Investor Trading and Book-Tax Differences

Transcription

1 Investor Trading and Book-Tax Differences Benjamin C. Ayers University of Georgia (706) Stacie K. Laplante University of Georgia (706) Oliver Zhen Li University of Arizona (520) Casey Schwab University of Georgia (706) October 2010 Ayers and Laplante gratefully acknowledge the support of the Terry College of Business and the J.M. Tull School of Accounting. Li gratefully acknowledges the support of the Eller College of Management. We thank Michelle Hanlon, Sean McGuire, John Robinson, and Terry Shevlin for helpful comments.

2 Investor Trading and Book-Tax Differences Abstract: Extant research suggests that book-tax differences are useful measures in evaluating firm performance and finds using long-window association tests that equity prices are associated with book-tax differences. There is little evidence, however, regarding whether investors actually trade based on the information in book-tax differences. Using short-window tests of buy and sell orders on the days surrounding earning announcements, this study investigates whether investor trades vary with book-tax differences in a predictable manner. Consistent with large book-tax differences signaling decreased earnings quality (e.g., less informative earnings), we find that the association between investor trades and earnings surprises declines for firms with large changes in book-tax differences. Additional analyses suggest that both institutions (large trades) and individuals (small trades) incorporate information in book-tax differences in their trades, although the effect is much more pronounced for institutions. JEL Classification: G38, H25, H32, M41 Key Words: Book-tax differences, trading volume, information content

3 Investor Trading and Book-Tax Differences I. INTRODUCTION This study investigates whether book tax differences are associated with investor trading. Specifically, we analyze actual trading data to assess whether investors trade shares based on book-tax differences, and whether investor sophistication affects this activity. Extant research suggests that book-tax differences are useful measures in evaluating firm performance and finds that equity prices are associated with book-tax differences using long-window association tests. For example, Lev and Nissim (2004) find that the ratio of taxable income to book income is systematically related to future earnings growth, and since 1992 that the ratio of taxable income to book income is associated with contemporaneous earnings-price ratios but not future stock returns. Likewise, Hanlon (2005) finds (a) that firms with either large positive or large negative book-tax differences have less persistent earnings and (b) no relation between book-tax differences and future returns. Both studies conclude that equity prices incorporate the information contained in book-tax differences, but neither study provides insight into whether investors actually incorporate the information in book-tax differences in their trading decisions, what types of investors incorporate this information, if any, and the size of these effects. 1 Financial accounting texts, extant literature, and anecdotal evidence suggest that book-tax differences can inform investors regarding earnings quality. For example, 1 Gleason and Mills (2008) provide evidence that the market discounts the reward for beating analysts forecasts for firms using a fourth quarter decrease in effective tax rate to beat forecasted earnings. Their evidence suggests that the market infers managed earnings via the tax decrease, which influences investors valuation of the firm (i.e., the price reaction to beating analysts forecasts). Our study more broadly addresses the question of how book-tax differences influence investor trading by not limiting the analysis to the subset of firms that beat the analyst benchmark with effective tax rate decreases, analyzing actual trading data, and distinguishing between types of investors trading on this information. 1

4 Palepu et al. (2005, 3-10) suggests that a widening book-tax difference represents a potential danger as it might indicate deteriorating earnings quality. Consistent with this contention, Hanlon (2005) finds that firms with large positive or large negative book-tax differences have less persistent earnings (i.e., current earnings is less informative regarding future earnings). This evidence suggests that, ceteris paribus, investors should have a more muted response to the earnings surprises of firms with large book-tax differences. 2 Specifically, if investors understand that earnings are less informative for firms with large book-tax differences and they trade on this information, we anticipate that the association between investor trades and earnings surprises is mitigated (i.e., declines) for firms with large changes in book-tax differences. Given the potential complexity in interpreting the information contained in booktax differences, it is likely that investor trading on book-tax differences varies with investor sophistication. Previous studies provide evidence that individual investors trade based on a less sophisticated or incomplete information set. For example, prior research (Bhattacharya 2001; Battalio and Mendenhall 2005) finds that individual investors are more likely to trade based on a less sophisticated earnings expectation model (seasonal random walk) than institutional investors. To the extent that investor trading on the information in book-tax differences varies with investor sophistication, we expect that the association between investor trades and earnings surprises is more negative for institutional investors. We separately analyze small trades (individual investors) and large trades (institutional investors) to test this hypothesis. 2 Consistent with this theory, Ayers, Laplante, and McGuire (2010) predict and find that the association between credit rating changes and earnings changes is mitigated for firms with large changes in book-tax differences. 2

5 We use short-window tests of buy and sell orders on the days surrounding earning announcements to investigate whether investor trades vary with book-tax differences in a predictable manner. 3 We measure buy-sell activity daily as shifts in supply and demand through changes in buy-sell order imbalances from the Trade and Quote (TAQ) database. We separate the order imbalances and designate those for small trades as originating from individual investors and those for large trades as originating from institutional investors. We follow Lee and Ready (1991) and Lee (1992) and classify trades into buy- or sellinitiated trades and use dollar-based cutoffs to define individual (small) and institutional (large) trades (Lee and Radhakrishna 2000). We assume that trades below $10,000 represent individual trades and that trades above $30,000 represent institutional trades. 4 We calculate the abnormal buy-sell imbalance around the earnings announcement (days - 1 through +1) as the average buy-sell imbalance during the announcement period less the average buy-sell imbalance during the control period (days -45 through -6), deflated by the average trading volume (average buy plus sell orders) during the control period. To test the effect of book-tax differences on investor trading around earnings announcements, we regress the announcement period abnormal buy-sell imbalance on analyst-based earnings surprises, the decile rank of the absolute value of a firm s change in book-tax difference for the year, and the interaction between analyst-based earnings surprises and the decile rank of the absolute value of a firm s change in book-tax difference. We use the absolute value of the change in book-tax differences because Hanlon (2005) suggests that firms with either large positive or large negative book-tax 3 As discussed in Section II, we assume that investors have the information available at the earnings announcement date to estimate book tax differences (Francis, Schipper and Vincent 2002). 4 In sensitivity analyses, we use a variety of trade-size cut-offs to distinguish individual and institutional trades (e.g., below (above) $10,000, $20,000, $30,000, etc.). Conclusions are robust to these alternative specifications. 3

6 differences have less persistent earnings. We utilize a changes measure of book-tax differences to be consistent with our short-window changes model for investor trading and earnings. 5 We expect a negative association between abnormal buy-sell imbalance and the interaction between analysts-based earnings surprises and the decile rank of the absolute value of a firm s change in book-tax differences if investors have a more muted response to the earnings surprises of firms with large book-tax differences. A more pronounced negative association for trades initiated by institutional investors (large trades) would be consistent with sophisticated investors being more likely to trade on the information contained in book-tax differences. Consistent with large book-tax differences signaling less informative earnings, we find a negative association between the abnormal buy-sell imbalance around earnings announcements and the interaction between analysts-based earnings surprises and a firm s change in book-tax differences. Additional analyses indicate that both institutions (large trades) and individuals (small trades) incorporate book-tax differences in their trades, although the effect is much more pronounced for institutions. For example, the effect of book-tax differences on investor trading is nine times larger for institutional (large) trades than for individual (small) trades. However, institutional trades also have a much higher association with analyst-based earnings surprises than individual trades (i.e., eight times larger). Thus, the difference in trading on book-tax differences across institutional and individual investors is similar to the trading differences based on analyst-based earnings surprises. Regarding the magnitude of the effects of book-tax 5 We also examine the effect of book-tax difference changes that move firms closer to or further away from the industry median because firms that move closer to (further away) from the industry median could arguably be engaging in less aggressive (more aggressive) behavior. Results are consistent with our primary findings for H1 reported in Table 4. For H2, however, the effect of book-tax differences on the association between buy-sell order imbalances and earnings surprises is more negative for large trades than for small trades only when firms move further away from the industry mean. 4

7 differences on investor trading, we find that the association between buy-sell order imbalance and analyst-based earnings surprise is completely eliminated for firms in the largest decile of book-tax differences. We find a similar relation for both institutional and individual trades, although the base relation between individual trades and analystbased earnings surprises is much smaller. In sensitivity analyses, we test whether the results are robust to controls for other factors that influence investor trades based on earnings surprises (e.g., large abnormal accruals) or book-tax differences (e.g., tax planning). Conclusions are the same, and we find some evidence that tax planning influences how institutional investors trade on book-tax differences (i.e., that the effects of book-tax differences on institutional investor trading is mitigated for high tax planning firms identified using the cumulative effective tax rates in Dyreng et al. 2008). This study makes the following contributions. Our results indicate that investor trades vary with book-tax differences in a predictable manner. Specifically, consistent with large book-tax differences signaling less informative earnings, we find that the association between investor trades and earnings surprises declines for firms with large changes in book-tax differences. Among other implications, these results suggest that investors view the information contained in book-tax differences as an important mechanism for evaluating earnings surprises and, at least in part, incorporate the information contained in book-tax differences in their trading (i.e., make portfolio allocation decisions). Further analyses are consistent with both individual and institutional investors incorporating the information contained in book-tax differences into trading, though the effects are more pronounced for institutional investors. 5

8 Our paper proceeds as follows. Section II develops hypotheses, while Section III describes the sample data and research methods. Section IV presents results, Section V presents sensitivity analyses, and Section VI concludes. II. PRIOR RESEARCH AND HYPOTHESES Information in Book-Tax Differences Managers prepare two summary measures of firm performance annually, one for financial reporting purposes as determined by generally accepted accounting principles (GAAP), and one for tax purposes in accordance with the Internal Revenue Code (IRC). Because book and taxable income serve different purposes and stakeholders, managers have different incentives in reporting book and taxable income. Managers generally have incentives to report higher income for financial reporting purposes because of bond covenants, compensation contracts, regulatory capital requirements, etc. 6 In contrast, for tax purposes, managers generally have incentives to report lower taxable income to reduce tax expenditures. Given that managers face different incentives in reporting book and taxable income and each measure is a summary measure of a firm s performance based on a unique set of rules, book-tax differences may be useful in evaluating firm performance. Anecdotal and empirical evidence confirm this expectation. For example, using long window (annual return) tests, Shevlin (2002), Hanlon, Laplante, and Shevlin (2005), and Ayers, Jiang, and Laplante (2009) find that book and estimated taxable income each summarize unique information reflected in stock returns. In other words, book and 6 In certain circumstances firms have incentives to report lower income e.g., due to political costs (Watts and Zimmerman 1986) and compensation contracts (Healy 1985). 6

9 estimated taxable income are incrementally informative and both measures of firm performance contain information. Indeed, IRS audit manuals instruct internal revenue agents to reconcile book and tax amounts and question any differences when auditing corporate tax returns. 7 Mills (1998) finds that IRS audit adjustments increase as the book-tax gap widens (i.e., as book income exceeds taxable income), providing empirical support that book-tax differences are informative regarding tax aggressiveness. From a financial reporting perspective, financial statement texts argue that an increasing book-tax difference is potentially a sign of deteriorating earnings quality (e.g., Palepu et al. 2005). For example, large book-tax differences may be evidence of income increasing accruals that artificially increase financial statement earnings or income decreasing accruals that artificially decrease financial statement earnings, both of which should lead to large future reversals (i.e., low persistence). Consistent with book-tax differences being informative regarding earnings quality, recent empirical literature finds that book-tax differences are systematically related to earnings persistence, earnings growth, and contemporaneous earnings-price ratios (Hanlon 2005; Lev and Nissim 2004). Likewise, other studies suggest that the information in book-tax differences is associated with information used by buy-side analysts (Weber 2009), auditors (Hanlon and Krishnan 2006) and credit analysts (Ayers, Laplante and McGuire 2010). Hypothesis 1 Book-Tax Differences and Investor Trading In spite of the growing evidence that the relation between book and taxable income is useful in evaluating firm performance, prior research provides little evidence regarding investor trading on book-tax differences. Indeed, Gleason and Mills (2007) 7 See Audit Technique Guides for various industries at 7

10 provide the only short-window analysis of the use of book-tax differences by firm stakeholders. They find that the market discounts the reward for beating analysts forecasts for firms using a fourth quarter decrease in effective tax rate to beat forecasted earnings. Their evidence suggests that the market infers opportunistic earnings management via the tax rate decrease and influences investors valuation of the firm (i.e., the price reaction to beating analysts forecasts). Our study more broadly addresses the question of how book-tax differences influence investor trading. Unlike Gleason and Mills (2007), we do not limit our analyses to the subset of firms that beat the analyst benchmark with effective tax rate decreases. In addition, we analyze actual trading data to assess whether investors liquidate shares based on the information contained in book-tax differences and distinguish between the types of investors trading on this information. Consistent with Hanlon (2005) who finds that current earnings is less informative regarding future earnings for firms with either large positive or large negative book-tax differences we argue that, ceteris paribus, investors should have a more muted response to the earnings surprises of firms with large book-tax differences. 8 Specifically, if investors understand that current earnings is less informative for firms with large booktax differences and they incorporate this information into their trading decisions, we anticipate that the association between investor trades and earnings surprises is mitigated (i.e., declines) for firms with large changes in book-tax differences. Our first hypothesis, stated in the alternative is: 8 Using a regression of year t+1 earnings on year t earnings and the interaction between year t earnings and the decile rank of the absolute value of the change in book-tax difference for year t, we confirm that the association between year t earnings and year t+1 earnings declines for firms with large changes in book-tax differences in year t. 8

11 H1: The association between buy-sell order imbalance and earnings surprises declines for firms with large book-tax differences. It is important to note that we assume investors have the information available at the earnings announcement date to estimate taxable income. Given that our sample period covers , a period largely characterized by increased disclosure of detailed income statement information with earnings announcements (Francis, Schipper 9, 10 and Vincent 2002), this assumption appears reasonable. To the extent that it does not hold, we would not expect to find a relation between investor trading and book-tax differences. Hypothesis 2 Book-Tax Differences and Investor Sophistication Prior evidence implies that individual investors trade based on a less sophisticated or incomplete information set. For example, prior research (Bhattacharya 2001; Battalio and Mendenhall 2005) finds that individual investors are more likely to trade based on a less sophisticated earnings expectation model (seasonal random walk) than institutional investors. This is especially pertinent in our setting for two reasons. First, tax returns are private so investors need to understand how to interpret information in the financial 9 Gleason and Mills (2007) also argue that there is sufficient information at the earnings announcement date to compute effective tax rates. 10 We examined a random sample of 90 firm-years (from 29 firms) during our sample period and found that they consistently disclosed total tax expense in their earnings announcements, but that other tax-related disclosures were inconsistent. For example, many firms disclosed the current tax liability (or current taxes payable) as well as current tax assets, but failed to differentiate between current tax and deferred tax expense. Also, tax disclosures appear to be more comprehensive in later years of the sample. While firms do not necessarily disclose current or deferred tax expense, it is possible to approximate taxable income using total tax expense. The correlation between current tax and total tax expense for the entire Compustat database during our sample period is approximately 92.9%, while it is 93.2% for our sample. In addition, the correlation between our primary independent variable (the decile rank of the change in taxable income) calculated with current tax expense versus total tax expense is 51.98%. These significant correlations suggest that investors are able to discern taxable income from disclosures made at the earnings announcement date. Given additional measurement error, however, it is not obvious that our results will hold if we use total tax expense to approximate taxable income. As noted later, we repeat the tests underlying Tables 5 and 6, and our primary results remain. 9

12 statements related to taxes (e.g., in our setting, we assume investors derive an estimate of taxable income from the financial statements). While Hanlon (2003) documents some of the issues associated with using financial statements to derive an estimate of taxable income, most prior empirical studies utilize estimates of taxable income from the financial statements under the assumption that sophisticated users of financial statements understand book-tax differences. Second, the rules underlying the calculation of taxable income are complex implying that some level of sophistication is necessary to understand what differences in book and estimated taxable income mean. 11 Given the potential complexity in interpreting the information contained in booktax differences, it is likely that investor trading on book-tax differences varies with investor sophistication. To the extent this holds, we expect that the effects of book-tax differences on large trades (our proxy for sophisticated institutional investors) will be more pronounced. Our second hypothesis, stated in the alternative is: H2: The effect of book-tax differences on the association between buy-sell order imbalances and earnings surprises is bigger for large trades than for small trades. Measuring Buy-Sell Order Imbalance III. EMPIRICAL METHODOLOGY We use the Lee and Ready (1991) algorithm to separate trades into buyer-initiated or seller-initiated trades for each firm. 12 Specifically, we compare traded prices with quotes at least five seconds earlier. If the traded price is above the mid-point of the bid- 11 For example, assuming length is correlated with complexity, the Internal Revenue Code (IRC - the part written by Congress) is approximately 3,387 pages long, while the Federal Regulations that help explain the IRC (the part written by the Treasury) is 13,454 pages long (United States Government Accounting Office 12 We suppress the firm subscript in Section III to simplify exposition. 10

13 ask spread, we define the trade as a buy [BUY]. If the traded price is below the mid-point of the bid-ask spread, we define the trade as a sell [SELL]. For Day k, we add up all the shares in buys and all the shares in sells, where M and N are the total number of buy and sell transactions on Day k, respectively. We then compute daily trading volume BPS k (BUY plus SELL) and daily order imbalance BMS k (BUY minus SELL). We use the 40-day period from Day -45 to Day -6 relative to the earnings announcement Day t as the control period. The normal daily trading volume is the average daily total buy and sell trades during the control period computed as:, (1) where T is the total number of days with data available for BPS and BMS. Similarly, the normal daily unscaled buy-sell order imbalance is the average daily net buy trades computed as:. (2) We require T to be at least 20 to compute NBPS t and NBMS t. We use a three-day event window and compute the excess buy-sell order imbalance during the event window for earnings announcement Day t as:, (3) where j indexes days relative to the earnings announcement during the event window. This measure captures the excess buy-sell order imbalance around the earnings announcement. Buy-sell order imbalance during the control period is:. (4) The literature on trades and quotes uses trade size to classify trades. Large trades are usually initiated by institutional investors and small trades are usually initiated by 11

14 individual investors (Lee 1992; Bhattacharya 2001). To analyze trading behavior of individual (less sophisticated) and institutional (more sophisticated) investors, we compute EXBMS for all trades, small trades and large trades. Specifically, if the dollar value of a round lot trade is below $10,000, we classify the trade as a small trade, one likely initiated by an individual. If the dollar value of a round lot trade is above $30,000, we classify the trade as a large trade, one likely initiated by an institution. Note that there are no exact cutoffs for defining individual versus institutional trades. The $10,000 and $30,000 cutoffs are used here to capture the likelihood of an individual or institutional trade. Regression Model We investigate whether investors trade based on the information in book-tax differences by regressing the excess buy-sell order imbalance, EXBMS it, (described above), for firm i during the three day event window surrounding the earnings announcement for period t on analysts-based earnings surprises, the decile rank of the absolute value of a firm s change in book-tax difference for the year, and the interaction between analysts-based earnings surprises and decile rank of the absolute value of a firm s change in book-tax difference as follows: EXBMS it = β 0 + β 1 ABS BTD Rank it + β 2 UE it + β 3 UE it ABS BTD Rank it + β 4 logmve it + ε it. (5) In Equation (5), ABS BTD Rank it is the decile rank of the absolute value of the change in the difference between book income and estimated taxable income ( BTD) for firm i from period t 1 to period t. Values of ABS BTD Rank it range from zero to nine. We calculate BTD it as the change in firm i s book-tax difference from year t 1 to year t, 12

15 where book-tax difference is book income minus estimated taxable income deflated by the market value of equity at the beginning of period t (data 199 data 25). 13 We examine the absolute value of the change in total book-tax differences for three reasons: Hanlon (2005) suggests that firms with large positive or large negative book-tax differences have less persistent earnings; a change specification is consistent with our short-window changes model for investor trading and earnings; and both temporary and permanent book tax differences may be informative for stakeholders (i.e., both may be indicative of lower quality earnings) because each capture managerial incentives and discretion underlying the calculation of book and taxable income. 14 For example, temporary differences include discretion for items that are reflected in one set of books in a different period than the other set of books. Depreciation is a common temporary difference. For book purposes, firms estimate the useful life of an asset, select a method of depreciation, and estimate an asset s salvage value. For tax purposes, the only decision is whether to use an accelerated method of depreciation (Revenue Procedure provides the depreciable life of every asset, and salvage value is ignored). Permanent items capture, among other things, discretion in reporting events that are reflected in one set of books but never in the other set, including some forms of off-balance sheet financing and tax shelters. 15 Following Hanlon et al. (2005), we calculate taxable income as the sum of federal tax expense (data 63) and foreign tax expense (data 64) divided by the top U.S. statutory 13 Unless otherwise noted, data refers to Compustat data items. 14 As discussed in Section II, managers generally have incentives to report higher book income and lower taxable income. 15 For example, different gain or loss treatment for book and tax purposes arising from the initial asset transfer to create synthetic leases or securitizations. See Mills and Newberry (2005) for more details on off-balance sheet financing, and Treasury (1999), Graham and Tucker (2006) and Wilson (2009) for additional details on specific tax shelters. 13

16 tax rate less change in net operating loss carryforward (data 52). 16 If federal or foreign tax expense is missing, we measure tax expense as the difference between total income tax expense (data 16) and deferred income tax expense (data 50). 17 Book income is pretax book income (data 170) less minority interest (data 49). Consistent with prior research, we define unexpected earnings, UE it, as firm i s year t actual earnings per share minus the single most recent analyst forecast available prior to the annual earnings announcement, both available from the Unadjusted I/B/E/S Detail History File, deflated by beginning of period t price per share (data 199) (Brown and Kim 1991; Brown and Caylor 2005; Ayers, Jiang and Yeung 2006). We expect that the interaction between UE and ABS BTD Rank, UE ABS BTD Rank, to be negative if investors have a more muted response to the earnings surprises of firms with large booktax differences. Institutions tend to invest in large stocks (Badrinath, Kale, and Noe 1995) and thus, net buying by institutions and individuals may be systematically related to firm size. We represent firm size as the logarithm transformation of market capitalization at the beginning of the year (data 199 data 25), denoted as logmve it. We expect net institutional buying to be positively related to logmve it and net individual buying to be negatively related to logmve it. 16 The statutory tax rate (STR) equals 35% for all the tax years in our sample ( ). 17 Although Hanlon (2003) identifies seven major problems in estimating a firm s tax liability and similarly taxable income from financial statements, estimated taxable income is the appropriate measure in our setting because it is the measure available to the market (as opposed to actual taxable income, which is not publicly available). 14

17 IV. RESULTS Sample Our primary sample consists of the intersection of the 2006 Compustat database, the Unadjusted I/B/E/S Detail History File, and the Trade and Quote (TAQ) database. We begin our sample period in 1993 (i.e., the effective date of SFAS No. 109) and exclude financial institutions (SIC codes ), public utilities (SIC codes ), and firms incorporated outside of the United States because these firms face different tax, regulatory, and/or financial reporting issues than the remaining Compustat population. We require that each firm-event observation have sufficient data to allow us to estimate the variables in Equation (5). These criteria result in a final sample consisting of 18,037 firm-event observations from 1993 to Descriptive Statistics Table 1 provides variable definitions, while Table 2 provides statistics on the daily frequency of shares traded and the total number of quotes that we process for our analysis. To obtain a normal level of buy-sell order imbalance, we processed a total of 312 million quotes (161 million buys and 151 million sells) during the 40-day control period prior to the event window. There are on average 229 buys and 211 sells per day during the control period. For small trades there are 118 buys and 111 sells per day, and for large trades there are 49 buys and 43 sells per day. The sum of small and large trades does not equal the total for all trades because we have a buffer of $20,000 between the trade sizes identifying small and large trades (i.e., we classify trades of less than $10,000 as small trades and trades of more than $30,000 as large trades). The daily frequencies of 15

18 trades increase during the three-day event period. For example, the average number of daily buys increases from 229 during the control period to 260 on Day -1, to 331 on day 0, and to 407 on Day 1. The average number of daily sells increases from 211 during the control period to 233 on Day -1, to 299 on Day 0, and to 370 on Day 1. The patterns for small and large trades are similar. Table 3 provides descriptive statistics for our sample. The mean capitalization, MVE, for our sample firms is $3.24 billion, and the median capitalization of $0.40 billion. The mean unexpected earnings is , and the median is The mean absolute value of the change in book-tax difference is , and the median is During the three-day event window, the mean and median daily excess buy-sell order imbalances for all trades are and , respectively. During the control period, the mean and median daily buy-sell order imbalances are and , respectively. Thus, as a whole, investors increase their net buying activities during earnings announcements. Similarly, for small trades, the mean and median daily buy-sell order imbalances increase from and during the control period to and , respectively, during the event period. For large trades, the mean and median daily buy-sell order imbalances increase from and during the control period to and , respectively, during the event period. Table 4 reports correlation coefficients between variables. Earnings surprise, UE, and the absolute value of the change in book-tax difference, ABS BTD, are negatively correlated (ρ = , p = , Pearson correlation). Market capitalization, MVE, and the absolute vale of the change in book-tax difference, ABS BTD, are also negatively correlated (ρ = , p = , Pearson correlation). Excess buy-sell order 16

19 imbalances, EXBMS, for all trades, small trades and large trades are all positively correlated with each other. EXBMS for all trades, small trades and large trades are all positively correlated with earnings surprise, UE, based on Spearman correlation statistics but are not correlated with UE based on Pearson correlation statistics. Main Regression Results Table 5 provides results for our primary regression analysis. 18 For all trades, the coefficient on earnings surprise, UE, is positive and significant (β 2 = , t = 2.21), suggesting that for investors as a whole, positive (negative) earnings innovations cause them to buy (sell) more shares around earnings announcements. Results for each group (small and large trades) are similar, although the coefficient for UE is much bigger for large trades (β 2 = , t = 3.66) than for small trades (β 2 = , t = 2.52). This evidence is consistent with prior research (Bhattacharya 2001; Battalio and Mendenhall 2005) that suggests that institutional investors (denoted by large trades) are more likely to trade based on analysts forecasted earnings than individual investors (denoted by small trades). To test our first hypothesis, we report the results of a pooled regression of Equation (5) for all trades in column one of Table 5. Hypothesis 1 predicts that the association between buy-sell order imbalance and earnings surprises declines for firms with large book-tax differences. Consistent with this prediction, the coefficient for UE ABS BTD Rank is negative and significant (β 3 = , t = -2.08) for all trades. 18 In untabled sensitivity analyses, we replicate our tests reported in Table 5 including firm and year fixed effects to control for cross sectional dependence that can result from having multiple observations from the same firm or period. All inferences remain. 17

20 To test our second hypothesis, we report whether the coefficient for UE ABS BTD Rank significantly differs for small and large trades by running a pooled version of Equation (5) that includes both trade types and a dichotomous variable (INST) equal to one for large trades and zero otherwise. We include INST separately in the model to capture any intercept effect and interact INST with each term in Equation (5). The last column of Table 5 reports these tests. Hypothesis 2 predicts that the effect of book-tax differences on the association between buy-sell order imbalances and earnings surprises is bigger for large trades than for small trades. Consistent with this prediction, the coefficient on UE ABS BTD Rank is more negative for large trades (β 3 = , t = -3.74) than for small trades (β 3 = , t = -2.33). The difference between the two coefficients ( ) is significant (p-value = , one-tailed test). In sum, we find that, on average, large changes in book-tax differences mitigate the association between analysts-based earnings surprises and abnormal buy-sell pressure, and this effect is more pronounced for large trades. To the extent that investors who initiate large trades are sophisticated institutional investors, the above results suggest that sophisticated investors, at least in part, are able to see through the valuation implications of book-tax differences and thus trade accordingly. We note that the differences in trading on book-tax differences across large and small trades are similar to the differences in trading on analysts-based earnings surprises. Specifically, the coefficient for UE ABS BTD Rank for large trades is 9.11 times that of the coefficient for UE ABS BTD Rank for small trades (i.e., / = 9.11), whereas the coefficient for UE for large trades is 8.26 times that of the coefficient for UE for small trades ( / = 8.26). In combination, this evidence implies that 18

21 those investors that are more likely to trade on a sophisticated earnings expectation model (analysts forecasted earnings) are also more likely to trade on book-tax differences. 19 Regarding the magnitude of the effects of book-tax differences on investor trading, we find that the association between buy-sell order imbalance and analysts-based earnings surprises is completely eliminated for firms in the largest decile of book-tax differences. Specifically, for firms in the tenth decile of the absolute value of changes in book-tax differences, the sum of the (a) coefficient for UE, , and (b) product of the book-tax difference decile rank times the coefficient for UE ABS BTD Rank, ( ), is not significantly different from zero (p-value = 0.48, two-tailed test). We find a similar relation for both large and small trades (although again, the base relation between small trades and analysts-based earnings surprises is much smaller). For large trades for firms in the tenth decile of book-tax differences, the sum of the (a) coefficient for UE, , and (b) product of the book-tax difference decile rank times the coefficient for UE ABS BTD Rank, ( ), is not significantly different from zero (p-value = 0.34, two-tailed test). For small trades for firms in the tenth decile of book-tax differences, the sum of the (a) coefficient for UE, , and (b) product of the book-tax difference decile rank times the coefficient for UE ABS BTD Rank, ( ), is not significantly different from zero (p-value = 0.49, two-tailed test). These results indicate that book-tax differences have a significant effect on how investors trade on analysts-based earnings surprises. 19 In sensitivity analyses, we replicate our analyses defining UE as the forecast error from a seasonal random walk model. Consistent with sophisticated investors being less likely to trade on a seasonal random walk expectations model, we find no association between buy-sell imbalance for large trades and UE or UE ABS BTD Rank. For small trades, we find an association between buy-sell order imbalance and UE and UE ABS BTD Rank though the coefficient for UE ABS BTD Rank ( ) is quite modest. 19

22 V. SENSITIVITY ANALYSIS We perform sensitivity analyses to address four research design issues including: (1) alternative methods for classifying large and small trades and calculating the buy-sell order imbalance; (2) nonlinearity in the relation between unexpected earnings and excess buy-sell order imbalance; (3) potential correlation between book-tax differences and abnormal accruals; and (4) the effect of tax planning on our results. In addition, we triangulate our trading analyses with additional short-window return tests. Alternative Methods for Calculating Buy-Sell Order Imbalances Because the cutoff point for classifying large versus small trades is an empirical issue, we conduct additional analyses using a variety of trade-size cut-offs to distinguish individual and institutional trades (e.g., below (above) $10,000, $20,000, $30,000, etc.). Results from these alternative specifications of small and large trades are similar to those presented in the tables. We also replicate our tests after redefining the buy-sell order imbalance based on the number of buy or sell orders, instead of shares in buy and sell orders. Results are similar. Finally, we repeat all tests using bid and ask prices (instead of the mid-point of bid and ask prices) to classify trades as buy or sell, i.e., if the traded price is at the ask (bid) price, we classify the trade as a buy (sell). All inferences remain. Nonlinearity in the Relation between Unexpected Earnings and Buy-Sell Order Imbalance In our primary tests, we assume that the relation between unexpected earnings and the buy-sell order imbalance is linear. Nonlinearity can arise because larger values (both positive and negative) of unexpected earnings are negatively correlated with earnings 20

23 persistence, i.e., larger absolute values of unexpected earnings more likely represent transitory earnings (Freeman and Tse 1992). To allow for this possibility, we use the arctangent transformation of unexpected earnings, arctanue, and run the following regression: EXBMS it = β 0 + β 1 ABS BTD Rank it + β 2 arctanue it + β 3 arctanue it ABS BTD Rank it + β 4 logmve it + ε it, (6) where all variables are as previously defined. All inferences remain. Inclusion of Abnormal Accruals We add the absolute value of abnormal accruals to Equation (5) to ensure that book-tax differences do not simply capture the same construct represented by abnormal accruals (i.e., to ensure that book-tax differences provide unique information). We define abnormal accruals using the Jones (1991) model as modified by Kothari, Leone and Wasley (2005) and calculated using the statement of cash flows (Hribar and Collins 2000). Specifically, we estimate abnormal accruals, ABACC, as the difference between total accruals and modified Jones model normal accruals (Kothari et al. 2005). We estimate modified Jones model normal accruals using the following equation crosssectionally by year for each two-digit SIC industry with at least ten observations: (7) TACC is total accruals measured as income before extraordinary items (data 123) less the difference between net cash flows from operating activities (data 4) and extraordinary items and discontinued operations (data 124), all from the statement of cash flows for firm i in year t. ΔSALES it is the change in sales (data 12) from year t 1 to year t. PPE it is firm i s gross property, plant and equipment (data 7) for year t, ROA it-1 is income 21

24 before extraordinary items (data 18) divided by total assets (data 6) in year t 1, and TA it-1 is total assets (data 6) at the end of year t 1, all for firm i. We include an intercept term in Equation (9) because Kothari et al. (2005, 173) advocate that this term helps alleviate heteroskedasticity, mitigate concerns regarding an omitted size variable, and generate a more symmetric measure of abnormal accruals. We interact the decile ranking of the absolute value of abnormal accruals, ABSABACC Rank, with unexpected earnings, UE, and report the results from this analysis in Table 6. When we include ABSABACC Rank and UE ABSABACC Rank in Equation (5), both coefficients are insignificant for all trades and large trades, while the coefficient on the interaction term, UE ABSABACC Rank, is significant but positive for small trades. This suggests that individuals (as represented by small trades) engage in more trading when abnormal accruals are large. In addition, all results are consistent with our main analyses. For example, the coefficient for UE is positive and significant (β 3 = , t = 1.82 for all trades; β 3 = , t = 1.82 for small trades; β 3 = , t = 3.39 for large trades), the coefficient for UE ABS BTD Rank is negative and significant (β 5 = , t = for all trades; β 5 = , t = for small trades; β 5 = , t = for large trades), and the coefficient for UE ABS BTD Rank is significantly more negative for large trades than small trades (p-value = , onetailed test). Tax Planning Prior research suggests that taxable income is less informative as a performance measure to equity investors when firms engage in tax planning (Ayers, Jiang and 22

25 Laplante 2009). This evidence is consistent with tax planning resulting in taxable income that obscures the firm s actual performance and suggests that equity investors, at least in part, distinguish sources of book-tax differences. To the extent that changes in book-tax differences are attributable to tax planning and traders incorporate the causes of book-tax difference changes in their analysis, we anticipate that tax planning will mitigate the effect that book-tax differences have on the relation between unexpected earnings and investor trading (i.e., because tax planning obscures the information content of taxable income and thus, book-tax differences). 20 To investigate this possibility, we expand our model to include an indicator variable for high tax planning firms (TaxPlan) and interact this indicator with our measures of changes in book-tax differences (ABS BTD Rank) and unexpected earnings (UE) as follows: EXBMS it = δ 0 + δ 1 TaxPlan it + δ 2 ABS BTD Rank it + δ 3 ABS BTD Rank it TaxPlan it + δ 4 UE it + δ 5 UE it TaxPlan it + δ 6 UE it ABS BTD Rank it + δ 7 UE it ABS BTD Rank it TaxPlan it + δ 8 logmve it + υ it. (8) We identify high tax planning firms in Equation (10) as firms in the lowest quintile of accumulated effective tax rates (ETR) among Compustat firms for each year and two-digit SIC industry calculated as follows:. (9) We calculate ETR two ways. In our first measure, Trad ETR, the numerator, X, is current tax expense calculated as total tax expense (data 16) less deferred tax expense (data 50) for firm i summed over the five year period from t 5 through t 1. The denominator is the difference between PTBI, pre-tax book income (data 170), and Special 20 Hanlon (2005) also argues that aggressive tax planning should not be associated with lower persistence in financial accounting earnings. 23

26 Items (data 17) accumulated for firm i over the five year period from t 5 through t 1. If Special Items is missing, we set it equal to zero. In our second measure of ETR, Cash ETR, the numerator, X, in Equation (12) is cash taxes paid (data 317) from the cash flow statement for firm i summed over the five year period from t 5 through t 1, while the denominator remains the same. If cash taxes paid is missing for a particular year, we set it equal to current tax expense for that year. Both measures of ETR capture the effects of tax planning strategies that defer or permanently avoid taxable income (i.e., deferred tax expense is excluded from the calculation of Trad ETR, and Cash ETR only reflects taxes once they are paid) (Dyreng, et al. 2008). In Equation (11), TaxPlan it equals one for firms in the lowest quintile of accumulated ETR for each year and two-digit SIC industry, and zero otherwise. A significantly positive coefficient on UE it ABS BTD Rank it TaxPlan suggests that tax planning attenuates the association between buy-sell order imbalance and earnings surprises for firms with large book-tax differences. Untabled results indicate that tax planning does not affect our primary results for small traders. For large traders, however, the coefficient on UE it ABS BTD Rank it TaxPlan is positive and significant (not significant) for the Trad ETR (Cash ETR) measure of tax planning. These analyses provide some evidence that tax planning influences how institutional investors (as denoted by large trades) trade on book-tax differences. Announcement Period Returns To triangulate evidence that investors trade on book-tax differences in a predictable manner around earnings announcements, we investigate whether the 24

27 association between announcement period equity returns and analysts-based earnings surprises declines as book-tax differences increase. Specifically, we modify Equation (5) such that our dependent variable is announcement period abnormal returns (RET) instead of excess buy-sell order imbalance, as follows: RET it = β 0 + β 1 ABS BTD Rank it + β 2 UE it + β 3 UE it ABS BTD Rank it + β 4 logmve it + ε it (10) where RET is calculated as the difference between firm i's actual return and the CRSP value-weighted market index return from day t 1 through day t + 1. If investor trading on book-tax differences is sufficient, on average, to influence price reactions around earnings announcements, we would expect the coefficient on UE ABS BTD Rank to be negative. Results (not tabulated) confirm this expectation. The coefficient for UE is positive and significant (β 2 = , t = 2.63) and the coefficient for UE ABS BTD Rank is negative and significant (β 3 = , t = -1.76). Although this analysis does not allow us to identify the types of investors (large vs. small) who trade on book-tax differences around earnings announcements, these results provide additional comfort in the conclusion that investors incorporate book-tax information into their trading decisions around earnings announcements. VI. CONCLUSION We use short-window tests of buy and sell orders on the days surrounding earning announcements to investigate whether investor trades vary with book-tax differences in a predictable manner. We measure buy-sell activity daily as shifts in supply and demand through changes in buy-sell order imbalances from the Trade and Quote (TAQ) database. 25