Countercyclical capital buffers and credit supply: Evidence from the COVID-19 crisis

This paper examines how European banks adjusted their lending subsequent to the release of the countercyclical capital buffers (CCyB) during the COVID-19 pandemic. At its onset in 2020Q1, being exposed to a higher ex-ante countercyclical capital buffer led to a reduction in banks’ lending. Yet the relief of the CCyBs removed this negative effect from 2020Q2 onwards. We find that the reduction in CCyBs led to a significant relative increase in the average bank’s lending by about 5.6 percentage points of their total assets. This increase happened mainly in retail mortgage loans and was stronger for poorly-capitalized banks. These results imply that the release of the CCyBs was effective in promoting bank lending during the pandemic.


Introduction
Countercyclical capital buffers aim to soften the cyclical effects of standard capital requirements by slowing down the credit supply growth during economic booms and by providing additional capital to continue loan provision during crises. This novel macroprudential tool was implemented in Europe in January 2016, but only became fully effective in January 2019. Until the beginning of the COVID-19 pandemic, however, the countercyclical capital buffer was constantly increased and it has never been released. Only at the onset of the COVID-19 pandemic, and for the first time ever, the countercyclical capital buffers (CCyBs) have been released in almost all participating jurisdictions. While it is generally well understood that banks reduce lending after increases in CCyB requirements across many samples (e.g., Jiménez et al., 2017;Basten, 2020;Auer et al., 2022 ), the question remains open whether banks R We are grateful for valuable comments from Ozan Güler and Loriana Pelizzon. We also would like to thank the participants at the 28th Annual Meeting of the German Finance Association for helpful comments. Dursun-de Neef gratefully acknowledges funding by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) -448659867 . The paper was written during Dursun-de Neef's time at Goethe University Frankfurt. * Corresponding author.
In general, our results confirm the expectation that a release of previously built-up CCyBs can help banks provide more loans during a crisis. However, besides this overall affirmation, this analysis contributes to the understanding of how CCyBs work by highlighting that the timing matters crucially: On the one hand, banks subject to high CCyBs sharply reduced their lending at the onset of the pandemic. On the other hand, subsequent to their releases, from 2020Q2 onwards, the negative effect of higher ex-ante CCyBs disappeared, and a significant positive effect prevails. Thus, policymakers should not release buffers "too late" as this could potentially outweigh any support for credit supply given the strong negative effect that persists when entering a crisis with high CCyBs.
In this paper, we use the fact that CCyB releases became immediately effective after the national authorities' release decisions, and they can thus be seen as a largely unanticipated and exogenous shock to the financial sector. Hence, the COVID-19 pandemic provides a unique setting to study their effectiveness. As shown in Fig. 1 , the average CCyB rate increased as countries started to make use of this macroprudential tool from the end of 2016. This continued until 2020Q1 when the pandemic started. The jurisdictions reacted quickly and started to release the CCyBs towards the end of March and the beginning of April 2020. As a result, the av- erage CCyB rate dropped from 0.91% to 0.23% from the first quarter to the second quarter in 2020.
We exploit this sharp reduction in CCyBs to study whether banks that were exposed to these releases provided more loans during the pandemic. To do so, we measure each bank's exposure to the CCyBs implemented in each jurisdiction where it has credit operations using the distribution of branches as a proxy for the credit exposure of banks in each country they operate in. We calculate the bank-level countercyclical capital buffers as the weighted average CCyB rates of all jurisdictions in which the bank has a branch where the weights are the fraction of branches in each country. We subsequently use the reduction in this bank-level CCyB rate from 2019Q4 to 2020Q2 to measure the exposure of each bank to the CCyB reliefs.
To quantify the effect of the CCyB releases, we use a difference in differences (DiD) estimation methodology where the treatment intensity is defined as the reduction in banks' CCyBs. Given that the CCyB releases were unanticipated, just as the pandemic itself, we expect that banks with different exposures to the capital reliefs had similar growth in their loans before the pandemic and would have had parallel loan growth during 2020 in the absence of the pandemic. To illustrate this, we first divide our sample into two groups as banks that experienced a reduction in the CCyBs, which we call treatment group, and the remaining banks as control group. Figure 2 shows the loan growth of these two types of banks from the beginning of 2018 until the end of 2020: Both types had similar loan growth during the pre-pandemic period, i.e., the parallel trends assumption holds. As shown in the figure, at the onset of the pandemic, in 2020Q1, treatment banks experienced a significant drop in their loans compared to control banks. The reason behind this is that the high levels of capital buffers, which treatment banks had to comply with at the onset of the pandemic, made these banks more hesitant to provide loans at this stage. Starting from 2020Q2, however, treatment banks increased their loans while control group's lending stayed constant over time.
Following this, in our analysis, we divide the overall pandemic into two sub-periods: 2020Q1 and the following three quarters of 2020. Once we control for the effect of the ex-ante capital buffers, we find that banks with a higher reduction in their CCyBs increased their lending significantly in both pandemic periods. At the onset of the pandemic, in 2020Q1, banks with a 1 percentage point higher ex-ante capital buffers had a significant reduction in their lending by about 7.2 percentage points of their total assets. In the same quarter, a 1 percentage point reduction in the CCyBs increased banks' lending by about 6.3 percentage points of their total assets. This implies that the negative effect of holding high capital buffers on bank lending was mostly removed by the CCyB releases at the onset of the pandemic. Over the following quarters, from 2020Q2 to 2020Q4, having a higher ex-ante capital buffer did not have a significant effect on banks' lending anymore. However, the positive effect of the CCyB releases continued to be significant: A 1 percentage point higher reduction in the CCyBs resulted in a significant increase in banks' lending by about 1.9 percentage points of their total assets. This suggests that the release of the CCyB buffers was effective from the second quarter of 2020.
When we study the average effect over the pandemic period, we find that a 1 percentage point reduction in the CCyBs led to a significant increase in banks' lending by about 5.6 percentage points of their total assets. In a more stringent specification, we include country × year-quarter fixed effects to control for loan demand. The impact remains significant at the 10% significance level with a similar magnitude: A 1 percentage point of CCyB reduction results in an increase by around 4.8 percentage points of total assets.
One may expect that the reduction of capital buffers benefits poorly-capitalized banks more as they are closer to their minimum required capital ratios. Once the buffers are released, these banks would be less capital constrained, and, as a result, they might provide more loans to their customers. When we analyze the differential impact of the capital buffer reliefs on banks with different capital ratios, we find that the significant positive effect is much larger and more significant for poorly-capitalized banks: A 1 percentage point reduction in the CCyBs resulted in a sig- nificant increase in poorly-capitalized banks' lending by 8.4 percentage points of total assets whereas the increase was only almost 4 percentage points for well-capitalized banks. In addition, we show that poorly-capitalized banks experienced a larger increase in their risk-weighted assets as a response to the buffer releases.
Last, we study the impact on different types of loans. Our results highlight that banks that experienced a reduction in their capital buffers increased their retail loans. According to the reported coefficient estimate, a 1 percentage point reduction in the CCyBs led to a significant increase in banks' retail loans by 5.1 percentage points of their total assets. This increase mainly comes from an increase in their retail mortgage loans. On the other hand, corporate loans increased only for the banks that experienced large amounts of loan commitment drawdowns during the pandemic. According to our results, the increase was around 2.3 percentage points of their total assets. This implies that CCyB releases helped these banks to honor their loan commitments during the pandemic.
Overall, our results show that the release of the previously built-up CCyBs helped banks to provide more loans during the pandemic. At the onset, being exposed to a higher ex-ante CCyB led to a reduction in banks' lending, yet the relief of the CCyBs removed this negative effect. From the second quarter of 2020 onwards, the negative effect of higher ex-ante CCyBs disappeared, and we find a significant positive effect of the countercyclical capital releases. These findings imply that CCyB releases were effective in promoting bank lending during the pandemic.
Literature review: While the literature on the general effects of capital ratios on bank lending is extensive (see, e.g., Carlson et al., 2013;Célérier et al., 2017;Kim and Sohn, 2017;Naceur et al., 2018;Gropp et al., 2019;De Jonghe et al., 2020;Fraisse et al., 2020;De Marco et al., 2021 ), research on the release of countercyclical capital requirements is scarce. This is mainly due to the fact that the countercyclical capital buffers have not been released until 2020. To over-come this issue and gauge the potential implications of such a release, researchers use conceptually similar regulatory requirements that were implemented in specific countries. In 20 0 0, Spain introduced the dynamic provisioning scheme which demands banks to build up a dynamic provision fund in good times to cover credit losses in bad times. This particular policy tool acts against the observation that banks tend to provision more in crisis times which increases the procyclicality of lending ( Laeven and Majnoni, 2003 ) and banks' risk taking ( Illueca et al., 2022 ). Unlike the Basel III countercyclical capital buffer that uses CET1 capital, the dynamic provision fund uses Tier 2 capital. Jiménez et al. (2017) exploit this policy to analyse the effects on credit supply and find that procyclical regulation reduces credit supply in good times. However, after the introduction of the new policy, banks that faced higher requirements concentrated more on clients with higher leverage and on those that pay higher interest on their loans. Those clients also exhibit higher default rates thereafter. According to their results, this behaviour hints at increased risk-taking and search for yield as a response to higher capital requirements, which constitutes an undesired outcome from the perspective of the policymaker. A second countercyclical policy tool that has been in place before the CCyB is the temporary deduction item "prudential filter" used in the capital calculation for Slovenian banks. It was introduced in 2006 and resulted in an average capital buffer of 0.8% of risk-weighted assets (RWA). Chen et al. (2019) study its release during the Great Financial Crisis and find that banks that held 1 percentage point higher capital buffers expanded their lending by 11 percentage points more compared to banks that were not subject to the regulation.
In 2013, Switzerland became the first country to initiate a sectoral CCyB that is targeted at mortgage loans financing of residential real estate. While several recent papers have analyzed its introduction, i.e., the build-up phase, and the effects on e.g., the composition of credit supply (e.g., Basten, 2020;Behncke, 2020;Auer et al., 2022 ), these studies were also not able to examine its deactivation, since this only happened at the onset of the pandemic.
Hence, the question of whether the release of CCyBs is able to "do its job" and helps mitigate the negative consequences of a crisis remains unanswered.
We are aware of only two other contemporaneous studies that examine the release of the general European countercyclical capital buffers. First, there is ongoing work by Avezum et al. (2021) , who apply a counterfactual analysis to investigate the effect of a countercyclical capital buffer as well as systemic risk buffer release on aggregate credit supply to households in the Euro area. In contrast to this paper's methodology, the authors use a synthetic control method at the country level and show that aggregate lending to households grew by 0.90 percentage points more between March and August 2020 in countries that released the buffer. Second, published as part of the Financial Stability Review (November 2021) of the ECB, the report by Couaillier et al. (2021) show the effects of several different buffer releases on credit supply by comparing banks with a smaller capital headroom on top of their combined buffer requirements with those that had a larger one. While such setting does not allow for an estimation of the specific effect that the CCyB might have on credit supply, the report demonstrates a positive effect on credit supply for both groups of banks. To the best of our knowledge, there are no other studies that deal with the release of the Basel III countercyclical capital buffer during the COVID-19 pandemic in Europe. Since the first study investigates the effect on aggregate lending without taking into account the bank-level effects and the second does not specify the stand-alone effect of a CCyB release on lending, our paper contributes to the literature as the first study on the impact of the CCyB releases on banks' lending during crises by examining the effect of the Basel III countercyclical capital buffer releases on European banks' lending during the COVID-19 crisis.
This paper is structured as follows. Section 2 provides a brief review on the countercyclical capital buffers and what happened during the COVID-19 pandemic. Section 3 discusses the empirical strategy and data, and Section 4 presents our main results and robustness checks. Last, Section 5 concludes.

Countercyclical capital buffers
In response to the lessons learned during the Great Financial Crisis, capital requirements were increased significantly in the new Basel III accord to make banks more resilient to shocks. In times of crisis, these requirements typically become more binding and might lead to a credit crunch since recapitalization or fire sales are no viable options. This can lead to the cyclicality of credit supply by capital-constrained banks (see e.g., Repullo and Suarez, 2013) . Moreover, an extensive credit supply growth in boom phases can be associated with the emergence of systemic risks to the banking sector as more defaults occur during recessions, which weigh on banks' balance sheets if they have relaxed their lending criteria during the boom.
To soften the cyclical effect of standard capital requirements, a new type of capital requirement (countercyclical capital buffer) was introduced with the Basel III accord published in 2010, which takes into account the procyclical lending behaviour of the banking sector. The buffer can be increased or decreased conditional on the risk of excessive credit supply and economic conditions. Its flexibil-ity allows the CCyB to alleviate the risk of a credit crunch in economic downturns via a relaxation of the binding capital requirements. Additionally, it has the benefit of reducing the likelihood of a credit driven crisis by imposing higher capital requirements that limit excessive credit ex-ante. For this paper, only the first effect is of interest. In a situation of a sudden shock to the economy, risks typically materialize and deplete a bank's capital. Coming closer to their capital requirements set by the supervisors, banks have the option to raise equity or to deleverage either by selling their assets or by tightening their credit supply. This response in turn can amplify the initial shock in a phase when companies need additional liquidity to overcome the reduction of their cash flows as it happened during the COVID-19 pandemic. The release of the previously built-up countercyclical capital buffers in such circumstances decreases banks' capital requirements and enables them to absorb higher losses while maintaining their capital ratios. This should alleviate the reduction in credit supply to the real economy.
Each national authority of the respective jurisdiction sets its buffer independently, taking into account its credit growth and the build-up of systemic risks. The credit-to-GDP gap serves as a common reference point in individual buffer decisions ( BCBS, 2011 ). Figure 3 shows the countries that have had positive CCyBs. 1 While banks are granted a lead time of up to twelve months to prepare for a countercyclical buffer increase, reductions are taking effect immediately after the announcement of the competent authority. The countercyclical capital requirement in each jurisdiction can vary between 0 and 2.5% of RWA. A bank's final capital requirement is the credit exposure weighted average of the buffers applied across the jurisdictions in which it has a private-sector exposure. The bank-specific buffer rate is applied on a consolidated level and consolidated total RWA is used for the calculation of all risk-based capital ratios. It has to be met with Common Equity Tier 1 (CET1) capital. The CCyBs were phased in from the beginning of 2016 to the end of 2018 and became fully effective on 1 st January 2019 ( BCBS, 2011 ).

Background on regulatory relief during the onset of the pandemic
During the onset of the COVID-19 pandemic in Europe, several measures were taken to counteract the impacts of the crisis. One was the announcement by the European Central Bank that it will allow significant institutions to temporarily use their capital buffers and operate below their Pillar 2 Guidance (P2G) and their capital conservation buffer (CCB) without any supervisory malus. The objective of this extra capital relief by the supervisor was the same as releasing the CCyB. However, in contrast to the CCyB release, banks as well as market participants know that any usage of capital below the P2G and CCB inevitably needs to be rebuilt at a later point in time. Additionally, using those buffers might be considered as a bad signal since in normal times dipping into those buffers triggers regulatory restrictions on dividend payouts, share buybacks and bonuses. Hence, banks refrained from using the P2G or CCB to support their credit supply as documented by Abboud et al. (2021) . On the contrary, after the release of the CCyBs, the previous capital levels do not have to be rebuilt later until a new decision is made by authorities. Furthermore, the CCyB releases affect the whole sector, leading to no supervisory restrictions and no bad signal when one single bank uses the freed cap- The figures provide an overview of the CCyB rates in several European countries between the end of 2015 and the end of 2020. We plot only those countries that are part of the sample and that had a non-zero CCyB rate at some point (other countries such as Netherlands, Finland, Poland, Austria, Italy, Romania, Spain, Portugal, Hungary, Germany, Belgium, and Slovenia had zero effective CCyB rates throughout the sample period).
ital. This implies that there should be no major distorting side effects from other capital releases that interfere with the effect of the CCyB releases.
Another action that was taken by several national competent authorities with regards to capital requirements was to lower the systemic risk buffer (SyRB) and the capital buffer for other systemically important institutions (O-SII). The O-SII buffer applies only to some (important) banks in each country and aims to mitigate contagion effects stemming from systemic risks in national banking systems. The SyRB intends to mitigate systemic risks of a longterm, non-cyclical, nature that are not covered by all other capital requirements. The buffer may apply either to all credit exposures, to domestic or foreign exposures, or to specific sector exposures. Importantly, both the SyRB and O-SII buffer requirements are connected. The SyRB is cumulative with the O-SII buffer when the systemic risk buffer applies to domestic exposures only while the higher of both buffers is used when the SyRB applied to all exposures. Given this interaction, in any analysis of capital requirements, both buffers must be considered in a combined manner. Since the SyRB and O-SII buffers were also released in some countries during the onset of the pandemic, we control for this addi-tional effect on capital requirements to reliably estimate the effect of a CCyB release.
A last important aspect when assessing the effect of changes in capital ratios on lending is that those changes are endogenous to a large extent. The banks' own decisions, for example, business expansions or recapitalizations, influence the capital ratio and hence may have a different effect on lending. The release of the CCyB, however, has been a truly exogenous shock to banks. It has been the decision of the respective regulatory authorities for which banks could not prepare as the release becomes effective immediately. The only exemption to this could be if certain banks are exempt from holding a CCyB due to their size (i.e., if they are small or medium-sized), which has been implemented in eight countries. Over the whole observation period five out of 301 banks in the sample with a positive CCyB rate are affected by the exemptions and for these banks, the CCyB rates are manually set to zero in our analysis. 2

Bank-specific countercyclical capital rate
Bank-specific countercyclical capital requirements are calculated by using the credit exposure weighted sum of each jurisdiction's CCyB rate. Unfortunately, there exists no information on the credit exposure of each bank across countries which is why the branch network is used as a proxy. The idea is that the distribution of branches should serve as a reliable proxy for the credit exposure each bank has in the countries they operate in. We thus implicitly assume that banks provide more loans in countries where they have a higher number of branches. 3 Intuitively, banks are making more business, i.e. they supply more credit to borrowers such as households or firms, in those countries in which they have a higher presence in terms of branches.
This relationship is used to calculate the bank-specific CCyB rate for bank i via the branch exposure weighted sum of the countercyclical buffer rates in each jurisdiction as follows: Branch i, j is the proportion of bank i 's branches in country j . C C yB j,t is the set countercyclical capital buffer rate in country j at time t . The number of bank branches (as of May 2021) across the world is downloaded from SNL Financial.

Empirical methodology
The objective of this study is to examine the effect of the CCyB releases on the credit supply of banks during the pandemic. To do this, we employ a DiD estimation methodology where the treatment intensity is defined as the reduction in banks' CCyBs. The important assumption of the DiD analysis is the exogenous allocation of treatment intensity across banks. The CCyB releases were random just like the pandemic itself. As a result, one expects to see a similar trend in the pre-pandemic loan growth for banks with different exposures to the capital reliefs and assume that they would have been similar during 2020 as well in the absence of the pandemic. To show evidence for this, we start by dividing our sample into two groups as treatment banks that experienced a reduction in their CCyBs and control banks that did not. Figure 2 shows the loan growth of these two types of banks from the beginning of 2018 until the end of 2020: Both, the treatment and control group, show an increasing trend in loan growth until 2019Q4, i.e., the parallel trends assumption holds, which is a requirement for the DiD analysis. This was followed by a reduction in loans in 2020Q1 for treatment banks while the control group's loan level remained constant. The reason behind this reduction is that treatment banks had high levels of CCyBs which made these banks more hesitant to provide loans at the onset of the pandemic. Subsequently, in 2020Q2, treated banks began to increase their credit supply, while the control group's loan level remained again very constant.
Following this preliminary, descriptive, analysis, we divide the pandemic period into two to differentiate between what happened in 2020Q1 as well as from 2020Q2 to 2020Q4. We estimate the following regression: The dependent variable is defined as the first difference of total loans divided by total assets at the beginning of the period, following Li et al. (2020) and Dursun-de Neef and Schandlbauer (2021) .
C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. In this period, almost all of the jurisdictions released their countercyclical capital buffers. Post 2020 Q1 is a dummy variable that is equal to one in 2020Q1 and zero otherwise, and Post 2020 Q 2 −2020 Q 4 takes the value one from 2020Q2 onwards, and zero otherwise. C C yB 2019 Q4 i is the level of CCyB as of the last quarter of 2019. C C yB i · Post 2020 Q1 and C C yB i · Post 2020 Q 2 −2020 Q 4 are the main independent variables. 4 β 1 captures the relation between the reduction in the CCyB rates and credit supply at the onset of the crisis, whereas β 2 focuses on how the reduction of CCyBs affected banks' lending as the crisis evolved.
To control for the impact of the CCyB rates on banks' lending for both post periods, we include the level of the CCyB rate for each bank as well as its interaction with the post periods. This enables us to control for the possible negative effect of having a high capital buffer on banks' lending especially at the onset of the crisis. As illustrated in Fig. 2 , we expect that banks with higher CCyB rates would decrease their lending more when pandemic started as higher capital requirements would discourage these banks to take additional risk.
X i,t−1 is a set of bank characteristics lagged by one period. This consists proxies for the CAMELS supervisory rating. We follow Dursun-de Neef and Schandlbauer (2021) and use the same proxies for the rating components: total equity for capital adequacy, loan loss reserves for asset quality, net interest income for management quality, return on assets for earnings, cash and cash equivalents for liquidity, and deposits for the sensitivity to market risk. Additionally, the amount of unused loan commitments is added as a control variable. Country i,t−1 is a set of economic and demographic variables that should serve as a proxy for loan demand in the markets the banks are operating in. Controls include the overall population, the GDP per capita, and the median household income.
In addition, the COVID-19 incidence rates, the economic policy responses from the Government Response Tracker, the systemic risk buffer and the capital buffer for other systemically important institutions, and a set of monetary policy controls, Y i,t , are included as well. All these controls are created by applying a weighted average of the geographic exposure of each bank's branches in the same way as our CCyB variable as described in Section 2.3 . Monetary policy controls consist of the amount of asset purchases, the level of the systemic risk buffer or the capital buffer for other systemically important institutions, liquidity facilities, and the overnight lending rate conducted by the respective central banks. Liquidity facilities include all non-standard monetary policy liquidity providing measures with a maturity of more than three months such as the Long-term Refinancing Operations by the ECB. For asset purchases and liquidity facilities, we form an indicator variable that is equal to one if they are bigger than zero and zero otherwise. Finally, bank fixed-effects, δ i and quarteryear fixed-effects, δ t , are included to control for unobserved heterogeneity. All standard errors are clustered at the bank level to control for heteroskedasticity. We next analyze the average effect during the pandemic by using the post dummy Post 2020 Q 1 −2020 Q 4 that is equal to one in 2020 and zero otherwise: (3)

Data sources and sample selection
We analyze European banks during the period of 2018Q1 to 2020Q4. Due to the similarity of the economic region, all banks that are domiciled in either the European Single Market, Great Britain, and Norway are included. The data on bank and country characteristics are retrieved from SNL Financial which is part of S&P Global. As the CCyB is applied on a consolidated basis, subsidiaries are excluded from the dataset. The countercyclical capital buffer rates, set by each country, are taken from the Bank for International Settlements' website. Additionally, the number of COVID-19 cases for each country, expressed by the quarterly incidence rate per 10 0 0 people, is retrieved from "Our World in Data". The information on monetary policy measures is obtained from the respective central banks. The amount of a central bank's asset purchases covers the purchases under all ongoing programmes in the specific quarter. Liquidity facilities include all extraordinary liquidity providing measures to the banking sector of the longer term. 5 To measure economic response of governments, we use the "economic support index" of the Oxford COVID-19 Government Response Tracker that collects information of individual government responses across 19 indicators, and that can be downloaded from github.
All banks with missing data on the number of branches or with a deposits-to-assets ratio of less than 2% were excluded. After all data cleaning, the final unbalanced sample consists of 3624 observations from 302 banks. As of Q4 2019, the dataset covers roughly 72% of the combined total sector assets held by banks in the European Single Market, Great Britain, and Norway.

Descriptive statistics
The dataset includes 3624 bank-quarter observations from 302 banks,though not all banks report in at all times. 6 Table A1 in the Internet Appendix presents the composition of banks across countries in terms of the number of banks and the corresponding share of total banking assets in each country. Although Germany has the largest number of banks with 49, it has the third largest corresponding share of total assets with 17%. France has the largest share of total assets of 23% with 10 banks, which is followed by the United Kingdom with 20%. Over the whole observation period, the average bank has total assets of EUR 150 billion while the largest has EUR 2.7 trillion (BNP Paribas SA (France) in 2020Q1) and the smallest only EUR 138 million (Hvidbjerg Bank A/S (Denmark) in 2017Q1). In pre-pandemic quarters depicted in panel A of Table 1 , loans amount to 65.3% of total assets and stand against financing 5 The liquidity facility should at least have a maturity of three months. Standard measures such as the ECB's main refinancing operations are not regarded as extra-ordinary measures. In some cases, the amounts of asset purchases and liquidity facilities were available only in monthly frequency. In these instances, the data was aggregated on a quarterly level, and if necessary, converted to Euro currency, using the historical exchange rates published by the European Central bank. The time series of Euro foreign exchange reference rates can be obtained via https://www.ecb.europa.eu/stats/policy _ and _ exchange _ rates/euro _ reference _ exchange _ rates/html/index.en.html . 6 This decreases to 804 observations with 99 banks when we concentrate on the ones that report all relevant bank control variables as shown in column 5 in Table 2 . via deposits of 60.8% of total assets. CET1 capital of the median bank stands at 17.3%, and, on average, 3.5% of total loans are provisioned for. Panels (B) and (C) show the summary statistics for the first quarter of 2020 and for the three remaining, pandemic, 2020 quarters (2020Q2 -2020Q4) separately. What stands out in the statistics is the main variable of interest, the quarterly change in total loans, normalised by the lagged total assets. In pre-pandemic times, this ratio is on average 0.9%. It drops to −2.8% in the first quarter of the pandemic and swings back to 1.0% in Q2-Q4 of 2020, a level that is very similar to the pre-crisis average, despite higher total assets. The share of loan loss reserves compared to total loans is not negatively affected in the first quarter of 2020. Instead, the share of provisions has increased given the negative change in total loans. In the subsequent quarters, the coverage ratio is decreasing. Table A2 in the Internet Appendix shows a correlation table for bank balance sheet characteristics in 2019Q4 as the pre-crisis quarter and the release in the CCyB rate in 2020Q1 and 2020Q2. Cash, deposits and the natural logarithm of assets have the highest correlation with the main independent variable. Also, countries with a higher real GDP per capita or disposable income decreased the CCyB rate more. In the regression analysis all these bank-and country-level characteristics are controlled for.
Turning finally to the bank-specific CCyB rates one can see the increase over time as shown in Fig. 1 as jurisdictions started to make use of this macroprudential tool. Also, the gradual phasein can be observed in the relatively larger increases in the first quarters of 2017, 2018 and 2019. The average CCyB stood at 0.91% in 2020Q1 and dropped to 0.23% in 2020Q2, when the following countries decreased their CCyBs: Czech Republic, Denmark, France, Iceland, Ireland, Lithuania, Norway, Slovakia, Sweden, and the United Kingdom.

Empirical results
We start our empirical analysis by examining whether the reduction in CCyB rates affected banks' lending. We subsequently discuss cross-sectional differences in the effect across bank capital characteristics and we differentiate between different types of loans. Last, we show that our results are robust to several robustness checks.

Bank lending
The main research question of this paper focuses on the effect of the countercyclical capital buffer release on credit supply. Table 2 summarizes the results for different model specifications. In column (1), we include only the reduction in CCyBs (the treatment intensity) interacted with the two post dummies, and we find a negative effect in 2020Q1 and a positive one in the quarters thereafter. However, the negative effect of the CCyB reduction is a result of the correlation between the reduction in the CCyB and the level of the CCyB: Banks that faced higher levels of CCyBs were the ones that experienced a larger reduction in the buffers. To show that this is indeed the case, we estimate Eq. (2) with the 2019Q4 level of CCyB and its interaction terms with the two post dummies in column (2). In this specification, the coefficient of the C C yB · Post 2020 Q1 becomes positive and the negative effect on lending in 2020Q1 is captured by the interaction term with the CCyB ratio, C C yB 2019 Q4 · Post 2020 Q1 , which is highly significant. This supports our argument that banks with high CCyB rates experienced a reduction in their lending in 2020Q1 and once we control for that we find that banks with a higher reduction in CCyBs had an increase in their lending in 2020Q1 as well. In column (3), we add bank specific control variables, while column (4) adds further country controls, COVID-19 cases, and policy controls, such as the size of central banks' asset purchases, liquidity facilities, and the overnight lending rate, as well as bank fixed-effects. In this last specification, the effect of the CCyB reduction, the coefficient of C C yB · Post, is positive for both 2020Q1 and 2020Q2-2020Q4. As reported in column (4), at the onset of the pandemic, in 2020Q1, banks with a 1 percentage point higher ex-ante capital buffer had a significant reduction in their lending by about 7.16 percentage points of their total assets. In the same quarter, a 1 percentage point reduction in the CCyBs increased banks' lending by about 6.29 percentage points of their total assets. This implies that the negative effect of holding high capital buffers was mostly removed by the CCyB releases at the onset of the pandemic. Over the following quarters, from 2020Q2 to 2020Q4, having a higher capital buffer ex-ante did not have a significant effect on banks' lending anymore. However, the positive effect of the CCyB reduction continued to be significant: A 1 percentage point higher reduction in the CCyBs resulted in a significant increase in banks' lending by about 1.93 percentage points of their total assets. This suggests that the release of the CCyB buffers was effective from the second quarter of 2020 onwards.
(3) where we report the average effect in the pandemic period by using all four quarters in 2020 as the post dummy. Since both C C yB · Post 2020 Q1 and C C yB · Post 2020 Q 2 −2020 Q 4 have the same sign, we continue with this specification as our baseline regression for the remainder of the paper. Our results show that, banks that experienced a higher CCyB relief had an increase in their lending. According to the coefficient estimate, a 1 percentage point reduction in the CCyBs led to a significant increase in banks' lending by about 5.63 percentage points of their total assets. However, at the same time, entering the pandemic with a higher CCyB led to a decrease in banks' lending: A 1 percentage point higher ex-ante capital buffer resulted in a significant reduction in banks' lending by about 4.55 percentage points of their total assets.
In our analysis, we study the changes in banks' total loans. In order to argue that the impact we capture comes from a change in the loan supply rather than the loan demand, we further use a more stringent specification and include country × year-quarter fixed effects in our analysis. Focusing on the banks that are headquartered in the same country in each quarter enables us to (partially at least) control for loan demand. We report the results in Table 8 . As shown in column (6), the positive impact on total loans remains with 10% significance level. The magnitude is similar to our baseline results: A 1 percentage point reduction in banks' Table 2 CCyBs and bank lending. This table shows how the reduction in CCyBs affected banks' lending in 2020. The dependent variable is defined as the first difference of total loans divided by total assets at the beginning of the period. C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. Column (1) includes only the reduction in CCyB interacted with the two post dummies (Post 2020 Q1 and Post 2020 Q 2 −2020 Q 4 , which are one in the respective time period and zero otherwise), whereas column (2) adds the level of CCyB in 2019Q4 and its interaction terms with the two post dummies. In column (3), bank specific control variables are added, while column (4) further adds country controls, COVID-19 cases, and policy controls, as well as bank fixed effects. Column (5) reports the average effect in the pandemic period by using all quarters in 2020 as the post dummy (Post 2020 Q 1 −2020 Q 4 ). The standard errors are clustered at the bank level. The symbols * * * , * * , and * denote significance at the 1%, 5%, and 10% levels, respectively.

Loans
(1) (2) (3) (4) CCyB × Post 2020 Q1 −3 .901 * * * 6 .033 * * * 6 .284 * * * 6 .285 * * (0 . CCyBs results in a significant increase by around 4.8 percentage points of their total assets. These findings provide useful evidence that a release of previously built up CCyBs, which reduces regulatory capital buffer requirements, can countervail banks' incentives to restrict credit supply in response to a negative shock. While such reaction may have been anticipated, or at least hoped for by e.g., regulators, there has nevertheless been some uncertainty prior to the pandemic about what exactly a potential CCyB release can achieve. For example, Haakon Solheim from the Norwegian Central Bank stated in a speech in December 2017 that "a number of challenges remain: it is still uncertain how banks will respond to a release of the CCyB" ( Solheim and Bank, 2017 ). Hence, our analysis shows that the additional space, which is created by the CCyB release, between the capital requirements and the level of actual bank capital, enables banks to continue lending. Stated differently, we find that banks Table 3 CCyBs and bank lending: Further controls. This table shows how the reduction in CCyBs affected banks' lending in 2020. The dependent variable is defined as the first difference of total loans divided by total assets at the beginning of the period. C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. Post 2020 Q 1 −2020 Q 4 is a dummy variable that is equal to one in 2020 and zero otherwise. All controls from column (5) in Table 2 are included. CE T 1 ratio, E conomic response , SyRB Osii , Asset purchases , Liquidit y f acilit ies and Weight ed C ov id per 10 0 0 interacted with the post dummy are included as further controls. We use the "economic support index" of the Oxford COVID-19 Government Response Tracker to measure the economic response of each government, Economic response , during the pandemic. SyRB Osii is the Systemic risk buffer that include both the systemic risk buffer (SyRB) and the capital buffer for other systemically important institutions (O-SII). Asset purchases is equal to one if the bank is operating in a country with asset purchases under all ongoing programmes in the specific quarter and zero otherwise. Liquidity f acilities is an indicator variable that is equal to one if the bank operates in a country with any extraordinary liquidity facilities provided to the banking sector. The standard errors are clustered at the bank level. The symbols * * * , * * , and * denote significance at the 1%, 5%, and 10% levels, respectively.

Loans
(1) (2) (3) (4) (5) (6) (7) use the capital that is freed up by CCyB releases for lending. Moreover, our findings add to the discussion about the timing of capital relief during crises. As the above results suggest, there might be the risk that the strong negative effect from entering the crisis with a high buffer in 2020Q1 could outweigh any positive lending effect from the buffer release itself if policy-makers do no react timely. This would imply that the general existence of a CCyB as policy tool could be debated. Thus, our finding provides a useful piece of information for policy decisions in favour of a quick reaction.

Further responses to the pandemic
As a next step, we analyze whether other bank characteristics or potentially confounding responses at the country level had an influence on the impact of the CCyB reduction on banks' lending behavior. Table 3 shows the results.
We control for the interaction of the post dummy with the following characteristics: CET1 ratios to control for possible effects of banks' regulatory capital ratios on their lending during the pandemic, economic responses of the countries, systemic risk buffers, asset purchases by the central banks, liquidity facilities, and COVID-19 cases per 10 0 0 people to control for the severity of the pandemic. We use the "economic support index" of the Oxford COVID-19 Government Response Tracker to measure the economic response of each government during the pandemic. Systemic risk buffers include both the systemic risk buffer and the capital buffer for other systemically important institutions. To examine the effect of asset purchases by the central banks by defining an indicator variable called "Asset purchases" that is equal to one if the bank is operating in a country with asset purchases under all ongoing programmes in the specific quarter and zero otherwise. To control for the effect of liquidity facilities, we again define an indicator variable that is equal to one if the bank operates in a country with any extraordinary liquidity facilities provided to the banking sector. Columns (2) to (7) report the results. According to the coefficient estimates, two controls have a significant effect on banks' lending during the pandemic: CET1 ratios and asset purchases. As reported in column (2), 1 percentage point higher CET1 ratios led to a significantly more bank lending by about 0.18 percentage point of total assets. This finding is in line with the literature that higher capital ratios result in higher lending during crises. In addition, as shown in column (5), banks operating in countries with an asset purchase program had significantly more lending by about 1.8 percentage points of their total assets. This implies that asset purchases helped banks to issue more loans during the pandemic. More importantly, however, controlling for these characteristics does not change the significant and positive effect of the CCyB reduction on banks' lending in this period.
Overall, controlling for these bank-and country-level characteristics does not lead to a change in the significant positive effect of the CCyB reduction on banks' lending. Once we include all interaction controls in column (8), we find that a 1 percentage point reduction in the CCyBs led to a significant increase in banks' lending by about 5.28 percentage points of their total assets, which is very similar to the coefficient estimate of 5.67 without the controls. This implies that the captured effect of the CCyB reductions on banks' loan supply is not affected by other relevant bank characteristics or country-level responses.

Economic magnitude
According to the coefficient estimate in column (5) of Table 2 , a 1 percentage point reduction in the CCyBs led to a significant increase in banks' lending by about 5.63 percentage points of their total assets. This corresponds to around EUR 8.10 billion additional loans for the average bank in our sample with an average size of EUR 144 billion during the pandemic. Given that the average re- Table 4 CCyBs and bank lending: Poorly-vs. well-capitalized banks. This table shows how the reduction in CCyBs affected banks' lending, their risk taking and capital in 2020 for poorly-and well-capitalized banks separately. The dependent variable is defined as the first difference of total loans (and risk-weighted assets respectively) divided by total assets at the beginning of the period. C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. Post 2020 Q 1 −2020 Q 4 is a dummy variable that is equal to one in 2020 and zero otherwise. A bank is defined as a well-capitalized bank if its capital headroom was above the median in 2019Q4 and poorly-capitalized is one for below the median. All controls from column (5) in Table 2 are included. The standard errors are clustered at the bank level. The symbols * * * , * * , and * denote significance at the 1%, 5%, and 10% levels, respectively. duction in CCyBs was 0.58 percentage point, this implies that the average bank in our sample supplied around EUR 4.7 billion additional loans as a result of the reduction in CCyBs, which is equivalent to around 3.26 percentage points of its total assets. 7 However, at the same time, entering the pandemic with a 1 percentage point higher ex-ante CCyB buffer resulted in a significant reduction in banks' lending by about 4.55 percentage points of their total assets. Given that the average bank-level CCyB buffer rate at the end of 2019Q4 was 0.83 percentage point, this suggests that the average bank experienced around EUR 5.44 billion reduction in its loans due to the counter-cyclical capital requirements. This implies that the net effect of the CCyB regulation on average banks' lending during the pandemic was a reduction of EUR 0.74 billion. When we study the onset of the pandemic, 2020Q1, and the following periods separately, as shown in column (4) of Table 2 , we find that the reduction happened mainly in the first quarter of 2020. The average bank experienced a decrease of EUR 8.55 billion in its loans as a result of holding high CCyBs at the onset of the pandemic. At the same time, the release of the CCyBs led to about EUR 5.25 billion increase in its loans. This implies a net effect of EUR 3.3 billion reduction. However, the negative impact of entering the pandemic with high CCyBs disappeared from 2020Q2 onwards and the positive effect of the release of CCyB was significant. According to the coefficient estimate of the CCyB reliefs, a 1 percentage point reduction in the CCyBs resulted in a significant increase in bank lending by about 1.93 percentage points of total assets from 2020Q2 to 2020Q4. This implies that the average bank increased its lending by about EUR 1.62 billion during this period. This suggests that the release of the CCyBs led to a positive effect on banks' lending from 2020Q2 onwards.

Well-versus poorly-capitalized banks
Poorly-capitalized banks are expected to be more capital constrained during the pandemic and, as a consequence, the reduction in CCyB rates would help these banks more to continue the provision of loans to their customers. To study this conjecture and to examine the differences in the effect on banks with higher versus 7 As a comparison, Lewrick et al. (2020) estimate a global extra lending potential of USD 5.3 trillion or 6% of total loans when releasing all possible capital buffers. Assuming usable capital buffers of USD 0.8 trillion, the CCyB contributes USD 0.1 trillion, which yields an extra lending potential of 0.75% of total loans in this backof-the-envelope calculation. lower headroom above their minimum capital requirements, we divide our sample of banks into two: banks with 2019Q4 capital headroom above the median ("High capital ratio") and those with below the median ("Low capital ratio"). Capital headroom is calculated as the difference between the CET1 capital ratios and the CET1 capital requirements. While the exact capital requirements of each individual bank is unfortunately not reported in the data, the formula for calculating them is as follows: CET1 capital requirements = Minimum CET1 ratio + Basel III capital conservation buffer + G-SIB buffer + SyRB and/or O-SII buffer + CCyB + Pillar 2 requirement For all banks, we know all these items except the Pillar 2 requirement, which is subject to supervisory discretion. Following this, we calculate an estimate of each bank's CET1 capital requirement by adding up the first five items: the minimum CET1 ratio (4.5%), the capital conservation buffer required under Basel III (2.5%), the capital buffer for global systemically important banks (G-SIB), the systemic risk buffer (SyRB), the capital buffer for other systemically important institutions (O-SII), and the CCyB of each bank. We use this estimate as a proxy for each bank's total CET1 capital requirement. As shown in columns (1) and (2) of Table 4 , although both poorly-and well-capitalized banks experienced a significant increase in their lending, the impact is much larger for poorlycapitalized banks that have capital ratios closer to the regulatory minimum requirements. We find that a 1 percentage point reduction in the CCyBs increased poorly-capitalized banks' loans by 8.35 percentage points of total assets at the 1% significance level whereas the increase corresponds to only almost 4 percentage points for well-capitalized banks at the 10% significance level.
We next study changes in banks' risk-taking by analyzing the change in banks' risk-weighted assets. To examine this, we change the dependent variable to be the change in banks' risk-weighted assets divided by last quarter's total assets. As reported in columns (3) and (4) of Table 4 , we find that both types of banks experienced an increase in their risk-weighted assets, where the increase is much stronger for poorly-capitalized banks. This is consistent with the findings that these banks had a significant increase in Table 5 CCyBs and bank lending: Different loan types. This table shows how the reduction in CCyBs affected banks' lending for different loan types as retail, corporate, retail mortgage, and other retail loans. Each dependent variable is defined as the first difference of loans divided by total assets at the beginning of the period. C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. Post 2020 Q 1 −2020 Q 4 is a dummy variable that is equal to one in 2020 and zero otherwise. All controls from column (5) in Table 2 are included. The standard errors are clustered at the bank level. The symbols * * * , * * , and * denote significance at the 1%, 5%, and 10% levels, respectively. their loans as a result of the buffer releases. This increase in their loans resulted in an increase in their risk-weighted assets.
Overall, our results imply that CCyB releases helped particularly the banks that were more capital constrained to continue their loan provision role during the pandemic. Release of the capital buffers provided extra capital space for these banks which enables them to continue their lending activity.

Different types of loans
The next interesting question is whether the increase in banks' loans happened in some specific type of loans. We differentiate between corporate loans and retail loans (the latter includes credit card lending). To investigate this, we again use the regression specification in Eq. (3) , but we change the dependent variable to be the change in the relevant type of loans divided by last quarter's total assets. The results are reported in Table 5 .
As shown in columns (1) and (2), we find that only retail loans increased significantly as a result of the buffer releases: A 1 percentage point reduction in CCyBs led to a significant increase in banks' retail loans by 5.05 percentage points of their total assets. When we analyze whether banks issued more retail mortgage loans or other types of retail loans, we find that only retail mortgage loans increased: A 1 percentage point reduction in banks' CCyBs resulted in a significant increase in their retail mortgage loans by 4.69 percentage points of their total assets.
On the other hand, banks on average did not experience an increase in their corporate loans. 8 To investigate this further, we examine the role of unused loan commitments. As documented e.g., by Li et al. (2020) and Acharya and Steffen (2020) , there was a significant increase in the loan commitment drawdowns by firms during the pandemic, which was much higher for banks that had higher unused loan commitments in 2019Q4, right before the pandemic started. One might thus expect that the banks that had to honor large amounts of loan commitment drawdowns could use the CCyB releases to issue the loans to their corporate customers. To analyze this, we use the 2019Q4 median value of unused loan commitments divided by total assets ratios to divide our sample: banks with high versus low unused loan commitments. According to our results reported in Table 6 , banks that had higher unused Table 6 CCyBs and corporate lending: High vs. low unused loan commitments. This table shows how the reduction in CCyBs affected banks' corporate lending for banks with high versus low unused loan commitments. High (low) unused loan commitments refers to the situation where the banks' 2019Q4 unused loan commitments divided by total assets is above (below) the median. C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. Post 2020 Q 1 −2020 Q 4 is a dummy variable that is equal to one in 2020 and zero otherwise. All controls from column (5) in Table 2 are included. The standard errors are clustered at the bank level. The symbols * * * , * * , and * denote significance at the 1%, 5%, and 10% levels, respectively.

High commitments
Low commitments (1) loan commitments, i.e., which were more exposed to larger drawdowns during the pandemic, experienced significant increases in their corporate loans. We find that a 1 percentage point reduction in the CCyBs led to a significant increase by 2.23 percentage points of their total assets. This implies that CCyB releases could help the banks that were in need of capital to honor the large amounts of loan commitment drawdowns during the pandemic.

Robustness checks 4.4.1. Alternative weights
In our main specification, we use the weighted average CCyB rates in each jurisdiction, where the weights are the proportion of bank branches in each country, to measure banks' CCyB rates. To examine whether our results are robust to using different weights, we repeat our analysis with three alternative measures: the GDPweighted country average, the equally-weighted country average, and the CCyB rate of the banks' headquarter country. Table 7 presents the results. All three measures yield very similar results to our baseline specification. The impact of the CCyB releases becomes less significant in the last specification where the headquarter country's CCyB ratios are used. This makes intuitive sense as banks have credit exposure outside of their headquarter countries, Table 7 Robustness check: Alternative weights. This table shows how the reduction in CCyBs affected banks' lending. The dependent variable is defined as the first difference of total loans divided by total assets at the beginning of the period. C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. CCyB GDP is the bank-specific weighted average CCyB rate where the weights are the GDPs of each country. CCyB equal is the bank-specific equally-weighted CCyB rate across countries. CCyB headquarter is the CCyB rate of the banks' headquarter country. Post 2020 Q 1 −2020 Q 4 is a dummy variable that is equal to one in 2020 and zero otherwise. All controls from column (5) in Table 2 are included. The standard errors are clustered at the bank level. The symbols * * * , * * , and * denote significance at the 1%, 5%, and 10% levels, respectively.

Loans
(1) which should be taken into account in the measurement of their CCyB rates. We further analyze the change in corporate loans with these alternative CCyB measures. As shown in Table A4 in the Internet Appendix, we find no significant effect which is consistent with the main results.

Alternative regression set-ups
We next show the results with different regression specifications as robustness checks. In our main specification, we use changes in the amount of loans divided by lagged total assets. We first document in column (1) of Table 8 that our results are stable when using the logarithmic changes of loans instead. Moreover, column (2) highlights that using the change in total credit supply (defined by total loans plus unused commitments) instead of the change in loans leads to similar results. Next, columns (3) - (5) show that using different pre-pandemic periods leads to very similar results.
One further potential concern of our analysis could be that the cross-country variation in CCyB releases may not be truly exogenous. This may cause a problem, for example if loan demand in one country was low during the pandemic, or if other unobservable factors were different, a CCyB release may potentially not have been required. This could correlate with lending differences across countries and lead to spurious results. As mentioned above, to overcome this worry, we add country × quarter fixed effects (to the existing bank fixed effects) in column (6). Hence, in this specification we compare banks that are headquartered in the same country in the same quarter but have a differential exposure to CCyB releases through their cross-country branch network. Stated differently, this setting allows us to account for time-varying unobservable effects at the country level, which may be seen as a proxy for loan demand. Our results survive this much more strin-gent specification: A 1 percentage point reduction the CCyBs leads to a significant increase in banks' lending by about 4.8 percentage points of their assets, which is of similar magnitude as our main specification but has a lower significance at the 10% level.
One further aspect is that our sample is unbalanced where not all banks report in all quarters. As the next robustness analysis, we repeat our regression for the subsample of banks that report in all quarters. Column (7) presents the results for the balanced subsample of banks. Although the sample size decreases, the results remain very similar. Next, we include house price indices in our analysis as an attempt to control for mortgage demand and our results remain very similar as reported in column (8).
Last, during the pandemic, some countries released the Pillar 2 capital requirements. As the final robustness check, we include a triple interaction term with a Pillar 2 dummy, where the dummy is one for countries that released the requirements, and zero otherwise. As shown in column (9), the significant positive impact of the CCyB releases on banks' lending is not affected by the Pillar 2 releases during the pandemic.

Further cross-sectional results
One might expect that banks with higher risk-weighted assets ratios would be more exposed to CCyBs and as a result the impact might depend on banks' risk-weighted assets. To study this, we divide our sample into two as banks with above median 2019Q4 risk-weighted assets (high risk density) and below the median (low risk density) and repeat our regression separately for these banks. As reported in Table 9 , we find that banks with a higher risk density reduced their lending more if they were exposed to higher CCyBs when they entered the pandemic (the coefficient in column (1) is double of that in column (2)). However, both types of banks (with high and low risk density) increased their lending similarly with the release of the CCyBs during the pandemic. Table 8 Robustness check: Alternative regression set-ups. This table shows how the reduction in CCyBs affected banks' lending using different regression specifications. C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. Post 2020 Q 1 −2020 Q 4 is a dummy variable that is equal to one in 2020 and zero otherwise. All controls from Column (4) in Table 2 are included. Column (7) focuses on a subsample of banks for which we have a balanced panel, and Column (8) includes the log of HPI (weighted at the country level) as an additional control variable. Last, Column (9) adds a dummy variable (P2R) that is equal to one if countries changed their Pillar 2 requirements in 2020, and zero otherwise. The standard errors are clustered at the bank level. The symbols * * * , * * , and * denote significance at the 1%, 5%, and 10% levels, respectively.

Panel A)
ln ( Table 9 Robustness check: Risk density. This table shows how the reduction in CCyBs affected banks' lending for banks that are exposed to a different risk density (the ratio of risk weighed assets to total assets). High (low) risk density refers to the situation where the banks' 2019Q4 risk density is above (below) the median. C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. Post 2020 Q 1 −2020 Q 4 is a dummy variable that is equal to one in 2020 and zero otherwise. All controls from column (5) in Table 2 are included. The standard errors are clustered at the bank level. The symbols * * * , * * , and * denote significance at the 1%, 5%, and 10% levels, respectively.

Loans
High risk density Low risk density (1) In addition, we further analyze whether banks that are active in countries with higher government loan guarantees during the pandemic would react to the CCyB release differently. We utilize the announcement-level database, which tracks the policies that countries adopted as a response to the pandemic, that is introduced by Kirti et al. (2022) . This database includes the loan guarantees that are announced in each country during the pandemic. To study the impact, we first aggregate the total amount of loan guarantees, focusing on the ones that affect the lending side of banks, during the pandemic for each country (scaled by the GDP). We then, similarly to above, calculate the weighted average loan guarantees for each bank in our sample, where the weights are the proportion of bank branches in each country. We divide our sample at the median and compare countries with high versus low government loan guarantees. According to our results reported in columns (1) and (2) of Table 10 , banks that are active in countries with lower loan guarantees experienced a significantly larger increase in their loans. We find that a 1 percentage point reduction in banks' CCyBs led to a significant increase by 12.3 percentage points of total assets for banks from countries with lower guarantees, whereas the impact is not significant for banks from countries with higher guarantees. Studying corporate and retail loans separately reveals similar results as reported in columns (3) to (6). This might indicate that loan guarantees decrease the riskiness of loans and that banks can issue loans independent of the reduction in the CCyBs. As a result, the CCyB reduction does not have a significant effect on banks' lending in countries with high loan guarantees. This finding suggests that the loan guarantees and the CCyB reduction can be used as substitutes by governments during economic downturns as banks seem to not utilize the CCyB reduction if the governments provide loan guarantees. Table 10 Robustness check: Government loan guarantees. This table shows how the reduction in CCyBs affected banks' lending depending on high or low government loan guarantees (as measured by guarantees applied to the lending side of financial institutions of each country, split by the median). C C yB i is the treatment intensity defined as the reduction in the bank-specific CCyB rate from the last quarter of 2019 to the second quarter of 2020. Post 2020 Q 1 −2020 Q 4 is a dummy variable that is equal to one in 2020 and zero otherwise. All controls from Column (4) in Table 2 are included. The standard errors are clustered at the bank level. The symbols * * * , * * , and * denote significance at the 1%, 5%, and 10% levels, respectively.

Conclusion
As a response to the unprecedented COVID-19 pandemic, almost all participating jurisdictions in Europe released their countercyclical capital buffers. Since their establishment in January 2016, CCyBs have only increased and this was the first time that this novel macroprudential tool was released. In this paper, we thus examine how European banks adjusted their lending subsequent to the release of their CCyBs. Stated differently, we ask whether the general goal of the CCyBs, that is to help maintain the supply of credit to the economy and to mitigate the downturn of the financial cycle, was achieved.
Using a difference-in-differences analysis, we show that the release of the previously built-up countercyclical capital buffers helped banks to provide more loans during the pandemic. While at the onset of the pandemic, being exposed to a higher ex-ante CCyB led a reduction in banks' lending, the relief of the CCyBs removed this negative effect. From the second quarter of 2020 onwards, the negative effect of higher ex-ante CCyBs disappeared, and we find a significant positive effect of the countercyclical capital releases. Overall, our results suggest that the release of the CCyB buffers was effective from the second quarter of 2020 onwards.
The contribution of this paper is therefore twofold: First, while the existing literature has focused on the effects of phasing in CCyBs in economic boom periods, we add novel insights of how the buffer release affects credit supply during a crisis. Our results hence complete the aggregate picture and allow to draw an overall conclusion on the effectiveness of the CCyB as a macroprudential policy tool. Second, our findings add to the discussion about the timing of capital relief during crises. Being exposed to higher CCyBs leads to a slowdown in banks' lending. While beneficial in boom times in order to avoid excessive lending and overheating of the economy, the problem may arise during a crisis when there might be the risk that such undesired effect could outweigh any positive lending effect arising from the buffer release itself, if policy-makers do no react timely. This would imply that the existence of a CCyB as policy tool could be debated. Thus, our finding provides a useful piece of information for policy decisions in favour of a quick reaction.

Data availability
The authors do not have permission to share data.

Supplementary material
Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.jbankfin.2023.106930 .