Static: Corporate and International
Corporate Data
Unfortunately, there is no available microdata for business entities. PWBM uses the aggregate SOI data to forecast the line items on business tax forms. These forms include the various 1120s as well as the 1065 and the associated schedules. PWBM utilizes the available SOI data to distinguish among entities by certain characteristics. These characteristics include major and minor industry defined by twodigit and threedigit North American Industry Classification System (NAICS) code and size by both total assets and business receipts. PWBM creates crosstabulations by merging the available size and industry data to create distinctions between firms by size, industry and size and industry. This disaggregation allows PWBM to use models that forecast business activity and tax liability for each of these subaggregate groups. For passthrough entities (1065, 1120S, 1120REIT, 1120RIC), the aggregate amounts are shared out to ultimate taxpayers when the data is available. PWBM also split the subaggregate data by 1120 filers excluding REITs, RICs and S corporations. The data is available from the IRS tax stats website.
Corporate Liability
Corporate income tax is estimated using the subaggregate data. Each subaggregate model simulates a single representative corporation’s behavior. Total corporate liability is calculated by combining the results from each subaggregate model. In this way, the PWBMTM allows for heterogeneity across firms in forecasting income and deductions.
Depreciation
An important aspect of the corporate model is the benefit of depreciation. PWBM uses a model that forecasts the usage of 15 different classes of investments with differing depreciation schedules across the subaggregate groups. The corporate tax module uses this model to adjust the amount of depreciation deductions as the incentives for investment change. For example, under a temporary expensing provision, the expectation is that corporations would move investment corporations would have undertaken in later years into years where immediate expensing is allowed. The amount of timing shift is estimated by PWBM using historical data.1
Net Operating Losses (NOLs)
NOLs are also an important aspect of the corporate tax code because it allows corporations to smooth income across a number of years (two years back and up to 20 years forward). As such, it is important to model the use of NOLs. In modeling the use of NOLs, PWBM relies upon the literature on the differential tax treatment of losses for corporations.2 PWBM uses the analysis in the literature to model limitations on the use of NOLs for each subaggregate group.3
Other Deductions
PWBM forecasts other deductions on the subaggregate business returns by estimating the relationship between the deductions and macroeconomic variables forecasted by PWBMsim. These deductions include the net interest and the research and experimentation deductions among others.
Effective Tax Rates
PWBM’s corporate tax module produces average effective tax rates (ETR). The nature of the subaggregate models based on PWBM’s merging of available SOI data allows for a calculation of ETRs by size, industry or size and industry. Each subaggregate model projects income and expenses and therefore, liability (as noted here).
International Income Shifting
The U.S. international tax code, prior to the passage of the Tax Cuts and Jobs Act, was often described as a ‘worldwide’ system where U.S. individuals’ and corporations’ foreignsourced income was taxed. This system was in contrast to the ‘territorial’ system all other major countries use where foreignsourced income is not taxed. After the passage of the TCJA, the U.S. system can best be described as a modified territorial system. The system under TCJA attempts to induce U.S. multinational corporations to locate intellectual property in the U.S. rather than in other tax jurisdictions. In doing so, it also induces multinational corporations to locate tangible assets in foreign countries. As such, these new tax provisions change the incentives firms face in their location of asset decision. This decision is further complicated by the reduction of the corporate rate from 35% to 21%. PWBM’s tax module uses available SOI data on foreign tax credits to impute multinational firms’ foreign holdings. These holdings and their associated income are used to calculate both Global Intangible LowTaxed Income (GILTI) and ForeignDerived Intangible Income (FDII) tax liabilities. PWBM also uses available OECD data on corporate income to determine tax liability of foreign holdings. These calculations inform the optimal location decision for firms. They also allow PWBM to estimate the revenue effects of any policy decisions made by foreign countries in response to TCJA.
Profit Shifting Incentive for Multinational Firms
Introduction
In this documentation, we first briefly review the literature on profitshifting incentive that multinational firms face. We then describe our method of estimating the semielasticity of pretax profits with respect to tax rates, using publicly available data and report the results. Next, we discuss some implications of the corporate rate reduction from 35 percent to 21 percent under the Tax Cuts and Jobs Act (TCJA) on profitshifting behavior of a foreign multinational enterprise (MNE). We conclude by proposing some potential avenues for future work.
Literature Review
Dharmapala (2014) provides an excellent review of the empirical literature on taxmotivated income shifting within multinational firms.4 A major finding is that the earlier studies that use aggregate countrylevel data generate semielasticity of pretax profits with respect to tax rates that is significantly higher than the more recent ones that take advantage of the availability of affiliatelevel datasets. The latter analyses are able to control for both observable and unobservable confounding factors that could potentially bias the estimate. For example, in their baseline specification, Hines and Rice (1994), a seminal paper on the topic of profit shifting, find a semielasticity of 2.25.5 On the other hand, Heckemeyer and Overesch (2017) identify a consensus estimate of 0.8 based on 25 studies which include many affiliatelevel studies.6 However, as Dowd, Landefeld and Moore (2017) point out, most of these analyses that use financial reporting data exclude subsidiaries located in tax haven jurisdictions, and therefore, underestimates the semielasticity.7 In addition, most of the datasets that provide information on individual firms’ responses are confidential. As a result, in this documentation, we only present the results based on aggregate countrylevel data. In the near future, we will improve upon our current estimates by using firmlevel data such as the Compustat database, which reports foreign and domestic sales, income and tax expense.
Methods
The main dataset used in the analysis is provided by the Bureau of Economic Analysis (BEA). The information includes the financial structure and operations of U.S. MNEs and their foreign affiliates. We use the 2016 annual survey of U.S. direct investment abroad that is publicly available.8 For our models, we focus on the majorityowned foreign affiliates, which are defined as those in which the combined direct and indirect ownership interest of all U.S. parents exceeds 50 percent. Focusing on these firms allow us to use the richer available data such as the amount of tangible capital that is used and foreign income taxes that are paid. Since the data are aggregated at the country level, we follow the methodology in Hines and Rice (1994) and model a representative U.S. MNE’s decision of where to book profits in response to tax rate differentials. This approach is summarized in equation 1:
Here, π_{i} is pretax profits reported by U.S. MNEs in country i, L_{i} and K_{i} are labor and capital inputs used in the production in country i, A_{i} is aggregate productivity of country i and τ_{i} is foreign tax rate in country i. To understand the rationale behind this equation, consider the production function of its foreign affiliates in country i: . L_{i}, K_{i} and A_{i} control for the real economic activities that generate profits while the other component of π_{i} is shifted income with β_{5} estimating its elasticity with respect to tax rate. We expect β_{5} < 0 since a lower tax rate creates an incentive for a U.S. MNE to shift income towards country i in order to reduce its tax liability. Recall that before the passage of the TCJA, foreign earnings were not taxed by the U.S. government until they were paid to the U.S. parent corporation as a dividend (“repatriated”) and a lot of corporations chose to defer indefinitely. As noted in a blog post, the accumulation of untaxed profits in U.S. MNEs’ foreign affiliates were estimated to be $2.8 trillion in 2017. In another blog post, we discussed the incentives created by the two key international tax provisions of the TCJA: the tax on Global Intangible LowTaxed Income (GILTI) and the reduced tax rate on Foreign Derived Intangible Income (FDII). The GILTI provision taxes future foreign earnings of U.S. MNEs net of a presumptive 10 percent return on tangible assets in order to reduce the potential erosion of the U.S. tax base. However, as discussed in Clausing (2019), the GILTI, as it stands, decreases the motivation for U.S. MNEs to shift profits to tax havens but increases their incentive to shift profits to other foreign countries.9
π_{i} is measured using “profittype return” in the BEA survey Table II.F 1. It is an economic accounting measure of profits from current production. Unlike net income, it is gross of foreign income taxes, excludes capital gains/losses and income from equity investments, and reflects certain other adjustments needed to convert profits from a financial accounting basis to an economic accounting basis. As discussed in Wright and Zucman (2018), this is the internationally agreed concepts of income and wealth and it avoids double counting the profits of indirectlyheld affiliates.10 We use “compensation of employees” in the same table as a measure of L_{i} and use “net plant, property, and equipment” in Table II.B 12 as a measure of K_{i}. These three variables are all reported in millions of dollars. A_{i} is proxied by GDP per capita from the World Bank (2016).11 To compute τ_{i}, we use the corporate statutory tax rates at the country level reported by KPMG.12 Since these rates may not reflect the actual amount of taxes paid by the foreign affiliates in the host countries, we also calculate the average tax rates by dividing foreign corporate income taxes paid (Table II.D 1) by pretax profits in country i. For most of the countries in our sample, the average rate is higher than the statutory rate, which is partially due to the fact that negative earnings of corporations lead to an upward bias in the measure of tax rates. For the sake of completeness, we report our estimation results using both rates.13
In order to capture the nonlinear effects of tax rates on a U.S. MNC’s profitshifting behavior, we include a quadratic tax term in equation 2:
The marginal effect of a decrease in τ_{i} is allowed to vary across countries with different levels of τ_{i}, more specifically, . As before, we expect β_{5} < 0. In addition, we expect β_{6} > 0 because a decrease in the tax rate in a foreign jurisdiction with a relatively high level of tax rate is likely to induce less profit shifting. The importance of considering its nonlinearity is highlighted in Dowd, Landefeld and Moore (2017).14
The ordinary least squares estimation assumes that τ_{i} and u are uncorrelated. However, if tax rates are endogenously chosen, then estimates of the regression coefficients are biased. Hines and Rice (1994) propose using the log of a country’s population as an instrument for its tax rate based on the observation that they are positively correlated and the former has significantly higher explanatory power compared to log GDP, log GDP per capita and other aggregates. The underlying assumption is that log population affects reported pretax profits only through tax rates once the other control variables are included. Similarly, (log population)^{2} is used as an instrument for τ_{i}^{2}.
We adopt the method of TwoStage Least Squares (2SLS) to calculate the instrumental variables (IV) estimates. Take the IV counterpart of equation 1 as an example. The firststage regression that we run is the following:
The secondstage regression is specified in equation 3:
Where is the predicted tax rate from the first stage:
Results
Columns (1)(2) of Table 1 report the OLS estimates of equations 1 and 2. In column 1, we find a negative coefficient of τ_{i} and it is statistically significant. It implies that if the tax rate in country i decreases by 1 percentage point, then the pretax profits reported there increases by 2.83 percent. In column 2, we find a positive coefficient of τ_{i}^{2}, although it is not statistically significant. Recall that in this case, the marginal effect of a decrease in τ_{i} is β_{5} + 2β_{6}τ_{i}. To illustrate the nonlinear effects of tax rates, compare Italy (31.4 percent) to Ireland (5.7 percent). The coefficients suggest that a percentage point decrease in the average tax rate in Italy would increase the pretax profits reported there by (5.06 + 2 * 3.25 * 0.314) = 3 percent. On the other hand, it would increase those in Ireland by (5.06 + 2 * 3.25 * 0.057) = 4.7 percent. All coefficients of the tax rate variables have the expected signs.
Table 1: Effect of Tax Rates on Location of Profits
Dependent variable: log π_{i}  

Ordinary least squares  Instrumental variables  
(1)  (2)  (3)  (4)  
Constant  1.45  1.08  1.96  4.02**  
(1.14)  (1.24)  (1.20)  (2.05)  
log L_{i}  0.02  0.06  0.02  0.28*  
(0.15)  (0.18)  (0.14)  (0.16)  
log K_{i}  0.89***  0.85***  0.86***  1.21*  
(0.15)  (0.18)  (0.17)  (0.17)  
log A_{i}  0.19  0.19  0.25*  0.20  
(0.12)  (0.12)  (0.14)  (0.16)  
τ_{i}  2.83***  5.06***  1.68  13.77  
(0.64)  (2.68)  (1.66)  (8.61)  
τ_{i}^{2}  3.25  25.17**  
(3.70)  (12.08)  
Observations  48  48  48  48 
Note: Standard errors are reported in parentheses. ***, **, and * indicate significance at the 1, 5, and 10% levels respectively.
Columns (3)(4) of Table 1 report the corresponding IV estimates. The firststage Fstatistic indicates that log population is a weak instrument for τ_{i}.15 In addition, DurbinWuHausman tests for endogeneity in IV estimation fail to reject the OLS specification. As a result, we choose the OLS coefficients as our preferred estimates.
Table 2 reports the estimation results using statutory corporate income tax rates. The estimate of the semielasticity is larger and suggests that a 1 percentage point decrease in the statutory rate in country i increases the pretax profits reported there by 4.03 percent. Column 2 implies that such a decrease in Italy would lead to a (15.64 + 2 * 21.37 * 0.314) = 2.2 percentage point increase in its reported profits.16 On the other hand, the increase in Ireland would be (15.64 + 2 * 21.37 * 0.125) = 10.3 percent.
Table 2: Effect of Tax Rates on Location of Profits
Dependent variable: log π_{i}  

Ordinary least squares  Instrumental variables  
(1)  (2)  (3)  (4)  
Constant  0.70  1.19  0.72  9.68  
(1.72)  (1.71)  (2.53)  (22.01)  
log L_{i}  0.28  0.36  0.28  0.12  
(0.27)  (0.30)  (0.24)  (0.85)  
log K_{i}  0.63***  0.54**  0.63***  1.11  
(0.23)  (0.25)  (0.21)  (1.35)  
log A_{i}  0.15  0.12  0.15  0.27  
(0.18)  (0.17)  (0.17)  (0.47)  
τ_{i}  4.03*  15.64***  3.98  67.72  
(2.30)  (4.76)  (8.06)  (229.18)  
τ_{i}^{2}  21.37***  164.69  
(6.95)  (593.5)  
Observations  53  53  53  53 
Note: Standard errors are reported in parentheses. ***, **, and * indicate significance at the 1, 5, and 10% levels respectively.
Implications
Although we obtain estimates of semielasticity of pretax profits with respect to tax rates that are higher than the studies that use affiliatelevel datasets due to data limitations, our results still illustrate the importance of taking into consideration its nonlinear effects when assessing the profitshifting behavior of U.S. MNEs. Once again, ignoring them would lead us to overestimate the responsiveness of reported pretax profits to tax rate changes in hightax jurisdictions while underestimating it in lowtax ones. Under the recent TCJA, the U.S. reduced its corporate rate from 35 percent to 21 percent. However, since its previous rate was one of the highest in the world, a 1 percentage decrease from that high level would have a much smaller effect compared to other countries with a significantly lower corporate rate to start with. Although the 14 percent decline was considerable, we would expect the increase in reported pretax profits by foreign MNEs in the U.S. to be considerably smaller than assuming a constant semielasticity.
Suppose that foreign MNEs decision making on profit booking is similar to U.S. MNEs, then we can apply the estimated semielasticity in Dowd, Landefeld, and Moore (2017) to gauge the amount of pretax profits that is shifted to the U.S. in response to its lower tax rate. They take the nonlinear effects seriously and are able to use a unique panel data set of U.S. tax returns over the period 20022012 which provides information on the activities of U.S. corporations and their subsidiaries. In the linear case, they estimate β_{5} = 1.437 using the statutory rates, which implies that reported pretax profits by foreign MNEs in the U.S. would increase by 1.437 * 14% = 20.118% while in the nonlinear case, they estimate β_{5} = 10.73 and β_{6} = 8.089, which suggests a 12.85% increase.17
Future Work
Note that so far, we have focused on the profitshifting incentive created by tax rate changes after controlling for the real economic activities. The next step is to investigate how tax rates affect the location choice of MNEs in terms of production and employment. Some argue that lower corporate tax rates benefit countries by attracting mobile capital and new investment and consequently higher wages and more jobs. Others disagree on the grounds that the race to the bottom on corporate tax would be harmful to the U.S. economy. It is both interesting and important to study whether a reduction in its corporate rate makes the U.S. a more attractive destination for investment compared to other foreign jurisdictions and the quantitative effect on economic growth that follows. In addition, the methodology described here can be applied to examine the impact of the GILTI tax on the reduction in pretax profits that U.S. MNEs shift to tax havens such as Bermuda and the Cayman Islands since these profits are now taxed immediately instead of when they are repatriated. Note also that the HinesRice method captures total profit shifting, but it would be insightful to understand how much of that is between the parent company and its affiliates and how much is between foreign affiliates themselves. Finally, we can improve upon our current estimates by including additional countrylevel control variables and take advantage of the fact that the BEA survey is publicly available every year between 1983 and 2016. This allows us to include year dummies, for example, to control for aggregate effects that are timevarying and unobservable. A comparison of the results based on the activities of foreign MNEs and U.S. MNEs will also shed light on how applicable the estimated semielasticity from each study is to MNEs incorporated in other countries.

The analysis PWBM performed also benefited greatly from conversations with Matthew Knittel and from the work of Kitchen and Knittel (2016): https://www.treasury.gov/resourcecenter/taxpolicy/taxanalysis/Documents/WP110.pdf. ↩

See “The Implications of Tax Asymmetry for U.S. Corporations,” Michael Cooper and Matthew Knittel, National Tax Journal, March 2010, 63(1), 3362 for a full discussion of the relevant literature. ↩

PWBM benefited greatly from conversations with Matthew Knittel of Pennsylvania’s Independent Fiscal Office as well as Treasury experts. ↩

Dharmapala, D. (2014). “What Do We Know about Base Erosion and Profit Shifting? A Review of the Empirical Literature,” CESIFO Working Paper No. 4612. ↩

Hines, J. R., Jr. and E. M. Rice (1994). “Fiscal Paradise: Foreign Tax Havens and American Business” The Quarterly Journal of Economics 109, 149182. ↩

Heckemeyer, J. H. and Overesch, M. (2013). “Multinationals’ profit response to tax differentials: Effect size and shifting channels,” ZEW Discussion Paper No. 13045. ↩

Dowd, T, P. Landefeld and A. Moore (2017). “Profit Shifting of U.S. Multinationals,” Journal of Public Economics 148, 113. ↩

Data can be retrieved from https://www.bea.gov/international/usdia2016p. ↩

Clausing, Kimberly (2019). “Fixing the Five Flaws of the Tax Cuts and Jobs Act,” Working Paper. ↩

Wright, T. and G. Zucman (2018). “The Exorbitant Tax Privilege,” Working Paper. ↩

Data can be retrieved from http://data.worldbank.org/indicator/ny.gdp.pcap.cd. ↩

Data can be retrieved from https://home.kpmg.com/xx/en/home/services/tax/taxtoolsandresources/taxratesonline/corporatetaxratestable.html. Dowd, Landefeld and Moore (2017) also use the annual summaries of corporate tax systems by KPMG as their main sources for corporate statutory tax rates. ↩

Out of the 57 countries in our sample, foreign affiliates of U.S. MNEs report negative “profittype return” in three countries, which we exclude in the estimation due to the loglinear specification. As Dharmapala (2014) points out, incentives for profitshifting activity are diminished for lossmaking firms. In addition, there is insufficient information to calculate average tax rates for three additional countries. We also exclude the three countries where average tax rates are negative or exceed 1. These cases are possible because the reported “profittype return” and “foreign income taxes” are not constrained to be nonnegative and the former is not restricted to be higher than the latter. ↩

Another variant of the estimating equation that we can take to the data is the following: log π_{i} = β_{1} + β_{2} log L_{i} + β_{3} log K_{i} + β_{4} log A_{i} + β_{5} (1 / (1  t_{i})^{2}) + u. This equation is derived by taking a firstorder Taylor expansion of the reported profitability equation in (1  t_{i})^{2}, around the point at which (1  t_{i})^{2} = λ^{2}, where λ is the Lagrange multiplier corresponding to the constraint. Since 1 / (1  t_{i})^{2} is an increasing function of t_{i} on the interval [0,1] and a lower tax rate encourages profit shifting towards country i, we expect β_{5} < 0 as well. See Hines and Rice (1994) for more detail. ↩

In a number of auxiliary regressions that are not reported in Hines and Rice (1994), they find that log population significantly outperforms log GDP, log GDP per capita, and other aggregates in explaining tax rates based on the 1982 BEA survey. As future work, we will find a better instrument for tax rates according to information in more recent years. ↩

Italy’s statutory corporate income tax rate happens to be the same as its average rate. ↩

Different from Hines and Rice (1994), Dowd, Landefeld and Moore (2017) estimate the semielasticity of log profits with respect to net of tax rate, which is defined as one less the statutory (average) tax rate for the country of the affiliate. Therefore, the impact of a 14 percent reduction in the corporate rate in the U.S. is calculated as follows: 5 * 1  0.21 + 6 * 1  0.212  5 * 1  0.35 + 6 * 1  0.352 = 0.1285. ↩