IPOs and Corporate Tax Planning

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) IPOs and Corporate Tax Planning Christine L. Dobridge, Rebecca Lester, and Andrew Whitten 2021-058 Please cite this paper as: Dobridge,ChristineL.,RebeccaLester,andAndrewWhitten(2022). “IPOsandCorporate Tax Planning,” Finance and Economics Discussion Series 2021-058r1. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2021.058r1. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

IPOs and Corporate Tax Planning † ‡ § Christine L. Dobridge, Rebecca Lester, and Andrew Whitten November 2022 Abstract Doesgoingpublicaffecttheamountandtypeofcorporatetaxplanning? Usingapanel of U.S. corporate tax return data from 1994 to 2018, we show that IPO completion is associated with the implementation of multinational income shifting strategies central to the current international tax policy debate. Specifically, firms (i) expand their foreign tax haven presence, (ii) enter into cross-border agreements that accompany intangible asset transfers to foreign subsidiaries, and (iii) increase their level of foreign related-party payments around the time that they go public. The effects are strongest among firms that switch to more sophisticated tax advisors in the years preceding the IPO. In contrast, we observe little domestic tax planning because large stock option deductions, which increase as a consequence of the IPO, provide large domestic tax shields. ThepapercontributestothenascentliteraturestudyingIPOsbydocumenting thetypesandtimingofspecifictaxstrategiesthatenablepublicfirmstoremainlightly taxed in the post-IPO period. Furthermore, the findings imply that U.S. tax policies targetedatearly-stageinnovativefirmsarecriticalforretainingdomesticallydeveloped IP – and the income earned on such assets – for the U.S. tax base. Keywords: Initial public offering (IPO), tax planning, multinational tax JEL Codes: D12, E21, H24 ∗WearegratefulforcommentsfromEricAllen,JenniferBlouin,TravisChow,AlexEdwards(discussant), NathanGoldman,MichelleHanlon,ShaneHeitzman,MichelleHutchens(discussant),PaulLandefeld,Saskia Kohlhase (discussant), Pete Lisowsky (discussant), Michael Love, Carmen Payne Mann, Neviana Petkova, Josh Rauh, Max Risch, David Samuel, Bridget Stomberg (discussant), Juan Carlos Suarez Serrato, IESEG Reading group, conference participants at the NBER Spring 2021 Public Economics Meeting, NBER 2022 Conference on Business Taxation, 2022 OTA Conference, 2021 UNC Tax Symposium, 2021 ATA Midyear Meeting, 2022 Norwegian Center on Taxation conference, 2019 National Tax Association Annual Meeting, and seminar participants at Carnegie Mellon University, Hong Kong University, SMU, the University of Washington,ErasmusAccountingWorkshop,andtheFederalReserveBoardofGovernors. WethankPreston Harry, Rebecca John, Sasha Ruby, and Shooka Saket for research assistance. †Dobridge: Board of Governors of the Federal Reserve System. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the Federal Reserve research staff or the Board of Governors. ‡Lester: Stanford Graduate School of Business. §Whitten: U.S. Department of the Treasury, Office of Tax Analysis. This research was conducted while Andrew was an employee of the U.S. Department of the Treasury. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors and do not necessarily reflect the views or the official positions of the U.S. Department of the Treasury. Any taxpayer data used in this research was kept in a secured Treasury or IRS data repository, and all results have been reviewed to ensure that no confidential information is disclosed.

1 Introduction Multinational firms’ low tax burdens have attracted substantial academic, policy, and press attention (e.g., Dyreng et al. 2017; Grinberg and Kysar, 2021; Drucker, 2021; Rubin and Walker, 2022). Current domestic and multilateral policy proposals address multinational tax planningbyincreasingtaxratesanddiscouragingcross-borderincomeshifting(OECD,2021; Biden, 2022). However, critical questions remain unanswered about multinational firms and their tax planning. For example, what drives companies to initially implement tax planning strategies? At what point do companies invest in foreign tax structures? This paper finds that IPOs motivate firms to implement the global tax planning strategies central to the current international tax policy debate. Understanding the relation between IPOs and tax planning is important for at least two reasons.1 First,considerableacademicworkfocusesoncorporatetaxplanningbylargepublic corporations (Hanlon and Heitzman, 2010; Wilde and Wilson, 2018). However, there is less evidence about the tax strategies of smaller, innovative private firms such as those that file to go public (Badertscher, Katz and Rego, 2013; Edwards, Hutchens and Rego, 2019; Edwards and Hutchens, 2020; Lisowsky and Minnis, 2020). Some evidence shows that a substantial number of private corporations are insensitive to tax considerations due to large, reported tax losses (Cooper and Knittel, 2006, 2010), whereas sophisticated public multinationals arbitrage tax rates and incentives across jurisdictions to minimize worldwide tax burdens (Blouin and Krull, 2018; De Simone and Olbert, 2021; Hanlon, Lester and Verdi, 2015). Documenting whether and to what extent the IPO is the point at which firms alter their approach to tax planning helps to bridge between these findings. Second, current tax policy 1We use the terms “tax planning” and “tax avoidance” interchangeably throughout the paper, where we definebothbroadlyasthereductionofexplicittaxes(HanlonandHeitzman,2010;WildeandWilson,2018). 1

proposals target tax avoidance by established multinational companies.2 Understanding the genesis of these tax strategies is informative for crafting the appropriate policy mechanisms to preserve the U.S. tax base. Prior literature offers competing predictions for the effect of an IPO on firm tax planning, with some work suggesting increased levels of tax avoidance. First, the transition to public markets increases capital market scrutiny of corporate tax obligations. Badertscher et al. (2019) and Hoopes et al. (2020) show that private firms engage in certain types of “conforming” tax avoidance given that financial reporting and tax incentives are aligned in private firms. However, upon a transition to capital markets, these incentives diverge, affecting managers reporting and tax choices (Graham et al., 2014). These findings imply that IPO firms will increase or change their tax planning strategies to align with that of other public firms around the time that the firm goes public. Second, IPOs provide discrete opportunities to implement unique tax structures, such as “supercharging” tax benefits (Edwards, Hutchens and Rego, 2019). The cost of certain foreign strategies is based on the valuation of firm assets used in such structures; thus, firms have incentives to establish these structures in anticipation of the IPO, when a lower or wider range of valuations is plausible. The IPO valuation implications, as well as the increased capital market scrutiny, are reasons why firms may increase their tax planning around the IPO. On the other hand, for many firms, an IPO may have little effect on tax planning. We show that the majority of private firms report substantial domestic tax losses in years prior to the IPO. These firms may simply be insensitive to corporate taxes or unwilling to invest in corporate tax planning. In this case, we anticipate little to no change in firms’ tax avoidance. Thus, the empirical relation between IPOs and corporate tax planning is unclear ex ante. 2For example, both the U.S. Corporate Minimum Tax and the OECD’s Pillar 1 provisions are estimated to impact approximately 80 of the largest publicly-traded companies (EconPol, 2021; Rubin and Francis, 2022). 2

Despite the importance of this topic and the centrality of IPOs in prior research (Ritter and Welch, 2002; Lowry, Michaely and Volkova, 2017), there is limited work on the relation between corporate taxes and IPO completion. The limited evidence is attributable to two factors. The first is an econometric challenge, as the decision to complete an IPO is endogenous. Indeed, a common critique of the IPO literature is that it is difficult to disentangle the effects of capital infusion from the managerial decision to go public, due to the lack of an appropriate control sample against which to measure effects. The second factor is a data challenge: there are few data sources on private firms, impeding researchers’ ability to observe changes in tax planning between the pre- and post-IPO periods. To address the econometric challenges, we compare tax planning by firms that complete an IPO (the treatment group) to three different control groups. We first follow Bernstein (2015) and construct a control sample of firms that filed for an IPO but ultimately withdrew andremainedprivate. Thesefirmsareagoodcontrolgroupbecausetheyarethemostsimilar in their life cycle and their incentives to go public. However, because these firms ultimately chose to forego the IPO and thus could differ in ways correlated with tax planning, we construct two additional control samples of private and public firms. To align these samples withIPOfirms, werequirethepublicandprivatecompaniestobeofasimilarsizeandtoalso issue equity of a comparable magnitude. However, unlike the IPO firms, these companies do not change their private or public status during the sample period. The use of multiple control samples permits us to evaluate the robustness of any observed effects and estimate a range of magnitudes across groups. Furthermore, use of the private and public control groups permits us to distinguish the IPO event from other external financing events, thereby isolating the distinct impact of going public. For all comparisons between IPO and control firms, we employ inverse probability weighting on observable firm characteristics to further mitigate endogeneity concerns. 3

We address the data availability challenge by using a panel of confidential U.S. corporate taxreturnsofpublicandprivatefirms. Ourdataspan1994through2018, withafocusonthe seven years surrounding each firm’s IPO filing. These data permit us to construct measures ofbothdomesticandforeignfirmtaxavoidanceandtoobservefirmsforseveralyearspriorto the IPO.3 Furthermore, the data permit construction of the three control samples, improving upon the approach in recent literature that generally relies on only withdrawn firms as a benchmark. Our first empirical analyses test whether domestic tax planning changes after the IPO. We examine book-tax differences (BTDs), which capture the extent to which financial reporting and taxable incomes diverge (Mills, 1998; Desai and Dharmapala, 2006; Chen et al., 2010). We also examine domestic cash effective tax rates (ETRs), which capture the amount of cash taxes paid per dollar of financial reporting income. BTDs (ETRs) are increasing (decreasing) in tax avoidance. We employ a stacked cohort difference-in-difference design in which we compare IPO-completing firms with each of the three control samples. We observe large increases in BTDs and decreases in ETRs in the years immediately following IPO completion relative to all three groups. Although these effects are consistent with increased tax planning in the post-IPO period, additional tests reveal that the results are not due to explicit or intentional tax planning strategies. Instead, we find that the effects are largely driven by increased stock option deductions; the stock option deductions are sufficiently large to minimize domestic tax obligations for several post-IPO years. Additional evidence suggests that some effects are also attributable to increased earnings management activity, which increases financial reporting income but is largely reversed on firms’ tax returns. Documenting these true factors driving the domestic tax planning effects is important for the 3Prior literature generally studies worldwide tax planning using consolidated tax measures from publicly available financial statements. Use of the tax data allows us to separately examine domestic and foreign tax planning, enabling a more precise analysis of the tax strategies employed by firms. 4

growing IPO literature that draws inferences based on similar measures constructed from public data (e.g., Chyz et al., 2022; Chen et al., 2021). We next study international tax planning, which is of particular importance given the considerable academic and policy focus on firms reducing their tax obligations through shiftingbothincomeandphysicalpresencetolower-taxedjurisdictions. Weexaminethreeforeign tax planning outcomes. We consider legal strategies for firm tax planning, including a commonlystudiedstrategyoflocatingsomecorporateactivitiesinjurisdictionswithloweffective tax rates—so-called corporate “tax havens” (Hines and Rice, 1994). We show that, relative to IPO withdrawers, IPO completers are 57 percent more likely to own a subsidiary in a tax haven – and own a greater number of such subsidiaries – after going public. When separately comparing to public firms, we observe strong effects consistent with firms responding to capital market pressures for tax avoidance: IPO firms’ likelihood of having a tax haven presence converges to that of the publicly traded companies by the third year after the IPO. Next,wetestwhetherIPOfirms’foreignstructuresfacilitatecross-borderincomeshifting. We measure the incidence of a cost-sharing agreement, which is a specific income shifting strategy that enables firms to source sales revenue to lower-taxed foreign jurisdictions after moving U.S.-developed IP offshore (De Simone and Sansing, 2019). Because the U.S. charges an exit tax on the fair value of the assets at the time of the offshore transfer, firms have incentives to strategically time this transfer prior to the IPO. We find sizeable IPO effects: the incidence of these agreements within IPO-completing firms more than doubles in the post-IPO period. Third, we use foreign subsidiary tax data on the amount, type, payor, and payee of related-party transactions to study income shifting payments. We observe that foreign subsidiaries of IPO firms report large increases in the level of receipts from both the U.S. parent and other related domestic entities (“outbound” payments from the U.S.). Because higher 5

levels of cross-border payments could be attributable to post-IPO growth within a vertically integrated multinational firm, we conduct several additional tests to link these findings to explicit foreign tax planning. For example, we also measure changes in the amount of payments to the U.S. (“inbound” payments). Although we observe that these payments increased after the IPO as well, the economic magnitude is smaller than the outbound payments, mitigating concerns that all cross-border payments increased regardless of tax incentives. Additionally, we refine the measures of outbound related-party receipts and payments to include amounts most commonly associated with income shifting: revenues from inter-company sales of intangibles, management service fees, royalty payments and license fees on intangible assets, and interest on related party loans (Grubert, 1998, 2003; Faulkender, Hankins and Petersen, 2019). We find that the total outbound payments primarily relate to these income-shifting transactions, not to routine intercompany sales of goods and services. Finally, we expect that the foreign planning effects should be the greatest in high-intangibles firms because income attributable to intangible assets can be more easily shifted across borders (De Simone, Mills and Stomberg, 2019). Results are consistent with (weakly) larger effects in firms with greater R&D expenses pre-IPO. To assess how firms accomplish this tax planning, we examine the use of professional tax advisors. This analysis is motivated by recent work documenting that tax advisors play an important role in firms’ tax planning decisions (e.g., Zwick, 2021; Bustos et al., 2022). We find that the proportion of firms using a “sophisticated” tax advisor (defined as those signing more than 4,000 tax returns) increases substantially for IPO completers from three years prior to the IPO through the year of the IPO. We again observe convergence with the publicly-tradedfirmcontrolsample: IPOfirms’useofsophisticatedadvisorsincreaseestothe level of the public firms by year t-1, consistent with IPO firms responding to capital market pressures for tax planning. We then link this professionalization with the foreign planning 6

effects, showing that these effects (i.e., having a tax haven subsidiary or participating in a cost-sharing agreement) are greatest among firms that switch to using a sophisticated tax advisor in the years leading up to IPO. Additional tests show that the results are not otherwise explained by pre-IPO changes in management or by venture capital ownership. This paper contributes to the tax literature in finance, economics, and accounting studying the drivers of firm tax behavior. A key innovation of our paper is examining the transition from private to public status, thereby bridging between the extensive literature on public firms’ tax avoidance (e.g., Dyreng et al., 2017) and tax considerations of private firms (Hoopes et al., 2020; Badertscher, Katz and Rego, 2013; Badertscher et al., 2019; Olbert and Severin, 2022; Lisowsky and Minnis, 2020). Specifically, we provide new evidence about the origins of foreign tax structures, which are inherently difficult to observe without access to the subsidiary tax data we use in this study. We show that the IPO motivates firms to substantially increase their foreign haven presence and cross-border transactions, thereby also contributing to the work studying multinationals’ foreign tax planning activities (e.g., Blouin and Krull, 2009; Graham et al., 2010; De Simone and Olbert, 2022). Second, we contribute to the literature studying the consequences of public listing (Feldman et al., 2021; Michaely and Roberts, 2012; Gao, Harford and Li, 2013; Brav, 2009; Dobridge, Gilje and Whitten, 2022). We contribute in two ways. First, we broaden the identification strategy used in Bernstein (2015), which leverages withdrawn firms as a control sample, by also using equity-issuing private and public firms as control samples. Second, we contribute to the nascent literature examining the tax implications of IPOs. This literature studies how corporate taxes affect IPO pricing (Edwards and Hutchens, 2020), industry tax planning (Chyz et al., 2022), transaction structures (Edwards, Hutchens and Rego, 2019), and compensation planning (Dambra, Gustafson and Quinn, 2020). Not only do we examine a range of domestic and foreign tax outcomes for our sample firms, but we help explain the 7

mechanism driving such effects by evaluating the use of professional advisors. Finally, this paper provides policy-relevant insights. We show that sophisticated foreign planning begins relatively early in public firms’ lifecycles. This finding means that policies targeting profit shifting will be most effective if they not only address large, well-established firms, but also young, growing, high-R&D firms. For example, incentives to retain intangible assets in the U.S. may be particularly effective in stemming the tide of foreign asset transfers that facilitate large amounts of income shifting. We look forward to future research that evaluates the effectiveness of recent incentives, such as the foreign-derived intangible income (FDII) provision in the 2017 Tax Cuts and Jobs Act. 2 Empirical strategy 2.1 Research Design To test the relation between IPO completion and firm tax planning, we use a stacked-cohort difference-in-differences design in which we compare tax planning for firms that complete an IPO to three sets of control firms. For each IPO year, we construct a cohort of IPOcompleting firms and control firms. We include all observations in a seven-year window centered around the IPO, requiring firms to be observable for at least two years pre- and post-IPO. We stack the cohorts in event time and estimate the following specification to measure the average effect of IPO completion on tax planning: 𝑌 = 𝛽 ·𝐼𝑃𝑂𝐶𝑜𝑚𝑝𝑙𝑒𝑡𝑒𝑑 ·𝑃𝑜𝑠𝑡 +𝛾 +𝛿 +𝜖 , (1) 𝑖𝑗 𝑖 𝑐𝑗 𝑐𝑗 𝑖 𝑖𝑗 𝑌 is one of several tax planning measures for firm 𝑖 in tax year 𝑗, described further in 𝑖𝑗 Section 2.2. 𝐼𝑃𝑂𝐶𝑜𝑚𝑝𝑙𝑒𝑡𝑒𝑑 is an indicator variable equal to one for firms that complete 𝑖 an IPO transaction, and 𝑃𝑜𝑠𝑡 is an indicator variable equal to one if 𝑗 ≥ 𝑡, where 𝑡 is 𝑐𝑗 8

the year of IPO filing or completion for cohort 𝑐 (depending on the control sample used, as discussed below). We include year-by-cohort fixed effects 𝛾 , which control for cohort- 𝑐𝑗 specific time series trends. We also include firm fixed effects 𝛿 to control for time-invariant 𝑖 firm characteristics. Each IPO firm is included only once in the sample, and therefore, firm fixed effects are equivalent to firm-by-cohort fixed effects. The firm and year-by-cohort fixed effects subsume the main effects of 𝐼𝑃𝑂𝐶𝑜𝑚𝑝𝑙𝑒𝑡𝑒𝑑 and 𝑃𝑜𝑠𝑡 . The term 𝜖 is an additive 𝑖 𝑐𝑗 𝑖𝑗 error term. Our coefficient of interest, 𝛽, measures the difference in tax planning between IPO firms and control firms after an IPO, relative to any difference in the pre-period. We exclude time-varying firm controls from Equation (1) because these variables are likely also affected by the IPO and thus could introduce bias due to a “bad controls” problem (Angrist and Pischke, 2008; Roberts and Whited, 2013). We cluster standard errors by three-digit NAICS industry, reflecting that industry-specific productivity shocks contribute to IPO waves.4 To further study year-specific effects of IPO completion and also evaluate the parallel trends assumption, we estimate the following equation: 𝑡+3 ∑︁ 𝑌 = 𝛼+ 𝛽 ·𝐼{𝑘 = 𝑗}·𝐼𝑃𝑂𝐶𝑜𝑚𝑝𝑙𝑒𝑡𝑒𝑑 +𝛾′ +𝛿′ +𝜖′ , (2) 𝑖𝑗 𝑘 𝑖 𝑐𝑗 𝑖 𝑖𝑗 𝑘=𝑡−3,𝑘̸=𝑡−1 where 𝑡 remains the year of IPO completion for cohort 𝑐, 𝐼{𝑘 = 𝑗} is an indicator equal to one if 𝑘 = 𝑗, 𝛾′ are year-by-cohort fixed effects, 𝛿′ are firm fixed effects, and 𝜖′ is an additive 𝑐𝑗 𝑖 𝑖𝑗 error term. We omit 𝛽 , and thus all effects captured by 𝛽 are measured relative to the 𝑡−1 𝑘 difference between the IPO firms and control firms in the year prior to IPO completion. 4Section4.4presentstwoanalysesforalternativelyestimatingEquation(1)includingtime-varyingcontrol variables and clustering at the firm level. Inferences are unchanged. 9

2.2 Tax Planning Measures We use several measures of tax planning. We first construct two measures of domestic tax planning from the prior literature. BTD is the difference between pre-tax financial statement income (“book” income) and taxable income, divided by beginning-of-year assets (Mills (1998); Desai and Dharmapala (2006); Chen et al. (2010)).5 We also use the firm’s domestic effective tax rate, Cash ETR, which is equal to domestic cash taxes paid divided by domestic book income. Although this measure requires firms to report positive pre-tax income and thus cannot be calculated for the large number of loss firms in our sample, we use this measure because it is commonly used in prior work (see Hanlon and Heitzman (2010)). Prior literature shows that greater foreign presence is associated with greater tax planning (Rego (2003); Foley et al. (2007); Hanlon, Lester and Verdi (2015)). Therefore, we next construct several measures of foreign tax planning using 2004-2018 data on financial performance and intercompany transactions of foreign subsidiaries.6 We construct three measures of tax haven presence. Havens (0/1) is an indicator variable equal to one if a firm owns at least one subsidiary in a foreign tax haven, and zero otherwise. #Havens is the count of tax haven subsidiaries; #DotHavens is a refined count capturing only those subsidiaries in “small as a dot” havens where firms locate primarily for tax reasons. We follow Hines 5For firms required to file IRS Form Schedule M-3 with their tax return, we define book income as the sum of Net income (loss) per income statement of includible corporations, U.S. current income tax expense, and U.S. deferred income tax expense. However, Schedule M-3 was not introduced until 2004 and is only required for firms with at least $10 million in assets. Thus, for tax returns that do not include Schedule M-3, we define book income as the sum of Net income (loss) per books and federal income tax per books from Schedule M-1. Taxable income is measured before tax loss carryforwards and other special deductions. AppendixAprovidesthespecificU.S.corporateincometaxreturnlineitemsusedtoconstructallvariables. 6We use data from Form 5471, Information Return of U.S. Persons with Respect to Certain Foreign Corporations, to construct these measures. We identify a foreign subsidiary’s jurisdiction using address information listed on Form 5471. Because these data are only available for even years beginning in 2004, we limit the analysis to the post-2004 period and impute the presence of subsidiaries in the odd years based on whetherthesubsidiaryreportedpositivebeginning-of-yearcashorotherassetsonitsbalancesheet(Schedule F) in the subsequent even year. For example, we assume a subsidiary was in existence in 2005 if the 2006 beginning balance sheet for that subsidiary had non-zero values. For firms that do not report beginning-ofyear values, we impute the count of tax haven subsidiaries in a given odd year by taking the minimum of the counts in the prior and succeeding years. 10

and Rice (1994) and Dyreng and Lindsey (2009) to determine a county’s haven status; see Appendix A. We also use foreign subsidiary data to measure the firm’s participation in tax-motivated income shifting structures. We first identify whether a foreign subsidiary was party to a cost sharing agreement, a specific income shifting strategy in which a foreign subsidiary shares in intellectual property (IP) development costs (De Simone and Sansing (2019)). Sharing in these costs entitles a foreign subsidiary to a portion of the subsequent IP-related revenue. Thus, by establishing a cost sharing agreement with a low-taxed foreign subsidiary, the U.S. parent effectively shifts income out of the U.S. to a low-taxed jurisdiction. Cost sharing (0/1) is an indicator equal to one if at least one foreign subsidiary participates in such an agreement during the year.7,8 Next, we construct measures of intercompany payments between the U.S. and its foreign subsidiaries. Total amounts received from U.S. and Total amounts paid to U.S. are equal to the natural logarithm of one plus the total amounts received from, or paid to, the U.S. parent and other domestic corporations, respectively (De Simone, Mills and Stomberg (2019)).9 Because a vertically integrated firm could naturally have higher levels of intercompany payments, we also construct two refined measures that include only those line items most closely related to tax-motivated income shifting: intercompany sales of property rights 7The specific question we use to identify this strategy is (based on the 2012 Form 5471 Schedule G), “During the year, was the foreign corporation a participant in any cost sharing arrangement?” 8A 2011 Bloomberg article provides one public example of a cost sharing arrangement used by Cisco. It states,“Beginningin1995,theNetherlandssubsidiarybeganpayingforpartofCisco’songoingresearchinthe U.S. under a cost sharing agreement, according to company records. Like other U.S. technology companies, Cisco qualifies for U.S. tax benefits by doing most of its research and development domestically. By paying for some of it, the overseas unit can remove a chunk of any subsequent profits from the U.S. and claim them offshore” (Drucker, 2011). Though Cisco went public in 1990 (prior to our sample period), this example is illustrative of the type of internal tax-motivated agreements that have previously been understudied in the prior literature. 9These measures enable us to not only study whether total related-party payments increase around the IPO, but if the amounts paid to foreign subsidiaries exhibit a different effect relative to the amounts paid to the U.S. parent. Observing different effects would be suggestive of tax-motivated income shifting out of higher tax countries (like the U.S.) to lower-taxed foreign jurisdictions. 11

(such as patents and trademarks), platform contribution transaction payments, cost sharing transaction payments, managerial and other service-related compensation, royalties and license fees, interest income (expense), and insurance premiums (Grubert, 1998, 2003; Faulkender, Hankins and Petersen, 2019). Total income-shifting receipts from U.S. is equal to the natural logarithm of one plus the total amounts received from the U.S. parent and other related domestic entities for these items. An additional measure also captures related-party payments among foreign subsidiaries. 3 Data and Summary Statistics 3.1 Sample construction: IPO filers We construct the sample of IPO filers using data from Thompson Financial, the SEC’s EDGAR platform, and administrative U.S. corporate income tax records. IPO filers include IPO-completing firms, as well as firms that file but ultimately withdraw their IPO (the first control group). Table 1 shows the sample selection steps. We draw the dataset of U.S. IPO filers from Thompson Financial’s SDC Platinum New Issues database. We retain IPOs between 1996 (the first year of requisite EDGAR IPO filing) and 2015 (22,484 events). Following standards in the IPO literature (e.g., Lowry, Michaely and Volkova, 2017), we excludefinancialfirmsandutilities,aswellasspin-offs,realestateinvestmenttrusts(REITs), American depository receipts (ADRs), closed-end funds, blank-check firms, and unit funds. We keep only common stock issuances, as well as IPO filings with requisite tax identifiers (tax employee identification numbers or EINs) and filing date information.10 For the six percent of withdrawing firms with multiple filings, we discard all but the first IPO filing. The final IPO sample includes 3,593 unique IPO events, of which 2,411 (1,182) are completed 10EINsarenecessaryformergingwiththetaxdata. BecausetheseareunavailableformanyfirmsinSDC, we scrape EINs from the IPO filing on EDGAR when available. 12

(withdrawn). We use EINs to merge the IPO sample with a stratified random sample of U.S. corporate tax return data, retaining 83 percent of the IPO firms: 2,125 IPO-completing firms (21,985 firm-years) and 841 IPO-withdrawing firms (6,550 firm-years).11 We retain observations for this merged sample from 1994 to 2018 so as to include pre- and post-period observations around the 1996-2015 IPO events. We discard observations that do not have C-corporation filing status, those with zero or negative values of gross receipts or total assets, and those missing industry information.12 We drop the 1996 IPO cohort due to the lack of withdrawn firms after imposing the prior sample restrictions. To mitigate concerns about sample attrition within each cohort, we require all firms to be present in the sample for at least years 𝑡−2 through 𝑡+1 relative to the IPO filing.13 We drop observations outside of the 𝑡−3 to 𝑡+3 window to isolate the tax planning effects immediately surrounding the IPO. The final sample includes 5,683 observations for 900 firms, of which 75 percent complete the IPO. 3.2 Sample construction: Public & private control firms TomitigateconcernsthatIPO-completingand-withdrawingfirmsdifferonunobservedtimevarying characteristics that drive tax planning, we use two additional control samples. These samples include similarly-sized private and public firms that also receive external financing but, unlike IPO firms, do not switch their private or public status. These samples permit us to assess the robustness of any observed IPO effect, and to difference out tax planning effects attributable to capital infusions. We follow similar procedures to construct these two control groups, also outlined in Table 11This stratified random sample is constructed, cleaned, and edited each year by the Statistics of Income (SOI) division of the Internal Revenue Service; see also Zwick and Mahon (2017) and Feldman et al. (2021). 12If industry information is missing or invalid, we instead identify industry affiliation using the firm’s earliest-reported valid NAICS code in the tax data. Thus, we only drop firms for which we never observe any industry information. 13Section 4.4 presents results relaxing this restriction. Inferences are unchanged. 13

1. The starting samples include all unique parent-level C corporations in the tax data from 2004 to 2018 (157,943 private firms and 10,771 public firms).14 As with the IPO sample, we exclude financial and utilities firms and require firms to have positive gross receipts and assets. To further align the samples, we retain only those firms with equity issuances of comparable size to the equity issuances reported by the sample IPO firms.15 We drop a small number of firms with a change in private-public status; the count is omitted from Table 1 due to tax disclosure restrictions. Imposing additional restrictions related to industry affiliation and cohort presence results in a sample of 2,905 private control firms (18,588 firm-years) and 1,691 public control firms (11,014 firm-years). Figure 1a shows the number of completed and withdrawn IPOs by year of filing. Figures 1b and 1c show the number of completed IPOs relative to private control firms and public control firms, respectively, by year of equity issuance. The number of completed and withdrawn IPOs peaks in 2000, prior to the 2001 recession. The sample includes approximately 300 private firms each year from 2006 to 2010; this number dips in 2011 to 2013 before increasing to over 350 firms in 2015. In contrast, the public firm sample includes the most observations in 2006 and 2007; after fluctuations around the Financial Crisis, the count stabilizes at 100-150 firms in the later years. Figures 2a, 2b, and 2c present equity issuance trends for the three samples. Equity Issuance is defined following Yagan (2015) as the dollar value of non-negative changes in equity from the prior to the current year. Figures 2d, 2e, and 2f present trends for Equity Financing, which is defined as Equity Issuance scaled by the sum of total equity and debt. 14We start in 2004 because we require information from U.S. tax form Schedule M-3 (introduced in 2004) to identify whether a firm is public or private. We report results using the withdrawn IPO firm sample over this shorter time period for comparison purposes; see Section 4. 15Specifically, we require that firms report an equity issuance at or above the 10th percentile of the level change in equity in our IPO sample, as well as the 10th percentile of the percentage change in equity in our IPO sample. If a firm has more than one equity issuance of this size, we use the first equity issuance. As part of this step, we also drop firms with assets larger than the largest IPO-completing firm observation to ensure size comparability. 14

Panels a and d show that most IPOs are completed in the same year of filing (𝑡), with some additional equity issuance in 𝑡+1. For the private and public control groups (Panels b, c, e, and f), we show trends relative to the year of equity issuance. The private firm control group exhibits similar patterns as the IPO firms for both the level and proportion of equity issued, whereas the absolute (relative) amount of Equity Issuance (Equity Financing) is larger (smaller) for public firms than for IPO firms. 3.3 Sample weighting To further mitigate concerns about differences between the IPO and control samples, we inverse probability weight (IPW) each sample. Within each cohort, we first predict the likelihood of IPO completion in the year preceding the IPO filing/equity issuance. The covariates included in the probit model are firm age (ln(Age)), size (ln(Total assets)), firm growth opportunities (Asset growth), financial performance (Taxable income/Sales), asset tangibility (Net PPE/Assets), and leverage (Debt/Assets); all variables are defined in Appendix A. We then weight the IPO and control firms using the inverses of the predicted likelihoods derived from the first-stage estimation. To eliminate the influence of outliers, we winsorize variables within each of the three samples at the five-percent level following prior work using corporate tax data (Cohn, Mills and Towery, 2014; Yagan, 2015; Zwick and Mahon, 2017).16 3.4 Summary statistics Table 2 displays summary statistics for the three samples. Average BTD is negative in each sample, meaning that taxable income exceeds book income. Average domestic Cash ETR 16Section 4.4 shows that inferences are unchanged if we exclude weights in estimation. Section 4.4 also presents results after winsorizing at the 1% level. We generally find larger effects, consistent with large outlier values affecting point estimates derived from Eq. (1). 15

is approximately 20.0 percent, with median values of 17.2 percent, 11.8 percent, and 16.2 percent for the withdrawn IPO, private firm, and public firm samples, respectively. Tax haven utilization is, unsurprisingly, the largest in the public firm control sample; firms have 3.1 tax haven subsidiaries and report a 42.0 percent likelihood of tax haven use. Forcomparison, thewithdrawnIPOfirmandprivatesamplesreport0.79and0.77taxhavens (29.3 and 27.1 percent incidence of haven ownership), respectively. Approximately 1.2 to 4.0 percent of the sample participate in a cost-sharing agreement. The bottom portion of the table shows that firms increase their external Equity Financing by 8.9 to 14.7 percent. The sample including withdrawn firms has the smallest firms based on Sales and ln(Total Assets), but it also reports the highest Asset Growth. Table 3 compares IPO firms to control firms in the year prior to IPO. Panel A presents a comparison of IPO completers and withdrawn firms. Panels B and C compare IPO firms to the private and public firm control samples, respectively; the number of IPO firms is smaller due to estimation on the shorter 2004-2018 sample period. IPO completers and control firms are statistically similar across almost all firm characteristics after weighting. The private control firms are 2.5 years older than the IPO completing firms on average; while this difference is statistically different, the logged transformation of age is not. The next section further demonstrates the similarity of IPO firms and control samples based on pre-period tax planning trends. 4 Results 4.1 Domestic Tax Planning Table 4 presents results from estimating Eq. (1) using domestic tax planning measures. Columns (1) through (4) of Panel A include results using BTD as the dependent variable; 16

Columns (5) through (8) report results using Cash ETR. Columns (1) and (5) report results for withdrawn firms for the full sample period (1994-2018), while Columns (2) and (6) report results for the withdrawn firms corresponding to the sample period of the private and public controls(2004-2018). Columns(3)and(7)((4)and(8))reportresultsfortheprivate(public) firm control sample. BTD (Cash ETR) is increasing (decreasing) in tax avoidance; thus, across all eight columns, we observe results consistent with increased tax avoidance post-IPO. The coefficients in the first four columns mean that IPO completion is associated with an increase in BTD of 2.1 to 4.8 percentage points. Figure 3 plots year-specific treatment effects obtained from estimation of Eq. (2); Figures 3a, 3b, and 3c correspond to the withdrawn IPO firm (over the 1994 to 2018 period), private firm, and public firm control groups, respectively. These graphs first confirm that the three samples generally exhibit parallel trends in BTD in the years preceding the IPO. We also observe statistically significant increases in BTD after the IPO, consistent with the Table 4 results. Columns (5) through (8) present results using domestic Cash ETR as the measure of tax avoidance. The effects are estimated on a smaller sample due to dropping loss firms. Despite this smaller sample, we observe results consistent with increased tax planning: Cash ETR declines by 3.9 to 12.5 percentage points after the IPO, equivalent to about a 15 to 50 percent decline.17 Figures 3d and 3e graph the decline in Cash ETR for completed IPOs, consistent with the Table 4 results. We observe similar effects when using the public firm control sample, but we acknowledge that the pre-period trends are dissimilar and thus we cannot attribute the effects in this sample to the IPO necessarily. Theeconomicmagnitudesimplylargeincreasesintaxplanning. However, bothBTD and 17The 15 (50) percent change is calculated for the public control sample (withdrawn firm sample over the 2004 to 2018 period) based on the pre-IPO untabulated average Cash ETR of 0.29 (0.23). All magnitudes are based on average values from the inverse probability weighted sample but are similar if calculated using unweighted descriptive statistics. 17

Cash ETR capture any divergence between book and taxable income (Hanlon and Heitzman (2010)). Thus, to assess whether the effects relate to explicit tax planning strategies that companies implement around the IPO, we conduct several additional tests using the underlying tax data. We find that the results are largely attributable to two factors not considered “explicit” planning. First, we decompose BTD using the specific book-tax difference line items from the corporate income tax return and find that the effects are largely explained by increased stock option deductions; see Columns (1)-(3) of Panel B.18 Second, we find that a portion of the increased difference between book and taxable income appears attributable to increased earnings management activity, which drives higher financial-reporting income but not taxable income.19 Additional untabulated analyses reveal that no other particular book-tax difference exhibits similar effects as these items; furthermore, when we adjust BTD to exclude the effects of both stock compensation and earnings management, we observe insignificant effects for the public and private firm control samples and substantially diminished effects in the withdrawn IPO sample. In summary, firms appear to engage in little explicit domestic tax planning after the IPO, likely because stock option deductions are sufficiently large to absorb U.S. taxable income for several post-IPO years. Documenting the true factors driving changes in domestic tax 18Specifically,weexamineeachlinereconcilingbookincometotaxableincomeonIRSScheduleM-3. Stock option deductions naturally increase after the IPO, as IPOs generally trigger stock option vesting. This in turn accelerates the exercise of options and the corresponding stock option deduction. Consequently, due to the difference in book and tax accounting for stock options (and not explicit tax planning activities), IPO firms will naturally report higher stock option deductions and thus larger BTD after the IPO. 19Prior work shows that earnings management incentives may be particularly strong around the IPO, as managers intend to both maximize the offering price at the time the firm goes public and meet subsequent earnings benchmarks to sustain their post-IPO stock price (Teoh, Wong and Rao, 1998; Teoh, Welch and Wong, 1998). However, because tax accounting methods are generally more “cash-based,” tax law may require firms to reverse certain income-increasing accruals when calculating taxable income. In untabulated analyses, we find evidence consistent with this. Specifically, we find that the increase in BTD varies based on two financial reporting incentives: (i) whether the company was covered by an analyst post-IPO and (ii) whether the company reported high discretionary accruals. We then examine the specific book-tax difference line items to measure whether the income-increasing earnings management activity is excluded for tax purposes and indeed find large BTD accrual adjustments by IPO firms. As expected, this effect is largest when comparing IPO firms to private firms, given that these firms had no incentive to increase their earnings management behavior. 18

avoidance measures is important for the literature, particularly because concurrent work also uses these measures to study IPO-related tax planning effects (e.g., Chen et al., 2021; Chyz et al., 2022). 4.2 Foreign Tax Planning We next test whether the IPO is associated with foreign tax planning activity. There are two factors unique to IPOs that motivate foreign tax planning: the requirement to disclose tax information via public financial reporting, and IPO valuation implications for firm intangibles. First, public disclosure requirements should impact tax planning, likely motivating firms to engage in similar levels and types of avoidance as other public firms. Because firms have little reason to engage in additional domestic tax planning (in light of the large stock option deductions discussed above), firms may instead use foreign tax planning to converge to the level exhibited by public companies. Second, the IPO provides a clear, observable signal of firm value that impacts the cost of establishing foreign income-shifting structures. For example, certain income-shifting strategies entail transferring intangible assets from the U.S. to lower-taxed foreign jurisdictions, triggering a U.S. “exit tax” calculated on the intangible’s fair market value.20 Consequently, firms have incentives to shift intangibles offshore in years preceding the IPO (when the value is likely lower and less certain), to both minimize the exit tax and position the firm to use these strategies in the post-IPO period. This section documents evidence consistent with both factors. We first provide a graphical analysis of trends in tax haven use around the IPO. Figures 4a and 4b plot the average levels of Havens (0/1) and #Havens for IPO firms as compared to the private and public firm control groups in the years around equity issuance (year 𝑡). 20Specifically,InternalRevenueCodeSection367imposesataxwhenU.S.companiestransferappreciated assets, such as internally developed intellectual property, to foreign subsidiaries. See technical discussions in Hicks and Sotos (2008), Fuller and Halpern (2012), and Murillo et al. (2016). 19

The top graph provides striking evidence of increased foreign tax planning. At the beginning of the sample period – three years prior to the IPO – 17 percent of IPO completers own a tax haven subsidiary. This level is similar to that of the private firm sample. Tax haven ownership among IPO completers then climbs, increasing to 36 percent by the IPO filing year. The level converges near the level of public firms at 47 percent by 𝑡 + 3. Panel B confirms that IPO firms also increase the number of haven subsidiaries around the IPO, with trends tracking those of public firms. While descriptive, both figures show that IPO firms increase their haven use around the IPO to more closely resemble that of public firms. Although Figure 4 provides strong evidence of changes in foreign planning around the IPO,thesegraphsalsoshowthatIPOcompletingfirmsdonotexhibitparalleltrendswiththe private or the public control samples for these outcomes. Thus, we use only the withdrawn IPO control sample for the next empirical tests. Figures 5a and 5b confirm that indeed the IPO completers and withdrawers exhibit similar pre-IPO trends for both Haven (0/1) and #Havens. Furthermore, across both figures, we observe increases by IPO completing firms in the post-IPO period. Table 5, Panel A presents empirical results corresponding to these figures, with results for Havens (0/1) in Column (1) and #Havens in Column (2). The probability that an IPO-completing firm has a haven subsidiary increases by 16.1 percentage points; based on the average haven incidence of 28.3 percent for IPO firms in the year prior to going public (untabulated), this is equivalent to a 57 percent increase. The number of haven and dot haven subsidiaries increases by 0.27 and 0.10, respectively, translating to a 34 percent and 41 percent increase compared to (untabulated) pre-IPO levels of 0.79 and 0.25, respectively. Observing statistically significant increases in dot havens underscores that the effects are largely attributable to tax motivations, not simply to growth and expansion of IPO firms. Having established an increased tax haven presence, we next test the extent to which 20

firms engage in cross-border income shifting within this entity structure. We first study whether the IPO is associated with increased use of cost-sharing agreements. The coefficient of 0.038 on Cost sharing agreement (0/1) in Panel B, Column (1) means that the likelihood of having a subsidiary participate in such an arrangement increases by 3.8 percentage points after the IPO. Based on the average value of 3.1 percent in this sample in the year prior to the IPO (Table 3), this effect is roughly a 125 percent increase after the IPO. Results presented in Columns (2) through (5) further show that these agreements indeed facilitate cross-border payments. We observe a coefficient of 2.179 in Column (2) for Total amounts received from U.S. While higher levels of cross-border payments could be attributable to post-IPO growth within a vertically integrated multinational firm, two additional results link this increase to tax-motivated income shifting. First, the Total amounts paid to U.S. also increased (see Column (3)), but the coefficient is 1.061. Although the coefficient on inbound payments is within the 90 percent confidence interval of the coefficient on theoutboundpayments(theconfidenceintervalis0.750to3.609), thegreatereconomicmagnitude for amounts paid from (rather than to) the U.S. is suggestive of shifting income into lower-taxed jurisdictions after the IPO. Second, when refining the measures to only include those line items most related to income shifting in Column (4) (i.e., the sale of intangibles, royalty payments, licensing fees, etc.), we observe a similar coefficient as in Column (2). This means that the increased payments in Column (2) primarily relate to income shifting transactions, not to routine intercompany sales of goods and services that are explicitly excluded in Column (4). In Column (5), we observe a similar coefficient even after augmenting the measure to include intercompany payments from other foreign subsidiaries, meaning that the primary effect in Column (4) is attributable to shifting out of the U.S. In summary, Table 5 shows that IPOs are associated with (i) increased haven ownership, (ii) higher likelihood of income shifting agreements, and (iii) increased intangibles-related 21

payments to foreign subsidiaries. The evidence is consistent with firms creating and increasingly using multinational tax planning structures around the time that they go public. 4.3 Heterogeneity in Effects 4.3.1 Domestic R&D-Intensive Firms We expect that the foreign planning and income shifting effects documented above should be the greatest in high intangibles firms. This is because intangible assets generate “mobile” income that can be more easily shifted via intercompany and cross-border transfers of internally developed intangibles (Grubert, 2003; De Simone, Mills and Stomberg, 2019; Bilicka et al., 2022). Thus, in this section, we test whether we observe heterogeneity in the foreign planning effects based on companies pre-IPO research and development (R&D) spending.21 Specifically, we calculate firms’ average R&D spending over the period t-3 to t-1 using R&D expenses reported on U.S. tax returns.22 We then partition the sample based on the median level of pre-IPO domestic R&D expense and re-estimate Eq. (1). Table 6 Panels A and B report results for Havens (0/1) and Cost sharing agreement (0/1), respectively. In both panels, the coefficients suggest a larger economic magnitude for high R&D firms in Column (1) as compared to low R&D firms in Column (2). For example, the incidence of havens increases by 21.4 percentage points among high-domestic-R&D firms, as compared to 13.3 percentage points among low-domestic-R&D firms. This difference is marginally significant (p=0.101). In Panel B, we similarly observe a 1.6 percentage point difference for Cost sharing agreement (0/1), although this difference is not statistically significant (p=0.34). 21Under U.S. accounting rules, internally developed intangibles are not recorded as assets but instead expensed as research and experimentation. Thus, the most appropriate way to test for this heterogeneity is to examine domestic R&D expense. 22We use information reported on Form 6765 for the U.S. Research and Experimentation Tax Credit. 22

Observing these patterns further supports the inference that the effects in Table 5 are attributable to tax planning and not simply to growth among vertically integrated firms (whichwouldotherwisesuggestnovariationintheresults). Furthermore, theseresultsreveal the extent of worldwide tax planning for high-intangibles firms: firms develop intangibles in the U.S., often reducing or minimizing their domestic taxable income through the R&D credit, and then prior to the IPO valuation event, shift domestically developed IP offshore to source subsequent income to the foreign jurisdiction. These effects suggest that firms may pay minimal tax—to either the U.S. or foreign jurisdictions—in the years surrounding the IPO. 4.3.2 Mechanism: Effects of Professionalization We next examine the mechanism by which firms engage in increased foreign tax planning. Recent work studies the role of professional tax advisors in business tax decisions, including responsiveness to tax reform changes (Bustos et al., 2022), increased take-up for tax-based subsidies (Goodman, 2021), and tax avoidance activities (Battaglini et al., 2020). For example, Zwick (2021) shows that, among firms eligible to claim refunds for tax losses, those that have more sophisticated tax advisors are more likely to do so. We build on this recent literature and test whether greater tax professionalization affects firms’ IPO-induced foreign tax planning. We obtain preparer information using identifying information about the accounting firm that prepares and signs IPO firms’ income tax returns. As tax preparer data for our sample are most complete since 2005, our analysis covers the period 2005 to 2018.23 In a similar vein to Zwick (2021), we define a sophisticated tax preparer as a firm that prepares tax returns 23Unlike the other data used in the manuscript, tax preparer information is not included in the sample of cleanedandeditedIRSdata;rather,weobtainthisinformationfromtheuneditedpopulationoftaxreturns. To construct this variable, we use preparer EIN data reported on the bottom of Form 1120, Page 1. 23

for more than 4,000 C corporations in any given tax year. During the 2005 to 2018 period, 53 percent of the IPO sample consists of tax returns prepared by a sophisticated preparer (Table 2). Figure 6 provides descriptive evidence about the proportion of IPO firms that use professional advisors. Figure 6a shows the fraction of IPO completing firms relative to the public and private samples. We observe similar patterns as the earlier haven ownership graph (Figure 4a). That is, IPO firms’ use of professional advisors is similar to private firms in year t-3 (41 percent by IPO firms versus 36 percent by private firms). The proportion then increases at a faster rate than private firms, converging to the same level as public firms immediately preceding the IPO (year t-1) and then slightly exceeding that level by year t+2 (53 percent for IPO firms versus 50 percent for public firms). In contrast to the IPO firms, public firms’ use of sophisticated tax advisors remains constant throughout the window. Figure 6b compares the use of professional advisors for IPO completing firms relative to withdrawn firms. Withdrawn firms have similar levels and increases in professional advisor use in year t-3 and t-2, but these firms then shift to less sophisticated advisors around the time of the withdrawal. Table6linksincreasedprofessionalizationtoincreasedforeigntaxplanning. Wepartition IPO firms based on whether the firm switched to a sophisticated tax preparer at any point between year t-2 and t relative to the IPO. We then re-estimate Eq. 1 for Haven (0/1) and Cost-Sharing (0/1) andreportresultsinPanelAandPanelB,respectively. AsinTable5, we use the withdrawn sample because of the similarity of pre-treatment foreign planning trends. We find evidence consistent with larger foreign planning effects among those firms that switched to larger tax advisors; for example, the likelihood of haven ownership increased by 30.8 percentage points in Panel A, Column (3) as compared to 11.0 percentage points among those that did not switch (Column (4)). This 19.8 percentage point difference is statistically 24

significant (p=0.00). Similarly, the likelihood of participating in a cost sharing program increased by 8.3 percentage points (Panel B, Column (3)) as compared to 2.9 percentage points among non-switchers (Column (4)), and the 5.4 percentage point difference is also statistically significant (p=0.04). We perform three additional untabulated analyses to confirm the foreign planning effects can be attributed to increased professionalization. First, if IPO firms hire new executives in the years prior to the IPO, and if the executives select different tax advisors, then the foreign planning effects may ultimately be attributable to the change in executives. We test this using tax data on the highest-paid individuals at the firm reported on Form 1125-E; use of these data assumes that the highest paid individuals are likely top executives. We identify firms with an executive change based on if any of the five highest-paid individuals in year t-2 are no longer listed as of year t. Due to reporting changes in the tax data, we can unfortunately only measure these changes for a tiny fraction of IPO-completing firms. Among this sample, we observe that less than 20% of the IPO-completing firms change executives, mitigating concerns that the observed professionalization effects are driven by underlying changes in management. Second, we study whether the effects are attributable to venture capital ownership. One possible explanation for the observed results is that VCs encourage firms to implement tax planning, and that such planning would occur regardless of the IPO. We obtain data from SDC on whether the IPO was VC-backed and partition the sample of both the IPOcompleters and withdrawers based on this variable. We then test whether the cost sharing and foreign tax haven results vary across these partitions. We observe two results that mitigate the concern that the foreign planning effects are driven by VC ownership. First, within the VC-backed partition, we observe increased use of haven subsidiaries and cost sharing agreements by VC-backed IPO completing firms, relative to VC-backed IPO withdrawing 25

firms. This result suggests that the effects occur even after controlling out VC ownership. Second, we observe statistically significant effects across both the VC-backed and non-VC backed IPO completing firms. Observing that the results do not vary by VC ownership – in contrast to the variation observed for tax advisors – further suggests that the foreign planning effects are attributable to professionalization. Third, we study whether the domestic tax planning effects varies based on professional advisors. We find little effect of increased professionalization among the domestic tax planning effects, consistent with the fact that increases in BTD are largely driven by stock option deductions (Table 4), which can be claimed with little external professional advice. Observing this variation further suggests that tax professionalization is a plausible explanation for the increased foreign tax planning effects. The collective results suggest that more sophisticated advisors facilitated the implementation of international tax structures. It is not possible to determine whether a higher level of tax professionalization leads firms directly to increase tax planning activity or whether pre-IPO managers seek out more sophisticated tax preparation firms. However, the trends and observed heterogeneity imply greater investment in and salience of foreign tax planning around the IPO. 4.4 Additional Analyses 4.4.1 Robustness Tests We perform several additional analyses to check the robustness of our results to alternative samples and specifications. Table 7 presents results for the 2004-2018 sample using withdrawn companies as control firms; results for domestic BTD are presented in Panel A and for foreign Havens (0/1) in Panel B. Results are generally robust across a number of tests. For ease of comparison, we repeat 26

the baseline results from Table 4 and Table 6 in Column (1) of Panel A and Panel B, respectively. Column(2)presentsresultswithoutinverseprobabilityweighting, confirmingthatthe results are not driven by this approach. For BTD, the baseline coefficient is unchanged; for Havens (0/1), the coefficient is slightly smaller but remains statistically significant. Column (3) shows results clustering standard errors by firm rather than by industry, following Bernstein (2015). Standard errors increase, but results remain statistically significant. Column (4) reports results after estimating Eq. 1 including time-varying controls for Debt/Assets, ln(Age), NetPPE/Assets, ln(Sales), and Pre-NOL taxable income/Sales. These controls have little effect on our baseline results. In Column (5), we relax the requirement that a firm must have four years of observations to be included in the sample (from 𝑡−2 to 𝑡+1). We instead require a firm to only be present for two years (year 𝑡−1 and 𝑡), resulting in a 25 percent larger sample across both panels. Estimation on these larger samples yields a somewhat smaller estimated coefficient for Havens (0/1) but little change for BTD. Columns (6) through (9) of Panel A present three additional tests for BTD. First, we report results after Winsorizing at the 1 percent level, rather than 5 percent as in the baseline. We observe a slightly larger coefficient, suggestive of outliers indeed influencing the point estimates.24 Column (7) reports results from alternatively scaling BTD by sales to mitigate concerns that effects are driven by large increases in assets around the IPO. We observe a modest decrease in coefficient size and significance. Column (8) reports results after limiting the sample to only those firms that file the requisite tax forms to construct the foreign planning sample (Schedule M-3). This restriction has little effect on the coefficient, confirming that results are unaffected by sample construction. Finally, Column (9) reports results using an alternative tax-based measure of BTD, which is equal to the difference in financial reporting tax expense and actual cash taxes paid (Mills, 1998). Given that the measure is 24We omit this column from Panel B because we do not Winsorize indicator variables. 27

based on differences in tax liabilities (and not differences in book and tax income), we find the expected smaller coefficient. Observing that this coefficient is statistically significant confirms that the inferences extend to this alternative measure. 4.4.2 U.S. Statutory Limitations on NOL Use One concern is whether the documented increase in BTD is attributable to statutory restrictions imposed on the use of tax losses after large changes in ownership. U.S. tax law imposes a statutory limitation that reduces the amount of loss carryforwards a firm can use to offset post-IPO income.25 While BTD should be relatively unaffected by this limitation because the measure is constructed using pre-NOL taxable income, we conduct two additional tests to further mitigate concerns about this limitation mechanically driving results. First, we construct an indicator equal to one for those firms affected by this limitation in the post- NOL period; that is, we identify those firms that claim lower levels of tax loss carryforwards post-IPO than they should otherwise be able to report.26 Columns (1) to (4) of Panel C presents results from regressing this indicator on 𝐼𝑃𝑂𝐶𝑜𝑚𝑝𝑙𝑒𝑡𝑒𝑑*𝑃𝑜𝑠𝑡 to test whether the incidence of the limitation changes after the IPO. We find no statistically significant change, meaning that the limitation does not substantially affect the firms included in our sample, 25Specifically, InternalRevenueCodeSection382limitstheabilityofafirmtouseitsnetoperatinglosses (NOLs) if the firm has had a greater than 50% ownership change over a three year rolling period. Briefly, thelimitationisequaltothefirm’smarketvalueatthetimeoftheownershipchange,timesthefederallongterm tax-exempt rate; the limit can be increased for certain “built in gains” under IRS Notice 2003-65. See Erickson and Heitzman (2010), Erickson, Heitzman and Zhang (2013), and Sikes, Tian and Wilson (2014) for more discussion of this limitation. This U.S. tax limitation should have essentially no effect on firms’ propensity to implement foreign planning. 26Specifically, the indicator variable is equal to one if the firm NOL deduction (Form 1120, line 29a) is lessthanthemaximumallowableNOLdeduction. ThemaximumallowableNOLdeductionisdefinedasthe minimum of pre-NOL taxable income (Form 1120, line 28) minus special deductions (Form 1120, line 29b) and the firm net operating loss stock (Form 1120: Schedule K, line 12). 28

at least within the three years following the IPO.27 Second, we drop the small number of firms for which the statutory limitation appears to bind in Columns (5) to (8). The effect of the IPO on BTD declines only slightly in magnitude and remains statistically significant at the 1% level across all four samples. 4.4.3 Instrumental Variables Analysis Our empirical design uses three distinct samples of control firms to isolate the effect of the IPO on corporate tax planning. In additional untabulated tests, we employ the instrumental variables (IV) approach of Bernstein (2015), who uses the 2-month NASDAQ composite return as an instrument for IPO completion. Specifically, we use this instrument and then adapt his approach to our setting by using the stacked cohort empirical design in the second stage. We confirm that this is a valid instrument for the sample period 1994 through 2018 (F-statistic of 18.5), which includes the sample period that overlaps with Bernstein’s sample (1985-2003). We find consistent, although slightly weaker results for BTD in the second stage; the coefficient is 0.085 (t-statistic of 1.613). For the period with requisite foreign data from 2004 to 2018 (which does not overlap with Bernstein’s sample period), we observe that the instrument in the first stage is weak; the F-statistic is 4.6, lower than the critical value of 8.96 (Larcker and Rusticus, 2010; Stock, Wright and Yogo, 2002). Consequently, this weak instruments problem precludes us from using the instrument for the most recent subsample with available foreign data. 27One explanation for the lack of effect is that companies continue to generate losses in the post-IPO period, such that the limitation does not bind until later years outside of the sample window. The second is that IPO firms have large built-in gains, which are equal to the difference in asset value (high at the time of the IPO) and the company’s tax basis (likely low). Large built-in gains increase the annual limitation, thereby relaxing the constraint on tax loss utilization. 29

5 Conclusion This paper studies the extent to which domestic and foreign tax planning activity changes around U.S. IPOs. Using confidential corporate tax data for IPO firms, we document three findings. First,theIPOmotivatesfirmstosetupandexpandtheirinternationaltaxstructure to facilitate post-IPO income shifting. Second, professional advisors play an important role in this foreign planning. Finally, firms do not appear to engage in domestic tax planning, as IPO-related stock option deductions appear to provide sufficient tax shields to absorb domestic taxable income. The foreign planning effects are consistent with firms converging to the level and types of tax avoidance exhibited by the public firms in the sample. When and why firms implement multinational structures is an open question in the literature. We provide direct evidence showing that the IPO is an event that motivates such activity. Beyond informing the academic and policy work related to addressing firms’ cross-border income shifting activity, we also provide evidence relevant for domestic revenue projections. There has been a striking decline in publicly listed firms in the United States in the past two decades, which has raised questions about the reasons for the decline and the potential consequences of fewer public firms. Our results document the transition from private to public status and demonstrate how this transition is associated with changes in firm tax planning activity. We look forward to future work that continues to evaluate the tax implications of IPO transactions. 30

References Angrist, Joshua D, and J¨orn-Steffen Pischke.2008.Mostly harmless econometrics.Princeton university press. Badertscher, Brad A, Sharon P Katz, and Sonja O Rego. 2013. “The separation of ownership and control and corporate tax avoidance.” Journal of Accounting and Economics, 56(2- 3): 228–250. Badertscher, Brad A, Sharon P Katz, Sonja Olhoft Rego, and Ryan J Wilson. 2019. “Conforming tax avoidance and capital market pressure.” The Accounting Review, 94(6): 1–30. Battaglini, Marco, Luigi Guiso, Chiara Lacava, and Eleonora Patacchini. 2020. “Tax professionals and tax evasion.” NBER Working Paper No, 25745. Bernstein, Shai. 2015. “Does going public affect innovation?” The Journal of finance, 70(4): 1365–1403. Biden, J. 2022. “This Tax Day, the President is Fighting to Reward Work, Not Wealth, While Republicans Want to Increase Taxes on the Middle Class.” Bilicka, Katarzyna A, Michael P Devereux, Irem Guceri, et al. 2022. “Tax Avoidance Networks and the Push for a’Historic’Global Tax Reform.” Tax Policy and the Economy, 37. Blouin, Jennifer, and Linda Krull. 2009. “Bringing It Home: A Study of the Incentives Surrounding the Repatriation of Foreign Earnings Under the American Jobs Creation Act of 2004.” Journal of Accounting Research, 47(4): 1027–1059. Blouin, Jennifer, and Linda Krull. 2018. “Does tax planning affect organizational complexity: evidence from the check-the-box.” Working paper. Brav, Omer.2009.“Accesstocapital,capitalstructure,andthefundingofthefirm.”The Journal of Finance, 64(1): 263–308. Bustos, Sebastian, Dina Pomeranz, Juan Carlos Suarez Serrato, Jose Vila-Belda, and Gabriel Zucman. 2022. “The race between tax enforcement and tax planning: evidence from a natural experiment in Chile.” NBER Working paper, 30114. Chen, Huimin, Bill Francis, Qiang Wu, and Yijang Zhao. 2021. “Strategic Reaction and Tax Avoidance: Evidence from the Effect of Large IPOs on Peers.” Working paper. Chen, S., X. Chen, Q. Cheng, and T. Shevlin. 2010. “Corporate tax avoidance and highpowered incentives.” Journal of Financial Economics, 95(3): 41–61. Chyz, James, Erin Henry, Thomas C Omer, and Biyu Wu.2022.“Mimickingtaxstrategies: Evidence from IPOs.” Available at SSRN 3463499. Cohn, Jonathan B, Lillian F Mills, and Erin M Towery. 2014. “The evolution of capital structure and operating performance after leveraged buyouts: Evidence from US corporate tax returns.” Journal of Financial Economics, 111(2): 469–494. Cooper, Michael, and Matthew Knittel. 2006. “Partial Loss Refundability: How are Corporate Tax Losses Used?” National Tax Journal, 59(3): 651–653. Cooper, Michael, and Matthew Knittel. 2010. “The Implications of Tax Asymmetry for U.S. Corporations.” National Tax Journal, 63(1): 33–62. Dambra, Michael J, Matthew Gustafson, and Phillip J Quinn. 2020. “Tax-advantaged trust use among IPO executives: Determinants and implications for valuation and future performance.” The Accounting Review, 95(3): 145–175. Desai, M., and D. Dharmapala. 2006. “Corporate tax avoidance and high-powered incentives.” Journal of Financial Economics, 79(1): 145–179. De Simone, Lisa, and Marcel Olbert.2021. “Real effects ofprivatecountry-by-country reporting.” The Accounting Review. De Simone, Lisa, and Richard Sansing. 2019. “Income shifting Using a cost-sharing arrangement.” Journal of the American Taxation Association, 41(1): 123–136. De Simone, Lisa, Lillian F Mills, and Bridget Stomberg. 2019. “Using IRS data to identify income shifting to foreign affiliates.” Review of Accounting Studies, 24(2): 694–730. Dobridge, Christine L, Erik P Gilje, and Andrew Whitten. 2022. “The secular decline in private firm leverage.” National Bureau of Economic Research. 31

Drucker, J.2011.“GlobaltaxloopholessaveCiscobillions-anditwantsmore.”Bloomberg News. Drucker, J. 2021. “An Accidental Disclosure Exposes a $1 Billion Tax Fight with Bristol Myers.” New York Times. Dyreng, Scott D, and Bradley P Lindsey. 2009. “Using financial accounting data to examine the effect of foreign operations located in tax havens and other countries on US multinational firms’ tax rates.” Journal of Accounting Research, 47(5): 1283–1316. Dyreng, Scott D, Michelle Hanlon, Edward L Maydew, and Jacob R Thornock. 2017. “Changesincorporateeffectivetaxratesoverthepast25years.”JournalofFinancialEconomics, 124(3): 441–463. EconPol. 2021. “OECD Tax Reform Affects Only 78 of tthe World’s Largest 500 Companies.” EconPol.edu, July 5. Edwards, Alexander, and Michelle Hutchens. 2020. “Taxes and IPO pricing: Evidence from US tax reform.” Available at SSRN 3582280. Edwards, Alexander, Michelle Hutchens, and Sonja Olhoft Rego. 2019. “The pricing and performance of supercharged IPOs.” The Accounting Review, 94(4): 245–273. Erickson, Merle, and Shane Heitzman.2010.“NOLpoisonpills: Selecticav.Versata.”Versata (June 23, 2010). Tax Notes, June, 23. Erickson, Merle M, Shane M Heitzman, and X Frank Zhang. 2013. “Tax-motivated loss shifting.” The Accounting Review, 88(5): 1657–1682. Faulkender, Michael, Kristine Hankins, and Mitchelle Petersen. 2019. “Understanding the rise in corporate cash: Precautionary savings or foreign taxes.” Review of Financial Studies, 32(9): 3300–3334. Feldman, Naomi, Laura Kawano, Elena Patel, Nirupama Rao, Michael Stevens, and Jesse Edgerton.2021.“Investmentdifferencesbetweenpublicandprivatefirms: Evidencefrom US tax returns.” Journal of Public Economics, 196: 104370. Foley, Fritz, Jay Hartzell, Sheridan Titman, and Gary Twite. 2007. “Why do firms hold so much cash? A tax-based explanation.” Journal of Financial Economics, 86: 579–607. Fuller, James, and Adam Halpern.2012.“NewRulesforCodeSec.304/CodeSec.367Overlap Transactions: The IRS Comes Full Circle with Notice 2012-15.” International Tax Journal, 38(5): 5–10. Gao, Huasheng, Jarrad Harford, and Kai Li.2013.“Determinantsofcashholdings: Evidence from private firms.” Journal of Financial Economics, 109(3): 623–639. Goodman, Lucas. 2021. “Take-up of Payroll Tax-Based Subsidies During the COVID-19 Pandemic.” Working paper. Graham, John R., M. Hanlon, and T. Shevlin. 2010. “Barriers to Mobility: the Lockout Effect of U.S. Taxation of Worldwide Corporate Profits.” National Tax Journal, 63: 1111–1144. Graham, John R., M. Hanlon, T. Shevlin, and N. Shroff.2014.“IncentivesforTaxPlanning and Avoidance: Evidence from the Field.” The Accounting Review, 89(3): 991–1023. Grinberg, I., and R. Kysar. 2021. “Why the United States Needs a 21% Minimum Tax on Corporate Foreign Earnings.” U.S. Department of Treasury, September 7. Grubert, Harry. 1998. “Taxes and the division of foreign operating income among royalties, interest, dividends, and retained earnings.” Journal of Public Economics, 68: 269–290. Grubert, Harry. 2003. “Intangible income, intercompany transactions, income shifting, and the choice of location.” National Tax Journal, 61(1): 221–242. Hanlon, Michelle, and Shane Heitzman.2010.“Areviewoftaxresearch.”Journal of accounting and Economics, 50(2-3): 127–178. Hanlon, Michelle, Rebecca Lester, and Rodrigo Verdi. 2015. “The effect of repatriation tax costs on U.S. multinational investment.” Journal of Financial Economics, 116: 179–196. Hicks, Hal, and David Sotos. 2008. “The Empire Strikes Back (Again) - Killer Bs, Deadly Ds, and Code Sec. 367 as the Death Star Against Repatriation Rebels.” International Tax Journal, 34-64: 37–. Hines, James, and Eric Rice. 1994. “Fiscal paradise: Foreign tax havens and American business.” The Quarterly Journal of Economics, 109(1): 149–182. 32

Hoopes, Jeffrey L, Patrick Langetieg, Edward L Maydew, and Michele Mullaney. 2020. “Is tax planning best done in private?” Available at SSRN 3420362. Larcker, David F, and Tjomme O Rusticus. 2010. “On the use of instrumental variables in accounting research.” Journal of Accounting and Economics, 49(3): 186–205. Lisowsky, Petro, and Michael Minnis. 2020. “The silent majority: Private US firms and financial reporting choices.” Journal of Accounting Research, 58(3): 547–588. Lowry, Michelle, Roni Michaely, and Ekaterina Volkova. 2017. “Initial public offerings: A synthesis of the literature and directions for future research.” Forthcoming Foundations and Trends in Finance. Michaely, Roni, and Michael R. Roberts. 2012. “Corporate dividend policies: Lessons from private firms.” The Review of Financial Studies, 25(3): 711–746. Mills, Lillian F. 1998. “Book-tax differences and Internal Revenue Service adjustments.” Journal of Accounting Research, 36(2): 343–356. Murillo, Jose, David Waimon, Russell Carr, Gary Scanlon, and Gunnar Haugen. 2016. “TaxationofOutboundTransfersofForeignGoodwillorGoingConcernValueunderSecs.367(a) and (d).” Tax Adviser, January 1. OECD. 2021. “Statement on a Two-Pillar Solution to Address the Tax Challenges Arising from the Digitalisation of the Economy.” October 8. Olbert, Marcel, and Peter Severin. 2022. “Private equity and local public finances.” Available at SSRN 3287687. Rego, Sonja. 2003. “Tax avoidance activities of U.S. multinational corporations.” Contemporary Accounting Research, 20(4): 805–833. Ritter, Jay R, and Ivo Welch. 2002. “A review of IPO activity, pricing, and allocations.” The Journal of Finance, 57(4): 1795–1828. Roberts, Michael R, and Toni M Whited. 2013. “Endogeneity in empirical corporate finance1.” In Handbook of the Economics of Finance. Vol. 2, 493–572. Elsevier. Rubin, R., and J. Walker. 2022. “Amgen Fights IRS Over $10.7 Billion Tax Bill.” Wall Street Journal. Rubin, R., and T. Francis. 2022. “Amazon, Berkshire Hathaway Could be Among Top Payers of New Minimum Tax.” Wall Street Journal. Sikes, Stephanie A, Xiaoli Shaolee Tian, and Ryan Wilson. 2014. “Investors reaction to the use of poison pills as a tax loss preservation tool.” Journal of Accounting and Economics, 57(2-3): 132–148. Stock, James H, Jonathan H Wright, and Motohiro Yogo. 2002. “A survey of weak instruments and weak identification in generalized method of moments.” Journal of Business & Economic Statistics, 20(4): 518–529. Teoh, Siew Hong, Ivo Welch, and Tak J Wong. 1998. “Earnings management and the longrun market performance of initial public offerings.” The Journal of Finance, 53(6): 1935–1974. Teoh, Siew Hong, Tak J Wong, and Gita R Rao. 1998. “Are accruals during initial public offerings opportunistic?” Review of Accounting Studies, 3(1): 175–208. Wilde, Jaron H, and Ryan J Wilson. 2018. “Perspectives on corporate tax planning: Observations from the past decade.” The Journal of the American Taxation Association, 40(2): 63–81. Yagan, Danny. 2015. “Capital tax reform and the real economy: The effects of the 2003 dividend tax cut.” American Economic Review, 105(12): 3531–63. Zwick, Eric. 2021. “The costs of corporate tax complexity.” American Economic Journal: Economic Policy, 13(2): 467–500. Zwick, Eric, and James Mahon. 2017. “Tax policy and heterogeneous investment behavior.” American Economic Review, 107(1): 217–48. 33

Appendix A: Variable definitions The table below provides our variable definitions. All data are sourced from the IRS unless otherwise noted, and all dollar-denominated variables are measured in thousands. Table A1: Variable definitions Variable Definitiona Age Tax year minus year of incorporation (Form 1120: box C) Assets Form 1120: Schedule L line 15(d) (when missing prior year Assets, use beginning-of-year assets from Form 1120: Schedule L line 15(b) and when missing beginning-of-year assets, use current year assets) Asset growth Percentage change in Assets from the prior year to the current year Book income (loss) When a firm reports attaching a Schedule M-3 to the Form 1120 (Box A4): Net income (loss) per income statement of includible corporations (Schedule M-3, Part I, line 11). When a firm does not report attaching a Schedule M-3 or in the years the Schedule M-3 was not utilized: Net income (loss) per books (Schedule M-1, line 1). Book-tax differences [Book income (loss) + Federal income tax expense - Pre-NOL taxable (BTD) income (loss)]/prior year Assets BTD tax [Federal income tax expense - Taxes Paid]/prior year Assets Cash ETR Taxes Paid/[Book income (loss) + Federal income tax expense] Cost sharing agree- Indicator variable equal to 1 if firm answers “Yes” to Form 5471, ment (0/1) Schedule G, Question 4 (“During the tax year, was the foreign corporation a participant in any cost-sharing arrangement?”) for at least one controlled foreign corporation (CFC) Debt Form 1120: Schedule L line 17(d) + line 20(d) Debt/Assets Debt/Assets Domestic book income Book income (loss) + Federal income tax expense Domestic taxable in- Pre-NOL taxable income (loss) come # Dot havens Number of CFCs reported on Form 5471 and located in countries designated with asterisks in the list of tax havens later in this table. Equity financing Equity issuance/[Debt + Total paid-in capital] aLine numbers reference the 2012 tax forms unless otherwise noted. 34

Table A1: Variable definitions (continued) Variable Definition Equity issuance Change in Total paid-in capital from the prior year minus the change in Treasury stock from the prior year. Federal income tax ex- When a firm reports attaching a Schedule M-3 to the Form 1120 (Box pense A4): U.S. current income tax expense (Schedule M-3, Part III, line 1) + U.S. deferred income tax expense (Schedule M-3, Part III, line 2). When a firm does not report attaching a Schedule M-3 or in the yearstheScheduleM-3wasnotutilized: federalincometaxperbooks (Schedule M-1, line 2). Foreign income items (Schedule M-3, Part II, line 1 + line 2 + line 3 + line4 + line5 + line 10) # Havens Number of CFCs reported on Form 5471 and located in one of the following countries: Andorra*; Antigua and Barbuda*; Aruba*; Bahamas*; Bahrain*; Barbados*; Belize*; Bermuda*; Botswana; British Virgin Islands; Brunei Darussalam; Cape Verde; Cayman Islands*; Cook Islands*; Costa Rica*; Cyprus*; Dominica*; Gibraltar*; Grenada*; Guernsey and Alderney*; Hong Kong; Ireland; Isle of Man*; Jersey*; Latvia; Lebanon; Liberia; Liechtenstein*; Luxembourg*; Macau*; Maldives; Malta*; Marshall Islands; Mauritius*; Monaco; Monserrat; Nauru*; Netherlands Antilles*; Niue*; Palau; Panama; San Marino; Samoa; Seychelles; Singapore; St. Kitts and Nevis*; St. Lucia Island*; St. Vincent and the Grenadines*; Switzerland; U.S. Virgin Islands; Uruguay; Vanuatu*. * indicates a “Dot” haven. Listfollows(DyrengandLindsey,2009;HinesandRice,1994), adding Costa Rica as a “dot.” Havens (0/1) Indicator variable equal to 1 if # Havens > 0 Industry First two digits of Form 1120: Schedule K, line 2a IPO Completed Indicator variable equal to 1 for firms that complete an IPO, sourced from SDC Platinum and SEC filings Net PPE Form 1120: Schedule L line 10b(d) Net PPE/Assets Net PPE/Assets Pre-NOL taxable in- Form 1120: line 28 come (loss) Pre-NOL taxable in- Pre-NOL taxable income/prior year Assets (when missing prior year come/Assets Assets, use beginning-of-year assets from Form 1120: Schedule L line 15(b)) 35

Table A1: Variable definitions (continued) Variable Definition Pre-NOL taxable in- Pre-NOL taxable income/Sales come/Sales Pretax financial in- When a firm reports attaching a Schedule M-3 to the Form 1120 (Box come A4): Net income (loss) per income statement of includible corporations (Schedule M-3, Part I, line 11). When a firm does not report attaching a Schedule M-3 or in the years the Schedule M-3 was not utilized: Net income (loss) per books (Schedule M-1, line 1). R&D expense Form 6765: maximum of line 9, line 53 and line 28 Sales Form 1120: line 1c Sales growth Percentage change in Sales from the prior year to the current year Stock option compen- (Schedule M-3, Part III, line 9) sation Taxes paid Form 1120: Schedule L line 31 Total amounts re- ln(Form 5471: Schedule M, line 12(b+c) + $1,000) ceived from U.S. Total amounts paid to ln(Form 5471: Schedule M, line 24(b+c) + $1,000) U.S. Total income-shifting ln(Form 5471: Schedule M, line 15(b+c) + line 16(b+c) + line receipts from U.S. 17(b+c) + line 18(b+c) + line 20(b+c) + $1,000) Total income-shifting ln(Form 5471: Schedule M, line 15(b+c+d) + line 16(b+c+d) + line receipts from U.S. and 17(b+c+d) + line 18(b+c+d) + line 20(b+c+d) + $1,000) foreign Total paid-in capital Form 1120: Schedule L lines 22b(d) + 23(d) 36

Figure 1: Firm counts (a) Withdrawn IPO control group (b) Private firm control group (c) Public firm control group Thefigureshowscountsoffirmsincludedinoursample. AllthreepanelsshowIPOcompletersinnavy(dark). Inorange(light), Figures1a,1b,and1cshowthecontrolgroupsofIPOwithdrawers,privatefirmswithequityinfusions,andpublicfirmswith equityinfusions,respectively. In1a,toprotecttaxpayerprivacy,weblurthecountsofIPOwithdrawersfortheyearsmarked by asterisks: 2003, 2012, 2013, and 2015. Specifically, we calculate the average IPO-withdrawer firm count pooling across all suchyears,andpresentthisaveragevalueforeachyear. AllpanelswerecreatedbytheauthorsusingIPOdatasourcedfrom SDCPlatinum,theSEC’sEdgarplatform,andIRSadministrativetaxdata. 37

Figure 2: Equity issuance and financing trends around IPO filing and equity infusion (a) Withdrawn IPO (b) Private firms (c) Public firms (d) Withdrawn IPO (e) Private firms (f) Public firms The figure presents averages of equity issuance (Figures 2a, 2b, and 2c) and equity financing (Figures 2d, 2e, and 2f) around an IPO filing event or equity infusion event. The charts show amounts for firms that complete an IPO and for three sets of controlfirms: firmsthatfiledforanIPObutlaterwithdrawtheirIPOfiling(2aand2d),privatelyheldfirmsthatissuedequity of comparable magnitudes to the IPO firms and remained private afterwards (2b and 2e), and publicly held firms that issued equityofcomparablemagnitudesandremainedpublic(2cand2f). AmountsforIPOcompletingfirmsareshownwithasolid line;amountsforeachcontrolsampleareshownwithadashedline. Year𝑡istheyearofIPOfilingin2aand2dandistheyear ofequityissuanceintheotherpanels. AllpanelswerecreatedbytheauthorsusingIPOdatasourcedfromSDCPlatinum,the SEC’sEdgarplatform,andIRSadministrativetaxdata. VariablesaredefinedinAppendixA;Section3describesthesample selectionsteps. 38

Figure 3: Domestic tax planning around the IPO (a) Withdrawn IPO (b) Private firms (c) Public firms (d) Withdrawn IPO (e) Private firms (f) Public firms This figure presents results from difference-in-differences regressions studying domestic Book-Tax Differences (Figures 3a, 3b, and 3c) and domestic Cash Effective Tax Rates (Figures 3d, 3e, and 3f) around the time of IPO filing. Firms that complete anIPOarecomparedtothreesetsofcontrolfirms: firmsthatfiledforanIPObutlaterwithdrawtheirIPOfiling(3aand3d), privatelyheldfirmsthatissuedequityofcomparablemagnitudestotheIPOfirmsandremainedprivateafterwards(3band3e), andpubliclyheldfirmsthatissuedequityofcomparablemagnitudesandremainedpublic(3cand3f). Year𝑡istheyearofIPO filinginPanels3aand3dandistheyearofequityissuanceintheotherpanels. Thechartplotsyear-specifictreatmenteffect coefficients (relative to year 𝑡−1) and 95-percent confidence intervals from estimating Equation (2) for firms that complete an IPO and the relevant control sample; see Section 2. All panels were created by the authors using IPO data sourced from SDC Platinum, the SEC’s Edgar platform, and IRS administrative tax data. Variables are defined in Appendix A; Section 3 describesthesampleselectionsteps. 39

Figure 4: Tax haven trends around the IPO (a) Havens (0/1) (b) # Havens ThefigurepresentstrendsinforeignsubsidiaryactivityforfirmsthatcompleteanIPOaswellascontrolfirmsthatreceivedan equity issuance around the time of the equity issuance. Figure 4a shows the average likelihood of having a controlled foreign corporation(CFC)inataxhaven;Figure4bshowstheaveragenumberofCFCsintaxhavens. Bothpanelsdepicteffectsfor firms that complete an IPO (solid line) as well as the constant-private (short dashed line) and constant-public (long dashed line)controlsamples. Year𝑡ineachchartistheyearofequityinfusion. AllpanelswerecreatedbytheauthorsusingIPOdata sourced from SDC Platinum, the SEC’s Edgar platform, and IRS administrative tax data. Variables are defined in Appendix A;Section3describesthesampleselectionsteps. 40

Figure 5: Tax haven use around the IPO (a) Havens (0/1) (b) # Havens This figure presents results from difference-in-differences regressions studying controlled foreign corporation (CFC) presence in tax havens around the time of IPO filing. Dependent variables include an indicator for having a CFC in a tax haven (5a) and the number of tax haven CFCs (5b). Firms that complete an IPO are compared to those that filed for an IPO but later withdraw their IPO filing. Year 𝑡 is the year of IPO filing. The chart plots year-specific treatment effect coefficients (relative toyear𝑡−1)and95-percentconfidenceintervalsfromestimatingEquation(2);seeSection2. Allpanelswerecreatedbythe authors using IPO data sourced from SDC Platinum, the SEC’s Edgar platform, and IRS administrative tax data. Variables aredefinedinAppendixA;Section3describesthesampleselectionsteps. 41

Figure 6: Sophisticated preparer usage (a) IPO completers compared to public and private firms (b) IPO completers as compared to withdrawers The figure presents trends in sophisticated preparer use for firms that complete an IPO as compared to control firms. In Figure6a,thecontrolfirmsincludeprivateandpublicfirmsthatissuedequityofcomparablemagnitudestotheIPOfirmsand remainedprivateorpublicafterwards. In6b,thecontrolfirmsareIPOwithdrawers. Year𝑡istheyearofequityissuancein6a and is the year of filing in 6b. Both panels depict the average likelihood of using a sophisticated tax prepared, defined as an accountingfirmthatpreparedover4,000taxreturnsduringtheyearfollowingZwick(2021). Bothpanelswerecreatedbythe authors using IPO data sourced from SDC Platinum, the SEC’s Edgar platform, and IRS administrative tax data. Variables aredefinedinAppendixA;Section3describesthesampleselectionsteps. 42

airetirc elpmaS :1 elbaT noitcelesOPI:AlenaP stneveOPI noitcarF stnevE gnitelpmoc gniniamer OPI 72.0 541,22 5102-6991sraeygnilfi,munitalPCDSmorfsOPI.S.U 52.0 537,71 seirtsudniseitilitu/ecnanfiedulcxE 06.0 648,5 ,sRDA,STIER,sdnufdne-desolc,sreffotinuedulcxE 06.0 147,5 sffonipsdna,seinapmockcehcknalb,sPL 85.0 455,5 kcotsnommocnonfoecnaussiedulcxE 76.0 308,3 ragdECESmorfetadgnilfiOPIdnaNIEevresbO 76.0 395,3 mrfireptneveOPItsrfiniateR 76.0 395,3 atadxat.S.UhtiwegremotsOPIfoelpmaslaniF selpmaslortnocdnatnemtaertrofatadxaT:BlenaP elpmasmrficilbuP elpmasmrfietavirP elpmasOPInwardhtiW elpmasOPIdetelpmoC sraey-mriF smriF sraey-mriF smriF sraey-mriF smriF sraey-mriF smriF gniniamer gniniamer gniniamer gniniamer gniniamer gniniamer gniniamer gniniamer selpmasOPInwardhtiwdnadetelpmoC 281,1 114,2 evobaelpmasOPIehtgnitneserpersmriF 055,6 148 589,12 521,2 atadxatIOSothctaM 137,5 838 589,12 521,2 OPIdetelpmocretfa .sborewardhtiw-OPIporD 915,5 518 306,12 701,2 sutatsnoitaroproc-CeriuqeR 497,4 687 912,91 420,2 stessalatotdnastpiecerssorgevitisoperiuqeR selpmasmrficilbupdnaetavirP 075,96 177,01 748,867 349,751 8102-4002,elpmasIOSnisnoitaroprocCllA 008,26 837,9 872,036 296,721 stessalatotdnastpiecerssorgevitisoperiuqeR 596,74 294,7 183,205 498,401 seirtsudniseitilitu/ecnanfiedulcxE 508,02 239,1 148,13 454,3 airetircruogniteemecnaussiytiuqeeriuqeR 200,02 )dettimo( 175,03 )dettimo( degnahcsutatsetavirp/cilbupfi .sboporD selpmasllA 599,91 196,1 625,03 509,2 567,4 567 771,91 099,1 noitailffiayrtsudnieriuqeR 599,91 196,1 625,03 509,2 567,4 567 125,81 419,1 )srewardhtiwon(trohocOPI6991porD 599,91 196,1 625,03 509,2 482,2 822 964,8 276 1+𝑡ot2−𝑡tneserpsmrfieriuqeR 410,11 196,1 885,81 509,2 663,1 822 713,4 276 3+𝑡ot3−𝑡sraeynisnoitavresboniateR 410,11 196,1 885,81 509,2 663,1 822 713,4 276 selpmaslaniF dna ,elpmas mrfi etavirp eht ,elpmas OPI nwardhtiw eht :selpmas lortnoc eerht eht dna sOPI detelpmoc fo elpmas tnemtaert eht rof spets noitceles eht stneserp elbat sihT avokloVdnayleahciM,yrwoLgniwollofmunitalPCDSmorf5102ot6991morfsmrfilaicnanfi-nonrofsOPI.S.UyfitnediotspetsehtstneserpAlenaP .elpmasmrficilbupeht cilbupdnaetavirpforebmunehtsallewsa,atadxatemocnietaroproc.S.UehthtiwegremtahtsOPInwardhtiwdnadetelpmoctcnitsidforebmunehtswohsBlenaP .)7102( lenaP .3 noitceS ni debircsed sa ,selpmas lortnoc mrfi cilbup dna etavirp eht etaerc ot raey nevig a ni ecnaussi ytiuqe rof tnemeriuqer dlohserht muminim eht gniteem smrfi ataD .ycavirp reyapxat tcetorp ot dettimo era stnuoc owT .snoitcirtser elpmas etisiuqer gnisopmi retfa snoitavresbo raey-mrfi dna smrfi fo rebmun eht htob stneserp osla B .snoitaluclac’srohtuadna,ecivreSeuneveRlanretnIehtfonoisividemocnIfoscitsitatSeht,munitalPCDSerasecruos 43

Table 2: Summary statistics Samples used in empirical analyses IPO completing & IPO completing & IPO completing & withdrawing firms private firms public firms Obs. Mean Median Obs. Mean Median Obs. Mean Median Outcome variables (1) (2) (3) (4) (5) (6) (7) (8) (9) BTD 5,683 -0.068 -0.021 20,346 -0.032 -0.013 12,975 -0.016 0 Cash effective tax rate 2,006 0.204 0.172 8,252 0.194 0.118 6,954 0.190 0.162 Stock option compensation BTD 2,972 -0.005 0.000 19,606 -0.005 0.000 12,576 0.000 0.000 # Havens 2,960 0.788 0.000 19,448 0.771 0.000 12,570 3.142 0.000 Havens (0/1) 2,960 0.293 0.000 19,448 0.271 0.000 12,570 0.420 0.000 # Dot havens 2,960 0.217 0.000 19,448 0.232 0.000 12,570 1.110 0.000 Cost sharing agreement (0/1) 5,683 0.012 0.000 20,346 0.022 0.000 12,975 0.040 0.000 Total amounts received from U.S. 1,551 5.668 0.000 10,036 5.049 0.000 6,540 6.979 0.000 Total amounts paid to U.S. 1,551 4.966 0.000 10,036 5.246 0.000 6,540 9.095 13.546 Total income-shifting receipts from U.S. 1,551 4.450 0.000 10,036 3.786 0.000 6,540 4.808 0.000 from U.S. & foreign 1,551 5.056 0.000 10,036 4.465 0.000 6,540 6.460 0.000 Other firm characteristics Equity issuance 5,683 30.952 5.209 20,346 41.998 6.064 12,975 85.855 10.786 Equity financing 5,683 0.147 0.040 20,307 0.121 0.026 12,965 0.088 0.020 Pretax financial income 5,683 -8,483 -4,847 20,346 -3,623 -4,619 12,975 40,151 6,400 Federal income tax per books 5,683 2,527 0.000 20,346 3,815 0.000 12,975 23,182 1,352 Pre-NOL taxable income 5,683 -3,825 -3,544 20,346 1,822 -3,834 12,975 60,621 9,936 Taxes paid 5,683 1,775 0.000 20,346 2,982 0.000 12,975 17,563 188 BTD, numerator 5,683 -2,904 -1,046 20,346 -1,864 -1,466 12,975 879 -129 BTD tax, numerator 5,683 361 0.000 20,346 -253 0.000 12,975 2,957 0.000 Total assets 5,683 303,379 96,250 20,346 720,991 189,630 12,975 2,231,565 782,597 Asset growth 5,683 0.512 0.149 20,346 0.297 0.076 12,975 0.206 0.057 ln(Total assets) 5,683 11.553 11.475 20,346 12.294 12.153 12,975 13.553 13.570 Sales 5,683 229,053 72,305 20,346 471,130 139,950 12,975 1,375,872 482,849 Sales growth 5,197 0.668 0.219 18,645 0.400 0.141 12,050 0.201 0.110 ln(Sales) 5,683 11.037 11.189 20,346 11.720 11.849 12,975 12.888 13.087 Net PPE/Assets 5,683 0.148 0.098 20,346 0.144 0.084 12,975 0.168 0.102 Debt/Assets 5,683 0.242 0.131 20,346 0.283 0.177 12,975 0.286 0.233 Pre-NOL taxable income/𝐴𝑠𝑠𝑒𝑡𝑠 5,683 -0.227 -0.041 20,346 -0.153 -0.019 12,975 -0.031 0.016 𝑡−1 Pre-NOL taxable income/Sales 5,683 -1.135 -0.051 20,346 -0.597 -0.025 12,975 -0.252 0.023 Age 5,683 8.420 7.000 20,346 11.283 9.000 12,975 21.774 15.000 ln(Age) 5,683 2.062 2.079 20,346 2.249 2.303 12,975 2.827 2.773 Uses a sophisticated tax preparer 2,524 0.533 1.000 16,061 0.505 1.000 8,609 0.554 1.000 The table presents unweighted counts as well as weighted means and medians for selected variables. The weights are inverse probability weights based onobservable characteristics in year 𝑡−1. Columns(1), (2), and (3)present statisticsfor the sampleof IPO completingand withdrawing firmsfrom1994-2018(forIPOsfiledfrom1997to2015). Columns(4),(5),and(6)presentstatisticsforthesampleofIPOcompletingandprivate firmsfrom2004-2018(forIPOsfiledfrom2006to2015). Columns(7),(8),and(9)presentstatisticsforthesampleofIPOcompletingandpublic firms from 2004-2018 (for IPOs filed from 2006 to 2015). Variables are defined in Appendix A. All dollar-denominated variables are reported in thousands. Data sources are SDC Platinum, the Statistics of Income division of the Internal Revenue Service, and authors’ calculations. All percentileestimatesareaveragesofthetenobservationsaroundthepercentilecutofftopreservetaxfilingconfidentiality. 44

Table 3: Pre-IPO-filing firm characteristics, with IPW weights Completed IPO Control firm Diff. in means Obs. Mean Std. Dev Obs. Mean Std. Dev Diff. P-value (1) (2) (3) (4) (5) (6) (7) (8) Panel A: Control sample, withdrawn IPOs (1994-2018) Assets 672 224,324 394,437 228 227,681 382,144 -3,357 0.83 Asset growth 672 0.66 0.95 228 0.63 0.92 0.03 0.68 ln(Assets) 672 11.25 1.42 228 11.28 1.45 -0.03 0.68 Sales 672 176,609 297,450 228 210,325 331,347 -33,716 0.25 Sales growth 672 1.61 3.08 228 1.55 2.91 0.06 0.99 ln(Sales) 672 10.78 1.82 228 10.87 1.91 -0.09 0.34 Net PPE/Assets 672 0.16 0.14 228 0.15 0.13 0.01 0.37 Debt/Assets 672 0.27 0.29 228 0.26 0.27 0.01 0.95 Pre-NOL taxable income/𝐴𝑠𝑠𝑒𝑡𝑠 672 -0.28 0.58 228 -0.29 0.60 0.02 0.54 𝑡−1 Pre-NOL taxable income/Sales 672 -1.10 2.64 228 -0.96 2.45 -0.13 0.38 Age 672 7.26 5.09 228 7.38 5.19 -0.12 0.95 ln(Age) 672 1.93 0.62 228 1.93 0.64 0.00 0.95 Panel B: Control sample, private firms (2004-2018) Assets 321 600,330 1,108,623 2,905 555,809 1,011,781 44,522 0.78 Asset growth 321 0.24 0.46 2,905 0.20 0.48 0.04 0.26 ln(Assets) 321 12.00 1.57 2,905 11.99 1.61 0.01 0.88 Sales 321 384,344 698,702 2,905 414,605 687,863 -30,261 0.69 Sales growth 321 0.63 1.33 2,905 0.72 1.60 -0.09 0.08 ln(Sales) 321 11.49 1.80 2,905 11.57 1.91 -0.07 0.79 Net PPE/Assets 321 0.15 0.15 2,905 0.16 0.17 -0.01 0.34 Debt/Assets 321 0.37 0.34 2,905 0.35 0.35 0.01 0.74 Pre-NOL taxable income/𝐴𝑠𝑠𝑒𝑡𝑠 321 -0.17 0.38 2,905 -0.17 0.40 0.00 0.90 𝑡−1 Pre-NOL taxable income/Sales 321 -0.67 1.60 2,905 -0.62 1.66 -0.06 0.74 Age 321 8.89 6.87 2,905 11.47 10.72 -2.58 0.00 ln(Age) 321 2.10 0.61 2,905 2.17 0.86 -0.07 0.46 Panel C: Control sample, public firms (2004-2018) Assets 321 1,887,677 2,115,738 1,691 1,877,226 3,137,368 10,451 0.50 Asset growth 321 0.20 0.41 1,691 0.22 0.41 -0.02 0.75 ln(Assets) 321 13.42 1.71 1,691 13.22 1.64 0.20 0.89 Sales 321 1,303,390 1,614,595 1,691 1,211,548 2,042,667 91,841 0.70 Sales growth 321 0.35 0.53 1,691 0.31 0.62 0.04 0.20 ln(Sales) 321 12.95 1.86 1,691 12.57 1.98 0.38 0.47 Net PPE/Assets 321 0.19 0.19 1,691 0.15 0.16 0.04 0.36 Debt/Assets 321 0.37 0.30 1,691 0.25 0.25 0.12 0.21 Pre-NOL taxable income/𝐴𝑠𝑠𝑒𝑡𝑠 321 -0.01 0.22 1,691 -0.04 0.21 0.02 0.68 𝑡−1 Pre-NOL taxable income/Sales 321 -0.16 0.77 1,691 -0.25 0.84 0.08 0.48 Age 321 22.59 18.74 1,691 18.48 14.99 4.12 0.61 ln(Age) 321 2.81 0.87 1,691 2.69 0.76 0.12 0.97 ThetablecomparesfirmcharacteristicsforIPO-completingfirmsrelativetothethreecontrolsamplesafterinverseprobabilityweighting. InPanels A, B, and C, the control samples are IPO-withdrawing firms, private firms, and public firms, respectively. Panel A describes firms in the year priortoIPOfiling,whileinPanelsBandCitistheyearpriortoequityissuance. Columns(1),(2),and(3)shownumbersofobservations,mean values,andstandarddeviations,respectively,forthecompletedIPOsample. Columns(4),(5),and(6)showthesameinformationforthecontrol samples. Columns (7) and (8) give the difference in means and a within-cohort p-value for that difference. Variables are defined in Appendix A. DatasourcesareSDCPlatinum,theStatisticsofIncomedivisionoftheInternalRevenueService,andauthors’calculations. 45

gninnalp xat citsemod OPI-tsoP :4 elbaT )sRTE( setar xat evitceffe hsac citsemod dna )sDTB( secnereffid xat-kooB :A lenaP sRTE hsaC sDTB :elbairav tnednepeD cilbuP etavirP nwardhtiW nwardhtiW cilbuP etavirP nwardhtiW nwardhtiW smrfi smrfi smrfi OPI smrfi OPI smrfi smrfi smrfi OPI smrfi OPI )81’-40’( )81’-40’( )81’-40’( )81’-49’( )81’-40’( )81’-40’( )81’-40’( )81’-49’( )8( )7( )6( )5( )4( )3( )2( )1( **930.0- ***560.0- ***521.0- ***380.0- **120.0 ***130.0 ***840.0 ***440.0 tsoP × detelpmoC OPI ]642.2-[ ]623.5-[ ]777.2-[ ]034.3-[ ]851.2[ ]120.3[ ]750.4[ ]868.3[ + + + + + + + + sEF trohoC × raeY + + + + + + + + sEF mriF 459,6 252,8 038 600,2 579,21 643,02 084,2 386,5 snoitavresbO 295.0 965.0 375.0 765.0 483.0 663.0 543.0 724.0 derauqs-R sDTB fo noitisopmoceD :B lenaP noitasnepmoc noitpo kcotS :elbairav tnednepeD cilbuP etavirP nwardhtiW smrfi smrfi smrfi OPI )81’-40’( )81’-40’( )81’-40’( )3( )2( )1( ***800.0 ***300.0 ***610.0 tsoP × detelpmoC OPI ]352.6[ ]885.3[ ]525.4[ + + + sEF trohoC × raeY + + + sEF mriF 675,21 606,91 034,2 snoitavresbO 324.0 045.0 714.0 derauqs-R raey radnelac a ot sdnopserroc trohoc hcae erehw ,strohoc dekcats no detamitse si )1( noitauqE .ytivitca gninnalp xat citsemod mrfi no stluser noisserger stneserp elbat sihT )1( snmuloC( stessa deggal fo erahs a sa secnereffid xat-koob citsemod :gninnalp xat citsemod fo serusaem owt rof stluser stneserp A lenaP .gnilfi OPI ro ecnaussi ytiuqe fo .stessadeggalfoerahsasasecnereffidxat-koobnoitasnepmoc-noitpo-kcotsrofstluserstneserpBlenaP .))8(ot)5(snmuloC(etarxatevitceffehsaccitsemodehtdna,))4(ot morf atad htiw ,dedulcni era smrfi OPI nwardhtiw ,)5( dna )1( snmuloC nI .desu era sthgiew ytilibaborp esrevni dna dedulcni era smrfi gnitelpmoc-OPI ,snoisserger lla nI dna )4( snmuloC nI .8102-4002 morf atad htiw ,dedulcni era smrfi etavirp ,)7( dna )3( snmuloC nI .8102-4002 ot detcirtser era atad eht ,)6( dna )2( snmuloC nI .8102-4991 ytiuqeehtdnuora3+𝑡ot3−𝑡morfsnoitavresbosedulcnitrohochcaedna,trohocenoylnonidedulcnisimrfihcaE .8102-4002morfatadhtiw,dedulcnierasmrficilbup,)8( nidetropererascitsitats-T .levelyrtsudnirojamSRIehttaderetsulcerasrorredradnatS .stceffedexfitrohoc-yb-raeydnamrfiedulcnisnoissergeR .raeygnilfiOPIroecnaussi ,munitalPCDSerasecruosataD .ylevitcepser,ecnacfiingistnecrep01dna,tnecrep5,tnecrep1fosleveletacidni*dna**,***.AxidneppAnidenfiederaselbairaV .stekcarb .snoitaluclac’srohtuadna,ecivreSeuneveRlanretnIehtfonoisividemocnIfoscitsitatSeht 46

Table 5: Post-IPO foreign tax planning and activity Panel A: Haven ownership Dependent variable: Havens (0/1) # Havens # Dot havens (1) (2) (3) IPO Completed × Post 0.161*** 0.266** 0.101*** [6.360] [2.548] [2.967] Year × Cohort FEs + + + Firm FEs + + + Observations 2,394 2,394 2,394 R-squared 0.766 0.884 0.915 Panel B: Cross-border activity Total income Cost sharing Total amounts Total amounts Total income shifting receipts Dependent variable: agreement received paid shifting receipts from U.S. & (0/1) from U.S. to U.S. from U.S. Foreign (1) (2) (3) (4) (5) IPO Completed × Post 0.038*** 2.179** 1.061** 2.320*** 2.199** [6.165] [2.508] [2.453] [2.768] [2.558] Year × Cohort FEs + + + + + Firm FEs + + + + + Observations 2,480 1,249 1,249 1,249 1,249 R-squared 0.426 0.801 0.793 0.787 0.806 This table presents regression results on firm foreign tax planning activity. Equation (1) is estimated on stacked cohorts, where each cohort corresponds to a calendar year of IPO filing. Panel A presents results for tax haven ownership. Panel B presents results for cross-border related-party agreements and payments. In all regressions, IPO-completing and IPO-withdrawing firms are included, with data from 2004 to 2018, and inverse probability weights are used. Each firm is included in only one cohort, and each cohort includes observations from 𝑡−3 to 𝑡+3 around the IPO filing year. Regressions include firm and year-by-cohort fixed effects. Standard errors areclusteredattheIRSmajorindustrylevel. T-statisticsarereportedinbrackets. VariablesaredefinedinAppendixA.***,**and* indicatelevelsof1percent,5percent,and10percentsignificance,respectively. DatasourcesareSDCPlatinum,theStatisticsofIncome divisionoftheInternalRevenueService,andauthors’calculations. 47

Table 6: Heterogeneity in foreign tax planning Panel A: Haven ownership Dependent variable: Havens (0/1) High Low Switched to Didn’t switch to Partition: Domestic Domestic sophisticated sophisticated R&D R&D preparer preparer (1) (2) (3) (4) IPO Completed × Post 0.214*** 0.133*** 0.308*** 0.110** [5.168] [4.322] [5.853] [2.347] Year × Cohort FEs + + + + Firm FEs + + + + Observations 1,169 1,225 277 1,966 R-squared 0.761 0.783 0.839 0.759 Panel B: Cross-border activity Dependent variable: Cost sharing agreement (0/1) High Low Switched to Didn’t switch to Partition: Domestic Domestic sophisticated sophisticated R&D R&D preparer preparer (1) (2) (3) (4) IPO Completed × Post 0.045*** 0.029*** 0.083*** 0.029*** [3.735] [2.990] [3.470] [3.281] Year × Cohort FEs + + + + Firm FEs + + + + Observations 1,224 1,256 288 2,033 R-squared 0.483 0.389 0.529 0.422 This table presents regression results on heterogeneity in firm foreign tax planning activity. Equation (1) is estimated on stacked cohorts, where each cohort corresponds to a calendar year of IPO filing. In Panels A and B, the dependent variables are indicators for owning a subsidiary in a tax haven and having a cost-sharing agreement in place with a foreign subsidiary, respectively. In all regressions, IPO-completing and IPO-withdrawing firms are included, with data from 2004 to 2018, and inverse probability weights are used. In each panel, Column (1) presents results for the subsample of firms with an abovemedianvalueofaveragedomesticR&Dexpendituresinthepre-IPO-filingyears(t-3tot-1);Column(2)restrictstofirmswith a below-median value. Similarly, Column (3) presents results for the subsample of firms that switched to a sophisticated tax advisorinayearleadinguptoorincludingtheIPOfilingyear(t-3tot);Column(4)restrictstofirmsthatdidnot. Eachfirm isincludedinonlyonecohort,andeachcohortincludesobservationsfrom𝑡−3to𝑡+3aroundtheIPOfilingyear. Regressions include firm and year-by-cohort fixed effects. Standard errors are clustered at the IRS major industry level. T-statistics are reportedinbrackets. VariablesaredefinedinAppendixA.***,**and*indicatelevelsof1percent,5percent,and10percent significance, respectively. Data sources are SDC Platinum, the Statistics of Income division of the Internal Revenue Service, andauthors’calculations. 48

sisylanA ssentsuboR :7 elbaT 8102-4002 elpmas lortnoc smrfi OPI nwardhtiw ,secnereffid xat-kooB :A lenaP sDTB :elbairav tnednepeD desab-xaT ot timiL elacS ezirosniW eriuqeR htiW gniretsulC dethgiewnU enilesaB sDTB srelfi 3-M selas yb %1 ta 𝑡 dna 1-𝑡 slortnoc mrfi yb setamitse noitacfiiceps )9( )8( )7( )6( )5( )4( )3( )2( )1( *300.0 ***340.0 **430.0 **950.0 ***840.0 ***840.0 ***840.0 ***840.0 ***840.0 tsoP × detelpmoC OPI ]537.1[ ]889.3[ ]971.2[ ]466.2[ ]753.4[ ]057.3[ ]726.3[ ]403.3[ ]750.4[ - - - - - + - - selbairav lortnoC + + + + + + + + + seF trohoC × raeY + + + + + + + + + sEF mriF 084,2 034,2 084,2 084,2 421,3 084,2 084,2 084,2 084,2 snoitavresbO 982.0 843.0 254.0 582.0 293.0 353.0 543.0 253.0 543.0 derauqs-R 8102-4002 elpmas lortnoc smrfi OPI nwardhtiW ,ecneserp nevaH :B lenaP )1/0( snevaH :elbairav tnednepeD eriuqeR htiW gniretsulC dethgiewnU enilesaB 𝑡 dna 1−𝑡 slortnoc mrfi yb setamitse noitacfiiceps )5( )4( )3( )2( )1( ***431.0 ***641.0 ***161.0 ***811.0 ***161.0 tsoP × detelpmoC OPI ]208.6[ ]923.5[ ]223.4[ ]598.3[ ]063.6[ - + - - selbairav lortnoC + + + + + sEF trohoC × raeY + + + + + sEF mriF 020,3 493,2 493,2 493,2 493,2 snoitavresbO 577.0 177.0 667.0 367.0 667.0 derauqs-R 49

)deunitnoc( sisylanA ssentsuboR :7 elbaT sesylana 283 noitceS :C lenaP smrfi detimil 283 gnipporD elpmas lluF :elpmaS sDTB elbawolla xam naht ssel noitcuded LON :elbairav tnednepeD cilbuP etavirP nwardhtiW nwardhtiW cilbuP etavirP nwardhtiW nwardhtiW smrfi smrfi smrfi smrfi smrfi smrfi smrfi smrfi )81’-40’( )81’-40’( )81’-40’( )81’-49’( )81’-40’( )81’-40’( )81’-40’( )81’-49’( )8( )7( )6( )5( )4( )3( )2( )1( *910.0 ***720.0 ***240.0 ***040.0 520.0 810.0 400.0 210.0 tsoP × detelpmoC OPI ]319.1[ ]396.2[ ]564.3[ ]734.3[ ]572.1[ ]781.1[ ]941.0[ ]167.0[ + + + + + + + + seF trohoC × raeY + + + + + + + + sEF mriF 725,01 542,81 672,2 771,5 579,21 643,02 084,2 386,5 snoitavresbO 593.0 763.0 943.0 724.0 046.0 865.0 885.0 145.0 derauqs-R stneserp A lenaP .gnithgiew ytilibaborp esrevni retfa selpmas lortnoc eerht eht ot evitaler smrfi gnitelpmoc OPI eht rof scitsiretcarahc mrfi fo nosirapmoc stneserp elbat sihT raey eht ni scitsiretcarahc stneserp B lenaP .OPI eht wardhtiw taht smrfi fo tes lortnoc eht dna OPI na etelpmoc taht smrfi rof gnilfi OPI ot roirp raey eht ni scitsiretcarahc ni scitsiretcarahc stneserp C lenaP .ytiuqe deussi taht smrfi etavirp fo tes lortnoc eht dna smrfi gnitelpmoc OPI fo elpmas eht rof ecnaussi ytiuqe ro noitelpmoc OPI ot roirp fo)3(dna,)2(,)1(snmuloC .ytiuqedeussitahtsmrficilbupfoteslortnocehtdnasmrfignitelpmocOPIfoelpmasehtrofecnaussiytiuqeronoitelpmocOPIotroirpraeyeht lenaphcaefo)6(dna,)5(,)4(snmuloC .elpmasOPIdetelpmocehtrof,ylevitcepser,snoitaiveddradnatsehtdna,seulavnaemeht,snoitavresboforebmunehtwohslenaphcae nmuloCdna,snaemniecnereffidehtstneserp)7(nmuloC .selpmaslortnocehtrof,ylevitcepser,snoitaiveddradnatsehtdna,seulavnaemeht,snoitavresboforebmunehtwohs fo scitsitatS eht,munitalP CDS era secruos ataD .A xidneppA ni denfied era selbairaV .snaem elpmas owt eht neewteb ecnereffid eht rof eulav-p trohoc-nihtiw a stneserp )8( .snoitaluclac’srohtuadnaecivreSeuneveRlanretnIehtfonoisividemocnI 50