Kicking the Can Down the Road: Government Interventions in the European Banking Sector

Abstract

Governments in an economy whose banking sector exhibits systemic distress have two types of interventions at their hand: systemwide measures affecting the banking sector as a whole and single-bank measures aimed at banks most in need (Farhi and Tirole 2012). In most cases, systemwide measures are performed by monetary authorities in the form of conventional policy (i.e., lowering interest rates) and/or unconventional policy (e.g., larger-scale asset purchase programs, such as TARP). In contrast, single-bank measures are usually conducted by the fiscal authority with either immediate incidence of fiscal costs or using government guarantees. Following Pazarbasioglu et al. (2011), bank- level measures in recent banking crises can be grouped into three categories, which are usually implemented sequentially as a crisis worsens: (A) guarantees, (B) capital injections, and (C) asset restructuring and/or resolution. While step A implies a short-run fiscal cost close to zero, steps B and C typically require governments to run a higher fiscal deficit, which has to be financed with higher debt or taxes. Therefore, fiscally constrained governments may deploy guarantees and/or engage in some form of forbearance (e.g., relax capital requirements or and asset quality recognition norms), and, in particular, decide not to implement step B.¹

We investigate these government interventions choices in the context of the Global Financial Crisis (GFC) and its impact on the European banking sector. While banks across all European countries were in distress, no centralized scheme at the European level existed to provide aid to individual banks. Therefore, bailout decisions were subject to the discretion and the fiscal constraints of the national governments.

Our analysis of government interventions builds on a novel, hand- collected data set of all aid measures granted to eurozone banks during the 2007 to 2009 period. A key measure of fiscal capacity is the country’s ratio of government revenues to gross domestic product (GDP) (e.g., Dincecco and Prado 2012). Higher revenues increase the capacity to recapitalize banks in distress (Stavrakeva 2020). Another widely used measure for fiscal strength is the total debt stock as a percentage of GDP (e.g., Demirgüç-Kunt and Huizinga 2013). A high government debt level can imply a tight budget, especially if debt is short term and has to be refinanced in the near future. We thus also include the proportion of maturing debt of a country as a relevant fiscal metric. In addition, we employ the current account surplus versus deficit as a potential determinant as fiscal constraints are likely to become more binding when a country borrows from abroad.

We use a bank-level hazard model to analyze the time until the first government intervention for a distressed bank. We show that banks located in countries with lower fiscal capacity were at least as likely to receive any form of government support as banks located in countries with stronger public finances.² However, consistent with the hypothesis that capital injections are costly in the sense that they tighten the government budget constraint in the short run, fiscally constrained governments delayed or suspended capital injections more than fiscally stronger countries. The effect is economically significant. For instance, the likelihood that a bank is recapitalized increases by about 30% when the sovereign’s revenues-to-GDP ratio increases by 1 percentage point (pp). The result is robust across different measures of fiscal capacity and holds after controlling for an array of bank-level, banking sector-level, and macro-level variables, as well as political control variables, such as CAMEL-type bank-level controls, too-many-to-fail effects (Acharya and Yorulmazer 2007; Brown and Dinç 2011), election cycles (Brown and Dinç 2005), and other factors.

In a next step, we investigate portfolio and lending decisions of banks that remained undercapitalized at the end of 2009, that is, after the GFC. A key identification challenge is that undercapitalization itself is endogenous and depends on both precrisis bank characteristics, that is, banks’ predisposition to require a bailout, and the ability and willingness of governments to bail out banks. To address this challenge, we use an econometric method developed by Hirano, Imbens, and Ridder (2003), and used, among others, by Jordà and Taylor (2016), called “inverse probability weighting.”

This method does not produce a classification of undercapitalized banks, rather it requires this information as an input. To that end, we classify a bank as undercapitalized if one of the following three conditions is met: (1) the Tier 1-capital ratio is below 8%, or, if these data are not available, (2) the equity-to-assets ratio is below 3% (the BCBS³ leverage ratio requirement) or (3) the nonperforming loans (NPL)-to- total-loans ratio is in the top 5% of all banks in our sample at the end of 2009.⁴ We then regress this indicator variable of a bank being undercapitalized on a set of bank and country characteristics (and their interaction terms) that we found to be important determinants in our first test of whether or not a bank is recapitalized. Since this regression captures the factors that were on average important in determining banks’ capitalization status in 2009, the difference between the prediction of the model and the actual outcome for a banking sector can be interpreted as the degree of the country’s governmental discretion. For example, consider two banks that are similarly weak on the basis of their bank-level characteristics, one located in Germany and the other one in Ireland, neither of which received a bailout, and both end up undercapitalized. Given the higher level of fiscal capacity, the degree of forbearance for the bank in Germany would be considered higher than the degree of forbearance for the bank in Ireland, where fiscal capacity was more likely the binding constraint.

Reweighting the sample with weights corresponding to the extent of governmental discretion implied by the bank’s outcome then allows us to put the spotlight on those banks that were still undercapitalized because of forbearance. From a statistical point of view, the weights allow us to reduce (or even eliminate) the bias from endogenous treatment assignment in the subsequent treatment effects models.⁵

Armed with these weights, we investigate the effect of being undercapitalized on bank-level outcomes during the period 2009 to 2012. Our second main result is that banks that were undercapitalized at the end of 2009 were more vulnerable or financially unstable. Over the next 3 years, undercapitalized banks lost further equity capital, reduced lending, but increased their loan loss provisions compared to their better-capitalized peers. Undercapitalized banks also increased their short-term borrowing from the European Central Bank (ECB) using its 3-year long-term refinancing operation (LTRO) facility introduced in 2011. Undercapitalized banks did not, however, have a higher probability to default, likely because their liquidity was secured via ECB funding.

We then analyze individual lending decisions by banks using the intersection of loan-level data from the Thomson Reuter DealScan and Bureau van Dijk’s Amadeus database to identify both banks and firms. Using a regression framework similar to Khwaja and Mian (2008), which captures firm demand for loans using firm fixed effects, we find that undercapitalized banks reduced loan supply to nonfinancial firms, both for relationship borrowers (intensive margin) and for new borrowers (extensive margin). Investigating the effect on lending as a function of borrower risk, we find that undercapitalized banks significantly reduced their loan supply to risky firms relative to better-capitalized banks.

This is intuitive as riskier lending binds (regulatory) capital. Interestingly, however, we do not find evidence that undercapitalized banks reduced lending to risky relationship firms, suggesting an evergreening of loans to “zombie” firms. We test this hypothesis directly using the definition of “zombie” firms in Acharya et al. (2018). This definition requires a low-credit-quality firm to be receiving subsidized credit, that is, paying an interest rate below the average for highly rated borrowers of the same industry. We find that undercapitalized banks increased the supply of credit to “zombie” firms while reducing lending to “nonzombie” firms, relative to better-capitalized banks. We also provide evidence that the “zombie” firms that are matched with banks that become undercapitalized in the lending market, perform worse than similar firms matched with better-capitalized banks.

Moreover, we investigate a behavior we call “search-for-yield lending”. Within a risk class, and therefore within a regulatory risk weight and capital cost category, undercapitalized banks gamble for resurrection by seeking borrowers accepting a higher interest rate. That is, for the same cost, undercapitalized banks prefer the potential upside relative to the (within-rating category) risk more than better-capitalized banks do. We find strong evidence for such behavior in the intensive margin of lending of undercapitalized banks.⁶

Finally, we examine the composition of assets on undercapitalized banks’ balance sheets. We observe that they shift a considerable part of their portfolio from real sector lending to government bonds during the period from 2009 to 2012. The average undercapitalized (better-capitalized) bank reduces (increases) its lending portfolio by 1 pp (2 pp) of total assets and increases (also increases) its security exposure by 5.5 pp (0.25 pp) of total assets. These heightened government bond purchases by undercapitalized banks are particular to the years 2011 and 2012, when several government bond yields spiked at the onset of the European sovereign debt crisis.

Our results are robust to alternative inverse probability weights that are based on other fiscal variables or timings. Not using any weights and thus treating the undercapitalization status as exogenous, however, reveals that our loan-level results could not have been uncovered. It therefore seems to be the case that the perverse lending incentives are particularly strong for banks whose undercapitalization is a consequence of forbearance. Hence, the fact that governments “kicked the can down the road” on banking sector repair affected subsequent outcomes in two ways. First, they delayed fiscal costs, which had to be borne in the subsequent sovereign debt crisis in the form of larger amounts of quasi-fiscal central bank support and a weakened credit supply to the real economy. Second, they provided perverse lending incentives to banks on which they exerted forbearance, leading to misallocation of capital in the form of “zombie” lending and search-for-yield behavior.

1. Related Literature

Our paper contributes to several strands of literature. First, it relates to the literature on regulatory forbearance that dates back to at least the 1980s and the discussion of “zombie thrifts” in the United States by Edward Kane and other authors, showing that regulatory forbearance is not a new phenomenon but one that has played out over decades (see, e.g., Kane 1989 and references therein). More recent research in this area focuses on the drivers of (regulatory) forbearance, for example, why governments do not intervene in the banking sector, even though it would be optimal from a general welfare perspective. Governments postpone the resolution of distressed banks if the banking sector has many weak banks (Acharya and Yorulmazer 2007; Acharya and Yorulmazer 2008; Kroszner and Strahan 1996; Hoshi and Kashyap 2010; Brown and Dinç 2011) or for political economy reasons, such as timing in electoral cycles (Brown and Dinç 2005; Imai 2009; Bian et al. 2017). Our paper highlights empirically—as posited by some of this literature—that fiscal capacity is an additional driver behind (regulatory) forbearance. Like most of the previous literature, we investigate the implications of (regulatory) forbearance. Gropp, Rocholl, and Saadi (2017), for example, show that regulatory forbearance in the United States—because of the Federal Deposit Insurance Corporation’s (FDIC) decision not to let banks fail— affects growth and employment in some regions; we show that a sovereign’s debt overhang can significantly impede an undercapitalized banking sector’s recovery after a financial crisis, especially its financial stability, credit supply, and risk-taking incentives.

Second, our paper adds to the growing literature investigating the cost- benefit trade-offs involved in government interventions in the banking sector. The main benefit is that recapitalizations help alleviate negative externalities from failing the severely undercapitalized banks (Diamond 2001). Costs mainly comprise large fiscal outlays (Acharya, Drechsler, and Schnabl(2014)) and moral hazard arising from bailout expectations (Mailath and Mester 1994; Dam and Koetter 2012; Fischer et al. 2014).⁷ Several papers analyze this trade-off during the GFC, focusing predominantly on the Capital Purchase Program in the United States (Veronesi and Zingales 2010; Bayazitova and Shivdasani 2012; Li 2013; Duchin and Sosyura 2014; Berger, Makaew, and Roman (2019); Black and Hazelwood 2013). Evidence from the United States suggests that recapitalizations stabilized bank lending growth, but also increased lending to riskier borrowers. In contrast, we investigate government interventions during the European financial and sovereign debt crisis.⁸Homar (2016) shows that timely bank recapitalizations reduce the duration of recessions using an international sample of banking crises. We provide new evidence that government interventions need to be large enough to overcome banks’ debt overhang problems, a theme reminiscent of the work of Caballero, Hoshi, and Kashyap (2008), Diamond (2001), Gianetti and Simonov (2013), and Brei, Gambacorta, and von Peter (2013).

Finally, our paper relates to the literature on the role of bank capital, particularly during financial crises. Berger and Bouwman (2013) document the importance of capital for banks, particularly medium- and large-sized banks, during crises. Several studies document that higher capital was associated with lower probability of bank failure during the 1990 credit crunch (Cole and Gunther 1995; Estrella, Park, and Peristiani 2000; Wheelock and Wilson 2000) and during the 2008–2009 financial crisis (e.g., Cole and White 2012; Berger, Imbierowicz, and Rauch (2016)). Beltratti and Stulz (2012) find that bank capital is key to understanding bank performance during the subprime crisis, and Fahlenbrach, Prilmeier, and Stulz (2012) show that poorly capitalized banks during the Russian debt crisis also performed poorly during the subprime mortgage crisis. By evaluating aggregated time series of sovereign and financial shocks simultaneously, Manzo and Picca (forthcoming) show that fiscal capacity of governments is an important determinant of sovereign shocks, which in turn spill over to the financial sector. We show on a disaggregate level that banks that were left undercapitalized by their governments during the GFC were more likely to eventually require greater government support, performed worse, lent poorly, and searched for yield in portfolio composition decisions.

2. Data

2.1 Government interventions

This paper builds on a novel, hand-collected data set comprising all government interventions for eurozone banks over the 2007 to 2012 period. Our primary data source is the State Aid Register of the European Commission (EC), which contains detailed information on government interventions in the European banking sector. The Treaty on the Functioning of the European Union (TFEU) generally prohibits government support to individual companies but government support can be admissible in exceptional cases, such as to “remedy a serious disturbance in the economy of a Member State” (TFEU Article 107(3.b)). Any such exception must be reviewed and approved by the EC on a case-by-case basis and is documented in the State Aid Register.

While the State Aid Register collects government interventions in the entire EU, we restrict our sample to eurozone banks to ensure that all banks in our sample have equal access to the ECB facilities (including nonstandard monetary policy measures, such as the LTRO).⁹ Since the LTROs were provided with full allotment in our sample period, there was no heterogeneity in the access to the ECB funding across banks. Thus, we do not expect our results to be biased by the existence of LTRO.

We start building our database by manually extracting information from all State Aid cases listed in European Commission for the 2007 to 2012 period.¹⁰ Government support can be approved for one of two cases: (1) as an ad hoc support measure to an individual bank or (2) as a sectorwide scheme making available a maximal amount for a certain aid measure and being accessible to eligible banks.¹¹

To ensure confidentiality, the Commission does not make available all details of government support measures in the State Aid Register. Also, decisions on sectorwide schemes do not contain information on individual beneficiaries. Therefore, when necessary, we augment the data with information from banks’ press releases, information from banks’ regular reporting activities, regulators’ and central banks’ reports, and newspaper articles. For every State Aid case number, we further cross- check whether approved intervention measures have been implemented.

As in Laeven and Valencia (2008), we classify government support into four categories: (1) recapitalizations, (2) guarantees, (3) other liquidity support, and (4) troubled asset relief.¹² Recapitalizations comprise all measures involving government-funded capital increases and conversions of existing capital or hybrid instruments into higher-order capital instruments.¹³ Guarantees comprise all government guarantees on nondeposit liabilities, including both existing and newly issued liabilities. Other liquidity support comprises all interventions other than guarantees that are targeted at stabilizing a bank’s liquidity.¹⁴ Finally, troubled asset relief programs are government interventions targeted at removing impaired or defaulted assets from a bank’s balance sheets by means of asset sales or guarantees.¹⁵

2.2 Bank-level and macro-level data

2.2.1 Sample construction

We obtain bank-level financial data for the 2007 to 2012 period from the Bureau van Dijk Bankscope database. Consistent with the literature (e.g., Sufi 2007), all information is aggregated to the ultimate parent level using shareholder information from Bankscope and various other sources. We remove all banks that receive a government intervention but cannot be matched to the Bankscope database. We also drop banks whose ultimate parent is not incorporated in a eurozone country, as the propensity of a bailout for these banks likely depends on the parent’s home country. The data set is further constrained to large banks and those of domestic importance, those whose failure creates a threat of financial contagion or has a large negative impact on the domestic economy. That is, we keep banks with a market share larger than 1% (measured in bank size/size of the national banking sector), with size of at least 10% of GDP, balance sheets larger than €1 billion, or banks that are among the five largest banks in the country.

We further exclude banks with very high Tier 1 ratios (⁠|$>$|30%) or equity-to-assets ratios (⁠|$>$|20%). All those cleaning steps leave us with a sample of 830 banks, of which 76 received at least one form of government intervention. Finally, we augment our data with country-level variables from Eurostat, the World Bank, and the International Monetary Fund (IMF).

2.2.2 Summary statistics

Panel A of Table 1 for the baseline year 2007 shows cross-sectional summary statistics for bank-level variables. Banks show considerable variation in their overall condition prior to the financial crisis. For example, the equity-to-assets ratio (⁠|$Equity/TA$|⁠) has a cross-sectional mean of 6.51% with a standard deviation of 2.75%. There is also considerable variation in other variables, such as loan loss provisions (⁠|$LLP/Loans$|⁠) and NPLs (⁠|$NPLs/Loans$|⁠). Cross-sectional summary statistics for macro-level variables in 2007 are shown in panel B of Table 1. The variation in current account balances is striking: it ranges from a current account deficit of -14.0% to a current account surplus of 9.9%. Similarly, the maturing government debt as a share of GDP ranges from 1.2% to 18.1%.

2.3 Loan-level and firm-level data

We obtain loan-level data from the Thomson Reuters LPC DealScan database, which provides detailed information on European syndicated loans including information on lenders as well as loan contract terms. For banks to be included in the sample, we follow the previous literature (e.g., Ivashina 2009; Heider, Saidi, and Schepens 2019) and require that banks must serve as lead arranger in the syndicate.¹⁶ If the loan allocation between syndicate members is unknown, we divide the loan facility equally among syndicate members. Also following the previous literature (e.g., Acharya et al. 2018; Gropp et al. 2019), we transform the data and calculate the annual outstanding exposure of bank |$b$| in country |$c$| to nonfinancial firm |$j$|⁠, using the maturity information on each loan at the end of each year.

We hand-match DealScan lenders to Bankscope at the ultimate parent level and match DealScan borrowers in our sample to firms in the Amadeus database. The final loan-level sample comprises 209 banks that arrange loans to 8,321 nonfinancial firms.¹⁷

3. Do Weak Governments Delay Interventions?

Governments may postpone recapitalizations by issuing rolling guarantees, by injecting just enough capital to avoid immediate insolvency, or by allowing banks to hide their losses. This section investigates the determinants of a government’s decision not to resolve a bank’s debt overhang immediately, but to practice (regulatory) forbearance. We use Cox regression models to formally investigate the role of a country’s fiscal capacity and the overall capitalization of the banking sector for the timing and type of an intervention.

3.1 Methodology

Theory suggests that forbearance and postponing costly capital interventions is an attractive alternative for fiscally constrained governments as new debt can only be issued at the expense of the sovereign’s creditworthiness (Acharya, Drechsler, and Schnabl 2014). Based on this theory, we ask two questions. First, are fiscally constrained governments as likely as unconstrained countries to provide recapitalizations? Second, are they equally likely to support distressed banks, when we do not take into account the type of support (recapitalization, liquidity support)?

Fiscal capacity. The main determinants for our model are measures of fiscal capacity. Different proxies have been proposed in the literature. One key measure is a country’s tax revenues (⁠|$Government Revenue$|⁠) expressed as a percentage of GDP (Dincecco and Prado 2012). A larger income increases the capacity to recapitalize banks in distress (Stavrakeva 2020). Another widely used factor is the total debt stock (⁠|$Debt/GDP$|⁠) also measured in units of GDP (e.g., Demirgüç-Kunt and Huizinga 2013). However, high total debt is only a potential problem if government revenues are low and/or if debt has to be refinanced. We thus also include a measure for maturing debt, by dividing the stock of outstanding government debt by its average maturity (⁠|$Maturing Debt (%GDP)$|⁠). This allows us to distinguish between countries that have a high outstanding debt stock but a low current debt service and countries with a low stock but a high current debt service, since only the latter should be relevant for forbearance. Lastly, we use the current account surplus/deficit (⁠|$CA Balance$|⁠) as a potential determinant as fiscal constraints might become binding when a country borrows from abroad (Freund and Warnock 2007).¹⁹

Figure 1 shows that these metrics of fiscal constraints show substantial cross-sectional variation, especially between GIIPS and other eurozone countries. In case of debt-to-GDP, government revenues-to-GDP, and current account balance, the differences were large at the time of the onset of the global financial crisis; interestingly, debt-to-GDP (and also maturing debt-to-GDP) worsen for GIIPS relative to other eurozone countries after the global financial crisis, whereas the current account balance improves from being at a deficit toward neutrality for other eurozone countries.

Figure 1

Developments of fiscal capacity: GIIPS versus non-GIIPS countries

GIIPS refers to Greece, Ireland, Italy, Portugal, and Spain, whereas non-GIIPS refers to eurozone countries other than GIIPS. Sources: IMF, OECD, and World Bank.

Open in new tabDownload slide

A further measure that we consider is a country’s current budget balance. This short-term flow measure quickly reacts to dynamics that might be relevant for bailout provision, such as the deterioration of the macroeconomic environment, which the government counteracts with a fiscal stimulus package. Moreover, it might be subject to reverse causality, if some bank aid measures were provided in 2007 as this would immediately and significantly affect the budget balance, while it would only have minor effects on government revenues or the stock of outstanding government debt. Hence, we decided not to include the budget balance in our study.

Banking sector. As a banking-sector-specific variable, we include the average book equity-to-assets ratio (⁠|$Avg. Equity Ratio$|⁠). Brown and Dinç (2011) show that governments are less likely to intervene if the banking sector as a whole is undercapitalized (too-many-to-fail effect). As a further relevant determinant of the need for bailouts, we include the level of household debt over GDP (⁠|$HH Debt/GDP$|⁠). Mian, Sufi, and Verner (2017) show that a high level of household debt is a strong predictor of economic downturns because of the sensitivity of mortgage credit to house price busts, which is exactly what was observed during the GFC. We also include the number of banks that have already received a bailout (⁠|$Number Bailouts$|⁠).

Bank characteristics. We include bank-level characteristics to control for bank health and their differential probability of becoming distressed. Specifically, we include bank size (⁠|$Total Assets/GDP$|⁠), equity-to-assets ratio (⁠|$Equity/TA$|⁠), wholesale funding dependence (⁠|$ST funding/TA$|⁠) and profitability (⁠|$ROAA$|⁠). We hypothesize that larger banks with lower capital ratios are more likely to obtain support. Short-term funding dependence, in addition, renders banks vulnerable to interbank funding freezes. ROAA might be an indicator for a sound business model as well as high precrisis risk-taking. All variables are measured at the end of the year preceding day |$t$| in the hazard model.

Elections. As in Brown and Dinç (2011), we include proxies that relate to the political environment in each country. We include the logarithm of the time until the next election (⁠|$log time to election$|⁠) and an index indicating to what extent the current parliament is supporting the European Union (⁠|$Pro-EU$|⁠).

3.2 Determinants of government bailouts

Table 2 reports the main results. We only show the measures of fiscal capacity; the full specifications are reported in an Internet Appendix to this paper. The dependent variable is |$h_{Recap}$| in panel A of Table 2. While the control variables are included in all regressions, we sequentially include proxies for fiscal capacity.²⁰

As the main determinant of a country’s fiscal strength, we include government revenues in all four regressions. Throughout all specifications, this variable is an economically and statistically significant predictor of recapitalizations. In columns 1 to 3, we add different additional proxies for the financial well-being of a country. First, we add the total debt-to-GDP ratio. While the coefficient shows the right sign—higher debt makes bailouts less likely—the result is not statistically significant. However, for any given debt level, a higher share of debt that has to be repaid in the current year induces immediate budgetary constraints and thus should reduce the incentives to recapitalize banks. We find that a large proportion of maturing debt (in the same year as the bailout decision) decreases the hazard rate and thus the likelihood of a recapitalization (column 2). When looking at the current account balance in column 3, the coefficient again shows an intuitive sign (higher current account surplus predicts higher likelihood of bailouts), but is not significant. In column 4, we run a horse race of all those three explanatory variables on top of the government revenues. We observe that maturing debt stays highly significant and important in size. Moreover, both the current account balance and the total debt level become significant in this specification. A higher current account is associated with a higher likelihood of recapitalizations. A higher debt level is, too, suggesting that after partialling out the effect of the debt service burden, a higher debt level corresponds to a country’s willingness to spend, thus making a recapitalization more likely. Overall, the results consistently show that countries that had more fiscal capacity when entering the GFC in 2008 to 2009 were more likely to recapitalize their banks.²¹

Panel B of Table 2 shows the results. There the dependent variable is Any; that is, we predict the likelihood of any kind of government intervention (recapitalization or liquidity support). None of our measures of fiscal strength is a significant predictor of government interventions.

4. Identifying Undercapitalized Banks at the End of 2009

4.1 Methodology

In the following we want to study the implications of banks leaving the GFC period in a status of undercapitalization. Naturally, this status is by no means exogenous. It depends on the capitalization of the bank before the crisis shock hits, its lending portfolio, profitability, and many other bank-specific factors. Moreover, as shown above, it strongly depends on whether the bank was located in a country whose government was able to provide a recapitalization if needed. That is, the probability of being undercapitalized at the end of 2009 depends on a bank’s performance in precrisis years as well as the fiscal capacity of the local government. To obtain a plausibly exogenous measure of undercapitalization, we therefore have to purge these factors from our measure.

To formalize this idea, we rely on an inverse probability weighting method developed by Hirano, Imbens, and Ridder (2003) and recently used in a time series context by Jordà and Taylor (2016), among others. The basic idea of this method is to remove all observable factors that are associated with the treatment assignment. For this purpose, the treatment probability is estimated and used to reweight the sample in all subsequent treatment effect models to reduce, or in the optimal case even eliminate, the bias from endogenous treatment assignment.

In our case, the bank-level treatment is “being undercapitalized at the end of 2009,” that is, after the GFC. To run a logit model for estimating the treatment probability, we need to construct a binary indicator for undercapitalization. Hence, we define a bank as being undercapitalized if one of the following three conditions holds at the end of 2009: (a) its Tier 1-capital ratio is below 8%²², and if Tier 1 ratio data are not available, (b) its equity-to-assets ratio is below 3% (the BCBS leverage ratio requirement), or (c) its NPL-to-loans ratio is in the top 5% of all banks in our sample. Our results are not very sensitive to the choice of these criteria. Removing the third criterion, or varying the thresholds for the first two criteria, does not qualitatively alter our results.

We then estimate a logit model predicting the treatment status based on bank-level and macro-level determinants (including our measures of fiscal capacity) and interactions of bank-level and fiscal variables as of the end of 2007. The results are robust to using inputs as of the end of 2006 (see Table A1 in the appendix). However, we use bank and government characteristics as of 2007 as they are arguably better predictors of post-crisis outcomes.

The distinction in the weight calculation between treated and nontreated is important. To reduce the endogeneity bias, a higher weight needs to imply a less predictable treatment status. Hence, if our model failed to predict that a bank is going to be treated, we want a higher weight than if the treatment was predicted. Thus, the weight formula for treated banks is a decreasing function of the treatment likelihood obtained from estimating Equation (2), and vice versa for non-treated banks.²³

4.2 Descriptive statistics

We report descriptive statistics in Table 3. Panel A of Table 3 shows which banks received government support during the 2008 to 2009 GFC and which are classified as undercapitalized. Our measure classifies around 10% of the banks in our sample (81 of 830) as undercapitalized. Of those 81 banks, 8 actually received a recapitalization, which therefore seemed to have been insufficient to stabilize those banks. The other 27 recapitalizations we observe were successful in that the receiving banks are not classified as undercapitalized in 2009. These numbers suggest that recapitalizations were, on average, a very prolific tool to stabilize banks.

Panel B of Table 4 shows the share of undercapitalized banks by country at the end of 2009. The countries with the largest share of undercapitalized banks are Ireland, Slovenia, and Italy, whereas France and Germany exhibit the lowest share of undercapitalized banks, which is reasonable as these countries implemented large-scale recapitalization measures and both belong to the group of fiscally strong countries.

4.3 Likelihood of being undercapitalized

Table 4 reports the results of the logit model described in Equation (2) to calculate the IPW. Table 4 shows that larger and better-capitalized banks are less likely to be undercapitalized post-GFC in countries with higher (tax) revenues, that is, in which governments have more fiscal space.

Similarly, columns 1 to 3 show that our other measures for fiscal capacity—total debt, maturing debt, and the current account balance— all interact with bank variables in determining the likelihood of a bank being undercapitalized. Column 4, similar to Section 3, runs a horse race of all fiscal variables, highlighting that all of them have their distinct importance. However, the average variance inflation factor for those fiscal variables in column 4 is around 25, which is beyond any acceptable threshold. Because of the possible multicollinearity of the covariates used in the specification reported in column 4, we use column 2 as our baseline regression model.²⁴

We want to stress the finding that the interplay of fiscal capacity and bank characteristics is a very important determinant of banks’ capitalization outcome. Hence, specifying the inverse probability weighting model the way we did is crucial to fully capture the endogeneity in the undercapitalization status driven by differences in fiscal capacity across countries.²⁵

Overall, the results reported in Table 4 show that being undercapitalized in 2009 is not an exogenous event, but depends on a variety of factors. To purge these factors, we use the IPW obtained from the baseline logit in Table 4, column 2, when evaluating the effect of being undercapitalized on real economic outcomes in the following sections.

Panel A of Figure 2 shows the difference of the average IPW per country and one.²⁶ The higher these values, the less the outcomes can be linked to the observable factors used as explanatory variables in Table 4. This difference therefore helps us to assess the extent of discretion applied by national governments in their recapitalization decision. We find the five GIIPS (Greece, Ireland, Italy, Portugal, and Spain) countries in the top six of the ranking in panel A of Figure 2. Similarly, panel B of Figure 3 shows a scatterplot of the same country-level average of IPW and the government revenue-to-GDP ratio. The plot clearly suggests a negative relationship, that is, fiscally stronger governments exerted less discretion.

Figure 2

Inverse probability weights: Descriptives

IPW = inverse probability weights; AT = Austria, BE = Belgium, DE = Germany, ES = Spain, FI = Finland, FR = France, GR = Greece, IE = Ireland, IT = Italy, LU = Luxembourg, NL = Netherlands, PT = Portugal, SI = Slovenia.

Open in new tabDownload slide

Figure 3

Stylized depiction of endogeneity in recapitalizations

This graph shows a stylized depiction of how recapitalization (“bailout”) of banks depends on the national government. Germany could afford to bailout banks with higher equity-to-asset ratio than Spain and Ireland. Our statistical approach allows us to eliminate this endogeneity, while ignoring this induces estimation bias.

Open in new tabDownload slide

4.4 Understanding inverse probability weights

Figure 3 uses an example to demonstrate how inverse probability weighting can be understood in our setup. We document above that countries provided bailouts to their banks as a function of their capitalization. This is represented by the vertical black lines in the graph, where Ireland, Spain, and Germany are ranked by their respective government revenues from low to high. The vertical black lines imply that Spain and Germany bailed out banks that had higher capital ratios compared to Ireland. That is, comparing an Irish bank (with a 4.25% capital ratio in 2007) to a German bank (also with a 4.25% capital ratio in 2007) at the end of 2009 ignores that Irish banks have never been bailed out with such a high capital ratio; however, a bailout was quite likely for a German bank, arguably because of Germany’s fiscal strength. Therefore, the inverse probability weighting allows us to remove the differences represented by the vertical black lines, which would induce a bias in subsequent treatment effect estimations.

By removing all the bank-specific and country-specific drivers of undercapitalization, as well as their interactions, the resultant weights then give us an estimate of the extent of discretion applied by the governments when deciding about recapitalizations. An undercapitalized bank with a high weight is a bank that should have been bailed out (given its situation) and could have been bailed out (given its governments’ situation), but was not, and vice versa for better-capitalized banks. As highlighted in the section above, these cases of elevated discretion are themselves negatively correlated with fiscal stability. Moving forward, we thus interpret our bank results as the impact of banks being undercapitalized due to governmental discretion linked to fiscal weakness.

5 Undercapitalization and Bank Balance Sheets

Delaying government interventions might cause distressed banks’ health to further deteriorate, as necessary recapitalizations are either omitted or severely limited. Undercapitalized banks likely have insufficient capital buffers to withstand future shocks. Moreover, a debt overhang might increase agency costs due to moral hazard, including risk-shifting (Meckling and Jensen 1976; Diamond 2001) and zombie lending (Peek and Rosengren 2005; Gianetti and Simonov 2013; Blattner, Farinha, and Rebelo 2019).

5.1 Methodology

The dependent variable |$\Delta Y_{i,09-12}$| is the log change in characteristic |$Y$| of bank |$i$| over the 2009 to 2012 period. |$Y_i$| are outcome variables measured at the end of 2012. The set of |$Y_i$| comprises equity-to-assets ratio, Tier 1-capital ratio, gross loans, loan loss provisions, share of NPLs, return on average asset, net interest margin, and the risk weighted assets-to-total assets ratio. Bank-level variables |$X_{i,09}$| include total assets to domestic GDP (⁠|$Total Assets/GDP$|⁠), the equity-to-assets ratio (⁠|$Equity/TA$|⁠), the loans-to-deposits ratio (⁠|$Loans/Deposits$|⁠), and return on average assets (⁠|$ROAA$|⁠), as of the end of 2009. These measures are supposed to capture the state of each bank at the beginning of the evaluated period with respect to its size, health, funding structure, and profitability.

5.2 Undercapitalization and bank balance sheets

Panel A of Table 5 summarizes the results of the balance sheet impact regressions. Columns 1 and 2 highlight that while undercapitalized banks’ equity-to-asset ratios further declined in the years 2009 to 2012, their risk-weighted Tier 1 capital ratio increased. This suggests that these banks did not build up additional equity but instead pushed the risk weights downward by lending less to risky borrowers.

Column 4 shows that the level of loan loss provisions of undercapitalized banks went up distinctly from 2009 to 2012, indicating a badly performing lending portfolio inherited from the GFC period. All these results are highly significant and robust to using alternative IPWs.

Interestingly, NPLs did not increase over the 2009 to 2012 period despite the increase in loan loss provisions (column 5). A possible interpretation is that undercapitalized banks continued to extend loans to these borrowers to avoid writing down their exposures. The return on assets is somewhat lower for undercapitalized banks (column 6). Net interest margins and risk intensity (columns 7 and 8) are unaffected by undercapitalization.

How do undercapitalized banks perform during the sovereign debt crisis? In particular, we investigate three dimensions: whether a company needs a recapitalization, files for insolvency or needs funding from the LTRO introduced in December 2011 and continued in March 2012. Panel B of Table 5 reports the results.

While we find an economically meaningful (but statistically insignificant) positive effect of undercapitalization on the likelihood to be recapitalized, and a lower likelihood to survive, we find a highly economically and statistically significant effect on LTRO in that undercapitalized banks borrow substantially from the LTRO facilities compared to better-capitalized peers. This shows that leaving banks in a state of undercapitalization, by not providing sufficient recapitalizations during the GFC, induced higher funding needs for these banks down the line, again to be borne by governments’ budgets, showing that governments just kicked the can down the road.

The results are not sensitive to the chosen weighting scheme, as panels A and B in Table A2 (contd.) show. An alternative weighting scheme and no weighting scheme at all provide similar albeit economically weaker results with lower explanatory power (⁠|$R^2$|⁠).

6 Undercapitalization and Bank Lending Decisions

Figure 4 shows the “excess reduction” in lending by undercapitalized banks, that is, the reduction in lending relative to all other banks, for all firms (left bars), for “high-risk” firms (middle bars) and for “zombie” firms, in particular (right bars). The grey bars show the descriptive differences in the lending behavior, while the black bars show the estimated coefficients from regressions described later in the text. We leave the formal definition of “high-risk” and “zombie” firms to the main text below. The differences are striking. While undercapitalized banks significantly reduce their lending more relative to other banks, especially to riskier borrowers, they increase lending to “zombie” firms. In other words, the remaining equity capital of already constrained banks appears to be withdrawn from risky, “nonzombie,” firms and tied up in lending to “zombie” firms.

Figure 4

Excess reduction in lending by undercapitalized banks relative to better-capitalized ones

This graph shows the difference between the reduction in lending between undercapitalized and better-capitalized banks (“excess reduction”). Positive values refer to negative loan growth, and vice versa. “Analytical” refers to the coefficient estimates from the regression models in Section 6. “Descriptive” refers to the purely descriptive difference between the lending reductions in the sample. Section 6 defines “Overall,” “High Risk,” and “Zombie.”

Open in new tabDownload slide

6.1 Loan volume

So far we have seen evidence for undercapitalized banks cutting back lending (Figure 4, left bars; Table 5, column 3). In this section, we want to drill down further into the lending decisions by undercapitalized banks by investigating the lending behavior at a more granular level.

We present the results for the baseline specification in column 1 of panel A of Table 6. We find that undercapitalized banks significantly reduce their loan supply, which is consistent with the balance sheet regressions shown above. Undercapitalized banks reduce their loan supply by 14 pp more than better-capitalized banks.