Avoiding taxes: banks’ use of internal debt

Financing in the banking sector relies heavily on debt: Banks usually have little equity relative to debt. The Bank for International Settlements reports an equity-to-total-assets ratio of only 6.9% for banks worldwide in 2015 (Bank for International Settlements 2017). German banks, constituting the sample in this paper, had on average an equity-to-total-assets ratio of 7.0% in 2015, compared to 28.2% in the non-financial sector (Deutsche Bundesbank 2016). Berg and Gider (2017) find that different asset risks can largely explain this capitalization gap between banks and non-banks. The main outside determinants of banks’ capital structure are taxation and regulation, which we will now address in turn.

First, consider taxation. In almost all countries, interest payments on debt are tax deductible while dividend payments are not. Thus, the tax system encourages the use of debt over equity to finance operations—the debt tax shield or debt bias. Thus, third-party debt lowers taxable profits (i.e. the tax base) and therefore implies a lower tax burden. Multinational firms, however, have an additional way to use debt to lower their tax payments: They can use intra-company lending to shift interest payments from affiliates in high-tax countries to affiliates in low-tax countries—debt shifting. Internal debt keeps the profits of the bank group as a whole constant, but lowers the tax burden by shifting profits to a low-tax country.⁵

Therefore, the tax consequences of external and internal debt are quite different: The first lowers the total tax base, while the second shifts profits from high- to low-tax countries. But do the decisions to take on external and internal debt influence each other? Møen et al. (2019) model theoretically the optimal choice of internal and external debt in a multinational firm. While they base their analysis on a generic model of a multinational firm and do not analyse banks, the main motives should be identical in the financial sector. A fundamental insight from their analysis is that the choice of internal and external debt is independent of each other if each has different costs for the firm. These costs are indeed very different: External debt has well-known costs, such as the use of debt as a disciplining device for overspending managers, and the need to balance indebtedness against the probability of costly bankruptcy. Internal debt neither affects the risk of bankruptcy, nor disciplines managers, nor does failure to repay trigger outside enforcement. The costs of internal debt arise, e.g. from compliance costs with tax rules, such as thin capitalization rules and/or controlled-foreign-company (CFC) rules.⁶ Therefore, we treat the decisions to take on internal and external debt as separate and focus exclusively on internal debt in this paper.⁷

We start by discussing the tax rules that may limit debt shifting. Many countries have implemented thin-capitalization rules or earnings stripping rules. These rules restrict the tax deductibility of interest payments to a certain amount or to a defined debt-to-equity ratio. Interest payments exceeding these thresholds do not reduce the tax base and therefore make debt shifting less attractive.⁸ However, most countries either exclude banks from the scope of these rules (e.g. Spain and Italy) or apply more generous thresholds that effectively exclude banks from the scope (e.g. Japan and UK). Germany applies its thin-capitalization rule also to financial corporations. However, the restriction of interest deductibility only applies on net interest expenses, i.e. total interest expenses net of total interest income. Since interest income is a main source of income for financial corporations, the focus on net interest expenses effectively also excludes banks from the scope of the thin capitalization rule. Debt shifting may further be less attractive for tax avoidance if the source country levies a withholding tax on interest payments. Among affiliates in European countries, there are no withholding taxes on interest payments due to the Interest and Royalties Directive (Council Directive 2003/49/EC of 3 June 2003). Additionally, most double tax treaties reduce or eliminate the withholding tax on interest payments.

In addition, several countries have controlled-foreign-corporation (CFC) rules that add passive income (e.g. interest income) in low-taxed affiliates to the tax base of the parent company (see e.g. Ruf and Weichenrieder 2012), allowing for a tax credit for the taxes already paid abroad. If binding, these rules would prevent debt shifting. However, some countries such as, e.g. Japan, the United Kingdom and the United States completely or in large part exclude income from banking from being affected by CFC rules. According to the prevailing view,⁹ also Germany completely excludes income from banking under the relatively loose condition of having a ‘commercially organized business operation’ in the low-tax country.¹⁰ This exclusion of banks from CFC legislation in some major countries provides additional scope for debt shifting compared to multinationals in other sectors.

Second, let us consider banking regulation. Since a large part of bank regulations addresses the capital structure of a bank, debt shifting may be more difficult for banks than for non-financial firms. Merz and Overesch (2016) find evidence suggesting that capital regulations affect profit shifting activities. In most countries, the capital requirements base on Basel III, which implemented a mandatory minimum equity-to-total-assets ratio of 3% in 2018 (with variable mark-ups for globally systemically relevant banks).¹¹ At the level of the parent company, the equity-to-total-assets ratio is calculated on a consolidated basis. Thus, as this ratio contains both the equity and the total assets of all consolidated group members, it does not constrain internal debt shifting (as long as the consolidated equity-to-total-assets ratio fulfils the minimum requirement).¹² However, foreign subsidiaries additionally have to fulfil the capital requirements in their home country. In contrast, foreign branches are only regulated on a consolidated basis in the home country of the parent company. These rules also apply to Germany, the home country of all multinational banks in our sample.¹³

Another potential issue at the border of regulation and taxation that might affect debt shifting in the banking sector is the implementation of bank levies in several countries in the aftermath of the financial crisis. In most countries, internal liabilities are also subject to the levy, increasing the costs of debt shifting. Germany introduced a bank levy in 2011 with progressive tax rates. However, the German levy includes a size threshold of 300 million euros, exempting 77% of German banks (Buch et al. 2016). Given this exemption of the majority of banks and the relatively low bank levy rates (also in other countries, see Devereux et al. 2019, for an overview), especially compared to the potential tax savings from internal debt, the German levy does not seem to affect debt shifting substantially. Furthermore, since the adoption of European bank levy standards in 2015 bank levy rates on intragroup liabilities are reduced by half.

A threat to the empirical identification of internal debt shifting are conduit entities that simply pass through liabilities, by taking up a loan from a related affiliate and passing it on to another affiliate. In these conduit affiliates interest income from conduit claims offsets interest expenses due to conduit liabilities. Thus, using internal gross liabilities as proxy for profits shifted out through internal debt leads to biased estimates. Nevertheless, previous empirical studies on debt shifting have not considered the existence of conduit entities and their potential impact on the estimation of debt shifting.

This paper accounts for conduit entities in internal debt financing. We define conduit affiliates as entities that simply pass-through debt from one related affiliate to another affiliate. Figure 1 illustrates the simplest example of such an internal conduit debt scheme: The tax haven affiliate faces a corporate tax rate equal to \(t_{Ha}\) and lends \(K_{C}\) units of money to the conduit affiliate which is taxed by \(t_{C} > t_{Ha}\). Through the related interest payments profits are shifted from the conduit affiliate to the tax haven affiliate. Moreover, also the headquarter wants to borrow \(K_{HQ}\) from the tax haven affiliate. Instead of directly taking out a loan from the haven affiliate, it can channel this loan via the conduit affiliate. In the headquarter the interest payments for \(K_{HQ}\) are tax-deductible. In the conduit entity the pass-through is completely tax-neutral (assuming that the loans are subject to the same interest rates) as interest expenses to the haven affiliate offset the interest income from the headquarter. In the tax haven affiliate interest income is taxed at rate \(t_{Ha}<t_{HQ}\). Hence, from a tax perspective, taking out the loan through the conduit entity is equivalent to direct lending.

While passing the loan through the conduit entity is neutral regarding the tax treatment, it may be preferable for several reasons. First, additional debt streams offer additional scope for mispricing of internal loans. This form of transfer pricing is a profit shifting channel different from internal debt shifting and is not the subject of this paper. Second, passing internal debt through conduit subsidiaries can simply reflect real structures: the conduit entity can serve as a financial hub that plays the role of a capital coordinator for the group and distributes capital from tax havens to affiliates. This also allows to re-bundle debt, for instance by taking out loans from several low-taxed subsidiaries and distributing them to several high-taxed affiliates through the hub. Third, multinationals might also use conduit entities to conceal the real origin of internal loans. As tax avoidance schemes of several multinationals were recently addressed in the media, multinationals might be interested in making these schemes increasingly opaque, although they are legal. Additionally, tax administrations may focus on loans directly from tax havens, and are less likely to check in detail loans from a conduit country, as the pass-through nature of these loans is hidden from them. Thus, the use of conduits may lower the probability of detailed checks by the tax administration.

How does the use of conduit subsidiaries affect the estimation of internal debt responses to tax rates? In the simple example in Fig. 1, passing \(K_{HQ}\) through the conduit affiliate increases the internal debt levels of both the conduit affiliate and the headquarter. However, \(K_{HQ}\) does not shift any profit out of the conduit affiliate. This double-counting of internal debt effectively assigns artificially high internal debt levels to these conduit entities. If the location of the conduit is correlated with tax rates, this leads to a bias in classical debt shifting regressions employed in earlier literature (which uses internal liabilities as the dependent variable). In Sect. 3.2 we elaborate on the sign of this bias.

Apart from internal debt shifting, some studies consider the use of conduit entities by multinationals in different contexts. Mintz (2004) models that multinationals give equity to a low-taxed conduit entity which then passes the capital as a loan to another higher-taxed affiliate. While the first transaction in most countries is not related with profit shifting (as dividends are usually largely tax-exempt), the loan shifts profits from the high-taxed affiliate to the lower-taxed conduit entity. Johannesen (2014) models how conduit entities can be used for cross-border hybrid instruments intended to avoid taxes. Mintz and Weichenrieder (2010) empirically investigate factors determining why multinationals might use intermediate entities for investing in their subsidiaries. In the same direction, Garcia-Bernardo et al. (2017) identify five countries that are most important for passing-through investments. Dyreng et al. (2015) find that U.S. multinationals systematically supply equity to their subsidiaries through conduit entities located in countries with low taxes on equity distributions. Literature on internal debt shifting so far has not considered the use of conduit entities.

In addition to these tax-related reasons for using conduit entities, banks also use conduit entities for non-tax reasons. First, multinational banks use such financial networks to manage liquidity. Cocco et al. (2009), e.g. find that banks with a larger imbalance in their reserve deposits tend to borrow funds from related banks and pay a lower interest rate on these funds than they would pay on funds from non-related banks. Second, multinational banks use their financial network to allocate systemic risk within the bank. Elliott et al. (2014), e.g. find that a more diversified financial network provides a better insurance against the failure of one specific bank in the network.

Analogously to earlier literature on internal debt shifting in non-financial sectors, we first estimate the effect of corporate tax rates on internal debt-to-asset ratios, using variation in tax rates within a multinational bank group across countries and across time. Accordingly, the baseline regression equation iswhere \(InternalLiabilities_{ikt}\) are internal liabilities in affiliate i of bank group k in period t. We assume that these internal liabilities refer to plain vanilla debt which classifies as debt for tax purposes in all involved countries.¹⁴\(TA_{ikt}\) are total assets. \(CTR_{ikt}\) is the statutory corporate income tax rate affecting affiliate i and \(X_{ikt}\) is a vector of control variables described below. \(\gamma _t\) are time fixed effects, \(\delta _k\) are bank group fixed effects and \(u_{ikt}\) is the error term. If multinational banks indeed shift profits via internal debt, we expect a positive estimate for \(\beta _1\).

To capture the size of an affiliate, we include the natural logarithm of total assets as a bank-specific control variable into \(X_{ikt}\).¹⁵ Earlier literature uses further subsidiary-level controls such as collateral, tangibility, payroll expenses, and off-balance sheet items to decrease the variance.¹⁶ We are unable to do so as our dataset does not include this information. As these subsidiary-level control variables are supposed to proxy for the capacity to take on external debt and to give collateral, they are clearly relevant for analysing the overall debt-to-asset ratio of a bank. However, their influence on internal debt is less clear. We thus find omitted variable bias unlikely to arise from the omission of these variables.

As the magnitude of a bank group’s engagement in a country and thereby also the use of internal debt might be influenced by the macroeconomic situation, we further control for GDP growth, consumer price inflation rates and the natural logarithm of the host country’s nominal GDP. A further control is a country’s share of the financial sector in its gross value added, which should account for countries that act as important financial centres.

Moreover, we include two regulatory variables that potentially influence a bank group’s activities and financing decisions in a country: First we incorporate the minimum regulatory capital requirement for banks and second we control for the capital regulatory index that is provided by Barth et al. (2013) based on the World Bank (2011) survey on bank regulation. This index captures whether a country’s capital requirement is adjusted for individual risks of banks, whether the regulatory capital is adjusted for certain market value losses, and whether certain funds may be used to capitalize a bank. It ranges from 0 to 10, with higher values indicating greater stringency of capital regulation. These variables also indirectly control for financial networks that are set up in response to regulatory requirements.

Another issue with the sample in this paper is that all bank groups are headquartered in Germany. As profit shifting is found to be less intense out of headquarters (see Dischinger et al. 2014), as a robustness check we also exclude all German headquarters from the sample and re-estimate the regressions. The results we find are very similar.

As outlined in Sect. 2.2, the simple internal-liabilities-to-total-assets ratio also includes conduit liabilities that are only passed through and hence do not reflect actual profit shifting. To solely capture internal debt that effectively shifts profits out of the respective affiliate, we have to subtract such pass-through loans: The ratio of internal debt net of pass-through loans divided by i’s total assets is the appropriate measure for debt shifting out of affiliate i. A straightforward debt shifting regression with this ratio as dependent variable isHere \(InternalDebt^{*}_{ikt}\) denotes internal debt net of pass-through loans, and the other variables are as defined above. Previous internal debt shifting regressions, as in regression equation (1), do not subtract pass-through loans in the dependent variable: The common dependent variable is \(\frac{InternalLiabilities_{ikt}}{TA_{ikt}} = \frac{InternalDebt^{*}_{ikt} + e_{ikt}}{TA_{ikt}}\), where \(e_{ikt}\) is debt that is passed through to other affiliates. The regressions therefore estimateIf the choice of the conduit affiliate’s location is correlated with the corporate tax rate, there is a bias in the estimate for \(\beta _1\) similar to the bias that arises with a systematic measurement error in the dependent variable. As pass-through debt always increases the amount of internal gross liabilities, the correlation between the dependent variable in Eq. (3) and \(\frac{e_{ikt}}{TA_{ikt}}\) is positive by definition. Therefore, the sign of the bias is equal to the sign of the covariance between \(\frac{e_{ikt}}{TA_{ikt}}\) (the ‘left-out variable’ here) and \(CTR_{ikt}\):where N is the number of affiliates, T is the number of sample periods and \({\overline{CTR}}\) is the sample mean of \(CTR_{it}\). For the sake of brevity we drop the bank group indicator k from here on, as it is fully included in the bank indicator i. In all subsidiaries that do not serve as conduit entities \(e_{it}\) is equal to zero. Therefore, one can rewrite (4):where subsidiaries \(i=1,\ldots ,H\) (with \(H \le N\)) serve as conduit affiliates. Rearranging givesEquation (6) is negative if the weighted average tax rate of conduit affiliates is lower than the average tax rate of all affiliates in the sample, with weights being equal to the pass-through-debt-to-total-assets ratio \(\frac{e_{it}}{TA_{it}}\). Hence, if conduit entities systematically face lower tax rates than the average of affiliates, classical debt shifting regressions estimate a downward biased coefficient for the corporate tax rate. If conduit entities are, vice versa, located in higher-taxed affiliates, there is an upward bias in estimates for \(\beta _1\) in Eq. (3). As Sect. 4.2 shows, banks in our sample locate their conduit entities systematically in low-tax countries, resulting in a downward biased estimate for \(\beta _1\) when using the classical dependent variable.

To account for the use of conduit affiliates in internal debt financing, we additionally use internal net debt (relative to total assets) as the dependent variable. This variable is defined aswhere \(InternalLiabilities_{ikt}\) denotes affiliate i’s internal liabilities and \(InternalClaims_{ikt}\) are claims to related parties of bank group k in period t. Therefore, the difference is the effective amount of internal debt that shifts profits out of affiliate i, accounting for the potential existence of conduit debt. If internal claims of an affiliate are larger than its internal liabilities, effectively no profits are shifted out via the internal debt channel and \(IntNetDebt_{ikt}\) is zero. The empirical specification for estimation with the ratio of \(IntNetDebt_{ikt}\) to total assets as the dependent variable is equivalent to Eq. (1):The explanatory variables are as defined in Sect. 3.1. With internal debt shifting we expect a positive estimate for \(\beta _1\) in Eq. (8). As argued above, the estimated tax rate coefficient is expected to be higher with internal net debt as the dependent variable compared to internal liabilities if the conduit entities are located in low-tax countries, and to be lower if conduit affiliates are located in high-tax countries. As a robustness check we again re-estimate equation (8) with exclusion of German headquarters to account for the sample’s idiosyncrasy that all bank groups are headquartered in Germany.

Another issue is that country characteristics other than the corporate tax rate influence a bank affiliate’s volume of internal net debt. To account for this, we conduct a robustness check by additionally including bank affiliate fixed effects into regression equation (8). We then only exploit corporate tax rate changes in the sample period to identify the tax effect on internal debt, measuring basically how banks adjust the volume of effectively profit-shifting debt in response to changes in the corporate tax rate.

As stated in Sect. 2.2, there are other, non-tax reasons for intrafirm lending within a multinational bank. Since we do not have data on liquidity or risk allocation, we cannot control for these factors in our model. The change in tax rates should, however, not affect the attractiveness of a conduit entity in these other dimensions. Therefore, insofar as these factors are constant over time (or changes are not correlated with tax rate changes), in the specifications with affiliate fixed-effects, we effectively control for such non-tax related determinants of internal debt.

Lastly, note that while our focus in this paper is on internal debt, similar problems may occur when studying total debt. Consider, for example, a case where a conduit entity takes on external debt and passes it on internally to the affiliate. In this case, we observe increasing external (or total) debt (and internal claims) at the level of the conduit, and internal liabilities (and external claims, if the money is lent to customers) at the level of the subsidiary. Internal debt is not double counted, but total debt is: The same financial flow shows up as (external) debt for the conduit, and as (internal) debt at the subsidiary. Both are part of total debt in unconsolidated data.

Starting from June 2014, the External Positions of Banks database of the Deutsche Bundesbank (2015) also splits up internal liabilities and internal loans by the country of the related affiliate from which the loan is taken or to which the loan is given. This allows to regress bilateral internal net debt on precise bilateral tax rate differentials that unambiguously identify the tax incentive to shift profits between two affiliates. For a subset of German non-financial multinationals, Overesch and Wamser (2014) show a positive effect of such precise tax rate differentials on bilateral debt stocks. So far no study has used bilateral data for estimating debt shifting in banks. Here we use internal liabilities net of internal claims that affiliate i takes out from related affiliates in country j as our dependent variable:where \(InternalLiabilities_{ijkt}\) are liabilities of affiliate i to other affiliates of the same bank group k in country j and \(InternalClaims_{ijkt}\) are claims of affiliate i to related affiliates in country j. We then estimate the following equation for the full sample of German multinational banks and their foreign affiliates:The main variable of interest is \(CTR_{ikt} - CTR_{jkt}\) which denotes the bilateral tax rate differential between the host country of affiliate i and the country of the internal creditor j. \(X_{ikt}\) is the same vector of control variables as above. \(Y_{jkt}\) contains the macroeconomic control variables also for the internal net creditor’s country. \(\gamma _t\) are monthly time fixed effects and \(\delta _k\) are bank group fixed effects. \(u_{ijkt}\) is the error term. Under the hypothesis that banks shift profits from higher-taxed to lower-taxed affiliates via internal debt we expect a positive estimate for \(\beta _1\). Also in this bilateral setting we include bank affiliate fixed effects as a robustness check.

We use the External Position of Banks database of the Deutsche Bundesbank (2015), a unique dataset provided by the German central bank on assets and liabilities in foreign affiliates of German multinational banks and in the respective German headquarters. As this is an administrative dataset to which all German banks with foreign activities are obliged to report monthly, it provides a complete and high quality sample of all German multinational banks. We observe separate records for all subsidiaries, whereas for branches we observe an aggregate figure per bank group and country.

As dependent variables we use internal liabilities held in an affiliate, and internal net debt which is calculated from internal liabilities and internal claims data. For the estimation of Eqs. (1) and (8), these variables are available from June 2010 to December 2015 on a monthly basis. More precise data on bilateral internal loans and liabilities, separated by the country of the internal counterpart, are available from July 2014 until December 2015. Although the sample period for this bilateral data is relatively short, the variation over affiliate/counterparty-country-pairs and over time allows the estimation of Eq. (10) and an identification of the effect of precise corporate tax rate differentials on bilateral internal net debt.

The literature on internal debt shifting usually scales internal debt by total assets. As data on total assets is not available, we proxy total assets by total external assets, which comprise claims against non-residents as well as money market papers, bonds, and shares issued by non-residents.¹⁷ To control for an affiliate’s size, we take the natural logarithm of total external assets as a bank-level control variable. We collect the statutory corporate tax rates on a monthly basis from the Worldwide Corporate Tax Guides of Ernst & Young (2011, 2014). We take country-level controls from the International Monetary Fund’s (IMF) International Financial Statistics, the World Development Indicators of the World Bank, the United Nations Conference on Trade and Development (UNCTAD) statistics and the online data centre of the Organisation for Economic Co-operation and Development (OECD). For some countries we have to complement the data with information provided by national statistical offices (see Appendix “Variable definitions” section for an overview of variables and data sources). As nominal GDP is only available quarterly, we transform it to monthly frequency with the proportional Denton method for flow series as described in Bloem et al. (2001). Also the share of the financial sector in a country’s gross value added is only available with quarterly frequency and we transform it to monthly frequency by cubic spline interpolation. To calculate annual GDP growth rates with monthly frequency, we use interpolated GDP values. Minimum capital requirements are taken from the World Bank (2011) survey on bank regulation. Based on several other questions in this survey, Barth et al. (2013) provide an index on the stringency of capital regulation. As the World Bank provided the most recent version of the survey only in 2011, in our sample these two variables are constant over time. Table 1 shows the basic descriptive statistics of all variables.

Figure 2 illustrates the geographical distribution of German bank affiliates.¹⁸ Most affiliates are located in Europe, probably due to the proximity to the home country and the common regulation in the European banking union that facilitates foreign activities. The most important foreign market for German banks is Luxembourg with 42 affiliates, followed by the United Kingdom with 32 affiliates (in 2013). Outside Europe the United States (20 affiliates) and Singapore (19 affiliates) are the most important markets.

Furthermore, Fig. 2 illustrates the location of the top 5 countries for conduit debt, defined as the sum of each bank affiliate’s \(\text {min}(InternalClaims_{it} ; InternalLiabilities_{it})\) per country. First note that these most important conduit countries are distributed around the world, suggesting that they serve as regional hubs for different world regions in which German banks are active. Second, three of the five most important conduit countries (Cayman Islands, Luxembourg, Singapore) are classified as tax havens by both Dharmapala and Hines (2009) and Johannesen and Zucman (2014), and also the United Kingdom (the most important conduit country) offers a relatively low tax rate. This observation already suggests that in the sample of German multinational banks conduit entities tend to be located in low-tax countries. The high amount of conduit debt in the United States probably reflects the important role of this financial market. Note that also conduit entities in the United States may face lower effective tax rates than the average U.S. tax rate, as banks can locate their foreign affiliates in Delaware, a well known domestic tax haven in the United States (Dyreng et al. 2013).¹⁹

For a further descriptive investigation of the use of internal debt, Table 2 ranks countries according to the mean of the internal debt-to-assets ratios of German bank affiliates in the respective country in 2013. As expected there are several high-tax countries at the top. German bank affiliates in Japan (corporate tax rate of 38.0% in 2013) had the highest internal debt-to-asset ratios (88.6% on average). Also the internal debt-to-assets ratio of German bank affiliates in France and Spain (two further high-tax countries) are on a relatively high level around 80%. Also some tax havens appear in the ranking: German bank affiliates in Hong Kong have a similar internal debt ratio as in Portugal, despite the substantially lower corporate tax rate that would suggest that banks shift profits into affiliates in Hong Kong rather than out of them. The last column in Table 2 explains this finding: It reports for each country the average conduit share in internal debt that is passed through an affiliate (formally defined as the country average of \(\text {min}\left[ \frac{InternalClaims_{it}}{InternalLiabilities_{it}}; \; 1\right] \) in each affiliate). In Hong Kong on average 94.9% of internal liabilities are merely passed through the affiliates, whereas in Portugal the average conduit share is only 25.7%. Hence, even though German banks have similar internal debt ratios in both countries, the tax-effective internal debt ratio is substantially higher in Portugal. Also bank affiliates in Singapore and the Cayman Islands have similar internal debt ratios as affiliates in high-tax countries (e.g. Italy), which can be explained with substantially larger conduit shares of internal debt.

Both Fig. 2 and Table 2 suggest that the conduit affiliates in the sample of German multinational banks tend to be located in tax havens and low-tax countries, implying an underestimation of debt shifting with the classical dependent variable (the internal-liabilities-to-total-assets ratio). Regressing the conduit share of internal debt in an affiliate on the corporate tax rate and controlling for other macroeconomic variables (see regression results in Appendix “Conduit share regressions”) indeed leads to a significantly negative tax coefficient. Hence, in the sample used in this paper, the conduit entities are systematically located in low-tax countries. From a debt shifting perspective this assigns too high internal liabilities to low-taxed affiliates, leading to an underestimation of internal debt shifting with the classical internal-liabilities-to-total-assets ratio as the dependent variable. We therefore expect a larger tax coefficient with the internal-net-debt-to-total-assets ratio as the dependent variable.

We now turn to some descriptives for our newly constructed dependent variable, internal net debt. Figure 3 depicts the development of internal net debt in our data over time. The share of internal net debt to total assets is relatively stable over the observation period and amounts to approximately 20%. Bilateral streams of internal net debt are largest between German headquarters and their subsidiaries/branches in Germany, France, the Cayman Islands, the Netherlands, and Spain (see Table 3). Therefore, internal net debt is especially relevant both in countries with many affiliates (see also Table 2) and in an important tax haven country.