Firm Valuation and the Uncertainty of Future Tax Avoidance

Abstract

Using a valuation framework, we show that two dimensions of tax avoidance, uncertainty and the level of expected future tax rates, are jointly related to firm value and need to be expressed as a ratio. We confirm the importance of a composite measure of tax avoidance adjusted for tax uncertainty in our empirical tests based on a sample of U.S. firms. Our findings indicate that shareholders jointly consider the level and uncertainty of future tax avoidance when valuing firms.

Keywords:

• Firm valuation
• Tax avoidance
• Tax uncertainty

JEL codes:

• M41
• G12
• H25

Acknowledgments

We sincerely thank Beatriz Garcá Osma (editor) and two anonymous reviewers, Alissa Brühne, Igor Goncharov, Jochen Hundsdoerfer, Ed Maydew, Maximilian Müller, John Robinson, Leslie Robinson, Martin Ruf, and seminar participants at the WHU – Otto Beisheim School of Management, the University of Tübingen, the X Workshop on Empirical Research in Financial Accounting in A Coruna, Spain, and the 37th Annual Congress of the European Accounting Association for valuable feedback and suggetions.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental Data and Research Materials

Supplemental data for this article can be accessed on the Taylor & Francis website, doi:10.1080/09638180.2019.1642775.

1. Appendix A. Simulation of OLS Estimations.

2. Table A1. Simulation Results.

3. Appendix B. Additional Tables.

4. Table B1. Pearson and Spearman Correlations between Sample Variables.

5. Table B2. Tax Avoidance and Firm Valuation—Without Financial Firms.

6. Table B3. Tax Avoidance and Firm Valuation—Winsorized Variables.

7. Table B4. Tax Avoidance and Firm Valuation—10-Year Values for Controls.

8. Table B5. Determinants of the TPS and Components—Concurrent Values

Notes

1 We use the standard deviation of annual cash ETRs (CETRs) to measure the tax uncertainty underlying a firm's ETR (see also Guenther, Matsunaga, & Williams, 2017). This approach reflects the assumption that the historical level (volatility) of tax rates is a predictor of future tax rates (uncertainty). Dyreng et al. (2008) show that long-run CETRs are potential proxies for future tax avoidance. In this paper, we also document that historical tax uncertainty is a potential proxy for future tax uncertainty.

2 The intuition for this argument is as follows. Although prices are typically assumed to follow a random walk without an upper or lower bound, taxes follow a different process, since they are typically sticky over a long period and then experience a jump in the case of a policy change (e.g., Hassett & Hubbard, 2002; Hassett & Metcalf, 1999). Even if the tax laws remain unchanged, large tax settlements in tax audits can lead to large spikes in tax expenses (e.g., Bauer & Klassen, 2014; Saavedra, 2018).

3 Most of the prior empirical work has typically assumed a linear relation between tax avoidance, measured via ETRs, and the market-to-book ratio (e.g., Desai & Dharmapala, 2009).

4 The term tax uncertainty also refers to tax provisions that will not be disputed by the tax authorities, such as bonus depreciation or the usage of net operating losses. Uncertainty in proxying for future tax rates in firm valuation arises for two reasons. First, it is uncertain whether future tax laws will allow for bonus depreciation. Second, it is uncertain whether a firm will utilize bonus depreciation opportunities. Hence, while bonus depreciation is a ‘safe’ tax reduction strategy from a legal perspective, there is uncertainty regarding whether it will be allowed in the future and whether a firm will use this strategy. Similarly, one can consider the usage of net operating losses as uncertain because loss offset provisions can change (e.g., recently, as part of the Tax Cuts and Jobs Act of 2017) and because firms may or may not end up having losses.

5 Tax uncertainty can affect firm value also indirectly via its influence on a firm's internal decisions. For example, tax uncertainty reduces the level of investments and delays large investments (Jacob et al., 2018). Another potential channel is proposed by McGuire et al. (2013), who argue that the sustainability of a firm's tax strategy reflects its management's expectations of future earnings such that more sustainable tax rates are associated with more persistent pre-tax earnings. While we do not specifically model these indirect effects in our theory, these channels point in the same direction.

6 The only requirement is clean surplus accounting, that is, all non-dividend changes in equity go through the income statement. While a staple framework in the accounting literature (e.g., Barth, Elliott, & Finn, 1999; Dechow, Hutton, & Sloan, 1999; Kothari, 2001), the residual income model has also been applied in the financial literature. For example, Lee, Myers, & Swaminathan (1999), Dong, Hirshleifer, Richardson, & Teoh (2006), and Dong, Hirshleifer, & Teoh (2012) use the residual income model to compute measures of misvaluation. Similarly, Loh & Mian (2006) and So (2013) use the residual income model to judge analyst forecast accuracy and Bris, Koskinen, & Nilsson (2009) apply it to infer a firm's cost of capital. We use an earnings-based valuation model to illustrate the interaction effect between pre-tax earnings and expected tax rates and to be consistent with Modigliani & Miller (1985), who use retained earnings. However, the logic still holds for a discounted cash flow model.

7 This simplification is intended to focus on how to best incorporate tax avoidance and tax uncertainty into a regression framework. We add further controls to the regression design to account for the possibility that the firm-specific long-run tax rate could be correlated with the persistence of shocks to value creation.

8 As a consequence, ${\delta }_t$ and $R{I}_t^p$ are also uncorrelated.

9 Generally, expectations of ratios of two random variables can be undefined and are difficult to compute unless the distributions of the variables are known. Thus, to properly specify ${\mathbb{E}}_t\left[R{I}_{t+i}/\right(1+r{)}^i]$, a full model of a firm's cost of capital would be necessary. We use the approximation to focus on highlighting the link between conditional expectations and uncertainty in past tax rates. An alternative argument for the approximation is that, if the forward-looking beta is constant and an appropriate estimate will be used in the regression, then r is essentially fixed for the purposes of the conditional expectation (Lambert, Leuz, & Verrecchia, 2007; Thomas & Zhang, 2014).

10 As Lambert et al. (2007, p. 389) note, ‘One way to justify the inclusion of additional information variables in a cost of capital model is to note that empirical proxies for beta, which for instance are based on historical data alone, may not capture all information effects. In this case, however, it is incumbent on researchers to specify a “measurement error” model or, at least, provide a careful justification for the inclusion of information variables, and their functional form, in the empirical specification.’

11 In addition, Sikes & Verrecchia (2016) argue that widespread tax avoidance increases the market risk premium. Since this is an effect common to all firms, we do not explicitly account for it but will take care of it using year fixed effects.

12 This is a reduced-form description of the estimation equation. In later stages, we add different proxies for $Growt{h}_{i,t}$ and interactions with either ${\delta }_{i,t}^{\ast }/Vo{l}_{\delta ,i,t}$ or $P{I}_{i,t}$ for a more complete model.

13 The main arguments are that, first, annual CETRs are not very predictive of long-run CETRs. Second, GAAP ETRs include both current and deferred taxes and will thus not reflect certain forms of tax avoidance. In addition, simply averaging a series of single-year ETRs tends to overweight years with unusually large or small ETRs, a problem that is mitigated by using the average of the sums. Our results are robust to using a shorter horizon, for example, a five-year horizon to proxy for the level and uncertainty of tax rates.

14 One cannot simply invert a variable in a regression without altering its functional form. To see why, assume a true form of $y=a_0+a_{1}x+e$ and an estimated regression $y=b_0+b_1\left(1/x\right)+e$. Then $b_1=\mathrm{c}\mathrm{o}\mathrm{v}(1/x,y)/\mathrm{v}\mathrm{a}\mathrm{r}\left(1/x\right)=a_1(\mathrm{c}\mathrm{o}\mathrm{v}(1/x,x)/\mathrm{v}\mathrm{a}\mathrm{r}\left(1/x\right)$. The term $\mathrm{c}\mathrm{o}\mathrm{v}(1/x,x)$ is not trivial to determine. For example, if x can be zero, the expectation and variance of 1/x are unbounded and it is not clear what the true value of $b_1$ should be. Depending on the correlation with other variables in the regression, this issue becomes even more complex.

15 Untabulated results show that the results are qualitatively similar when earnings growth is used. We pick sales growth to be consistent with prior literature.

16 We obtain similar results when using a 10-year window for these control variables as for TPS (see Table B4 in the Online Appendix).

17 In contrast to most other studies, we keep financial firms to maintain a sample close to that of Dyreng et al. (2008). We obtain very similar results when excluding financial firms (see Table A.2).

18 We note that our results are robust to winsorizing all variables, with the exception of CETR, at the 1% and 99% levels (see Table A.3). In this test, CETRs are winsorized at zero and one.

19 Table B1 in the Online Appendix provides the Pearson and Spearman correlations for the variables in this study.

20 In Table B5 om the Online Appendix, we show that our results are similar when using only concurrent values of our control variables, as in prior literature.

21 The results are similar when using a quintile, tercile, or median split.

22 To address concerns that our results are driven by the low number of observations in each portfolio, we sort firms into 25 groups that result from the quintiles of the CETR and of VolCETR. We find a similar positive relation between TPS and the estimated ${\beta }_1$ coefficients.

23 Since some firms can report very small UTB balances, we only use firms with UTBs of at least 1% of the pre-tax income. Our results are not sensitive to the choice of this arguably arbitrary cutoff. We continue to find significant results for UTBs of at least 0.2%, 0.5%, 0.75%, 2%, 3%, or higher relative to pre-tax income.