Interactions between Bank Levies and Corporate Taxes: How is Bank Leverage Affected?

Regulatory bank levies set incentives for banks to reduce leverage. At the same time, corporate income taxation makes funding through debt more attractive. In this paper, we explore how regulatory levies affect bank capital structure, depending on corporate income taxation. Based on bank balance sheet data from 2006 to 2014 for a panel of EU-banks, our analysis yields three main results: The introduction of bank levies leads to lower leverage as liabilities become more expensive. This effect is weaker the more elevated corporate income taxes are. In countries charging very high corporate income taxes, the incentives of bank levies to reduce leverage turn insignificant. Thus, bank levies can counteract the debt bias of taxation only partially.

tax on liabilities. At the same time corporate income taxation makes funding through debt more attractive, because interest on debt is tax-deductible in most countries while return on equity is not. The goal of this paper is to investigate how bank leverage is aected by the introduction of bank levies depending on corporate income tax rates.

Contribution
By providing evidence regarding the impact of bank levies on bank behavior conditional on corporate income taxation, we contribute to the assessment of regulatory reforms and potential interaction eects like the European Banking Union using bank levies to fund the Single Resolution Fund. Given the crucial role of bank leverage in the global nancial crisis, a better understanding of the impact of bank levies on banks' capital structure also yields valuable implications for nancial stability. In addition, our analysis bridges and thus contributes to two strands of the literature considering the individual eects of bank levies and corporate income taxes (CIT) on banks' capital structure.

Results
First, we conrm ndings from previous literature. Banks in countries where a levy is introduced, such that debt nancing becomes more expensive, show lower leverage than banks not subject to a levy. Second, in countries with higher CIT rates, an introduction of a bank levy reduces leverage less than in countries with lower tax rates. Third, in those countries with the most elevated tax rates, the positive incentives of bank levies on capitalization are not large enough to counteract the debt bias of taxation; the eect of a bank levy turns statistically insignicant. In a broader context, our results imply that before introducing new regulation to target a specic outcome, regulators should take possible interaction eects with (non-) regulatory measures. Otherwise, regulatory eectiveness cannot be guaranteed. Regulatory bank levies provide incentives for banks to reduce leverage, as they are typically designed as a tax on liabilities. At the same time, corporate income taxation makes funding through debt more attractive, because interest on debt is tax-deductible in most countries while return on equity is not. In this paper, we ask how eective regulatory bank levies are in reducing bank leverage, depending on the corporate income tax (CIT) rate. Moreover, we study how the design of bank levies aects their impact upon leverage. As the European Banking Union also uses bank levies to nance the Single Resolution Fund, evidence regarding the impact of bank levies on bank behavior conditional on the corporate tax environment contributes to the assessment of such regulatory reforms.
The literature shows that corporate income taxes aect bank capital structure (De Mooij and Keen, 2016) and that banks exposed to regulatory levies strengthen capitalization (Célérier, Kick, and Ongena, 2018;Devereux, Johannesen, and Vella, 2019). However, empirical evidence on the interaction eects between regulatory and corporate taxes is so far missing. Against the background of the evaluation of changes in banking regulations and potential interactions between dierent policy interventions (FSB, 2017), this paper aims at lling this gap. Our goal is to investigate what role regulatory bank levies play in counteracting the debt bias of taxation. A better understanding of the impact of bank levies on bank capital structure, depending on corporate taxation, is crucial given that the debt bias of taxation is shown to not only increase leverage of both non-nancial and nancial rms, but also the probability of systemic banking crises (De Mooij, Keen, and Orihara, 2013).
In the aftermath of the global nancial crisis, many European countries introduced regulatory levies, the goal being to internalize banks' contribution to systemic risk. On the one hand, bank levies are aimed at establishing funds to nance the restructuring and resolution of banks in distress. On the other hand, banks' funding composition 1 should be inuenced by taxing non-deposit liabilities of banks, thereby setting an incentive for lower leverage and funding risk. Given the opposite incentives for higher bank leverage that result from corporate taxation, the goal of this paper is to better understand the interaction eects between regulatory and corporate taxes, thus understanding the consequences for the eectiveness of bank levies as a tool to increase nancial stability through a less risky bank capital structure.
Using bank-level balance sheet data for EU-countries over the 2006 2014 period, we investigate how bank leverage is aected by the introduction of regulatory levies, depending on CIT rates. The regression analysis yields three key insights. First, we conrm ndings from previous literature (Célérier et al., 2018;Devereux et al., 2019) that the direct eect of bank levies on leverage is negative and statistically signicant.
Banks in countries where a levy is introduced, such that debt funding becomes more expensive, show lower leverage than banks that are not subject to a levy. Second, higher CIT rates mitigate the leverage-reducing eect of bank levies. In countries with higher CIT rates, an introduction of a bank levy reduces leverage less than in countries with lower tax rates. At the average CIT rate (30.2%), introducing a bank levy reduces leverage by 0.4 percentage points. The equity-to-assets ratio, in turn, increases by 4% for the average bank. For the countries with the lowest CIT rates in the sample (10%), the corresponding marginal eect on leverage amounts to minus 3 percentage points and hence to a rise in the equity ratio of about 30% for the average bank. Third, and lastly, for the most elevated CIT rates, the positive incentives of bank levies on capitalization are not large enough to counteract the debt bias of taxation. Indeed, the eect of a bank levy turns statistically insignicant in high-corporate income tax countries, such that the goal of fostering nancial stability through lower leverage cannot be fullled by the regulatory tax.
Our analysis bridges and, thus, contributes to two strands of the literature. A rst set of related studies deals with the implications of the introduction of regulatory bank levies since the global nancial crisis. Exploiting variation in bank levies in the 2 European Union (EU) across countries, banks and time, Devereux et al. (2019) present empirical evidence that banks exposed to regulatory levies increase their equity ratio, thus reducing funding risk. At the same time, portfolio risk is shown to increase. Concentrating on dierent bank-level outcome variables, Buch, Hilberg, and Tonzer (2016) show that loan supply and deposit rates were, on average, not signicantly aected by the introduction of the bank levy in Germany. However, the most aected banks reduced loan supply and deposit rates while raising lending rates. An increase in lending rates is also found after the introduction of the Hungarian levy by Capelle-Blancard and Havrylchyk (2017). For a sample of EU banks, Kogler (2019) nds that banks pass the levies through to customers via higher lending rates while keeping deposit rates constant. This eect is weaker for the well-capitalized banks that are less exposed to the levies. 1 Our analysis diers from these studies as we focus, besides the direct impact of levies on bank leverage, on the interactions between bank levies and the CIT.
A second strand of literature investigates the relationship between corporate income taxation and leverage. As summarized in a meta-study by Feld, Heckemeyer, and Overesch (2013), the design of the corporate tax system is an important determinant of non-nancial rms' capital structure. Typically, tax systems incentivize leveraging since interest paid on debt is tax-deductible whereas the return on equity is not. To lower their tax burden, rms tilt their capital structure more toward debt than they would in the absence of this tax bias in advantage of debt. The positive eect of the CIT on leverage is well established in the literature.
2 Findings by Heider and Ljungqvist (2015) suggest asymmetric eects of tax rates on leverage: U.S. rms' leverage responds to tax increases, but not to tax cuts.
1 Kogler (2019) discusses theoretically the interaction eects between corporate taxation and levies for the pass-through of bank levies to customers in terms of lending rate increases. If the levy payment is not tax deductible, as in Germany or the UK, the pass-through is expected to be stronger than in countries where the levy payment can be deducted so that double taxation is prevented.
2 For an overview, see Bremus and Huber (2016). Another but less related strand of literature analyzes whether and how much corporate income taxes are shifted to customers (see e.g. Banerji, Chronopoulos, Sobiech, and Wilson (2017) or Capelle-Blancard and Havrylchyk (2014)) and how securitization is aected by the CIT (Gong, Hu, and Ligthart, 2015).

3
As banks face dierent funding decisions than non-nancial rms and are subject to regulatory capital requirements, they were typically excluded from the analysis of capital structures pre-crisis. Yet, Gropp and Heider (2010) show that, as long as banks hold more capital than required by regulation, the drivers of capital structure are similar for nancial and non-nancial rms. Still, banks tend to be more leveraged than non-nancial rms. Berg and Gider (2017) nd that this dierence is largely explained by lower asset side risk of banks due to diversication.
Regarding the role of CIT for bank capital structure, a small but growing literature concludes that the debt bias of taxation also aects nancial rms. Comparing the tax sensitivity of banks' and non-banks' capital structure, Heckemeyer and De Mooij (2017) nd similar values for both groups of rms. However, the tax sensitivity diers across rm size and leverage. While larger and capital-tight banks react less to tax changes, the relationship between tax rates and the size of non-banks is found to be Ushaped. De Mooij and Keen (2016) argue that capital buers that are typically above regulatory capital requirements leave scope for taxes to aect bank leverage. Based on bank balance sheet data for 82 countries, they conrm that banks' reaction to taxation is, on average, similar to that of non-nancial rms and that large banks are less tax-sensitive than small ones.
3 Related studies for the United States (Milonas, 2018;Schandlbauer, 2017) conrm a signicant impact of tax changes on bank leverage, which diers across bank characteristics like capitalization and size. Using Italian data, Gambacorta, Ricotti, Sundaresan, and Wang (2017) provide evidence that banks reduce leverage following tax reductions and that non-deposit liabilities decline more than deposits. Focusing on the capital structure of multinational banks, Gu, de Mooij, and Poghosyan (2015) show that the debt bias of taxation also aects bank subsidiaries and that international tax dierences lead to debt shifting to countries with high taxes.
Shifting the focus from CIT to the eects of bank levies and of Allowances for Corporate Equity (ACE), Célérier et al. (2018) nd that tax reforms that make leverage 3 Hemmelgarn and Teichmann (2014) nd smaller, but also statistically signicant, eects of CITchanges on bank leverage. 4 more expensive increase bank capitalization, while simultaneously promoting lending.
Regarding tax reforms, they exploit, on the one hand, that several countries have reduced the tax discrimination against equity by allowing for a deduction of a notional interest rate for equity through ACEs, while others have not. On the other hand, they also exploit the introduction of bank levies that increase the total cost of capital, since liabilities are taxed, thus becoming more expensive. In a similar vein, Schepens (2016) presents evidence that the capitalization of Belgian banks signicantly increased after implementation of an ACE in 2006.
While the discussed studies analyze the eects of CIT and of regulatory taxes separately, we contribute to the literature by estimating the eects of introducing bank levies, depending on CIT rates. By examining the interaction eects between regulatory and corporate income taxes, we aim at gauging how far bank levies can counteract the debt bias of taxation. The remainder of the paper is structured as follows. In the following Section 2, we explain the theoretical link between bank leverage and taxes, both corporate income taxes and bank levies. Section 3 describes both the data used and its sources as well as our empirical model specication. We discuss the regression results and several robustness tests in Section 4, while Section 5 concludes and presents potential policy implications. Bank levy expenses = Levy rate * (T otal liabilities − Customer deposits − Equity) 4 See Appendix B for a detailed overview on which countries use wholesale liabilities as a tax base for their bank levy and Section 3.2 for more information on the data.
Consequently, the cost of debt (or: leverage) increases, making debt funding less attractive. Bank levies target exclusively the nancial sector, especially credit institutions.
In the aftermath of the global nancial crisis, bank levies were introduced as an instrument to establish resolution funds to nance the resolution and restructuring of banks in distress (e.g. Cyprus, Germany, Latvia, and Sweden). In addition, countries opting for a bank levy that taxes wholesale funding aimed at reducing systemic risk by providing incentives for banks to shift from an over-reliance on short-term interbank nancing to more stable funding sources such as customer deposits and equity capital (Kogler, 2019). Along these lines, Devereux et al. (2019) present a theoretical model of bank leverage, a tax on liabilities, and bank capital requirements where banks maximize the expected return to shareholders by choosing, among others, the optimal level of total debt (or leverage, as the amount of total assets is kept xed for simplicity).
In that framework, banks react to an increase in the tax on debt by reducing leverage. Similarly, in the model by Keen (2018), optimal leverage falls the higher the levy is, since the cost of leverage increases. These considerations lead to our rst testable hypothesis: Hypothesis 1: A bank levy on debt incentivizes banks to reduce leverage.
In contrast to bank levies, corporate income taxes are a general instrument targeting the non-nancial as well as the nancial sector. The main objective is to generate revenues for the public sector. Given that interest payments on debt are tax deductible, expenses due to corporate income taxes amount to CIT expenses = CIT rate * (N et income bef ore taxes & interest − Interest payments on debt). ( There is no explicit aim to target the behavior of taxed entities as concerns their capital structure. Nevertheless the empirical and theoretical literature documents that higher CIT rates set incentives for both non-nancial rms and banks to increase leverage in order to lower tax expenses (Feld et al., 2013;Gropp and Heider, 2010; 6 De Mooij and Keen, 2016;Langedijk, Nicodème, Pagano, and Rossi, 2015). This debt bias of taxation results from the fact that interest rate costs for external debt are generally tax deductible, and thereby reduce the taxable net income of a company, whereas interest on equity is not.
As shown in the model of corporate income taxes and bank leverage presented by De Mooij and Keen (2016), if banks optimally choose total debt in a world with capital requirements, they borrow up to the point where the expected cost of violating the capital requirement equals the tax advantage of debt. The model implies that higher tax rates result in banks increasing their optimal amount of debt. The marginal tax benet of debt increases in the corporate income tax rate, thus increasing the optimal level of debt if tax rates rise. We can thus form the second hypothesis: Hypothesis 2: Bank leverage is higher the more elevated corporate income tax rates are.
Due to opposing eects of corporate income tax rates and bank levies on leverage, the question arises of whether there is an interconnection between corporate income taxes, bank levies and bank leverage. If this is the case, it bears important policy implications. In particular, considering the case that the leverage-reducing eect of a bank levy taxing wholesale funding interacts reversely with the debt bias of taxation of the corporate income tax. In such a context, the eectiveness of the regulatory policy instrument cannot be guaranteed.
According to the trade-o theory of debt, rms choose the optimal debt level such that the marginal benet of debt equals the marginal cost of debt (see e.g. Feld et al. 2013, Graham 2003, Heider and Ljungqvist 2015. The value of the tax shield of debt and hence the benet of debt increases in the CIT. The cost of debt is generally modeled as an increasing, non-linear function of debt that mirrors, e.g., bankruptcy cost. As argued above, the cost of debt rises once a levy on bank debt is introduced. Following Heider and Ljungqvist (2015), Figure 1 illustrates the optimal choice of debt according to the trade-o theory. When comparing the reduction in optimal leverage in response Figure 1: Changes in optimal leverage at dierent CIT rates This gure illustrates optimal leverage according to the trade-o theory of debt. Following the illustration by Heider and Ljungqvist (2015), we posit that the benet of debt positively depends upon the CIT and it linearly increases in debt (B = CIT · r d D). The cost of debt increases with debt in a non-linear way (C = a + bD + cD 2 ). If a bank levy is introduced (levy = 1), the cost of debt is higher for each level of debt compared to the case without a levy (levy = 0). Firms increase debt up to the point where the marginal benet of debt equals the marginal cost of debt. D l debt if a levy is in place and the CIT rate is low (high), whereas D l 0 (D h 0 ) is optimal debt if there is no levy and the CIT rate is low (high). Source: Own illustration.
to a bank levy if the CIT is high to the reduction in leverage when the CIT is low, it appears that rms adjust leverage less in response to a levy in countries where the CIT is high (and hence optimal leverage is high) compared to rms in countries where the CIT is low (and hence optimal leverage is lower). Thus, we suspect that the negative eect of bank levies on leverage is potentially smaller the higher the prevailing CIT rate is: Hypothesis 3: The leverage-reducing eect of bank levies is counteracted by the size of the corporate income tax rate.
In what follows, we empirically analyze the nexus between regulatory and corporate taxation and its eect on bank leverage. Fund. We next describe our dataset and some key features of the variables of interest, before discussing our estimation and identication strategy.

Bank-level data
Annual balance sheet and income statements for banks in 27 EU member states were obtained from Bankscope by Bureau van Dijk for the 2006 2014 period.

In order
to clean the data from misreporting and outliers, we apply several standard screens.
We eliminate bank observations if negative or zero values of equity, assets, liabilities or loans are reported or when the loans-to-assets or the equity-to-assets ratio exceeds one.
Further, only banks with at least three observations across the sample period are kept.
Following De Mooij and Keen (2016) and Kogler (2019), we consider unconsolidated accounts that end at national borders and to which national tax rates and in general also regulatory bank levies apply. That is, we include observations with Bankscope consolidation codes U1 (unconsolidated statement with no consolidated companion) and U2 (unconsolidated statement with a consolidated companion). In terms of bank business models, our baseline sample includes bank holdings and holding companies, commercial banks, cooperative banks, and savings banks. In order to prevent outliers from aecting our results, we winsorize all bank-level variables at the top and bottom 5 We do not cover all 28 EU-countries as Croatian banks do not report all control variables included in the regression equation and, therefore, drop out of the sample. 9 1%-percentile. 6 Following the banking literature, our dependent variable, leverage of bank i in year t, is dened as liabilities divided by total assets (Berg and Gider, 2017;Gropp and Heider, 2010;Gu et al., 2015). Figure 2 illustrates that at the sample median, leverage has followed a slight upward trend between 2007 and 2013, with the highly leveraged banks (75th percentile) showing a rather stable leverage ratio, while leverage trended upwards for lower-leverage banks (25th percentile). 7 The standard bank-level control variables that gauge bank size, protability, and risk are also sourced from Bankscope. Appendix A provides a detailed data description of all variables used in the regression analysis. Table 1 reports summary statistics for our baseline regression sample. The sample mean of bank leverage, as measured by total liabilities to total assets, is 90%, varying between 16 and 98%. 8 Regarding the unconditional correlations between the bank-level variables included in the regression model below, Table 2 reveals that leverage is higher for larger banks and lower for more protable banks in our sample.

Country-level data
Information on bank levies for our sample period, like the year of the introduction and the tax base, is taken from Devereux et al. (2019) and double-checked with the ECB's Macroprudential Policies Evaluation Database by Budnik and Kleibl (2018). We also verify whether countries have implemented a bank levy in those years not covered by Devereux et al. (2019). Detailed information on the data source by country is provided in Appendix B.

Figure 2: Evolution of bank leverage
This gure illustrates the evolution of bank leverage as measured by total liabilities to total assets (in %) for the sample median as well as the 25th and 75th percentile. Source: Bankscope, own calculations.  In our baseline regressions, we include 27 EU-countries and construct a dichotomous variable that equals one if a bank levy is in place in a given country and year, and zero otherwise. Appendix B contains detailed information on the countries that implemented the levy, the implementation year, and the tax base. The majority of countries implemented a levy in 2011, while others adopted it earlier or later. As shown in Table   3, prior to 2009, no banks included in our sample were subject to a levy, whereas in 2011, about one-third of the banks had to pay levies. The share of aected banks increased to 73% at the end of our sample period. The timing is in line with the start of policy discussion about the implementation of levies to nance restructuring funds and internalize banks' contribution to systemic risk after the nancial crisis (IMF, 2010). Table 3 reveals that the number of banks covered in our sample signicantly increases across time. This is predominantly due to the improving coverage of the non-performing loans ratio used as a control variable for bank risk.
Among the 17 countries that have introduced a bank levy within our sample period, the majority implemented the levy design as suggested by the IMF (2010), namely as

Regression model
In order to analyze how the introduction of bank levies aects bank capital structure, depending on the prevailing CIT, we estimate the following regression equation using a panel xed-eects estimator. The dependent variable, bank leverage of bank i in country c at time t, is dened as the ratio of liabilities (total assets minus equity) to total assets (LA ict ). The main explanatory variables of interest are Levy ct , a dummy variable that equals one if a bank levy is in place in country c at time t, and CIT ct , the corporate income tax rate in country c at time t. Capturing bank levies by a country-specic dummy variable is a very crude proxy and ignores that some countries implement dierent levy designs and exclude, for example, small banks from the levy.
Thus, in Section 4.2, we assess in more detail the role of the levy design and, in further robustness tests, we account for the role of bank size.
Based on theoretical considerations and empirical results from previous literature, we expect the direct eect of a bank levy on leverage, β 1 , to be negative, whereas the direct eect of CIT, β 2 , is supposed to be positive. The total eect of bank levies on leverage, depending on the CIT, is given by β 1 + β 3 * CIT . To investigate how eective bank levies are at counteracting the debt bias of taxation, our coecient of interest is β 3 , i.e. the interaction eect between the bank levy and the corporate income tax rate.
Supposed that leverage is lower for banks that are aected by a levy relative to banks that are not (β 1 < 0), then the larger and positive β 3 is, the more the leverage-reducing eect from the levy is mitigated with higher CIT rates.

15
The vector X ict−1 contains bank characteristics, all lagged by one year to account for potential simultaneity concerns. 12 Following the literature on bank capital structure and taxation (Devereux et al., 2019;De Mooij and Keen, 2016), we include the log of total assets (in million USD) and the square of the log of total assets to control for bank size and for the fact that large banks display higher leverage, the return on assets (in %) to measure protability, and the ratio of non-performing loans to gross loans (in %) as a measure of bank risk. The term Y ct summarizes annual GDP growth, ination, and regulatory variables, that is, country-level controls.

Regression results
This section discusses estimation results for the baseline sample including bank observations from EU-countries, using banks from those countries that introduced a levy as the treatment group and the remaining banks as the control group. We then limit the sample to countries with a more homogenous levy design, before testing the robustness of our ndings with respect to additional changes in the sample composition. Finally, we test the sensitivity of our results to changes in the Basel regulatory regime as well as individual banks' level of capital.  relative to total assets (in %). Bank levy is a country-level dummy variable that is one if a bank levy is in place and zero otherwise. Corporate tax rate is a continuous variable, also dened at the country level. Bank-level controls are included with a lag and standard errors are clustered at the bank level.
(1) The estimated direct eect of the bank levy dummy in column 4 implies that for banks in countries with a levy in place, the liabilities to assets ratio is 4.  The estimated coecients on the control variables are in line with the related literature. Bank leverage increases with bank size, but this eect levels o and turns negative for the largest banks. Higher protability allows banks to accumulate equity, such that leverage declines. Bank risk, as measured by the ratio of non-performing loans to gross loans, ination, and institutions mitigating moral hazard, do not seem to systematically aect leverage, whereas leverage tends to be higher during booms but lower in countries where supervisors have less discretion if banks violate the laws (higher values of the variable supervisory forbearance discretion).
To rule out pre-event trends and to investigate how leverage evolves after the introduction of bank levies, we study the dynamics of leverage relative to the introduction of a bank levy. Following Beck, Levine, and Levkov (2010), we dene dummies that equal zero except for one year before or after the introduction of the levy in a country.
We consider four years pre-introduction given that our sample starts in 2006 and most countries start implementing the levy in 2010/11. The post-event window spans three years given that our sample ends in 2014. Leverage is then regressed on this set of dummies, as well as on bank and year xed eects. Standard errors are clustered at the bank level. Figure 4 shows that pre-introduction no signicant trend in leverage emerges. Immediately after the levy introduction, we nd signicant eects revealing a downward trend in leverage. The importance of the levy design Since the design of bank levies diers across countries, in a next step, we split the regression sample according to the tax base of the levy. For those banks subject to a levy designed as a liabilities tax (L-D), theory predicts a negative link with leverage as a liabilities tax makes debt nancing more expensive (Devereux et al., 2019). However, for banks aected by levies with a dierent tax base, like risk-weighted or total assets (Finland, Hungary, Slovenia), deposits (Cyprus, Ireland, Spain), or minimum capital requirements (France), the impact on leverage is not clear. To account for dierent levy regimes, we exclude, for example, bank observations of those countries that implemented a levy but did not design it as a liabilities tax over the whole sample period (compare also Appendix B).
The estimation results in Table 6   Bank levy is a country-level dummy variable that is one if a bank levy is in place and zero otherwise.
Corporate tax rate is a continuous variable, also dened at the country level. Column 1 repeats the baseline results from Table 5, column 4. Columns 2-5 show the estimates for subgroups with regard to the levy design. The estimation sample covers countries with the respective levy tax base and countries that never implemented a levy. Bank-level controls are included with a lag and standard errors are clustered at the bank level. ***, **, and * indicate signicance at the 1%, 5%, and 10% level, respectively.
(1) in these cases. Thus, the leverage-reducing eect of bank levies is more pronounced for the L-D design, i.e. for pure liabilities taxes. Our ndings that bank levies only matter less in countries with a high CIT rate and that the eect of levies is stronger for countries implementing the L-D-design strengthen the argument brought forward in the related literature that bank levies reduce leverage through increased costs of debt (e.g. Célérier et al. (2018)).
In terms of economic signicance, the estimated eects of the levy for the L-D-sample are unsurprisingly a bit larger compared to the eects for the full sample including all levy types: Table 6, column 2 reveals that leverage is 5 percentage points lower in countries with a liabilities tax in place. For those countries with the lowest CIT rates in the sample (10%), a levy leads to a reduction in leverage of 3.9 percentage points, whereas under the highest CIT rates (37%), a tax on liabilities still somewhat mitigates leverage (-0.9 percentage points relative to banks not subject to a levy).
Thus, when comparing the results from Tables 5 and 6, it appears that bank levies that are designed as a tax on liabilities are more ecient in incentivizing a more stable bank funding structure, even for higher CIT rates.
Overall, the estimation results point to a favorable eect of bank levies on capitalization and this is the more so, the smaller the debt bias of taxation. For very high CIT rates, the resulting incentives for debt nancing exceed the incentives from the bank levy to reduce leverage, such that the overall eect of the levies turns insignicant in these countries. Not surprisingly, the strengths of the levy-eect and, hence, its eectiveness to foster nancial stability through lower leverage depends on levy design.

Potential sources of endogeneity
Regarding potential endogeneity issues, one could be concerned about reverse causality, meaning that high bank leverage drives the introduction of bank levies. However, this would imply a positive link between leverage and the introduction of bank levies, whereas we nd a negative relationship between the two variables. Thus, our estimates would be biased downwards, such that they reect a conservative estimate of the eect of levies on leverage if we do not fully control for reverse causality. Additionally, many countries did not primarily aim at inuencing bank capital structure with the introduction of bank levies, but rather at lling bank resolution and restructuring funds. Lastly, we consider leverage at the bank level but control for the introduction of the levy at the country level. This approach lowers concerns about reverse causality 26 as individual banks might not drive the outcome of the regulatory process.
A further concern could be related to anticipation eects. For example, anticipating the introduction of bank levies, banks might, pre-introduction, lower leverage ratios in order to reduce regulatory costs. However, as bank levies were introduced quickly in most countries after rst political discussion (see Section 3.2) and partially refer to balance sheets of years preceding the introduction (see e.g. Buch et al. (2016), Devereux et al. (2019)), it is unlikely that banks already adjusted their capital structure before the introduction. Again, such anticipatory adjustments would rather bias our results downwards because we would underestimate the full decline in leverage.
Finally, with respect to confounding factors that inuenced bank capital structure at the same time as levies, we control for a large set of potential candidates. Disruptions due to the nancial crisis, the European sovereign debt crisis, and expansionary monetary policy aecting all banks alike are captured by time xed eects. Country-level macroeconomic developments, which obviously diered across the sample countries, and dierences in the stance of regulation in the banking sector are controlled for by including a corresponding set of variables and country-level regulatory controls as described above.
As a response to the nancial crisis, the regulatory framework has been reformed substantially with potential eects on banks' capital structure. However, our sample ends in 2014, whereas regulatory changes with respect to capital and liquidity requirements under Basel III were subsequently phased-in. Also, as concerns the establish-

Robustness tests
We run several robustness checks in order to test the sensitivity of our results with respect to the timing and limits of bank levies (Table 7). Moreover, we analyze the sensitivity of our results to bank capital regulation (Table 8).
We rst assess whether changes in the levy rate, the timing of the levy introduction, and the included banks and countries impact on our ndings. Table 7 reveals that our baseline results (Table 5, column 4) are driven by banks in countries with an increasing levy rate over time (column 2) and by banks that were subject to bank levies early on (2012 or earlier, columns 3 -4). 17 The latter result might reect that levies have been most eective in countries implementing them relatively quickly after rst political discussions such that banks could not adjust ex ante.
To test whether the composition of banks and countries matters for our results, we account for the fact that in many countries, smaller banks face exemptions from the levy. For the four countries which have an explicit lower limit on the tax base, namely Austria (1 billion Euros), Germany (300 million Euros), the Netherlands (20 billion Euros), and the UK (20 billion Pounds), we use the Bankscope data to calculate the tax base as closely as possible. We then exclude those banks from the sample that fall below the respective limit and hence do not have to pay a levy. The results remain robust (Table 7, column 5) -only the magnitude of the coecients slightly changes.
Alternatively, we adjust the levy dummy and set it to zero for those banks falling below the limit while including them in the sample (column 6). The coecients stay highly signicant but again change slightly in magnitude.
17 Sweden increased its levy rate in 2011 from 0.018% of non-deposit liabilities to 0.036%. Austria, Cyprus, France, Hungary, Latvia, and the United Kingdom also increased their levy rates (Budnik and Kleibl, 2018). The dependent variable is total liabilities relative to total assets (in %). Bank levy is a country-level dummy variable that is one if a bank levy is in place and zero otherwise. Corporate tax rate is a continuous variable, also dened at the country level. For comparison, the baseline results from Table 5 are reported in column 1. Column 2 restricts the sample to banks in countries increasing the levy rate over time, while columns 3-4 present results for subsamples of countries that introduced levies relatively early. Column 5 excludes banks in Austria, Germany, the Netherlands and the United Kingdom that fall below the tax base limit. In column 6, the levy variable is set to zero for banks in countries that implemented a levy but which fall under the lower limit of the tax base. Column 7 adds a triple interaction with a size indicator being one if a bank has a mean value of total assets larger than the country mean and zero otherwise. All models include bank-level and country-level controls, as well as bank and time xed eects. Bank-level controls are included with a lag and standard errors are clustered at the bank level. ***, **, and * indicate signicance at the 1%, 5%, and 10% level, respectively.
(1) As a more sophisticated test for the role of smaller versus larger banks across countries, we add a triple interaction term of the bank levy, the corporate tax rate and an indicator variable being one for large banks and zero otherwise. The indicator variable is dened at the country level to account for country-specic designs of levies. It turns one if a bank has an average value of total assets that is larger than the country mean.
Results show that large banks do not respond dierently than smaller banks (column 7) -the coecient of the triple interaction term is statistically insignicant.
Second, we extend our tests to capture dierent regulatory regimes (Table 8)  The dependent variable is total liabilities relative to total assets (in %). Bank levy is a country-level dummy variable that is one if a bank levy is in place and zero otherwise. Corporate tax rate is a continuous variable, also dened at the country level. For comparison, the baseline results from Table 5 are reported in column 1. Column 2 excludes the year 2014. Columns 3-6 add controls for regulatory changes. Column 7 adds a triple interaction with an equity indicator being one if a bank has a mean value of the equity ratio larger than the sample mean and zero otherwise. In column 8, we add interactions of bank group and time xed eects to the baseline model. Bank groups are based on the quartiles of bank equity ratios. All models include bank-level and country-level controls, as well as bank and time xed eects. Bank-level controls are included with a lag and standard errors are clustered at the bank level. ***, **, and * indicate signicance at the 1%, 5%, and 10% level, respectively.
(1) The results presented in columns 3 -6 remain remarkably robust, while tighter capital or macroprudential regulation stand in a negative relationship with the total liabilities to assets ratio. However, only the coecient of the variable indicating that a countercyclical capital buer is in place shows a signicant and negative coecient.

18
The reason for the robustness of our results when adding controls for capital regulation is most likely that capital regulation was applied quite uniformly across European countries and is thus captured by time xed eects. Moreover, Basel III was slowly phased-in only at the very end of our sample period. Hence, signicant eects on leverage might only be seen in the following years.
To control for dierences in capitalization at the bank level, we created an indicator variable that takes on a value of one for all banks that have an average equity ratio larger than the sample mean and zero otherwise. This indicator captures whether a bank is less capital constrained compared to the sample mean (and thus less under pressure once capital regulation is tightened). In Table 8, column 7, we add a triple interaction of our interaction term of interest (C IT t * Levy t ) and the high equity indicator. For low equity banks, we nd signicant results similar to the baseline case. Comparing results to high equity banks, the coecient of the triple interaction term is signicantly positive implying that banks with an equity ratio above the sample mean show a dierent sensitivity than banks with lower equity ratios.
As before, the leverage reducing eect of the bank levy is lower with higher corporate income taxes. This is the more so for banks with higher equity ratios. Thereby, the total eect of the bank levy on leverage is close to zero for banks with a low equity ratio for an average CIT rate of 30% and approximately minus one for banks with a high equity ratio. Hence, banks with an equity ratio above the sample mean are more responsive to bank levies, which is similar to the result by Devereux et al. (2019) that low-risk banks increase equity ratios more than high risk banks in response to a levy.
One potential reason could be that safer banks, i.e. those with a higher equity ratio, 18 The insignicance of the coecients of these variables is in line with Gropp and Heider (2010) who nd that capital regulation is not that important when it comes to banks' capital structure.
nd it easier to adjust to new regulation by deleveraging as raising capital is less costly than for risky banks.
Since adjustments in leverage due to changes in nancial regulations or responses to the global nancial crisis may have been heterogeneous across banks with dierent capitalization, we follow Devereux et al. (2019) and Kogler (2019) and account for bank group-specic time trends. For that goal, dummy variables are computed for each quartile of the equity ratio for the entire sample and are then interacted with year dummies (column 8). Adding these bank group-time xed eects does not aect the sign or statistical signicance of the baseline results, but reduces the size of the coecient on the bank levy dummy as well as the coecient on the interaction with the CIT rate.
Appendix C provides additional estimations to test the sensitivity of our results with respect to model specication.    (3). The dependent variable is total liabilities relative to total assets (in %). Bank levy is a country-level dummy variable that is one if a bank levy is in place and zero otherwise. Corporate tax rate is a continuous variable, also dened at the country level. For column 1 and columns 4-6, the estimation period covers the years 2006-2014. Columns 2-3 show estimates for alternative sample periods. Column 4 interacts the bank levy with the average corporate income tax at the country level. Column 5 adds as a control the tax to pre-tax prot ratio and column 6 adds the countries' implicit tax rate on corporate income. All models include bank-level and country-level controls, as well as bank and time xed eects. Bank-level controls are included with a lag and standard errors are clustered at the bank level. ***, **, and * indicate signicance at the 1%, 5%, and 10% level, respectively. (1) (2) (3) (4) (5)   (3). The estimation period covers the years 2006-2014. The dependent variable is total liabilities relative to total assets (in %). Bank levy is a country-level dummy variable that is one if a bank levy is in place and zero otherwise. Corporate tax rate is a continuous variable, also dened at the country level. Column 2 includes a triple interaction with government ownership, column 3 with a GIIPS indicator. In column 4, bank holdings are excluded. Columns 5 and 6 add controls. All models include bank-level and country-level controls, as well as bank and time xed eects. Bank-level controls are included with a lag and standard errors are clustered at the bank level (excluding the last column). ***, **, and * indicate signicance at the 1%, 5%, and 10% level, respectively. (1)