own.

1

Electronic copy available at: https://ssrn.com/abstract=3149271

Abstract The 2007 corporate tax reform in China engendered a convergence in tax rates for firms. We find a stronger reaction in terms of tax avoidance among firms with a larger tax increase than firms with a smaller tax increase. Firms with a 1 percentage point lower estimated pre-reform tax rate (corresponding to a 0.6 percentage point higher tax increase) are associated with a 0.11 percentage point additional increase in the cost of goods sold plus expenses, as scaled by sales revenue. However, we do not find a similar effect within the possible tax cut group. Therefore, these results imply that firms perceive losses more strongly than gains, as suggested by Kahneman and Tversky (1979). Furthermore, these asymmetric reactions existed only in private firms and not in state-owned enterprises (SOEs) and collective firms, suggesting that SOEs and collective firms pursued policy objectives in addition to profit maximization. Using pretax profit scaled by sales revenue as the dependent variable yields similar results. We provide evidence for why abnormal changes in expense ratios and profit margins resulted from tax compliance rather than real investment activities or audit by tax authorities. Overall, our findings highlight the potential costs of increasing or reducing tax rates frequently. Key words: tax avoidance; corporate tax; loss aversion; behavioral economics; ownership JEL code: M41, H26, H32, G32, G38

2


Ⅰ.INTRODUCATION Corporate income taxes represent a vital cash outflow for firms, and high tax rates probably induce an increase in tax avoidance activities. Although corporate tax avoidance/compliance draws considerable attention in academic research, we note that research based on plausibly exogenous variations such as a tax reform remains scarce. Whether firms typically respond to tax cuts and tax increases in a symmetric manner warrants attention; specifically, whether such firms reduce tax avoidance when tax cuts and increase tax avoidance when tax increases should be investigated. Additionally, considering the widespread presence of state-owned enterprises (SOEs) in most parts of the world compared with private firms, a deviation from profit maximization might lead to different reactions to tax reforms between SOEs and private firms. In this paper, we explore these tax avoidance/compliance activities by using a plausibly exogenous variation resulting from China’s 2007 corporate tax reform. Prior to the reform, China’s corporate income tax rates typically varied between 15% and 33% by area and industry, and foreign-invested firms abided by a different corporate tax law and typically enjoyed a preferential corporate income tax rate of 15%. This special treatment of foreign-invested firms was designed to attract foreign capital in the early reform period (i.e., the 1980s), when the country had limited capital. In contrast to the corporate tax structure of the United States, China taxes firms’ profit on a constant schedule to ensure a constant marginal tax rate. A new corporate tax law passed in March 2007 harmonized the tax rates at 25% for most medium or large firms, in addition to discontinuing the preferential treatment of foreign-invested firms. The new law became effective in January 2008. Therefore, firms with a pre-reform tax rate of 25% or above enjoyed a tax cut, whereas those with a pre-reform tax rate of 25% or below faced a 3


tax increase, regardless of ownership. This overhaul in the corporate tax code is of comparable consequence to the 1986 tax reform in the United States. However, it is different from typical tax reforms because of the simultaneous and widespread increase and reduction of tax rates. Therefore, the large variation in pre-reform tax rates combined with the exogenous policy change provides a unique opportunity for us to identify the asymmetric reactions to the tax reform (Li and Cai 2011). An obstacle associated with the large variation in pre-reform tax rates is the undocumented statutory tax rates for specific firms. To overcome this obstacle, we use the average of corporate taxes payable divided by pre-tax profit between 2005 and 2007 as the estimated pre-reform tax rate (EPRT) and apply it to each firm by using a method similar to that by Lin, Mills and Zhang (2014). Chandra and Long (2013), and Chen (2017) also use the similar method and data to estimate the value-added tax rate for firms or counties. In the studies of Cai and Liu (2009) and Li and Ma (2015), a key finding is that their imputed “true” profit is positively correlated with the true profit. Therefore, in our study, it is crucial that the corporate taxes payable divided by pretax profit is positively correlated with the statutory tax rate applied to each firm. Although this is reasonable1, we will still provide further evidence to support this. We find a convergence in the estimated tax rates faced by different firms, as well as a strong negative correlation between the estimated pre-reform tax rate and the estimated tax rate change, consistent with the proposed change of the tax code.

1

Pre-tax profit is equal to taxable profit plus some adjustment. The adjustment may be positive or negative, including penalty imposed by the government or expense beyond the deductible amount if any. This adjustment is usually small compared with the pre-tax accounting profit. Unless the adjustment is very large so that the positive correlation between the imputed “true” profit and “true” profit is overturned, which seems impossible, our results are valid. The difference between pre-tax profit and taxable profit in China is comparable with the difference in the two sets of books in U.S. firms, one for IRS (tax books), and another for their annual reports (shareholders’ books). 4

Using an event study method, we find a substantial negative effect of the EPRT on the post-reform cost of goods sold plus expenses for the possible tax increase group (EPRT

0.25),

suggesting more tax avoidance activities by firms confronted with larger tax increases compared with those with smaller tax increases. Firms with a 1 percentage point lower EPRT (corresponding to a 0.6 percentage point higher post-reform tax rate) is associated with a 0.11 percentage point increase in the cost of goods sold plus expenses, scaled by sales revenue (hereafter termed expense ratio). By contrast, we do not find similar effects for the possible tax cut group (EPRT > 0.25). The findings thus indicate that firms with larger tax cuts did not reduce tax avoidance activities compared with those with smaller tax cuts. This corresponds to previous research suggesting that firms perceive losses more strongly than gains (Kahneman and Tversky 1979). These asymmetric reactions are predominantly driven by private firms, whereas SOEs and collective firms2 do not exhibit significant reactions, regardless of whether ownership is based on registration type or paid-in capital. Furthermore, we also find a strong positive effect of the EPRT on the post-reform pre-tax profit, scaled by sales revenue (frequently referred to as the profit margin) within the tax increase group but not the tax cut group. To the best of our knowledge, this is the first paper documenting the asymmetric reactions of firms to corporate tax changes. An advantage of this study is that it comprises a large sample that is mainly composed of non-publicly traded firms; by contrast, studies regarding tax avoidance in the United States typically use data on publicly traded companies, and hence substantially smaller in sample sizes. Additionally, publicly traded companies may differ from non-publicly traded companies

2

Nominally the collective firms are controlled by workers, however, this nominal controlling power is weak usually since the ownership is not really clear, hence these firms are much more likely to be influenced by the government. 5

regarding tax avoidance behavior, because these firms have more financial instruments at their disposal and are subject to stricter tax audits and often inflate earnings to boost stock prices. Lin et al. (2014) do not find an effect on publicly traded companies following China’s 2007 corporate tax reform, which is possibly a result of strict auditing and incentives related to stock prices. The price we pay for the unique tax reform and the large sample size is that we cannot construct a book-tax gap as a measure of tax avoidance. Therefore, we meticulously explain why the abnormal changes in expense ratios and profit margins result from tax avoidance/compliance rather than real investment activities. Regarding the tax increase group, we find an increasing expense ratio and a diminishing profit margin associated with the EPRT. If the changes in expense ratios result from real investment activities, groups with EPRTs close to 0.25 would likely use almost all the intermediate goods from firms with lower EPRTs, which in turn would use almost all the intermediate goods from firms with next lower EPRTs. This would require firms with the lowest EPRT to exhibit increases in expense ratios from real investment activities, which seems impossible. Moreover, if the changes in expense ratios and profit margins result from real expenditures on intermediate goods, we would likely observe industry-wide or prefecture-wide changes. We therefore add the interaction terms of industry fixed effects/prefecture fixed effects with the post-reform dummy (2008–2009) in a robustness check, and the estimates of major interest are identical, implying real investment activities have little impact on our results. Finally, an alternative explanation is audit by tax authorities. For example, this tax reform might impact tax audit differentially based on pre-reform tax rates such as what documented by Khan, Khwaja, and Olken (2016) in Pakistan. This explanation is not quite 6

plausible because it is hard to explain why there was not differential impact among the tax cut group. Furthermore, we divide the tax increase group into smaller firms and bigger firms. Although the literature usually suggests that bigger firms are more likely to be audited by tax authorities since tax audit incurs large fixed costs, our empirical results are actually identical for these two groups. Our findings are consistent with Carrillo, Pomeranz and Singhal (2017), who also find that audit toward tax avoidance is limited in developing countries. The contribution of this study to the literature is threefold. First, this is the first paper to document asymmetric reactions of firms to corporate tax changes. Firm decisions usually reflect an aggregation of opinions from numerous people and such decisions might differ from individual decisions; hence, whether loss aversion still exists for firms is unclear. The existence of asymmetric reactions has a strong policy implication. Although many economists suggest that a lower tax rate will reduce tax avoidance and hence do not lower government revenues significantly, our findings show that the reduction in tax avoidance is small. Especially, if the government initially increases the tax rate and reduce it later, as regularly happened, there could be a large loss in tax revenues and social welfare. Overall, our findings highlight the potential costs of increasing or reducing tax rates frequently. Second, we apply a continuous measure of the treatment group (i.e., logarithm of the EPRT) in an event study framework and verify its reliability, facilitating the analysis of asymmetric reactions and providing strong identification. Specifically, firms with certain EPRTs can be considered the treatment group, whereas those with all other EPRTs can represent the control group. Finally, our large dataset enables us to construct a large panel to provide strong statistical power, and non-publicly traded firms, which usually form a larger share of the economy, are more likely to avoid taxes. 7

The remainder of this paper is organized as follows. Section 2 provides the background of the tax policy change and a literature review, and develops hypotheses. Section 3 introduces the empirical strategy and discusses the identification. Section 4 describes the data. Section 5 reports the main empirical results and robustness check. Section 6 concludes. Ⅱ. BACKGROUND, LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT The 2007 Corporate Tax Reform In contrast to the United States, which has two levels of corporate income tax (federal and state)3, China has only one level of corporate income tax. In China, corporate income tax is collected by the local tax bureau instead of the national tax bureau, although the tax revenue is shared by the local and national governments. Enforcement of the corporate income tax code is more likely to be determined by the local government. Prior to the 2007 reform, the standard corporate income tax rate was 33% in China. However, the national government granted various preferential tax treatments (e.g., temporary tax reductions) to various types of firms (e.g., foreign firms, high-tech firms, and joint ventures). Local governments also granted tax holidays and preferential tax rates to promote economic development. Furthermore, tax collection and enforcement were quite discretionary, engendering the possibility of widespread distortion and bribery in exchange for tax reduction (Cai and Liu 2009). In addition, foreign-invested firms obeyed a separate corporate tax law and frequently enjoyed a preferential tax rate of 15%. However, because foreign-invested firms generally faced tax increases during the reform, we exclude them from our analysis. 3

Although some local governments (county, municipality and school district) in the U.S. also impose corporate income tax, the rate is usually very small and the tax is not important compared with property tax and sales tax. In China, property tax does not exist in most areas, and hence corporate income tax is very important for local governments. 8

The new law, passed on March 16, 2007, and enforced on January 1, 2008, heralded the beginning of a new era of investment in China. The salient features of the new law are a revised standard tax rate of 25% applicable to most native and foreign-invested firms, a preferential rate of 20% available to small businesses, and a 15% rate for certain new and high-tech enterprises. For firms with tax rates above 25% or nearly 25% prior to the reform, the 25% tax rate became effective immediately. For firms paying tax rates considerably below 25% (such as 15% or lower) prior to the reform, the new law allowed a gradual increase in the tax rate over a period of five years.4 Tax holidays granted before the reform were also allowed to carry over. 5 In our empirical study, we drop the possible small businesses based on the criteria specified in the law because they enjoyed a post-reform tax rate of 20% rather than 25%.6

Insert Figure 1 here

Firms with a pre-reform tax rate of 25% or below generally faced a tax increase, whereas those paying a pre-reform tax rate of 25% or above generally received a tax cut, regardless of ownership. Figure 1 presents the estimated tax rates for the two groups. The low EPRT group experienced a sharp increase from 2007 to 2008, while the high EPRT group experienced a 4

Notice that firms granted a grace period might also face different pre-reform rates. Firms with a lower pre-reform tax rate still faced a larger increase in tax rate than those with a higher pre-reform tax rate, even if they were both granted the same schedule for a grace period. Moreover, in our data, we find firms with a EPRT between 0.2 and 0.25 faced relatively stable estimated tax rates, while firms with a EPRT between 0.15 and 0.2 experienced a relatively smaller tax increase than firms with a EPRT less than 0.15. We also find a highly negative correlation between the estimated pre-reform rate and the estimated change in tax rate for the possible tax increase group (EPRT 0.25) as for the possible tax cut group (EPRT > 0.25), hence overall for firms with a lower tax pre-reform tax rate, the tax increase would be larger. Therefore, including firms which perhaps granted a grace period should not affect my results. 5 We exclude firms with an estimated income tax rate less than 0.01 in any year, so our sample should not contain firms having not used their tax holidays. 6 Although it is not feasible to drop possible high technological enterprises enjoying the preferential tax rate, a sketch of their main products (there is a variable recording the major products of each firm) shows that the number of these firms should be very small. 9

sharp drop over the same period. We note that the percentage point change for the low EPRT group was smaller on average; this tends to bias down our estimated asymmetric reactions. Additionally, the percent change corresponding to 1 percentage point change in tax rate is larger for the low EPRT group than the high EPRT group, and hence the percent change in tax rates for the low EPRT group is larger than for the high EPRT group. In a robustness check, we consider the logarithm of EPRT as a treatment status to address this issue, and the results are similar. Using the standard deviations of the estimated tax rates by year, Figure 2 shows the convergence in tax rates faced by the two groups, for both the whole sample and the subsamples such as private firms and SOEs and collective firms.

Insert Figure 2 here

Related Literature and Hypothesis Development Current studies on this tax reform focus on the consolidation of tax treatments for native firms and foreign-invested firms, which is the most salient feature of this new law, partly because of the difficulty in identifying statutory tax rates applied to different firms. An (2012) reveals that consolidation drove away foreign direct investment, particularly from Hong Kong, Macao, and Taiwan. Mao and Ji (2014) study corporate tax avoidance induced by the reform. They reveal a significant reduction in the reported profitability of foreign-invested firms, particularly those with pure foreign capital, compared with the native firms. They theorize that 10

these firms have more opportunity to misreport profits through transfer pricing. Our analysis differs from theirs because we consider tax avoidance/compliance for native firms by using a novel definition of treated and control groups, and we reveal asymmetric reactions to this reform. Our results imply that regardless of transfer pricing probably employed by foreigninvested firms, native firms also exhibit effective methods for reducing tax payments. Lin et al. (2014) discuss the differential reactions of publicly traded companies and nonpublicly traded private firms facing tax cuts engendered by the reform by using a similar measure of the pre-reform tax rates to ours, although they do not provide evidence to support this measure. They state that non-publicly traded private firms reported higher accrued expenses in 2007 than in 2008 to reduce tax burdens, and that such firms were more tax-aggressive compared with publicly traded companies. Their study focuses on extremely large non-publicly traded private firms and therefore does not explore most non-publicly traded firms. By contrast, we explore all firms and focus on asymmetric reactions within the possible tax increase group and tax cut group. Moreover, they regard changes in accrued expenses between 2007 and 2008 as a simple shift, whereas we find evidence that the tax reform altered firms’ calculations permanently. Our focus on the asymmetric reactions among the possible tax increase and tax cut groups is related to the literature on behavioral economics. Kahneman and Tversky (1979) suggest that people perceive losses more strongly than gains, implying that penalties are more effective than proportional bonuses. Abdellaoui, Bleichrodt, and Paraschiv (2007) provide a parameter-free measurement of loss aversion. Recent experimental evidence confirms this view. Homonoff (2013) investigates the effect of a five-cent shopping bag tax imposed in the Washington Metropolitan Area. Despite the small size of the incentive, she finds that the tax 11

substantially reduced the number of customers using a disposable bag. By contrast, a similar policy that offered customers a five-cent bonus for reusable bag use generated virtually no effect on behavior. Such loss aversion exists in monkeys as well, and is probably an instinctual animal characteristic (Chen, Lakshminarayanan and Santos 2006). Dyreng, Hanlon and Maydew (2010) suggest that individual executives significantly affect the level of tax avoidance determined. Managers’ loss aversion possibly affects firm behavior. However, most major strategic decisions by firms are made within a group. Economic and psychological studies indicate that groups are more rational than individuals (Bornstein and Yaniv 1998; Bornstein, Kugler and Ziegelmeyer 2004). Therefore, whether group decision making mitigates managers’ loss aversion and whether loss aversion applies to firm behavior are empirical questions. We therefore propose two competing hypotheses: Hypothesis 1a. Firms with larger tax increases react more strongly in terms of tax avoidance/ compliance than firms with smaller tax increases do. Firms with larger tax cuts do not react more strongly in terms of tax avoidance/compliance than firms with smaller tax cuts do. Hypothesis 1b. Firms with larger tax increases react more strongly in terms of tax avoidance/compliance than firms with smaller tax increases do. Firms with larger tax cuts react more strongly in term of tax avoidance/compliance than firms with smaller tax cuts do. The 1986 corporate tax reform in the United States has been sufficiently studied.7 This reform reduced the federal corporate income tax rate by approximately 12%. Through the use of data on publicly traded companies, several studies reveal significant tax-induced earnings

7

After then there is not an overhaul in the corporate tax law in the United States. 12

management (Dhaliwal and Wang 1992; Guenther 1994; Lopez, Regier and Lee 1998). Johannesen and Zucman (2014) investigate the Tax Haven Crackdown by G29 countries. In contrast, the substantially larger variation in pre-reform corporate tax rates and the two-sided changes in tax rates (cuts and increases) in China provide a unique opportunity for identifying various effects of the tax rates. An advantage to study non-publicly traded firms is that they provide a substantially larger sample compared with publicly traded companies. Moreover, without worrying stock price reactions, non-publicly traded firms have a stronger incentive to avoid taxes. Because the number of exogenous policy shocks is extremely limited, a major strand of the literature on corporate tax avoidance constructs effective tax rates (e.g., Dyreng et al. 2008; Lennox et al. 2013) or book-tax differences as proxies (e.g., Manzon and Plesko 2002; Desai and Dharmapala 2006, 2009; Chen, Chen, Cheng and Shevlin 2010). Lennox, Lisowsky and Pittman (2013) use multiple proxies to capture various dimensions of firms’ tax minimization behavior. Desai and Dharmapala (2006) further attempt to eliminate the effect of total accruals on this measure. Because of the lack of reported tax rates for non-publicly traded firms, constructing a similar book-tax gap in our study is infeasible, which is a consequence of using such a large dataset. Cai and Liu (2009) and Li and Ma (2015) analyze tax avoidance in Chinese firms by imputing the underlying “true” profit with the national income account technique, using the same dataset as this study. They then subtract intermediate input, financing expense, depreciation, and value-added tax from the gross output, and they consider the gap between the remaining term and pre-tax profit as a measure of tax avoidance. However, the post-reform (2008–2009) datasets do not contain information required for constructing such a measure. Therefore, we account for these influences by including firm fixed effects. In a robustness check, 13

we add the interaction terms of industry fixed effects or prefecture fixed effects with a postreform period (2008–2009) dummy.8 A key finding in the studies by Cai and Liu (2009) and Li and Ma (2015) is that the estimated “true” profit is positively correlated with the actual profit. Hence, in the current study, it is crucial that the corporate taxes payable divided by pre-tax profit is positively correlated with the statutory tax rate applied to each firm; Figures 1 and 2 support this. Although SOEs are nominally owned by the people, such firms largely serve the interests of insiders who benefit from increased cash flows from avoided taxes. SOEs often have minority shareholders, who likewise benefit from tax sheltering. Furthermore, the penalty imposed by tax authorities following a tax audit is relatively small because of political connections. For example, Faccio, Masulis and McConnell (2006) find that political connections increase the likelihood of government aid. Li and Ma (2015) reveal a strong government size– tax avoidance relationship in China’s SOEs. Annuar, Salihu and Obid (2014) also find that government ownership is associated with corporate tax avoidance among Malaysian listed companies. Nevertheless, SOEs often pursue policy objectives in addition to profit maximization (Cooper, Gong and Yan 2010; Cheng 2016). Moreover, top management in SOEs receive only a limited share of the benefits from tax avoidance, although they must assume major responsibility if their activities become uncovered. Therefore, SOEs may pursue less aggressive tax strategies compared with private firms. Chan, Mo and Zhou (2013) provide supporting evidence for this in a study of China’s publicly traded companies. Although most of the referenced studies use data on publicly traded companies, our study provides evidence of this theoretically ambiguous issue by using a sample mainly composed of non-publicly traded 8

The data on county government size, such as those used by Li and Ma (2015), are only available till 2006, so they are not sufficient for our analysis here. 14

firms. Our use of a plausibly exogenous policy change also provides substantial identifying power. Ⅲ. EMPIRICAL STRATEGY AND IDENTIFICATION Identifying Variation Our identification relies on a plausibly exogenous policy shock resulting from China’s 2007 corporate tax reform. Prior to 2007, China’s corporate income tax rates usually varied between 15% and 33% by area and industry, and foreign-invested firms abided by a different corporate taxation law and typically enjoyed a preferential tax rate of 15%. A new corporate taxation law passed in March 2007 harmonized the tax rates at 25% for most medium or large firms, in addition to discontinuing the preferential treatment of foreign-invested firms. The new law became effective in January 2008. Overall, firms with a pre-reform tax rate of 25% or above enjoyed a tax cut, whereas those with a pre-reform tax rate of 25% or below faced a tax increase, regardless of the ownership. We will exclude the foreign-invested firms from our analysis because few pre-reform tax rate variation existed in foreign-invested firms. We will also exclude small businesses because of their different tax treatment after the reform. A problem accompanied by the large variation in pre-reform tax rates is that the statutory tax rates for specific firms are undocumented. To solve this problem, we use a method similar to that by Lin, Mills and Zhang (2014). Chandra and Long (2013), and Chen (2017) also use the similar method and data to estimate the value-added tax rate for firms or counties. Specifically, we use the average of corporate taxes payable divided by pre-tax profit (ET) between 2005 and 2007 as the estimated pre-reform tax rate (EPRT) and apply it to each firm:

15

(1)

Using the average can mitigate the effects of fluctuations such as the temporary tax incentives or penalties. Moreover, China’s prefecture leaders, who influence local corporate tax rates heavily, leave their position in less than four years on average, which makes estimated tax rates based on longer periods less reliable. To check if there is a convergence after the reform, we also obtain an estimated post-reform tax rate (EPOT) as the average of corporate taxes payable divided by pre-tax profit between 2008 and 2009. A simple regression of (EPOT−EPRT) on the EPRT with standard errors clustered at the firms yields coefficients of −0.61 and −0.7 for the possible tax increase and tax cut groups, significant under the 1% level, This confirms a larger tax increase for lower EPRT firms in the tax increase group and a larger tax cut for higher EPRT firms in the tax cut group. Estimation Strategy We estimate the following event-study regressions for the whole sample, the tax increase group and the tax cut group, respectively, similar to Cabral, Geruso and Mahoney (2014) (2) where Y can be cost of goods sold plus expenses scaled by sales revenue (expense ratio), or pretax profit scaled by sales revenue (profit margin). Normalization with sales revenue should be better than normalization with total assets. The reason is that sales revenue is usually measured very accurately and can account for the effect of tax rate on profitability beyond manipulation appropriately, whereas the total assets for non-publicly traded companies probably reflect more on historical value rather than current value. Our method for detecting abnormal cost of goods sold and expenses is similar to that of Lin et al. (2014). However, trade credit is not added here 16

because we do not find any effect on trade credit. Notably, an increase in accounts receivable relative to accounts payable for firms facing tax cuts is not necessary for tax purposes. With increased cash flows, firms with greater tax cuts can extend more trade credit to their customers than firms with lower tax cuts or tax increases can; moreover, with diminished cash flows, firms with larger tax increases are more likely to delay payment to their suppliers than firms with smaller tax increases or tax cuts.9 We further consider an abnormal cost of goods sold as a signal of tax avoidance, because manipulation of the cost of goods sold is widespread in China, which is typically conducted through related-party transactions. The cost of goods sold is usually substantially higher than expenses; therefore, ignoring the cost of goods sold would engender a severe underestimation of tax avoidance. After controlling the interaction terms of industry fixed effects/prefecture fixed effects with the post-reform dummy, we can mitigate the possibility that the abnormal costs result from real investment activities. Pre-tax profit margins include non-operating profit and other manipulations without changes to expense ratios; therefore can capture these additional dimensions of tax avoidance. EPRT*Year are our main variables of interest, which tells us that the how much abnormal cost of goods sold plus expenses or profit would be reported after the reform by firms with high pre-reform tax rates, compared with those with a lower pre-reform tax rates. Year dummy variables include 2005, 2006, 2008, and 2009. We treat 2007 as the base year, and hence there is not an interaction term of EPRT and year dummy 2007. The use of EPRT as a treatment status is equivalent to 0.25−EPRT, the supposed change in tax rates. In this way, we can express the regression equation as:

9

For example, Love, Preve and Sarria-Allende (2007) find that firms that are financially more vulnerable to crises extend less trade credit to their customers. 17

(3)

Although the tax rates for the tax increase group did not converge to 0.25 perfectly, firms with lower EPRTs experienced a larger tax increase than firms with higher EPRTs generally. The percentage point change in tax rate corresponding to 1 percentage point difference in EPRT is smaller for the tax increase group than for the tax cut group, and we probably underestimate the asymmetric reactions. We also consider a setting with EPOT−EPRT as the treatment status, and the results are almost the same. This is not surprising because we already see there is a highly significantly negative relationship between EPRT and EPOT−EPRT. EPOT may be endogenous, because it may contain more measurement errors resulting from less periods for estimation, and there may be more fluctuations resulting from the transition to a new tax schedule or manipulations in the post-reform period. Therefore, we prefer EPRT or 0.25−EPRT to (EPOT−EPRT) as the treatment status. Lin et al. (2014) also compare their estimated pre-reform tax rate with the statutory post-reform tax rate of 0.25 to identify the tax cut and tax increase groups. X includes other covariates, including logged total assets, tangibility (fixed assets scaled by total assets), logged number of employees, logged firm age and export ratio (value of export scaled by total production). Firm size is measure by total assets. We control firm size since a large firm perhaps has the scale of economy or monopoly power, or has a lower growth than small firms. With more fixed assets as collateral, a firm can probably obtain loans more easily, hence tangibility can be an important determinant in obtaining loans. Furthermore, tangibility is more likely to be exogenous than interest payment used by some prior studies. Besides, it can also capture the differential treatments of depreciation (Chen et al. 2010). More employees may 18

make collusion in tax avoidance harder, as Kleven, Kreiner and Saez (2016) and Mao and Ji (2014) confirmed, although it is not the case in our study. Older firms may be more tax sophisticated than younger firms as documented by Chen et al. (2010), and also probably have better political connections or experience a lower growth than newly established firms, hence they may have a lower reported profit. Export may make transfer pricing easy. However, products exported by Chinese firms are mainly labor-intensive and hence are not very profitable. Moreover, the negative shock of 2008 global economic crisis may affect the profitability of China’s export. More discussions on the changes in international trade resulting from the 2008 crisis can be found in (Gawande, Hoekman and Cui 2015).

are year fixed effects.

are firm

fixed effects, which can account for the effects of all the time-invariant observed or unobserved factors. Therefore, we do not need to control industry fixed effects or prefecture fixed effects because they are almost time-invariant. In a robustness check, we add the interaction terms of industry fixed effects and the postreform dummy, in order to account for the possible industry-wide changes (or prefecture-wide) in expenses and profitability over time. Note that in this way we will probably lose some useful variations. Additionally, we also add the interaction term of prefecture fixed effects and the post-reform dummy. We find that the estimates of major interest are identical, implying real investment activities have little impact on our results. For identification, there should be not significant pre-reform trends based on treatment status. In our regressions, this means that the estimated coefficients of EPRT*2005 and EPRT*2006 cannot be significantly different from zero. As will be seen in the estimation results, this is indeed the case for almost all the regressions.

19

Finally, as posited by Slemrod (2001) and Lin et al. (2014), accounting manipulations are less costly than real investment changes. Real earnings management has direct consequences for cash flow and is, thus, costly to implement. For example, cutting marketing and advertising, canceling price discounts or tightening credit terms, although may save cash in the current period, will negatively affect future sales and cash flows. Although reporting lower cost of goods sold and expenses through overproduction can lead to higher operating margins temporarily, the firm is likely to incur other production costs and holding costs that will lead to higher annual production costs and expenses relative to sales, and lower cash flows from operations given the resulting sales levels. By contrast, some expenses may be managed through accounting methods without direct effects on cash flows. For example, classifying more fixed manufacturing overhead to be period cost, expensing rather than capitalizing items to the greatest extent possible, and increasing the bad debt provision to the maximum allowed will result in a lower level of reported profit for the current period, but have very limited consequences for cash flows. Therefore, the large change in profitability or expense ratio in a short period probably reflects accounting manipulation. Ⅳ. DATA AND DESCRIPTIVE STATISTICS The firm data comes from the proprietary Annual Survey of Industrial Firms collected by China’s National Bureau of Statistics (NBS). The 2009 survey is the most recent one available with necessary variables for this research. It covers all state-owned manufacturing firms regardless of size and non-state-owned firms with an annual sales revenue above 5 million yuan (approximately 700 thousand U.S. dollars). As Cai and Liu (2009) explain, the NBS collects the data to compute the GDP of China. To accomplish this purpose, every industrial firm in the dataset is required to file an annual report of production activities and accounting 20

information with the NBS. The NBS has implemented standard procedures in calculating the national income account since 1995 and has double checking procedures for consistency in the accounting information. Specifically, since we focus on over-reporting cost of goods sold and expenses and under-reporting profit for tax purpose in this paper, it is not necessary to obtain the underlying “true” profit as needed by many other papers. What we need is that there are not systematic deviations between the profits or expenses reported to the NBS and those reported to the tax authorities. There is no incentive for firms spending additional costs to provide inconsistent information to the NBS and the tax authorities. On the one hand, it is not necessary for firms to report a higher cost of goods sold plus expenses or a lower profit to the NBS than those reported to the tax authorities since it will not alleviate the tax burden. On the other hand, it is also not necessary for firms to report a lower cost of goods sold plus expenses or a higher profit to the NBS than those reported to the tax authorities because the NBS will not penalize the underreporting of profits, and firms will risk being penalizing by tax authorities once the gap is detected. Although firms probably keep a book for management accounting, they do not need to provide information in this book to either the NBS or the tax authorities. Additionally, there is no evidence documenting a systematic gap between the numbers reported to the NBS and those reported to the tax authorities. Following Mao and Ji (2014), we use the 2005–2009 datasets which contain the consistent variables that we need in this study. To obtain a clean sample, we use the similar criteria in Cai and Liu (2009) and Li and Ma (2015). We drop observations with missing covariates; or observations for which one of the following is true:

21

(i) the sales revenue is below 5 million yuan, (ii) the value of total assets is below 1 million yuan, and (iii) the number of employees is less than 8; or observations that have a negative value for one of the following variables: (i) total assets minus liquid assets, (ii) total assets minus total fixed assets, (iii) total assets minus net value of fixed assets, (iv) accumulated depreciation minus current depreciation. We drop possible small businesses because these firms enjoy a post-reform tax rate of 20% rather than 25%, based on the criteria specified in the law: pre-tax profit is less than or equal 300 thousand yuan, total assets are less than or equal 30 million yuan and the number of employees is less than or equal 100 in year 2008 and 2009. We also drop observations with an EPRT larger than 0.5 (usually resources industries or restricted industries), with an EPRT less than 0.01 (firms probably enjoying tax holidays which can be carried over after the reform), with an expense ratio larger than 1.5 (about 99 percentile), and with an export ratio larger than 2. Our results are not sensitive to these cutoffs, such as an expense ratio of 2, or export ratios of 1.5 and 3. Our sample does not include joint-stock companies and foreign-invested firms. For the former, it is not clear they are closer to private firms or SOEs/collective firms regarding their objectives; for the latter, the majority of them faced a very low pre-reform tax rate and hence is not feasible for our asymmetric analysis. Therefore, our sample contains two types of firms: private firms and SOEs/collective firms, taking up 90 percent and 10 percent of the 22

sample, respectively. The definition of ownership is based on registration type in 2006, although the definition based on paid-in capital does not change our results. Finally, we obtain a panel dataset between 2005 and 2009 with 48775 firm by year observations. The number of firms in my sample is 9755, which is approximately 5 fold the usual sample size of publicly traded companies in the United States used in prior literature. In Table 1, we show the descriptive statistics of the main variables for the possible tax increase group (EPRT

0.25). We note that the mean and median expense ratios increased by 1

and 0.6 percentage points respectively, implying possible increased tax avoidance activities. We do not see a comparable change in the profit margin. The logged total assets and number of employees increased, reflecting the secular expansion of survived firms. The export ratio declined, probably resulting from the global recession. Tangibility did not change after the reform. In our regressions, we notice that adding these covariates and firm fixed effects provide a large explanatory power as shown by the large adjusted R-square.

Insert Table 1 here

Table 2 shows the descriptive statistics of the main variables for the possible tax cut group (EPRT > 0.25). The mean and median expense ratios decreased by 0.5 and 0.8 percentage points respectively, while the mean and median profit margins increased by 0.9 and 0.7 percentage points, respectively. However, these effects are noisy and as we control more covariates and fixed effects, these effects disappear. Other covariates show a similar trend of changes as in Table 1. 23

Insert Table 2 here

Ⅴ. RESULTS Before presenting our main results, we refer to Figure 1, which presents the estimated tax rates for the two groups. The low EPRT group experienced a sharp increase from 2007 to 2008, whereas the high EPRT group experienced a sharp drop over the same period. We observe that the percentage point change for the low EPRT group is smaller on average; however, this may bias down our estimated asymmetric reactions. Additionally, the percent change corresponding to 1 percentage point change in tax rate is larger for the low EPRT group than the high EPRT group, and hence the percent change in tax rates for the low EPRT group is larger than for the high EPRT group. In a robustness check, we consider the logarithm of the EPRT as a treatment status to address this issue, and the results are similar. Moreover, a simple regression of (EPOT−EPRT) on the EPRT with standard errors clustered at the firms yields coefficients of −0.61 and −0.7 for the possible tax increase and tax cut groups, respectively, which are significant under the 1% level. This confirms a larger tax increase for lower EPRT firms in the tax increase group and a larger tax cut for higher EPRT firms in the tax cut group. Finally, we conduct a first-stage regression of estimated tax rates on the interaction terms of EPRT and four year dummies, year dummies and firm fixed effects. For both the tax increase and tax cut groups, we observe negative estimated coefficients of EPRT*2008 and EPRT*2009, which are significantly lower than the estimated coefficients of EPRT*2005 and EPRT*2006. Therefore, with higher EPRTs, firms in the tax increase group experienced small increases in the estimated 24

tax rate following the reform, and firms in the tax cut group experienced larger increases in the estimated tax rate following the reform. Main Results Table 3 presents the results for the tax increase group (EPRT

0.25), and tax cut group

(EPRT > 0.25), using the expense ratio as the dependent variable. For both groups, the estimated coefficients of EPRT*2005 and EPRT*2006 are not significant, which demonstrate no significant pre-reform trend. We find a strong effect of the reform on the expense ratio for the tax increase group. Firms with a 1 percentage point lower EPRT (corresponding to a 0.6 percentage point higher post-reform tax rate) were associated with a 0.11 percentage point increase in the expense ratio—approximately 0.12 percent of the average expense ratio prior to the reform. The average sales revenue in 2007 was 1.4 billion yuan. Considering two firms with an EPRT of 0.1 and 0.2, the increase in cost of goods sold plus expenses for the 0.1 firm relative to the 0.2 firm was approximately 15 million yuan. Therefore, this effect is large and economically meaningful. The results of the other covariates are generally as predicted in the Empirical Strategy part.

Insert Table 3 here

We do not find a significant effect for the tax cut group in 2008. Firms with large tax cuts did not reduce tax avoidance compared with firms with small tax cuts. This suggests that tax cut firms are more likely to take the windfall as granted, whereas tax increase firms attempt 25

to avoid losses. Specifically, firms perceive losses more strongly than gains, as suggested by Kahneman and Tversky (1979). Although a significant effect of EPRT*2009 is observed under the 5% level, the p-value is close to 0.05. This effect can be explained as follows: For firms experiencing a larger decline in tax rates (compared with those with a smaller decline in tax rates), tax authorities lost a large amount of revenue in 2008 from them. This likely motivated the authorities to conduct more tax audits on the firms. Notice that statistical power is not an explanation of our differential results for the tax increase and tax groups. The number of observations is larger for the tax group, but there are not significant estimates for it. To the best of our knowledge, this is the first paper documenting firms’ asymmetric reactions to corporate tax changes in terms of tax avoidance/compliance. Table 4 reports parallel results derived using the profit margin as the dependent variable. Similarly, we find a strong effect from the reform on profit margins for the tax increase group, but not for the tax cut group. Firms with a 1 percentage point lower EPRT is associated with a 0.09 percentage point drop in the profit margin, nearly 1.1 percent of the average profit margin prior to the reform. Because the average sales revenue was 1.4 billion yuan in 2007, firms with an EPRT of 0.1 reported 13 million yuan less than firms with an EPRT of 0.2. Additionally, we do not find any significant pre-reform trend for either the full sample or the two groups. Results reported in table 3 and table 4 support H1a.

Insert Table 4 here

26

Lin et al. (2014) consider changes in accrued expenses as a simple shift between 2007 and 2008. However, we do not find the effects of EPRT*2005 and EPRT*2006 to be different from 0, the normalized effect for 2007. Moreover, the effects of EPRT*2009 and EPRT*2008 are similar. This indicates that the tax reform changed firms’ calculation permanently. There are both benefits and costs associated with tax avoidance. When tax rates increased, the benefits of tax avoidance increased while the costs remained constant, thus rendering tax-aggressive activities more worthwhile. We note that, on average, firms with higher EPRTs tended to have higher expense ratios and lower profit margins prior to the tax reform. Therefore, firms with lower EPRTs had a relatively large space to report more cost of goods sold plus expenses. Lin et al. (2014) analyze a substantially smaller sample and do not use 2009 data, which may explain why our findings are different. Our findings also support the use of expense ratio and profit margin in detecting tax avoidance. Regarding the tax increase group, we find an increasing expense ratio and a diminishing profit margin associated with the EPRT. If the changes in expense ratios result from real investment activities, the groups with a EPRTs close to 0.25 would likely use almost all the intermediate goods from firms with lower EPRTs, which in turn would use almost all the intermediate goods from firms with next lower EPRTs. This would require firms with the lowest EPRT to exhibit increases in the expense ratios from real investment activities, which seems impossible. Moreover, we add the interaction term of industry fixed effects, prefecture fixed effects with the post-reform dummy (2008–2009) in a robustness check, and the results are still intact. Agent Problems and Audit: Subsample Analysis

27

Private firms usually regard profit maximization or similar economic achievements as their main objective, whereas SOE/collective firms generally pursue policy objectives in addition to profit maximization and hence are less likely to conduct tax avoidance activities. On the other hand, although SOEs are nominally owned by the people, such firms largely serve the interests of insiders, who benefit from increased cash flows from avoided taxes. SOEs often have minority shareholders, who likewise benefit from tax sheltering. Furthermore, the penalty imposed by tax authorities following a tax audit is relatively small for SOEs because of their political connections. Our findings in Table 5 compare the differential reactions to the tax reform regarding expense ratio and profit margin between these two types of firms in the tax increase group. We find the asymmetric reactions exist only in private firms. Since private firms take up 90 percent of the whole sample, the results for private firms mirror the results for the whole sample in Table 3. Importantly, the estimates of EPRT*2008 and EPRT*2009 for SOE/collective firms are of the “wrong sign” in terms of tax avoidance/compliance. We also consider a setting with triple interaction terms EPRT*Year*SOE, interaction terms EPRT*Year, EPRT*SOE, and SOE*Year, where SOE is an dummy variable equal to 1 for SOE and collective firms. We find that estimates of EPRT*2008*SOE and EPRT*2009*SOE are positive when using the expense ratio as the dependent variable with a p-value close to 0.1. Overall, these results mean that SOEs/collective firms are probably less likely to conduct tax avoidance activities.

Insert Table 5 here

28

To check the robustness of the results presented in Table 5, we also run the regressions for subgroups of firms defined by the controlling share of paid-in capital. The results shown in Appendix Table 1 are very similar to those in Table 5, since ownership based on the two kinds of definition is identical for most firms. An alternative explanation of our results is audit by tax authorities. For example, this tax reform might impact tax audit differentially based on pre-reform tax rates such as what documented by Khan, Khwaja, and Olken (2016) in Pakistan. This explanation is not quite plausible because it is hard to explain why there was not differential impact among the tax cut group. Furthermore, in Table 6, we divide the tax increase group into smaller firms and bigger firms. Although the literature usually suggests that bigger firms are more likely to be audited by tax authorities since tax audit incurs large fixed costs, our empirical results are actually identical for these two groups. Our findings are consistent with Carrillo, Pomeranz and Singhal (2017), who also find that audit toward tax avoidance is limited in developing countries. Therefore, our results are likely to be driven by tax avoidance/compliance by firms rather than by audit by tax authorities.

Insert Table 6 here

Further Robustness Check Using Logged EPRT as the Treatment Status

29

We notice that the percent change corresponding to 1 percentage point change in tax rate for firms with a lower EPRT is larger than that for firms with a higher EPRT, because of a lower base tax rate for the former group. To address this issue, we use logged EPRT as the treatment status in a robustness check. As shown in Table 7, we still find that firms in the tax increase group exhibit the strong reactions to the tax reform rather than those in the tax cut group, and only private firms exhibit such asymmetric reactions. Using profit margin as the dependent variable gives similar results, and the results are not reported here for brevity.

Insert Table 7 here

More Interaction Terms If the changes in expense ratios and profit margins result from real expenditure on intermediate goods, we should probably find such changes are industry wide or prefecture wide. In a robustness check, we add the interaction terms of industry fixed effects and the post-reform dummy, and the interaction term of prefecture fixed effects and the post-reform dummy, in order to account for the possible industry-wide changes (or prefecture-wide) in expense ratios or profit margins over time. As shown in Table 8, we find that the estimates of major interest are identical, implying real investment activities have little impact on our results. Notice that in this way we will probably lose some useful variations because some tax rate changes are industry (or prefecture) specific.

30

Insert Table 8 here

Anticipatory Reactions In studies using a policy shock, there is a regular concern about anticipatory reactions. For example, firms perhaps switch ownership type, change location or industry to enjoy a preferential tax rate after the reform. However, since a quite uniform tax treatment applies to firms nationwide regardless of ownership type, location or industry, there are very limited such arbitrage opportunities.10 Indeed, there is only a tiny share of firms changing their ownership, location or industry. Regressions adding the interaction terms of industry dummies and the postreform period dummy or the interaction terms of prefecture dummies and the post-reform period dummy give the similar results as already mentioned. Our results are also similar either using the ownership status in 2006 or time-varying ownership status. Ⅵ. CONCLUSION Considering a plausibly exogenous variation from a corporate tax reform in China, we find a strong negative relationship between the EPRT and the post-reform cost of goods sold plus expenses within the possible tax increase group (EPRT

0.25), indicating an increase in

tax avoidance activities by firms experiencing larger tax increases compared with firms with smaller tax increases. Firms with a 1 percentage point lower EPRT (corresponding to a 0.6 percentage point higher post-reform tax rate) is associated with a 0.11 percentage point increase in the expense ratio. By contrast, we do not find a similar effect within the possible tax cut group (EPRT > 0.25). Firms with larger tax cuts did not reduce their tax avoidance activities 10

Although some firms were granted a grace period, the schedule is identical across the nation. 31

compared with firms with smaller tax cuts. Therefore, these results imply that firms perceive losses more strongly than gains, as suggested by Kahneman and Tversky (1979). Furthermore, these asymmetric reactions represent private firms rather than SOEs and collective firms, suggesting that SOEs and collective firms pursue policy objectives in addition to profit maximization. Corporate tax avoidance and government tax collection involve large costs, particularly for private firms. These findings are valuable in understanding the behavior of firms with different incentives. Overall, our findings highlight the potential costs of increasing or reducing tax rates frequently. Although many economists suggest that a lower tax rate will reduce tax avoidance and hence do not lower government revenues significantly, our findings show that the reduction in tax avoidance is small. Especially, if the government initially increases the tax rate and reduce it later, as regularly happened, there could be a large loss in tax revenues and social welfare.

32

REFERENCES Abdellaoui, M., H. Bleichrodt, and C. Paraschiv. 2007. Loss aversion under prospect theory: A parameter-free measurement. Management Science 53 (10): 1659 – 1674. An, Z. 2012. Taxation and foreign direct investment (FDI): empirical evidence from a quasiexperiment in China. International Tax and Public Finance 19 (5): 660-676. Annuar, H. A., I. A. Salihu, and S. N. S. Obid. 2014. Corporate ownership, governance and tax avoidance: An interactive effects. Procedia-Social and Behavioral Sciences 164: 150160. Bornstein, G., T. Kugler, and A. Ziegelmeyer. 2004. Individual and group decisions in the centipede game: Are groups more “rational” players? Journal of Experimental Social Psychology 40 (5): 599-605. Bornstein, G., and I. Yaniv. 1998. Individual and group behavior in the ultimatum game: Are groups more “rational” players? Experimental Economics 1 (1): 101-108. Cabral, M., M. Geruso, and N. Mahoney. 2014. Does privatized health insurance benefit patients or producers? Evidence from Medicare Advantage: National Bureau of Economic Research. Cai, H., and Q. Liu. 2009. Competition and corporate tax avoidance: Evidence from Chinese industrial firms. The Economic Journal 119 (537): 764-795. Carrillo, P., D. Pomeranz, and M. Singhal. 2017. Dodging the taxman: Firm misreporting and limits to tax enforcement. American Economic Journal: Applied Economics 9(2): 144164. Chan, K. H., P. L. Mo, and A. Y. Zhou. 2013. Government ownership, corporate governance and tax aggressiveness: evidence from China. Accounting & Finance 53 (4): 1029-1051. Chandra, P., and C. Long. 2013. VAT rebates and export performance in China: Firm-level evidence. Journal of Public Economics 102: 13-22. Chen, M. K., V. Lakshminarayanan, and L. R. Santos. 2006. How basic are behavioral biases? Evidence from capuchin monkey trading behavior. Journal of Political Economy 114 (3): 517-537. Chen, S., X. Chen, Q. Cheng, and T. Shevlin. 2010. Are family firms more tax aggressive than non-family firms? Journal of Financial Economics 95 (1):41-61. Chen, X. 2017. The effect of a fiscal squeeze on tax enforcement: Evidence from a natural experiment in China. Journal of Public Economics 147: 62-76. Cheng, H., and Yuanyuan Ma. 2016. Enforcing Government Policy: Privatization and the Weakening Effects of China's One-child Policy. Paper presented at the 2016 AEA Meeting. Cooper, R., G. Gong, and P. Yan. 2010. Dynamic labor demand in china: Public and private objectives: National Bureau of Economic Research. Desai, M. A., and D. Dharmapala. 2006. Corporate tax avoidance and high-powered incentives. Journal of Financial Economics 79 (1): 145-179. ———. 2009. Corporate tax avoidance and firm value. Review of Economics and Statistics 91 (3): 537-546. Dhaliwal, D., and S.-w. Wang. 1992. The effect of book income adjustment in the 1986 alternative minimum tax on corporate financial reporting. Journal of Accounting and Economics 15 (1):7-26. 33

Dyreng, S. D., M. Hanlon, and E. L. Maydew. 2010. The effects of executives on corporate tax avoidance. The Accounting Review 85 (4): 1163-1189. Faccio, M., R. W. Masulis, and J. McConnell. 2006. Political connections and corporate bailouts. The Journal of Finance 61 (6): 2597-2635. Gawande, K., B. Hoekman, and Y. Cui. 2015. Global supply chains and trade policy responses to the 2008 crisis. The World Bank Economic Review 29 (1): 102-128. Guenther, D. A. 1994. Earnings management in response to corporate tax rate changes: Evidence from the 1986 Tax Reform Act. The Accounting Review 69(1): 230-243. Homonoff, T. A. 2013. Can small incentives have large effects? the impact of taxes versus bonuses on disposable bag use. Industrial Relations Section, Princeton University Working Paper (575). Johannesen, N., and G. Zucman. 2014. The end of bank secrecy? An evaluation of the G20 Tax Haven Crackdown. American Economic Journal: Economic Policy 6(1): 65–91. Kahneman, D., and A. Tversky. 1979. Prospect theory: An analysis of decision under risk. Econometrica 47(2): 263-291. Khan, A., A. Khwaja, and B. Olken. 2016. Tax farming redux: Experimental evidence on performance pay for tax collectors. Quarterly Journal of Economics 131 (1): 219–271 Kleven, H. J., C. T. Kreiner, and E. Saez. 2016. Why can modern governments tax so much? An agency model of firms as fiscal intermediaries. Economica 83 (330): 219-246. Lennox, C., P. Lisowsky, and J. Pittman. 2013. Tax aggressiveness and accounting fraud. Journal of Accounting Research 51 (4): 739-778. Li, L., and G. Ma. 2015. Government size and tax evasion: evidence from China. Pacific Economic Review 20 (2): 346-364. Li, O. Z., and N. Cai. 2011. Capital market research in taxation: Do it in China! China Journal of Accounting Research 4 (1): 1-7. Lin, K. Z., L. F. Mills, and F. Zhang. 2014. Public versus private firm responses to the tax rate reduction in China. The Journal of the American Taxation Association 36 (1): 137-163. Lopez, T. J., P. R. Regier, and T. Lee. 1998. Identifying tax-induced earnings management around TRA 86 as a function of prior tax-aggressive behavior. The Journal of the American Taxation Association 20 (2): 37-56. Love, I., L. A. Preve, and V. Sarria-Allende. 2007. Trade credit and bank credit: Evidence from recent financial crises. Journal of Financial Economics 83 (2): 453-469. Manzon, G. B., and G. A. Plesko. 2002. Relation between financial and tax reporting measures of income, Tax Law Review 55: 176-195. Mao, C., and L. Ji. 2014. Impact of the tax rate on foreign invested firms' tax avoidance. World Economy. Slemrod, J. 2001. A general model of the behavioral response to taxation. International Tax and Public Finance 8 (2): 119-128.

34

Figure 1: The median estimated tax rates for the possible tax increase group (EPRT

0.25) and

tax cut group (EPRT > 0.25)

EPRT is the estimated pre-reform income tax rate in 2005-2007, using the average of corporate taxes payable divided by pre-tax profit. This graph shows that there is a sharp increase from 2007 to 2008 for the low EPRT group and a sharp drop from 2007 to 2008 for the high EPRT group.

35

Figure 2: The standard deviations of estimated tax rates for each year

The estimated tax rate is the corporate taxes payable divided by pre-tax profit. This graph shows that the standard deviations of the estimated tax rates after the reform are smaller than those before the reform, for the whole sample and firms with different types of ownership.

36

Table 1 Descriptive statistics for the possible tax increase group (EPRT ≤ 0.25)

Variable Expense/Sales Profit/Sales Logged total assets Tangibility No. of employees Firm age Export ratio N

Before reform Mean(Std) Median 0.901(0.121) 0.919 0.080(0.079) 0.061 9.879(1.335) 9.680 0.367(0.210) 0.343 292(1740) 110 9.26(8.91) 7 0.116(0.279) 0 10098 10098

After reform Mean(std) Median 0.911(0.111) 0.925 0.081(0.078) 0.063 10.283(1.356) 10.104 0.361(0.221) 0.331 317(1825) 120 11.31(8.64) 9 0.107(0.265) 0 6732 6732

Note: Expense/Sales is the cost of goods sold plus operating expense, management expense and financial expense scaled by sales revenue; Profit/Sales is the pre-tax profit scaled by sales revenue; tangibility is fixed assets scaled by total assets; export ratio is export scaled by total production. Before reform is 2005-2007, and After reform is 2008-2009.

37

Table 2 Descriptive statistics for the possible tax cut group (EPRT >0.25)

Variable Expense/Sales Profit/Sales Logged total assets Tangibility No. of employees Firm age Export ratio N

Before Reform Mean(Std) Median 0.924(0.110) 0.948 0.057(0.063) 0.039 9.780(1.225) 9.680 0.359(0.201) 0.334 267(1930) 117 9.75(10.49) 7 0.151(0.315) 0 19167 19167

After reform Mean(std) Median 0.919(0.107) 0.940 0.066(0.074) 0.046 10.156(1.243) 10.003 0.350(0.208) 0.317 282(1991) 125 11.59(10.24) 9 0.143(0.304) 0 12778 12778

Note: Expense/Sales is the cost of goods sold plus operating expense, management expense and financial expense scaled by sales revenue; Profit/Sales is the pre-tax profit scaled by sales revenue; tangibility is fixed assets scaled by total assets; export ratio is export scaled by total production. Before reform is 2005-2007, and After reform is 2008-2009.

38

Table 3 The impact of the reform on the total expense scaled by sales revenue This table presents the regression results:

is Expense/Sales, calculated as the cost of goods sold plus operating expense, management expense and financial expense scaled by sales revenue. Column (1) reports the results of the full sample, column (2) reports the results of the tax increase group (EPRT ≤ 0.25) and column (3) reports the results of the tax cut group (EPRT >0.25). EPRT is the estimated pre-reform income tax rate in 2005-2007, using the average of corporate taxes payable divided by pre-tax profit. EPRT*2005, EPRT*2006, EPRT*2008, EPRT*2009 are interaction terms of EPRT and four year dummies. are firm fixed effects, are year fixed effects, includes logged total assets, tangibility (fixed assets scaled by total assets), logged number of employees, logged firm age and export ratio (export scaled by total production). The standard errors are clustered at the firms. *significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

EPRT ≤ 0.25

Dependent variable: Expense/Sales

(2) EPRT*2005

0.00407 (0.0420) 0.0535 (0.0353) -0.111*** (0.0371) -0.113*** (0.0386) -0.0116*** (0.0031) 0.0104 (0.0086) 0.0074** (0.0036) 0.0026 (0.0041) 0.0063 (0.0081) 16830 0.424

EPRT*2006 EPRT*2008 EPRT*2009 Logged total assets Tangibility Logged employees Logged firm age Export ratio N Adj. R-sq

39

EPRT >0.25 (3) 0.00182 (0.0584) -0.00289 (0.0506) -0.0644 (0.0521) -0.120** (0.0591) -0.0142*** (0.0024) 0.0012 (0.0062) 0.0070*** (0.0026) 0.0035 (0.0025) -0.0032 (0.0050) 31945 0.434

Table 4 The impact of the reform on the pre-tax profit scaled by sales revenue This table presents the regression results:

is Profit/Sales, calculated as the pre-tax profit scaled by sales revenue. Column (1) reports the results of the full sample, column (2) reports the results of the tax increase group (EPRT ≤ 0.25) and column (3) reports the results of the tax cut group (EPRT >0.25). EPRT is the estimated pre-reform income tax rate in 2005-2007, using the average of corporate taxes payable divided by pre-tax profit. EPRT*2005, EPRT*2006, EPRT*2008, EPRT*2009 are interaction terms of EPRT and four year dummies. are firm fixed effects, are year fixed effects, includes logged total assets, tangibility (fixed assets scaled by total assets), logged number of employees, logged firm age and export ratio (export scaled by total production). The standard errors are clustered at the firms. *significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

EPRT ≤ 0.25

Dependent variable: Profit/Sales

(2) EPRT*2005

0.0155 (0.0228) -0.0157 (0.0187) 0.0857*** (0.0183) 0.0950*** (0.0226) 0.0137*** (0.0020) -0.0086* (0.0050) -0.0023 (0.0020) -0.0015 (0.0026) -0.0050 (0.0037) 16830 0.626

EPRT*2006 EPRT*2008 EPRT*2009 Logged total assets Tangibility Logged employees Logged firm age Export ratio N Adj. R-sq

40

EPRT >0.25 (3) -0.000680 (0.0238) 0.0219 (0.0208) 0.00692 (0.0251) 0.0280 (0.0258) 0.0132*** (0.0016) -0.0065* (0.0034) -0.0039*** (0.0014) -0.0032** (0.0015) -0.0029 (0.0019) 31945 0.648

Table 5 The impact of the reform on Expense/Sales or Profit/Sales: Private firms versus SOEs/collective firms in the tax increase group This table presents the regression results:

is Expense/Sales or Profit/Sales. Expense/Sales is calculated as the cost of goods sold plus operating expense, management expense and financial expense scaled by sales revenue. Profit/Sales is calculated as the pre-tax profit scaled by sales revenue. EPRT is the estimated pre-reform income tax rate in 2005-2007, using the average of corporate taxes payable divided by pre-tax profit. EPRT*2005, EPRT*2006, EPRT*2008, EPRT*2009 are interaction terms of EPRT and four year dummies. are firm fixed effects, are year fixed effects, includes logged total assets, tangibility (fixed assets scaled by total assets), logged number of employees, logged firm age and export ratio (export scaled by total production). The standard errors are clustered at the firms. *significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

Dependent variable:

EPRT*2005 EPRT*2006 EPRT*2008 EPRT*2009 Other covariates Firm FE N Adj. R-sq

Expense/Sales

Expense/Sales

Profit/Sales

Profit/Sales

Private

SOE/collective

Private

SOE/collective

(1) -0.0109 (0.0441) 0.0382 (0.0365) -0.127*** (0.0387) -0.131*** (0.0397) Yes Yes 15185 0.398

(2) 0.133 (0.132) 0.197 (0.137) 0.0730 (0.133) 0.0991 (0.157) Yes Yes 1645 0.578

(3) 0.0327 (0.0213) -0.0035 (0.0194) 0.0958*** (0.0189) 0.102*** (0.0236) Yes Yes 15185 0.620

(4) -0.175 (0.157) -0.136* (0.0695) -0.0500 (0.0691) 0.0050 (0.0740) Yes Yes 1645 0.657

41

Table 6 The impact of the reform on Expense/Sales or Profit/Sales Smaller firms versus bigger firms in the tax increase group This table presents the regression results:

is Expense/Sales or Profit/Sales. Expense/Sales is calculated as the cost of goods sold plus operating expense, management expense and financial expense scaled by sales revenue. Profit/Sales is calculated as the pre-tax profit scaled by sales revenue. We divide the tax increase group (EPRT ≤ 0.25) into smaller firms and bigger firms based on the median total assets. EPRT is the estimated pre-reform income tax rate in 2005-2007, using the average of corporate taxes payable divided by pre-tax profit. EPRT*2005, EPRT*2006, EPRT*2008, EPRT*2009 are interaction terms of EPRT and four year dummies. are firm fixed effects, are year fixed effects, includes logged total assets, tangibility (fixed assets scaled by total assets), logged number of employees, logged firm age and export ratio (export scaled by total production). The standard errors are clustered at the firms. *significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

Dependent variable:


Expense/Sales

Expense/Sales

Profit/Sales

Profit/Sales

Smaller firms

Bigger firms

Smaller firms

Bigger firms

(1) 0.0098 (0.0600) 0.0663 (0.0525) -0.0967* (0.0552) -0.112* (0.0575) Yes Yes 8415 0.378

(2) -0.0393 (0.0228) 0.0440 (0.0564) -0.0958* (0.0570) -0.106* (0.0616) Yes Yes 8415 0.475

(3) 0.0164 (0.0277) -0.0169 (0.0244) 0.0578** (0.0231) 0.0867*** (0.0269) Yes Yes 8415 0.600

(4) 0.0141 (0.0461) -0.0104 (0.0329) 0.110*** (0.0317) 0.0904** (0.0373) Yes Yes 8415 0.663

42

Table 7 The impact of the reform on the total expense scaled by sales revenue (with logged EPRT as the treatment status) This table presents the regression results:

is Expense/Sales, calculated as the cost of goods sold plus operating expense, management expense and financial expense scaled by sales revenue. Column (1) reports the results of tax increase group (EPRT ≤ 0.25) of private firms, column (2) reports the results of the tax cut group (EPRT >0.25) of private firms, column (3) reports the results of tax increase group (EPRT ≤ 0.25) of SOEs/collective firms, and column (4) reports the results of the tax cut group (EPRT >0.25) of SOEs/collective firms. EPRT is the estimated pre-reform income tax rate in 2005-2007, using the average of corporate taxes payable divided by pre-tax profit. Logged EPRT*2005, Logged EPRT*2006, Logged EPRT*2008, Logged EPRT*2009 are interaction terms of Logged EPRT and four year dummies. are firm fixed effects, are year fixed effects, includes logged total assets, tangibility (fixed assets scaled by total assets), logged number of employees, logged firm age and export ratio (export scaled by total production). The standard errors are clustered at the firms. *significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

Dependent variable:

Expense/Sales Logged EPRT*2005 Logged EPRT*2006 Logged EPRT*2008 Logged EPRT*2009 Other covariates Firm FE N Adj. R-sq

EPRT ≤ 0.25

EPRT >0.25

EPRT ≤ 0.25

EPRT >0.25

Private

Private

SOE/collective

SOE/collective

(1) -0.000390 (0.00495) 0.00204 (0.00421) -0.0119** (0.00477) -0.0151*** (0.00441) Yes Yes 15185 0.398

(2) 0.00308 (0.0190) -0.00123 (0.0165) -0.0159 (0.0167) -0.0388** (0.0189) Yes Yes 28290 0.358

(3) 0.0184 (0.0146) 0.0263 (0.0182) 0.0140 (0.0173) 0.0104 (0.0200) Yes Yes 1645 0.578

(4) 0.00367 (0.0657) 0.0304 (0.0557) -0.0434 (0.0630) -0.0355 (0.0704) Yes Yes 3655 0.619

43

Table 8 The impact of the reform on the total expense scaled by sales revenue, adding more interaction terms This table presents the regression results:

is Expense/Sales, calculated as the cost of goods sold plus operating expense, management expense and financial expense scaled by sales revenue. Column (1) reports the results of tax increase group (EPRT ≤ 0.25) of private firms, column (2) reports the results of the tax cut group (EPRT >0.25) of private firms, column (3) reports the results of tax increase group (EPRT ≤ 0.25) of SOEs/collective firms, and column (4) reports the results of the tax cut group (EPRT >0.25) of SOEs/collective firms. EPRT is the estimated pre-reform income tax rate in 2005-2007, using the average of corporate taxes payable divided by pre-tax profit. EPRT*2005, EPRT*2006, EPRT*2008, EPRT*2009 are interaction terms of EPRT and four year dummies. are firm fixed effects, are year fixed effects, includes logged total assets, tangibility (fixed assets scaled by total assets), logged number of employees, logged firm age and export ratio (export scaled by total production). can be industry fixed effects or prefecture fixed effects or both. Post is the dummy for post-reform period (2008-2009). The standard errors are clustered at the firms. *significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

Dependent variable: Expense/Sales EPRT*2005 EPRT*2006 EPRT*2008 EPRT*2009 Other covariates Firm FE Industry FE*Post Prefecture FE*Post N Adj. R-sq

EPRT ≤ 0.25

EPRT >0.25

EPRT ≤ 0.25

EPRT >0.25

(1)

(2)

(3)

(4)

0.00342 (0.0421) 0.0529 (0.0355) -0.116*** (0.0375) -0.117*** (0.0388) Yes Yes Yes No 16830 0.427

0.000837 (0.0584) -0.00206 (0.0506) -0.0574 (0.0526) -0.119** (0.0597) Yes Yes Yes No 31945 0.437

0.00322 0.0542 0.0542 (0.0357) -0.107*** (0.0387) -0.108*** (0.0395) Yes Yes Yes Yes 16830 0.445

0.00208 -0.000357 -0.000357 (0.0508) -0.0837 (0.0531) -0.146** (0.0600) Yes Yes Yes Yes 31945 0.455

44

Appendix Table 1 The impact of the reform on Expense/Sales or Profit/Sales: Private firms versus SOEs/collective firms in the tax increase group (with ownership defined by paid-in capital) This table presents the regression results:

is Expense/Sales or Profit/Sales. Expense/Sales is calculated as the cost of goods sold plus operating expense, management expense and financial expense scaled by sales revenue. Profit/Sales is calculated as the pre-tax profit scaled by sales revenue. EPRT is the estimated pre-reform income tax rate in 2005-2007, using the average of corporate taxes payable divided by pre-tax profit. EPRT*2005, EPRT*2006, EPRT*2008, EPRT*2009 are interaction terms of EPRT and four year dummies. are firm fixed effects, are year fixed effects, includes logged total assets, tangibility (fixed assets scaled by total assets), logged number of employees, logged firm age and export ratio (export scaled by total production). The standard errors are clustered at the firms. *significant at the 10% level; **significant at the 5% level; ***significant at the 1% level.

Dependent variable:


Expense/Sales

Expense/Sales

Profit/Sales

Profit/Sales

Private (1) -0.0352 (0.0502) 0.0376 (0.0413) -0.171*** (0.0421) -0.141*** (0.0444) Yes Yes 11810 0.411

SOE/collective (2) 0.139 (0.149) 0.158 (0.164) 0.0660 (0.154) 0.0296 (0.186) Yes Yes 1155 0.513

Private (3) 0.0450* (0.0238) 0.0087 (0.0223) 0.108*** (0.0209) 0.0973*** (0.0259) Yes Yes 11810 0.634

SOE/collective (4) -0.283 (0.202) -0.186** (0.0742) -0.0581 (0.0812) -0.0059 (0.0830) Yes Yes 1155 0.636

45