The Impact of Policy Interventions on Systemic Risk across Banks

This paper connects several strands of literature related to studies assessing risk realizations and research focusing on forward-looking risk. First, our paper is linked to research on policy interventions and risk realizations. A number of studies analyze how regulatory policies can help control systemic risk. Considering several international financial crises, Weiß et al. (2014a), for example, find that global systemic risk is significantly influenced by the characteristics of regulatory regimes, explicit deposit insurance schemes and country-wide (macro-) prudential policy interventions (e.g., Karamysheva and Seregina 2020). Recapitalizations can reduce the systemic contribution of banks (López-Espinosa et al. 2012), while liquidity injections can temper the risk incentives of insolvent institutions (Cordella and Yeyati 2003). Buch et al. (2019) show that European banks that received state aid during the crisis are associated with increased systemic importance. However, the U.S. Troubled Assets Relief Program (TARP), based on injections of preferred equity, significantly reduced contributions to systemic risk, particularly for larger and safer banks (Berger et al. 2020). Anginer et al. (2014b) point to the stabilizing role played by deposit insurance arrangements during stress periods but also criticize their destabilizing role in normal times.

Second, our research is related to an extensive literature exploring the forward-looking nature of risk. Numerous papers investigate the moral hazard embedded in deposit insurance schemes (Demirgüç-Kunt and Detragiache 2002; Demirgüç-Kunt and Huizinga 2004; Gropp et al. 2014), recapitalizations (Kane 1995), or liquidity injections (Acharya and Yorulmazer 2007; Farhi and Tirole 2012). Financial help in general may deteriorate the liquidity situation of banks when regulators cannot and/or do not distinguish between illiquid and insolvent financial institutions (Freixas et al. 1999; Repullo 2005). Moreover, rescue packages provided to large banks may incentivize them to engage in highly risky operations (Mishkin 2006), invest in illiquid assets (Cao and Illing 2008) and take on excessive credit risk (Gropp et al. 2014). Additionally, state guarantees provided to efficient banks can enhance their exposure to systemic events (Myerson 2012). On the other hand, the adverse impact of a deposit insurance scheme on systemic risk may diminish if banks hold higher levels of Tier 1 capital (Bostandzic et al. 2014). Homar (2016) highlights the importance of the amount of interventions, presenting empirical evidence that banks that receive large enough capital injections boost the supply of credit, access supplementary funding and improve their balance sheets. Similarly, Giannetti and Simonov (2013) show that a reasonable level of capital injections helped banks increase lending and stimulate investments during the Japanese banking crisis of the 1990s. In the long run, bailouts can increase investors’ expectations of future bailouts (Bayazitova and Shivdasani 2012), especially when banks are considered too big to fail, too interconnected to fail and/or too many to fail (Acharya and Yorulmazer 2007; Brown and Dinç 2011). This could generate enormous costs for shareholders and taxpayers, impede a recovery, and distort competition. Rescued banks may obtain competitive advantages that increase their market power (Berger and Roman 2015), while sound banks that do not receive intervention may increase their loan rates and reduce depositor risk premiums (Koetter and Noth 2016).

We contribute to the extant literature in several ways. First, our foremost contribution is to assess to what extent policy interventions implemented by governments affect systemic risk. Although the isolated impact of the bailout mechanism has been addressed in several theoretical and empirical studies, in comparison with the approaches in other studies, our specifications include several types of interventions with details regarding their volume and their associated restrictions. A large spectrum of the most important policy interventions made by European member states and implemented at the bank level during 2008–2014 is examined: state guarantees, recapitalizations (capital injections) and liquidity injections. The dataset is collected manually from banks’ annual reports, financial statements, websites, and the State Aid Register of the European Commission. Our empirical approach is also different. First, we are interested in investigating the short-run response of systemic risk to policy interventions; in addition, we examine the long-run implications of this relation. This distinction is particularly important for European countries, as governments were not always perceived as trustworthy providers of financial support, which might lead to a delayed effect of interventions.

Second, we contribute to the literature on the restrictions associated with bailouts. In the long run, a tradeoff between too-big-to-fail (TBTF) protection and the regulatory burdens associated with bailouts might come into play. We employ a hand-collected dataset of regulatory restrictions and assess their impact on the relation between policy interventions and systemic risk. Specifically, we consider the following set of restrictions imposed by regulators on the banks receiving intervention during the duration of bailouts: supervisory board intrusions, management pay limitations, and capital payout bans.

Third, we add to the literature by examining the interaction between policy interventions and banks’ risk strategies. We explore how banks’ risk postures related to size, leverage, and profitability exacerbate or mitigate the relation between rescue actions and systemic risk. Especially in the long run, banks are more likely to manifest moral hazard behavior by adjusting their risk strategies.

To account for the selection of banks into treatment (government intervention) on their unobservable systemic importance, we employ the two-stage procedure of Heckman (1979). This approach is in line with the previous literature (Berger et al. 2020; Gerhardt and Vander Vennet 2017; Kick et al. 2016). In the first stage, we use a Probit model to estimate the probability of a bank receiving policy interventions, as represented by the equation: where P_{Intervened ij,t} represents the probability that bank i receives (one or more) policy interventions in quarter t, Identifying restrictions_ij,t-1 is a set of exclusion restrictions that explain the decision of government to provide financial assistance to bank i, and Bank controls_ij,t-1 represents differences in risk profiles among banks (size, leverage, credit risk, liquidity and profitability). To account for heterogeneity among different banking systems and economies, we include banking market controls (Market controls_j,t-1) and macroeconomic controls (Macro controls_j,t-1), which are at the country level. The specifications include year fixed effects (ʋ_t) to control for unobserved heterogeneity. ε_ij,t is an iid error term specific to bank i from country j in quarter t. The explanatory variables are lagged by one period. Variables are winsorized at the 1st and 99th percentiles. The results are corrected for heteroskedasticity and correlation using bank-level clustered standard errors.⁶ The empirical specification is run for the full sample of 83 banks, and the period accounts for 40 quarters during 2005–2014.

The Probit model includes several identifying restrictions that are excluded from the 2^nd-stage OLS specifications, i.e., political stability, vote share of nongovernment parties, legislative and executive elections dummy, and prevalence. The political stability index measures the stability of political institutions, with higher values indicating greater political stability. The vote share of nongovernment parties reflects the vote share of parties other than the governing parties. The legislative and executive elections dummy is a variable that takes the value 1 if a legislative or executive election took place in the quarter before bailouts were provided. Prevalence is the index developed by Braun and Raddatz (2010) and measures the prevalence of connectedness through the ratio of actual to expected number of political connections (i.e., in essence the number of cases within a country where a former politician later sits on a bank`s board).⁷ The political stability index is from the World Governance Indicators database, the vote share of nongovernment parties and the legislative and executive elections dummy are from the Database of Political Institutions of Cruz et al. (2016), and the prevalence index is from Braun and Raddatz (2010). The choice of the exclusion restrictions follows the literature suggesting that political institutions and electoral cycles influence banks' probability of receiving financial assistance (Brown and Dinc 2005; Liu and Ngo 2014; Behn et al. 2017; Berger et al. 2020).

An important condition that the identifying covariates must satisfy is that they should directly affect the probability of intervention (i.e., one or more of the exclusion variables) but not be correlated with systemic risk. To verify the validity of the exclusion variables, we assess their explanatory power for the likelihood of bailouts and their orthogonality with systemic risk. Following Dam and Koetter (2012), we employ two econometric tests. First, using a weak identification test, i.e., a pseudo Kleibergen‒Paap F test, we check whether the set of identifying restrictions have sufficient explanatory power for the likelihood of intervention in the 1^st-stage equation. Second, we examine the exogeneity of the excluded covariates, i.e., whether they are uncorrelated with systemic risk, employing a pseudo Hansen J test.⁸

In the second stage of the Heckman (1979) procedure, we examine the impact of emergency policy interventions on systemic risk using an OLS fixed effects model for the sample restricted to the 30 rescued banks. To control for sample selection bias, the inverse Mills ratio generated by the Probit model in Eq. (1) is included. The following baseline model specification is run for the restricted sample of banks that received bailouts:

The dependent variable is represented by the contribution of bank i from country j to systemic risk in quarter t. The data represent the values of bank-level systemic risk indicators estimated on a weekly basis using the MES methodology proposed by Acharya et al. (2017).⁹ The MES assesses the marginal contribution of a bank to the total capital shortfall of the system. Using this methodology, we estimate the average return on bank i’s market capitalization in the weeks in which the total market capitalization of the system experiences the worst 1% of its outcomes (i.e., the expected loss of bank i’s market equity returns conditional on the system’s market equity returns exceeding the Value at Risk limits). We start by determining the conditional Expected Shortfall (ES) of the system’s returns as follows: \({ES}_{t}^{sys}=\) \(E\left[R_{Market\;Equity,t}^{sys}\vert R_{Market\;Equity,t}^{sys}\leq{VaR}_t^{sys}\right],\) where \(R_{Market\;Equity,t}^{sys}\) is the return of the system’s market capitalization and \({VaR}_{t}^{sys}\) is the Value at Risk indicator that expresses the maximum possible loss that the system can register for a given confidence level α (i.e., 1%) over a specific period of time.¹⁰ To estimate the values of MES, we use a multivariate GARCH-DCC specification, a refinement proposed by Brownlees and Engle (2017) that accounts for time-varying volatility and correlation. Online Appendix 4.1 provides a detailed description of the estimations. To merge the quarterly balance sheet and macroeconomic variables, we transform the weekly values of the systemic risk indicators to quarterly frequency by summing them up for each bank within each quarter. We express the systemic risk indicators (losses) as positive numbers; hence, higher values denote greater systemic importance.¹¹

The main regressors of interest are represented by the emergency rescue measures taken by the national authorities of country j and implemented by bank i during the event window (Policy interventions_{ij, event window}). These take the form of state guarantees, recapitalizations, and liquidity injections. They are represented by continuous variables obtained by dividing the volume of bailouts by the size of the banks. First, we assess the short-run effects of policy interventions by employing an event window starting one quarter before the intervention and ending one quarter after the intervention. Interventions take the same value within this event window (i.e., from t-1 to t + 1). However, for some banks’ rescue packages, a longer period might be required for their effects to be fully realized. Therefore, in the next step, we analyze the long-run impact of bailouts on systemic risk by employing an event window approach starting one quarter before the intervention is implemented by bank i and ending the quarter in which the bailout is paid back (i.e., through t-1 to t + n). Under this framework, the volume of bailouts as a share of total assets takes the same value during the event window.¹² A negative β₁ coefficient is associated with a decrease in systemic risk contribution for the rescued banks after they receive the assistance package from government.¹³

IMR_ij,t-1 represents the inverse Mills ratio generated by the Probit model in Eq. (1). The descriptions of the other regressors are similar to those for Eq. (1). The specifications include bank fixed effects (φ_i), year fixed effects (ʋ_t), and/or country × year fixed effects (μ_ij) to control for unobserved heterogeneity and shocks that affected our sample countries, such as regulatory changes.¹⁴ε_ij,t is an iid error term specific to bank i from country j in quarter t. The explanatory variables are lagged by one period. Several alternative models that include other proxies for bank-level risk profiles are estimated to test the robustness of the results. Variables are winsorized at the 1^st and 99^th percentiles. The results are corrected for heteroskedasticity and correlation using bank-level clustered standard errors,¹⁵ especially in the context that the dependent variables are preestimated.

A limitation of our analysis is that causality cannot be established with certainty due to the lack of a natural experimental setting within our framework. To address this issue, we provide alternative explanations to the maximum extent possible.

Policy interventions can exert an immediate effect on systemic risk, a delayed effect, or no effect at all, as depicted in Fig. 1. Moreover, the beginning of our event windows likely precedes the arrival in the market of information about the intervention. Regulators aim to fix systemic distress through financial assistance packages and usually provide them after risk is realized. Thus, declines in the equity value of rescued banks are likely during the time before intervention. An “immediate effect” would imply a prompt drop in systemic risk. However, interventions might not succeed immediately and might lead to a “delayed effect” characterized by an increase in systemic risk first and then a reduction, or measured systemic risk may increase before the intervention and be reversed slowly. As our dependent variable incorporates pre-intervention risk realizations, interventions can be positively linked with systemic risk for a period after their implementation. In addition, European governments faced credibility concerns during the GFC and the ESDC; therefore, declines in equity value may still have occurred after the intervention. Finally, “no effect” might indicate that systemic risk increases before the intervention were fully reversed by the end of the window. The financial health of some banks might not have recovered after the intervention, leading their stock returns to continue to underperform and increasing their contribution to systemic risk even more. Nevertheless, regulators imposed restrictions on the rescued banks and imposed fees during the duration of the bailouts, which could result in even higher systemic risk.¹⁶

Our sample consists of 83 publicly listed European banks whose assets totaled more than 20 trillion euros at the end of 2014. They are internationally active and represent 22 European states. The interest in this portfolio is motivated by regulatory considerations, as the group includes large banks identified as G-SIBs (Global Systemically Important Banks) by Financial Supervisory Board but also small local banks that present systemic importance. Among them, 53% are included in the EBA’s stress testing exercise, while 39% are included in the ECB’s Single Supervisory Mechanism framework (Online Appendix 1). We focus only on banks because they are the most important financial intermediaries in Europe. The size variation is considerable within the sample, as total assets at the end of 2014 ranged from approximately 1 billion euros to approximately 2 trillion euros. The average coverage of total banking system assets of the analyzed countries is approximately 49%. We use consolidated statements to capture all cross-border business transactions of international banks.

We start with a sample of 351 active and publicly listed financial institutions from the EU28 area that are included in the Thomson Reuters Financial Datastream within the sector “Banks”. We consider publicly listed banks because the methodology that we use for the estimation of systemic risk indicators is based on market data, which restricts the sample to banks listed on a stock exchange. Furthermore, due to methodology constraints imposed by the systemic risk estimation, we apply several exclusion criteria. First, we exclude banks that do not have weekly market capitalization data available in Datastream for the whole period. Second, we eliminate banks with negative equity.²⁰ Third, we include banks that have more than 75% of observations for their quarterly balance sheet data available in Worldscope (Table 1). Finally, to have a balanced sample in terms of size and across countries, we eliminate banks with total assets below 1 billion euros at the end of 2014. These filters lead us to our final sample of 83 banks, out of which 30 institutions received public interventions (Table 2).²¹

The data required for our systemic risk estimations span 2005 to 2014. We chose this period because it allows us to track the evolution of systemic risk during two financial crisis, i.e. the 2008 global financial crisis and the 2010 European sovereign debt crisis. Following Brunnermeier and Oehmke (2013), we estimate systemic risk during the whole period to account for the buildup phase in the precrisis and the propagation phase during the crisis. The impact of policy interventions on systemic risk is analyzed during the same time span. Although there have been important regulatory changes after 2014, most of the bailout packages were provided to European banks during 2008–2012, according to the State Aid Register data of European Commission.

The systemic risk measures are estimated separately for each bank using weekly returns extracted from Datastream (see Table 1 and Online Appendix 4 for computation details). Figure 2 presents the evolution of our sample’s market capitalization. This corresponds to the 83 banks analyzed over 521 weeks (2005–2014). Due to the deteriorating economic conditions in international financial markets, market equity shows a downturn in the first phase of the crisis (2008–2009), decreasing by more than 70% in comparison with its maximum value, reached in the middle of 2007. There were signs of recovery during 2009–2010, but the market equity started declining again at the end of 2011 when the European sovereign debt crisis took off.

The summary statistics of the systemic risk indicators are reported in Table 3 (Panel A). The data corresponding to the Marginal Expected Shortfall model reveal that for 2005–2014, the quarterly average contribution to systemic risk of all banks translates to an approximately 69% loss of the banks’ market capitalization. The statistics resulting from the Conditional Value at Risk measure, which we employ as a robustness check, show that banks’ marginal contribution to systemic risk represents an approximately 39% loss of the system’s market capitalization within a quarter.

The main features are compared between unrescued banks and banks affected by rescue packages (Table 3 Panel B). Overall, the difference in means analysis shows that the mean contribution to systemic risk for the whole sample period is larger for banks that received state guarantees, capital injections or liquidity injections than for nonaffected banks.²²

Figure 3 presents the weekly average contribution to systemic risk of all banks in our sample during 2005–2014. The graph reveals an increase in risk during the Lehman collapse in September 2008 and the European sovereign debt crisis. Our systemic risk measure is very likely to reflect both risk realization and forward-looking risk. Even though the governments of European member states intervened promptly with financial assistance programs, interventions were usually applied after the realization of risk. Therefore, declines in the market equity of bailed-out banks were likely during the period before intervention and may have persisted for a period after. Similarly to us, Black et al. (2016) find that the systemic risk of the European banking system increased during the crisis, reaching a peak during the sovereign debt crisis in Europe.

The impact on systemic risk of the emergency measures taken by European member states during the global financial crisis and the European sovereign debt crisis is analyzed for each of several types of policy interventions. To limit the negative spillovers in the banking system and to ensure financial stability, supervisory authorities used a broad range of mechanisms which we group into three categories (1) state guarantees, (2) recapitalizations (capital injections), and (3) liquidity injections. These are described in Online Appendix 2, which provides details on the type, size, and time of implementation of the interventions. Banks from our sample received public guarantees for bond issues, senior notes or other forms of debt; recapitalizations in the form of hybrid capital, participation capital, preferred shares, deeply subordinated perpetual notes or contingent convertible subordinated bonds (CoCos); and liquidity injections consisting of loan facilities, swap facilities, illiquid asset back-up facilities or asset protection schemes (APSs). We hand-collect the dataset from banks’ annual reports, financial statements, websites and the State Aid Register of the European Commission.

All three types of bailouts were applied to the banks in our sample during 2008–2014 (Online Appendix 3). In terms of value, liquidity injections lead with an average size of approximately 4% of banks’ total assets, followed by state guarantees (3% of total assets) and recapitalizations (2% of total assets). The aim of guarantee schemes is to ensure the supply of liquidity in the interbank market or to prevent bank runs. Recapitalizations are intended to strengthen the capital base of banks. Liquidity injections are given to limit the probability of runs and to encourage bank participation in asset markets, thereby limiting financial instability. They are expected to generate higher liquidity and greater transparency. The guarantees and recapitalizations were more country-wide in nature, as the schemes were aimed at restoring confidence, while the liquidity measures targeted the exposure to losses by individual banks (Panetta et al. 2009).²³

Table 4 provides descriptive statistics of the financial support provided by the government for the entire sample in Panel A and for the sample restricted to banks with intervention events in Panel B. In sum, our sample was exposed to 106 policy intervention events: 36 events corresponding to state guarantees, 35 events related to recapitalizations, and 35 events linked with liquidity injections (among which 8 are related to APSs and 27 to other types of emergency liquidity schemes). Out of 83 banks, 30 implemented these types of policies: 6 banks received all three types of interventions, 14 banks applied two types of interventions, and 10 banks relied on a single intervention measure. The rescued banks represent 15 countries (out of the 22 included in our sample), among which in seven countries more than two banks were affected. From the 30 rescued banks in our sample, 18 were released from the bailouts before 2014.

In most cases, bailouts come at a cost for banks. For example, banks often paid a fee, usually linked to the risk of the financial institution (i.e., its rating), executive compensation is limited, and dividends could be distributed only to the government (Petrovic and Tutsch 2009). We account for these features by considering a set of restrictions imposed by regulators during the duration of bailouts: supervisory board intrusions, management pay limitations, and capital payout bans. They are expressed as dummy variables, and their definitions are provided in Table 1. Among the policy intervention events from our sample, 28% are associated with supervisory board intrusions, 67% have limitations on executive compensation, and 41% are linked with capital payout restrictions.²⁴

To account for different risk strategies, we control for the risk profiles of banks one quarter before intervention. Prior studies suggest that bank characteristics such as size, leverage, profitability, and credit and liquidity risk were key drivers of systemic risk during the most recent financial crisis (Demirgüç-Kunt and Huizinga 2010; Tarashev et al. 2010; Acharya et al. 2012; Mayordomo et al. 2014). In line with the literature, the following risk profile indicators are used: (1) size (logarithm of total assets); (2) asset growth (asset growth relative to average bank asset growth in the bank’s country); (3) leverage (common equity to total assets ratio and Tier 1 regulatory capital ratio in robustness exercises); (4) the quality of the loan portfolio (provisions for loan losses to gross loans); (5) liquidity ratio (liquid assets to deposits and short-term funding); (6) rollover risk ratio (deposits and short-term funding to total deposits and borrowings); and (7) profitability represented by the ROAA ratio (net profit to average assets). Our presumption is that larger size, asset growth, credit risk, and rollover risk are associated with a higher level of systemic risk, while higher capitalization, liquidity, and profitability are associated with a lower level. Additionally, we capture the orientation of banks’ business toward traditional and nontraditional activities by including the share of lending activity (gross loans to total assets) and the net noninterest margin (net noninterest income to gross revenues). Previous studies show that systemic risk is associated with a high share of nontraditional activity (Brunnermeier et al. 2020; Demirgüç-Kunt and Huizinga 2010). Additionally, we capture the comovement of bank value with financial system value by including the beta coefficient, computed as the covariance of bank i’s stock returns with the market’s stock returns divided by the variance of the market’s stock returns. The variables are extracted from Worldscope, and their definitions are given in Table 1.²⁵ The descriptive statistics presented in Table 4 show that on average during 2005–2014, banks from our sample had a Tier 1 ratio of 11%, liquidity ratio of 32%, credit risk ratio of 1% and gross loans shares of 61%. These statistics suggest that on average, the institutions are well capitalized and have a good liquidity situation and a high-quality loan portfolio. Additionally, they are more oriented toward traditional banking business.

Following previous studies (Girardi and Ergün, 2013; Anginer et al. 2014a; Weiß et al. 2014a, b), we control for the banking market and macroeconomic environment. Accounting for the particularities of each national banking sector, we consider the intensity of competition − or lack thereof − expressed by the Boone indicator,²⁶ which has been found to be associated with a reduced contribution to systemic risk (Anginer et al. 2014a). Next, we account for the strictness of prudential regulations regarding initial and overall capital held by banks. As a proxy, we use the capital regulatory index provided by the Bank Regulation and Supervision Database of the World Bank and calculated as in Barth et al. (2013).²⁷ Additionally, we employ the sovereign debt holdings of the banking system as a share in GDP and the inflation rate and GDP growth as macro controls. We expect systemic risk to increase with the deterioration of macro conditions. Finally, we include in our specifications a dummy variable that reflects the global financial crisis (Dummy GFC) and a dummy variable that accounts for the European sovereign debt crisis (Dummy SVG) to control for the dynamics of these crises that affected the evolution of systemic risk. The variables are extracted from the World Development Indicators, Global Financial Development and European Central Bank databases. Their definitions are given in Table 1, and the descriptive statistics are given in Table 4.

The output for the 1^st-stage Heckman (1979) selection model is presented in Table 5 Panel A. The results reveal that greater political stability significantly increases the probability of the government providing bailouts to financial institutions. The vote share of nongovernment parties enters the model significantly, with a positive sign, suggesting that in countries with higher political competition where the vote share of parties other than the government parties is higher, the regulator is more likely to implement policy interventions. Prevalence is significant as well. The F statistic associated with the pseudo Kleibergen‒Paap F test indicates that the set of identifying covariates has sufficient predictive power to explain the likelihood of the government providing financial assistance.²⁸ Additionally, the p value corresponding to the pseudo Hansen J test validates the orthogonality between the exclusion restrictions and systemic risk. Therefore, our set of instruments contains additional information on the likelihood of bailouts and meets the validity conditions.²⁹

Thus far, we have estimated the impact of policy measures on the “average” bank. This section presents the impact of restrictions imposed by regulators on the relation between emergency rescue actions and systemic risk. The model specification is introduced in subsection 3.3.2. We consider the following constraints: supervisory board intrusions, management pay limitations, and capital payout bans. Table 10 Panel A shows the empirical estimates for the short-run models, while Panel B shows the long-run specifications.

The results suggest that all three types of regulatory restrictions mitigate the harmful effect of liquidity injections on systemic risk in the short run, as captured by the negative and strongly significant coefficients associated with the interaction between restrictions and liquidity injections. These findings support the hypothesis that investors believe that liquidity injections enhance value when regulators impose tighter restrictions, leading to an increase in stock prices and therefore a reduction in systemic risk. It is likely that regulatory restrictions are perceived as effective tools in reducing portfolio risk by assuring stricter monitoring of the banks injected with liquidity. For the other type of financial assistance programs, the results show no significant effect of the interaction coefficient between bailouts and restrictions. A possible explanation is that in the short run, investors give greater importance to restrictions when they are associated with rescue packages that quickly fix banks’ distress than they give to other interventions that may need a longer time horizon to produce their effect.

In the long run, we obtain a similar effect for supervisory board intrusions in the case of state guarantees and management pay limitations associated with recapitalizations. Seats on the supervisory board ensure stricter supervision of investment and lending practices after governments guarantee the debt of the financial institutions. This may counterbalance the lack of credibility regarding the governments’ ability to guarantee banks’ debt, softening the positive impact of state guarantees on systemic risk. In the case of recapitalizations, ceilings on executive salaries and bonuses may temper managers’ appetite for risky projects, leading to an increase in market valuation. Therefore, executive compensation limits could further enhance the long-run beneficial impact of recapitalizations on systemic risk. We find no significant effect for liquidity bailouts, which indicates that the realization of distress continued in the long run at banks injected with liquidity, regardless of the type of regulatory restrictions applied.

Overall, applying restrictions to rescued banks while interventions are in effect can be an efficient policy tool to reduce the positive association of interventions with systemic risk. One limitation of our framework is that the interventions that we assess are specific to the global financial crisis and the sovereign debt crisis that affected European banks. Since then, a number of regulatory and legal changes have been implemented. For example, there have been several decisions by the European Commission that imposed stricter restrictions on the rescued banks, such as dividend bans, during the duration of the bailout.³⁷ Additionally, the overall legal framework changed after the adoption of the EU Bank Recovery and Resolution Directive (BRDD) in 2014.³⁸ However, governments can still choose from the bailout methods that we examined to save financial institutions in distress. In future research, it will be interesting to assess the effects of policy interventions on systemic risk considering the stricter restrictions imposed on rescued banks after 2014, especially in the context of the COVID-19 pandemic.

Finally, we examine the impact of banks’ risk profiles on the relation between emergency rescue actions and systemic risk. The model specification is introduced in subsection 3.3.2. We discuss the empirical results for the following risk profile indices: size, leverage, and profitability. Table 11 Panel A shows the empirical estimates for the short-run estimates, while Panel B shows those for the long-run specifications.

Regarding size, the findings presented in Panel A Column (2) show that in the short run, the influence of recapitalizations is significantly different (although only marginally so) for larger banks in comparison to smaller banks, as suggested by the coefficient on the interaction term Recapitalizations × Dummy Size (i.e., 16.117*). For the average bank, the positive link of recapitalizations with systemic risk becomes significant, and it is increasing in size for larger banks. A possible explanation is that the market valuation of large banks injected with capital is penalized more because of the dilution of preintervention shareholders’ earnings, which translates into an increase in systemic risk. In the long run, the positive association of liquidity injections with systemic risk is even more pronounced for large banks (Panel B Column (5)). The coefficient on the interaction term Liquidity injections × Dummy Size is positive and significant (i.e., 1.225**). This result is consistent with the literature on moral hazard embedded in government support programs for TBTF banks. Large banks are usually more focused on investment activities that are riskier than traditional lending activities; thus, rescue packages may incentivize them to increase portfolio risk, which enhances their systemic importance. In terms of policy implications, these findings support the restrictions suggested by the European Commission, which requested the downsizing of several large European banks that received bailouts during the crisis.

The link between interventions and systemic risk is also related to leverage in both the short and long run. The estimates show that the immediate influence of recapitalizations is negative for better capitalized banks (Column (3)). The interaction of Dummy leverage with recapitalizations enters the specifications with a negative sign (i.e., -19.833***).³⁹ The finding suggests that the contribution to systemic risk of banks injected with capital is lower immediately after the intervention for better capitalized banks than for less capitalized banks. The long-run results highlight that the realization of distress continues in the long run for banks injected with liquidity but is diminished when banks have a higher level of capitalization (Column (7)). These findings suggest that the provision of liquidity assistance should be oriented toward safer banks. This is in line with the actions of European governments, which provided liquidity assistance to safer financial institutions with regulatory capital above the minimum required threshold. They did not wait to intervene until bank capitalization deteriorated significantly, as it would have been costlier.

Finally, performance can be significantly associated with the relationship between recapitalizations and systemic importance in the long run, as shown by the interaction term Recapitalizations × Dummy Profitability in Column (8), which is positive and highly significant (i.e., 8.801***). The result indicates that recapitalizations fix the problems created by risk realizations, and the result is amplified for less profitable banks. In turn, for banks with higher performance, recapitalizations are linked with an increase in systemic risk. A possible explanation for this finding could be that more profitable banks are likely to borrow more and engage in risky operations, increase their portfolio risk, and therefore intensify their systemic importance.

Overall, the results suggest that the relation between policy interventions and systemic contribution varies significantly with the risk profiles of banks. Characteristics such as size, leverage, and profitability can significantly shape the relationship between bailouts and banks’ systemic importance in the long run, while the immediate link between governmental assistance programs and systemic risk is heterogeneous among banks with different sizes and levels of leverage. Most importantly, from supervisors’ perspective, the efficiency of emergency rescue measures can be mitigated or enhanced by banks’ risk strategies.