Investor valuations of Japan’s adoption of a territorial tax regime: quantifying the direct and competitive effects of international tax reform

For a description of optimal international tax systems, including the implications of international taxation for capital export, import, and ownership neutrality, see, e.g., Musgrave (1969), Desai and Hines (2003) or Devereux (2008).

Ten OECD countries have adopted territorial tax systems since 2000 (Dittmer 2012), including Iceland (2003), the Czech Republic, Norway, Slovakia (2004), Estonia, Turkey (2005), Poland (2007), Japan, New Zealand, and the UK (2009). Only seven OECD countries still have worldwide tax systems: Chile, Greece, Ireland, Israel, Korea, Mexico, and the US. Brys et al. (2011) describe several such recent reforms, the net effect of which has been that the US’s share of real GDP among OECD countries with worldwide tax systems increased from 56.4 to 78.2% between 2005 and 2010.

Toder (2014) indicates that the timing of the Japanese reform was heavily affected by near-simultaneous passage of a similar law in the UK: “the United Kingdom’s adoption of its territorial system when it did may have been a tipping point, because Japanese policymakers always follow what is happening in other countries. They periodically send study groups out from the government [\(\ldots \)] These groups return to Japan, they draw up their comparison tables, and then consider what other countries are doing and why. They typically look at the United States, the United Kingdom, France, Germany, and a few other jurisdictions” (pp. 23–24).

See Egger et al. (2015), Feld et al. (2016) and Liu (2014) for consideration of the effects of territorial regime adoption on foreign investment.

For additional applications of event study methods to quantifying the perceived benefits of tax avoidance, see Desai and Dharmapala (2007), Hanlon and Slemrod (2009) or Bradley (2013). Sakurada and Nakanishi (2011) also examine investor reactions to news of the Japanese dividend exemption, albeit for only a single event date and a small selected sample of large Japanese MNCs.

Germany’s territorial tax regime was adopted in 1920, and current law (in place since 2001) features a 95% dividend exemption similar to the system adopted in Japan, albeit with a lower corporate tax rate (i.e., 30.2% since 2008 versus 40.69% in Japan during this time period). Germany’s economy is also closest in size among all territorial regimes to the Japanese and US economies and thus presents a natural comparison group. Closer to Japan, only Singapore and Australia had territorial tax systems in 2008, but their economies are arguably very different.

Results for the 47 German firms included in our analyses provide no credible evidence of negative investor reactions surrounding the Japanese events. The main purpose of incorporating German data is to better account for global financial market developments. Full results showing German stock market responses are available from the authors upon request.

According to Japanese tax practitioners, there should have been no remaining doubts beyond the final January 2009 Cabinet meeting with respect to both passage of the proposal into law 2 months later and the details of the new policy: “draft legislation is delivered to the Cabinet in January for review and approval. The final legislation is completed in February. It’s delivered to the National Diet usually during late February. And, in March, it is virtually always passed to go into effect as of April 1” (Toder 2014, p. 25).

Including our historical estimation period, publicly listed Japanese and US firms of all types experienced cumulative losses of 40–60% of stock market capitalization between April 2007 and April 2009. Identification hence rests on our ability to distinguish event-driven market reactions from a profoundly negative and unstable market environment.

By comparison, US multinationals repatriated a total of $362 billion (about 2.9% of 2006 US GDP) under the terms of the American Job Creation Act’s temporary 85% dividends received deduction over the period 2004–2006, most of which occurred in 2005 (Redmiles 2008).

See, for example, Toder (2014, p. 24) and Altshuler et al. (2015, pp. 24–25) or Takashima (2009) who argue that (i) many Japanese corporations lack a full awareness of the importance of international tax and accounting strategies and thus tend to bear extra tax costs that they could otherwise avoid, (ii) Japanese companies tend to assume that taxes are unavoidable and are to be paid to the government as a matter of loyalty or patriotism, and (iii) Japanese corporations lack sufficient human resources for tax planning.

Among the set of foreign subsidiaries established between 2012 and 2015 by formerly domestic-only firms, those owned by Japanese parents were more than 5 times more likely to be located in tax havens than those owned by US parents.

Expenses corresponding to these 5% of repatriated dividends are assumed to be deducted from the taxable incomes of parent firms at the time of investment, and thus are not exempted upon repatriation under the new exemption system.

For more details about the reform, see, e.g., PricewaterhouseCoopers (2009), Deloitte Touche Tohmatsu (2009), or http://​japantax.​org/​?​p=​590.

Because of this situation, referred to as the “twisted” Diet, the passage of the bill for the tax reform of 2008 was delayed from April 1, 2008, to April 30, for example.

See MacKinlay (1997), Binder (1998) or Corrado (2011) for reviews of differing event study methodologies and associated statistical issues.

This approach allows simultaneous estimation of firm fixed effects and market co-movement parameters alongside event date-specific interaction effects involving firm- or industry-level characteristics.

To see this, consider the simplest case with a 2-period event window. Equation (3) may thus be rewritten (suppressing the idiosyncratic error term for brevity) as \(r_{it} = \alpha _i + \beta _i {\mathbf {R}_{\mathbf{t}}} + \mathrm{AR}_{i1}\cdot W_1 + \mathrm{AR}_{i2}\cdot W_2\), with \(W_s=1\) on date s and 0 otherwise. By definition of cumulative abnormal returns, \(\mathrm{CAR}_1 = \mathrm{AR}_1\) and \(\mathrm{AR}_s=\mathrm{CAR}_s-\mathrm{CAR}_{s-1}\) for all subsequent dates in the event window, such that this last expression can be transformed into a function of CARs only: \(r_{it} = \alpha _i + \beta _i {\mathbf {R}_{\mathbf{t}}} + \mathrm{CAR}_{i1}\cdot W_1 + \left( \mathrm{CAR}_{i2}-\mathrm{CAR}_{i1}\right) \cdot W_2=\alpha _i + \beta _i {\mathbf {R}_{\mathbf{t}}} + \mathrm{CAR}_{i1}\cdot \left( W_1-W_2\right) + \mathrm{CAR}_{i2}\cdot W_2.\) Specifying \(D_1\equiv W_1-W_2\) and \(D_2\equiv W_2\) completes the desired transformation.

Smith et al. (1986) estimate Eq. (3) as a system of equations in order to address potential cross-sectional correlation among firm ARs. Lack of contemporaneous (daily) variation in firm characteristics precludes our ability to employ such a technique, which would moreover be constrained by limits on the number of cross-equation restrictions that may be imposed in order to recover average CARs for a large sample of publicly traded firms. The most popular approach to addressing cross-sectional correlation in the event study literature—estimation of aggregate portfolios of stock returns (Kolari and Pynnönen 2010)—assumes away the possibility of heterogeneous policy effects and is consequently equally uninteresting for our purposes. Our panel estimation approach can instead be viewed as a hybrid of these techniques, whereby conditioning on firm characteristics may be viewed as yielding a set of flexibly defined portfolios and should as such largely mitigate—if not eliminate—concerns associated with cross-sectional correlation.

Historically, an additional econometric concern in event studies has been the issue of nonsynchronous trading, whereby the timing of realized market returns and individual stock returns differ. Scholes and Williams (1977) show, for example, that this can yield biased and inconsistent estimates of the degree of co-movement with the market, with the direction of the bias depending on the relative frequency of trading. An extension of this is unavoidable in the present context. Brown and Warner (1985) present evidence that this does not preclude valid inference in the case of the basic market model.

For reasons owing partly to data availability, we are unable to credibly identify the existence of any such negative impacts on German firm valuations. We cannot consequently distinguish firm competitiveness effects from other sources of remaining variation in ARs, despite our intent.

Datastream does not provide information on ex-dividend returns. Returns based on changes in market capitalization may therefore be influenced by dividend payouts. For this reason (among others) outlying stock market returns and corresponding abnormal return estimates derived from the standard market model are winsorized to the 1st and 99th percentile values from their respective daily distributions. Basic tests suggest only modest sensitivity to the choice of cutoff or outright exclusion of outlying return observations.

Treasury bill rates are courtesy of Kenneth French and interpolated over US stock market holidays for the purposes of stripping out risk-free returns for Japanese and German listings.

We also include analogous information for German parent corporations. The resulting matched sample of publicly traded German firms with complete financial statement information is roughly only one tenth the size of our Japanese or US samples, thereby hindering inference involving German firms. For expositional purposes, we exclude Germany from further discussion.

Due to the static nature of Bureau van Dijk’s ownership database, these links are based exclusively on fiscal 2012 ownership information from the time that we accessed the data. We are constrained to treat firm ownership structures as though these were not systematically affected by the Japanese reform in the 3 years following enactment.

We exclude finance and real estate firms from our analysis because of their distinct tax treatment and special sensitivity to market events over the 2008–2009 period.

See Appendix C for further details on variable construction.

This designation includes MNCs whose foreign operations are deemed immaterial relative to their domestic operations for financial reporting purposes. “Domestic” firms in our sample may therefore include companies with small foreign operations. Conversely, the use of 2012 ownership information due to data limitations implies that certain multinationals may not yet have had foreign operations in 2009. Both sources of mismeasurement will tend to attenuate differences in investor responses vis-à-vis domestic versus multinational firms.

We describe various alternative measures of foreign tax rates in Appendix C.1, which aim to address different concerns associated with possible missing data (e.g., the lack of detailed coverage of minority-owned subsidiaries) and speak to different valuation channels depending on investor expectations about future tax avoidance. Our preferred definition, \(\mathrm{TS} = \mathrm{AETR}\_\mathrm{DOM}-\mathrm{AETR}\_\mathrm{FOR}\), thus reflects the average cost of repatriating pre-reform earnings and implicitly assumes an unchanged future allocation of reported profits between foreign subsidiaries. In contrast, computing tax savings on the basis of the lowest statutory tax rate facing any foreign subsidiary within the MNC group, for example, effectively assumes the adoption of a more aggressive tax planning strategy, whereby future foreign profits are reallocated entirely to the parent’s lowest-taxed subsidiary.

Redmiles (2008) and Gravelle and Marples (2011) report that approximately one half of earnings repatriated under the American Jobs Creation Act of 2004’s temporary tax holiday originated in the pharmaceutical and other high technology sectors and were repatriated from predominantly low-tax countries or tax havens. More generally, Altshuler and Grubert (2003) find that about half of the difference in profitability between high- and low-tax countries is due to transfers of intellectual property.

Our preferred measure is adapted from Corrado et al. (2005) and Chen and Dauchy (2017) and is based on aggregate investment in intangible and tangible assets for 19 two-digit industries. See Appendix D for a description of this and alternative proxies for intangible intensity. Measures based on industry-level asset stocks, or defined as asset-weighted averages across subsidiary industries yield qualitatively similar results. Due to the limitations of financial statement information, all such measures are superior to the use of firm-level measures, which necessarily capture only acquired intangibles and may be a noisy proxy for firm-level intangible intensity (Lev 2003a, b).

An alternate measure of tax savings based on MNCs’ lowest subsidiary tax rate similarly reflects opportunities for future income reallocation (exploiting existing MNC structures) and yields qualitatively similar results. See Appendix C.1 and Appendix F.

Use of an alternate measure of tax haven presence based on the share of all foreign affiliates operating in tax havens yields a similar pattern of results, albeit with less statistical precision.

This is despite the fact that Japan has not historically restricted parent corporations from borrowing from their foreign subsidiaries without triggering a repatriation tax on “deemed” dividends, contrary to the United States’ use of I.R.C. §956.

Our average AETR measures within domestic or multinational groups are very close to those found by Markle and Shackelford (2012) using Orbis data for a comparable time period.

For a brief (truncated) timeline of the financial crisis, see, e.g., http://​www.​washingtonpost.​com/​wp-srv/​business/​economy-watch/​timeline/​.

All results for the full set of events are available upon request. Consistent with the discussion of confounding factors in Appendix E, CARs surrounding these other dates show little significant association with firm attributes.

As illustrated in the 5-day event windows in Fig. 1, a majority of statistically significant abnormal returns occurred within plus or minus 1 day of their respective event dates. As a result, we focus hereinafter on 3-day event windows, as in Hanlon and Slemrod (2009). Allowing longer 5-day event windows does not qualitatively alter our main results (not shown).

By design, these last figures are equivalent to ACAR estimates obtained from estimation of the standard market model for identical firm groups, albeit with standard errors from the single-step approach that account for intertemporal correlation and cross-group correlation directly. In contrast to the preliminary results presented in Fig. 1, our single-step estimates apply exclusively to the top quartile of domestic and multinational firms by market capitalization.

Note that we estimate the full complement of matching terms for each country in our analysis (including, e.g., \(\zeta ^\mathrm{DE}\)), the purpose being to control for and test—rather than assert—certain logical outcomes, such as the tax savings rate of German firms having zero effect on German CARs or the share of US MNC consolidated assets held in Germany having no influence (beyond the tax savings rate) on US stock market reactions to the Japanese reform.

Altshuler et al. (2015) note that “A notable feature of the Japanese tax environment is a compliant international tax-planning culture. [\(\ldots \)] Although changes in attitudes are occurring, many Japanese companies consider paying taxes a matter of loyalty, and the amount of taxes paid are considered a measure of the company’s success” (pp. 24–25). Tax practitioners point out, for example, that the Japanese tax system does not restrict parent firms from borrowing from foreign subsidiaries, contrary to the US’s treatment of “deemed” dividends under I.R.C. §956, but instances of Japanese corporations availing themselves of this tax avoidance opportunity are unheard of.

Ownership data for our full 100% sample of parent firms from Orbis indicate that 29% of formerly domestic Japanese firms in the 2012 sample had become MNCs by 2016, suggesting that the reform facilitated international expansion at a higher rate for the larger domestic firms in our top-quartile sample. This contrasts with the fact that only 12% of formerly domestic US firms in the full sample had become MNCs by 2016, such that international expansion opportunities were more nearly uniform across the market capitalization distribution of US domestic firms over the same period.

See, for example, “Promote Dividend Repatriation,” by Joseph M. Calianno and Fred F. Murray, Tax Analysts, 2009. http://​www.​taxanalysts.​com/​www/​freefiles.​nsf/​Files/​CALIANNOandMURRA​Y-8.​pdf/​$file/​CALIANNOandMURRA​Y-8.​pdf. The US government implicitly recognized this argument by temporarily relaxing restrictions on US parent borrowing from foreign subsidiaries at the peak of the financial crisis. Faulkender and Petersen (2012) examine the extent to which liquidity constraints played a role in the repatriation and reinvestment behavior of US firms following the enactment of the American Jobs Creation Act of 2004. Constrained firms were more likely to exploit the temporary repatriation tax holiday for the purpose of reinvesting in the US, yet unconstrained firms were the primary beneficiaries of the policy in terms of total remittances.

These latter responses are also more closely aligned with Tajika et al. (2014), who find that liquidity-constrained Japanese MNCs responded to the enactment of the dividend exemption system by increasing dividend receipts from their foreign subsidiaries throughout 2009 to a greater degree than other MNCs.

This approach assumes a full realization of investor expectations over the course of the 12 days defined by the May 9, June 27, August 18, and January 23 events, without allowing for offsetting investor reactions on excluded dates. This assumption is broadly validated in the results of a comparable analysis applied to the full sequence of nine event dates (not shown). We remain reluctant to place too much weight on dates that were so clearly impacted by major global market developments in the context of the financial crisis, and though we cannot fully exclude the occurrence of even more gradual dissemination of information beyond the dates considered, this nevertheless provides some assurance that incorporating the set of most likely additional events to our analysis does not change the overall estimated impact of the reform through the January 23 event.

The corollary to this observation is the aforementioned result that 45% of domestic firms in the top quartile sample became new MNCs between 2012 and 2015, with attributes resembling those of the smaller 75% of MNCs included in our full sample.

Indices for the US auto industry (DJUSAT) and the financial industry (Dow Jones Global Financial Index) are obtained from Global Financial Data (available at https://​www.​globalfinanciald​ata.​com).

See http://​www.​rieti.​go.​jp/​en/​database/​. We use tables on “Capital inputs,” and “Investment and capital stock in intangible assets.”

KLEMS data can be found at www.​euklems.​net.

KLEMS data are available for a number of countries including Japan and Germany, but not the US.