Curbing risk in the financial system: Early evidence of the countercyclical capital buffer

50  Download (0)

Full text


Curbing risk in the financial system:

Early evidence of the countercyclical capital buffer

Author: Miguel Krol, 11046538 Supervisor: dhr. Prof. Dr. A.C.F.J. Houben Master thesis MSc Economics: Monetary policy & Banking, 15-08-2021 Faculty of Economics and Business, University of Amsterdam




In the recent decade, advanced economies have experienced a wider implementation of macroprudential policies to curb large swings in the financial cycle. This thesis studies the effect of the countercyclical capital buffer on the development of two bank risk indicators, namely the growth of leverage and lending. Its effectiveness is assessed by using a two-step system GMM model, using data from bank balance sheets in 28 advanced economies over the 2006-2020 period. The results indicate that the countercyclical capital buffer has a strong and significant impact in curbing bank risk. Its effect is especially strong during the upward phase of the financial cycle, signalling this policy acts as a binding constraint during a boom.


This document is written by Miguel Krol, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.




1. Introduction ... 4

2. The countercyclical capital buffer ... 7

2.1 Basel III’s capital framework ... 7

2.2 Institutional setup and CCyB objectives ... 8

2.3 Credit-to-GDP as a reference indicator ... 9

2.4 Rate setting and geographical implementation ... 10

2.5 Alterations to Basel III’s countercyclical capital buffer ... 11

3. Existing literature on the effects of time-dimensional macroprudential policies ... 13

3.1 Expected bank behaviour upon CCyB implementation ... 13

3.2 Empirical studies on the countercyclical capital buffer ... 14

3.3 Dynamic provisioning, loan-to-value and debt-to-income caps ... 15

4. Hypotheses ... 17

4.1 Bank risk indicator: Leverage growth ... 17

4.2 Bank risk indicator: Loan growth ... 19

4.3 Effects of the CCyB by phase of the cycle ... 19

5. Methodolodgy ... 21

5.1 Model specification ... 21

5.2 Robustness specifications ... 23

5.2.1 Adding macroeconomic variables ... 23

5.2.2 Adding government debt and banking crisis variables ... 24

5.2.3 Adding financial industry variables ... 24

6. Data description ... 25

7. Results ... 30

7.1 The countercyclical capital buffer on bank risk... 30



7.2 Robustness analysis ... 34

8. Discussion ... 36

9. Conclusion ... 38

References ... 39

Appendices ... 46

Appendix A ... 46

Appendix B ... 48

Appendix C... 49




In the past decade, the economic ramifications of the Great Financial Crisis (GFC) sparked the urge to adopt a new framework targeting system wide financial risk. After all, it became apparent that microprudential and monetary policy alone had not been enough to contain the build-up of systemic risk. Since then, a growing consensus among academics (Lim et al., 2011; Galati & Moessner, 2012; Claessens, Ghosh & Mihet, 2013) and policymakers (BCBS, 2010; ERSB, 2014) recognized the need to implement a set of macroprudential regulations. While a broader implementation was previously confined to emerging economies, since then the design and implementation of these policies in developed economies gathered speed (Cerutti, Claessens, & Laeven, 2017).

Macroprudential regulation in general is intended to mitigate the build-up of financial vulnerabilities over time, and improve the resilience of the banking system at any given point in time (BCBS, 2010). In other words, it targets systemic risks that are both time-dimensional and structural in nature. When assessing time-dimensional risk, regulators often examine a number of areas like the growth in total credit, housing prices or foreign currency mismatches (IMF-FSB-BIS, 2016). Some macroprudential policy instruments are designed to be calibrated depending on developments in these risk areas. Such calibration often requires building up buffers as systemic risk builds up, and releasing them in times of stress. In the case of structural risk, regulators monitor contagious linkages between financial intermediaries, market infrastructures and the impact of individual banking failure on the system as a whole.

The effectiveness of policy instruments that target structural risk can be assessed whenever a financial crisis occurs, and thus their effectiveness is difficult to measure on a year-to-year basis. For instruments that target the accumulation of financial imbalances, and are more intensely calibrated, this annual assessment is more straightforward. As such, this thesis focusses on measuring the effectiveness of these type of macroprudential policy instruments.

Besides better measurability of policy effectiveness, studying the effects of instruments that deal with the build-up of systemic risk is also highly important. After all, financial imbalances accumulate over a long period of time, while they are quick to unravel and have long lasting economic costs (Borio, 2014). Thus, it is important for policymakers to explicitly incorporate both phases of the financial cycle and not solely focus on high-frequency shocks in the markets.


5 Even though Basel III’s countercyclical capital buffer (CCyB) is by nature a policy instrument that targets time-dimensional risk, its main objective is to cushion the downward phase of the financial cycle (BCBS, 2010). It follows the reasoning that if the CCyB rate is set high enough preceding a crisis, a buffer release could ensure the system can cope with a downturn. Practices like fire-sales and credit restrictions that exacerbate the downturn can then be mitigated (Borio, 2010). The BCBS considers the CCyB’s dampening impact during the build-up phase of the cycle a “potential side-benefit”. That is, banks that are requested to increase their capital buffers, either reduce their risk-weighted assets or increase their equity (Solheim, 2016). This will serve as a brake during the upturn of the financial cycle.

The case of the United Kingdom serves as a good illustration. It implemented a countercyclical capital buffer of 1% in November 2018, but decided to release this buffer in March 2020 to soften the economic shock of Covid-19. In a press statement the Bank of England announced that its release would support up to £190 billion of bank lending, equivalent to 13 times banks’ net lending to businesses in 2019 (Bank of England, 2020). The following questions arise: Did the countercyclical capital buffer in the United Kingdom restrict banks’ lending activities when the policy was implemented in 2018? Two years later, did the buffer release support banks’ lending activities?

This thesis performs panel data regressions for macroprudential policies on financial system vulnerabilities in 28 countries. In the regression models, two financial system vulnerability indicators are considered. First, the leverage growth rate indicates the change of a bank’s indebtedness, and an adjustment to the maximum amount of losses it can incur before becoming insolvent. Second, the growth rate of outstanding loans reveals a bank’s confidence to expand its main business operation. Across 28 countries, 200 banks are identified during the period from 2006 to 2020, to help answering the following research questions: Did the countercyclical capital buffer reduce the cyclicality of bank leverage and lending activity? Is the effect of the countercyclical capital buffer different in the financial upturn, than in the downturn?

Besides the countercyclical capital buffer, other policy instruments are included in the regressions as well. Dynamic provisioning (DP), another instrument that is designed to be calibrated according to the phase of the financial cycle, is expected to behave in similar fashion to the CCyB. Policy instruments which constrain borrowers, namely loan-to-value (LTV) and debt-to-income (DTI) caps, are regressed on bank risk variables too. According to Cerutti and


6 Claessens (2015), LTV and DTI policies have been actively adjusted in intensity over time and have thus been able to consistently affect credit growth and housing prices. As a result, the implementation of these policies can behave countercyclically too.

This research paper extends the existing literature in three ways. First, most research that studies the effect of macroprudential policies on the build-up of financial imbalances, does so on the aggregate level (Lim et al., 2011; Cerutti, Claessens, & Laeven, 2015). By studying bank behaviour instead, a better understanding is formed on how key actors in the financial system react to certain macroprudential policy decisions. In addition, by studying banks a larger sample is obtained and fewer endogeneity concerns arise. Second, most papers study the effects of macroprudential policies from a global or emerging markets perspective (Lim et al., 2011; Claessens, Gosh & Mihet, 2013; Tantasith et al., 2020). The reason is that emerging markets initially outpaced advanced economies in terms of actual macroprudential policy usage (Cerutti, Claessens, & Laeven, 2017). Since the GFC however, policy implementation in advanced economies has steadily caught up. Studies that cover a sample period ending around the GFC, are not able to account for these effects in advanced economies only. Third, few studies consider the intensity of policy implementation (Lim et al., 2011; Claessens, Gosh & Milhet, 2013; Cerutti, Claessens, & Laeven, 2015; Olszak & Kowalska, 2019). The different implications of setting a CCyB rate at either 0.5% or 2.5% is thus ignored.

In particular, the property of dummy variables prevents existing research from incorporating the impact of policy variability. To the best of knowledge, no other study has researched the impact of changes in the effective CCyB rate in a cross-country framework.

The remainder part of this thesis is structured as follows. Section 2 introduces the role of the CCyB and describes its implementation so far. Section 3 discusses the effects of macroprudential policies on curbing systemic risk. Section 4 formulates three hypotheses for this study. The methodology is laid out in section 5 and section 6 describes the data used for analysis. Section 7 presents the results, which are discussed in section 8. Section 9 concludes this thesis.




2.1 Basel III’s capital framework

In December 2010, the Basel Committee on Banking Supervision (BCBS) published an international regulatory framework document for building a more resilient banking system (BCBS, 2010). Its main purpose was to strengthen bank capital requirements, develop sophisticated liquidity standards, improve risk assessments and promote market discipline. In the aftermath of the global financial crisis, many banks resorted to bankruptcy and governments across the world bailed-out banks that were considered “too big to fail”.

Apparently, the banking system was unable to absorb the losses resulting from a systemic collapse. The Basel III capital framework addresses these capital gaps.

The core capital requirement for banks determines how much of their total funding must originate from common equity, meaning shareholders equity or retained earnings. Other funding arises from liabilities, like deposits or short-term market borrowing. Additional Tier 1 (AT1) capital includes preferred shares and contingent convertible securities. Both types of capital measure a bank’s ability to absorb losses and to remain a “going-concern”. Among other funding sources, Tier 2 (T2) capital includes subordinated debt and general provisions and can be labelled as a “gone-concern” form of capital. In the case of bankruptcy, it can act as protection for deposit holders, or governments in case of a bail-out outcome. The capital adequacy ratio is computed by dividing the eligible bank capital by the total amount of risk- weighted assets. Introduced under Basel I in 1998, assets were assigned a certain weight from 0% to 100% depending on their perceived riskiness. Basel II introduced a more tailorized approach six years later. Banks were allowed to follow a more precise standardized approach (SA) or develop an internal-ratings-based (IRB) approach to determine risk-weighted assets.

Basel III requires banks to hold a minimum amount of common equity of 4.5% to risk-weighted assets, and a minimum of 6% including additional T1 capital. Tier 2 capital has a regulatory minimum of 2%.

In addition to these minimum requirements, Basel III introduced four new type of capital buffers. First, the countercyclical capital buffer that can be set by the national regulatory authority up to 2.5% of risk-weighted assets. It is the only capital buffer designed to vary in accordance with different stages of the financial cycle, and as a result being explicitly


8 macroprudential. Since 2015, 15 countries have put into effect a non-zero rate for the CCyB.

Second, the capital conservation buffer sits on top of minimum capital requirements and acts as an additional buffer equal to 2.5% of risk-weighted assets. This buffer stays constant over time and when breached, regulatory action limits the amount of dividends and bonus payments a bank can make. It is required for all type of banks that fall within the jurisdiction of the BCBS. Third, systemically important banks are required to have additional core capital on their balance sheet. For Globally Systemically Important Banks (G-SIBs), additional capital requirements range from 1% to 3.5% of their risk-weighted assets. The Financial Stability Board publishes a list of eligible G-SIBs each year and makes systemicness assessments based on the bank’s level of cross-jurisdictional activities, size, complexity and substitutability. In 2020, the number of identified G-SIBs remained at 30 (Financial Stability Board, 2020). The minimum requirement for Domestically Systemically Important Banks (D-SIBs) complements the G-SIB requirement, and its implementation is based on assessments by local authorities.

The number of banks that are required to comply with D-SIB regulation amounted to 132 in 2018, and where mostly located in Asia, Europe and North America (FSB, 2020).

Fourth, in Europe a regulatory package known as CRD IV/CRR introduced another type of capital buffer in 2014, called the systemic risk buffer. At national discretion and without an existing binding cap, the buffer is intended to mitigate systemic risks of a long-term and non- cyclical nature. The buffer amounts to 2.5% of risk-weighted assets, and 14 European countries have activated the buffer since 2014.

2.2 Institutional setup and CCyB objectives

The authority for monitoring the relevant financial indicators and setting the appropriate countercyclical buffer rate, falls either under the national central bank or financial supervisory authority. In some countries a special inter-agency committee holds the main responsibility over the CCyB implementation. Committee members usually are high-level officials from the central banks, financial supervisory authorities, government agencies or academia (Arbatli-Saxegaard & Muneer, 2020). These committees reside outside the central bank in Finland and France, while they operate within the central bank in the United Kingdom and Iceland. The BCBS has set out international guidelines so that policies are conducted everywhere under a shared set of principles (Basel Committee on Banking Supervision, 2010).

The European Systemic Risk Board has also laid out its recommendations for setting the CCyB


9 (European Systemic Risk Board, 2014). Its countercyclical capital buffer regime follows the principle of “guided discretion”. It is primarily a rules-based approach together with the exercise of national discretionary powers when setting an appropriate buffer rate. In addition, the ECB has a topping up discretion. This one-sided policy discretion reflects concern on the inaction bias national authorities may be prone to. In practice, the ECB has never used its topping up powers.

Guidelines mainly relate to policy objectives, rate setting and communication. The BCBS (2010) explains that the main objective of the countercyclical capital buffer is to provide the banking system with additional loss absorbing capacity, so that pro-cyclical bank behaviour in a down-turn is mitigated. The rationale is that if banks hold on to enough capital before system-wide financial stress occurs, they can maintain the flow of credit to the private sector without jeopardizing their solvency. Thus, the main objective of the countercyclical capital buffer focuses on release phase of the buffer. A side benefit comes from notion that it may lean against the build-up of excessive risk taking in the first place. The ERSB (2014) also argues that the CCyB can dampen excessive credit growth during the upswing of the financial cycle.

For both institutions, moderating the financial cycle is not seen as a secondary objective, but rather as an additional positive outcome. In Belgium, the United Kingdom, Sweden, Hong Kong, Estonia and the Czech Republic it is explicitly stated that the countercyclical capital buffer is not implemented to “lean against the wind” during the build-up phase of the financial cycle (Arbatli-Saxegaard & Muneer, 2020). In other jurisdictions however, the impact of the buffer during the build-up phase is considered as a policy objective too. In France for example, dampening the upswing is regarded as a secondary objective, while in Spain higher relative importance is placed on the upward phase of the cycle (ERSB, 2017).

2.3 Credit-to-GDP as a reference indicator

The common reference guide for setting the CCyB rate is the BIS credit-to-GDP gap.

Based on total credit to the non-financial sector, the gap can be calculated as the difference between the credit-to-GDP ratio and its long-term trend. This long-term trend is computed with a one-sided Hedrick-Prescott filter, and the lambda parameter that smoothens historical observations is set to 400,000 (ERSB, 2014). Consequently, this HP filter assigns high weights to the most recent observations, making it capable of dealing effectively with structural breaks. The size of the CCyB buffer add-on is zero when the credit-to-GDP gap is below a


10 certain threshold and is to be linearly increased when a higher threshold is reached. Detken et al. (2014) argue that the BCBS benchmark gap is likely to perform well in signalling crises, but believe it is unlikely to be suitable for each individual country. In the European Union for example, there are persistent differences in the credit-to-GDP gap between countries due to other reasons than credit cycles, like differences in financial structure or financial development. Therefore, to avoid the risk of misleading signals, countries should assess a broad set economic indicators which can include various asset prices, CDS spreads, credit condition surveys and real GDP growth.

2.4 Rate setting and geographical implementation

It matters at what time during the build-up phase of the financial cycle the CCyB is activated. Arbatli-Saxegaard and Muneer (2020) give two reasons. First, a buffer set at an early stage can protect against uncertainty associated with measuring financial imbalances and the lags between buffer announcement and rate implementation. Second, it provides room for more gradual rate setting and thus reducing the costs for banks when capital requirements are increased. Ireland, Denmark and Norway have an early build-up policy in place that requires activating the buffer before excessive credit takes place. Even more prudent policy exists if the regulator sets a non-zero default level for the buffer, when systemic risks are neither elevated nor subdued. The Bank of England’s Financial Policy Committee (FPC) announced a 1% default CCyB rate in 2016. Before the Covid-19 crisis was commonly understood to have a severe impact on the economy, the FPC announced it would raise the default rate to 2% in December 2019. The Czech Republic and Lithuania have also set a non- zero default level for the buffer during moderate risk level periods. According to Stojkov (2020), a positive default CCyB rate improves overall buffer usability. Except for soon after release, it allows regulators to release capital requirements at any given point in time. This improves the resistance of a financial system to a greater variety of shocks, especially for those which do not originate from within the financial system and for which leading indicators are not able to warn ahead of time. The Hong Kong protests, Brexit and the outbreak of the Covid- 19 pandemic are good examples where a positive non-zero CCyB rate made regulators better able to lower capital requirements.

Concerning the release of the CCyB, it can be either immediate or gradual. Most countries do not have formal frameworks in place regarding the type of release that certain


11 circumstances would warrant in absence of a deep recession. Not imposing gradual cuts in the buffer rate when financial imbalances wane, implies that policymakers regard the CCyB as an asymmetrical policy instrument. It could also be explained by the fact that the policy instrument is a new phenomenon, and that countries have little experience with releasing the CCyB. Also, a more immediate release during crisis periods constitutes to the BCBS’s primary objective, ensuring that bank can maintain lending operations during periods of financial distress. During the Covid-19 crisis, most countries with non-zero CCyB rates followed this notion and lowered their effective rate to zero percent (Appendix A). However, the Czech Republic, Hong Kong, Slovakia and Norway keep non-zero effective rates as of December 2020. Part of the reason is that these countries had relatively high effective rates before the crisis broke out. Luxembourg even increased its effective CCyB rate from to 0.25% to 0.5% in 2020.

Jurisdictional reciprocity is applied to internationally active banks. Domestic authorities set a specific buffer requirement that only applies to credit exposures located in their jurisdiction. They are obliged to inform foreign banks on their buffer decisions and responsible for supervising domestic banks on their foreign exposures in countries with positive CCyB rates (BCBS, 2010). The relevant authority should inform all banks 12 months prior to the date CCyB comes into effect. When the CCyB is released, new regulations apply with immediate effect.

2.5 Alterations to Basel III’s countercyclical capital buffer

Besides a countercyclical capital buffer targeted towards all risk-weighted assets and applicable to all active banks, Switzerland and Canada have introduced two alternatives. First, Switzerland became the first country to enforce a form of countercyclical capital regulation in 2013, which is exclusively targeted towards the mortgage market. The buffer is activated when domestic mortgage credit growth is perceived to be excessive and requires banks to build-up capital relative to their residential mortgage related RWA. The Swiss application of the CCyB has the advantage of being more impactful towards a sector of concern, which is particularly useful when excesses take place in a confined area of the economy (Behncke, 2020). On the other hand, this application has smaller system-wide effects than the Basel III CCyB. A similar transmission to the economy can be expected like the regular CCyB, as its two goals are to increase loss absorbing capacity and lean against excessive credit growth (SNB, 2014). Second,


12 Canada has implemented a domestic stability buffer (DSB) since 2018. It is similar to Basel III’s CCyB, but applies only to its six domestic systemically important banks and targets their foreign exposures too. Given the large scale of their foreign operations, the DSB requires these Canadian banks to build up more capital than the equivalent domestically oriented CCyB would do (Stojkov, 2020).




3.1 Expected bank behaviour upon CCyB implementation

As Solheim (2016) summarizes, banks can comply with a higher effective CCyB rate by doing either of two things. They can increase the level of CET1 capital by reducing dividends or raising new equity. On the other hand, banks can also reduce the relative size of risk- weighted assets, by reducing lending or shifting towards assets with lower risk-weights.

Solheim (2016) states that historically, except for the global financial crisis, the biggest contributions to meeting higher overall capital requirements have been retained earnings.

When regulatory shocks apply, banks can adjust dividends without creating large disruptions to their stock price. Cohen and Scatigna (2014) also find that retained earnings account for the bulk of higher capital ratio adjustments, while reductions in risk-weights play a lesser role.

Lower dividend pay-outs and higher interest rate spreads have allowed banks to build up their capital after the GFC. They argue that banks continued the growth of their lending activities in most places, though in Europe higher capital requirements slowed lending. The latter is confirmed by Imbierowicz et al. (2019), who find that higher capital requirements in Germany have resulted in immediate decreases in the growth of domestic and cross-border bank lending. On the other hand, Stojkov (2020) mentions there are reasons to believe that an increase in the CCyB may not pull down bank lending during a financial boom. Lending during this phase of the financial cycle yields high returns and banks are thus willing to internally generate the required capital and expand their balance sheet. In addition, normally all banks hold voluntary levels of capital during each phase of the cycle. This means that a gradual rise in the CCyB rate is not immediately a binding constraint, so that in the short run lending is not necessarily to be reduced. Similar skepsis holds for the CCyB’s effect during a sudden buffer release during a crisis. Publicly traded banks might be disincentivized to lower their capital ratio due a pushback by the market. As the financial environment is uncertain and banks face capital adjustment costs, they might prefer holding on to additional voluntary buffers instead (Gambacorta & Mistrulli, 2004).


14 3.2 Empirical studies on the countercyclical capital buffer

The broad-based Basel III CCyB is a new macroprudential tool at the disposal of national authorities that was first implemented in 2015. Only in 2020 did countries release this buffer for the first time. Partly due to the short time horizon, the empirical effect of the CCyB has not been much studied. Instead, most research has focused on measuring its effect on bank activity by using either a capital requirement proxy or performing macroeconomic modelling (Drehman & Gambacorta, 2012; Akram, 2014; Tayler & Zilberman, 2016; Chen, Sivec & Vector, 2018). Drehmann and Gambacorta (2012) provide a counterfactual simulation to investigate whether the so called “side-benefit” of the CCyB, namely the containment of excessive credit growth during the build-up phase, may indeed exist. Using available bank capital as a proxy for the CCyB, they find that a time varying capital buffer can have a material impact on decreasing bank pro-cyclicality. Tayler and Zilberman (2016) apply a DSGE model to identify the interaction between credit markets and the real business cycle. Following a credit supply shock, an aggressive countercyclical regulatory response is found to have a more significant impact on financial stability than monetary policy. Akram (2014) finds that time- varying capital requirements have had a significant impact on house prices and credit supply in Norway. He argues that the same holds for the CCyB, by using a VECM model that assumes the same directional responses between regular and time-varying capital requirements. Chen, Sivec and Volk (2018) have experimented with an exogenous buffer release to the Slovenian banking system. Due to the country’s switch to International Financial Reporting Standards (IFRS) in 2006, banks were suddenly allowed to hold less loan loss provisions than under the previous regime. The deduction item was called a prudential filter and amounted to 0.8% of risk-weighted-assets. As the new regime went into immediate effect, it mimics the release phase of the CCyB. Their findings suggest the CCyB helps to mitigate sharp decreases in the supply of credit during a crisis.

Two studies come closest to a real empirical study on the CCyB’s short-run effects (Behncke, 2020; Basten, 2020). The mortgage targeted CCyB in Switzerland came into effect at the end of September 2013, and early bank-level data could potentially indicate what effects are to be expected from Basel III’s CCyB. Behncke (2020) concludes the CCyB in Switzerland has guarded against a further accumulation of mortgage risks without negative side-effects. A treatment group of capital constrained banks, for which the introduced policy


15 is binding, increases the share of low risk-weighted mortgages relative to high risk-weighted ones upon buffer tightening. Other banks reduce the total amount of mortgage RWA through a decline in their mortgage growth rate. Basten (2020) finds no evidence for explicit credit rationing. However, banks that were more capital constrained increased their mortgage pricing. As a result, growth of their mortgage lending slowed and capital cushions were rebuilt.

System-wide effects were less significant, but the composition of mortgage lenders shifted towards banks that were beforehand less exposed to these types of loans.

3.3 Dynamic provisioning, loan-to-value and debt-to-income caps

The countercyclical capital buffer is only one of the many policy instruments that targets time-dimensional systemic risk. For these type of policy instruments, a distinction is made between borrower-based and financial institutions-based policies (Cerutti, Claessens &

Leaven, 2017). The former is designed to restrict borrowing capacity, usually for individuals or households. It includes a maximum loan amount relative to the value of a property (LTV) or a fixed multiple of household income (DTI). In most cases, the purpose of these instrument is to limit household leverage and to prevent property markets from overheating. On the other hand, financial institutions-based instruments are set up to limit the capacity of banks to expand operations. Examples are capital requirements, dynamic provisioning and limits on inter-bank exposures.

Many studies have found that macroprudential policies in general have been effective in reducing the pro-cyclical nature of credit (Lim et al., 2011; Claessens, et al., 2013; Cerutti, Claessens, & Laeven, 2015). However, for a broad set of instruments, Cizel et al., (2019) find evidence for substitution toward nonbank credit. This substitution effect is especially large if these policies are binding, and if a country’s nonbank capital market is well developed. These findings pose limits on a regulator’s ability to control the supply of total credit in the economy within the current macroprudential framework.

At the bank level, measures that are aimed at borrowers are found to be effective in reducing leverage, assets and noncore to core liabilities growth during the build-up phase of the financial cycle (Claessens, Gosh & Milhet, 2013). They can also limit the share of risky mortgages, and slow the growth of high-LTV mortgages, and as a consequence making economies more resilient (Arena et al., 2020). The effect of an LTV cap can be even stronger when complemented with a DTI ratio, like in Poland, or when its calibration varies according


16 to the conditions in the residential real estate market, like in Korea and Hong Kong (Brockmeijer, 2012). Important to note is that borrower-based policies can be either targeted or broad-based and a cap can be either binding or flexible. In Thailand for example, there exists a large variation of LTV measures. There, the effect of these policies did not manifest itself in a slower pace of credit growth, but rather in changes to the LTV distribution of new loans (Tantasith et al., 2020).

Besides Basel III’s countercyclical capital buffer, dynamic provisioning (DP) is another financial institutions-based policy that is designed to curb credit cycles. Most commonly, it requires banks to make provisions against a benchmark rate that is tied to the credit cycle, or it is set with discretion by the national regulator (Lim et al., 2011). Spain has become the first country to adopt the former DP framework in 2000, and there it has smoothened the credit cycle and supported firm performance during crisis times (Jiménez et al., 2017). These buffers were large enough to cover almost half of all loan losses by Spanish banks in the two years after the crisis begun in 2008, not including delinquencies.

According Altunbas et al., (2018), bank specific characteristics also play an important part in the transmission of macroprudential policy on bank risk indicators. They find that banks with lower capital ratios, small size and fewer deposits react stronger to changes in these policies. When policies are tightened, these balance sheet characteristics cause banks to suffer more from a high degree of informational frictions in capital markets. As a result, they face higher costs in raising wholesale funding and are required to reduce their lending more than other banks.




Summing up the collected literature, there exists a significant and often strong relationship between macroprudential policy and mitigating financial cycles. This effect is not exclusively valid during financial downturns when these policies have improved bank resilience a posteriori, but can also be effective in taming excessive risk taking before the cycle turns. This research paper formulates three hypotheses that help determine whether these expectations are to be justified in the given sample. The first two hypotheses differ only in the nature of the dependent variable, as the first examines the effect of the CCyB on the growth of Basel III’s definition of the leverage ratio, and the second on loan growth. The third hypothesis tests if the results in the previous two differ according to the phase of the financial cycle. With respect to the four available macroprudential policies in this study, only the countercyclical capital buffer is admitted to the three hypotheses. As dynamic provisioning, the loan-to-value and debt-to-income caps are represented only with dummy variables, there exists the possibility that actual policies are not well represented by these variables.

Nevertheless, in the results section empirical findings on these policies are discussed.

4.1 Bank risk indicator: Leverage growth

Prior to the GFC, banks cranked up their leverage both on and off their balance sheets.

Often banks increased their leverage to precariously high levels, and by virtue of maintaining seemingly adequate risk-weighted capital buffers, their financial position was believed to be acceptable. Several studies describe how financial booms have the tendency to compress risk- weights with respect to capital requirements (Borio & Zhu, 2012; Vallacas & Hagendorff; 2013;

Altunbas, Gambacorta & Marques-Ibanez, 2014). According to Kashyap, Stein & Wilcox (1993), favourable economic conditions result in the increase in available profitable projects in the economy, and higher demand for loans. As a result, most banks increase their leverage in terms of exposure during a financial boom. Basel III’s capital framework introduces a backstop to this phenomenon, with a regulatory leverage minimum of 3% (BCBS, 2014). In contrast to capital ratios, Basel III’s definition of the leverage ratio compares a bank’s Tier 1 capital relative to total exposure. It is expected to behave countercyclically, in the sense that it is binding during a financial boom, when banks are keen on increasing their exposure beyond the regulatory limit. In addition, there exists evidence that banks have consistently


18 underpredicted internal risk estimates in order to lower capital requirements (Behn et al., 2016). A non-risk-weighted prudential backstop therefore complements existing capital regulation.

When considering a bank’s leverage ratio at a given point in time, Brei and Gambacorta (2014) explain this is the maximum loss that a bank can absorb given its amount of available capital. As such, it is believed to be a good indicator for estimating the risk of banks. Following this reasoning, it can be tested whether macroprudential policies can mitigate the growth of leverage. The empirical evidence explains well how bank leverage has reacted to the financial cycle in the past. Financial sector leverage in the United States before the financial crisis has behaved pro-cyclically (Adrian & Shin, 2010). An important distinction is that commercial banks opt for a leverage ratio that is fixed, while other financial intermediaries like investment banks increase the leverage ratio when the economy is booming. Commercial banks are imposed more regulatory constraints, while the latter group is not subject to similarly binding regulation and has more capacity to increase leverage whenever confidence in the economy is strong. The authors describe the mechanism of pro-cyclical leverage as a positive feedback loop between rising asset prices and credit supply to the private sector. In Europe’s prevailing banking model of the “universal bank”, where commercial and investment activities are grouped together, leverage also reacted pro-cyclically to the financial cycle (Baglioni et al., 2013). Also, bank leverage between 2008 and 2012 seems to have been more pro-cyclical than in the decade before the crisis (Brei & Gambacorta, 2014).

The Basel III countercyclical capital buffer has been implemented only recently in advanced economies and bank leverage is not a primary objective, as long as it remains well clear of the back-stop minimum. This explains why no empirical studies have yet researched whether Basel III’s CCyB has had an influence on the growth of leverage. However, studies do expect this relationship to materialize (Faria-e-Castro (2021); Borsuk, Budnik & Volk (2020)).

This study considers Basel III’s definition of the leverage ratio as an indicator for bank risk.

Unlike the term “leverage”, a higher outcome implies stronger bank resilience. As such, the following hypothesis is formulated:

H1: In countries where the countercyclical capital buffer has been implemented, a change in the effective CCyB rate positively impacts the growth of the leverage ratio and dampens its cyclical component.


19 4.2 Bank risk indicator: Loan growth

One procyclical element of the economy to which financial intermediaries contribute is bank lending (Adrian & Shin, 2010). When banks expand collateralized lending, the newly released funding is used to buy up assets, which in turn drives up asset prices. When the underlying collateral becomes regarded as a more popular asset, the value of the collateral rises and banks become more willing to increase credit. When the financial cycle turns, capital constrained banks that are unable or unwilling to raise new capital are forced to reduce their bank lending, amplifying the downturn (Haselmann & Watchel, 2016). In relation to this reasoning, it can be tested whether macroprudential reduces lending whenever policy is tightened, or increases lending when policy is loosened.

This relationship has been investigated in earlier research papers, which mostly rely either on a proxy for the CCyB or on a counterfactual simulation (Drehman & Gambacorta, 2012; Akram, 2014; Tayler & Zilberman, 2016, Chen, Sivec & Volk, 2018). These studies find a significantly negative effect between implementing a time varying buffer and bank lending.

However, the cyclical impact of this macroprudential tool is not incorporated in their frameworks. An interaction term between the cycle and macroprudential policy could have be added in order to measure this effect. The following hypothesis summarises this reasoning:

H2: In countries where the countercyclical capital buffer has been implemented, a change in the effective CCyB rate negatively impacts bank lending growth and dampens its cyclical component.

4.3 Effects of the CCyB by phase of the cycle

As the BCBS defines the primary objective of the CCyB as increasing the resilience of banks, it explicitly places its importance on taming the downward phase of the cycle. However, it is not clear whether this policy instrument is actually more effective in absorbing bank risk and sustaining bank lending after the bust, or mitigating increases in bank risk during the build- up of financial imbalances. In addition, different policy objectives across countries requires understanding under which circumstances it is most effective in taming bank risk. The fact that Sweden’s Finansispektionen only mentions the objective of building bank resilience, while Spain’s Banco de España attaches highest importance to taming excessive risk, may indicate


20 that internationally different expectations on policy outcomes exist.

Claessens, et al., (2013) find that a broad set of binding macroprudential policies are much more effective in booms than during bust phases. More specifically, they find that during booms policies like limits on foreign lending, dynamic provisioning, limits on profit distribution, DTI and LTV caps are more effective in containing the growth of leverage and assets. Country level research by Schryder and Opitz (2021) studies the impact of macroprudential policies on credit-to-GDP ratios. They find that their effects are stronger during credit cycle upturns than in downturns. Even though no study has yet differentiated for the financial cycle when determining the CCyB’s effect in particular, these studies can be used as reference point for macroprudential policy. As such, the following hypothesis is formulated:

H3: The CCyB is more effective in reducing bank risk during the expansionary phase of the financial cycle, than during the contractionary phase.




5.1 Model specification

The empirical model used in this research paper helps to analyse whether macroprudential policies have an effect on the change in bank risk variables. As the build-up of systemic risk in the banking sector is often a reason to implement macroprudential policies in the first place, there could exist endogeneity bias issues when performing regular OLS regressions (Claessens, 2014). To overcome problems of endogeneity in panel regressions, a two-step system GMM model is applied. In case there is unobserved correlation between explanatory variables and the error term, this econometric method can apply appropriate weights to lagged instrumented variables, so that the asymptotic variance of the estimator is minimized. That is why lagged dependent variables are added as instruments to the level equation (Bond et al., 2001) and are their validity is assessed with a second order autocorrelation test. If the null hypothesis of no autocorrelation is not rejected, it can be assumed the GMM estimator is consistent. Furthermore, a Hansen J-test is performed to assess the validity of the instruments. If the null hypothesis is not rejected, the assumption that overidentifying instruments are not endogenous holds. However, according to Roodman (2009), when the p-value for Hansen J-test falls outside the 0.1 – 0.25 boundary, it signals potential signs of endogeneity as well.

The original empirical model that studies the impact of macroprudential policy on bank risk is designed by Claessens et al., (2013) and the specification used in this research paper is designed as follows:

∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡 =𝛼 + 𝜆 ∗∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡−1+ 𝜇 ∗ 𝐿𝐸𝑉𝐸𝑅𝐴𝐺𝐸𝑖,𝑐,𝑡−1+𝜂 ∗ 𝐷𝐸𝑃𝑂𝑆𝐼𝑇𝑖,𝑐,𝑡−1

+ 𝛽 ∗𝑀𝑎𝑃𝑃𝑐,𝑡+𝛿 ∗𝑀𝑎𝑃𝑃𝑐,𝑡∗ ∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡−1+ 𝜀𝑖,𝑐,𝑡(1)

In model specification (1), the variable ∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡 features leverage ratio growth and loan growth as its two main risk variables for each bank i in country c in year t. Instead, the original model by Claessens et al., (2013) uses leverage growth, assets growth and non-core liabilities growth as dependent bank risk indicators. Loan growth is preferred over asset growth in this research paper, as this higher risk-weighted asset class can give a more precise


22 indication of the effect of macroprudential policies on bank risk behaviour. Non-core liabilities growth is excluded from the regressions because the literature does not provide an explanation for a relationship between this indicator and the macroprudential policies of interest. In terms of scope, the original model features a wider range of macroprudential policy tools and incorporates banks from both rich and poorer countries.

In order to allow for natural convergence, each regression includes a lagged risk variable, denoted by ∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡−1. It is instrumented by lagged observations of the dependent variable, the BIS credit-to-GDP gap and both country and time fixed effects.

Individual bank conditions are controlled for by a lagged leverage ratio and customer deposit ratio through 𝐿𝐸𝑉𝐸𝑅𝐴𝐺𝐸𝑖,𝑐,𝑡−1 and 𝐷𝐸𝑃𝑂𝑆𝐼𝑇𝑖,𝑐,𝑡−1, respectively. After all, banks that are highly leveraged may have less room to decrease the proportion of its capital in relation to total exposures, and banks with a high fraction of market funding are more cautious to increase their risk profile. The matrix 𝑀𝑎𝑃𝑃𝑐,𝑡 represents four macroprudential policies that can have an effect on the change in bank risk variables. First, the countercyclical capital buffer variable is recorded as a year-on-year change in the effective rate at the end of the year. If countries have not adopted a Basel III CCyB framework or have not gone further than only announcing future rates, then the variable is assigned a value of zero. Dummy variables for dynamic provisioning, loan-to-value and debt-to-income caps are included as well. Only if domestic regulators have put such policies into effect, variables are assigned 1 and zero otherwise. At first, four regressions are performed for each dependent variable to test whether individual macroprudential policies have an effect on the change in the bank risk variable. Additionally, four regressions are performed in which the interaction variable 𝑀𝑎𝑃𝑃𝑐,𝑡 ∗ ∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡−1 is included. These interactions give an estimation to what extent the effect of macroprudential policies vary according to the intensity of the cycle. If coefficients for this variable turn out to be significant, this indicates that policies are either countercyclical or pro-cyclical, depending on the dep risk variable and the sign of the coefficient. In other words, then policies are more effective when the financial cycle is more intense.


23 5.2 Robustness specifications

Three alternative model specifications are added as robustness checks for the baseline regressions, similar to the research by Brei & Gambacorta (2014). To avoid endogeneity issues, most robustness variables are lagged one period. While the regressions control for country fixed effects in all regressions through the instrumented variable, including these variables ensures that differences between countries and their impact on the change in bank risk can be directly investigated. The first robustness specification includes cyclical variables that can move either in the same or opposite directions to changes in bank risk. The latter two robustness regressions measure country specific characteristics of a financial industry, which are expected to be more static by nature. For this reason, the interactions between the lagged dependent variable and the countercyclical capital buffer are only included in the first robustness regression.

5.2.1 Adding macroeconomic variables

∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡 = α + λ ∗∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡−1+𝜇 ∗𝐿𝐸𝑉𝐸𝑅𝐴𝐺𝐸𝑖,𝑐,𝑡−1

+𝜂∗𝐷𝐸𝑃𝑂𝑆𝐼𝑇𝑖,𝑐,𝑡−1+𝛽 ∗𝑀𝑎𝑃𝑃𝑐,𝑡+𝛿 ∗ 𝑀𝑎𝑃𝑃𝑐,𝑡∗ ∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡−1

+ 𝜁 ∗∆% 𝑅𝐸𝐴𝐿𝐺𝐷𝑃𝑐,𝑡−1+𝜎 ∗∆% 𝐼𝑅𝑐,𝑡−1 + 𝛾 ∗ 𝐹𝑋𝑆𝑌𝑆𝑇𝐸𝑀𝑐,𝑡−1+ 𝜀𝑖,𝑐,𝑡 (2)

In model specification (2), the state of the real business cycle is proxied by the growth in real GDP and lagged by one period with ∆% REALGDP𝑐,𝑡−1. If a country’s real GDP growth is high, banks are more likely to expand their lending operations and increase their leverage. Including a lagged variable for the growth of the central bank policy rate, denoted by ∆% IR𝑐,𝑡−1, allows the model to control for monetary policy. As far as asset prices react to changes in the interest rate, this variable can be expected to be correlated with the financial cycle. Controlling for a country’s foreign exchange arrangement in the previous period, expressed by FXSYSTEM𝑐,𝑡−1 ,is important for two reasons (Brei & Gambacorta, 2014). It determines to what extent central banks are free to conduct monetary policy to stabilize the economy. Also, when comparing bank balance sheets in different currency zones that are denominated in US dollars, fluctuations in the dollar to local exchange rate can be absorbed by this variable for flexible exchange rate regimes.


24 5.2.2 Adding government debt and banking crisis variables

∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡 = α + λ ∗∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡−1+𝜇 ∗𝐿𝐸𝑉𝐸𝑅𝐴𝐺𝐸𝑖,𝑐,𝑡−1 + 𝜂∗ 𝐷𝐸𝑃𝑂𝑆𝐼𝑇𝑖,𝑐,𝑡−1

+ 𝛽 ∗𝑀𝑎𝑃𝑃𝑐,𝑡+ +π ∗ GOVDEBT𝑐,𝑡−1 +ρ ∗ CRISIS𝑐,𝑡−1+ 𝜀𝑖,𝑐,𝑡 (3)

In model specification (3),the variable GOVDEBT𝑐,𝑡−1 stands for the lagged public debt-to- GDP ratio. It could potentially measure a governments ability to perform countercyclical fiscal policies, which in turn affect banks’ balance sheets. Also, a lagged financial crisis variable is included with CRISIS𝑐,𝑡−1. A systemic banking crisis is expected to influence the dynamics between macroprudential policies and bank risk variables. If there exist disparities between countries regarding the impact of the GFC or the Covid-19 crisis on the banking system, and subsequently a government’s fiscal response, these variables may control for them.

5.2.3 Adding financial industry variables

∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡 =α + λ ∗∆%𝐵𝐴𝑁𝐾𝑅𝐼𝑆𝐾𝑖,𝑐,𝑡−1+𝜇 ∗𝐿𝐸𝑉𝐸𝑅𝐴𝐺𝐸𝑖,𝑐,𝑡−1 + 𝜂 ∗ 𝐷𝐸𝑃𝑂𝑆𝐼𝑇𝑖,𝑐,𝑡−1

+𝛽 ∗𝑀𝑎𝑃𝑃𝑐,𝑡+κ ∗ STRUCTURE𝑐,𝑡−1+ψ ∗ 𝐿𝑂𝐴𝑁𝑆𝐹𝑂𝑅𝐸𝐼𝐺𝑁𝑐,𝑡−1+ ε𝑖,𝑐,𝑡(4)

In model specification (4), the variable STRUCTURE𝑐,𝑡−1 measures the financial structure of a country in the previous period. As macroprudential policy is mainly targeted towards the banking system, its effectiveness depends in large part on the structure of the financial system. For example, if a country’s shadow banking system or public bond market has a relatively large footprint on the economy, it may be that financial institution-based policies are of lesser importance to regulators. When other policies outside of the empirical model are implemented instead, including a financial structure variable may control for this omitted variable bias. The variable 𝐿𝑂𝐴𝑁𝑆𝐹𝑂𝑅𝐸𝐼𝐺𝑁𝑐,𝑡−1 measures the lagged share of total loans from a country’s financial sector that is directed towards foreign residents. If in a subgroup of countries, predominantly large banking groups with high overseas exposures are located, this variable allows the model to control for jurisdictional limits of macroprudential policies.




A panel dataset is constructed in order to test whether macroprudential policies affect vulnerabilities within the banking sector. The first bank risk indicator is measured by the annual percentage change in Basel III’s definition of the leverage ratio, that is Tier 1 capital divided by the sum of total assets and off-balance sheet exposures. The second bank risk indicator, the growth of loans, is simply the annual percentual change in a banks’ net outstanding loans. This measure is unlike annual gross lending, as it also considers loan repayments and write-offs. However, this growth rate gives an accurate description of a bank’s lending activity during a given year. Leverage ratio growth and loan growth are fundamentally different variables, as the former may have an expected value of zero in the steady state, while the latter can be expected to grow in tandem with the economy in the long run. For all banks in the data sample, average leverage ratio growth and loan growth are 2,4%

and 4,9%, respectively (Table 1). The positive average leverage ratio growth supports earlier empirical evidence that banks have been deleveraging and de-risking after the fallout of the GFC (Caseli et al., 2016).

Annual data from bank balance sheets are accessed through Bankscope, which is a commercial database managed by International Bank Credit Analysis Ltd and Bureau van Dijk.

Rather than studying cross-country level effects, using bank level data ensures there is less risk from endogeneity when studying the effects of macroprudential policies (Claessens et al., 2013). Unless there exists a very high degree of industry concentration, it is less likely that individual banks risk variables are used over aggregate variables as input for macroprudential policy. The GMM setup should help to remove the residual endogeneity that still exists between bank level data and macroprudential policies.

To ensure broad coverage of a country’s banking industry, banks are selected by descending order of size to cover at least 70% of domestic bank assets. As such, the sample comprises 200 banks that together account for 2410 observations. While Claassen et al., (2013) use unconsolidated banking statistics and end up with 2820 banks in 48 countries, there is one key advantage of using consolidated banking statements. That is, decisions on capital adequacy are mainly taken by management at the consolidated level (Brei & Gambacorta, 2014). However, for some macroprudential policies, like the CCyB or LTV cap, compliance only applies to banking activities within a particular jurisdiction. This study primarily considers



Table 1

Summary of all regression variables

Variable Obs. Mean Std. Dev. Min Median Max

Banks 200 (unique)

Countries 28 (unique)

Years 2006-2020

Bank level

Leverage growth YoY (%) 2219 2.36 11.41 -18.20 1.04 28.30

Loan growth YoY (%) 2255 4.86 12.19 -14.32 3.26 33.21

Assets (USD $B) 2392 385.72 568.94 0.39 137.49 3096.33

Leverage ratio (%) 2392 5.20 2.49 -0.14 4.71 23.40

Deposit ratio 2390 55.32 17.58 0.34 57.92 92.29

Country level

Credit-to-GDP gap (%) 2232 -2.37 19.38 -100.5 0.1 82

Real GDP growth (%) 2392 1.38 3.02 -14.84 1.95 25.30

Exchange rate classification 2240 6.13 1.94 1 7 7

LIR growth YoY (%) 2262 -2.30 59.66 -100 0 900

Public debt to GDP (%) 2392 72.29 49.46 0.01 68.83 234.86

Systemic banking crisis 2392 0.12 0.33 0 0 1

Foreign loans (%) 1953 25.16 16.95 1.12 24.36 79.33

Bond market to GDP (%) 1033 51.69 42.64 0.01 39.62 192.76

CCyB difference YoY 2392 0.02 0.31 -2.5 0 1.5

LTV 1844 0.40 0.49 0 0 1

DP 1844 0.09 0.28 0 0 1

DTI 1844 0.02 0.41 0 0 1

Notes: The credit-to-GDP gap is retrieved from the BIS statistics website. Data on monetary policy can be found at the International Financial Statistics database from the IMF. The classification on the exchange rate regime can be found in at the IMF’s AREAER database, and its classification ranges from 1 to 7, where low rankings indicate such a regime is more fixed, while higher rankings suggest it is more flexible. Real GDP growth statistics are derived from April 2021 World Economic Outlook by the IMF. Data on the share of loans to foreign residents and the measure of financial structure, captured by the capitalization of the public bond market as a share of GDP, is retrieved from the World Bank’s Financial Structure Database. The data for systemic banking crisis dummies and central government debt-to-GDP is sourced from the World Bank’s Global Financial Development and World Development Indicators databases, respectively.



Table 2

Summary of bank variables statistics per country Country Banks Obs. Leverage


Leverage growth

Loan growth

Leverage growth

Loan growth

Leverage growth

Loan growth 2006-2020 Expansion phase Contraction phase

AT 5 55 5,4 1,4 2,9 3,0 7,7 0,7 1

AU 9 114 4,7 2,6 5,4 3,5 10,2 2,2 3,6

BE 6 73 3,6 4,7 3,7 7,0 5,4 2,1 1,9

CA 5 50 3.6 0,2 3.8 0,2 3.8

CZ 4 44 5,9 3,8 6,4 4,2 6,6 -0,3 10,2

DE 10 135 3,4 4,6 1,3 0,5 2,3 5,7 0,9

DK 7 98 6,9 1,5 3,7 1,0 6,2 1,4 2,2

EE 3 28 10,5 2,2 8,9

ES 13 142 5,0 1,2 2,1 5,9 8,3 0,4 1,2

FI 3 32 5,4 -0,6 9,2 1,7 11,6 -4,3 7,5

FR 12 166 3,2 2,4 6,0 2,4 6,0

GB 14 172 3,7 1,0 2,1 1,1 7,5 0,9 0,9

HK 14 200 5,6 2,9 10,1 2,9 10,1 -5,2 3,8

HU 5 42 6,3 8,7 -0,35 8,1 0,7

IE 3 33 6,2 1,4 -1,9 1,8 -0,2 3,1 -3,1

IS 4 46 14,4 3,0 5,6

IT 8 87 4,7 2,0 7,1 1,74 11,1 2,6 4,7

JP 11 118 4,6 1,6 3,6 1,4 1,4 0,9 12,2

KR 8 76 5,2 0,1 6,9 -1,02 5,8 4,6 11,3

LT 2 30 8,6 4,1 9,1

LU 3 38 5,1 0,8 4,5 1,4 3,2 0,6 4,8

NL 4 51 4,2 3,8 0,4 4,1 -4,0 3,5 3,3

NO 8 104 7,1 2,7 6,0 4,1 7,8 0,9 3,2

NZ 4 41 5,7 1,4 4,3 0,8 10,8 2,5 3,6

PT 6 68 4,1 4,4 -1,7 8,7 2,7 2,2 -3,7

SE 5 70 3,6 2,8 5,2 2,0 10,6 3,3 0,8

SK 4 39 6,5 1,3 7,3

US 20 213 5,7 1,2 7,1 4,6 7,1 0,0 7,1

Notes: For the Basel III credit-to-GDP gap, there is no data available for Estonia, Iceland, Lithuania and Slovakia. The consequences are manageable, as these countries do not account for a large bulk of the observations in the sample. Canada and France have not experienced a negative credit-to-GDP gap between 2006 and 2020, while Hungary has not experienced a contractionary phase of this indicator.


28 consolidated banking statements. If the data is available however, a local banking subsidiary is admitted instead of its larger parent holding company, so that spillover effects from international exposures are limited.

To make sure the dataset is mostly confined to universal banks or commercial banks, a sample criterion is the share of customer deposits to total assets. Only about 5% of bank observations have a customer deposit ratio lower than 25%. Balance sheet data is recorded in US dollars. Large foreign exchange fluctuations vis-à-vis the dollar could potentially bias the results, however time and country fixed effects could control for this issue. To prevent estimation bias by having outliers in the dataset, the dependent variables are winsorized at the five percent level at both sides of the distribution. The sample covers the time period from 2006 to 2020, a period covering high volatility in financial markets and for some countries several economic cycles. According to Brei and Gambacorta (2014), quarterly data could potentially provide better insights into the financial cycle. However, they refer to other studies that have compared frequencies in rich bank level datasets and have not found a significant difference between them.

The sample only considers banks in developed economies. After all, Basel III’s countercyclical capital buffer has mostly been implemented in rich countries. Including poorer countries as a counterfactual would impose strong omitted variable bias. For example, these countries dedicate a larger share of their policy effort towards capital account related risk, while advanced economies target indebted households through borrower-based instruments more often (Cerutti, Claessens & Leaven, 2017). The distribution of banks between countries depends on the size of the economy and the availability of banks that meet the sample criteria.

The United States is the country that accounts for most banks and Lithuania for the least, at 20 and 2 respectively (Appendix A). Considering the leverage ratio, most countries are within the range of 3% to 6%. Estonia and Iceland have average leverage ratios higher than 10%, but together only represent 7 banks in the entire sample. The average leverage ratio growth rate is highest in Hungary and Belgium, while Finland records negative means. The average loan highest in Hong Kong and Finland, while Ireland, Portugal and Hungary report negative means.

Finland ranking among the highest in loan growth and lowest in leverage ratio growth, again illustrates that both risk variables are different in nature. Furthermore, it can be noted that the distribution of bank variables is not biased in one way or another, so that we have a balanced set of countries that share similar characteristics.




Related subjects :