Does Tax Aggressiveness Reduce Financial Reporting Transparency?

Transcription

1 Does Tax Aggressiveness Reduce Financial Reporting Transparency? Karthik Balakrishnan Phone: (215) Jennifer Blouin* Phone: (215) Wayne Guay Phone: (215) All authors are at the Wharton School, University of Pennsylvania 1300 Steinberg Hall-Dietrich Hall First draft: October 24, 2010 This draft: September 20, 2011 Abstract This paper investigates whether aggressive tax planning firms have less transparent information environments. Although tax planning provides expected tax savings, it can simultaneously increase the complexity of the organization. And, to the extent that this greater complexity cannot be adequately communicated to outside parties, such as equity investors, creditors, and analysts, transparency problems can arise. Our investigation of the association between a newly developed measure of tax aggressiveness and measures of information uncertainty, information asymmetry and earnings quality suggests that aggressive tax planning increases the opacity of a firm s information environment. We find some evidence, however, that managers increase the volume of disclosure in an attempt to mitigate these transparency problems. Overall, our results suggest that firms face a trade-off between financial transparency and aggressive tax planning thereby potentially explaining why some firms appear to engage in more conservative tax planning than would otherwise be optimal. JEL Classification: H20; M41 Keywords: Tax aggressiveness; tax planning; information content; earnings quality * Corresponding author. We appreciate comments from seminar participants at the London Business School, University of Auckland, University of Colorado, University of Connecticut, and the University of Melbourne.

2 1. Introduction Corporations engage in various forms of tax planning to reduce expected tax liabilities. These expected benefits, however, do not come without costs. Such costs include direct labor and information systems necessary to carry out the tax planning, as well as expected costs of negotiation and penalties stemming from interactions with taxing authorities. In this paper, we examine a previously unexamined cost of tax planning related to financial transparency. Specifically, we argue that aggressive tax planning can increase organizational complexity, which can, in turn, reduce financial transparency. As an illustration of this point, consider the following Double Irish tax planning technique used by pharmaceutical firm Forest Laboratories. Forest Laboratories Irish subsidiary, Forest Laboratories Holdings Ltd., reorganized in 2005 by creating a new Irish subsidiary, Forest Laboratories Ireland, and relocating itself to Bermuda. Although the new Irish subsidiary handled manufacturing, its new Bermudian entity was responsible for the licensing of patents. Forest Laboratories Ireland paid the Bermudian firm a royalty fee for the use of the patents and, since Bermuda does not have an income tax, this organization structure reduced Forest Laboratories Ireland s tax rate to 2.4% from 10.3%. To further reduce Forest s worldwide tax liabilities, the royalty payment made to the Bermudian entity is paid to Forest Finance BV, a Dutch affiliate. By routing the royalty payment through the Netherlands, Forest Laboratories Ireland avoided paying a 20% withholding tax that would be necessary if the royalty was paid to an entity outside of the EU (the Netherlands has no such withholding requirement). In the year of the reorganization, foreign operations reduced Forest s effective tax rate by 21.8% (see Drucker 2010a for a detailed description of the transaction). 1

3 Although the tax planning described above appears to provide expected tax savings to the parent company, it simultaneously increases financial and organizational complexity. And, to the extent that this greater complexity cannot be adequately communicated to outside parties, such as equity investors, creditors, and analysts, transparency problems can arise. Bushman, Chen, Engel, and Smith (2004, p. 175) describe this transparency potential problem as follows: Operational complexities can arise as firms act to arbitrage institutional restrictions such as tax codes and financial restrictions (Bodnar et al., 1998). For example, firms may employ complex transfer pricing schemes to shift profits to low tax jurisdictions that can complicate efforts by shareholders and board members to understand firms foreign operations. Of course, managers may respond to this greater financial complexity by augmenting financial reporting and disclosures in an attempt to maintain a transparent information environment. At the same time, managers (and shareholders) may have countervailing incentives to provide limited disclosure regarding aggressive tax planning activities. In particular, although US corporations are required to disclose details about material operations in other tax jurisdictions, managers may be hesitant to transparently disclose the details of these subsidiaries if doing so would provide a roadmap for an audit by the tax authorities. Hence, when aggressive tax planning increases financial and organizational complexity, managers financial reporting and disclosure choices may not serve to mitigate the increased opacity. Thus, we view the extent to which aggressive tax planning reduces transparency to be an empirical issue. Although our Forest Laboratories example above describes a specific tax planning technique that most would agree is aggressive, we are aware of no universally accepted definition of aggressive tax planning. Frank, Lynch, and Rego (2009 p. 468) define aggressive tax reporting as, downward manipulation of taxable income through tax planning that may or 2

4 may not be considered fraudulent tax evasion. However, Slemrod (2004) argues that tax aggressiveness is a more specific activity, being encompassed by transactions where the primary purpose is to lower the firm s tax liability. For our study, we define a tax aggressive firm as one that pays an unusually low amount of tax given the firm s industry and size. Although our measure of tax aggressiveness does not rely on or capture any specific tax planning technique, it does incorporate the notion that, all else equal, similar firms should have similar tax planning opportunities. And, among firms with similar planning opportunities, firms with lower tax liabilities can be considered more tax aggressive. As an alternative measure of tax aggressiveness, we use the number of tax haven countries in which a firm has operations. By introducing additional foreign subsidiaries through which intercompany transactions flow, financial and organizational complexity is increased. We document that tax aggressive firms are characterized by lower transparency. Specifically, we find that firms with unusually low tax liabilities or more tax haven operations have larger analysts forecast errors, greater analysts forecast dispersion and a higher adverse selection component of the bid-ask spread. Tax aggressive firms also exhibit lower accruals quality, measured using several approaches advanced in the earnings quality literature. In addition to our main tests, we subject our analysis to a variety of sensitivity analyses including robustness to the treatment of loss firms, alternative measures of tax aggressiveness, and alternative controls for organizational complexity. Overall, our results suggest that the benefits of tax aggressiveness come at a cost of lower financial transparency. We also investigate whether managers at tax aggressive firms recognize that tax planning can give rise to transparency problems, and respond by augmenting their disclosures. Consistent with this conjecture, we find that the management discussion and analysis (MD&A) section of 3

5 the 10-K report is, on average, lengthier for tax aggressive firms. Further, tax aggressive firms that provide additional disclosure in the MD&A exhibit significantly smaller spreads than tax aggressive firms that do not provide additional disclosure (our disclosure results for analyst forecast errors and dispersion are directionally similar but not statistically significant). Our results advance the literature on the relation between financial transparency and taxrelated decisions. It is well documented that differences between book and tax income provide information to market participants (see Hanlon and Heitzman 2010). For example, Lev and Nissim (2004) and Weber (2009) find that the ratio of taxable income to book income is useful in predicting earnings growth, and Hanlon (2005) documents that extreme book-tax differences provide a signal on the persistence of accruals. Further, when earnings management increases the spread between book and tax income, the ability of accruals to provide information about future cash flows can be constrained (e.g., Dhaliwal et al and Comprix et al. 2010). These papers, however, do not investigate how aggressive tax planning, irrespective of book-tax differences, alters the firm s information environment. 1 We provide evidence suggesting that aggressive tax planning increases financial and organization complexity in a manner that increases investor uncertainty about future profitability as well as increases information asymmetry between investors. Overall, our findings highlight lower financial transparency as a potentially important cost of aggressive tax planning. These results may help explain why some firms appear to engage in more conservative tax planning that would otherwise be optimal. In addition, we develop new measures of tax aggressiveness that incorporate the notion that tax planning opportunities are 1 There is also a tangential line of research that investigates whether aggressive tax planning provides an opportunity for insiders to extract rents (Desai and Dharmapala, 2006; Desai, Dyck, and Zingales, 2007; Hanlon, Hoopes, and Shroff, 2010). Our analysis does not presume any managerial incentives to tax plan beyond their implicit alignment with shareholder interests. 4

6 expected to vary cross-sectionally, and over time. Our measures benchmark firms estimated tax burdens by size and industry to obtain a measure of the aggressiveness of a firm s tax planning relative to other firms in their operating environment. These measures may be useful to future researchers investigating the relation between tax aggressiveness and firm behavior. The paper proceeds as follows. Section 1 provides a summary of prior literature and develops our hypotheses on the relations between aggressive tax planning and financial reporting transparency. Section describes our sample selection procedures and descriptive statistics. We present our results on the relation between tax aggressiveness and transparency in Section 4, and our results on whether tax aggressive firms augment their disclosures in Section 5. Section 6 provides sensitivity analyses and in Section 7, we conclude. 2. Prior Research and Hypothesis Development A. Trade-offs between tax and non-tax costs In their seminal textbook, Scholes and Wolfson explain that managers face conflicts between financial reporting and tax planning. While managers often desire to report high levels of income to investors, they simultaneously desire to report low levels of income to the tax authorities. In the U.S., as in many other countries, tax reporting rules differ from financial reporting rules, allowing firms to report disparate levels of income to tax authorities and to investors. However, as many economic transactions are reported similarly for book and tax reporting, firms often face a trade-off between cash tax savings and lower reported earnings. 2 Of course, costs of lower reported earnings is only one of many potential direct and indirect costs of tax planning. Direct costs of tax planning include labor, information systems, 2 For example, see Scholes, Wilson and Wolfson (1990), Guenther, Maydew, and Nutter (1997); Maydew (1997); Matsunaga, Shevlin, and Shores (1992). 5

7 coordination among business units, expected audit costs, and expected penalties in the event that tax planning strategies are found to be inappropriate. Indirect costs include potential agency conflicts between managers and shareholders, as well as reputational costs of being an overly aggressive tax planning firm. A further potential indirect cost, and the focus of our study, is the effect of tax planning on organizational transparency. As illustrated in our Forest Laboratories example above, tax planning strategies can substantially increase organizational complexity. Other examples of how tax planning strategies can increase complexity include the creation of entities for multistate tax planning (e.g., captive REITs, intangible holding companies); net operating loss monetization; and capital loss utilization. As noted by Bushman, Chen, Engel, and Smith (2004), organizational complexity can, in turn, hinder efforts by investors to understand the firm s operations. Complexity can influence transparency through multiple channels. One possibility is the influence of tax planning-related complexity on the quality of financial accounting, say through the influence of tax planning on the accrual process. For example, consider that many planning opportunities require the bifurcation of business activities into separate legal structures (e.g., income that qualifies for treaty based withholding taxes, activity qualifying for the domestic manufacturers deduction). If the separation of the business activities increases the complexity of the accounting because either there is more accounting to be done or there is less understanding of how to record activity for separate activities, then the quality of the accruals process may decline. Even in the absence of financial accounting issues, however, the influence of tax planning on operational and financial strategies can render the firm more opaque (as the Forest Laboratories example illustrates). Thus, even the most well-intentioned management may have difficulty providing investors with 6

8 disclosures that reduce information uncertainty and/or information asymmetries between investor groups. In summary, we predict that aggressive tax planning will increase information uncertainty and information asymmetry, and will reduce the quality of financial reporting. Our research question is related to the call by Shackelford and Shevlin (2001) for further research on the drivers of cross-sectional variation in tax planning. If reduced corporate transparency is a cost of aggressive tax planning, and this cost varies cross-sectionally, then one would expect to observe variation in tax planning across firms. Our study may also shed light on the observation by Armstrong, Blouin and Larcker (2010) that aggressive tax planning appears to be underutilized by firms given large potential benefits and relatively small potential costs stemming from audits, interest and penalties. Transparency-related costs may explain some of this apparent underutilization of aggressive tax planning. Before moving on to our analysis, we note that our research question differs from existing work on book-tax differences. Differences between book and tax reporting have been used extensively to study the earnings quality of firms (e.g., Lev and Nissim 2004; Hanlon 2005) and whether taxable income contains information incremental to pre-tax income (e.g., Hanlon, LaPlante, and Shevlin, 2005; Hanlon, Maydew and Shevlin, 2008). 3 Hanlon (2005) and Comprix et al. (2010) document that large book-tax differences are associated with less 3 These papers have led to what is known as the book-tax conformity. Many have argued that allowing firms to report separate incomes for book and tax is precisely what leads to aggressive tax and GAAP reporting (e.g., Desai, 2005). If firms were forced to conform their book and tax reporting, then firms would be relatively less incentivized to undertake earnings management or extreme tax planning. Economists propose that firms should report their GAAP income on their tax returns (so, taxable income should confirm to book). Hanlon, LaPlante, and Shevlin (2005) and Hanlon, Maydew, and Shevlin (2008) argue that conformity would result is the loss of information to the capital markets. To date, the book-tax conformity debate continues with little consensus. Although many in the academic community agree that conforming tax reporting and book reporting is not a good idea (see Shackelford and Slemrod 2004), there is less agreement as to whether conformity ultimately leads to a loss of information to the capital market s (e.g. Raedy, Seidman, and Shackelford, 2010; Atwood, Drake, and Myers, 2010). 7

9 persistent earnings and lower quality of accruals. In a similar vein, Dhaliwal et al. (2008) examines whether large book-tax differences are associated with a higher cost of equity capital. The extant book-tax difference literature implies that book-tax differences are informative to the capital markets because they capture earnings management behavior; not because they include information regarding tax planning. Our interest is determining whether tax planning itself can leave its imprint on firms accounting quality. We conjecture that firms entering into complex tax transactions potentially lessen the transparency (or increase the opacity) of their financial statements. Consequently, firms are effectively trading off earnings quality for cash tax benefits. B. Measuring Aggressive Tax Planning There is no well-accepted measure of tax aggressive. Lisowsky et al. (2010) provide a continuum of the ability of specific measures of firms tax attributes to capture tax aggressiveness but they do not empirically test whether their conjectures are true. Although several recent papers have attempted to develop measures intended to specifically capture tax aggressiveness, each suffers from several weaknesses. Wilson (2009) and Lisowsky (2010) create measures of the estimated probability that a firm has entered into tax shelter. Both measures are derived by estimating a probability from the coefficients of a logit model of attributes of firms discovered engaging in shelter activity. The trouble with both of these measures is two-fold. First the measures rely on a very small sample of firms whose shelter behavior was a) detected by the tax authorities and b) litigated. 4 Wilson (2009) and Lisowsky (2010) rely on 59 and 211 of shelter-engaging firms, respectively, to generate the parameters for 4 Wilson (2009) relies on court records discussed in the popular press to identify his shelter firms. Lisowsky (2010), which is an extension of Wilson (2009), identifies shelters using proprietary IRS data. His sample of shelter transactions come from the Office of Tax Shelter Analysis, which identifies shelters through the audit process. 8

10 estimating sheltering probabilities. Frank, Lynch, and Rego (2009) develop a measure of discretionary permanent differences, DTAX, which relies on the premise that permanent differences are more aggressive than timing differences. Although anecdotal evidence suggests that the optimal tax planning opportunity is one that creates permanent differences because of their financial statement benefits, there is little evidence to support this conjecture (see Hanlon and Heitzman, 2010). Much of the work on tax planning is focused on understanding cross-sectional variation in tax aggressiveness. Papers have found evidence that extreme tax planning is associated with executive compensation and risk taking (e.g., Rego and Wilson 2010; Brown, Drake, and Martin 2010; Li et al. 2010). However, these papers do not reach a consensus on how to measure tax aggressiveness. Furthermore, we are unaware of any measures that attempt to capture firms aggregate level of tax aggressiveness. Although the shelter probabilities and DTAX are likely correlated with aggressive tax planning, neither satisfactorily captures all potential tax planning. The shelter probabilities rely on detecting specific law breaking tax-related transactions, and although we would agree that such transactions are aggressive, the shelter probability measure does not capture the type of legal tax planning described by the Forest Industries example above. Rather, our objective is to develop a measure of tax aggressiveness that includes legal strategies that lack any business purpose beyond the potential tax benefits as well as any specific tax shelter activity. Further, our intention is to study variation in aggressive tax planning that stems from both timing and permanent differences. DTAX only includes aggressive permanent differences. In summary, although some proposed measures of aggressive tax planning exist, none seems to encompass the aggregate level of tax aggressiveness of a particular firm. Existing 9

11 measures also fail to measure aggressiveness relative to some benchmark of a normal level of tax planning. For example, some industries have far more extensive foreign operations than others (computer manufacturing as compared to food distributers). These industries may well have greater ability to take advantage of various tax planning strategies, but presumably investors or analysts that follow these industries will be aware of such strategies. When investors and analysts are aware of common industry practices, it seems plausible that such strategies do not create substantial transparency problems. Thus, for our purposes, the ideal measure would capture a) variation in firms total tax planning (timing and permanent) and b) some notion of how a given firm compares to other firm s with a similar underlying production function. We construct our measure in two steps. First, we use the GAAP and Cash effective tax rates (ETR) as proxies for each firm s aggregate tax burden. GAAP ETR (Cash ETR) is the total tax expense (total cash tax paid for income taxes) over the pre-tax income. 5 These measures are bottom-line measures of tax burden, and therefore will reflect the firm s total tax planning efforts. To avoid year-to-year noise in our measures of tax planning, we estimate the effective tax rates by aggregating three years of data (see Dyreng, Hanlon, and Maydew 2008). So, GAAP ETR (Cash ETR) is the sum of the past three years (t to t-2) of total tax expense (total tax paid for income taxes) over the sum of the past three years of pre-tax income. To obtain a measure of whether a firm engages in an unusual (i.e., aggressive) amount of tax planning, we then adjust each firm s average three-year ETR based on industry (48 Fama and French 1997) and size (quintile of total assets). We adjust the ETRs for size and industry by sorting independently on industry and size, and then subtracting the mean ETR over the last three year for the size-industry matched bin. Our measures, TA_GAAP and 5 We truncate the GAAP and Cash ETR to be between 0 and 1. 10

12 TA_CASH, are the industry-size matched GAAP ETR and Cash ETR less the firm s GAAP ETR and Cash ETR, respectively. Thus, positive values of TA_GAAP and TA_CASH suggest that the firm pays less tax than its size-industry peers, and greater values for this measure imply greater tax aggressiveness. In appendix 1, we discuss tax planning strategies at Google and Forest Laboratories to illustrate of how our measure captures tax aggressiveness. Although TA_GAAP and TA_CASH are bottom line measures that reflect the results of aggressive tax planning, they do not speak to any specific tax planning strategies. As noted above, strategic choices related to geographic operating and financing activities are one common tax planning strategy for large corporations. As a proxy for these geographic tax planning activities, we examine the number of tax haven countries in which a firm has operations. Tax havens are countries that are well known to offer firms various tax advantages for locating certain operating and financing activities within the countries borders. Hines and Rice (1994) and Dyreng and Lindsay (2009) both provide evidence consistent with firms using tax havens to reduce their tax obligations. We discuss our measurement of our tax havens variable in more detail in Section 4. C. Transparency Measures and Predictions If aggressive tax planning increases financial and organizational complexity, we predict that aggressive tax planning will increase information uncertainty and information asymmetry, and will reduce the quality of financial reporting (notwithstanding efforts by management to clarify tax planning via augmented clarifying disclosures, an issue we discuss in more detail below). Therefore, we construct measures of each of these three facets of transparency. As proxies for information uncertainty, we use absolute analyst forecast errors and dispersion in analyst forecasts. By adding tax related opaqueness to the financial statements, we 11

13 expect that aggressive tax planning will increase uncertainty about expected earnings (see, for example, the Google example in Appendix 1). Thus, we predict that aggressive tax planning will be positively related to analyst forecast errors in absolute terms, and positively related to analyst forecast dispersion. We measure analyst forecast errors as the average absolute analyst forecast error over the three years corresponding to the measurement of our tax aggressiveness measures (Gu and Wu 2003). Each year, the forecast errors are the absolute value of the difference between median analyst estimate of forecasts issued immediately before the fiscal year-end and the actual earnings for that fiscal year scaled by the price at the end of previous year. Forecast dispersion is estimated as the standard deviation of the analysts forecasts issued immediately before the fiscal year-end scaled by lagged price over the same three years as the tax aggressiveness measure. We use the adverse selection component of the bid-ask spread as a proxy for information asymmetry. This variable measures the extent to which prices are affected by unexpected order flow and is increasing in the level of information asymmetry among investors. We estimate the adverse selection component of the bid-ask spread, Spread, following Madhavan, Richardson, and Roomans (1997) as described in Armstrong, Core, Taylor, and Verrecchia (2010) to take into account cross-sectional differences in firm size. 6 To estimate Spread, we gather trade-bytrade and quote data from the ISSM and TAQ databases. We match trades and quotes using the Lee and Ready (1991) algorithm with a five second lag to infer the direction of the trade (i.e., buy or sell). Once trades are classified as either buyer- or seller-initiated, we estimate the following firm-specific regression using all transactions available during the month: p t /p t-1 = ψ D t + λ (D t ρd t-1 ) + u t, (2) 6 This measure is sometimes referred to as lambda in the literature. 12

14 where p t is the transaction price, D t is the sign of trade (+1 if buy and -1 if sell), and ρ is the AR(1) coefficient for D t. We measure Spread (λ) at a monthly level using all intra-day data for that month to estimate equation for each firm in the sample. We use the average over the three years corresponding to the measurement of the tax aggressiveness measures in our tests. We predict that aggressive tax planning is positively related to information asymmetry as measured by Spread. As a proxy for the quality of financial reporting, we use a measure of accruals quality (AQ) that follows Francis et al. (2005). We construct the AQ measure by first estimating annual cross-sectional regressions for each of the 48 Fama and French (1997) industries (at least 20 observations are required for each industry-year regression): TCA i, t 1 β β β β β β β + ε = 0t + 1t + 2tCFOi, t 1 + 3tCFOi, t + 4tCFOi, t t REVi, t + 6t PPEi, t i, t (2) ATAi, t where, for year t and firm i, TCA is total current accruals and is calculated as the difference between income less the cash flow from operations, ATA is average total assets, CFO is cash flow from operations scaled by average total assets, REV is the change in sales less the change in accounts receivables scaled by average total assets, PPE refers to the property, plant and equipment scaled by average total assets. The accruals quality measure is estimated for each firm i and each year t as the standard deviation of residuals from the above cross-sectional regression over the period t-5 to t-1. We expect that if aggressive tax planning confounds the ability of accrual accounting to resolve timing and matching problems with cash flows, then accruals quality will be lower. Because the Dechow and Dichev(2002) measure is decreasing with accruals quality, we expect that aggressive tax planning will be positively related to the accruals quality measure. 13

15 Further, recognizing the lack of a well-accepted measure of accruals quality in the literature, we examine three alternative measures: 1) Alt_AQ1 is calculated as AQ scaled by the mean absolute value of TCA over the period t-5 to t-1; 2) Alt_AQ2 is a modified version of the accruals quality measure proposed by Wysocki (2008), who argues that the accruals-based measure derived in Dechow and Dichev (2002) does not reliably capture high-quality accruals because of a confounding relation with opportunistic earnings management. He proposes a modification to the AQ measure that aims to extract the contemporaneous association between accruals and cash flows, and better capture the incremental association between current accruals and past and future cash flows over the association between current accruals and current cash flows. This measure is estimated in two steps. First, we estimate two variations of the Dechow and Dichev (2002) model. The first model is a regression of working capital accruals on current cash flows. The second model is the original Dechow and Dichev model that regresses working capital accruals on lagged, current, and future cash flows. We then compute the standard deviation of the residuals of each model during the years t-5 to t-1 and measure Alt_AQ2 as the ratio of the standard deviation of the residuals from the first model to the second model; and, 3) Asset_AQ modifies AQ by first estimating the cross-sectional regression indicated above for each year and each of the 48 Fama and French (1997) industries by quintiles of asset size. This measurement technique is adopted so as to be consistent with the measurement procedure adopted for the tax aggressiveness measures. D. Control Variables We include controls for factors that are expected to influence the quality of a firm s information environment. To control for the size of the firm, we include Size, the log of market value of equity. Larger firms typically have a higher quality information environment. We 14

16 include Leverage, the ratio of long-term debt to total assets, to control for firms debt service needs and capital structure, and Age, the natural logarithm of the number of years the firm has been listed on Compustat, to control for a relation between firm age and the quality of the information environment. We also control for the geographical complexity of the organization as we are interested in the incremental firm complexity created by aggressive tax planning. Complexity is Bushman et al. s (2004) revenue-based Hirfindahl-Hirschman index calculated as the sum of the squares of each geographic segment s sales as a percentage of the total firm sales. We anticipate that more geographically complex firms are relatively more opaque (note that lower values of the index imply more complexity). To control for firms growth opportunities, we include Mkt to Book which is the ratio of the market value of assets to the book value of assets. We also include an indicator variable, Loss, which is equal to one if the firm s income before extraordinary items is less than zero in the current year and zero otherwise. Loss firms typically have lower earnings quality and have higher levels of information asymmetry. In addition, loss firms will typically appear to be very aggressive tax planners, when, in actuality, they merely have very low income. Finally, we include industry as well year fixed effects in all specifications. To ensure that our inferences regarding aggressive tax planning are not confounded by the influence of book-tax differences, we include the absolute value of the mean of the past threeyear s Book-Tax Gap, measured as the difference between pre-tax income less taxable income (defined as current federal tax expense grossed up by the maximum federal statutory tax rate (i.e., 35%) plus pre-tax foreign income less the annual change in NOLs) scaled by total assets. The book-tax gap has been shown to be correlated with firms earnings management activities, and we include the absolute value of the book-tax gap because any earnings management, 15

17 regardless of direction, could affect earnings quality. 7 Comprix et al (2010) and Dhaliwal et al. (2008) both provide evidence that larger book-tax gaps adversely affect firms earnings quality and cost of equity capital, respectively. Finally, with respect to our accruals quality tests, we follow Liu and Wysocki (2008) and include additional control variables for general uncertainty in the firm s operating environment when investigating the association between accruals quality and tax aggressiveness: the standard deviation of operating cash flows (Std Dev of Cash Flows) and the standard deviation of sales (Std Dev of Sales). We expect a positive coefficient on these two variables. 3. Sample We obtain our main sample from Compustat s annual database. The sample period spans from fiscal year 1990 through We retain firms for which we are able to compute the tax aggressiveness measures as well as all of the control variables used in the regression specifications. 8 This results in a sample of 29,970 firm-year observations. Analyst forecast estimates are obtained from I/B/E/S. NYSE TAQ database is used to compute the transaction weighted bid-ask spread. The additional data requirement reduces the sample size to 19,674 firm-year observations in regressions involving analyst data and to 24,470 firm-year observations in regressions involving bid-ask spreads. The tests relating to length of MD&A and tax haven information from Exhibit 21 involves extracting the MD&A section as well as Exhibit 21 from a firm s annual 10-K report. Exhibit 21 reports all of the firms material subsidiaries as well as their jurisdiction (typically their subsidiaries location of incorporation). We obtain these data 7 Note that all of our inferences are identical if we use temporary book-tax differences (i.e., deferred tax expense grossed up by 35%) or the signed book-tax gap, in place of the book-tax gap. All inferences also hold if we remove the book-tax gap from the regression model. 8 We exclude REITS from our analysis as they are not typically subject to entity level taxation. 16

18 from SEC EDGAR. We were able to extract and match MD&A data for 16,382 firm-year observations and Exhibit 21 data for 8,528 observations. 9 Descriptive statistics for our sample are presented in Table 1. The median TA_GAAP is 3.1%, indicating that more than half of our sample firms have a GAAP ETR that is less than the mean of its size-industry peers. At the same time, the median TA_CASH is -2.0%, indicating that more than half of our sample firms have a CASH ETR that is greater than the mean of its size-industry peers. 10 The GAAP and Cash ETR of 27.8% and 23.5%, respectively, are much less than the 35% top statutory corporate tax rate, which could be attributable to either extensive foreign operations and/or tax planning. With respect to other firm characteristics, our sample is comprised of fairly large, mature, and profitable firms (only 26.5% of firms report negative net income). 4. Results We begin our analysis by examining the relation between aggressive tax planning and our three sets of financial transparency measures. Table 2 reports regressions of absolute analyst forecast errors on our ETR-based proxies for tax aggressiveness. The regressions include controls for firm and earnings characteristics discussed above. Consistent with our conjecture that aggressive tax planning increases uncertainty faced by investors with respect to forecasting future profitability, we find that absolute earnings forecast errors are significantly larger for firms with relatively low effective tax rates. In addition, we find that the dispersion of forecast errors 9 MD&A data is available from However, for Exhibit 21 we focused on the post K forms for ease of extraction. Post-2001 companies filed 10-K using the XHTML format that enables easier identification of the Exhibit.. 10 Note that the means of our TA_GAAP and TA_CASH measures are not zero because we use all firms with available ETR data to estimate three-year ETRs in the size-industry bins. Data limitations for our information environment variables, however, further restrict the sample. 17

19 is also higher for tax aggressive firms. Consistent with analysts being less able to accurately forecast earnings in the presence of managerial earnings management, we find that larger booktax differences are associated with higher forecast errors and greater forecast dispersion. In Table 3, we examine whether aggressive tax planning is related to information asymmetry between investors, as measured by the adverse selection component of the bid-ask spreads. Our regressions indicate that the GAAP and Cash-based ETR measures of tax aggressive measures are positively associated with Spread. Taken together with the results in Table 2, this finding suggests that aggressive tax planning not only increases investor uncertainty about future profitability, but also increases the information gap between informed and uninformed investors. Thus, it appears that some investors have a relative advantage in processing information about corporate tax planning. Table 4 explores whether accrual quality is adversely affected by aggressive tax planning. Panel A reports results for regressions of AQ on the ETR-based tax aggressiveness proxies. The findings indicate that GAAP_ETR and CASH_ETR are positively related to AQ (recall that because AQ is the standard deviation of residuals from the accruals regression model, greater values of AQ imply lower quality accruals). Panel B of Table 4 shows that the results in Panel A are robust to using our three alternative proxies for accruals quality. Controlling for book-tax differences, the results in Table 4 indicate that, across an array of accruals quality measures, more aggressive tax planning distorts the relation between accruals and cash flows. We next explore whether our results are robust to an alternative measure of tax planning aggressiveness: the number of haven countries with material subsidiary operations. Firms use of haven countries for financial and operating activities has been linked with evidence of extensive transfer pricing activity (Hines and Rice 1994) and lower effective tax rates (Dyreng and Lindsay 18

20 2009). We collect data on the number of haven countries reported in firms Exhibit 21 in their 10-K. Exhibit 21 is a required element of a firm s 10-K and includes a listing of all of the firms subsidiaries with material operations. Tax havens are jurisdictions that structure a tax regime to take advantage of firms desire to reduce their tax burdens. Generally, tax havens have low or no tax rates and have very little information sharing of tax information with other jurisdictions thereby making it more difficult for one jurisdiction to determine whether a firm artificially is stripping its earnings into the haven. However, a haven could also include countries that have modified their tax laws in order to attract foreign capital (e.g., Ireland). Either type of haven could be utilized for tax avoidance purposes. Because there is no consensus on which countries are considered havens, we rely on two sources to determine whether a country is considered a tax haven (1) Table 1 in Dyreng and Lindsay (2009) and (2) the seven havens identified in Hines and Rice (1994). TAX_HAVENS (TAX_HAVENS_BIG7) is the number of reported subsidiaries in Dyreng and Lindsay s (Hines and Rice s) list of haven countries. Because havens are used for tax avoidance purposes, we predict that greater haven usage is consistent with more aggressive tax planning. The mean number of material operations in havens is 2.3 and 1.2 (see Table 1) for TAX_HAVEN and TAX_HAVEN_BIG7, respectively. However, the median is zero for both measures suggesting that havens are not used by the majority of firms. In Table 5, we replicate the results in Tables 2, 3 and Panel A of Table 4 using the tax haven measures of tax aggressiveness in place of the ETR-based tax aggressiveness variables. The results are similar to those in the earlier tables. Specifically, firms with operations in a greater number of tax haven countries have larger absolute analyst forecast errors, greater forecast dispersion, and larger bid-ask spreads. The haven measure does not appear to explain 19

21 accruals quality. One possible explanation for this latter result is that firms use of tax havens may frequently involve financial activities as opposed to real economic activity, and the former may have relatively little influence on the accrual process. 5. Management disclosure decisions in the presence of aggressive tax planning In light of our findings, it is interesting to consider whether managers increase disclosure to at least partially mitigate the difficulties that investors have in understanding the financial and organizational complexity induced by aggressive tax planning. The predictions regarding augmented disclosure, however, are not unambiguous. On one hand, shareholders and other investors may demand increased disclosure to mitigate the reduced transparency stemming from tax aggressiveness. On the other hand, managers may be reluctant to publicly reveal too many details about their tax planning if this increasing the likelihood that tax authorities will take action against the firm. Shareholders may also recognize these potential costs and resign themselves to accept lower transparency in return for more profitable tax planning activities. In Table 6, Panel A, we explore whether managers who make aggressive tax planning choices increase the volume of disclosure in the Management Discussion & Analysis (MD&A) section of the annual report, as measured by the number of words. 11 After controlling for firm and industry characteristics, we find that the MD&A is significantly longer for firms with more aggressive tax planning, as measured by both TA_GAAP and TA_CASH. In Table 6, Panel B, we also explore whether the firms that provide additional disclosure in the MD&A are successful in reducing some of the transparency problems that are generated by aggressive tax planning. We construct our test by first developing dummy variables for firms 11 Admittedly, the number of words is a noisy proxy for tax planning disclosure, and we are in the process of developing richer measures to proxy for tax planning disclosures. 20

22 that have above- and below-median MD&A disclosure, and then interacting these dummy variables with the TA_GAAP and TA_CASH variables. We expect that if additional MD&A disclosure mitigates transparency problems related to tax aggressiveness, the relations between tax aggressiveness and the transparency proxies will be weaker for high MD&A disclosure firms than for low MD&A disclosure firms. Consistent with expectations, we find that in five of the six specifications in Panel B of Table 6, the coefficients on TA_GAAP and TA_CASH are less positive for high MD&A disclosure firms than for low MD&A disclosure firms. These differences, however, are only statistically significant in the two Spread regressions, and in TA_GAAP version of the analyst forecast dispersion regression. Thus, these results are somewhat mixed. 6. Robustness Tests A. Loss Firms Other things equal, loss firms typically have a more opaque information environment than profitable firms (Dechow and Dichev 2002). Although we include a dummy variable for firm years with a reported book loss in all of our analyses, we undertake several additional tests to allay concerns that loss firms are exerting undue influence on our measures of tax aggressiveness. In Table 7, we report the sensitivity of our analyst forecast error results to three alternative forms of our tax aggressiveness variables partitioned as a function of estimated taxable income (see definitions in Appendix 2). In columns (1) and (2), we limit our analyses to only the sample of firms that report positive estimated taxable income. In columns (3) and (4), we limit our size-industry benchmark firms to only those firms with estimated positive taxable 21

23 income. Finally, in columns (5) and (6), we only include firms in our analyses that have nonmissing pre-tax foreign income and positive estimated taxable income. This last approach isolates the sample to those firms that have the greatest opportunity to engage in foreign tax planning. In each set of sensitivity tests, we find that analysts forecast errors continue to be increasing in the applicable measure of tax aggressiveness. For brevity, we have only tabulated the robustness of our analysts forecast error results. However, the inferences from the sensitivity analyses in Table 7 also hold for our analyst forecast dispersion, information asymmetry, and accruals quality results reported in Tables 2, 3 and 4. B. Other proxies of tax aggressiveness In Table 8, we investigate whether our tax aggressiveness measures capture information about tax aggressiveness incremental to that contained in the raw ETR. Not surprisingly, because TA_GAAP and TA_CASH are a function of the underlying ETRs, they are correlated significantly with the raw ETRs. And, although other things equal, one expects that firms with lower ETRs are more tax aggressive, one may be curious as to whether our industry-size adjustment to the ETR adds incremental explanatory power for explaining transparency. To investigate whether TA_GAAP and TA_CASH measure incremental tax aggressiveness, we partition the ETRs using an indicator variable equal to one when the firm is in the top half of the distribution of tax aggressiveness, and zero otherwise. The results in Table 8 indicate that, even in the presence of the raw ETR variable, the tax aggressiveness indicator variable is significantly related to the transparency proxies. Notice further that in columns (7) and (8), accruals quality is not related to the raw ETR, but is related to the incremental tax aggressiveness indicator variable. 22

24 As discussed in Section 2.B., there are several alternative measures of tax aggressiveness that have been developed in the accounting literature. Although we conjecture that these measures potentially have some weaknesses, they could still capture some notion of tax aggressiveness. For our robustness tests, we focus on two measures: Frank et al. s (2009) DTAX and Wilson s (2009) SHELTER. We begin by computing Frank et al. s (2009) DTAX measure of discretionary permanent tax differences. Frank et al. (2009) compute DTAX as the residuals from a regression of an estimate of permanent differences on measures of intangible assets, income of unconsolidated subsidiaries, minority interest, state tax burdens, changes in NOLs, and lagged permanent differences. Positive levels of DTAX indicate that our sample firms have discretionary permanent differences that reduce taxable income relative to the population of Compustat firms that are used to estimate DTAX. We do not find any evidence that high levels of discretionary permanent differences (DTAX) are related to firms information environments. Specifically, DTAX is insignificant in each of the four specifications (Columns 2, 4, 6, and 8). Finally, we estimate Wilson s (2009) SHELTER measure which was developed using a sample of approximately 60 firms identified in court documents as having participated in illegal tax shelters. Wilson suggests estimated the probability that a firm is engaging in a tax shelter as follows: SHELTER = *BookTax Differences *Discretionary Accruals *Leverage *Size *ROA *ForeignIncome *R&D. SHELTER is not associated with either analysts forecast errors or dispersion (columns 1 and 3). Contrary to our expectations, firms with a high probability of entering into tax shelters have relatively lower spreads. We also find that SHELTER is significantly positively related to 23

25 firms level of accruals quality. However, this relation could be mechanical because SHELTER as well as the accruals quality measure are a function of discretionary accruals. C. Additional proxies for organizational complexity As a proxy for organizational complexity, our analyses use Bushman et al. s (2004) revenue-based Hirfindahl-Hirschman index calculated as the sum of the squares of each geographic segment s sales as a percentage of the total firm sales (Complexity). In Table 10, we re-estimate our analyst forecast error regressions using three additional measures of complexity. In columns (1) and (2), we measure complexity as an asset-based Hirfindahl-Hirschman index, calculated as the sum of the squares of each geographic segment s assets as a percentage of the total assets. In columns (3) and (4), complexity is estimated using the number of geographic segments the firm operates in. Lastly, in columns (5) and (6), we measure complexity as a revenue-based Hirfindahl-Hirschman index, calculated as the sum of the squares of each business segment s sales as a percentage of the total sales. The inclusion of none of the additional proxies for complexity in our analyses alters our inferences. As in Table 7, for brevity, we have only tabulated the robustness of our analysts forecast error results. However, the inferences from the sensitivity analyses in Table 10 also hold for our analyst forecast dispersion, information asymmetry, and accruals quality results reported in Tables 2, 3 and Conclusion Corporations engage in various forms of tax planning to reduce expected tax liabilities. These expected benefits, however, do not come without costs. Such costs include direct labor and information systems necessary to carry out the tax planning, as well as expected costs of negotiation and penalties stemming from interactions with taxing authorities. In this paper, we 24