Harald Amberger Eva Eberhartinger Matthias Kasper. Tax Rate Biases in Tax-Planning Decisions: Experimental Evidence

Transcription

1 WU International Taxation Research Paper Series No Tax Rate Biases in Tax-Planning Decisions: Experimental Evidence Harald Amberger Eva Eberhartinger Matthias Kasper Editors: Eva Eberhartinger, Michael Lang, Rupert Sausgruber and Martin Zagler (Vienna University of Economics and Business), and Erich Kirchler (University of Vienna)

2 Tax-Rate Biases in Tax-Planning Decisions: Experimental Evidence* Harald J. Amberger Vienna University of Economics and Business Department of Finance, Accounting and Statistics Tax Management Group Eva Eberhartinger Vienna University of Economics and Business Department of Finance, Accounting and Statistics Tax Management Group Matthias Kasper University of Vienna, Tulane University Faculty of Psychology, Department of Economics Keywords: heuristics, decision bias, salience, tax-rate information, rational inattention JEL-Classification: C91, D03, H32, K34, M21, M41 * The authors would like to thank Kathleen Andries, Sebastian Beer, Tobias Bornemann, Erich Kirchler, Saskia Kohlhase, Pete Lisowsky, Ben Lockwood, Michaele Morrow (discussant), Martina Rechbauer, Bill Rees, Timothy Rupert (discussant), Rupert Sausgruber, Alfons Weichenrieder, three anonymous reviewers for the 22 nd Journal of the American Taxation Association (JATA) Conference, participants at the 22 nd Journal of the American Taxation Association (JATA) Conference, the 108 th NTA Conference on Taxation in Boston, the 2016 Accounting and Finance Association of Australia and New Zealand (AFAANZ) Conference in Gold Coast, the 5 th Workshop on Current Research in Taxation in Prague, the 5 th Doctoral Meeting of the Oxford University Centre for Business Taxation in Oxford, the 2015 IAREP-SABE Joint Conference in Sibiu, the 1 st Doctoral Colloquium of the British Accounting and Finance Association SAG (BAFA) in Edinburgh, the 2 nd Doctoral Research Seminar in Vienna, the 2018 VHB Spring Meeting in Jena, and seminar participants at WU Vienna, UNSW Sydney, UTS Sydney, and ANU Canberra for valuable comments and suggestions. Amberger and Kasper gratefully acknowledge financial support from the Austrian Science Fund (FWF): W 1235-G16. Kasper gratefully acknowledges financial support from the Austrian National Bank (OeNB): Corresponding Author: Harald J. Amberger, Vienna University of Economics and Business, Vienna, Austria. He can be reached at harald.amberger@wu.ac.at.

3 Tax-Rate Biases in Tax-Planning Decisions: Experimental Evidence Abstract Recent empirical findings suggest that firms might not always engage in economically optimal tax planning. We conduct a series of four laboratory experiments with 223 students and 62 tax professionals to examine whether decision biases offer a behavioral explanation for tax outcomes. We find that individuals overestimate the importance of tax rates relative to the tax base when facing time pressure in tax-planning decisions. This systematic decision bias results in suboptimal choices. In line with the theory of rational inattention, we observe that increasing tax-burden differences between two tax-planning strategies weakly mitigate the decision bias. However, tax-planning behavior is unaffected by the level of experience: students and highly experienced tax professionals are similarly prone to biased decisionmaking. Overall, our findings suggest that overestimating the implications of salient tax-rate information can cause decision biases and contribute to heterogeneity in tax outcomes.

4 1. Introduction This paper investigates whether systematic decision biases could lead to economically suboptimal tax-planning and thus contribute to heterogeneity in tax outcomes. Standard economic theory assumes that agents behave economically optimally and incorporate all relevant information in their decisions. Thus, when firms decide on whether to implement a tax-planning strategy, decision makers balance the associated costs and benefits to maximize shareholder wealth (Scholes, Wolfson, Erickson, Hanlon, Maydew, & Shevlin, 2014). More specifically, they consider tax-rate effects and tax-base effects to make an economically optimal tax-planning decision (Huizinga, Laeven, & Nicodeme, 2008; Hines & Rice, 1994). In reality, however, tax outcomes differ considerably across firms. For example, Dyreng, Hanlon, & Maydew (2008) show that a surprisingly large share of firms persistently pays taxes in excess of the statutory U.S. corporate tax rate. Although a growing body of literature investigates this phenomenon (e.g. Gallemore, Maydew, & Thornock, 2014; Desai & Dharmapala, 2006), the causes of heterogeneity in firms tax outcomes are not yet entirely clear (Hanlon & Heitzman, 2010). As suboptimal tax planning impairs a firm s competitiveness, investment and growth (Donohoe, Lisowsky, & Mayberry, 2015), a more profound understanding of factors that drive tax-planning decisions is required. Prior research building on the work of Simon (1955) and Tversky & Kahneman (1981, 1974) shows that decisions, if based on simplifying decision strategies (i.e. heuristics), can result in economically suboptimal outcomes. Decision makers use heuristics when tasks are complex and resources limited (Kahneman, 2003; Tversky & Kahneman, 1974), which is typically the case in corporate tax-planning decisions. One heuristic is to focus on the most salient aspect of the decision environment where the importance of this aspect is overestimated in subsequent choices (Taylor & Thompson, 1982)

5 For tax-related decisions, prior studies found that focusing on salient information reduces information-processing costs (Powers, Schmidt, Seidman, & Stomberg, 2017), but negatively affects decision quality (Graham, Hanlon, Shevlin, & Shroff, 2017). These studies suggest that differences in the salience of average and marginal tax rates determine their relevance in corporate decisions. However, the financial consequences of a tax-planning decision depend on tax-rate effects and tax-base effects. Both effects are important for firms as changing tax rates and tax bases are the main channels for tax policy reform (Simons, 1938; Haig, 1921) and lowering tax rates while broadening the tax base has become a popular policy tool (Alm, 2018; OECD, 2010). There is initial indication that taxpayers might underestimate the financial implications of tax bases when choosing between stylized tax systems (Blaufus, Bob, Hundsdoerfer, Kiesewetter, & Weimann, 2013). However, Blaufus et al., (2013) find that learning effects mitigate decision bias. This result suggests that corporate decisions should be largely unbiased, which is inconsistent with the findings of prior empirical work (Powers et al., 2017; Graham et al., 2017). In addition, the causes of behavioral responses to differences in the salience of tax-rate effects and tax-base effects remain largely unknown. We carry out a series of four experiments to examine whether and under which conditions differences in the salience of tax-rate effects and tax-base effects guide decisionmaking and cause economically suboptimal tax-planning decisions. An experimental approach allows us to isolate causal effects and to determine the importance of specific sources of information in tax-planning decisions that are otherwise not observable (Alm & Jacobson, 2007). Specifically, we design an intra-group, cross-border financing scenario and vary statutory tax rates and tax bases. On this basis, we ask participants to choose the taxminimizing tax-planning strategy (i.e. equity or debt capital for a foreign subsidiary) for a total of 16 decisions. We model tax-rate effects via differences in statutory tax rates (i.e. the - 2 -

6 tax-rate differential) between the financed subsidiary and the financing parent company, and tax-base effects via the subsidiary s tax loss carry-forward. We expect, first, that the economic effects of tax rates are easier to identify and thus more salient than the effects of the tax-base. Consequently, decision makers should weigh tax-rate information more strongly in their decision-making. Second, we assume that time constraints aggravate the use of heuristics (Kocher, Pahlke, & Trautmann, 2013) and increase the share of economically suboptimal choices. Third, and in line with the theory of rational inattention (Abeler & Jaeger, 2015; Gabaix, 2014; Mackowiak & Wiederholt, 2009; Sims, 2003), we expect that an increasing tax-burden difference between the two tax-planning strategies reduces decision bias. In our first experiment, which includes 185 students, we find robust evidence that decision-making is less optimal than standard economic theory predicts. Specifically, under time constraints, decision makers systematically overestimate tax-rate effects and underestimate the economic consequences of tax bases. This pattern of decision-making suggests that decision makers use salient tax-rate information as a heuristic in tax-planning decisions, which leads to suboptimal tax-planning choices. In line with the theory of rational inattention, we find that an increasing tax-burden difference between tax-planning strategies weakly mitigates decision biases. Next, we re-run our initial experiment with 62 highly experienced tax professionals to alleviate concerns that professionals may behave differently from students and that a lack of experience drives our results. Prior studies suggest that student decision-making does not systematically deviate from that of professionals (Depositario, Nayga, Wu, & Laude, 2009; Liyanarachchi, 2007; Remus, 1996; Ashton & Kramer, 1980), if, as in our case, the difficulty of an experimental task is low (Elliot, Hodge, Kennedy, & Pronk, 2007). Indeed, our findings show that decision-making is unaffected by the level of professional experience. In fact, students and tax professionals are equally prone to decision biases in tax-planning decisions

7 These results corroborate the external validity of our initial experiment and suggest that professional experience, or a lack thereof, does not explain the use of salient tax-rate information as a heuristic. Finally, we conduct two additional experiments to address possible alternative explanations for our findings. First, we invert the tax-minimizing strategy for each taxplanning decision and show that unobserved preferences for equity or debt financing do not explain our results. Second, we change the order of tax parameters and provide information on the tax base before showing tax rates. Consistent with our main results, we continue to find that decision makers overestimate tax-rate effects while underestimating the economic consequences of tax bases. In exploring the relevance of tax-rate and tax-base changes in tax planning decisions, our study makes several contributions. Most significantly, we expand research on the processing and relevance of tax-related information in tax-planning decisions. Prior studies investigate how different types of tax rates affect corporate decisions (Graham et al., 2017; Powers et al., 2017) but less is known about the trade-off between tax-rate effects and taxbase effects. Decision biases might result from situational factors, such as time constraints, or depend on the financial consequences of a tax-planning decision. Prior work also does not disentangle competing explanations for the greater salience of tax rates such as order effects or information complexity. Finally, Blaufus et al., (2013) suggest that individuals with no educational or professional background in taxation might underestimate the financial implications of tax-base changes relative to tax-rate changes 1 and that learning effects mitigate decision bias. It thus remains unclear to what extent, if at all, individuals with academic training and professional experience in taxation show a similar behavior. Our paper addresses all these unresolved and critical issues. Our findings suggest that decision makers weigh information on tax rates more strongly than information on tax bases. 1 The study uses a non-gender-balanced sample of non-academic university staff

8 This leads to systematic decision biases. Further, we show that situational factors, such as time constraints, have the potential to systematically affect tax outcomes. We also contribute to research on the theory of rational inattention (Gabaix, 2014; Mackowiak & Wiederholt, 2009; Sims, 2003). While Abeler & Jaeger (2015) find no evidence for rational inattention in a tax-complexity experiment, our findings suggest that the propensity to make a biased tax-planning decision is affected by the tax effects of a taxplanning strategy. More specifically, our results indicate that the tendency to make biased taxplanning decisions is higher in situations with seemingly minor economic consequences. We also expand prior work which suggests that learning effects mitigate decision bias and that individuals overestimate tax-rate effects particularly when they are described prior to tax-base effects (Blaufus et al., 2013). In contrast to these findings, our results indicate that decision biases are driven by the complexity of tax information rather than order effects, and that a lack of experience does not explain biased decision-making. This suggests that individuals decision biases could lead to economically suboptimal tax-planning and thus contribute to heterogeneity in firms tax outcomes. Finally, our findings inform the ongoing discussion of the adequacy of student samples in laboratory experiments (Alm, Bloomquist, & McKee, 2015; Alm & Jacobson, 2007; Bloomquist, 2009). We find that individuals with substantial professional experience exhibit a similar systematic decision bias as students. This suggests that experienced students could be valid surrogates in experiments that investigate tax-planning decisions. Our study might encourage future experimental research in areas where archival data are unavailable. The remainder of this paper proceeds as follows: In Section 2, we outline a model for tax planning through intra-group financing, discuss research on decision biases in tax planning, and develop our hypotheses. Section 3 presents our empirical analysis, including (i) survey results on time pressure in corporate tax planning, (ii) the design of our experiment, (iii) results for the student sample, (iv) results for the professional sample, and (v) - 5 -

9 supplementary analyses. Section 4 discusses the implications of our findings and potential limitations. 2. Theoretical Background Tax Planning through Intra-Group Financing Tax planning is economically optimal if it increases shareholder wealth. A common tax-planning strategy is income shifting through intra-group debt (Heckemeyer & Overesch, 2017; Ramb & Weichenrieder, 2005). Internal capital markets enable firms to exploit tax-rate differentials between countries (Desai, Foley, & Hines, 2004) and to shift income via interest payments to low-tax jurisdictions (Buettner & Wamser, 2013). Thus, low-taxed group members of multinationals finance their high-taxed counterparts with intra-group debt (Overesch & Wamser, 2014; Huizinga et al., 2008). Aside from statutory tax rates that determine tax-rate differentials, several tax-system features affect the tax base of a firm and thus the choice of an optimal tax-planning strategy. An example of a tax-base effect is the opportunity to carry tax losses forward in time. This may reduce a firm s periodic tax base and turn a high-taxed entity into a low-taxed one. As a result, income shifting via interest payments on intra-group debt and a tax loss carry-forward may conflict with each other because the financed group member does not benefit from the interest deduction if it does not pay any taxes due to a tax loss carry-forward. At the same time, interest payments are taxed at the level of the financing group member. We outline this trade-off between statutory tax rates and a tax loss carry-forward to derive tax-planning preferences of a multinational for intra-group equity or debt assuming economically optimal decision-making. 2 Consider parent company A and its foreign 2 In our experiment, a tax-minimizing decision is equivalent to an economically optimal outcome as the taxplanning strategies do not differ in non-tax costs or benefits. Thus, we use both terms interchangeably but acknowledge that tax-minimizing behavior in reality does not necessarily imply economically optimal results. For example, firms differ in their financial reporting costs of tax planning (e.g. Frank, Lynch, & Rego, 2009; Badertscher, Philips, Pinco, & Rego, 2009), regulatory costs of tax planning (e.g. Mills, Nutter, & Schwab, 2013), or reputational costs of tax planning (e.g. Gallemore et al., 2014) which implies different levels of economically optimal tax planning across firms. To isolate the causal effect of tax-rate effects and tax-base effects on decision-making, we disregard these aspects in our experimental setting

10 subsidiary B. The firms earn π $ and π %, which is taxable income of A and B before deducting interest payments on intra-group debt and a tax loss carry-forward. The tax base after deducting interest payments and a tax loss carry-forward is taxable at statutory tax rates τ $ and τ %, respectively. To limit our analysis to the trade-off outlined above, we assume that τ ' > τ ), which implies a positive tax-rate differential between the subsidiary and the parent company. A may finance B via intra-group equity or debt. A is risk neutral and has no preference for either form of financing. If A chooses equity financing, dividends are neither tax deductible for B nor taxable at A. 3 If B is financed via debt, interest payments I are tax deductible at B and taxed at A. In a one-period setting, A selects the tax-planning strategy yielding the lowest group tax burden. First, the group tax burden for equity financing T, in a given period amounts to T, = π ) τ ) + (π ' L ' ) τ ', (1) where L ' denotes the amount of tax loss carry-forward from prior years available for B, which is equal to or less than π '. In order to focus on one single tax-base effect, we assume that any unused tax loss carry-forward will be forfeited in later periods. Second, the group tax burden for debt financing T 5 in a given period amounts to T 5 = (π ) + I) τ ) + (π ' I θl ' ) τ ', (2) where I indicates interest payments on intra-group debt, which are tax deductible at B and taxable at A. θ denotes the fraction of L ', which can be offset against taxable income after deducting I (θ = 7 89: ; 8 ). We assume that the tax loss carry-forward exceeds the income 3 This approach resembles a territorial tax system for foreign income earned by parent company A, which is the prevailing standard for most jurisdictions

11 of B after the deduction of I and that country B does not offer a tax refund in case of a loss; i.e. the deductible fraction of L ' is limited to the tax base. 4 A selects the form of intra-group financing which yields the lowest group tax burden, thus min {T,, T 5 }. (3) It follows that the group tax burden is a function of both the statutory tax rates and the tax loss carry-forward. More formally, A prefers intra-group debt over equity if π ) τ ) + (π ' L ' ) τ ' > (π ) + I) τ ) + (π ' I θl ' ) τ '. (4) Substituting τ ' for τ ) + d, where d denotes the tax-rate differential (d = τ ' τ ) ), and changing the order of preferences, yields a tax-planning preference for intra-group debt if (π ' L ' )(τ ) + d) I τ ) > 0. (5) The definition of θ implies that I > π ' L ' holds. Thus, given a positive tax-rate differential and a tax loss carry-forward that exceeds the income of B after deducting I, Equation (5) suggests that tax-planning preferences for either equity or debt depend on the size of (i) the positive tax-rate differential and (ii) the tax loss carry-forward. 5 All else being equal, the tax advantage of debt increases in the positive tax-rate differential, while equity becomes more preferable with a larger tax loss carry-forward. To make an economically optimal tax-planning decision, decision makers consider tax-rate effects and tax-base effects and balance the gain from income shifting via intra-group debt with the tax effects of foregoing the tax loss carry-forward. If decision makers are incentivized to make economically optimal tax-planning decisions, they should resolve this trade-off and minimize the group tax burden. 4 We also assume that π ) 0; π ' L ' > (π ' I); and π ' > I > 0. We disregard agency costs, information asymmetries, and tax-planning costs. 5 Equation (5) does not hold if d 0. In this case, intra-group equity is the tax-minimizing strategy. If θ = 1 (i.e. the income of B after deducting I exceeds its tax loss carry-forward), Equation (5) collapses to Id > 0. Thus, debt (equity) financing is preferable for a positive (negative) tax-rate differential while A is indifferent between both forms of financing if d = 0. As we model the trade-off between the positive tax-rate differential and the tax loss carry-forward in our experiment (see Section 3), we do not discuss these conditions in detail

12 Decision Biases in Tax Planning Empirical evidence suggests that firms differ in tax-planning behavior and many firms appear not to minimize their tax burden (Dyreng et al., 2008). While legal or economic factors partly explain heterogeneity in tax outcomes, certain behavioral patterns could trigger suboptimal decisions in tax contexts (Graham et al., 2017). 6 This argument is consistent with research in behavioral economics and economic psychology, which finds systematic biases in economic decision-making where decision outcomes are less optimal than theoretically assumed (for reviews, see Kirchler, 2007 and Ranyard, 2017). Building on early work by Simon (1955), Tversky & Kahneman (1974) identify cognitive principles that guide decision-making in complex situations. As humans have limited cognitive capacity, and as it is often difficult to assess the exact economic consequences of a choice, individuals seek simplifying decision strategies known as heuristics. These decision strategies affect the way information is processed (Gigerenzer & Gaissmaier, 2011). Heuristics alleviate the complexity of a task and reduce its cognitive load (Kahneman, 2003; Tversky & Kahneman, 1974). In contrast to extensive decision strategies, heuristics could imply decision biases and thus do not necessarily yield economically optimal choices. The literature discusses several types of heuristics that facilitate complex decisions. For example, individuals focus their attention on relatively salient, i.e. visible, aspects of a complicated decision environment and overestimate the importance thereof in subsequent choices (Taylor & Thompson, 1982; Bordalo, Gennaioli, & Shleifer, 2012). Chetty, Looney, & Kroft (2009), for example, show that the visibility rather than the magnitude of taxes drives decision-making, suggesting that tax-information salience could cause optimization errors. Similarly, Abeler & Jaeger (2015) find that individuals underreact to less salient changes in the tax environment. Blaufus et al., (2013) study the choice between hypothetical tax systems 6 A meta-study by Feld, Heckemeyer, & Overesch (2013), for instance, finds that a 10 percentage point change in statutory tax rates induces only a 3 percentage point reaction in the debt ratio of a firm

13 and find that individuals overestimate tax-rate effects. This effect is stronger when the tax rate is mentioned prior to the tax base and weaker after participants learn about the optimal choice. Overestimating the importance of salient information, however, might expose firms to decision biases. 7 Graham et al., (2017) empirically show that corporate decisions could be economically suboptimal, as executives overestimate the importance of easily accessible and thus more salient information (e.g. the average tax rate) relative to less accessible but more accurate information (e.g. the marginal tax rate) in capital structuring and M&A decisions. Similarly, Powers et al., (2017) find that investors use the salient U.S. statutory tax rate as a heuristic to reduce information-processing costs. Moreover, Dyreng, Hanlon, & Maydew (2010) show that individual decision-making affects corporate behavior as executives shape their firms tax planning decisions. Taken together, these results suggest that biased decision-making at the individual level could contribute to suboptimal tax-planning and heterogeneity in firms tax outcomes. Hypothesis Development In line with the literature on decision biases in tax planning (Graham et al., 2017; Powers et al., 2017), we expect that decision makers overestimate the importance of salient information. The economic consequences of tax rates are highly salient, as statutory tax rates are easily available and decision makers are well aware of their economic effects. Tax-base effects, on the other hand, are more complex and their economic consequences are less intuitive. As the implications of tax bases are more difficult to identify, they appear less salient to the decision maker. 8 As individuals weigh salient information more strongly, we expect decision makers to use salient tax-rate information as a heuristic when evaluating the 7 Some scholars argue that corporate decision-making is economically optimal. In this respect, DellaVigna (2009) posits that firms have measures to increase the optimality of their decisions. These include external consultants and feedback from the capital market. Others acknowledge that corporate decision-making could be biased if no feedback is provided to the decision maker (Camerer & Melmendier, 2007). Given the complexity of tax planning, the specificity of information required to make tax-planning decisions, and the opportunity to obscure relevant information in financial statements (Balakrishnan, Blouin, & Guay, 2018), the likelihood of identifying economically suboptimal tax-planning decisions is relatively low. 8 Equation (5), for instance, suggests that it requires fewer operations to determine the economic consequences of a tax-rate change on the group tax burden than to quantify the effect of a change in the tax loss carry-forward

14 economic consequences of a tax-planning decision. However, a simplified decision strategy based on the salience of tax-rate effects potentially induces a decision bias if tax-base effects do not receive adequate attention. Based on these arguments, we hypothesize: HYPOTHESIS 1. Tax-planning decisions are biased towards the tax-rate effect. Extensive decision-making strategies require time and cognitive effort, and often lead to accurate choices. Corporate decisions, however, are complex and resources for decision-making, such as time and information, are often limited. As a consequence, decision makers experience time pressure in many economic decisions (Kocher et al., 2013). When resources are limited, decision makers rely more strongly on heuristics in order to accelerate the decision-making process (Kahneman, 2011, 2003). Prior research in psychology documents that scarcity of resources, such as time constraints, stimulates the use of heuristics and thus affects decision-making (Hockey, 1997). More specifically, decision quality increases under moderate time constraints and decreases as time constraints become more onerous (Dror, Busemeyer, & Basola, 1999; Payne, Bettman, & Johnson, 1988; Rothstein, 1986). Several studies in auditing provide support for the negative effect of time constraints on decision quality (e.g. Low & Tan, 2011; Braun, 2000; Choo, 1995; Ponemon, 1992). These findings also apply to tax-planning decisions as auditors and tax consultants operate in similar decision environments. For example, both professions work with financial information and face similar information-processing costs due to billing their clients by the hour. As time constraints stimulate the use of heuristics, we assume that: HYPOTHESIS 2. Time constraints increase decision bias in tax-planning decisions. Next, we hypothesize that decision biases depend on the economic consequences of a tax-planning strategy, i.e. the resulting tax burden. In certain settings, economic consequences are large and therefore obvious. Consequently, decision makers should be less likely to overestimate the importance of salient information in such settings. This assumption is in line with the theory of rational inattention, which suggests that decision makers consider less

15 salient information if ignoring the information is costly (Abeler & Jaeger, 2015; Gabaix, 2014; Mackowiak & Wiederholt, 2009; Sims, 2003). For our setting, the theory of rational inattention predicts that the probability of making optimal decisions depends on the taxburden difference between the two tax-planning strategies (i.e. equity or debt). If, for instance, equity yields a significantly lower tax burden than debt, we expect a higher likelihood of choosing equity. However, if the tax effects of equity or debt differ only marginally, we expect fewer optimal decisions. In other words, the more obvious the benefits of a taxplanning strategy, the less likely decision biases occur. Based on these arguments, we posit: HYPOTHESIS 3. An increase in the tax-burden difference between tax-planning strategies mitigates decision bias in tax-planning decisions. Figure 1 summarizes our hypotheses and illustrates the theoretical framework of our experiment. INSERT FIGURE 1 HERE 3. Empirical Analysis Survey on Time Pressure in Corporate Tax-Planning Decisions We survey tax professionals to gain a better understanding of the relevance of time constraints in corporate tax-planning decisions. Using an online survey tool, we ask participants questions related to time pressure in their work environment and the amount of time they would require to make an economically optimal tax-planning decision absent any budgetary restrictions or time constraints. 9 We distributed the survey via to tax professionals that are members of the International Fiscal Association (IFA) in Austria. 10 In total, we sent the survey to 200 tax professionals. 97 individuals opened the link to the survey and 74 completed the questionnaire. Among respondents, 71.6 percent are male and the average age is 49.9 years (SD = 10.26). Respondents have, on average, 22.9 years of tax- 9 A copy of the survey is available on request. Respondents granted approval to publish anonymized results. 10 IFA is a not-for-profit association that supports research on international taxation. IFA has more than 12,000 members worldwide, including practitioners, researchers, and policy makers (see

16 related professional experience (SD = 10.17). 77 percent of the respondents are a partner in a tax-consulting firm or hold a senior tax position in the industry (i.e. Head of Tax/Accounting or CFO). Thus, our sample comprises tax professionals with substantial experience and the authority to make tax-planning decisions. Results suggest that time constraints are a critical issue in corporate tax-planning decisions. On a scale from 1 (I do not agree) to 7 (I fully agree) respondents indicate that they are often pressed for time (µ = 5.65, SD = 1.21), that a rapid work pace is generally required in their job (µ = 5.95, SD = 0.95), and that they usually have too little time to optimally prepare for tax-planning decisions (µ = 4.86, SD = 0.90). 11 Participants state that they would require much more time than generally available to prepare a tax-planning decision (on average 1.68 times the available time, SD = 0.46). Taken together, these results suggest that highly experienced tax professionals face severe time constraints in their tax-planning decisions. Experimental Setup This section describes the setup of our four experiments that aim to identify behavioral responses to tax-rate effects and tax-base effects. We systematically vary tax rates and tax loss carry-forwards and assess individuals tax-planning choices for a set of 16 decisions. We model time constraints by limiting the decision time in one experimental treatment (time-pressure treatment). Participants do not face time constraints in the other treatment (no-time-pressure treatment). Taken together, this setup yields a factorial design with between and within subject variation. The experiments are programmed in Z-Tree (Fischbacher, 2007) and comprise three stages: (i) a questionnaire, (ii) instructions for the taxplanning game, and (iii) a set of 16 tax-planning decisions T-tests suggest that means are significantly larger than the midpoint (i.e. 4) of the scale (all p < 0.01, twotailed). 12 A copy of the experimental instrument is available on request. Our participants granted approval to publish anonymized results of our study

17 To determine the decision time for our time-pressure treatment, we run pre-tests with 18 junior faculty members and track the decision time. On average, pre-test participants require 90 seconds to make one tax-planning decision. In the time-pressure treatment, we restrict the time to 45 seconds per decision so that participants have 50 percent of the average decision time per task. The resulting time pressure is consistent with our survey responses, which revealed that, on average, tax professionals have only 60 percent (= 1/1.68) of the time necessary to optimally prepare for tax-planning decisions. As survey respondents exhibit several decades of experience, the indicated time pressure likely constitutes a lower bound. Thus, the time constraints we introduce in our time-pressure treatment provide a realistic approximation of time pressure in real-world tax-planning decisions. We follow Low and Tan (2011) and inform participants about the time limitation before the first tax-planning task. We do not force the participants to make a decision within the given time limit to avoid upward bias. Once the decision time expires, participants have no opportunity to provide an answer and automatically proceed to the next decision. Experimental Procedure Questionnaire We initiate our study with a questionnaire the individual s background (e.g. age, professional/work experience, education). Further, we ask eight questions to identify attitudes towards debt financing. The questionnaire concludes with a test that measures reflexivity in decision-making (cognitive reflection test; see Frederick, 2005). 13 Instructions Next, we introduce participants to a tax-planning game, which entails 16 tax-planning decisions. The scenario follows the model laid out in Section 2 and presents a multinational that operates in two countries. The parent company is located in low-tax country A; the subsidiary operates in high-tax country B. Participants take the position of a CFO and decide on intra-group financing provided by the parent company to the subsidiary. Financing 13 We include this test to explore the role of reflectiveness in decision-making. As we run our study in German, we use the translated version of the cognitive reflection test by Piazolo (2007)

18 arrangements can take the form of intra-group equity or debt. We instruct participants to choose the form of financing that minimizes the total group tax burden in a one-period setting. After providing background on taxes and financing decisions, we outline a simple tax system that includes (i) statutory tax rates τ ), τ ' and (ii) the subsidiary s tax loss carryforward L '. As we are interested in the trade-off between a positive tax-rate differential and a tax loss carry-forward determined under Equation (5), τ ) is strictly smaller than τ '. 14 The subsidiary in country B reports a tax loss carry-forward L ', which can be offset against its taxable income. Any unused L ' will be forfeited in later periods. Dividends are tax-exempt in country A, while debt financing triggers interest payments that are tax deductible at the subsidiary and taxable at the parent company. Next, we outline a simple economic environment in which the parent company does not report any income beyond interest or dividend payments received from the subsidiary. The subsidiary earns a constant profit of ECU 4,000, and distributes any after-tax profit. The financed amount (ECU 50,000,000) and interest payments (ECU 3,000,000) do not vary. We apply three safeguards to ensure that participants understand their task. First, we present a table summarizing all information and depicting how to compute the group tax burden, conditional on the form of intra-group financing. On this basis, we present two model calculations which comprise all relevant factors, i.e. statutory tax rates, the tax loss carryforward, the financed amount, interest payments, the taxable profit of the subsidiary, and the group tax burden resulting from equity or debt financing. While one model calculation yields equity as the tax-minimizing strategy, the other implies debt financing. Second, we explain the tax effects of both strategies in written form. The third and final safeguard involves three questions (check questions) which cover (i) the effects of the tax-rate differential, (ii) the effects of the tax loss carry-forward and (iii) the taxation of dividends and interest payments. 14 In Experiment III, we introduce a lower statutory tax rate at the subsidiary (i.e. a negative tax-rate differential) and a tax loss carry-forward at the parent company. These changes do not affect our results. 15 ECU ( Electronic Currency Unit ) is a fictitious currency we use in the experimental testing

19 To ensure that only individuals who fully understand the task are included in the primary analysis, we exclude those who do not correctly answer the check questions. 16 Tax-Planning Game The main part of our experiment consists of a tax-planning game with 16 decision tasks (hereinafter: items). Each individual faces these 16 items in random order. At any time, participants may refer back to the general information and the model calculations described in the previous section. We provide a calculator, paper, and a pencil. During the experiment, participants do not receive feedback. Only at the end are they informed as to whether they made tax-minimizing decisions. Based on the trade-off determined under Equation (5), we vary the tax-minimizing solution by systematically changing (i) statutory tax rates and (ii) the subsidiary s tax loss carry-forward. 17 We keep the remaining parameters constant, which enables us to determine the decision relevance of (i) the tax-rate effect and (ii) the tax-base effect. 18 The 16 items are divided into four item groups with four items each. In the Baseline item group, debt financing is the tax-minimizing strategy, as the positive tax-rate differential dominates the low tax loss carry-forward. Relative to Baseline, we vary the tax base (ChangeTLCF) and the tax rates (ChangeTD), respectively, so that equity financing becomes the tax-minimizing strategy for these two item groups. For a fourth item group (Symmetry), we simultaneously alter the tax base and the tax rates so that debt financing remains the tax-minimizing strategy. Table 1, Panel A summarizes the tax parameters for the 16 items and Panel B for the four item groups. In the Baseline item group (items 1-4), the tax-rate differential is constant, while the subsidiary s tax loss carry-forward increases from item 1 to item 4. This reduces the tax advantage of debt across these items. In the ChangeTLCF item group (items 5-8), the taxrate differential is identical to Baseline. As the subsidiary s tax loss carry-forward increases 16 The check questions are provided in the Appendix. In robustness tests, we include individuals who failed the check questions and find similar results (see Table A1 of the Online Appendix). 17 We are aware that, in contrast to statutory tax rates, time effects alter the value of a tax loss carry-forward. While tax-rate differentials require long-term tax planning, a tax loss carry-forward is usually limited in time. We restrict the tax-planning game to a one-period setting to rule out that time effects affect decision-making. 18 Figure A1 in the Online Appendix provides a schematic overview of our experimental design

20 from items 5 to 8, the tax advantage of equity financing grows accordingly. In the ChangeTD item group (items 13-16), we lower the tax-rate differential relative to Baseline while the tax loss carry-forward decreases from item 13 to item 16. This leads to an increase in the tax advantage of equity capital. Finally, in the Symmetry item group (items 9-12), we set a constant but higher tax-rate differential than in Baseline while the tax loss carry-forward, and thus the tax advantage of debt financing, decreases from item 9 to item 12. We set the tax-rate differentials and the subsidiary s tax loss carry-forwards such that the tax-burden difference between equity and debt financing varies within each item group while yielding an identical pattern across item groups. In addition, the tax-burden difference varies such that one strategy is clearly preferable for some items but less so for others. Items 1, 8, 9, and 16, for example, result in a tax advantage of ECU 200,000 (Delta200000) for the tax-minimizing strategy, while we reduce the difference to roughly ECU 1,280 (Delta1280) for items 4, 5, 12, and Items 3, 6, 11, and 14 yield a tax-burden difference of ECU 25,000 (Delta25000) and items 2, 7, 10, and 15 of ECU 60,000 (Delta60000), respectively. To identify a decision bias, i.e. unequal responses to tax-rate effects and tax-base effects, we compare decision-making in the Baseline item group with ChangeTLCF and ChangeTD items. We expect individuals to choose debt financing in the Baseline item group, which is the tax-minimizing strategy for these items. In contrast, the tax-rate differential suggests debt financing for ChangeTLCF and ChangeTD items even though equity yields the tax-minimizing strategy. Thus, if individuals overestimate the importance of tax-rate information, we expect a smaller proportion of tax-minimizing decisions in the ChangeTLCF and the ChangeTD item group than in Baseline Due to rounding difficulties in obtaining readable statutory tax rates, tax-burden differences implied by equity or debt financing do not completely match for items 4, 5, 12, and 13 (i.e. Delta1280). The tax-burden differences match for the remaining item groups (Delta25000, Delta60000, and Delta200000). 20 Figure A1 suggests to also compare decisions in the ChangeTLCF item group to Symmetry and Baseline items. With this analysis, we explore the same decision bias but from a different angle. In line with H1, we expect individuals to choose debt in the ChangeTLCF item group, although equity is the tax-minimizing decision for these items. For Baseline and Symmetry items, however, the tax-rate differential suggests debt, which is the tax

21 We do not expect differences in decision-making when comparing Symmetry with Baseline items, as the tax-rate differential indicates debt financing, which is the taxminimizing strategy for these items. The main reason for including Symmetry items is that eight items yield debt (Baseline and Symmetry) while another eight items yield equity financing as the tax-minimizing strategy (ChangeTLCF and ChangeTD). 21 INSERT TABLE 1 HERE We complete our experiment with a short questionnaire and ask individuals to assess the importance of the tax parameters for their decisions on a scale from 1 (not important) to 9 (very important). We collect this information for the statutory tax rates of the subsidiary and the parent company, the subsidiary s tax loss carry-forward, the tax deductibility of interest payments, and the tax exemption of dividend payments. Dependent Variable and Remuneration We ask participants to choose the tax-minimizing form of intra-group financing. Thus, they face 16 binary choices between equity and debt. In our analysis, we use the variable Tax-Minimizing, which takes the value of one if an individual made an economically optimal decision. 22 We provide incentives for participation (Croson, 2005) and pay participants based on the performance in the tax-planning game. To achieve equivalent incentives, we vary payouts across samples. Participants in the first experiment (student sample) receive a showup fee of 6 and we reward tax-minimizing decisions with 0.50 each. In the second experiment (professional sample), we double the payout so that the show-up fee equals 12 minimizing solution. Thus, we expect a larger share of tax-minimizing decisions in the Symmetry and Baseline item group than for ChangeTLCF items. Results in Figure 2 are consistent with this expectation. 21 We do not include further combinations of the tax-rate differential and the tax carry-forward in the experiment as we do not expect any differences in decision-making (e.g. decision-making for items with a low tax loss carry-forward and a tax-rate differential of should be similar to Baseline and Symmetry items). Thus, the chosen design enables us to test our hypotheses while keeping the number of items as low as possible. 22 We provide a definition of variables in the Appendix

22 and each tax-minimizing decision yields We only remunerate economically optimal decisions. Thus, the maximum payout is 14 per person in the student sample and 28 in the professional sample. We do not allow participants to leave the laboratory until the last participant has completed the experiment. Experiment I: Student Sample Sample and Participants We conduct our first experiment with students of a public business school in Austria. We invite all students to participate via , and personally recruit graduate students in accounting, taxation, and/or finance who regularly have initial professional experience in these fields. We conduct the experiment in the computer laboratory of the university and test 185 individuals in 11 sessions. Observations of 44 individuals are excluded from the primary analysis, as 40 did not pass the check questions and another 4 did not provide decisions for any of the 16 items. We do not require a full set of 16 decisions per individual. 24 The final sample consists of 141 individuals and 2,024 decisions percent of the individuals in the final sample are male (Male) and the average age is 25.1 years (Age). 27 percent of the individuals study in a graduate program in accounting, taxation, and/or finance (Education) and 46.8 percent have work experience in these fields (WorkExp). On average, individuals solve 1.5 out of the 3 questions of the cognitive reflection test (CRT). 23 The payout structure for the professional sample is based on job vacancies posted on online career platforms in Austria. The average offer for tax consultants with several years of experience is 17 per hour. As this amount is based on collective bargaining agreements and actual salaries are higher, we offer a competitive compensation. 24 As a robustness test, we limit our sample to individuals who provided decisions on all 16 items. Results do not differ qualitatively from our primary analysis (see Table A1 of the Online Appendix). 25 Missing observations concern the time-pressure treatment, as we do not force individuals to make a decision. If an individual does not make a decision within 45 seconds, we regard the decision as unsolved and present the next item. By not taking a decision, participants forego the opportunity to receive remuneration for that item. We do not observe a systematic pattern of missing observations across items (c² = 0.93, df = 15, p > 0.99). Thus, missing observations are randomly distributed

23 We assigned 60 percent of the participants to the time-pressure treatment. 26 This treatment contains more individuals with education in accounting, taxation, and/or finance (z = -1.88, p = 0.06, two-tailed). Other than that, we do not observe significant differences between treatments (all p > 0.34, two-tailed). A session without time pressure lasts one hour and a time-pressure session 45 minutes. Accordingly, the average payout in the time-pressure treatment ( 10.76) is lower than the payout in the no-time-pressure treatment ( 12.82). Univariate Results We first show univariate results for the tax-planning game. Therefore, we calculate the mean proportion of tax-minimizing decisions per individual (Tax-Minimizing) for the full sample, the time-pressure treatment, and the no-time-pressure treatment percent of the individuals select the tax-minimizing strategy for all items. The mean proportion of taxminimizing decisions amounts to 68.8 percent per individual, which is significantly lower than the optimum of 100 percent tax-minimizing decisions (z = -9.98, p < 0.01, two-tailed) and significantly higher than the average of 50 percent tax-minimizing decisions that we would expect had individuals guessed throughout (z = 7.70, p < 0.01, two-tailed). 27 In the notime-pressure treatment, 40 percent of the individuals consistently make tax-minimizing decisions whereas only 5 percent do so under time pressure. This leads to a smaller mean of Tax-Minimizing in the time-pressure treatment than in the no-time-pressure treatment (z = - 26 We do not assign individuals to specific sessions but invite them to participate in a session that best fits their schedule. Thus, the number of individuals in each experimental session depends on (i) the number of registered individuals and (ii) the actual turnout of individuals. To ensure that the anticipated duration of the experiment is similar in each session, we test one treatment per session. Overall, we run five no-time-pressure sessions and six time-pressure sessions, having cancelled one scheduled no-time-pressure session due to a low number of registered individuals. This explains why the number of individuals differs between treatments. 27 We use non-parametric Wilcoxon-Mann-Whitney-Tests to test for differences between subsamples and Wilcoxon signed rank tests to test whether Tax-Minimizing differs from a specific value (e.g. the optimum of 100 percent). Shapiro-Wilk tests support this approach as the normality assumption for Tax-Minimizing is violated in several subsamples and non-parametric tests are less sensitive to the underlying distribution. Inferences are similar when using parametric tests (e.g. t-tests). This is plausible as t-tests that adjust for unequal variances are robust to a violation of the normality assumption if, as in our case, subsamples are relatively large (Scheffé 1959)