Stock Liquidity and Corporate Tax Avoidance: The Tale of Two. Tails

Transcription

1 Stock Liquidity and Corporate Tax Avoidance: The Tale of Two Tails Yangyang Chen Department of Banking and Finance Monash University Leon Zolotoy* Melbourne Business School University of Melbourne First draft: May 2013 Current version: January 2014 ABSTRACT We examine the relation between stock liquidity and corporate tax avoidance. Utilizing a quantile regression framework for the comprehensive sample of US firms, we document the following key results: (i) stock liquidity is positively (negatively) associated with the lower (upper) tail of the tax avoidance distribution; (ii) the effect of stock liquidity on both tails of tax avoidance distribution is stronger for firms with high levels of business uncertainty; and (iii) the effect of stock liquidity on the lower (upper) tail of tax avoidance distribution is mitigated (magnified) for the financially constrained firms. Our findings are consistent with the positive feedback channel where, by making share prices more informative to managers, greater stock liquidity affects firm s tax planning. JEL Classification: G10, G30, M40. Keywords: Stock liquidity; Tax avoidance; Information asymmetry; Financial constraints; Positive feedback. * Corresponding author. Address: 200 Leicester Street, Carlton VIC L.Zolotoy@mbs.edu. Tel:

2 I. INTRODUCTION Corporate tax avoidance has become a substantial issue in both finance and accounting research in recent years. The statutory corporate tax rate in the United States is one of the highest in the world. At the same time, there is striking variation in corporate tax payments among U.S. firms. At one extreme, a substantial share of profitable U.S. corporations pay very little or even no corporate taxes, which makes some observers view corporate tax avoidance as the most important compliance issue in the U.S. tax system today (Desai and Dharmapala, 2006). At the other extreme, approximately one-fourth of U.S. firms pay corporate taxes in excess of the statutory tax rate (Dyreng, Hanlon, and Maydew, 2008), suggesting that these firms engage in little or no tax avoidance. Why do some firms pay little corporate tax while others pay excessive levels? The determinants of corporate tax avoidance are still far from being well understood, and are a fertile ground for further research in this important area (Shackelford and Shevlin, 2001; Hanlon and Heitzman, 2010). We examine the effect of stock liquidity on corporate tax avoidance. Two strands of the previous literature motivate our research question. The first strand examines the determinants of tax avoidance and its implications for shareholder wealth. Studies in this stream of research view tax avoidance as an investment decision along the continuum of tax planning strategies available to managers, where in equilibrium a value-maximizing firm should invest in tax avoidance as long as marginal benefits of tax avoidance exceed its marginal costs (Hanlon and Heitzman, 2010; Edwards, Schwab and Shevlin, 2013). However, similar to other investment decisions, 2

3 managers may deviate from the optimal level of investment in tax avoidance (from the shareholders perspective) by investing either too little or too much in tax planning (Armstrong et al., 2013). On the one hand, overly-conservative tax avoidance (that is, being in the lower end of the continuum of tax planning strategies) results in loss of potential cash-savings (Graham and Tucker, 2006; Cheng et al., 2012). On the other hand, overly-aggressive tax avoidance (that is, being in the upper end of the tax planning continuum) facilitates managerial rent extraction (Desai and Dharmapala, 2006; Kim, Li and Zhang, 2011), and may also result in additional auditor fees, legal fines and reputational costs (Mills, 1998; Wilson, 2009; Graham et al., 2013), and thus will erode shareholder wealth 1. A second strand of related literature examines the effect of stock liquidity on firm s investment decisions. Prior studies suggest that firm managers make investment decisions that are contingent on the information revealed by the firm s stock price (e.g., Subrahmanyam and Titman, 2001; and Khana and Sonti, 2004), and that the sensitivity of firm s investments to stock price is increasing in stock price informativeness (Chen, Goldstein and Weng, 2007). Greater stock liquidity facilitates trading by informed traders (Easley and O Hara, 2004), thereby making share prices 1 Prior studies provide empirical evidence consistent with investors perceiving both overlyconservative and overly-aggressive tax avoidance as the value-destroying activities. For the overlyconservative tax avoidance, Cheng et al. (2012) find that, relative to matched firms, businesses targeted by hedge fund activists exhibit lower tax avoidance levels prior to hedge fund intervention, but experience increases in tax avoidance after the intervention. For the overly-aggressive tax avoidance, Hanlon (2005) document that investors perceive large book-tax differences (book income greater than taxable income) as a red flag and reduce their expectations of future earnings persistence for these firms. In a similar context, Hanlon and Slemrod (2007) find that, on average, a company s stock price declines when there is news about its involvement in tax sheltering. 3

4 more informative to managers and allowing firm management to make better investment decisions, a so-called positive feedback effect (Fang, Noe, and Tice, 2009). 2 High stock liquidity may also improve firm s investment decisions through the corporate governance channel. For instance, Maug (1998) shows that stock liquidity facilitates the formation of larger blockholdings and more liquid trading by these blocks, thereby enhancing blockholders voice and their ability to intervene. Alternatively, stock liquidity may discipline managers by enhancing the threat of blockholder exit, when managerial compensation is closely tied to firm s stock price (Edmans, 2009; Edmans and Manso, 2011; and Bharath, Jayaraman, and Nagar, 2013). Collectively, prior research suggests that (i) tax avoidance is a part of firm s overall investment strategy, (ii) there are beneficial effects of stock liquidity on firm s investment decisions, and (iii) there are adverse implications of both overlyaggressive and overly-conservative tax avoidance strategies for shareholder wealth. Integrating these three sets of results, we posit that firms with high stock liquidity will engage less in both overly-conservative and overly-aggressive tax avoidance, ceteris paribus. 2 A positive feedback perspective rests on the premise that the extent of share price informativeness is determined by stock liquidity, and not the other way around, consistent with what prior studies suggest (e.g., Grossman and Stiglitz, 1980; Kyle, 1985; Easley and O Hara, 2004; Chordia et al., 2008; and Jayaraman and Milbourne, 2012). In particular, Chordia et al. (2008) document that liquidity Grangercauses stock price informativeness, but find no statistical evidence of reverse causality. 4

5 Consistent with our predictions, we document a positive association between firm s stock liquidity and the lower tail of the conditional tax avoidance distribution, which is symptomatic of firm s underinvestment in tax avoidance. We also document a negative association between a firm s stock liquidity and the upper tail of the conditional tax avoidance distribution, which is symptomatic of firm s overinvestment in tax avoidance. The effect is economically significant and is robust across different measures of tax avoidance and stock liquidity. We use stock splits as a positive shock to stock liquidity and the instrumental variable method to address potential endogeneity concerns, and obtain consistent results. Next, we examine possible mechanisms for how stock liquidity may affect tax avoidance. We find that the effect of stock liquidity on both tails of the tax avoidance distribution is stronger for firms with high levels of business uncertainty. Further, the effect of stock liquidity on the lower (upper) tail of tax avoidance distribution is mitigated (magnified) for the financially constrained firms. These results provide support for the existence of a positive feedback channel. On the other hand, we find the effect of stock liquidity to be similar across firms with high versus low shareholder rights, and across firms with high versus low managerial pay-forperformance-sensitivities. These findings are inconsistent with either blockholder intervention or blockholder exit channels. Our paper contributes to the existing literature in several ways. First, at a broader level, our paper can be viewed as a contribution to the tax avoidance research. Prior studies on the determinants of tax avoidance mainly focus on firm fundamentals or 5

6 executive incentives (e.g., Desai and Dharmapala, 2006; Lisowsky, 2010; Armstrong, Blouin and Larcker, 2012; and Rego and Wilson, 2012). Little is known, however, about how stock market conditions, and in particular the liquidity of the firm's stock, affect firm s tax planning. To the best of our knowledge, our paper is the first to document the effect of stock liquidity on tax avoidance, thereby establishing an important link between corporate tax avoidance and firm s stock market conditions. Second, our paper also contributes to the corporate finance literature documenting beneficial effects of stock liquidity on the value of the firm (e.g., Fang et al., 2009; and Bharath et al., 2013). Our results suggest that greater stock liquidity mitigates both overly-aggressive and overly-conservative tax planning, and thus highlight a potential channel through which greater stock liquidity increases firm value. Finally, at a methodological level, our study contributes to the tax avoidance literature by further emphasizing the importance of examining the whole distribution of tax avoidance, instead of focusing on the mean effect; an essential point raised by Armstrong et al. (2013). The rest of the paper is organized as follows. In Section II we discuss the relevant literature and develop our hypotheses. Section III describes the data. Section IV outlines the methodology. In Sections V to VII, we report and discuss our empirical findings. Concluding remarks are presented in Section VIII. 6

7 II. RELATED LITERATURE AND HYPOTHESES DEVELOPMENT Corporate Tax Avoidance and Stock Liquidity We posit that high liquidity firms engage less in both overly-aggressive and overly-conservative tax avoidance. Our empirical predictions rest on the premise that tax avoidance is viewed by managers as one of the many investment decisions (Hanlon and Heitzman, 2010; Armstrong et al., 2013), and thus is considered as a part of the overall investment strategy of the firm (McGuire, Omer, and Wilde, 2013). Accordingly, factors that affect firm s overall investment strategy should also affect firm s tax avoidance as shown by McGuire et al. (2013). The above discussion suggests several causative mechanisms through which stock liquidity may influence corporate tax avoidance. First, greater stock liquidity may affect tax avoidance through the feedback channel from stock prices to firm s tax planning. Prior studies show that greater stock liquidity stimulates the entry of informed traders (Grossman and Stiglitz, 1980; Kyle, 1985; and Easley and O Hara, 2004). Informed investors factor the effect of their trades on firm s investment decisions, trade more aggressively, and consequently make stock prices more informative to firm s management (Subrahmanyam and Titman, 2001; Khana and Sonti, 2004; Foucault and Gehrig, 2008; Fang et al., 2009). Information uncertainty may cause managers to misestimate firm s potential tax savings over the course of tax avoidance investment, thereby resulting in over-or underinvestment in tax avoidance. Thus, by increasing share price informativeness to managers, greater stock liquidity is 7

8 predicted to facilitate better informed tax planning decisions, and therefore will influence firm s tax avoidance. In particular, our empirical predictions are most closely related to the studies by Khana and Sonti (2004), Chen et al. (2007), and Edwards et al. (2013). In Khana and Sonti s (2004) model higher share prices signal improved prospects to firm managers and relax firm s financial constraint. Informed traders factor this effect of share prices on firm investment decisions and manipulate share prices to get firm to take certain investments. Chen et al. (2007) show that the sensitivity of investment to share price increases in the amount of private information in stock price, suggesting that managers learn from the private information in stock price about their own firm s fundamentals (consistent with Khana and Sonti, 2004). Edwards et al. (2013) show that financially constrained firms resort to a more aggressive tax planning. Integrating these three sets of results, we posit that greater stock liquidity will facilitate more informed trading to relax the level of financial constraints for the firms whose tax avoidance strategies are considered by investors as overly-aggressive. Similarly, greater stock liquidity will facilitate more informed trading to tighten financial constraints for the firms whose investment in tax planning is perceived by investors as overly-conservative. Accordingly, a positive feedback perspective predicts that high liquidity firms will engage less in both overly-aggressive and overly-conservative tax avoidance. Second, a link between stock liquidity and tax avoidance can also be motivated based on the corporate governance theories. Separation of ownership and control may 8

9 lead to corporate tax planning that reflects the private interests of the manager. The resulting agency problems cause managers to engage in overly-aggressive or overlyconservative tax avoidance (e.g., Chen and Chu, 2005; Desai and Dharmapala, 2006; and Kim, Li and Zhang, 2011). Prior research highlights two mechanisms for the beneficial role of stock liquidity in corporate governance. One strand of literature in this vein suggests that stock liquidity enhances blockholders incentives to intervene by allowing blockholders to increase their stakes in the firm at lower cost (Maug, 1998). Another agency-based causative mechanism suggests that liquidity improves corporate governance by facilitating the threat of blockholder exit (Edmans 2009; and Edmans and Manso 2011). In these models, strategic dumping of firm shares by blockholders exerts a downward pressure on share price, and thus penalizes management for actions that are considered not to be in the best interests of blockholders. Prior studies provide empirical evidence consistent with greater stock liquidity improving corporate governance (e.g., Bharath et al., 2013; and Edmans et al., 2013). More directly related to our hypothesis is the study by Armstrong et al. (2013) who document that corporate governance increases extremely low levels of tax avoidance, and decreases extremely high levels of tax avoidance. Thus, a corporate governance perspective suggests that high stock liquidity will deter firm management from engaging in both overly-aggressive and overly-conservative tax avoidance by either enhancing blockholder intervention or amplifying the threat of blockholder exit. 9

10 To summarize, both the positive feedback and corporate governance perspectives predict that high liquidity firms will engage less in both overly-aggressive and overlyconservative tax avoidance. The above discussion leads to our main hypothesis. H1. Ceteris lower paribus, (upper) stock tail of liquidity tax avoidance is positively distribution. (negatively) associated with the lower (upper) tail of tax avoidance distribution. Is There a Positive Feedback Effect? Having established a link between stock liquidity and corporate tax avoidance, our next step is to examine the causative mechanism of this relation. Positive feedback perspective predicts that the effect of stock liquidity on firm s investment decisions will be stronger for the firms with high levels of business uncertainty (Subrahmanyam and Titman, 2001; Fang et al., 2009). Further, since positive feedback effect is predicted to relax financial constraints for the firms that engage in overly-aggressive tax avoidance, the effect of liquidity on the upper tail of tax avoidance is expected to be stronger for the financially constrained firms. On the other hand, since positive feedback effect is predicted to tighten financial constraints for the firms that engage in overly-conservative tax avoidance, the effect of liquidity on the lower tail of tax avoidance distribution is predicted to be weaker for the firms that already face financial constraints. This discussion leads to the following hypotheses. H1a: Ceteris paribus, the association between stock liquidity and both tails of corporate tax avoidance distribution is stronger for the firms with high levels of business uncertainty. 10 H1b: Ceteris paribus, the association between stock liquidity and the upper (lower) tail of tax avoidance distribution is stronger (weaker) for the financially constrained firms.

11 Is There Blockholder Intervention Effect? The effect of blockholder intervention depends on the balance of power between shareholders and managers. Therefore, if the association between stock liquidity and tax avoidance occurs through blockholder intervention channel, the effect of liquidity should be amplified for the firms with high shareholder rights (Fang et al., 2009; and Bharath et al., 2013). This insight leads to the following hypothesis. H1c: Ceteris paribus, the association between stock liquidity and both tails of corporate tax avoidance distribution is stronger for the firms with high shareholder rights. Is There Blockholder Exit Effect? The disciplining effect of the threat of exit comes through managers' wealth invested in firm s stock. Accordingly, the implications of blockholder exit will be more severe for managers, whose compensation is highly sensitive to changes in share price (Bharath et al., 2013; and Edmans et al. 2013). Thus, blockholder exit perspective predicts that the effect of stock liquidity will be magnified for the firms with high pay-for-performance sensitivity. This discussion leads to the following hypothesis. H1d: Ceteris paribus, the association between stock liquidity and both tails of corporate tax avoidance distribution is stronger for the firms with high managerial pay-for-performance sensitivity. 11

12 III. SAMPLE SELECTION AND VARIABLES Sample Selection Our main measure of stock liquidity is the relative effective spread, calculated using intra-day data from the Trade and Quote database (TAQ). This measure comes from Vanderbilt University's Financial Markets Research Centre (hereafter FMRC). 3 Because TAQ only provides data since 1993, we restrict our sample period to We obtain firm financial information from Compustat fundamental annual files. Our initial sample consists of all the firms in Compustat over the sample period of We drop observations without sufficient information for construction of the tax avoidance and /or stock liquidity measures, and those with negative pre-tax income. We winsorize all the variables at both the 1 st and 99 th percentiles to mitigate the effect of outliers. Variables Tax Avoidance In our analysis we adopt Hanlon and Heitzman s (2010) broad definition of tax avoidance as the reduction of explicit taxes that reflects all transactions that have any effect on firm s tax liability. Our primary measure of tax avoidance is GAAP effective tax rate (GETR) which is computed as follows: TXT GETRit PI it it (1) 3 We are grateful to Hans Stoll and Christoph Schenzler from Vanderbilt University for providing the effective spread data. 12

13 where i denotes firm, t denotes year, is firm i s total tax expenses and is firm i s pre-tax income. This measure has been extensively adopted by prior research (e.g., Dyreng, Hanlon and Maydew, 2010; Armstrong et al., 2012; and Cheng et al., 2012). In particular, our choice of GETR as the primary measure of tax avoidance is motivated by the results of Graham at al. s (2013) survey, who find that management at public companies places a higher weight on the GAAP effective tax rates compared to other tax metrics, such as cash taxes paid. Further, Armstrong et al. (2012) conclude that GETR is a more informative measure of tax director actions, compared to other tax avoidance metrics. Lower values of GETR indicate more aggressive tax avoidance. Stock Liquidity Our primary measure of stock liquidity (LIQ) is relative effective spread, computed as the ratio of the difference between the trade price and the midpoint of the bid-ask quote over the trade price. The relative effective spread is regarded as one of the best measures of stock liquidity and has been extensively used by prior studies (e.g., Fang et al., 2009; and Fang, Tian and Tice, 2013). FMRC computes daily relative effective spread for the stock as the trade-weighted average of the relative effective spread of all the trades during the day. To obtain the annual relative effective spread, we take the arithmetic mean of the daily relative effective spreads over the firm's fiscal year. Since relative effective spread is highly skewed and higher value indicates lower liquidity, we take the natural logarithm of the ratio to correct for the 13

14 skewness following Fang et al. (2009) and then multiply it by minus one for ease of interpretation. Accordingly, higher values of LIQ indicate higher stock liquidity. Control Variables The selection of control variables follows prior literature (e.g., Rego, 2003; and Dyreng et al., 2010) and captures different aspects of firm operations. We include return on assets (ROA) and ROA volatility (STDROA) to control for firm profitability and the uncertainty of the firm's operation. PPE assets (PPE) and intangible assets (INTANG) are included to control for the nature of the firm's business. We also include leverage ratio (LEV), loss carry forward dummy (NOL) and change in loss carry forward (DNOL_AT) to capture the debt and non-debt tax shields. Firm size (SIZE) is included to control for political costs of tax avoidance (e.g., Rego, 2003). Change in goodwill (POSGDWL) is included to capture the merger and acquisition activities of the firm. Equity income in earnings (EQINC) and new investments (NEWINV) are included to control for firm's investment activities. We also include foreign assets (FRGNAT) to control for the economies of scale in tax planning. Last, we include market-to-book (MB) to control for growth opportunities, cash holdings (CASH) to control for financial conditions, and abnormal accruals (ABACC) to control for earnings management. Detailed definitions of all the variables are presented in Appendix A. 14

15 Descriptive Statistics Table 1 presents the summary statistics of the data. The mean (median) GAAP effective tax rate (GETR) of our sample firms is (0.348) and the mean (median) stock liquidity (LIQ) is (5.724). These numbers are well in line with those reported in prior studies (e.g., Fang et al., 2009; and Dyreng et al., 2010). As far as the control variables are concerned, the mean cash holdings, ROA, and ROA volatility are 0.186, 0.099, and 0.117, respectively. On average, our sample firms have change in goodwill of 0.021, new investments of 0.085, leverage ratio of 0.172, foreign assets of 0.44, equity income in earnings of 0.001, and change in loss carry forward of These firms also have PPE assets of 33.1% and intangible assets of 13.3% of total assets. 29% of the firms have loss carry forward and 34% of them have foreign income. Last, the sample firms have mean firm size of 6.182, mean market-to-book of 2.882, and mean abnormal accruals of [Insert Table 1 about here] Table 2 presents the correlation matrix of the variables. The table shows that GETR exhibits weak positive correlation with stock liquidity. GETR is also negatively correlated with cash holdings, ROA volatility, loss carry forward dummy, new investments, market-to-book, leverage, foreign assets, equity income in earnings, change in loss carry forward, and abnormal accruals; while positively correlated with return on assets, PPE assets, change in goodwill, firm size, intangible assets, and foreign income dummy. [Insert Table 2 about here] 15

16 IV. RESEARCH DESIGN As discussed in Section II, high liquidity firms are predicted to engage less in both overly-aggressive and overly-conservative tax avoidance. Accordingly, the effect of stock liquidity on tax avoidance is expected to be stronger at the tails while having potentially little or no effect on the mean or the median of the tax avoidance distribution. To test our predictions, we follow Armstrong et al. (2013) by utilizing a quantile regression approach (Koenker and Basset, 1978; Koenker and Hallock, 2001). Below we briefly outline the quantile regression method and discuss its advantages over the standard regression methods in the context of our research question. Let y be the response variable (in our case, tax avoidance), x be the explanatory variable of interest (in our case, stock liquidity), and Z be the vector of control variables (including a constant term). Assume that a conditional -th quantile of given values of x and Z,, is given by (2) for. The -th quantile coefficients, and, are estimated by solving the following optimization problem { } (3) where is the so-called check function which weighs positive and negative values asymmetrically, and is a dummy variable which takes value of 1 if is negative and 0 otherwise. The key advantage of the quantile regression method is that it allows the effect of the variable of interest, to vary across different parts of the distribution of the 16

17 response variable, y. This feature is particularly important in the context of our study, where greater stock liquidity is predicted to have a positive (negative) effect on the lower (upper) tails of the tax avoidance distribution. This effect cannot be captured by the standard ordinary least squares (OLS) regression method, which focuses solely on the effect of the variable of interest on the mean of distribution of the response variable. In contrast to the standard regression methods, quantile regression is designed to characterize changes in both the location and shape of the distribution of interest (Armstrong et al., 2013), and thus provides an appealing analytical framework for our analysis. 4 V. STOCK LIQUIDITY AND CORPORATE TAX AVOIDANCE Univariate Evidence Figure 1 presents the univariate relation between stock liquidity and tax avoidance. The x-axis plots decile ranks of GETR and y-axis plots mean values of stock liquidity lagged one year that correspond to these deciles. The relation between the two variables exhibits a distinct inverted U-shape, with firms in the centre (tails) of the GETR distribution having higher (lower) stock liquidity. Specifically, the average relative effective spreads for the firms in the highest and lowest deciles of GETR are 60 and 40 basis points, respectively. In contrast, the relative effective spread for the firms in the 5-th decile of GETR is only 25 basis points. The difference is also 4 This important point is emphasized by Mosteller and Tukey (1977, p.266), who note that...regression often gives a rather incomplete picture. Just as the mean gives an incomplete picture of a single distribution, so the regression curve gives a corresponding incomplete picture for a set of distributions 17

18 statistically significant. These results support H1, which predicts that high liquidity firms will engage less in both overly-aggressive and overly-conservative tax avoidance. [Insert Figure 1 about here] Multivariate Evidence Turning to our multivariate tests, we estimate a quantile regression with the baseline model specification as follows GETR it PPE 5 LEV 10 LIQ it 1 it 1 DNOLAT 14 0 it 1 it 1 POSGDWL 6 1 INTANG 11 DFI 15 CASH 2 it 1 it 1 it 1 it 1 SIZE ROA it 1 FRGNAT 12 ABACC it 1 it 1 it 1 NOL 8 STDROA t it 1 EQINC 13 4 i it 1 YR IND it it 1 NEWINV 9 it 1 (4) where i denotes firm, t denotes year, and ε is the error term. GETR is GAAP effective tax rate and LIQ is stock liquidity measured by negative log relative effective spread. IND is the industry fixed effects based on two-digit SIC codes and YR is the year fixed effects. All the independent variables are lagged one year to mitigate potential endogeneity concerns following Fang et al. (2009). We estimate our baseline model separately for the 10-th, 50-th (median) and 90-th GETR percentiles. We present the regression results in Columns (1) - (3) of Table 3. The coefficient for stock liquidity is positive and statistically significant for the 10-th GETR percentile (z-statistic=6.26, p-value<0.01, two-tail) and is negative and statistically significant for the 90-th GETR percentile (z-statistic=-7.94, p-value<0.01, two-tail). These results suggest that stock liquidity is negatively associated with both overlyaggressive and overly-conservative tax avoidance, thereby providing support for H1. 18

19 The coefficient for stock liquidity for the median GETR is positive and statistically significant (z-statistic=3.98, p-value<0.01, two tail) 5. To facilitate economic interpretation of our results, we rank stock liquidity for each year, and then partition the resulting ranks into deciles labelled from 1 (lowest decile) to 10 (highest decile). Next, we estimate our baseline quantile regressions with the decile ranking of stock liquidity and present the results in Columns (4) - (6) of Table 3. The results further confirm the positive (negative) relation between the 10-th (90-th) percentile of GETR and stock liquidity. Further, the coefficient for stock liquidity in Column (4) suggests that moving from the lower to the upper decile of stock liquidity increases 10-th percentile of GETR by 0.011*(10-1) = 0.099, or 9.9% of pre-tax income. The coefficient for stock liquidity reported in Column (6) suggests that moving from the lower to the upper decile of stock liquidity decreases 90-th percentile of GETR by 0.004*(10-1) = 0.036, or 3.6% of pre-tax income. Therefore, we conclude that the effect of stock liquidity on the tails of tax avoidance distribution is not only statistically significant but is also economically meaningful. For comparison, stock liquidity has a relatively moderate effect on the median GETR, where moving from the lower to the upper decile of stock liquidity increases median GETR by only 0.001*(10-1) =0.009 or 0.9% of pre-tax income. [Insert Table 3 about here] To further examine the relation between stock liquidity and tax avoidance, we reestimate our baseline model for the 10-th, 15-th, 20-th, and 90-th percentiles of 5 For robustness purposes, we repeat our analyses with financial firms (SIC codes ) excluded from our sample. The results (untabulated) remain qualitatively similar to those reported in our paper. 19

20 GETR. Next, we plot in Figure 2 the estimated coefficients for stock liquidity along with the 95% confidence intervals against the corresponding percentiles of GETR. Figure 2 reveals several interesting findings. Stock liquidity associates positively with GETR (i.e., associates negatively with tax avoidance) when moving from the centre to the upper end of the tax avoidance distribution (i.e., lower percentiles of GETR) On the other hand, stock liquidity associates negatively with GETR (i.e., associates positively with tax avoidance) when moving from the centre to the lower end of the tax avoidance distribution (i.e., upper percentiles of GETR). Furthermore, the magnitude of the effect monotonously increases when moving from the centre of distribution to the extreme percentiles of GETR. Overall, these results provide further support for H1, suggesting that high liquidity firms engage less in both overlyaggressive and overly-conservative tax avoidance. [Insert Figure 2 about here] As far as the control variables are concerned, since prior studies use least squares regression method in their research design the results for the control variables are not directly comparable to those reported in prior literature. Nonetheless, examination of the coefficients for control variables reveals several interesting items. First, the effects of control variables on the median GETR are largely in line with those reported in prior studies (e.g., Rego, 2003; and Chen et al., 2010). Specifically, median GETR is positively related to intangible assets (INTANG) and the return on assets (ROA), and is negatively associated with equity in earnings (EQINC), proportion of foreign assets (FRGNAT), loss carried forward (NOL), and firm size (SIZE). Second, while for some 20

21 variables the effect on GETR is uniformly positive or negative across different parts of GETR distribution (e.g., EQINC and NOL), other variables, such as leverage (LEV) and cash holdings (CASH), experience shifts in the coefficient sign as one moves from the upper to lower tail of GETR distribution. While in-depth examination of the effects of control variables is out of scope of a current study, these observations further emphasize the importance of controlling for changes in both location and shape of the tax-avoidance distribution. VI. ROBUSTNESS TESTS In this section, we conduct a variety of robustness checks to validate our main findings. For the sake of brevity, we only report the coefficients for stock liquidity. Alternative Measures of Tax Avoidance and Stock Liquidity We commence by examining the robustness of our results to alternative measures of tax avoidance. We use the following measures of tax avoidance employed in prior studies (Desai and Dharmapala, 2006; Dyreng, 2008; Armstrong et al., 2012; and Cheng et al., 2012): (1) annual effective cash tax rate, (2) long-term effective cash tax rate, and (3) the residual book-tax difference 6. The results are presented in Panel A of Table 4. The coefficient for stock liquidity is positive and significant for the 10-th percentile regression (smallest z-statistic=2.76, p-value<0.01, two-tail), and is negative and significant for the 90-th percentile regression (smallest z-statistic=-2.23, p-value<0.05, two-tail) for all three measures of tax avoidance. 6 For ease of interpretation we multiply residual book-tax difference by

22 Next, we examine the sensitivity of our results to using alternative measures of stock liquidity. We use the following stock liquidity measures: (1) the Amihud (2002) measure, modified by Gopalan and Pevzner (2012), (2) implicit bid-ask spread measure of Hasbrouck (2009), and (3) the percentage of zero returns measure of Lesmond (2005). We take natural log of both Amihud (2002) and the Hasbrouck (2009) measures, as these two measures are heavily skewed and multiply by the minus one for ease of interpretation. Similarly, we use Lesmond (2005) measure as one minus percentage of zero returns. Following Jayaraman and Milbourn (2012) we combine these three measures into a composite measure using principal components. The results are reported in Panel B of Table 4. The coefficient for stock liquidity is positive and significant for the 10-th percentile of GETR (z-statistic=8.27, p- value<0.01, two-tail), and is negative and significant for the 90-th percentile of GETR (z-statistic=-5.95, p-value<0.01, two-tail) Hence, we conclude that our results are robust to alternative measures of tax avoidance and stock liquidity. Alternative Regression Specifications Next, we examine the robustness of our results to alternative model specifications. We conduct two analyses. In our first analysis, we consider the possibility that our results are driven by the time-trends in tax-avoidance and/or stock liquidity. To address this concern, we estimate our baseline model (Eq.4) cross-sectionally (yearby-year) and report the time-series averages of the coefficients for stock liquidity. The results are reported in Panel C of Table 4. The coefficient for stock liquidity is positive and significant for the 10-th percentile of GETR (t-statistic=2.85, p- 22

23 value=0.01, two-tail), and is negative and significant for the 90-th percentile of GETR (t-statistic=-2.39, p-value=0.03, two-tail). Hence, we conclude that our findings are not likely to be driven by the time-trends in variables. In our second analysis, we augment our baseline model with additional control variables which were shown by prior literature to be correlated with stock liquidity and which may also affect firm s tax avoidance. Specifically, in addition to the control variables specified in Section III we also include the following variables: (1) institutional ownership, (2) analyst coverage, (3) board independence (proportion of independent directors on the board), (4) CEO delta (sensitivity of CEO compensation to changes in stock price), (5) CEO vega (sensitivity of CEO compensation to changes in stock volatility), (6) implied firm s cost of equity, and (7) probability of informed trading (PIN). The results are reported in Panel C of Table 4. The coefficient for stock liquidity remains positive and significant for the 10-th percentile of GETR (z-statistic=4.14, p-value<0.01, two-tail), and negative and significant for the 90-th percentile of GETR (z-statistic=-2.91, p-value<0.01, two-tail). Overall, based on the results of these two analyses we conclude that our findings are robust to alternative regression specifications. [Insert Table 4 about here] 23

24 Endogeneity Stock Splits as a Shock to Stock Liquidity Given the cross-sectional design of our analysis, our results may suffer from endogeneity concerns. To address this issue, we conduct two analyses. First, we use stock splits as a shock to stock liquidity and study its effect on tax avoidance. Prior studies show that stock splits are associated with increases in stock liquidity (Lin et al., 2009; Jayaraman and Milbourn, 2012). Furthermore, Chemmanur, Hu, and Huang (2013) document that stock splits bring in more institutional trading and a consequent reduction in information asymmetry (consistent with more informative share prices). Since stock splits are not clustered in calendar time and have no effect on the underlying firm fundamentals (e.g., Easley et al., 2001), they provide an appealing framework to examine the causal effect of stock liquidity on corporate tax avoidance. We collect the data on stock splits from CRSP database. Excluding stock splits with missing data on GAAP effective tax rate, stock liquidity and control variables leaves us with a sample of 987 stock split events. We proceed as follows. First, we estimate our baseline model (Eq.4) excluding stock liquidity (that is, with control variables only) for the 10-th and 90-th percentiles of GETR. Next, we construct a dummy variable, DUM1, which takes value of one if firm i s GETR was below the estimated conditional 10-th percentile of GETR for the firm i in the year prior to stock split. We construct a second dummy variable, DUM10, which takes value of one if firm i s GETR was above the estimated conditional 90-th percentile of GETR for the firm i in the year prior to stock split and zero otherwise. Last, we construct a dummy 24

25 variable, DUM_MID, which equals to one minus the sum of DUM1 and DUM10. Our prediction is that changes in liquidity around stock splits will be positively (negatively) associated with changes in GETR for the firms that engaged in overly-aggressive (overly-conservative) tax avoidance prior to stock split, that is, the firms with DUM1=1 (DUM10=1). To test our prediction, we perform regression analysis on the changes in the effective tax rate against changes in stock liquidity around the stock splits. The dependent variable is the change in GETR over the two years around the stock split event and the independent variables are the change in stock liquidity, interacted with DUM1, DUM10 and DUM_MID dummy variables, and the changes in the control variables 7. The results are presented in Table 5. Consistent with our prediction, the coefficient for the interaction term of stock liquidity with DUM1 dummy is positive and significant (t-statistic=2.29, p-value<0.05, two-tail), and the coefficient for the interaction term of stock liquidity with DUM10 dummy is negative and significant (tstatistic=-1.84, p-value=0.066, two-tail). These results are consistent with greater stock liquidity resulting in less overly-aggressive and overly-conservative tax avoidance. [Insert Table 5 about here] 7 The advantage of using firm-specific changes regression is that it provides time-series evidence of the link between the effective tax rate and stock liquidity as well as alleviates concerns about firm-specific omitted correlated variables (Jayaraman and Milbourn, 2012). 25

26 Instrumental Variable Quantile Regression In our second analysis, we employ the instrumental variable quantile regression method of Chernozhukov and Hansen (2007). Following Fang et al. (2009) and Jayaraman and Milbourn (2012) we use stock liquidity lagged two years and mean stock liquidity for the industry as the instruments. The use of stock liquidity lagged two years further helps mitigate concerns that our results are driven by the omitted variable correlated with both effective tax rate and stock liquidity in fiscal year t. As for the industry liquidity, the portion of firm i s liquidity that is correlated with the liquidity of its industry is less likely to be correlated with potential firm-specific omitted variables (Jayaraman and Milbourne, 2012). The results are reported in Table 6. The coefficient for stock liquidity for the 10- th percentile of GETR remains positive and significant (z-statistic=6.26, p- value<0.01, two-tail). The coefficient for stock liquidity for the 90-th percentile of GETR remains negative and significant (z-statistic=-5.45, p-value<0.01, two-tail). Overall, based on the results of these two analyses we conclude that our findings are robust to potential endogenous effects. [Insert Table 6 about here] VII. HOW DOES STOCK LIQUIDITY AFFECT TAX AVOIDANCE? Is There a Positive Feedback Effect? To test for the positive feedback channel, we examine the effects of business uncertainty and financial constraints on the association between stock liquidity and tax avoidance. Specifically, H1a and H1b predict that (1) the effect of stock liquidity 26

27 on both tails of tax avoidance distribution is stronger for the firms with high levels of business uncertainty, and (2) the effect of stock liquidity on the upper (lower) tail of tax avoidance distribution is magnified (mitigated) for the financially constrained firms. First, consider the effect of business uncertainty. Following prior studies (e.g., Lang and Lundholm, 1996; Diether, Maloy and Scherbina, 2002; and Zhang, 2006) we employ the dispersion in analyst earnings forecasts (DISP) as a proxy for business uncertainty. The results are reported in Columns (1) - (3) of Table 7. The coefficient for the interaction term of stock liquidity with the analyst forecast dispersion is positive and significant for the 10-th percentile of GETR (z-statistic=2.45, p- value<0.05, two-tail), and is negative and significant for the 90-the percentile of GETR (z-statistic=2.96, p-value=0.01, two-tail). Since stock liquidity is positively (negatively) associated with the 10-th (90-th) percentile of GETR, these results suggest that the effect of liquidity on both tails of tax avoidance distribution is stronger for the firms with high levels of business uncertainty, thereby supporting H1a. Next, consider the effect of financial constraints. We use the Whited-Wu index (WW) proposed by Whited and Wu (2006) as the proxy for firm financial constraints. Higher values of the Whited-Wu index indicate firms that are more financially constrained. The results are reported in Columns (4) - (6) of Table 7. The coefficient for the interaction term of stock liquidity with the Whited-Wu index is positive and significant for both 10-th and 90-th percentiles of GETR (smallest z-statistic=4.35, p- 27

28 value<0.01, two-tail). 8 Since stock liquidity is positively (negatively) associated with the 10-th (90-th) percentile of GETR, these findings imply that the effect of stock liquidity on the upper (lower) tail of the tax avoidance distribution is magnified (mitigated) for the financially constrained firms, and thus provide support for hypothesis H1b 9. Overall, we conclude that our findings are consistent with the positive feedback channel. [Insert Table 7 about here] Is There Blockholder Intervention Effect? To test for the blockholder intervention channel, we examine the effect of shareholder rights on the association between stock liquidity and tax avoidance. Specifically, H1c predicts that the effect of stock liquidity on both tails of the tax avoidance distribution is stronger for the firms with high shareholder rights. We consider two measures of shareholder rights. Our first measure is the governance index (GINDEX) of Gompers, Ishii, and Metrick (2003). Following Gompers et al. (2003) we classify firms with GINDEX<=5 (GINDEX>=14) as the firms with high (low) shareholder rights. Our second measure is CEO power (CEO_POWER) which is the principal component of three board governance measures suggested by prior literature (Jensen, 1993; Mehran, 1995; and Core, 8 For robustness purposes we also try a principal component of the three financial constraints measures suggested by prior literature: Kaplan-Zingales index (Kaplan and Zingales, 1997), Whited-Wu index (Whited and Wu, 2006), and Hadlock-Pierce index (Hadlock and Pierce, 2010). The results are qualitatively similar to those reported for the Whited-Wu index measure. 9 The coefficient for the Whited-Wu index is negative and significant for all three regression models (smallest z-statistic=-5.00, p-value<0.01, two-tail). These results suggest that financially constrained firms, on average, engage more in tax avoidance, consistent with Edwards et al s. (2013) results. 28

29 Holthausen and Larcker, 1999). These measures include board size, CEO-Chairman duality, and board independence. Higher values of CEO_POWER indicate lower shareholder ability to influence CEO decision making. The data used to construct these measures were obtained from RiskMetrics. 10 First, consider the results using GINDEX as a proxy for shareholder rights. The results are reported in Columns (1) - (3) of Table 8. For the 10-th percentile of GETR the coefficient for the interaction term of stock liquidity with the LOW_GINDEX dummy (GINDEX<=5) is negative and significant (z-statistic=-1.94, p-value=0.053, two-tail). The coefficient for the interaction term of stock liquidity with the HIGH_GINDEX dummy (GINDEX>=14) is not significant (z-statistic=-0.34, p- value=0.73, two-tail). The difference between the two coefficients is not significant (z-statistic=-1.03, p-value=0.30, two-tail). For the 90-th percentile of GETR the coefficient for the interaction term of stock liquidity with the LOW_GINDEX dummy is negative and marginally significant (z-statistic=-1.74, p-value=0.08, two-tail). The coefficient for the interaction term of stock liquidity with the HIGH_GINDEX dummy is not significant (z-statistic=0.29, p-value=0.76, two-tail), and so is the difference between the two coefficients (z-statistic=-1.36, p-value=0.18, two-tail). Next, consider the results for the CEO_POWER measure. The results are reported in Columns (4) - (6) of Table 8. The coefficient for the interaction term of stock liquidity with CEO_POWER is not significant for both 10-th and 90-th percentiles of GETR (largest z-statistic=-1.60, p-value=0.11, two-tail). Overall, since the 10 The RiskMetrics database covers S&P 1500 firms and contains information of the governance index since 1990 and the board of directors since

30 blockholder intervention channel predicts the effect of liquidity to be stronger for the firms with high shareholder rights, the results reported in Table 8 do not support H1c. [Insert Table 8 about here] Is There Blockholder Exit Effect? To test for blockholder exit channel, we examine the effect of pay-forperformance sensitivity on the association between stock liquidity and tax avoidance. Specifically, H1d predicts that the effect of stock liquidity on both tails of tax avoidance distribution is stronger for the firms with high pay-for-performance sensitivity. We employ two measures of pay-for-performance sensitivity. The first measure, outlined in Core and Guay (2002) (PPS_CG), is computed as the natural logarithm of the dollar value change in managers' stock and option holdings with respect to 1% change in the firm's stock price. We obtain the data used to construct this measure from the ExecuComp database. 11 The second measure, outlined in Edmans, Gabaix and Landier (2009) (PPS_EGL), is defined as the dollar change in the CEO wealth for a one hundred-percentage point change in firm value divided by annual flow compensation. We obtain the data from Alex Edmans website. 12 The results are reported in Table 9. Columns (1) to (3) report the results for PPS_CG. The coefficient for the interaction term of stock liquidity with the pay-for 11 The ExecuComp database covers the stock and option compensation information of the management of S&P 1500 firms since

31 performance sensitivity is negative and significant (z-statistic=-2.14, p-value<0.05, two-tail) for the 10-th percentile of GETR, and is not significant for the 90-th percentile of GETR (z-statistic=1.14, p-value=0.25, two-tail). 13 Columns (4) to (6) report the results for the same sample when using scaled pay-for-performance sensitivity, PPS_EGL. The coefficient for the interaction term of stock liquidity with scaled pay-for-performance sensitivity is not significant for both the 10-th and 90-th percentiles of GETR (largest z-statistic=-0.58, p-value=0.59, two-tail). These results do not support H1d which predicts that the effect of stock liquidity on tax avoidance is stronger for the firms with high pay-for-performance sensitivity, and thus are inconsistent with the threat of exit channel. [Insert Table 9 about here] VIII. SUMMARY AND CONCLUSIONS In this study, we investigate the effect of stock liquidity on corporate tax avoidance. Greater stock liquidity makes share prices more informative to managers, and thus allows managers to make better informed tax planning decisions. Greater stock liquidity may also affect tax avoidance through the corporate governance channel by enhancing blockholders voice or amplifying the threat of blockholder exit. We document a positive (negative) effect of stock liquidity on the lower (upper) tails of the (conditional) tax avoidance distribution, suggesting that high liquidity 13 Given that PPS_CG is significantly correlated with firm size (Edmans et al., 2009), it is possible that the negative coefficient for the interaction term of PPS_CG with stock liquidity for the 10-th percentile of GETR captures the effect of firm size. In our sample the Pearson correlation coefficient between PPS_CG and firm size is 0.58 while that between PPS_EGL and firm size is only