A. Prior Literature

Modigliani and Miller (Reference Modigliani and Miller1963) showed that there are benefits to higher leverage as a result of tax policies. Most countries allow interest payments to be deducted from income, while similar deductions are not allowed on payments to equity holders. Empirically, the effect of tax rates on leverage seems small (Graham (Reference Graham1996), (Reference Graham2000), (Reference Graham2006), Hanlon and Heitzman (Reference Hanlon and Heitzman2010)).

DeAngelo and Masulis (Reference DeAngelo and Masulis1980) theoretically show that when there are non-debt tax shields, it can reduce the tax benefits of debt. Companies have to decide on their non-debt tax shields and tax shields without knowing what the state of the world will be. An increase in non-debt tax shields decreases taxable profit, which leads to more states of the world in which this profit is negative. Since governments do not credit taxes when profits are negative, any additional deductions from taxable profits do not generate a tax benefit. Therefore, the value of the debt tax shield decreases when a company has non-debt tax shields. This means that non-debt tax shields can crowd out the debt tax shield. While this effect can be mitigated by the ability for companies to carry losses forward to future years, the value of the tax shield still declines due to the time value of money.Footnote ¹ Moreover, unless the number of bad states of the world declines in the following year, they once again will in expectation not benefit from the debt tax shield. This suggests that in the long run, maintaining levels of non-debt tax avoidance and a debt tax shield that creates many states of the world with negative profits is not beneficial.

Cobham and Janský (Reference Cobham and Janský2018) show that worldwide tax avoidance benefits exceed half a trillion. While little is known about the choice of tax avoidance method, every taxed euro can only be avoided once. Huizinga and Laeven (Reference Huizinga and Laeven2008) show that multinationals avoid taxes by shifting profits from high-tax firms to low-tax firms within their corporate structure.

B. Hypotheses

I predict that when more profit is shifted out of a country, there will be more states of the world in which taxable profit is negative. Since there are now more states of the world in which there is no benefit to be gained from the debt tax shield, the expected value of the debt tax shield is lowered, relative to a world without non-debt tax avoidance. The reduction in the value of the debt tax shield results in lower leverage.

Huizinga and Laeven (Reference Huizinga and Laeven2008) show that while profit is shifted out of the high-tax countries, it is shifted into low-tax countries. This results in a reduction in taxable profit in high-tax countries, while increasing taxable profit in low-tax countries. Therefore, profit shifting leads to more states of the world with negative profit for high-taxed companies. This reduces the benefit of the debt tax shield and, therefore, results in lower leverage, relative to a world without profit shifting.

However, it simultaneously leads to fewer states of the world with negative profit for low-taxed companies, since the profit is moved into these companies. Therefore, the benefits of the tax shield increase in the low-taxed company.

Since Huizinga and Laeven (Reference Huizinga and Laeven2008) predict that high-taxed companies shift profit out, and low-taxed companies shift profit in, Hypotheses 1 and 2 contrast strongly with the predictions of trade-off theory (Modigliani and Miller (Reference Modigliani and Miller1963), Kraus and Litzenberger (Reference Kraus and Litzenberger1973)). The outbound profit shifting by high-taxed companies will reduce their taxable profit, lowering the value of the debt tax shield in exactly those countries in which trade-off theory predicts it to be high. The inbound profit shifting in the low-taxed companies increases the taxable profit, increasing the value of the debt tax shield, in the countries’ trade-off theory, which predicts this value to be low. Therefore, when testing trade-off theory on a broad sample of firms, we underestimate the effect of tax rates on leverage in the presence of non-debt tax avoidance. Once adequately taking into account the role of non-debt tax avoidance, the relationship between debt and tax rates should be stronger.

C. Empirical Strategy

1. Profit Shifting

In order to determine the effect of tax avoidance on leverage, I first need to identify tax avoidance. Tax avoidance is hard to observe, but multinational’s profit shifting is largely observable. Therefore, I use multinationals in this article.

I estimate profit shifting using the methodology from Huizinga and Laeven (Reference Huizinga and Laeven2008), who use an incentive-to-shift variable. The construction of this variable is discussed in the data section. In line with the literature, I indicate the incentive-to-shift with the letter “C” (Huizinga and Laeven (Reference Huizinga and Laeven2008)). A positive value of C indicates that the tax rate in that country is high, relative to tax rates faced by the same multinational group in the other countries it is active in. A negative value indicates that it wants to move profit into that country, because the tax rate is relatively low. Following Huizinga and Laeven (Reference Huizinga and Laeven2008), I use the natural logarithm of EBIT as a measure of profit shifted. This methodology will show the degree of profit shifting resulting from the incentive to shift, by using the following regression equation:

(1) $$ \ln {(EBIT)}_{ic t}={\beta}_1{C}_{ic t}+{\beta}_x{\mathbf{X}}_{ic t}+{\beta}_z{\mathbf{Z}}_{ct}+{\phi}_{ic}+{\xi}_t+{\unicode{x025B}}_{ic t}, $$

where “i” indicates the multinational group, “c” the country, and “t” the year. Control variables are similar to those used in Huizinga and Laeven (Reference Huizinga and Laeven2008), which include ln(labor costs), ln(fixed assets), interest rates, and ln(GDP per capita). Contrary to the usual industry fixed effects, I include a multinational fixed effect and year-fixed effects. This allows me to estimate how the differences in incentive-to-shift across the firms within the multinational’s group structure affect the profit shifted in the multinational. The coefficient of interest is $ {\beta}_1 $ . I expect a negative coefficient, which implies that profit is shifted out of firms that face relatively high-tax rates and into relatively low-taxed firms.

As a second method of determining profit shifting, I use an exogenous shock caused by a ruling of the European Court of Justice (2006). Schenkelberg (Reference Schenkelberg2018) shows that this ruling impacted profit shifting. This ruling was related to the so-called Controlled Foreign Company (CFC) rules. These rules impose an additional tax on profits made by foreign subsidiaries in low-tax countries. For instance, the U.K. government will tax profits made by an Irish subsidiary (12.5% tax rate) of an English parent (30% tax rate) with an additional 17.5% tax rate, but an Austrian subsidiary (25% tax rate) of the same parent will not face any additional tax. The purpose of these rules is to limit the benefits of shifting profit to low-tax countries. However, the European Court of Justice (2006) ruled that these CFC rules violate the free movement of capital in the EU. Therefore, EU countries can no longer apply these rules to subsidiaries in other EU countries.

I use a difference-in-difference approach to estimate the effect of the shock on profit shifting. The treated companies are the low-taxed EU subsidiaries of EU parent companies from countries with CFC rules. Denmark, Germany, Lithuania, Norway, Portugal, the United Kingdom, and Sweden had such rules in place.Footnote ² I expect an increase in profit shifted into these low-taxed subsidiaries, as the CFC rules are now lifted. The European Court of Justice (2006) ruling only applies to relations between EU parent and EU subsidiaries; therefore, the control companies are subsidiaries in the same low-taxed EU countries but owned by parent companies from non-EU countries with CFC rules (Australia, Canada, Japan, New Zealand, the United States, and South Korea).

2. Leverage

Hypothesis 1 poses that non-debt tax avoidance crowds out leverage. Using the estimates for profit shifting, I employ two different methodologies to estimate the effect on leverage. First, I take the point estimates on the incentive-to-shift ( $ {\beta}_1 $ in equation (1)) and multiply it by the value “C” that takes for that firm. $ {\beta}_1 $ is the within multinational estimate of the effect that “C” has on profit shifted. Multiplying this by “C” gives me the profit shifted by that multinational in and out of each country it operates in, which is unique to each firm in each multinational. I then estimate the effect of this profit shift on leverage. A positive value of profit shifted indicates that profit is shifted into the firm, while a negative value indicates profit was shifted out. Hypothesis 1 suggests that inbound profit shifting results in higher leverage, while outbound profit shifting results in lower leverage. Therefore, I expect a positive effect of the profit shift on leverage. I use the following regression equation:

(2) $$ {Leverage}_{ict}={\beta}_1{\mathrm{Profit}\ \mathrm{shifted}}_{ict}+{\beta}_x{\mathbf{X}}_{ict}+{\beta}_z{\mathbf{Z}}_{ct}+{\gamma}_i+{\zeta}_t+{\unicode{x025B}}_{ict}, $$

where $ c $ indicates the country, $ t $ time, and $ i $ the multinational. $ {\beta}_1 $ captures the effect predicted in Hypothesis 1. $ \mathbf{X} $ is a vector of company-level control variables which includes tangibility, profitability, size, and depreciation (Rajan and Zingales (Reference Rajan and Zingales1995)). $ \mathbf{Z} $ is a vector of country-level controls, which includes the domestic corporate tax rates, ln(GDP per capita), and interest rates. I define leverage as interest-bearing debt over total assets. Following Faulkender and Smith (Reference Faulkender and Smith2014), I include year and multinational fixed effects to capture how within multinational differences in in- and outbound profit shifting affect leverage in the multinational. Moreover, Faulkender and Smith (Reference Faulkender and Smith2014) argue that both tax rates and variation in international operations are highly autocorrelated, making the use of multinational fixed effects preferable.Footnote ³ Standard errors are clustered at the multinational level, as tax avoidance decisions within a multinational are correlated (Huizinga et al. (Reference Huizinga, Laeven and Nicodeme2008)). I also use an alternative specification using bootstrapped errors clustered at the same level.

Second, I use the shock caused by the European Court of Justice (2006). The ruling made it easier to shift profit into low-tax subsidiaries. Hypothesis 1 suggests that this increase in inbound profit shifting should cause an increase in leverage. Therefore, I investigate the effect of the shock on leverage, expecting a positive coefficient.Footnote ⁴

Hypothesis 2 states that the relationship between leverage and tax rates is different for different levels of non-debt tax avoidance. Therefore, I split my sample into four groups based on the level of profit shifting obtained using equation (1). These are highest quartile inbound shifters, middle quartile inbound shifters, middle quartile outbound shifters, and highest quartile outbound shifters. I interact these categories with the domestic tax rate and estimate the effect of tax rates on leverage for the different subgroups.

As an additional test, I use one-off changes in tax rates to show how these changes impact inbound and outbound profit shifters differently. This is effectively a difference-in-difference approach, using tax rate changes as shocks. Firms in countries with a tax change are treated, while firms in countries without a tax change are control companies. Hypothesis 2 suggests that the change in tax rates should impact inbound profit-shifting companies more strongly than outbound profit-shifting companies.

A. Accounting Data

In order to observe profit shifting, I map the group structure of a large sample of European multinationals.Footnote ⁵ One of the major benefits of using EU multinationals is that they are part of a common internal market in which the taxable effects of actions between parent and subsidiary firms are regulated by EU laws. This ensures uniform withholding taxes and double taxation rules for the companies in my sample.

Mapping the group structure and the profit shifting of these multinationals requires accounting and ownership data, which I obtain from the Orbis database. I use the historical files from Orbis and link every parent company to its subsidiaries, and each of those is then linked with its own subsidiaries, and so on (Kalemli-Ozcan, Sorensen, Villegas-Sanchez, Volosovych, and Yesiltas (Reference Kalemli-Ozcan, Sorensen, Villegas-Sanchez, Volosovych and Yesiltas2015)). The data set starts in 2005 and ends in 2011. The ownership data quality greatly improved from 2005 onward. Moreover, ending the sample in 2011 ensures that the period of high uncertainty about the future of profit shifting, caused by the OECD research committee on Base Erosion and Profit Shifting OECD (2014) is not included in the sample. Moreover, this period includes the shock caused by the European Court of Justice (2006), which will allow for a clean identification of the crowding out of debt by non-debt tax avoidance.

Following Huizinga and Laeven (Reference Huizinga and Laeven2008) and Brok (Reference Brok2024), I aggregate all accounting data at the multinational-country-year level to ensure that multinationals with many subsidiaries do not dominate the data sample.Footnote ⁶ Because I use this method of aggregation, I use unconsolidated accounts to observe profit being shifted within the multinational.

For the shock caused by the European Court of Justice (2006), the important unit of observation is the subsidiary of a treated or control parent. Therefore, a slightly different data set is used for the difference-in-difference estimates. This data set looks at the individual firms, rather than the data aggregated at the multinational-country level. This allows me to identify the particular low-taxed firms that have a parent in a country that had a CFC rule. To establish pre-trends, I add the year 2004 to the data.Footnote ⁷ Any firms not observed during the year of the shock and the two prior years are dropped from the sample. If a subsidiary has multiple parent companies, it is included in the treated group if one of those parents is from a treated country.

B. Variable Definition

The Huizinga and Laeven (Reference Huizinga and Laeven2008) measure of incentive-to-shift (“C”) is unique for each multinational-country-year observation and captures how high the domestic tax rate is relative to the foreign tax rates faced by that multinational. The authors define it as follows:

(3) $$ C=\frac{1}{\left(1-{Tax}_{ct}\right)}\times \frac{\sum \limits_{c\ne k}^N\left(\frac{sales_{ict}}{1-{Tax}_{ct}}\times \left({Tax}_{kt}-{Tax}_{ct}\right)\right)}{\sum \limits_{c=1}^N\left(\frac{sales_{ict}}{1-{Tax}_{ct}}\right)}, $$

where $ c $ indicates the country, $ t $ time, and $ i $ the multinational. The measure is a sales-weighted average of tax differences within a multinational. C is positive for the firms that have an incentive to shift profit out and is negative for the firms that should receive that foreign profit.

Control variables are defined in Table 1.

TABLE 1Variable Definitions

TABLE 1 Long description

The table consists of three columns: Variable, Description, and Source.

1. Variable: l n ( E B I T ) sub i c t. Description: The natural logarithm of earnings before interest and taxes in the entity. Source: Orbis Database.

2. Variable: Leverage sub i c t. Description: Measures liabilities compared to total assets, calculated as the fraction: non-equity liabilities sub i c t minus non-interest carrying debt sub i c t, all over total assets sub i c t. Source: Orbis Database.

3. Variable: Tangibility sub i c t. Description: Measures tangible assets as a proxy for collateral, calculated as the fraction: tangible fixed assets sub i c t all over total assets sub i c t. Source: Orbis Database.

4. Variable: Depreciation sub i c t. Description: Normalized by sales to proxy non-debt tax shields, calculated as the fraction: depreciation sub i c t all over sales sub i c t. Source: Bureau van Dijk’s Orbis Database.

5. Variable: Size sub i c t. Description: The natural logarithm of total assets. Source: Orbis Database.

6. Variable: Profitability sub i c t. Description: Return on assets, calculated as the fraction: E B I T sub i c t all over total assets sub i c t. Source: Orbis Database.

7. Variable: Fixed assets sub i c t. Description: The natural logarithm of fixed assets. Source: Orbis Database.

8. Variable: Labor costs sub i c t. Description: The natural logarithm of labor costs. Source: Orbis Database.

9. Variable: Interest rates sub c t. Description: Country-level risk-free interest rate. Source: Thomson Reuters Datastream.

10. Variable: l n ( G D P capita ) sub c t. Description: Natural logarithm of the annual G D P per capita. Source: World Bank Data.

11. Variable: Domestic tax rate sub c t. Description: Marginal corporate tax rate. Source: E & Y Worldwide Corporate Tax Guide.

12. Variable: Incentive-to-shift sub i c t. Description: Used for profit shifting equations. Calculated as the fraction 1 over ( 1 minus Tax sub c t ) multiplied by a weighted average of tax differentials across all other entities k in the group. Source: E & Y Worldwide Corporate Tax Guide.

C. Summary Statistics

The variables used in this article are defined in Table 1. Table 2 shows the summary statistics for the two different samples. Panel A shows the statistics for the sample using the incentive-to-shift (Huizinga and Laeven (Reference Huizinga and Laeven2008) methodology for estimating profit shifting).

TABLE 2Summary Statistics

TABLE 2 Long description

The table contains six columns: Variable, No. of Obs., Mean, Std. Dev., Min, and Max.

Panel A. Incentive-to-Shift Sample (49,732 observations):

* Leverage: Mean 0.502, Std. Dev. 0.239, Min 0.013, Max 0.963.

* Domestic tax rate: Mean 0.304, Std. Dev. 0.068, Min 0.1, Max 0.406.

* C: Mean minus 0.022, Std. Dev. 0.083, Min minus 0.320, Max 0.483.

* Size: Mean 9.821, Std. Dev. 2.827, Min 4.403, Max 17.194.

* Tangibility: Mean 0.175, Std. Dev. 0.197, Min 0, Max 0.863.

* Depreciation: Mean 0.012, Std. Dev. 0.042, Min 0, Max 0.474.

* Profitability: Mean 0.109, Std. Dev. 0.108, Min 0.000, Max 0.549.

* l n (E B I T): Mean 7.087, Std. Dev. 2.830, Min 1.135, Max 15.048.

* l n (Fixed assets): Mean 7.786, Std. Dev. 3.663, Min 0, Max 16.596.

* l n (Labor costs): Mean 8.069, Std. Dev. 2.465, Min 2.890, Max 14.645.

Panel B. Shock Sample (10,246 observations, except for log variables which have 8,682):

* Leverage: Mean 0.333, Std. Dev. 0.253, Min 0.008, Max 0.985.

* Domestic tax rate: Mean 0.183, Std. Dev. 0.024, Min 0.125, Max 0.34.

* Size: Mean 8.290, Std. Dev. 1.568, Min 2.303, Max 10.405.

* Tangibility: Mean 0.284, Std. Dev. 0.244, Min 0, Max 0.929.

* Depreciation: Mean 0.040, Std. Dev. 0.063, Min 0, Max 0.479.

* Profitability: Mean 0.108, Std. Dev. 0.157, Min minus 0.378, Max 0.795.

* l n (E B I T): Mean 6.071, Std. Dev. 1.953, Min 0, Max 10.825.

* l n (Fixed assets): Mean 6.729, Std. Dev. 2.313, Min 0, Max 11.709.

* l n (Labor costs): Mean 6.711, Std. Dev. 1.625, Min 0.693, Max 11.311.

The incentive to shift is, on average, close to 0. This is by design, as the firms that make up a multinational have an incentive to shift profit to other firms in the same multinational.

The sample covering the shock created by the European Court of Justice (2006) ruling consists of companies that are smaller. This makes sense since this sample is not aggregated at the multinational-country-year level. Moreover, the tax rate faced by the companies in this sample is, on average, 12 percentage points lower. This can be explained by the fact that the shock exclusively affected subsidiaries in low-tax countries. Despite being located in low-tax countries, these companies are not just letterbox firms; they have 11 percentage points higher levels of tangibility compared to the incentive-to-shift sample.

A. Endogenous Entry

A potential concern is that multinationals can enter into specific countries for purposes related to tax avoidance. To ensure such endogenous entry is not driving my results, I drop any observation that entered the sample after the start. Moreover, for the purposes of calculating the incentive-to-shift variable, I use the sales the multinational made in each country at the start of the sample, instead of updating it annually. These two adjustments ensure that there is no endogenous entry or expansion that drives the results. Table 10 shows the results. The results are qualitatively unaffected.

TABLE 10Endogenous Entry

TABLE 10 Long description

A table with four numbered columns representing different regression models. Column 1 has the dependent variable l n E B I T. Columns 2, 3, and 4 have the dependent variable Leverage.

Key coefficients and standard errors in parentheses:

* Constant C: -1.474 in Model 1.

* Profit shifted: 0.455 in Model 3.

* Highest quartile outbound asterisk domestic tax rate: -0.542 in Model 4.

* Domestic tax rate: -0.093 in Model 2, 0.940 in Model 3, and -0.231 in Model 4.

* Middle quartile inbound asterisk domestic tax rate: 0.415 in Model 4.

* Highest quartile inbound asterisk domestic tax rate: 0.571 in Model 4.

* Highest quartile outbound: 0.176 in Model 4.

* Middle quartile inbound: -0.150 in Model 4.

* Highest quartile inbound: -0.193 in Model 4.

* Size: 0.002 in Model 2, 0.001 in Models 3 and 4.

* Tangibility: -0.052 in Models 2 and 3, -0.049 in Model 4.

* Depreciation: -1.043 in Models 2 and 3, -1.039 in Model 4.

* Profitability: -0.401 in Model 2, -0.399 in Model 3, -0.391 in Model 4.

* l n G D P per capita: -0.106 in Model 1, 0.053 in Model 2, 0.048 in Model 3, 0.040 in Model 4.

* Interest rate: 0.020 in Model 1, 0.006 in Model 2, 0.009 in Model 3, 0.015 in Model 4.

* l n Fixed assets: 0.225 in Model 1.

* l n Labor costs: 0.623 in Model 1.

Summary statistics for all models:

* Number of observations: 39,249.

* R-squared: 0.846 for Model 1, 0.465 for Models 2 and 3, 0.469 for Model 4.

* Year F E and Multinational F E are included in all models.

For the shock, I have imposed that companies need to be present for the 2 years prior to the shock, which rules out that entry drives the results in the difference-in-difference estimation.

B. Lobbying

A potential concern is that companies can lobby for a tax rate change. Companies might not be reacting to the tax rate changes as much as the tax rate changes are reacting to these companies. This reverse causality concern is mitigated by the fact that I measure leverage at the end of the year and tax rates at the start of the year.

However, to formally test for this concern, I use insight from the literature on the determinants of lobbying. Hill et al. (Reference Hill, Kelly, Lockhart and Van Ness2013) show that the main determinant for lobbying is the size of the company. Neretina (Reference Neretina2020) shows that only a small subset of companies can effectively lobby for policy changes, and trade associations are not effective at representing smaller companies in the lobbying process.

Detailed data on which companies lobbied on specific laws are not available in most European countries. To ensure my results are not driven by lobbying, I omit the largest third of firms from my sample and rerun the regressions. Table 11 shows that the results are qualitatively unaffected.

TABLE 11Robustness

TABLE 11 Long description

The table consists of six numbered columns grouped into three categories: Lobby (1 and 2), Thin-Capitalization (3 and 4), and C F C (5 and 6).

Row-by-row data:

- Profit shifted: Column 2 is 0.227**, Column 4 is 0.163***, Column 6 is 0.154***. Standard errors are in parentheses below each value.

- Domestic tax rate: Values range from 0.008 in Column 1 to 0.488*** in Column 4.

- Size: Consistently positive and significant across all columns, ranging from 0.005*** to 0.008***.

- Tangibility: Consistently negative and significant, ranging from -0.036*** to -0.113***.

- Depreciation: Negative and significant in Columns 1 and 2 (-0.030***), but positive and non-significant in others.

- Profitability: Consistently negative and significant, ranging from -0.063*** to -0.241***.

- ln(G D P per capita): Positive and significant across all columns, ranging from 0.041*** to 0.055***.

- Interest rate: Positive and significant across all columns, ranging from 0.007** to 0.019***.

- Number of observations: 22,196 for Lobby; 36,611 for Thin-Capitalization; 41,484 for C F C.

- R-squared: 0.572 for Lobby; 0.483 for Thin-Capitalization; 0.472 for C F C.

- Year F E and Multinational F E: Marked as Yes for all columns.

C. Tax Base

Kawano and Slemrod (Reference Kawano and Slemrod2012) argue that not only tax rates, but also differences in the tax base can be important to take into account. In the context of leverage, several anti-tax avoidance tools are important to discuss. A popular anti-tax avoidance tool used by governments is Thin-capitalization and Earning-stripping rules. These rules target debt-based tax avoidance by putting a hard cap on the deductibility of interest when a company has a leverage ratio beyond a given cap, or when interest payments exceed a percentage of earnings.

Germany switched from a leverage ratio to interest/ebit restriction in 2007. I exclude Germany from this year onward. In 2006, France made a similar change, and Belgium introduced a tax deduction on equity, putting it on a comparable footing with debt. Both are excluded.

Moreover, the 2006 court case by the European Court of Justice (2006), which I use in the difference-in-difference, could drive the other results. Therefore, I exclude affected firms using the method and group structure data from Brok (Reference Brok2024). Table 11 shows that the results are qualitatively unaffected.

None of the subsidiaries in the shock sample were affected by these tax base changes.

D. Additional Tests

4. Bankruptcy Law

Countries can differ in the strength of creditor rights. Prior literature has shown that this can affect leverage (La Porta, de Silanes, Shleifer, and Vishny (Reference La Porta, de Silanes, Shleifer and Vishny1997)). To control for this, I use a measure based on La Porta et al. (Reference La Porta, de Silanes, Shleifer and Vishny1997) and Djankov, McLiesh, and Shleifer (Reference Djankov, McLiesh and Shleifer2007), which measures the strength of creditor rights in a country. I use the adjusted version published by the World Bank and used in Safavian and Sharma ((Reference Safavian and Sharma2007), Doing Business Report, (no longer) available at http://www.doingbusiness.org). This version is more granular and therefore has time series variation. The main results are qualitatively unchanged.

Table 12 shows the results. The main results are qualitatively unchanged.

TABLE 12Additional Controls

TABLE 12 Long description

A table with three columns. The first column lists independent variables, the second column is l n (E B I T) labeled 1, and the third column is Leverage labeled 2. Standard errors are in parentheses below coefficients.

Key results for Leverage (Column 2):

- Profit shifting: 0.100*** (0.034)

- Domestic tax rate: 0.343*** (0.085)

- Size: 0.004*** (0.001)

- Tangibility: minus 0.037*** (0.009)

- Depreciation: minus 0.063 (0.041)

- Profitability: minus 0.292*** (0.013)

Key results for l n (E B I T) (Column 1):

- C: minus 1.534*** (0.181)

- l n (Fixed assets): 0.224*** (0.007)

- l n (Labor costs): 0.657*** (0.011)

Shared variables (Column 1, Column 2):

- l n (G D P per capita): 0.316*** (0.042), 0.003 (0.009)

- Interest rate: 0.028** (0.012), 0.007*** (0.002)

- Impartial courts: minus 0.149*** (0.020), minus 0.025*** (0.003)

- Integrity of the legal system: 0.051*** (0.018), 0.005* (0.003)

- Legal enforcement of contracts: minus 0.093*** (0.014), 0.017*** (0.003)

- Creditor rights: 0.103*** (0.008), 0.001 (0.001)

- Legal system and property rights: minus 0.042 (0.029), 0.046*** (0.005)

- Hiring regulations and minimum wage: 0.028*** (0.006), minus 0.016*** (0.001)

- Centralized collective bargaining: 0.071*** (0.010), minus 0.021*** (0.002)

- Labor market regulations: minus 0.193*** (0.022), 0.048*** (0.004)

- Regulatory burden: minus 0.016* (0.010), 0.001 (0.002)

- Tax compliance: 0.008 (0.010), 0.016*** (0.002)

- Business regulations: 0.060 (0.045), minus 0.041*** (0.008)

Summary Statistics:

- Number of observations: 49,667 for both.

- R-squared: 0.864 for Column 1, 0.480 for Column 2.

- Year F E and Multinational F E: Yes for both.

E. Standard Errors

The main specifications cluster the standard errors at the multinational level. In an alternative specification, I cluster at the multinational-year level. Tables 13 and 14 show that the results are qualitatively unaffected.

TABLE 13Firm Clustered Standard Errors

TABLE 13 Long description

The table consists of three columns labeled Leverage 1, 2, and 3.

Profit shifted: Model 1 is blank; Models 2 and 3 show 0.124 with standard errors of 0.035 and 0.034 respectively.

Domestic tax rate: Model 1 shows 0.043 (0.033); Models 2 and 3 show 0.292 (0.078).

Size: All models show 0.004 (0.001).

Tangibility: All models show minus 0.038 (0.009).

Depreciation: Model 1 shows minus 0.064 (0.046); Models 2 and 3 show minus 0.063 (0.046).

Profitability: Model 1 shows minus 0.296 (0.013); Models 2 and 3 show minus 0.295 with standard errors of 0.013 and 0.014.

l n (G D P per capita): Model 1 shows 0.055 (0.003); Models 2 and 3 show 0.053 with standard errors of 0.003 and 0.004.

Interest: Model 1 shows 0.016 (0.002); Models 2 and 3 show 0.017 (0.002).

Summary statistics for all models: Number of observations is 49,732; R-squared is 0.468; Year F E and Multinational F E are both Yes.

TABLE 14Firm Clustered Standard Errors

TABLE 14 Long description

The table consists of three columns labeled Leverage 1, 2, and 3.

Key variables and their coefficients (standard errors in parentheses):

* Highest quartile outbound asterisk domestic tax rate: Model 2 and 3 show minus 0.398*** (0.096 and 0.110).

* Domestic tax rate: Model 1 is 0.043 (0.033); Models 2 and 3 are minus 0.135 (0.085 and 0.096).

* Middle quartile inbound asterisk domestic tax rate: Models 2 and 3 show 0.477*** (0.083 and 0.098).

* Highest quartile inbound asterisk domestic tax rate: Models 2 and 3 show 0.543*** (0.097 and 0.115).

* Highest quartile outbound: Models 2 and 3 show 0.121*** (0.033 and 0.038).

* Middle quartile inbound: Models 2 and 3 show minus 0.173*** (0.027 and 0.032).

* Highest quartile inbound: Models 2 and 3 show minus 0.182*** (0.031 and 0.036).

* Size: All models show 0.004*** or 0.003*** (0.001).

* Tangibility: All models show minus 0.038*** (0.009).

* Depreciation: All models show approximately minus 0.06 (0.045).

* Profitability: All models show approximately minus 0.29*** (0.013).

* l n (G D P per capita): Model 1 is 0.055*** (0.003); Models 2 and 3 are 0.042*** (0.004).

* Interest: Model 1 is 0.016*** (0.002); Models 2 and 3 are 0.022*** (0.002).

Summary statistics for all models:

* Number of observations: 49,732.

* R-squared: 0.468 for Model 1; 0.472 for Models 2 and 3.

* Year F E and Multinational F E: Yes for all models.