Adjustments in capital buffers and bank risk taking: The
European setting
Arlette Brouns1 Thesis MSc. Finance Supervisor: Dr. P.P.M. Smid
Thesis Internship Deloitte Supervisor: M.M.A. Kersten
University of Groningen January 2016
1 University of Groningen
Faculty of Economics & Business Student number: s2013363
2
Adjustments in capital buffers and bank risk taking: The
European setting
Abstract
This study investigates the relationship between adjustments in capital buffers and bank risk over the time period 2008-2014. Building on an unbalanced dataset of all significant banks in Europe, results show this relationship is positive. Interestingly, results indicate that well-capitalised banks increase capital buffers and decrease risk taking more than other banks. However, undercapitalised banks show to have a higher speed of adjustment towards target levels of capital buffers. The current Supervisory framework of Basel II has been effective to restore stability in the financial sector since the credit crunch in 2008. Nevertheless, undercapitalised banks fall behind compared to their well-capitalised peers.
3
1. Introduction
The recent banking crisis raises the question if the current financial system is safe enough and stresses the importance of bank supervision, regulatory capital and control mechanism to monitor bank risk taking. In response to this banking crisis, the first Basel Accord is revised to improve the safety of the financial system. Whereas Basel I already includes regulatory capital requirements, Basel II relates these capital requirements to underlying risks. In recent years, the second pillar of Basel II, supervisory review, receives increasing attention. Supervisory measures became stricter and supervisory control shifted towards the Single Supervisory Mechanism (henceforth: SSM)2. The SSM consists of both the European Central Banks (henceforth: ECB) and national authorities, who directly supervise banks in Europe. Together, they ensure the safety of the financial system through constant monitoring of bank capital levels and risk taking (ECB, 2014). Peura & Keppo (2006) argue that the risk-based capital requirements and supervisory framework of Basel II cause banks to hold additional levels of capital, so called capital buffers. Capital buffers in this particular context can be defined as the amount of capital in excess of the required regulatory minimum. Banks hold this buffer to prevent regulatory penalty for falling below the minimum, to keep track of risky assets and to be able to recapitalise (Peura & Keppo, 2006).
The aim of this paper is to examine the relationship between adjustments in capital buffers and bank risk taking for all significant banks supervised by the ECB. Moreover, variables that influence target capital buffers and risk taking levels are examined. Lastly, a distinction is made between well-capitalised and undercapitalised to examine if differences exist in terms of capital and risk taking adjustments, but also in adjustment speed towards both target levels. The research questions are as follows: (i) What is the relationship between adjustments in capital buffers and bank risk taking? (ii) Does any difference exists between well-capitalised and undercapitalised banks with regard to both adjustments? (iii) Is the speed of adjustment towards target capital buffers and bank risk taking dependent on the degree of bank capitalisation?
An unbalanced dataset of balance sheet data over the years 2008-2014 is constructed to examine the research questions. For the analyses, a simultaneous equation model is adopted and examined using a two-stage least squares (henceforth: 2SLS). Results suggest that the relationship between capital buffers and risk taking adjustments is positive. Moreover, well-capitalised banks increase capital buffers and decrease risk taking more than other banks.
4 Lastly, undercapitalised banks show to have higher speed of adjustments towards target capital buffers.
This paper contributes to the growing area of research on capital buffers, risk taking and capitalisation by simultaneously assessing these variables for supervised banks in Europe. So far, the empirical literature generally focusses on individual countries, in particular the United States3. Also, most of the previous work has a focus on time periods prior to Basel II4. Examining banks in Europe during the Basel II period is of interest since results might be different in a distinct regulatory environment and time period. Furthermore, the results of this paper provide new insights into the field of supervisory review as part of Pillar II. Different results due to the degree of bank capitalisation require regulators to assess banks differently. Also, Lindquist (2004) suggests it is important to examine if capital buffers fluctuate with bank risk taking for a longer time period to improve regulatory standards. Moreover, supervisory review remains a key element in Basel III. The newest Basel Accord, reform of bank capital regulation, evolves in response to deficiencies in capital regulation and risk taking balances and will be fully implemented by 2019 (Basel Committee, 2010).
The remainder of this paper is structured as follows. Section 2 gives a brief overview of the theoretical as well as empirical literature on capital buffers and bank risk taking. Additionally, hypotheses are developed. Section 3 describes the empirical methodology followed by a data and variable description in section 4. Results are presented in section 5 whereas section 6 provides robustness checks. Section 7 draws conclusions.
2. Capital buffers and bank risk taking: review of the theory
The purpose of this section is twofold. First, the regulatory background of the European Union is outlined. Second, relevant theoretical models as well as empirical results on capital buffers and bank risk taking are reviewed that result in hypotheses.
2.1 Regulatory background for the European Union
Basel II came into force in 2008 and, consists of three pillars; (i) minimum capital requirements (ii) supervisory review and (iii) market discipline. For the first pillar (i), banks are required to hold a minimum amount of regulatory capital that is equal to at least 8% of the
3 Aggarwal and Jacques, 2001; Jokipii and Milne, 2011; Shim, 2012.
4 Schrieves and Dahl, 1992; Jacques and Nigro, 1997; Aggarwal and Jacques, 2001; Rime, 2001; Jokipii and
5 risk weighted assets (henceforth: RWAs). The RWAs assign different risk weights to all assets and therefore capture risks involved in balance sheet items. This minimum amount of regulatory capital can be seen as an absolute minimum and is therefore constantly monitored by supervisory authorities. The third pillar (iii) of Basel II, market discipline, consists of disclosure requirements. These requirements oblige banks to provide key information about the total risk exposure together with matching controls. Because this paper focusses on the implications of pillar II, this pillar is more thoroughly discussed.
Pillar II (ii) can be seen as a control mechanism for capital under the first pillar and is called supervisory review. The purpose is to improve the relationship between risk profiles, risk mitigation, risk management frameworks and capital for banks (EBA, 2014). Supervisory measures as part of pillar II examine the financial condition of banks through comprehensive assessments. Comprehensive assessments aim to make sure that capital levels are adequate and that banks can survive potential economic downturns (ECB, 2015). As explained by the ECB, comprehensive assessments are held on a yearly or ad hoc basis. The first denotes that all significant banks in the Eurozone are evaluated yearly. The latter permits the ECB and national authorities to assess banks during exceptional market situations, in addition to the yearly assessments. Moreover, the sample of banks for the comprehensive assessments can always be expanded if considered to be necessary5. The comprehensive assessments include two main components; the European wide stress test (henceforth: EU wide stress test) and the asset quality review (henceforth: AQR). Supervising authorities conduct the EU wide stress test to examine the financial stability of banks in the Eurozone under different scenarios. The main goal of the AQR is to provide transparency about risk exposures and the quality of assets in the banking book. Another important part of pillar II is the supervisory review and evaluation process (henceforth: SREP). This supervisory process uses the outcomes of the comprehensive assessments to not only regulate bank risk taking, but also regulate capital and liquidity positions, the business model and the governance framework. Taken together, the SREP is defined as follows (EBA) (2014, p. 4): “the key purpose of SREP is to ensure that financial institutions possess adequate strategies, mechanisms, arrangements and processes together with sufficient liquidity and capital to ensure a sound management and coverage of their risks”.
5 For example; DNB Group in Norway, and four banks in the United Kingdom were obliged to participate in the
6 Although supervisory measures of pillar II remain consistent after the foundation of Basel II, supervisory responsibility has largely shifted towards the SMM. The importance of a consistent supervisory framework is put forward by the recent banking crisis. Hence, the SSM is brought into existence to safeguard the financial health of all banks in the Eurozone. Also, it increases financial integration and ensures stability throughout consistent supervisory practices (EBA, 2015). Collectively, the SSM conducts the comprehensive assessments and the SREP in cooperation with the European Banking Authority (henceforth: EBA). Whereas non-Euro countries can choose to participate in the SSM, countries that adopted the Euro automatically fall under its direct supervision6. Different types of banks are supervised including (mixed) financial holding companies, credit institutions and credit institution branches that are established in non-participating member states (EBA, 2015). In total, the ECB supervises 123 significant banks that hold 82% of the total banking assets in Europe (ECB, 2014). The national authorities supervise the non-significant banks. The SSM has established four criteria that determine if banks can be classified as significant or non-significant. At least one of these criteria should be met in order to be classified as significant. Criteria for significance contain the size of a bank, economic importance to the country, the amount of cross-border activities and obtained financial support as shown in appendix II (ECB, 2015).
2.2 Theoretical models on capital buffers and bank risk taking
In the literature, different theories and empirical evidence exist regarding the relationship between bank capital and risk taking. Collectively, three different theories emerged; the theory of moral hazard, the charter value theory and the capital buffer theory.
The moral hazard theory explains the relationship between capital and bank risk taking from a perspective of shareholder value maximization, causing banks to engage in excessive risk taking while lowering capital levels (Merton, 1977). However, the Basel Accords and in specific the capital adequacy regulation, serves as a constraint for shareholder maximization. It requires banks to hold capital levels in line with risk taking, shifting part of the downside risk back towards shareholders. Logically, it can be expected that supervisory review as part of pillar II causes the shift in downside risk towards shareholders to be even larger. Supervisory measures such as the comprehensive assessments and SREP, serve as a constant monitoring
6 As previously mentioned, the SSM is, in consultation with the ECB, always permitted to examine banks outside
7 mechanism that supervises bank risk taking. The current risk sensitive capital framework may thus daunt a bank to choose risky assets or risky loan portfolios.
A second, more evolved theory, is the charter value or franchise value theory. Charter value can be defined as the present value of future profits (Demsetz, Saidenberg and Strahan, 1996). The theory predicts that banks are less willing to take risk in order to protect their charter values (Keeley, 1990; Park and Peristiani, 2007). If charter values are high, banks attempt to protect these values through better diversified loan portfolios and less risky assets choice (Demsetz et al., 1996). Moreover, the charter value theory allows for the fact that banks could hold capital buffers on top of the minimum threshold. It predicts banks to either increase the amount of buffer capital or maintain current capital levels, to protect high charter values (Park and Peristiani, 2007). The opposite is true for lower charter values. In addition, the potential loss in charter value if banks decide to engage in more risky behavior, can be seen as an indirect form of bankruptcy cost (Demsetz et al., 1997). Therefore, if the charter value is high, incentives to avoid liquidation are also expected to be high. These potentially incurred bankruptcy costs cause banks to either increase capital levels or decrease risk taking (Jokippi and Milne, 2011). Thus, the charter value theory predicts that the relationship between capital and bank risk taking is likely negative.
Building on the charter value theory, the capital buffer theory receives increasing attention. As often shown by total capital ratios, banks hold average capital levels well above the required minimum. The buffer theory therefore separates regulatory capital levels from target capital levels of banks. Jokipii and Milne (2011) suggest that the trade-off between capital holding costs and the loss in charter value, expressed as costs of failure, decides how big this capital buffer is. There is a large volume of theories describing why banks hold capital buffers7. From a pillar II perspective, incentives to hold a capital buffer are driven by regulatory pressure. Marcus (1984) and Furfine (2001) argue that banks hold capital buffers in order to prevent costly regulatory interferences when falling below the minimum threshold. This argument is driven by two assumptions. The first assumption explains that banks may not be able to adjust capital and risk taking levels instantaneously. The second relies on losses in charter value. If charter values decrease, regulatory interferences are more likely since lower charter values imply higher bankruptcy costs. Peura and Keppo (2006) argue that regulatory interferences are costly to banks. Additionally, they state that altering the level of capitalisation
7 Jackson, Furfine, Groeneveld, Hancok, Jones, Perraudin, Radecki and Yonenyama, 1999; Peura and Keppo,
8 can be a costly process. The buffer theory takes both costs into account by allowing to examine the speed of adjustment (i.e adjustment costs) towards target capital levels (Jokipii and Milne, 2011). These adjustments are of interest since banks face two choices when falling below the minimum threshold; they can either increase capital buffers or decrease risk taking. Collectively, the capital buffer theory is able to make a distinction between the long and short term relationship between capital levels and bank risk taking (Jokipii and Milne, 2011). The long term relationship is likely to result the same as the one predicted by the charter value. The short term relationship allows for different degrees of bank capitalisation. The buffer theory expects this relationship to be positive for banks holding high levels of capital buffers (i.e. capital levels close to their targets) and negative for banks with capital levels close to the required minimum (Jokipii and Milne, 2008).
2.2 Empirical model on capital buffers and bank risk taking and hypotheses development A number of empirical papers examine the relationship between bank capital and risk taking adjustments based on theories as outlined in section 2.18. The largest part of these studies focus on time periods prior to Basel II9. Also, most of these papers use total capital levels instead of capital buffers10. Therefore, these studies often do not separate regulatory capital requirements from banks’ own target capital levels. Still, these studies are good indicators for the possible relationship between adjustments in capital buffers and bank risk taking.
Schrieves and Dahl (1992), show that adjustments in capital levels and risk taking are simultaneously related for commercial banks, carrying a positive sign. The study is conducted for the time period 1984 till 1986 and includes US banks having a FDIC11 deposit insurance. Additionally, they show that this relationship is positive for both high and low capitalised banks. As suggested by Schrieves and Dahl (1992), this result provides evidence that the positive relationship cannot only be explained by regulatory pressure but also by other factors. They suggest that private incentives of managers or managerial risk aversion could be factors. Managerial risk aversion causes managers to increase capital levels when bank risk increases (Saunders and Travlos, 1990). Similarly, Aggarwal and Jacques (2001) find a positive relationship when allowing for adjustments in capital levels and bank risk taking due to the
8 See amongst others; Schrieves and Dahl, 1992; Jacques and Nigro, 1997; Aggarwal and Jacques, 2001; Rime,
2001; Heid, Porath and Stolz, 2003; Jokipii and Milne, 2011; Shim 2012.
9 Schrieves and Dahl, 1992; Jacques and Nigro, 1997; Aggarwal and Jacques, 2001; Rime, 2001; Jokipii and
Milne, 2011; Shim, 2012.
10 Schrieves and Dahl, 1992; Jacques and Nigro, 1997; Aggarwal and Jacques, 2001; Rime, 2001.
11 The Federal Deposit Insurance Corporation is an independent organisation providing deposit insurance in the
9 impact of the prompt corrective action standards (Henceforth: PCA standards)12. For a set of US banks, they find both well-capitalised and undercapitalised banks to increase capital ratios while increasing portfolio risk during 1993-1996. Furthermore, Rime (2001) shows that regulatory capital standards causes banks to increase capital levels far above minimum requirements in the Swiss banking sector during 1989-1995. No such support was found for bank risk taking, indicating that regulatory capital standards did not influence bank risk taking. By contrast, Jacques and Nigro (1997) show that the relationship between adjustments in capital levels and bank risk taking is negative due to the capital standards as set by Basel I. This results suggests that the Basel Accord causes an increase in capital levels while portfolio risk decreases. They only find a significant result for well-capitalised banks. A more recent study by Jokipii and Milne (2011) that covers a longer time period, shows that the relationship between adjustments in capital buffers and bank risk taking is generally positive. The study is conducted for US commercial banks between 1986 and 2008. Furthermore, they allow for different degrees of bank capitalisation that results in a positive relationship when banks are well-capitalised and a negative relationship otherwise. Also Heid, Porath and Stolz (2003) found evidence for this relationship examining German banks from 1994 until 2002. They show that the ability of banks to adjust risk taking or capital buffers depends on the size of the capital buffer. This size can be expressed as the degree of capitalisation. When banks hold a relatively high capital buffer (i.e. well-capitalised banks), this buffer is restored by increasing risk taking as capital increases. Alternatively, when banks hold a relatively low capital buffer (i.e. undercapitalised banks), banks restore their capital buffers by decreasing risk taking behavior while increasing capital levels. In such case, capital buffers are close to the minimum regulatory requirement. Overall, previous empirical literature focusses on time periods prior to Basel II. Alaa, Van Son, Issouf and Fulbert (2013) suggest that the relationship between adjustments in capital buffers and bank risk taking shows a stronger positive relationship during the Basel II period. Their paper examines the impact of regulatory changes on capital levels, risk taking and performance of Canadian banks over the time period 1982-2010. The results of this study show a stronger positive relationship in the two years after the implementation of Basel II than before.
Different empirical results exist with regard to adjustments in capital buffers and bank risk taking. Overall, there seems to be more evidence that the relationship is positive. Also,
12 The PCA standards are an US law that ensure stricter regulation in terms of corrective action for banks that
10 previous papers suggest that this relationship is positive for well-capitalised banks and negative for undercapitalised banks. In line with the most recent findings, stricter supervisory regulation as part of pillar II, and characteristics of banks in this paper13, the following is hypothesised:
H10: There is no relationship between adjustments in capital buffers and bank risk taking.
H11: There is a positive relationship between adjustments in capital buffers and bank risk
taking.
Next to examining adjustments in capital buffers and bank risk taking, it is interesting to see if differences exist between well-capitalised and undercapitalised banks. By allowing for different degrees of capitalisation, the impact of the regulatory time period can be explored. Most of the empirical papers that focus on time periods prior to Basel II find evidence that undercapitalised banks increase capital levels more than their well-capitalised peers14. This result could be explained by the buffer theory that predicts low buffer banks, tending to fall below the minimum capital threshold, try to comply with regulatory standards. However, Jacques and Nigro (1997) show contradicting results, suggesting that well-capitalised banks increase capital levels more than other banks. They suggest that this finding can be attributed to banks preventing themselves against equity shocks from regulatory pressure. Well-capitalised banks are better able to prevent themselves against future equity shocks than their undercapitalised peers and therefore build capital buffers to protect themselves. In such case, well-capitalised banks ensure capital levels rise in order to be able to constantly meet regulatory standards. Regardless if banks currently comply with the minimum threshold.
With regard to differences in risk taking behavior for well-capitalised and undercapitalised banks, empirical papers show mixed results. Schrieves and Dahl (1992) and Heid et all., (2003) show that undercapitalised banks decrease risk taking more than other banks. This results is in line with the buffer theory that predicts undercapitalised banks to decrease risk taking to maintain or rebuild capital buffers. Contrary, Jacques and Nigro (1997) show that well-capitalised banks decrease risk taking more than their undercapitalised peers. They suggest that this finding is the result of constant regulatory pressure. Furthermore, this results could be caused by the fact that undercapitalised banks tend to be less risk averse than
13 Overall, banks in the sample are well-capitalised. Given results of previous empirical papers, this likely results
in a positive relationship between adjustments in capital buffers and bank risk taking. The bank characteristics are outlined in the data description of section 4.
11 their well-capitalised peers (Stolz and Widow, 2009). Also, Stolz and Widow (2009) suggest that poor risk management might be present in undercapitalised banks. The latter two results could explain the fact that undercapitalised banks decrease risk taking not as much as well-capitalised peers when experiencing regulatory pressure. Rime (2001) did not find any evidence of differences between well-capitalised and undercapitalised banks.
Although previous empirical papers show mixed results, the largest part indicate that undercapitalised banks increase capital levels and decrease risk taking more than other banks. This results in the following hypothesis:
H20: There is no difference between well capitalised and undercapitalised banks for
adjustments in capital buffer and risk taking levels.
H21: Undercapitalised banks increase capital buffers and decrease risk taking more than other
banks.
Lastly, various empirical papers show a difference exists in adjustment speed towards target capital buffers for well-capitalised and undercapitalised banks15. These papers show that undercapitalised banks have a higher adjustment speed than well-capitalised banks. This result can be contributed to the fact that undercapitalised banks have a greater incentive to reach target capital levels. Those banks are more in danger of falling below the minimum threshold and face higher regulatory costs. Interestingly, no similar empirical evidence exists for adjustment speed towards target levels of bank risk taking. Therefore, the following is hypothesised:
H30: There is no difference in adjustment speed towards target capital buffers for
well-capitalised and underwell-capitalised banks.
H31: Undercapitalised banks have a higher speed of adjustment towards target capital buffers
than well-capitalised banks.
12
3. Empirical framework and methodology
A review of the theory shows that capital buffers and risk taking decisions are simultaneously related. Hart and Jaffee (1974) and Marcus (1983) state that such a relationship can only empirically be examined if allowing for exogenous- and discretionary changes. By way of illustration, Schrieves and Dahl (1992) argue that regulatory obligation to increase capital levels is exogenous to a bank. For bank risk taking, this exogenous change could be an economic downturn. In order to empirically examine changes in capital buffers and bank risk taking, the simultaneous equations model is adopted in this paper. This model is developed by Schrieves and Dahl (1992) and later extended by Jacques and Nigro (1997), Rime (2001) and Jokipii and Milne (2011). It includes both exogenous factors as a discretionary component:
(1) ∆𝑏𝑢𝑓𝑗𝑡 = ∆𝑏𝑢𝑓𝑗𝑡𝑑𝑖𝑠𝑐𝑟 + ∈jt (2) ∆𝑟𝑖𝑠𝑘𝑗𝑡= ∆𝑟𝑖𝑠𝑘𝑗𝑡𝑑𝑖𝑠𝑐𝑟 + 𝜇𝑗𝑡
Where ∆𝑏𝑢𝑓𝑗𝑡 and ∆𝑟𝑖𝑠𝑘𝑗𝑡 are changes (i.e. adjustments) that can be observed in capital buffers and bank risk taking respectively, for bank j at time t. ∆𝑏𝑢𝑓𝑗𝑡𝑑𝑖𝑠𝑐𝑟 and ∆𝑟𝑖𝑠𝑘𝑗𝑡𝑑𝑖𝑠𝑐𝑟 depict the discretionary changes and 𝜖𝑗𝑡, 𝜇𝑗𝑡 are exogenous random shocks.
The discretionary changes ∆𝑏𝑢𝑓𝑗𝑡𝑑𝑖𝑠𝑐𝑟 and ∆𝑟𝑖𝑠𝑘𝑗𝑡𝑑𝑖𝑠𝑐𝑟 are examined using the partial adjustment model. This dynamic model is used by various empirical studies to examine the relationship between adjustments in capital levels and bank risk taking16. It is especially of interest in this paper because it is able to capture adjustments as well as the speed of adjustments (i.e. adjustment costs) towards target levels of capital buffers and bank risk taking. Furthermore, it acknowledges that banks may not be able to adjust capital buffers and risk taking instantaneously. The discretionary changes are in proportion to differences between existing capital levels at the previous time period t –1 and target capital levels such that: (3) ∆𝑏𝑢𝑓𝑗𝑡𝑑𝑖𝑠𝑐𝑟 = ɛ (𝑏𝑢𝑓𝑗𝑡∗ − 𝑏𝑢𝑓𝑗𝑡−1)
(4) ∆𝑟𝑖𝑠𝑘𝑗𝑡𝑑𝑖𝑠𝑐𝑟 = 𝜑 (𝑟𝑖𝑠𝑘𝑗𝑡∗ − 𝑟𝑖𝑠𝑘𝑗𝑡−1)
16 See amongst others; Schrieves and Dahl, 1992; Jacques and Nigro, 1997; Aggarwal and Jacques, 2001; Rime,
13 Where 𝑏𝑢𝑓𝑗𝑡−1 and 𝑟𝑖𝑠𝑘𝑗𝑡−1 depict the actual levels of buffer capital and risk taking in the
previous time period. 𝑏𝑢𝑓𝑗𝑡∗ and 𝑟𝑖𝑠𝑘𝑗𝑡∗ are the target levels of capital buffers and bank risk taking respectively. The coefficients ɛ (of 𝑏𝑢𝑓𝑡−1) and 𝜑 (of 𝑟𝑖𝑠𝑘𝑡−1) measure the speed (i.e. adjustment costs) with which banks are able to adjust capital buffers and risk taking towards target levels. High speed of adjustment implies lower adjustment costs (Flannery and Rangan, 2006). If bank have a high speed of adjustment, it shows a fast ability to reach target levels of. This is only possible when adjustment costs to reach these levels are low. To date, a number of studies investigate adjustment costs towards target levels of capital buffers and bank risk taking through a partial adjustment framework17. These papers suggest that the values of the coefficients should lie between 0 and 1 (i.e. 0 < ɛ, 𝜑 < 1). This can be explained as follows. When banks do not incur adjustment costs, both ɛ = 1 and 𝜑 = 1. In such case, banks are able to adjust current levels to target buffer and risk levels such that it is expected that 𝑏𝑢𝑓𝑗𝑡 = 𝑏𝑢𝑓𝑗𝑡∗ and 𝑟𝑖𝑠𝑘𝑗𝑡 = 𝑟𝑖𝑠𝑘𝑗𝑡∗. By contrast, when ɛ = 0 and 𝜑 = 0, capital buffers and risk taking are expected to be equal to the capital buffer and risk taking level of the previous time period. This is due to the presence of high adjustment costs. In such case, it is expected that 𝑏𝑢𝑓𝑗𝑡 = 𝑏𝑢𝑓𝑗𝑡−1
and 𝑟𝑖𝑠𝑘𝑗𝑡 = 𝑟𝑖𝑠𝑘𝑗𝑡−1. Although this methodology is commonly suggested in the literature, adjustment costs cannot fully be explained with certainty due to the presence of the exogenous random shocks ∈jt and 𝜇𝑗𝑡.
Subsequently, equation (1) and (2) can be substituted into equation (3) and (4) such that changes in the capital buffer and bank risk taking can be written as:
(5) ∆𝑏𝑢𝑓𝑗𝑡 = ɛ (𝑏𝑢𝑓𝑗𝑡∗ − 𝑏𝑢𝑓𝑗𝑡−1) + ∈jt (6) ∆𝑟𝑖𝑠𝑘𝑗𝑡= 𝜑 (𝑟𝑖𝑠𝑘𝑗𝑡∗ − 𝑟𝑖𝑠𝑘𝑗𝑡−1) + 𝜇𝑗𝑡
Overall, observed adjustments in the capital buffer and risk taking in time period t of bank j can be depicted as a function of the target capital buffer and risk levels, the lagged capital buffer and risk taking levels and exogenous changes.
The target capital buffer 𝑏𝑢𝑓𝑗𝑡∗ and risk taking levels 𝑟𝑖𝑠𝑘𝑗𝑡∗ are unobservable. Therefore, a set of bank specific explanatory variables are determined in order to predict these levels. To date, various explanatory variables are found that might influence target capital buffers and risk taking levels. Notably, only a few of these variables are commonly used. The variables
14 that are used in this paper are identified by Schrieves and Dahl (1992) and outlined in section 4.2. Taken together, the equation model of (5) and (6) can be specified as follows:
(7) ∆𝑏𝑢𝑓𝑗𝑡 = 𝜀0 + ɛ1𝐷𝑐𝑎𝑝𝑗𝑡 + ɛ2𝑠𝑖𝑧𝑒𝑗𝑡 + ɛ3𝑅𝑂𝐴𝑗𝑡 + ɛ4∆𝑟𝑖𝑠𝑘𝑗𝑡−
ɛ5𝑏𝑢𝑓𝑡−1− ɛ6𝐷𝑐𝑎𝑝𝑗𝑡∗ 𝑏𝑢𝑓𝑗𝑡−1 + 𝜏𝑗 + ∈jt
(8) ∆𝑟𝑖𝑠𝑘𝑗𝑡= 𝜑0 + 𝜑1𝐷𝑐𝑎𝑝𝑗𝑡 + 𝜑2𝑠𝑖𝑧𝑒𝑗𝑡 + 𝜑3𝑙𝑙𝑜𝑠𝑠𝑗𝑡 + 𝜑4∆𝑏𝑢𝑓𝑗𝑡−
𝜑5𝑟𝑖𝑠𝑘𝑡−1− 𝜑6𝐷𝑐𝑎𝑝𝑗𝑡∗ 𝑟𝑖𝑠𝑘𝑗𝑡−1 + 𝜏𝑗 + 𝜇𝑗𝑡
The set of bank specific exaplantory variables, that influence target levels of capital buffers, include the size of a bank (𝑠𝑖𝑧𝑒𝑗𝑡) and a measure of bank profitability 𝑅𝑂𝐴𝑗𝑡. In the risk
equation, target risk levels are influenced by the size of the bank (𝑠𝑖𝑧𝑒𝑗𝑡) and the ratio of loan loss reserves (𝑙𝑙𝑜𝑠𝑠𝑗𝑡). Moreover, ∆𝑟𝑖𝑠𝑘𝑗𝑡 and ∆𝑏𝑢𝑓𝑗𝑡 are incorporated as regressors in the buffer and risk equation respectively. This is done because the relationship between capital buffer and risk taking adjustments is mutually dependent (Shrieves and Dahl, 1992; Rime, 2001). Thus, banks adapt risk taking levels in line with the amount of capital hold. 𝐷𝑐𝑎𝑝𝑗𝑡 is a
dummy variable that equals 1 for banks with capital buffer higher than 2% and zero otherwise18. These percentages are consistent with the definition of the EBA for well capitalised and undercapitalised banks as shown in appendix III. Also, the coefficients to examine adjustment speed towards target levels of capital buffers and risk taking are now further specified. These coefficients are ɛ5 (of 𝑏𝑢𝑓𝑗𝑡−1) and φ5 (of 𝑟𝑖𝑠𝑘𝑗𝑡−1) respectively. In addition, an interaction term is added to each equation. The interaction term examines if any differences exist between adjustment speed of well-capitalised and undercapitalised banks19. This interaction term is
specified as 𝐷𝑐𝑎𝑝𝑗𝑡∗ 𝑟𝑖𝑠𝑘𝑗𝑡−1 in the buffer equation and 𝐷𝑐𝑎𝑝𝑗𝑡∗ 𝑟𝑖𝑠𝑘𝑗𝑡−1 in the risk equation. As shown, equation (7) and (8) maintain the same structure as equation (5) and (6) as a combination of the simultaneous equations and partial adjustment model. Therefore, both equations contain two negative signs; one for the lagged levels of capital buffers (bank risk taking) and one for the interaction terms. Consistent with previous literature, it is expected that the values of these coefficients give a negative result somewhere between 0 and -1. Lastly, τj depicts a vector to model time fixed effects through time dummies. Brooks (2008) explains
15 that such a model allows that the intercepts change over time but not cross-sectional. For example, the time fixed effects model can capture regulatory changes over the years. Those changes are likely to impact adjustment decisions in buffer capital and risk taking, but are the same for all banks (i.e. do not change cross-sectional). Contrary, the bank fixed effects model would allow intercepts to change cross-sectional, but not over time. Of special interest in this paper is not the variation between banks, but across years. Another panel technique could be the random effects model, where all cross-sectional intercepts are assumed to arise from a joint intercept (Brooks, 2008). However, the random effects model is not appropriate when the error term is correlated with the explanatory variables. Due to endogeneity in the model, the random effects model is not appropriate. This endogeneity can be contributed to the fact that adjustments in capital buffers and bank risk taking are mutually related20.
A 2SLS method is adopted in order to estimate the equations as given in (7) and (8). Schrieves and Dahl (1992) use this approach to simultaneously address changes in capital and risk taking. Due to the presence of endogeneity in both equations, the ordinary least squares (henceforth: OLS) procedure will generate inconsistent coefficients. As explained by Brooks (2008), the 2SLS first estimates the reduced form equation by an OLS to obtain the fitted values for the dependent variable. Thereafter, the endogenous variable is replaced with the fitted values of the first stage to estimate the structural equations in the second stage.
4. Data and variables description
The following section provides more background information about all regression variables used in equation (7) and (8). In addition, the type of measurement for each variable is shortly mentioned. More detailed information of how variables are measured is given in appendix IV and V. Also, these appendices show data definitions and data sources of the regression variables. Section 4.1 provides information about the main variables; the capital buffer and bank risk taking. In addition, all variables that influence target levels of capital buffers and bank risk taking are outlined in section 4.2. Those variables include bank size, profitability and the loan loss reserves ratio. Lastly, the dummy variable and interaction term are more thoroughly discussed in section 4.3. Those variables allow to examine differences in well-capitalised and underwell-capitalised banks. Lastly, section 4.4 gives a description of the dataset.
20 The Hausman test shows that endogeneity is present in the model. This can be attributed to the mutual
16 4.1 Variables concerning capital buffers and bank risk taking
The capital buffer: In the literature, a variety of measures are adopted for capital buffers. This paper focusses on the supervisory review framework of pillar II. Therefore, it is important to look at the required capital level as monitored by regulators. This minimum capital level as set by the Basel Accord is a ratio of total regulatory capital, consisting of Tier 1 and Tier 2 capital, to risk weighted assets and should be at least 8% (Basel Committee on Banking Supervision, 2010). Following the Basel II definition and previous literature, the definition of the capital buffer in this paper is given as the difference between the actual capital ratio to risk weighted assets and the required minimum capital ratio of 8% (Jokipii & Milne, 2008; Shim, 2012). This variable is given as ∆buf.
Risk: Looking at bank risk taking from a regulatory perspective, portfolio risk appears to be the most important. Regulators base their minimum capital requirements on the amount of risky assets within the total portfolio of a bank. However, no consistent measure of portfolio risk exists, capturing all risks in the asset portfolio. In order to examine bank risk taking in a consistent way, different risk measures are adopted in this paper. The first risk measure (∆risk(1)) is the ratio of risk weighted assets to total assets21. It is an often used risk measure
in the academic literature to account for asset risk22. The logic reasoning behind the usage of this measure is that it depicts portfolio risk as the distribution of assets to various risk buckets and therefore captures asset risk (Rime, 2001). However, Jokipii and Milne (2011) show that the assignment of the same weights to different assets in one portfolio could generate misleading results. Therefore, a second risk measure is adopted; the ratio of non-performing loans to gross loans (∆risk(2)). This ex-post ratio of risk is widely used in the finance literature to capture bank risk taking23. It increases rapidly when banks face bankruptcy and therefore serves as a proxy for asset risk.
4.2 Explanatory variables that influence target levels of capital buffers and bank risk taking Bank size: The size of a bank is included in both equations because it might influence both target levels of capital buffers and bank risk taking. Titman and Wessels (1988) argue that
21 In this paper, total assets are measured in book values.
22 See among others; Schrieves and Dahl, 1992; Aggarwal and Jacques, 2001; Rime, 2001, Jokipii and Milne,
2001; Shim 2012.
17 larger banks can more effectively diversify asset risk. This causes them to hold lower target capital buffers and experience lower asset risk than smaller banks. Therefore, banks size is expected to show a negative sign in the capital buffer and risk equation when using the risk weighted assets to total assets as a risk measure. In addition, capital buffers of larger banks are generally lower than those of smaller banks due to better accessibility to capital markets (Titman and Wessels, 1988). Moreover, the too-big-to-fail theory states that banks become too big and mutually connected. In case of failure, banks could therefore have a dramatic impact on the economy. This theory states that government bail-outs, in terms of additional capital receipts, are expected when banks are financially distressed. The too-big-to-fail hypotheses therefore predicts banks to hold lower target capital buffers. In addition, Dam and Koetter (2012) suggest that bank-bailouts lead banks to increase risk taking behavior. In line with the too-big-to-fail hypothesis, the moral hazard theory predicts larger bank to adopt riskier lending behavior (Gorton and Huang, 2004). This behavior can be explained by the presence of deposit insurance. The depositor insurance causes banks to shift wealth from insurers by increasing risk in loan portfolios (Keeley, 1990). In such case, banks would be more eager to provide loans to less qualified borrowers, leading to higher non-performing loan ratios (Dimitrios, Angelos and Vasilios, 2011). In such case, the expected sign for bank size would be positive in the risk equation when using the non-performing loans to gross loans as a risk measure. However, as shown by Keeley (1990), the presence of charter values diminishes excessive risk taking. To capture the overall size effect, the log of the total assets is included in both equations, and is given by size. Overall, it is expected to have a negative sign in the capital buffer equation and an ambiguous sing in the risk equation.
Profitability: Banks’ profitability might have an influence on target capital buffers due to the preference of retained earnings over external financing methods to attract financing (Myers and Majluf, 1984). Being more profitable likely increases the ability to attract capital through retained earnings to rise capital levels. On the other hand, as banks’ profitability increases, the desired capital levels might decrease (Jokipii & Milne, 2008). Therefore, the sign of profitability in the capital equation is ambiguous. Similarly to Titman and Wessels (1988) and Fama and French (2002), profitability is measured as the pre-tax profits to total assets. This measure is given as ROA.
non-18 performing loans to gross loans as a risk measure. Furthermore, Jokipii and Milne (2011) show that if banks expect higher loan losses, they are more risky in terms of risky asset choice. Thus, the ratio of loan loss reserves is also expected to show a positive sign when using the risk weighted assets to total assets to measure risk taking. The loan loss reserves ratio is measured as the loan loss reserves to gross loans and is expected to have a positive sing in the risk equation and is given as lloss.
4.3 Allowing for different degrees of bank capitalisation
Dummy variable: The dummy variable Dcap is included in both equations to examine if adjustments in buffer capital and bank risk taking levels are dependent on the degree of bank capitalisation. The buffer theory predicts that well-capitalised behave differently compared to undercapitalised banks in terms of these adjustments. Jokippi and Milne (2011) state that banks holding high levels of buffer capital keep capital levels stable. Contrary, banks holding low levels of buffer capital try to increase current capital buffers to comply with regulatory requirements. As stated in hypothesis H21, it is expected that undercapitalised banks increase
capital buffers and decrease risk taking more than other banks.
Interaction terms: The interaction term in equation (7) assesses if differences exist in terms of adjustment speed towards target capital buffers for well-capitalised and undercapitalised banks. In order to examine these differences, the dummy variable Dcap is interacted with lagged levels of capital buffers (𝑏𝑢𝑓𝑡−1). This interaction term examines
hypothesis H31, that predicts that undercapitalised banks have a higher adjustment speed
towards target capital buffers than other banks. Although not supported by current empirical papers yet, the interaction terms is also added to the risk equation. This interaction term examines if adjustment speed towards target levels of risk taking shows similar results. For the risk equation, Dcap is interacted with the lagged level of bank risk taking (𝑟𝑖𝑠𝑘(1)𝑡−1, 𝑟𝑖𝑠𝑘(2)𝑡−1).
4.4 The dataset
19 Eurozone. The sample is not restricted to the SSM banks since stress testing, as part of the comprehensive assessments, is done by both the EBA and SSM. In the EU-wide stress test, banks of every EU member state and Norway are included that hold 50% or more of the national assets (EBA, 2014). All banks supervised by the SSM hold 82% of the total assets in the Eurozone (ECB, 2014). Thus, combining both list of banks gives a better overview of past performance of the most significant banks in Europe.
All data is obtained from the Bureau van Dijk Bankscope database. If available, missing values of listed banks were solved using Bloomberg and Thompson’s Financial Datastream database. Differences between the databases in terms of reporting standards and currencies were taken into account to overcome inconsistencies. The database contains date over the years 2008 until 2014. The years are chosen from the start of the Basel II regulation in 2008. The observations are on a yearly basis, which is the best periodicity with regard to data availability. To date, various studies concerning adjustments in capital levels and risk taking base their examinations on yearly data24. Therefore, yearly observation are assumed to provide consistent results. Due to non-available data, 15 banks are excluded from the sample, leaving the total sample to 141 banks. The bureau van Dijk Banksope database eliminates all historical data for banks that cease to exist. For example in the case of bankruptcy, acquisitions or mergers. This could create a survivorship bias. To date, all banks in the sample are either part of the SSM or were included in the EU-wide stress test in 2014. Therefore, no banks cease to exist during the sample period. The largest part of the sample includes commercial banks followed by cooperative banks as shown by appendix VI. Germany (27 banks), Spain (16 banks), Italy (13 banks) and France (11 banks) are the largest representatives in the sample. More information about the bank distribution per country can be found in appendix VII.
Figure 1 gives and overview of the total capital level developments between 2008 and 201425. The figure shows that total capital levels rise after 2008 with a small decrease in 2011.
This trend can be explained by the aftermath of the financial crisis and the constant regulatory interference to restore confidence in the financial sector after 2011. The figure confirms that SSM and EU-wide stress test banks hold total capital levels well above the 8% required minimum as set by the regulators. Given the definition of the EBA, most banks were well-capitalised during the sample period.
24 Schrieves and Dahl, 1992; Aggarwal and Jacques, 2001; Rime, 2001; Heid, Porath and Stolz, 2003; Jokipii and
Milne 2011.
20
Figure 1: Total capital level evolution during 2008-2014
This figure shows the evolution of the average total capital levels over the time period 2009-2014 for 141 banks of the SSM and EU-wide stress test. Total capital levels consist of both Tier 1 and Tier 2 capital in accordance to the Basel II regulation. The x-as shows the years whereas the y-as shows the levels of capital in percentages.
Interestingly, big differences exist between capital buffers of various countries in the sample. This is shown in figure 2. On average, banks in the sample hold a total capital buffer of 7.1 % of the risk weighted assets across all years. Banks in Luxembourg and Estonia hold the largest average levels of capital buffers during the sample period. Banks in Portugal and Greece hold the lowest average capital buffer. Although figure 2 provides a clear overview of the sample, it might not fully represent true levels of capital buffers in these countries. This can partly be explained by the fact that the number of banks present per country are not equally divided. For example, Estonia shows high levels of capital buffers compared to its peers in the sample. However, only one bank in Estonia is included in the sample.
21
Figure 2: Average capital buffers during 2008-2014 per country
This figure gives an overview of the average level of capital buffer for banks in each country in the sample. The x-as shows all countries of the SSM and countries included in the EU-wide stress test of 2014. The y-as gives the buffer capital levels in percentages. The capital buffer is calculated as the total capital ratio, consisting of Tier 1 + Tier 2 capital, minus the required minimum of 8%.
Table 1 provides descriptive statistics of all 141 banks included in the sample. The table gives a summary of the number of observations, mean, standard deviation, minimum and maximum value of all regression variables. These variables are the capital buffer, the lagged capital buffer, the change in capital buffers, lagged levels of both risk measures, the change in both the risk measures, bank size, ROA, the loan loss reserve ratio, the dummy variable and both interaction terms. On average, banks hold a capital buffer of 7.1% with a standard deviation of 7.8%26. The change in capital buffers appears to be 0.6%, implying that banks positively adjusted their capital buffers compared to the previous year. The mean of the risk weighted assets to total assets is 48.6% with a standard deviation of 20.9%. The mean of changes in risk weighted assets to total assets appears to be negative (-1.5%), indicating that banks on average decreased this ratio compared to previous period. The second risk measure, non-performing loans to gross loans, has a mean of 7.8% with a standard deviation of 7.5%. The change in non-performing loans to gross loans is 1.1%.
26 Standard deviations are high in the sample. Therefore, both the capital buffer and the risk equation are examined
when correcting for outliers. Winsorising the data to the 95% percentile gives similar results for both equations. The robustness section 6.1 discusses these findings.
0% 2% 4% 6% 8% 10% 12% 14% 16% 18% Au stri a Be lg iu m Cy p ru s De n m a rk Esto n ia F in lan d F ra n c e Ge rm an y Gre e ce Hu n g ary Ire lan d Ita ly Latv ia Li th u an ia Lu x em b o u rg M atl a Ne th erlan d s No rwa y Po la n d P o rtu g al S lo v ak ia S lo v en ia S p a in S we d en Un it e d Kin g d o m C ap ital b u ff er s Countries
22
Table 1: Descriptive statistics of the regression variables
This table represents descriptive statistics in terms of number of observations, mean, standard deviation, minimum and maximum of the regression variables. An unbalanced panel dataset is used with 38,447 yearly observations for banks in the Single Supervisory Mechanism in 2014-2015. Only the lagged levels of capital buffers and bank risk taking are observed for the time period 2007-2013. All regression variables are observed for the period 2008-2014. In total 4293 observations were missing, counting to 11.2% in total. Buf is the capital buffer, 𝑏𝑢𝑓(𝑡−1) is the lagged capital buffer, ∆𝑏𝑢𝑓 represents the change in the capital buffer, risk(1) is the ratio of risk weighted-assets to
total assets, 𝑟𝑖𝑠𝑘(1)𝑡−1 is the lagged risk measure of total risk weighted assets to total assets, ∆risk(1) is the change in the risk weighted
assets to total assets, risk(2) is the ratio of non-performing loans to gross loans, 𝑟𝑖𝑠𝑘(2)𝑡−1 is the lagged level of non-performing loans to
gross loans, ∆risk(2) is the change in non-performing loans to gross loans, size is the natural log of total assets, ROA is a measure of profitability and lloss is a ratio of loan reserves to gross loans, 𝐷𝑐𝑎𝑝𝑗𝑡 is a dummy variable to capture time effects and 𝐷𝑐𝑎𝑝𝑗𝑡∗ 𝑏𝑢𝑓𝑗𝑡−1,
𝐷𝑐𝑎𝑝𝑗𝑡 ∗ 𝑟𝑖𝑠𝑘(1)𝑗𝑡−1, 𝐷𝑐𝑎𝑝𝑗𝑡 ∗ 𝑟𝑖𝑠𝑘(2)𝑗𝑡−1 are interaction terms to allow for differences in high and undercapitalised banks.
Observations Mean Stand. Dev. Minimum Maximum
buf 878 0.0712 0.0780 -0.131 1.0320 𝑏𝑢𝑓𝑡−1 858 0.0654 0.0761 -.1310 1.0320 ∆𝑏𝑢𝑓 829 0.0066 0.0374 -0.3100 0.3710 Risk(1) 763 0.4864 0.20935 0.0145 1.1079 𝑟𝑖𝑠𝑘(1)𝑡−1 743 0.5008 0.2124 0.0145 1.1079 ∆𝑟𝑖𝑠𝑘(1) 719 -0.0146 0.0732 -0.7330 0.6342 Risk(2) 768 0.0776 0.0751 0.0003 0.5181 𝑟𝑖𝑠𝑘(2)𝑡−1 745 0.0673 0.0651 0.0003 0.5181 ∆𝑟𝑖𝑠𝑘(2) 716 0.0111 0.0326 -0.2278 0.2219 size 939 17.9807 1.6427 12.4665 21.6458 ROA 912 0.00047 0.0174 -0.15674 0.1056 lloss 838 0.0423 0.0427 0.0002 0.3394 𝐷𝑐𝑎𝑝𝑗𝑡 878 0.9146 0.2797 0.0000 1.0000 𝐷𝑐𝑎𝑝𝑗𝑡∗ 𝑏𝑢𝑓𝑗𝑡−1 841 0.0625 0.0772 -0.0710 1.0320 𝐷𝑐𝑎𝑝𝑗𝑡 ∗ 𝑟𝑖𝑠𝑘(1)𝑗𝑡−1 734 0.4535 0.2435 0.0000 1.1079 𝐷𝑐𝑎𝑝𝑗𝑡 ∗ 𝑟𝑖𝑠𝑘(2)𝑗𝑡−1 694 0.0628 0.0664 0.0000 0.5181
Table 2: Correlation matrix
This table shows the pairwise correlations for all continuous variables. Buf is the capital buffer, 𝑏𝑢𝑓𝑡−1 is the lagged capital buffer, ∆buf represents the change in the capital buffer, risk(1) is the ratio of risk
weighted-assets to total weighted-assets, 𝑟𝑖𝑠𝑘(1)𝑡−1 is the lagged risk measure of total risk weighted assets to total assets, ∆risk(1) is the change in this risk measure, risk(2) is the ratio of non-performing loans to gross loans, 𝑟𝑖𝑠𝑘(2)𝑡−1
is the lagged level of non-performing loans to gross loans, ∆risk(2) is the change in the risk measure, size is the natural log of total assets, ROA is a measure of profitability and lloss is a ratio of loan reserves to gross loans. ***, **, and * show significance at 1%, 5% and 10% level.
5. Results
Results for the capital buffer and risk equation as given in equation (7) are presented in table 3 and 4. The simultaneous equations are examined using the 2SLS27. Table 3 shows the results when using the risk weighted assets to total assets as a measure for bank risk taking. Table 4 shows the results for the non-performing loans to gross loans as a risk measure. In both tables, the intercept and time dummies are given for all years of the sample period except one. For example; Dum09 sums to unity for all observations in 2009 and zero otherwise. The time dummies allow the coefficients to be different compared to 2008.28 Accordingly, these
dummies allow for regulatory changes or macro-economic shocks during the sample period that might influence adjustments in capital buffers and bank risk taking (Schrieves and Dahl, 1992 and Rime, 2001). Thus; the time dummies allow to correct for time fixed effects. As explained in the methodology section 3, this panel data technique is considered to be of best interest for the model in this paper. Largely, the dummy variables are not statistically significant indicating that no differences between time observations exist. Model I and II examine both equations without and with the interaction terms respectively. These interaction terms are added to the equations in order to examine if the speed of adjustment towards target capital buffer and risk taking levels differ across well-capitalised and undercapitalised banks. The result section reports abbreviations of the regression variables as reported in earlier tables of section 4.4. The results for the first hypothesis regarding the relationship between capital buffers and bank risk taking is discussed in section 5.1. Thereafter, section 5.2 provides the results with regard to explanatory variables that influence target capital buffers and bank risk taking. Lastly, section 5.3 discusses the results concerning differences between well-capitalised and underwell-capitalised banks that allow to draw conclusions about the second and third hypothesis.
27 It was decided that the best method to adopt for this study is the 2SLS. A large part of the empirical papers uses
the three-stage least squares (3SLS). This method would result in more efficient estimates of parameters than the 2SLS (Jacques and Nigro, 1997; Aggarwal and Jacques, 2001; Rime, 2001; Shim, 2012). Therefore, the analyses are also employed using the 3SLS. Although standard errors slightly reduced, coefficients and significance levels largely remain the same. The results of the 3SLS are available upon request.
26
Table 3: Results for the ratio risk weighted assets to total assets to measure bank risk taking
This table shows the results of the 2SLS estimation. The dependent variables are ∆buf (measured as the change in capital buffers) and ∆risk(1) (measured as the change in risk weighted assets to total assets). 𝑏𝑢𝑓𝑡−1 is the lagged capital buffer, 𝑟𝑖𝑠𝑘(1)𝑡−1 the lagged measure of risk, size is the natural log of total assets, ROA is the pre-tax profit to total assets and lloss is a ratio of loan reserves to gross loans. 𝐷𝑐𝑎𝑝 ∗ 𝑏𝑢𝑓𝑡−1 and 𝐷𝑐𝑎𝑝 ∗ 𝑟𝑖𝑠𝑘(1)𝑡−1 are the interaction terms. Model
I and model II present results of analyses without and with interaction terms respectively. Dum09, Dum10, Dum11, Dum12, Dum13 and Dum14 are dummy variables to allow for time effects. ***, **, and * show significance at 1%, 5% and 10% level.
Variable model I model II
∆𝐛𝐮𝐟 ∆𝐫𝐢𝐬𝐤(1) ∆𝐛𝐮𝐟 ∆𝐫𝐢𝐬𝐤(1) Intercept 0.0350 (0.0228) 0.1051** (0.0434) 0.0470** (0.0220) 0.1034* (0.054) Dcap 0.0384*** (0.0079) -0.0241** (0.0111) 0.0235*** (0.0085) -0.0227 (0.0377) ∆risk(1) 0.3388** (0.1348) 0.3053** (0.1289) ∆buf 0.1791 (0.2114) -0.1425 (0.1897) 𝑏𝑢𝑓𝑡−1 -0.3311*** (0.0474) -1.0567*** (0.1968) 𝑟𝑖𝑠𝑘(1)𝑡−1 -0.0912*** (0.0164) -0.0885 (0.0558) size -0.0199* (0.0079) -0.0037* (0.020) -0.0019* (0.0011) -0.0037* (0.020) ROA 0.1906* (0.1034) 0.1611 (0.0989) lloss 0.0685 (0.0740) 0.0662 (0.0743) 𝐷𝑐𝑎𝑝 ∗ 𝑏𝑢𝑓𝑡−1 0.7639*** (0.2017) Dcap ∗ risk(1)t−1 -0.0035 (0.0576) Dum09 -0.0055 (0.0082) 0.0279*** (0.0106) -0.0062 (0.0078) 0.0279*** (0.0107) Dum10 -0.0028 (0.0075) 0.0167 (0.0106) -0.0042 (0.0071) 0.0167 (0.0107) Dum11 -0.0106 (0.0076) 0.0182* (0.0184) -0.0112 (0.0072) 0.0187* (0.0109) Dum12 0.0088 (0.0069) -0.0024 (0.0105) 0.0047 (0.0066) -0.0024 (0.0107) Dum13 -0.0002 (0.0079) 0.0219** (0.0109) -0.0019 (0.0075 0.0220* (0.0110) Dum14 0.0003 (0.0077) 0.0170 (0.111) -0.0019 (0.0074) 0.0172 (0.0112) Observations 647 647 647 647 Adjusted R2 -0.574429 0.1161 -0.4297 0.1100
29 The R2 in the buffer equation for both model specification I and II is negative. In such case, the residual sum of squares (RSS) is lower than the total sum of
squares (TSS). Therefore, the model sum of squares (MSS) and R2 is negative. A negative reported R2 is common when the 2SLS methodology is used and does
27
Table 4: Results for the ratio of non-performing loan to gross loans to measure bank risk taking
This table shows the results of the 2SLS estimation. The dependent variables are ∆buf (measured as the change in capital buffers) and ∆risk(2) (measured as the change in non-performing loans to gross loans). 𝑏𝑢𝑓𝑡−1 is the lagged capital buffer, 𝑟𝑖𝑠𝑘(2)𝑡−1 the lagged measure of risk, size is the natural log of total
assets, ROA is the pre-tax profit to total assets and lloss is a ratio of loan reserves to total reserves. 𝐷𝑐𝑎𝑝 ∗ 𝑏𝑢𝑓𝑡−1 and 𝐷𝑐𝑎𝑝 ∗ 𝑟𝑖𝑠𝑘(2)𝑡−1 are the interaction
terms. Model I and model II present results of analyses without and with interaction terms respectively. Dum09, Dum10, Dum11, Dum12, Dum13 and Dum14 are dummy variables allow for time effects. ***, **, and * show significance at 1%, 5% and 10% level.
Variable model I model II
28 5.1 Results of the relationship between capital buffers and bank risk taking
Overall, ∆buf and ∆risk appear to be positively related, providing support for hypothesis H11.
The negative relationship between ∆risk(1) and ∆buf in table 3 (see model II) is not significant and therefore cannot be explained with certainty. In the buffer equation, results are only significant using ∆risk(1) to capture bank risk taking. These results indicate that if adjustments in risk taking increase with 1 percentage point, adjustments in capital buffers increase with 33.88 (30.53) percentage points. Significance in the risk equation for ∆risk(2) (see table 4) also supports the positive relationship. This results show that if adjustments in capital buffers increase with 1 percentage points, banks adjust risk taking with 28.90 (27.82) percentage points. The results match those observed in earlier studies during both Basel I and Basel II regulatory periods, stating that banks increase (decrease) risk taking while simultaneously increasing (decreasing) capital buffers30.
As shown in figure 1, the SSM and EU-wide stress test banks are adequately capitalised. Given this characteristic of banks in the sample, results are consistent with the capital buffer theory that predicts well-capitalised banks to adjust capital buffers and risk taking levels in the same way31. As suggested by the theory, this finding might be the result of supervisory regulation as specified in Basel II (Alaa et all., 2013). Furthermore, these results support the idea that banks became better capable to balance capital levels and risk taking behaviour because it shows that higher risk taking also causes banks to increase capital buffers. Such observed behavior is in line with targets of the pillar II supervisory review processes, such as SREP. Another possible explanation of the positive relationship might be banks’ private incentives (Schrieves and Dahl, 1992). For example, managerial compensation is dependent on risky claims which causes managers to be more risk averse then stockholders. Also, managers are more risk averse than stockholders in general. The observed increases in capital levels when bank risk taking increases can therefore be attributed to managers trying to protect their compensation levels (Saunders and Travlos, 1990).
30 Schrieves and Dahl, 1992; Rime, 2001; Jokippi and Milne, 2011, Shim, 2012.
31 The theory also suggests that the relationship between adjustments in capital buffers and bank risk taking is
29 5.2 Results of explanatory variables that influence target capital buffers and bank risk taking
In general, the explanatory variables carry the expected sign. The relationship between size and ∆buf is negative in both table 3 and 4. However, these results only show to be statistically significant when using ∆risk(1) as a risk measure. The results are consistent with the idea that larger banks are better able to diversify, leading them to hold lower capital buffers. In addition, larger banks have better access to capital markets which reduces the need to hold higher capital buffers on a constant basis. If necessary, large banks can increase capital through capital markets relatively easy. This results is also in line with the too-big-to fail hypothesis. This suggests that larger banks hold lower levels of capital buffers due to bail-out provisions during difficult financial times. With regard to the risk equation, the relationship between size and ∆risk(2) was observed to be positive and significant in Table 4. This result is also consistent with the too-big-to-fail hypothesis, stating that larger banks are more willing to engage in excessive risk lending behavior due to the safety net. In such case, banks are more likely to provide loan to less qualified borrowers that results in higher non-performing loan ratios. The relationship between size and ∆risk(1) in the risk equation is negative and significant in table 3. This indicates that if bank size increases, asset risk decreases and vice versa. Consistent with previous theoretical and empirical literature, this result could be explained by the fact that larger banks are better able to diversify asset risk. Also, this result is in line with the presence of constant supervisory review processes after the credit crunch in 2008. Supervisory measures such as comprehensive assessments and SREP, demand banks to be better in control of all risk faced, leading them to be more risk averse in general.
The relationship between ROA and ∆buf is positive. As explained in subsection 4.2, an ambitious sign was expected for ROA. The positive relationship can be explained by the fact that more profitable banks have higher retained earnings. This leads them have better access to financial markets. Higher capital buffers are the result. Although ROA and ∆buf seems to be positively related in both buffer equations (see table 3 and 4), results are not significant. Therefore, these results do not allow to be interpreted with certainty.
30 amount of asset risk in the portfolio also increases. This finding is consistent with previous literature, showing that banks who expect higher loan losses also choose riskier assets.
5.3 Results for well-capitalised and undercapitalised banks
The dummy variable Dcap is positive and significant for ∆buf as shown in table 3 and 4. Thus, highly capitalised banks in the sample adjust capital buffers by 3.84 and 3.29 percentage points more compared to other banks. This evidence contradicts expectations since well-capitalised banks are expected to raise capital levels less than their undercapitalised peers. Those banks are not likely to experience regulatory pressure and therefore do not fear costly interferences. However, results show the opposite. Furthermore, Dcap is negative related to ∆risk(1) and ∆risk(2), but only significant in model I (see table 3 and 4). This suggests that well-capitalised banks lower risk taking behavior with 2.41 and 1.69 percentage points more than other banks. Interestingly, the association between Dcap and ∆risk(1), ∆risk(2) is not significant when including the interaction term in model II. Including the interaction term, Dcap represents the association of only undercapitalised banks with the outcomes ∆risk(1), ∆risk(2). Now, undercapitalised banks did nog significantly predict to lower risk taking. Together, these results imply that hypothesis H21 should be rejected, stating that
undercapitalised banks increase capital buffers and decrease risk taking more than other banks. In contrast, results shows that well-capitalised banks increase capital buffers and decrease risk taking more than other banks. As suggested by Jacques and Nigro (1997), this results might be due to the fact that well-capitalised banks try to protect themselves against future equity shocks. Also, constant regulatory pressure of the Basel Accords could be an explanation. Furthermore, this result indicates that undercapitalised banks are less able to comply with possible regulatory actions. This an important finding for supervisory review processes.
As stated in equation (7) and (8), the coefficients 𝜑5 and ɛ5 are indicators of the speed
of adjustment towards target levels of capital buffers and risk taking. The coefficients in model I are averages for both well-capitalised and undercapitalised banks. These coefficients are given as 𝑏𝑢𝑓𝑡−1 for the buffer equation and 𝑟𝑖𝑠𝑘(1)𝑡−1, 𝑟𝑖𝑠𝑘(2)𝑡−1 for the risk equation.
