Master Thesis
University of Groningen Faculty of Economics and Business
MSc Economics & MSc Finance 23.06.2017
Taylor Rule Residual and Bank Risk-Taking
Author: Ronald Ruben Heins
Course code: EBM000A20
Mail: r.r.heins@student.rug.nl
Student number: 2353628
2
Abstract
This paper examines the relationship between the interest rate environment and bank risk-taking behavior using a dataset of 114 banks within the European Monetary Union for the time period 2000 and 2016. The Fixed Effects model indicates a negative relationship between the interest rate, described by the Taylor rule residual, and banks’ risk appetite, as proxied by loan growth and risk assets. Nevertheless, the results are not consistent for the dependent variables and the different specifications of the model. Furthermore, the risk-taking coefficient turns positive for the post-2011 period, suggesting that the current expansionary monetary policies do not trigger additional bank risk-taking.
3 Table of Contents Abstract ... 2 Table of Contents ... 3 Introduction ... 4 Literature Review ... 6
Impact of Bank-Risk Taking ... 6
Risk Taking proxies ... 7
Regulatory Framework ... 9 Data ... 10 Descriptive Statistics ... 13 Research Methodology ... 15 Results ... 17 Pooled OLS ... 18
Fixed Effects Model ... 19
Persistence of Risk-Taking Behavior ... 21
Discussion and Limitations ... 23
Policy Implications ... 25
Conclusion ... 26
Bibliography ... 27
Databases ... 28
Appendix ... 28
Appendix A: Variables Description ... 28
Figures ... 32
Appendix B: Explanation and Construction of Taylor Rule Residuals ... 33
Appendix C: Construction of Regulatory Index ... 34
4
Introduction
The low interest rate environment that was pursued and sustained by central banks in the United States and in Europe after the dotcom bubble is often blamed for fueling new financial imbalances, leading to the build-up of the world-wide financial crisis of 2008.1 The environment created pressures on the banking system’s core operating results via the net interest margins. What we observed in the market was a ‘search for yield’ that pushed financial institutions into more risk taking activities, facilitated by financial innovations and a relatively non-stringent regulatory framework. The accumulated asset bubble and risk exposures eventually manifested themselves after the Lehman Brothers default, putting the world in a big depression. Almost ten years later, Europe still finds itself dealing with the legacy of the financial crisis, as exemplified by the long-lasting, artificially low interest rate. In light of the current monetary policies, that resemble much of the sustained interest rate environment before the financial crisis, it is crucial to gain a profound understanding of the relationship between the interest rate and bank risk-taking behavior and the exact risk channels through which these exposures are accumulated.
Securitization activities took a prominent position in the interplay between the interest rate environment and bank risk-taking during the build-up to the financial crisis in 2008. A good way to understand this mechanism is by classifying the financial institutions into three main groups: commercial banks, investment banks, and the insurance groups/pension funds. First of all, the low interest margins incentivized commercial banks to lower their underwriting standards and expand their loan portfolio to offset the pressured margins on lending operations. The banks faced too much competition from other lenders on their interest rate activities as a result of excess capital, while there seemed to be a lower bound on the return that they pay on the liability side: the depositors, altogether putting pressures on the net interest margin. Consequently, the commercial banks were inclined to increase the riskiness of their loan portfolio or increase the credit that is provided. Simultaneously, insurance groups and pension funds were unable to generate their targeted return on capital, as the risk free investments were not profitable enough and below their projected return on assets. Investment banks were pleased to facilitate as a bridge between demand and supply, by re-packaging the historically illiquid loan portfolio into fixed income products.
5
The asset securitization, which was only one of the risk-taking channels, appeared very profitable for the banking system for a long period of time. On the back of a long tendency of deregulation and product innovation in the financial sector, it allowed banks to increase their leverage and accumulate huge risk exposures. Despite the wide focus on the interest rate in current economic research, its exact impact on banks’ risk-taking behavior has remained rather trivial, due to the wide range of dynamics related to risk-taking. Several empirical papers (Altunbas et al., 2010; Delis and Kouretas, 2011; Maddaloni and Peydro, 2011) show that the low interest rate environment increased the banking system’s willingness to take on more risk, while securitization activities seem to amplify this relationship. Despite the intuitive results, the existing literature often comprises one of the following econometrical concerns that may undermine the inference from their work. There is a major endogeneity problem in the relationship between the interest rate and bank risk-taking, since the central banks predominantly use the interest rate as their main monetary policy instrument to pursue price stability. In order to remove this endogeneity bias, one needs to examine the interest rate as an exogenous variable that is not affected by pro- or countercyclical measures from the monetary institutions. Furthermore, the existing literature tends to use risk-proxies that capture the risk exposure of banks rather than the actual risk-taking, as exemplified by the usage of risk-proxies such as non-performing loans, loan loss provisions, or the expected default frequency.
This paper therefore examines the relationship between Taylor rule residuals (TRR)2 and bank risk-taking for banks located in the European Monetary Union (EMU), where banks’ loan portfolio growth and their share of risky assets are used as proxies for risk-taking behavior. To my best knowledge, the existing literature has not examined the risk dynamics with the setting mentioned above, making this paper a valuable contribution. Using the TRR for EMU countries will remove potential endogeneity of the interest rate, as EMU countries face a uniform interest rate imposed by the ECB even though they differ in terms of inflation and economic performance. Different economic fundamentals subsequently translate into heterogeneous country-specific TRRs. The following research questions will take center stage in this paper; Do Taylor rule residuals explain the risk-taking behavior of banks? And if so, is the artificially low interest rate currently shaping new financial imbalances in the Eurozone countries with negative residuals, potentially leading to the onset of a new financial crisis? Examining the latter is essential, regarding the similarities between the mid-2000s’ and the current interest rate environment, and the incomplete coverage on the post-crisis dynamics.
6
Literature Review
Impact on Bank Risk-Taking
The existing literature provides several channels through which the interest rate can influence bank risk-taking. Altunbas et al (2010) stress the direct effects of monetary policy on asset price valuations and measured risk. In estimating credit risk exposure, banks typically make use of an expected loss function, which depends on the probability of default and its exposure. The probability of default is the likelihood that a loan will not be reimbursed and will fall into default. During times of a low interest rate environment, one generally finds that asset prices increase, which increases collateral values of loans and therefore reduces the probability of default. As banks aim to hold the same amount of credit risk, either because of internal targets or due to regulatory constraints, they tend to keep their expected loss constant. This is done by increasing the loan portfolio (exposure at default), till it has offset the lower probability of default. In the end, this scenario has had no effect on the banks’ credit risk exposure, which is dependent on the situation in the market, while the bank has implicitly taken on more risk by expanding its underlying credit risk.
A similar effect is observed on the market-risk spectrum of the bank, which is usually estimated with the Value-at-Risk (VaR) measurement. The VaR measures the maximum loss on the market portfolio at a certain confidence level (usually 99%), which depends on the price volatility and the total value of the portfolio. As a result of expansionary monetary policies, there are two effects that tend to decrease the VaR. First, interest rates have a direct negative effect on the present value of assets, captured in the discount factor. Secondly, asset price volatilities are often lower when asset prices go up, which decreases the VaR via the volatility term. Similar effects are observed for other risk segments being managed by banks. Nonetheless, this paper limits its focus to credit risk and market risk, since these comprise the main risk exposures of financial institutions.
7
behavior in a low interest rate environment. A considerable problem in this research is the proxy for bank risk-taking, as the EDF tends to measure the risk exposure, not the risk-taking behavior itself. Banks that are featured by a high EDF, therefore argued to have low risk-aversion, could simply face a higher risk exposure to market, while it may not have chosen this exposure in terms of risk-taking behavior.
A second way in which monetary policy can influence bank risk-taking is through the ‘search for yield’. Rajan (2005) argues that asset managers may take on more risk when interest rates are low, because of a so-called money illusion. Investors are focused on the nominal returns and seem to neglect that low interest rates decline to compensate for low inflation, which will partially offset the change in real interest rate return. Furthermore, financial institutions (especially pension funds and life insurers) may be subjected to regulatory constraints, as they are required to obtain a minimum nominal rate of return. These constraints may push them into more risky instruments, yielding a higher return, which aligns with the scenario presented in the Introduction. Buch et al (2013) validate this ‘search for yield’ mechanism and find that banks increase loan volumes, while loan spreads decline. Furthermore they argue that small domestic banks increase their exposure to risk more significantly than other banks.
A third channel is through habit formation. Campbell and Cochrane (1999) find evidence that the risk-aversion of investors decreases during good times, since their consumption levels increase relative to normal levels. Expansionary monetary policy is shown to decrease the required equity risk premium, which makes investors and financial institutions willing to invest in lower yielding return projects for a similar amount of risk. Several central banks have already defined the low risk premium, which is a direct result of the lack of investment alternatives, as a potential source of risk for the continuity of the systems stability in the current days.3 In light of the ongoing quantitative easing program from the ECB, which implies the montly purchase of EUR 60bn worth of investment grade corporate- and sovereign bonds, it is relevant to examine the exact dynamics through which the equity risk premium enters the risk-taking equation.
Risk-Taking Proxies
Andries et al. (2015) use the loan loss provisions, non-performing loans and the Z-score as proxies for banks’ risk appetite to measure the effect of monetary policy on risk-taking, where
3
8
the Z-score is computed as the return on assets plus the capital-asset ratio to the standard deviation of the assets returns. The authors take a primary focus on the financial crisis that is regarded as the period between 2008 and 2011. Using a dataset of 571 commercial banks in the Eurozone and the TRR for the respective entities, they find mixed results for the effect on the different risk-taking proxies. Interestingly, the authors find that the negative interest rate coefficient becomes more significant during periods of financial turmoil. A considerable problem with the proposed risk-taking proxies is that they all seem to measure the risk exposure and not the actual risk-taking behavior, since their values are mostly driven by external market movements. In other words, the three proxies are not able to observe the increased risk-appetite at the time the additional risk is taken, but only after a certain time. During times of increased financial instability, when risk proxies start to rise, it is hard to argue whether this is due to its idiosyncratic nature or because of the firm’s exposure to the market, which puts the results into question. Banks that showed unsustainably high NPLs and loan loss provisions did not necessarily take on the same amount of risk as others, but may simply be exposed to the wrong market segments. Despite to invalidity of the proxies from Andries et al. (2015), the empirical set-up is applicable to different risk-taking variables.
An important contribution to the existing literature was provided by Foos et al. (2009), who examine three hypotheses on the relation between abnormal loan growth and asset risk, bank profitability, and bank solvency. The results suggest that loan growth is one of the main drivers of bank risk taking as they find a significant effect on all three accounts. The evidence is backed by the work from Amador et al. (2013), who use a rich data panel of Colombian financial institutions. Abnormal credit growth for a prolonged period of time is found to reduce bank solvency and increase the non-performing loans ratio. The findings of the papers above are convincing enough to incorporate loan growth as a proxy for bank risk-taking, as it is reported on the exact moment the risk is acquired, while it is not associated with the level of risk exposure. Therefore this paper will make use of this proxy while having a research methodology that is in line with the approach from Andries et al (2015).
9
dependent variable. Besides controlling for bank- and country fixed effects, The authors also control for regulatory conditions, using the dataset of Barth et al. (2008) that comprises an extensive survey on bank regulation and supervision for 143 countries. Using the published versions from 2003 and 2007, the authors categorize the questions into three indices pertaining to capital stringency, official supervisory power and market discipline. One would expect that the regulatory indices have become more relevant throughout the years in determining banks’ risk-appetite, which stresses the importance to update the regulatory framework with more recent data.
Maddaloni and Peydró (2011) study the relationship between low interest rates on bank risk-taking and argue that low short term interest rates soften underwriting standards for households and corporate loans. Besides, this softening of lending standards seems to be amplified by asset securitization, weak capital regulations and, by a ‘too low for too long’ effect. In order to resolve the endogeneity of monetary policy, the authors use the TRRs in the Euro-zone and US. Another finding is that a softening of lending standards in the run-up to the financial crisis is associated with lower TRRs and worse economic performance in the period after the burst.
Dell’ Ariccia et al (2014) support the effects on leverage and risk-taking by means of a theoretical framework. They argue that low interest rates incentivize firms to increase their leverage and accumulate more risk exposure if they can endogenously determine their capital structure. If the capital structure is fixed, due to regulatory capital requirements, the effect on risk-taking depends on the initial leverage. Regarding the Basel III agreement one could argue that the capital structure has become more restricted and exogenously determined by the more stringent regulatory framework, which could possibly have changed the interaction between banks’ risk-taking behavior and the interest rate environment.
Regulatory Framework
10
present in the Basel II framework. The new capital requirements incorporated off-balance sheet exposure based on a Value at Risk measure. However these VaRs are derived from historical data, which underestimate the risk exposures during good times in which volatilities are low and correlation coefficients give a too rosy picture. It is often argued that these regulatory arbitrages have strongly contributed to the financial crisis in 2008.
Basel III was introduced in 2013 as a response to the financial crisis. The new regulatory framework comprises stricter capital adequacy, leverage ratios and implements stress tests based on extreme market movements. In accordance with the Basel III regulations, Dell’ Ariccia et al (2014) created a model in which only well capitalized banks increase their risk when the interest rate is low, as capital structures have become more exogenously determined. In other words, institutions with lower solvency metrics do not have the option to increase their leverage and risk behavior, because they need to comply with the stringent capital buffers and other regulatory requirements. Concerning this endogeneity of the capital structure, it is extremely relevant to research whether the effect of the interest rate on risk-taking behavior is similar for the pre- and post-crisis period.
All in all, the key question whether the interest rate environment has an effect on banks’ risk-taking behavior has remained partly unanswered by the existing literature. Many empirical studies tend to focus on the bank’s risk exposure to the market, rather than the actual appetite for risk. This common measurement error calls for the introduction of additional proxies for risk-taking behavior, in combination with an exogenous interest rate variable. Furthermore the dynamics need to be examined for the current interest rate environment, to examine whether the expansionary monetary policies are potentially fueling new financial imbalances.
Data
This section will serve as a brief discussion on the database that is used for this research. Appendix A provides a more elaborate discussion on the derivation of the data variables, the reasoning behind their inclusion and potential limitations to the data.
11
The data is collected for 114 publicly listed banks from 17 Eurozone countries4. The time frame considered ranges from the introduction of the Euro, year 2000, till the most recent figures, which optimally results in 16 observations of net loan growth for every bank entity. Note that some countries have entered the Eurozone at a later stage. This is corrected for by using 2005 as a starting point for Cyprus, Estonia, Malta and Slovenia, and 2006 for Latvia and Slovakia as this aligns with the respective year in which their currency was pegged to the Euro. In order to correct for potential seasonality of certain performance metrics used as control variables, and to foster the completeness of the dataset, preference was given to yearly data.
Besides the rather limited amount of data in comparison to other papers, there are at least two additional limitations to the data. First of all, there is a so-called ‘survivorship’ bias, as some banks do not exist in the sample over the complete time frame, because they were either acquired, merged or went into default in the meantime. Banks that did survive may have taken over other banks in the observed time period, which increased their loan portfolio, though this expansion was not necessarily driven by a higher risk appetite. Nonetheless, it would be beyond the scope of this research to disentangle the exact value of M&A activities from the acquirer’s loan portfolio. In an effort to correct for the survivorship bias, total assets are included as a control variable, and a dummy in constructed for each year a bank completed an acquisition, whose deal value was deemed sizeable to the banks’ total assets. I expect both coefficients to be significantly positive as a function of the first bank risk-taking proxy, which would absorb the major part of the implied size effect.
A second limitation is referred to as a ‘globalization’ bias. Since the ‘80s, considered as the start of major financial innovation and digitalization, one can observe a transition towards more international activities within the banking sector. Alongside the deregulation trend, it has allowed banks to expand their operations in foreign markets, which currently accounts for over 50% of total operations for several banks. This globalization process has made the relationship between the interest rate environment and bank’ risk-taking less clear cut, since many banks face heterogeneous interest rates because of their international characteristics. As Datastream only features financial data for listed firms, implying balance sheet items and performance metrics for the total group / parent, the bias mentioned in the above may have a significant impact on the results. Correcting for the international characteristics is not a possibility, since it would leave us with a sample that is too small to provide meaningful interpretations.
12 Appendix D provides a comparison between the consolidated banking group results and the unconsolidated results by examining total assets for all 114 banks included in the empirical analysis, using current data from Orbis Bank Focus. For Austria, France, Italy and Spain, there is a rather large mismatch between the sample banks’ unconsolidated and consolidated total assets, implying that risk-taking metrics and bank-specific control variables for several banks in these countries do not give a good reflection of the domestic operating environment. Furthermore, observing a low sample bias does not necessarily imply that the entity specific internationalization bias within that country is insignificant. The provided statistics simply serve as an indication in which country the banks’ group results differ most from their domestic stand-alone total assets, on an aggregated basis.
13
Table 1
Descriptive Statistics
Number of Obs Completeness (%) Mean Std. Deviation Min Max
ln (Loan growth) 1,513 82.9 4.049 21.111 -164.733 201.128 Risk assets 1,554 85.2 0.876 0.116 0.204 0.999 Total assets 1,565 85.8 1.618 3.578 29,465 2.399 ROE 1,483 81.3 5.042 21.960 -230.560 88.620 ROA 1,278 70.1 0.859 1.490 -23.838 10.247 DEBTEQ 1,558 85.4 6.431 13.896 -274.533 303.484 MRO 1,824 100 1.626 1.251 0 3.922 TRR1 1,824 100 -0.939 3.902 -17.913 18.938 TRR2 1,824 100 1.732 4.453 -19.268 21.773 Real GDP growth 1,824 100 1.200 2.865 -15.929 23.330 Regulatory index 1,824 100 8.855 2.688 5 16
14
Table 2
Correlation Matrix
Lnlgr Ra/ta Totas ROE_1 ROA_1 DEBTEQ MRO TRR1 TRR2 Rggdp Regu
Lnlgr 1.000 Risk assets -0.101 1.000 Total assets 0.136 0.137 1.000 ROE_1 0.467 -0.077 0.009 1.000 ROA_1 0.315 -0.101 -0.125 0.766 1.000 DEBTEQ 0.077 0.106 0.175 0.006 0.198 1.000 MRO 0.445 -0.127 0.091 0.480 0.556 0.265 1.000 TRR1 -0.229 0.180 -0.150 -0.460 -0.207 -0.042 -0.247 1.000 TRR2 -0.163 0.381 -0.116 -0.351 -0.142 0.003 -0.255 0.875 1.000 GDP growth 0.399 -0.054 0.087 0.254 0.110 0.192 0.356 -0.457 -0.372 1.000 Regu index 0.071 0.114 -0.158 0.004 -0.090 -0.150 -0.317 0.217 0.290 -0.007 1.000 Table 3
Descriptive statistics (mean) by country
AT BE CY DE EE ES FI FR GR IE IT LV MT NL PT SI SK Entities 10 3 3 17 2 5 3 24 7 2 18 2 3 5 4 2 4 Observations 127 47 25 226 10 80 48 323 103 31 274 18 32 70 52 14 33 Completeness 79.4 97.9 75.8 83.1 45.5 100 100 84.1 92 96.9 95.1 90 97 87.5 81.3 63.6 82.5 Lnlgr 2.406 3.423 6.773 1.571 25.14 6.885 5.746 3.645 6.678 0.698 6.590 7.837 4.169 -1.321 0.129 4.260 3.513 Ra/ta 0.835 0.817 0.812 0.827 0.732 0.938 0.861 0.899 0.850 0.926 0.917 0.909 0.861 0.845 0.921 0.902 0.880 Total Assets 5.317 2.968 3.766 2.148 4.196 3.638 1.617 2.388 5.177 1.398 1.118 2.166 4.876 3.638 4.437 4.886 6.586 ROE 5.715 5.429 0.100 1.169 -8.694 10.75 10.21 7.818 -1.543 5.501 4.101 1.812 12.61 10.64 1.635 -11.25 7.214 ROA 1.222 0.847 0.795 0.035 1.198 1.261 1.016 -0.558 0.513 0.947 0.312 0.856 1.122 0.742 0.248 0.906 DEBTEQ 8.219 7.345 1.024 11.14 0.585 7.040 8.233 5.055 3.284 6.662 5.290 3.548 1.808 5.109 11.75 4.442 2.797 TRR1 -1.192 -1.872 2.296 -1.640 -4.246 0.626 -2.570 -1.054 3.205 -1.540 -1.071 -2.477 -2.446 -0.368 1.278 -1.950 -4.31 Real ∆ GDP 1.354 1.381 1.486 1.167 3.185 1.497 1.138 1.121 -0.181 4.112 0.053 3.532 3.039 1.180 0.259 1.930 3.953 Regu Index 10.88 7.94 10.31 7.938 9 11.63 7.063 9.25 7.938 7.813 7.188 10.94 11.75 8.063 9.438 10.63 8.688
The table reports mean values of the variables included in the empirical analysis, decomposed by country. Entities describes the amount of banks included in the sample, and completeness describes the amount of available observations to the maximum amount of observations possible, expressed in percentage points.
15
Research Methodology
As stated before, this research aims to examine whether the interest rate has a negative effect on banks’ risk-taking behavior. If so, are there any indications that the ECB’s expansionary monetary policies are potentially fueling new financial imbalances in countries with an interest rate that is ‘too low’ given the underlying economic fundamentals. Moreover, it may be worthwhile to check for non-linearity in the relationship between the TRR and loan growth. In other words: is a bigger residual associated with bigger loan growth. Taking into account the aforementioned discussion, I formulate three research hypotheses:
Hypothesis 1: The TRR has a negative effect on banks’ loan growth, which would indicate that a low interest rate environment triggers banks to increase their risk appetite.
Hypothesis 2: After the introduction of a more stringent regulatory framework, the effect may have become less pronounced. Consequently, I expect to observe a less-negative coefficient after the implementation of the newly implemented regulatory requirements.
Hypothesis 3: The relationship between the TRR and loan growth is non-linear, where a larger residual increases the intensity of the coefficient.
The general empirical model to be estimated takes the following form:
r𝑖𝑡 = 𝑐 + 𝛽1 𝑇𝑅𝑅𝑖𝑡 + 𝛽2 𝑇𝑅𝑅𝑖𝑡∗ 𝐶𝑟𝑖𝑠𝑖𝑠𝑡+ 𝛽3 𝑇𝑅𝑅𝑖𝑡2 + 𝛽
4𝑀𝑅𝑂𝑡+ 𝛽5𝑏𝑖𝑡 + 𝛽6𝑐𝑡+ 𝜇𝑖𝑡 1)
16 Hypothesis 1: 𝐻0: 𝛽1 = 0; 𝐻1: 𝛽1 < 0
Hypothesis 2: 𝐻0: 𝛽2 = 0; 𝐻1: 𝛽2 > 0 Hypothesis 3: 𝐻0: 𝛽3 = 0; 𝐻1: 𝛽3 > 0
For completeness and robustness considerations, the data is regressed using two different risk variables. The first proxy for risk-taking, loan growth, reflects credit risk exposure faced by the bank, which usually comprises the primary source of risk. Nonetheless, a bank could also increase its risk by moving into a more subordinated security portfolio, by increasing its exposures to derivative instruments or by pursuing a less prudential risk management as a potential source of risk that is not covered within the loan portfolio. Therefore, the results are also examined using a second proxy for risk taking behavior, namely the ratio of risk assets. Regarding the relatively low variability in the risk assets among banks and its static development over time, as shown in table 1 and 3, I give preference to the loan growth as our primary risk-taking proxy.
Analyzing the effect of the interest rate on bank risk-taking raises three econometrical concerns, also frequently discussed in the related literature. The first concern relates to the endogeneity of the interest rate as it is often the main monetary policy tool used by the central bank to pursue their objective, e.g. price stability or economic stability. The interest rate therefore functions as a countercyclical monetary policy tool to weaken certain trends in the business cycle, potentially caused by banks’ behavior. As a remedy, this paper uses the country-specific TRR to describe the interest rate environment. Furthermore, one could argue that the ECB cannot take into account all the country-specific needs when formalizing the unique monetary policy, which even mitigates the potential endogeneity problem for the interest rate itself when studying the Eurozone countries.
17
potential unobserved heterogeneity. As the Random Effects model is a more efficient estimator, as it assumes the unobserved variables to be uncorrelated with the explanatory variables, I verify the preferred model using the Hausman Test for correlated coefficients.
Third, the existing literature provides several arguments that bank risk is persistent, which implies a deviation from the equilibrium in the short run. Keeley (1990) argues that fierce competition between banks encourages banks to increase asset risk and capital buffers. Especially before the implementation of the Basel requirements and the bail-in regime, the mechanism was invigorated by the fixed-rate deposit insurance system and the perception of moral hazard. Therefore, bank risk-taking is also argued to be persisted by regulation, as I have observed that deposit guarantees and capitalization requirements may have created or aggravated moral hazard, triggering banks to increase their risk appetite for a long period of time. In addition, it may take time before macro-economic policies have been fully anticipated by the banks, as their behavior seems to correspond quite closely to the business cycle. The interest rate environment could potentially have a lagged effect on bank-risk taking, even though the specifications and the dynamics of the empirical model are yet to be determined.
Empirical Results
18
Table 4
Taylor rule Residuals and bank risk-taking: Pooled OLS estimations
I loan growth II loan growth III loan growth IV loan growth V risk assets
TRR 1 -0.571*** (0.140) -0.049 (0.165) -0.454* (0.248) 0.037 (0.107) TRR 2 -0.094 (0.232) TRR_CRISIS 0.688** (0.315) 0.443 (0.272) -0.169 (0.136) MRO 2.934*** (0.598) 2.845*** (0.599) 3.295*** (0.622) 0.247 (0.257) ACQ 14.740*** (2.434) 14.68*** (2.431) 14.787*** (2.430) -0.193 (1.049) LNTOTAS -0.703** (0.303) -0.718** (0.302) -0.743** (0.304) -0.030 (0.130) ROA_1 0.932** (0.422) 1.021** (0.423) 1.135*** (0.421) -0.281 (0.182) DEBTEQ 0.020 (0.038) 0.029 (0.038) 0.028 (0.038) -0.002 (0.016) RGGDP 0.658*** (0.228) 0.617*** (0.228) 0.673*** (0.229) -0.055 (0.098) REGU -0.005 (0.248) 0.044 (0.249) 0.004 0.248) 0.036 (0.107) Obs 1,513 1,213 1,213 1,213 1,221 F-stat 16.61 16.51 15.25 15.21 0.50 P-value 0.000 0.000 0.000 0.000 0.87
*** Statistical significance at the 1% level ** Statistical significance at the 5% level * Statistical significance at the 10% level
Pooled OLS
The results obtained from the estimation of equation 1 using a pooled OLS are presented in table 4. Most control variables are significant for regressions I – IV, in which I use loan growth as the risk proxy, while the model becomes completely insignificant for the risk assets as a proxy for risk-taking behavior. Over the full time period, there seems to be no relationship between TRR and banks’ loan growth, as indicated by the insignificant TRR1 coefficient for regression II. However, regression III provides some evidence that the relationship is not consistent over time. Using an interaction dummy for the period after 2011 I find a negative relationship between the TRR and banks’ loan growth for the first period, while it turns positive in the period as of 2012. Though the latter seems counterintuitive, it may be related to the trend of increased regulatory stringencies implemented over time, predominantly as of 2012, forcing many banks to deleverage their lending activities in an indirect way to strengthen capitalization metrics.
19
with the Ramsey Reset test, an augmented Dickey-Fuller unit root rest, and the Breusch-Pagan test confirming the homoscedasticity of the standard errors. The Ramsey Reset test indicates that there are no signs for non-linearity in the relationship between TRR and bank risk-taking, as presumed in hypothesis 3. The Dickey-Fuller test, with the null hypothesis of having a unit root, is rejected for both dependent variables on a 1% confidence level, implying that both series are stationary. The main restriction of the pooled OLS estimation is that it does not acknowledge the presence of unobserved heterogeneity, meaning that banks may differ from each other in terms of underlying characteristics, not included in the model and potentially correlated with the control variables.
Table 5
Taylor rule residuals and loan growth: Fixed Effects and Random Effects estimations
I II III IV V VI - RE TRR1 -0.323** (0.155) -0.183 (0.221) -0.407 (0.299) -0.370 (0.302) -0.720*** (0.272) TRR2 -0.231 (0.318) TRR_CRISIS 0.671* (0.391) 0.432 (0.476) 0.323 (0.376) 0.601 (0.380) TRR_SQRD 0.036* (0.021) 0.023 (0.026) 0.002 (0.021) 0.033 (0.021) MRO 3.195* (1.796) 3.110* (1.799) 3.107* (1.799) 3.543** (1.810) -1.597 (1.635) ACQ 13.409*** (2.559) 13.555*** (2.557) 13.464*** (2.559) 13.597*** (2.560) 14.085*** (2.436) LNTOTAS 11.678*** (2.356) 11.538*** (2.358) 11.573*** (2.359) 11.393*** (2.378) -0.706** (0.302) ROA_1 1.027* (0.549) 1.038* (0.551) 1.083** (0.553) 0.962* (0.546) 1.101** (0.435) DEBTEQ 0.063 (0.041) 0.065 (0.041) 0.065 (0.041) 0.062 (0.041) 0.040 (0.038) RGGDP 0.670* (0.353) 0.471 (0.373) 0.541 (0381) 0.605* (0.362) 0.099 (0.328) REGU 0.468 (0.353) 0.558 (0.359) 0.534 (0.360) 0.489 (0.355) 0.078 (0.269) Obs 1,513 1,213 1,213 1,213 1,213 1,213 F-stat 9.43 9.01 9.01 8.65 8.52 P-Value U_i = 0 0.018 0.022 0.022 0.028 0.015
*** Statistical significance at the 1% level ** Statistical significance at the 5% level * Statistical significance at the 10% level
Fixed Effects Model
20
related to the control variable(s), making the FE model the preferred specification. All estimations in table 5 include entity- as well as year fixed effects.
In contrast to the β1 and β2 coefficients from the pooled OLS estimations (table 4) and RE
(table 5 V), the FE model fails to reject hypothesis 1 and 2 on a 5% significance level. The most complete specification is estimation IV, which indicates a negative relationship between TRR and banks loan growth, and a change in the relationship in the post-2011 period, though the coefficients are not is sufficiently large to allow for any inference. Furthermore, the very marginal β3 coefficient suggests that there is no non-linearity in the explanatory variable. The
null hypothesis 3 is therefore not rejected either. The conclusions remain the same when TRR2 is used as the banks’ risk-taking proxy, see estimation V. The results from estimation VI show a strong negative effect between TRR1 and loan growth, although the Hausman test has stressed the use of fixed effects instead of random effects.
In addition to the change in the key explanatory variables, there are noticeable differences in the coefficients for banks’ total assets and the interest rate term. Whilst the pooled OLS and RE estimations show a moderate negative relationship between a bank’s size and loan growth, the relationship turns strongly positive in the FE estimations. Given the substantial change between the RE (regression V) and FE output, which both account for unobserved heterogeneity, this effect is likely to be attributable to an omitted variable that is correlated with banks’ total asset base.
21
The other big difference between the FE and RE estimations is the monetary policy rate, as proxied by the MRO. Similar to the pooled OLS regressions, the coefficient takes a positive value for the FE estimations, which exemplifies the potential endogeneity of the MRO as a countercyclical monetary policy tool. In times of fast economic expansions, the central bank may be inclined to increase the interest rate, in an effort to slow down the economy. As a result, I observe a positive value for MRO, suggesting that a higher interest rate triggers extra lending operations from the banks. In the RE model, the MRO has an insignificant negative value. The mismatch between the outputs indicates the presence of an unobserved variable, correlated with MRO and impacting banks’ lending decisions. Another take away from table 5 is the insignificance of the GDP coefficient as a country specific control variables, which seems to be obviated by the time dummies.
Table 6
Taylor rule residuals and loan growth: Fixed Effects and persistence of bank risk-taking
I II III IV V - TRR2 VI - RE TRR -0.265 (0.447) -0.572 (0.430) -0.713* (0.406) -1.060*** (0.339) -0.194 (0.339) -0.995*** (0.340) TRR_1 -0.127 (0.403) 0.471 (0.528) 0.533 (0.512) 0.655 (0.506) -0.281 (0.393) 0.468 (0.511) TRR_2 -0.681* (0.369) -0.475 (0.496) -0.342 (0.492) 0.060 (0.376) -0.386 (0.500) TRR_3 -0.304 (0.345) -0.333 (0.345) -0.554* (0.298) -0.313 (0.339) TRR_CRISIS 0.712* (0.403) 1.020*** (0.383) 1.082*** (0.367) 1.077*** (0.360) 0.797** (0.314) 1.151*** (0.302) MRO 3.045* (1.806) 5.047** (2.290) 3.809* (2.185) ACQ 13.572*** (2.558) 13.606*** (2.449) 13.507*** (2.378) 13.482*** (2.382) 13.646*** (2.380) 14.964*** (2.259) LNTOTAS 11.530*** (2.359) 9.884*** (2.310) 11.922*** (2.305) 11.346*** (2.293) 11.244*** (2.326) -0.512* (0.282) ROA_1 1.037* (0.551) 1.339** (0.517) 1.405*** (0.493) 1.528*** (0.490) 1.295*** (0.491) 1.354*** (0.380) DEBTEQ 0.066 (0.041) 0.071* (0.037) 0.073** (0.035) 0.074** (0.035) 0.073** (0.035) 0.042 (0.034) RGGDP 0.550 (0.416) 0.510 (0.397) 0.556 (0.381) 0.815** (0.353) REGU 0.544 (0.361) 0.593 (0.343) 0.468 (0.335) Obs 1,213 1,138 1,062 1,062 1,062 1,062 F-stat 8.62 10.13 11.72 12.56 12.02 TRR F-Test 0.361 0.115 0.069 0.001 0.205 0.018 P-Value U_i = 0 0.023 0.003 0.000 0.000 0.000
*** Statistical significance at the 1% level ** Statistical significance at the 5% level * Statistical significance at the 10% level
Persistence of Risk-Taking Behavior
22
power to the relationship between the interest rate environment and bank risk-taking. Regression II shows a negative coefficient between the TRR of two periods ago, although the combined TRR effect is not significant, as shown by the TRR F-test statistic. Adding a third period lagged term increases the combined TRR significance to 6.9%, see estimation III. Dropping the potentially endogenous MRO variable and the country specific country variables in estimation IV gives the most explanatory power, both in terms TRR significance and the estimation-wide F-statistic. The results show that banks realize higher abnormal loan growth in a low interest rate environment, though this relationship has a structural breakdown in the post-2011 period, both effects being significant at a 1% confidence level. As a result, I reject the null hypothesis for hypothesis 1 as well as hypothesis 2 using α = 0.05. There should remain cautiousness on the economic inference of the aforementioned conclusions, given that the results do not hold for the secondary TRR or for the share of risk assets as a risk proxy. Regression V uses TRR2, and its lagged terms as the key explanatory variable and indicates that there is no strong significant relationship between the interest rate environment and abnormal loan growth, even though the crisis coefficient shows an upward change for the post-2011 period. The parameters in the RE estimation VI are very similar to the FE outcome, and provide evidence that the TRR is negatively related to loan growth, though the coherence breaks down in the post-2011 period.
23
Aggregating the results from the three different regression sets, I have obtained mixed findings for the relationship between the interest rate environment and bank risk-taking. Furthermore, the results provide moderate support for the use of TRR and abnormal loan growth as proxies for the interest rate environment and bank risk-taking respectively. Even though the heterogeneity of the regression outcomes strongly limits the potential economic inference from this paper, one can conclude that there seems to be a clear difference in terms of risk-taking dynamics in the pre- versus post-crisis period. As there appears to be a negative relationship between the interest rate and bank risk appetite in the build-up towards the financial crisis, this effects is completely unobserved, if not inverted, for the period after 2011. One would expect that this is the result of the more stringent regulatory framework, which simply impedes the banks for accumulation more risky exposures. This rationale is partly depicted in the results, once I account for the persistence of bank risk-taking, even though the inclusion of year effects and the inclusion of a regulatory index are not sufficient to capture the exact dynamics originating from the regulatory framework. Nonetheless, the results indicate that the current expansionary policies do not seem to translate into additional risk taking in the banking system, at least not in terms excessive credit growth or the relaxation of underwriting standards.
Discussion and Limitations
A valid question that should be raised after observing the results is why the relationship between TRR1 and loan growth seems to disappear, if not turn positive, for the period after the financial crisis in Europe, which is considered as of 2012. The first proposition is that this change in risk behavior has an intrinsic nature, meaning that banks have started to behave more responsively when it comes to risk-taking. Banks may have learned from their excessive risk-taking behavior before the crisis and could have repositioned their role in society into a more responsible and sustained way. Or it could be that the banks have adopted more sophisticated risk models that enables them to model- as well as retain their risk exposure more accurately.
24
activities, or accumulate additional capital in order to comply with the new regulations. With the introduction of minimum leverage ratio, liquidity requirements, and loss-absorbing requirements by 2018 and 2019 respectively, regulatory pressures are expected to intensify even further, especially for the banks classified as globally systemically important institutions (G-SIIs). The growing list of regulations is expected to increasingly restrain the ‘playing field’ of the bank. The regulatory trend described in the above is partly accounted for in the model, with the use of year fixed effects and a regulatory index. However, the fact that the index has relatively little variation and the fixed effects are yearly time dummies, not a time trend, does not rule out the regulatory framework as an explanation for the heterogeneous relationship between the TRR and abnormal loan growth over time.
Another scenario is that banks might be accumulating exposures via risk-channels not observed in the risk proxies used in this paper. While the risk exposures in the build-up to the previous financial crisis were predominantly observable in the volumes and quality of the loan portfolio, this does not necessarily apply to the next crisis. Historical events have taught us that every financial crisis is unique in its origination, which makes it unlikely that we would experience a similar bubble as in 2008, mainly driven by securitization activities. Regarding the active supervision and stress tests that the banking sector is subjected to nowadays, it is deemed implausible that banks have the freedom to take excessive risks outside the scope of the financial regulators.
25
Policy Implications
In line with the existing literature, this paper provides additional evidence that the interest rate environment does have an effect on banks’ risk appetite. Observing this relationship has relevant implications for monetary policy makers, as it stresses the importance of having an appropriate and dynamic monetary policy framework in an effort to maintain financial stability within the financial sector and the economy as a whole. The Taylor rule has been a proposed guidance for central banks and financial regulators in determining monetary policies, because it comprises the fundamental macro-economic indicators and it has proven to be a successful tool in pursuing short-term stabilization of the economy, while still maintaining long-term growth. In addition to being a benchmark for monetary policy, the Taylor rule is found to be a key determinant for the risk-taking behavior of the financial sector. Therefore, this paper advocates the adherence to the Taylor rule in formalizing the appropriate interest rate environment for policy makers, anno 2017.
Following the Taylor rule is not as straightforward within the European Monetary Union, because the heterogeneous macro-economic parameters for the member states. Germany and Spain form a good example to illustrate the difficulty of having a uniform interest rate in the Eurozone. At 31.03.2006, the Taylor rule residual, according to Bloomberg, stood at 3.5% for Germany and -7.5% for Spain, which means that the interest rate in the Eurozone was 3.5% too high for Germany and 7.5% too low for Spain according to the Taylor rule estimation. At YE2016, the relationship had completely inverted, as the TRRs stood at -7% and 0.95% respectively. In other words, it is difficult to find an appropriate compromise for the countries’ individual needs. However, one could still conclude that the interest rate is too low regarding the Eurozone weighted average TRR of -1.5% at YE2016.
26
Conclusion
This paper studies the relationship between the interest rate environment and bank risk-taking behavior. Using a dataset of 114 banks within the European Monetary Union for the time period between 2000 and 2016, I find a negative relationship between the Taylor rule residual and banks’ excessive loan growth, though the relationship is not consistent for all specifications of the model. Our findings are in line with the existing literature on this topic (e.g. Altunbas et al., 2010; Delis and Kouretas, 2011; Maddaloni and Peydro, 2011 find a negative relationship relation between monetary policy and bank risk-taking). The innovative feature of this paper is the usage of Taylor rule residuals in combination with risk proxies that describe the actual risk taking, and not the risk exposure that is related to the market conditions. Applying the Taylor rule residuals on the Euro-area countries has provided a natural experiment when it comes to potential endogeneity of the interest rate.
Furthermore, the results point out that the relationship contains a structural break for the post-2011 period, which is argued to be attributable to the regulatory stringencies, put in place with the implementation of the new Basel frameworks and the CRD IV directorate. The positive coefficient for the post-2011 period forms an indication that the current expansionary monetary policies do not seem to trigger additional bank risk-taking.
Despite the intuitive findings of this paper, one should remain cautious when it comes to the economic inference of the results. First of all, the dataset comprises a rather limited set of banks, because I was forced to use a non-preferred source. Using a better dataset, with preferably more banks, should reduce the internationalization and survivorship bias as discussed before, and strengthen the explanatory power of the regressions. Besides, the fixed effects estimations suggests that there might be endogeneity in the control variables, observed in the total assets term, and potentially causing biased results for the Taylor rule residuals and banks’ loan growth.
Bibliography
Altunbas, Y., Gambacorta, L., Marqués-Ibánez, D., 2010. Does monetary policy affect bank risk-taking. Unpublished working paper. European Central Bank, Frankfurt am Main.
27
Andries, A., Cocris, V., Plescau, I., 2015. Low interest rates and bank risk-taking: has the crisis changed anything? Evidence from the Eurozone. Review of Economic & Business Studies 8, 125-148.
Barth, J., Caprio, G., Levine, R., 2001. The regulation and supervision of banks around the world – a new database. Policy Research Working Paper No. 2588, The World Bank.
Barth, J., Caprio, G., Levine, R., 2008. Bank regulations are changing: but for better or worse? Policy Research Working Paper No. 4646, The World Bank.
Barth, J., Caprio, G., Levine, R., 2012. The evolution and impact of bank regulations. Policy Research Working Paper No. 6288, The World Bank.
Buch, C., Eickmeier, S., Prieto, E., 2011. In search for yield? Survey-based evidence on bank risk taking. Unpublished working paper. Deutsche Bundesbank, Frankfurt am Main.
Campbell, J., Cochrane, J., 1999. By force of habit: a consumption-based explanation of aggregate stock market behavior. The Journal of Political Economy 107, 205-251.
Delis, M., Kouretas, G., 2011. Interest rates and bank risk-taking. Journal of Banking & Finance 34, 840-855.
Dell’ Ariccia, G., Laeven, L., Marquez, R., 2014. Real interest rates, leverage, and bank risk-taking. Journal of Economic Theory 149, 65-99.
Dell’ Ariccia, G., Laeven, L., Marquez, R., 2017. Bank leverage and monetary policy’s risk-taking channel: evidence from the United States. The Journal of Finance 72, 613-654.
Foos, D., Weber, M., Norden, L., 2009. Loan growth and riskiness of banks. Journal of Banking and Finance 34, 2929-2940.
Hausman, J., 1978. Specification tests in econometrics. Econometrica. 46: 1251-1271
Kahn, G., 2010. Taylor rule deviations and financial imbalances. Unpublished working paper. Federal Reserve Bank of Kansas City, Kansas City.
28
Maddaloni, A., Peydró, J., 2011. Bank risk-taking, securitization, supervision, and low interest rates: evidence from the Euro-area and the U.S. lending standards. The Review of Financial Studies 24, 2121-2165.
Okun, A., 1962. Potential GNP: its measurement and significance. In: Alexandria, VA: American Statistical Association. pp. 89 – 104.
Rajan, R., 2005. Has financial development made the world riskier? Unpublished working paper. National Bureau of Economic Research, Cambridge.
Databases
Bloomberg Terminal. Download Taylor rule residuals and consolidated banking sector figures. Accessed at European Investment Bank, Luxembourg [Accessed 16 February 2017]. DataStream. Download balance sheet items and profitability metrics. Accessed at University of Groningen [Accessed 17 March 2017].
European Banking Authority (EBA). Download unconsolidated banking sector figures. Available at: http://www.eba.europa.eu/risk-analysis-and-data/risk-dashboard [Accessed at 7 May 2017].
Eurostat. Macroeconomic variables, e.g. HCIP and GDP. Available at:
http://ec.europa.eu/eurostat/web/hicp/data [Accessed 25 March 2017].
Financial Stability Reports of Austria, Belgium, Cyprus, Estonia, France, Germany, Italy, Latvia, Malta, Netherlands, Portugal, Slovakia, Slovenia and Spain.
Orbis Bank Focus. Quantifying internationalization bias. Accessed at the European Investment Bank, Luxembourg [Accessed 9 May 2017].
The World Bank. Construction of regulatory index. Available at:
http://econ.worldbank.org/WBSITE/EXTERNAL/EXTDEC/EXTRESEARCH/0,,contentMD K:20345037~pagePK:64214825~piPK:64214943~theSitePK:469382,00.html [Accessed 2 April 2017].
Zephyr. Download acquisition data. Accessed at the European Investment Bank, Luxembourg [Accessed at 20 May 2017].
Appendix
Appendix A. Variables description
29
account is defined as gross loans minus the accumulated account for loan loss provisions (allowance for loan and lease losses). Loan growth has been suggested as one of the key drivers of bank risk taking by several papers (e.g. Foos et al., Amador et al.) and seems to be more applicable than the proxies mentioned before in at least two ways: loan growth should give a better reflection of the additional risk taken on by a bank, as it is less subjected to risk exposure. In addition, the risk taking behavior is observed right away, while most other proxies have to include a dynamic model before risk appetite starts to reveal. This is especially the case for NPLs and loan loss provisions, which are heavily subjected to the macroeconomic environment.
It is important to realize that a bank’s decision to expand its loan portfolio can be demand and supply driven. Demand-side driven lending growth does not result from additional risk taking, but reflects the improved economic conditions, which incentivizes the retail- and corporate sector to increase their borrowings. Supply driven lending growth derives from the change in banks’ underwriting standards. As the lending standards of banks are usually not published and hard to quantify, I aim to measure this behavior implicitly by examining the growth in banks’ loan portfolio.
30
Secondly, I use the ratio of risk assets to total assets as a proxy for bank-risk taking behavior. This ratio is in line with the approach of Delis and Kouretas (2011), though this research does not exclude government securities from the risk assets, as sovereign bonds do carry a positive amount of risk, which differs between countries. After the formation of the Eurozone, there has been a long-lasting convergence of the sovereign yields, driven by the market perception that country specific default risk had disappeared. Soon after the onset of the financial crisis it became clear that the heterogeneity among countries had lasted, as the underlying risks turned out to be heavily underestimated. Consequently, many investors fled into northern-European sovereign debt, while yields of the GIIPS5 countries spiked to unsustainably high rates. Therefore, I consider sovereign debt as a potential way of risk-taking behavior, even though regulators currently apply a zero-risk weight to sovereign exposures. The ultimate solution to acknowledge risk behavior from government debt would be to quantify the relative exposure to every country for each bank-entity and applying positive weights for these exposures based on country-specific default probabilities. However, this is beyond the scope of this research. Nonetheless, several European regulatory bodies have made proposals to apply positive risk weights on sovereign debt based on a probability of default model.
Similar to the loan growth, the share of risk assets features the nice property to reflect risk appetite at the time the additional risk is taken by the bank. Furthermore, this share of risk assets is not affected by risk exposure resulting from macroeconomic shocks. During a time of financial distress, a bank would generally face losses on several balance sheet items, not related to the classification as a risk- or non-risk asset. I use both proxies in our regressions to check if the results are consistent.
Using the central banks’ key interest rate as our main explanatory variable would create a significant endogeneity bias, since the interest rate is used as the primary monetary policy tool to maintain price stability or pursuing other macroeconomic objectives. Changes in the interest rate often have a countercyclical motive, which potentially leads to reverse causation in our regression model, showed by an inverse coefficient. In periods of high loan growth, implying good economic performance, the central bank may be inclined to increase the interest rate and vice versa in the case of loan portfolio contractions. The regression model
5
31
would show correlation between the main explanatory variable and the error term, making the results biased. Other papers have tried to impose a lagged relationship between the interest rate and the risk-taking proxy, but this approach is rather subjective in our view. In an effort to correct for the endogeneity, I use the Taylor rule residual as the main independent variable, which is defined as the key interest rate imposed by the central bank minus the prescribed interest rate, determined by an inflation- and an output gap component. Regarding consistency and robustness of the results, I generate two Taylor rule residuals. See Appendix B for an elaborate explanation on how the residuals have been obtained.
The Eurozone offers a natural experiment in examining the effect of the TRR on bank risk-taking, since the same interest rate applies to all Euro members (in the form of the rate on Main Refinancing Operations (MRO) set by the European Central Bank), while the prescribed interest rate differs among the countries. The approach of using the TRR to describe the interest rate environment has been widely used in the existing literature (e.g. Altunbas et al., Andries et al., Dell’ Ariccia et al., Maddaloni and Peydró) and is argued to be an adequate remedy for the endogeneity of the interest rate.
Alongside the net loan portfolio, risk assets and total assets, I have obtained net loan losses, NPLs to gross loans, net interest margin, return on equity, return on assets, and debt to equity ratio from Datastream for all entities over the defined time period. I control for these bank-specific characteristics, as the existing literature has provided enough evidence to conclude that capitalization and profitability are of explanatory power in bank risk equations. Stock values have been adjusted for inflation, using the HICP index obtained from Eurostat. Growth rates and performance metrics have been manually converted towards logarithmic growth rates to improve the accurateness of the coefficient inferences. The MRO- and TRR data have been obtained from Bloomberg, see Appendix B.
32 -5 0 0 50 1 0 0 L o a n G ro w th -10 -5 0 5 10 TRR2 lnlgr Fitted values .2 .4 .6 .8 1 R is k As s e ts -10 -5 0 5 10 TRR1
rata Fitted values
.2 .4 .6 .8 1 R is k As s e ts -10 -5 0 5 10 TRR2
rata Fitted values
-5 0 0 50 1 0 0 L o a n G ro w th -10 -5 0 5 10 TRR1 lnlgr Fitted values
Figures 1 – 4: Bank interest rate environment and risk-taking. The figures report the pooled regression between bank risk-taking, measured by the loan growth for figure 1-2 and the share of risk assets to total assets for figure 3-4, and the bank interest rate environment, reflected by the TRR1 for figure 1 & 3 and TRR2 for figure 2 & 4. The trend lines provide rather mixed results for the estimated relationship. Figures 1 and 2 provide a negative correlation between the interest rate and bank risk-taking, which is in line with our hypothesis. Yet figures 3 and 4 provide an inverted picture, indicating a positive correlation.
Source: Stata
Figure 1: Loan Growth and TRR1 Figure 2: Loan Growth and TRR2
33 Appendix B. Explanation and construction of Taylor rule residuals
As stated before, this paper uses the Taylor rule residual to describe the interest rate environment within a country. Let us first derive the general form of the Taylor rule:
𝑖𝑡 = 𝑟 + 𝜋𝑡+ 𝛼 (𝜋𝑡− 𝜋∗) + 𝛽 (𝑦
𝑡− 𝑦𝑡∗) 2)
Where 𝑖𝑡 is the prescribed value of the policy interest rate, 𝑟 the natural real interest rate, 𝜋𝑡 is the actual inflation, 𝜋𝑡− 𝜋∗ the deviation of the actual inflation rate from the target in period
t and 𝑦𝑡− 𝑦𝑡∗ is the deviation of actual real output 𝑦𝑡 from the potential output 𝑦𝑡∗ in period t. The variables α and β are positive numbers that comprise the weight that is given to inflation- and output gap component, which are both assumed to be 0.5 according to the conventional Taylor rule estimation.
Using Okun’s law, we can convert the output gap into an unemployment gap: 𝑢𝑡− 𝑢𝑡∗ = 𝜃 (𝑦
𝑡− 𝑦𝑡∗) + 𝜀𝑡 , 𝜃 < 0 3)
Where 𝑢𝑡− 𝑢𝑡∗ is the deviation of the actual unemployment rate from the natural level of
unemployment. The variable 𝜃 is a conversion factor, which is assumed to be equal to -0.5. Rewriting equation 2 results in the following:
𝑦𝑡− 𝑦𝑡∗ = (𝑢𝑡−𝑢𝑡∗) 𝜃 + 𝜀𝑡 4) 𝑦𝑡− 𝑦𝑡∗ = (𝑢𝑡 ∗−𝑢 𝑡) −𝜃 + 𝜀𝑡 5)
Substituting equation 3 into the general form of the Taylor rule, we find the following relationship between the prescribed interest rate dependent on output and unemployment: 𝑖𝑡 = 𝑟 + 𝜋𝑡+ 𝛼 (𝜋𝑡− 𝜋∗) + 𝛽 (𝑢𝑡
∗−𝑢 𝑡)
−𝜃 6)
Bloomberg defines the natural level of unemployment as the ‘Non-accelerating inflation rate of unemployment’ (NAIRU), which is a specific level of unemployment in an economy that would not trigger any changes in inflation. The NAIRU is considered as the equilibrium between the state of the economy and the labor market and is also referred to as the ‘long-run Philips curve’.
34
unemployment, while the Taylor rule estimations from Bloomberg assumes persistence of differences in NAIRU among countries. Therefore this paper also tests the model based on Taylor rule estimations that have not assumed heterogeneity among countries and consider the NAIRU to be universal at 5%. This level of natural unemployment is widely used in the economic literature and should be a more objective proxy. Yearly seasonally adjusted unemployment- and inflation numbers, extracted from Eurostat, are used to estimate country-specific Taylor rule estimations based on homogeneous natural levels of unemployment.
Appendix C. Construction of the Regulatory Index
The bank regulation and supervision survey, provided by Barth et al. (2001, 2008 and 2012, World Bank), is used to construct a regulatory index, which captures regulatory- and capital stringencies over time, within a country. Out of the 270 questions in the initial survey, I have selected the following 20 questions, provided in table 7, which are perceived the most relevant regarding regulatory stringencies and could be answered by either yes or no:
Table 7: Selected questions
1) Is it legally required that applicants submit information on the source of the funds to be used as capital? 2) Are the sources of funds to be used as capital verified by the regulatory/supervisory authorities? 6 3) Does the minimum ratio vary as a function of an individual bank’s credit risk?
4) Does the minimum ratio vary as a function of an individual bank’s market risk? 5) Is subordinated debt required as part of capital? 6
Applied to question 6,7 and 8: Before minimum capital adequacy is determined, which of the following are deducted from the book value of capital?
6) • Market value of loan losses not realized in accounting books 7) • Unrealized losses in securities portfolios
8) • Unrealized foreign exchange losses
9) Are accounting practices for banks in accordance with International Accounting Standards (IAS)? 10) Are there explicit, verifiable, and quantifiable guidelines regarding asset diversification? 11) Are banks required to hold either liquidity reserves or any deposits at the Central Bank?
12) If a customer has multiple loans and one loan is classified as non-performing, are the other loans automatically classified as non-performing?
13) Are off-balance sheet items disclosed to the public?
14) Must banks disclose their risk-management procedures to the public? 15) Do regulations require external credit ratings for commercial banks? 16) Is there a single financial supervisory agency for the financial sector? 6
17) If an infraction of any prudential regulation is found by a supervisor, must it be reported? 6
18) Can the initial disbursement or subsequent injections of capital be done with assets other than cash or government securities?
19) Can initial disbursement of capital be done with borrowed funds?
20) Does accrued, though unpaid, interest/principal enter the income statement while the loan is still non-performing?
The country-specific regulatory index is determined by adding 1 if the answer is yes to question 1-17 and 0 if the answer is no. The opposite occurs for question 18-20, where I subtract 1 if the answer is yes and 0 otherwise. Theoretically, the regulatory index can take a value between -3 and 17, where a higher value reflects stricter regulations and better
6
35
enforcement regarding capital requirements and related affairs. The bank regulation and supervision surveys were carried out in 2003, 2007 and 2011 based on data of the respective years and published in the year after it was composed. As question 6, 8 and 16 were not examined in the last survey, I have used the country-specific score in 2007 for these inputs to maintain consistency and comparability.
The regulatory framework has become more universal among Eurozone countries with the newly created European Banking Union (EBU). The EBU was implemented by all Euro area countries in November 2014 and is currently based on two pillars: the Single Supervisory Mechanism (SSM) and the Single Resolution Mechanism (SRM), while a third pillar (the European Deposit Insurance Schemes) has been proposed by the European Commission. Furthermore, the pillars are based on a common legislative foundation, the Single Rulebook, which includes the Banking Resolution and Recovery Directive (BRRD).
The SSM regulation organizes the prudential supervision of all credit institutions in the participating Member States, implying that all participating countries face a common financial supervisory agency for the financial sector. The ECB directly supervises ‘significant’ banks, whereas the national competent authorities supervise less significant ones. The SRM regulation complements the SSM in managing the failure and the resolution of credit institutions with minimal costs to taxpayers. Resolutions are implemented with the financial resources from the Single Resolution Fund (SRF), which are collected from the contributions of the banking industry. The SRF can only contribute to the resolution of a bank if at least 8% of the total liabilities of the respective bank have been bailed-in.
36 Appendix D: Consolidated versus Unconsolidated Results
In an effort to quantify the size of the internationalization bias that is created by the dataset, I examine the country- and sample specific differences between the consolidated and unconsolidated results for total assets. The results are provided in table 8.
Table 8: Consolidated and unconsolidated total assets at YE2016, in EUR
Country Unconsolidated (m) Consolidated (m) Country bias Sample Bias
AT 598,800 711,141 84% 62% BE 1,016,000 1,013,011 100% 69% CY 42,500 72,568 59% 97% DE 4,116,100 6,399,575 64% 80% EE 16,500 17,973 92% 95% ES 3,307,300 2,727,804 121% 50% FI 415,400 489,875 85% 70% FR 6,918,800 7,488,736 92% 60% GR 288,600 209,442 138% 92% IE 301,300 952,374 32% 73% IT 2,277,500 3,334,424 68% 65% LT 12,700 14,655 87% 99% MT 19,300 41,804 46% 65% NL 2,114,200 2,174,872 97% 81% PT 284,000 421,278 67% 93% SI 22,500 29,924 75% 91% SK 40,200 50,042 80% 119%
37
