Christoph Reischmann S4086090
c.h.reischmann@student.rug.nl
Double Degree M.Sc. Economic Development and Globalization Rijksuniversiteit Groningen | Georg-August Universität Göttingen
Faculty of Economics and Business
Supervisor: Dr. A.C. Steiner
Co-assessor: Prof. Dr. Tino Berger Date: June 16, 2020
1
Abstract
Low interest rates have been named as one of the factors that contributed to the build-up of risks in the period before the global financial crisis. The relationship between loose monetary policy and increased bank risk-taking has since been subject to intense empirical
investigation and an additional transmission mechanism of monetary policy, the bank risk-taking channel, has been established. The prolonged period of ultra-loose monetary policy in the euro zone has once again raised concerns about the relationship of monetary policy and bank risk-taking. This thesis provides additional empirical evidence for the existence of the risk-taking channel in the Euro zone. Using bank survey data on the country level, it analyses the effect of a loose monetary policy stance on the change in lending standards of European banks between 2003Q1 and 2017Q4 using a difference GMM estimator. While a general effect is found providing support for the existence of the bank risk-taking channel during the whole period, no such effect is found for the period after 2012. Furthermore, the existence of potential non-linearities in the relationship of monetary policy and bank risk-taking are examined. Additionally, a differential effect of conventional interest rate policy and the unconventional measures adopted by the ECB on bank risk-taking is considered. Keywords: bank risk-taking, lending standards, monetary policy
Statutory Declaration
2
Table of Contents
1.Introduction ... 3
2. Literature review ... 5
3. Data and econometric methodology ... 10
3.1 Data ... 10
3.1.1 Dependent variables ... 10
3.1.2 Independent variables ... 11
3.2 Econometric methodology ... 14
4. Results ... 16
4.1 The impact of monetary policy on bank lending standards ... 17
4.2 The effect of monetary policy on bank risk-taking for different time periods .... 21
4.3 Non-linear effects of monetary policy on bank risk-taking ... 26
5. Conclusion ... 28
6. References ... 30
3
1.Introduction
The channels through which monetary policy transmits to the real economy has been of key interest for researchers for a long time. One particular channel which had received relatively little attention, however, was the effect of monetary policy on the risk-taking behaviour of banks. Since the global financial crisis, this has changed. In the build-up of the crisis, interest rates were quite low which made them the subject of being one of the contributing causes of the crisis (Claessens, 2013, Ahrend 2008). In particular Taylor (2009) was quick to point to policy rates being below those
suggested by historical evidence (i.e. deviations from a Taylor rule) as a key
explanation of the global financial crisis (Taylor, 2009). In following, the relationship between monetary policy and bank risk-taking were subject to intense academic debate. This gained further strength due to the sustained ultra-low interest rates as well as the adoption of unprecedented easing measures by central banks in the past years.
Borio and Zhu (2012) were the first to coin the term “risk-taking channel”.
(Neuenkirch, 2018) The risk-taking channel is hereby the change in behaviour of banks concerning their attitude and assessment of risk induced by a change in monetary policy. (Brana, 2019) It can thus be considered to be part of the credit channel of monetary policy and hence amplifies other transmission channels rather than being a separate channel (Bernanke, 1995). The two main explanations through which such a channel can be rationalized are the effect of monetary policy on the risk perception of banks on the one hand and on the incentives to take on existing risks on the other. In the first channel, eased monetary policy affects the ability of banks to take on more risk by easing constraints on bank lending imposed by risk measures. In the second channel, monetary easing strengthens the incentives of bankers to take on more risk. Empirical evidence for the existence of the bank risk-taking channel is growing. An easing stance of monetary policy is associated with higher bank risk-measures such as the expected default frequency or the z-score (Altunbas, 2014, Andries 2015, Brana, 2019), increases in the riskiness of new loans (Buch, 2014, Jimenez, 2014), decreasing lending standards (Maddaloni, 2010, 2013, Neuenkirch, 2018), negative effects on banks’ net interest margins and, to a lesser extent, bank profitability (Claessens, 2017, Borio, 2017, Butsch, 2015, Cruz-Garcia, 2019) and possibly even increases in systemic risk (Colletaz, 2018).
One fundamental difficulty hereby is the identification and measurement of bank risk-taking. Overall credit measures are not appropriate as they typically do not allow to make inferences about borrower quality or the pricing of risk by banks. Some studies have tried to overcome this problem by using extensive datasets available in certain countries that allowed for the discrimination of different types of loans (e.g. Jimenez, 2014) Others have focused on bank risk-indicators such as the z-score or the
distance to default (e.g. Brana, 2019).
4
zone in which banks indicated whether they have tightened or loosened their credit standards in the last three months. This allows me to take on a cross-country macro approach rather than being focused on a certain country. Using survey data
additionally brings the advantage of being able to observe the changes in bank behaviour not only concerning the loans that were given but also the ones that were not. The BLS of the ECB allows furthermore for a better identification of the bank risk-taking channel as banks not only indicate whether they changed lending standards but also why they did so. This allows me to differentiate between direct effects on banks’ balance sheets and indirect effects through their perception of risk due to changes in the economy. I will then use a dynamic panel analysis to determine the influence of monetary policy on the change in lending standards. In this approach, I follow studies such as Buch (2014) for the US. and Maddaloni (2010, 2013) for the Euro zone. My analysis will however represent an extension of their analysis on several fronts. Firstly, by analysing the period 2003q1-2017q4, I am able to provide more robust results by adding up to six more years to the analysis. Furthermore, this allows me to examine the relationship of monetary policy and bank risk-taking not only as an explanation of the build-up of the crisis and central banks’ ability to soften lending constraints during the crisis but also to look at the relationship after the crisis. The period after the crisis is in itself interesting for several reasons. The first one is the adoption of the ECB of unconventional monetary policy measures, characterized by a negative target rate and massive expansions of its balance sheet. Thus, as a second extension, I will additionally employ the shadow short rate by Krippner (2016) besides more conventional measures of monetary policy such as EONIA and Taylor rule residuals to explore whether such measures changed the relationship between monetary policy and bank risk-taking. Additionally, the period during and after the crisis was characterized by the adoption of macroprudential policies aimed at curtailing the excessive taking of risks by banks. I will assess whether such policies were associated with tightening lending standards and weakened the impact of monetary policy on these lending standards. As a last extension, while Maddaloni (2013) considered the change in lending standards due to bank balance sheet factors, they neglected the changes due to risk perceptions. I will provide evidence for this. The third extension is the consideration of non-linear effects. The banks’ change in risk-taking as a response to a change in monetary policy might differ depending on the general interest rate environment. This is motivated mostly by the non-linear effects interest rates appear to have on bank profitability (Claessens, 2017, Borio, 2017).
5
2. Literature review
The risk-taking channel of monetary policy was first termed as such by Borio and Zhu (2012) and is defined as the way in which changes in monetary policy affect the perception, management and pricing of risks by banks (Borio, 2012). 1More
specifically, banks can change their asset and their liability side in order to adjust their risk-taking. On their asset side, ex-ante banks can take on more risks by
granting loans for a longer period of time (liquidity risk) or to riskier borrowers (credit risk). Ex-post, banks can increase risk-taking by decreasing costly monitoring efforts. On their liabilities side, banks can take on risk by increasing leverage and by using more short-term funding. 2
There are two main channels through which monetary policy can affect bank-risk taking.
The first channel, as motivated by Borio et al. (2012) affects banks risk perceptions through the effect of monetary policy on valuations, incomes and cash flows. The fundamental idea is that measures of risk of banks and hence their risk perception are influenced by changes in monetary policy.
Adrian (2010) propose a model in which banks are risk neutral but face a value at risk constraint. Subject to this value at risk (VaR) constraint, the bank must hold a certain amount of equity to cover its losses. They now argue that if monetary policy is able to influence the net interest margin of banks and hence the banks’ profitability, it will increase the banks’ capital measures if they are forward looking. As a result, banks can increase their holdings of risky securities which lowers the price for risk in the market. (Adrian, 2010) The impact of monetary policy on the net interest margin can be rationalized by considering a situation in which banks face a constant or sticky long-term lending rate and a short-term funding rate which is determined by monetary policy. The rationale is that conventional monetary policy is mostly able to affect short-term interest rates and hence the banks funding side but not long-term interest rates. As a result, a change in the central bank policy rate directly affects the net interest margin which is defined as the banks interest income on its assets minus the interest paid on its liabilities. Thus, a lower short-term interest rate decreases the banks notational liabilities while leaving its notational assets unchanged (Adrian, 2019). Additionally, even if one considers an equilibrium model in which monetary policy only shifts the yield curve without altering its slope, such an effect might arise if
1 To avoid confusion, it must be noted that the risk-taking channel as such is not a separate transmission channel but part of both the overall bank lending channel (Disyatat, 2010, Gambacorta, 2011) and the balance sheet channel. It can be seen as an extension of both channels considering the effects of risk considerations in the supply of credit by banks (Köhler-Ulbrich, 2016).
6
bank funding rates reach the zero lower bound or if monetary policy is able to influence non-interest income and provisions (Borio, 2017, Demiralp 2019). 3
It must be noted that such an effect can also be motivated on the demand side.4 If
the net worth of borrowers increase due to an expansionary monetary policy shock, this will lower the external finance premium due to the financial accelerator
(Bernanke, 1995) if banks are risk-neutral. Assuming that banks are less risk averse with increasing wealth of borrowers, this effect will be even larger which then
constitutes an additional effect due to the risk-taking channel (Borio, 2012)
The second channel concerns the relationship between monetary policy and its effect on incentives of banks. It is thus different from the first channel in that it does not change the perceptions of banks on risk but changes the willingness to take on existing risk.
First, monetary policy affects the incentives of banks after the intervention of the central bank. The most commonly cited justification for this is the search-for yield channel of Rajan (2006). The idea put forward is that managers’ salary might depend upon meeting a nominal rate of return target. In a low interest rate environment, the incentive to take on risks is thus much more pronounced than in a high interest rate environment where renumeration is high even if little additional risk is taken on. A nominal rate of return target can be motivated by shareholders’ interests or the competition among banks for deposits. (Rajan, 2006) This search-for-yield concept again largely depends on the influence of monetary policy rates on bank profitability. Angeloni (2015) provides a similar rationale for the funding side. In their model,
banks choose between capital and short-term deposits to fund risky projects. Using short-term deposits have the advantage that they allow the banker to extract rent due to bargaining power but has the disadvantage of making bank runs more likely as the payoff of the risky project must be sufficient to pay more depositors. A fall in the policy rate then reduces the cost of short-term funding, increasing the possible rent extraction by using short-term debt instead of capital. The banker will then substitute short-term funding for capital more than is socially optimal as they do not internalize the whole costs of a bank run.5 (Angeloni, 2015)
3 Empiricial evidence indeed largely hints towards a negative relationship of expansive monetary policy and bank profitability. Cruz-Garcìa et al. (2019) find that for a panel of 32 countries between 2008 and 2014, expansionary monetary policy both reduced interest rates and flattened the yield curve with lowered net interest margins for banks. (Cruz-Garcia, 2019). Claessens et al. (2017) for a panel of 47 between 2005 and 2013 find a similar albeit non-linear effect on net interest margins but more mixed evidence for bank profitability, i.e. return on assets. These effects are economically significant (a 1 pp drop in the interest rate implies a 8 basis point drop in net interest margins) and stronger in a low interest rate environment. Busch and Memmel (2015) find a positive short-run and a negative long-short-run effect of monetary policy on net interest margins for German banks between 1968-2013. Borio et al. (2017) find that for a panel of 109 international banks 1995-2012 there is a positive effect if interest rates on net interest margins and a negative effect on provisions and non-interest income. The first effect however dominates, creating a positive relationship between interest rates and bank profitability. Genay (2014) finds similar results but argues that these effects are economically negligible compared to the influence of economic conditions on bank profitability.
7
Second, monetary policy can create moral hazard in banks if they anticipate future monetary policy intervention. If banks expect central bank intervention in case of a crisis, this lowers the cost of default for banks. In this case, banks might be willing to take on more risk. (Borio, 2012) Such an effect is incorporated in the model of
Diamond and Rajan (2012) in which demand deposits are used to solve the agency problem between banker and depositor. This will restrict bankers in their behaviour as otherwise depositors have the option to make a bank run which forces the bank to liquidate its assets. In a competitive market, banks will then compete for household deposits by offering the highest rate. There is then an externality which is not
internalized: On the one hand a higher deposit rate increases utility of the household, on the other hand it increases the likelihood of bank runs which however is not taken into account. In such a model, monetary policy intervention can then be used to prevent bank runs but weaken the discipline function of demand deposits. As a result, banks might make excessive promises on their liabilities which then increases the risk of bank runs (Diamond, 2012). As a result, anticipated ultra-low interest rates can potentially increase banks willingness to take on liquidity or credit risks (Rajan, 2010).
However, one can also argue theoretically in favour of an opposite effect. In principle, a tightening monetary policy shock can reduce the net worth of a bank by enough for it to be insolvent. In this case, a gambling for resurrection strategy of such a zombie bank might be used in which the bank takes on huge risks to increase its accounting measures of solvency (Kane, 1989, as cited in Maddaloni, 2013). Additionally, if banks net interest margins widen short-term due to a lowering of monetary policy, banks might pursue a less risky lending strategy to secure the additional gains (Dell ‘Ariccia, 2014). As a result, this effect of monetary policy on bank risk-taking is an empirical matter in the end.
Such empirical evidence largely confirms the existence of a risk-taking channel for the euro zone. Maddaloni and Peydró (2010) found that bank lending standards were weakened prior to the crisis and that looser monetary policy was associated to lower lending standards. They also found that securitization and weak supervision
8
the granting of loans to risky firms (on the extensive margin) and a 26 percent in the amount committed (on the intensive margin). In a follow-up paper using the same credit register for Spain they find a positive short term and a negative medium term effect on total credit risk of low interest rates. They argue that in the short-run the positive effect on the funding of outstanding loans outweighs the negative effect on the interest obtained by new loans. In the medium term, when the amount of new loans grows relatively to outstanding loans, this effect gets reversed. (Jímenez, 2017) Andries et al. (2015) explore the effect of interest rates being below the levels
suggested by a Taylor rate on three proxies of bank risk-taking, the NPL ratio, the z-score and the loan loss provisions ratio. They find that in the Eurozone between 1999 and 2011 lower interest rates were associated with higher risk-taking by banks.
Interestingly, they find that this effect was even stronger between 2008 and 2011. They interpret this as evidence for increased risk-taking in a low-interest rate
environment when interest rates were lowered as a response to the crisis. (Andries et al. (2015)). Brana et al. (2019) find that loose monetary policy increases both the market perception of risk, measured by the distance to default as well as the Z-score for banks in the Eurozone 2000-2015. This holds true both before and after the global financial crisis. Altunbas et al. (2014) find that monetary policy is transmitted through increased bank risk-taking. Using a panel of European and US banks they find that between 1999 and 2008 the expected default probability (the probability that within a year a bank will default) increases as a response to the policy rate being below the natural rate. Dell’ Ariccia et al. (2017) find similar results for the US 1997-2011. Using banks internal ratings on loans they find a negative relationship between bank risk-taking ex-ante and the FED funds rate.
A complementary approach is presented by Colletaz (2018) which attempted to measure the impact of monetary policy not on individual bank risk-taking but on systemic risk. Using the aggregate system risk indicator, defined as the bail-out amount of capital needed in a financial crisis, they find a causality from monetary policy to the build-up of system risk in the Eurozone 2000 to 2008. Buch et al. (2014) on the other hand found that only small domestic banks increased their risk-taking behaviour in response to an expansive monetary policy shock. However, for the overall banking system, they do not find evidence supporting the risk-taking channel for banks in the US 1997-2008. As a result, my first hypothesis is that I will find evidence for a positive relationship between monetary policy and bank risk-taking. More specifically, I expect to find tightening of monetary policy to be associated with the tightening of lending standards by banks.
A second strand of empirical investigations concerns the differential effect of
9
In line with the argument of non-linear effects, they find that this effect is even stronger if the indicators fall below a certain threshold. Pleşcău and Cocriş (2016) find unconventional monetary policy to have a similar effect on banks z-score than conventional monetary policy.
Additionally, I expect to find a stronger relationship to risk-taking for a measure that includes unconventional monetary policy than for the conventional measure of short-term interest rates. The rationale is that conventional monetary policy primarily influences short-term interest rates while unconventional monetary policy measures are rather directed towards the flattening of the yield curve. This is likely to have a more direct effect on bank profitability and through the search-for-yield effect. Thus, my second hypothesis is that unconventional monetary policy has a relatively
stronger effect and that this results in an overall stronger effect for the period after the crisis in which sustained ultra-low interest rates are combined with massive
expansions of the central banks’ balance sheet.
Furthermore, there is likely a non-linear effect of monetary policy on bank risk-taking. The reason is that the depression of bank margins is likely to be particularly grave in a low interest rate environment. As depositors have always the option of holding cash instead of deposits, which yields no interest and is associated only with some cost for storage, banks can lower deposit rates much below zero. Hence, my third hypothesis is that the effects measured are stronger for lower interest rates.
Thus, the hypotheses I will investigate, are:
H1: There is a negative association between loose monetary policy and bank risk-taking. Higher interest rates are associated with a net tightening of credit standards.
H2: This effect works by affecting banks risk perceptions, both through effects on their balance sheets as well as effects on the rest of the economy.
H3: Taking unconventional monetary policy into account will strengthen this relationship. H4: The relationship holds for the period before, during and after the crisis.
10
3. Data and econometric methodology
3.1 DataMy main dataset consists of quarterly data at the country level of the period 2003Q1- 2017Q4. The start date is the date at which the bank lending survey (BLS) of the ECB started and the end date is determined by the data availability of the
macroprudential index. The analysis is conducted for all Euro area countries, except for Finland for which results from the BLS are not published. Individual country data is only available as soon as a country joins the Euro, leading to an unbalanced panel. In 2003 the sample spanned around 90 banks in 12 Euro countries while in 2016 it included around 140 banks from 19 Euro area countries. Larger banks are
overrepresented in the sample while smaller banks are only included if their
behaviour constitutes a specific national feature of the respective banking system. Banks from this sample account for about 60% of outstanding loans to the private non-financial sector in the euro area (Köhler-Ulbrich, 2016).
3.1.1 Dependent variables
The dataset contains 3 main dependent variables, all obtained from the BLS. The first measure is the overall net tightening of overall credit standards (CS) applied to business loans. Credit standards are the ex-ante standards banks apply when they choose to whom to lend from a pool of borrowers they ranked according to their risk (De Bondt, 2010). As such they are a natural candidate to directly explore the change of risk-taking behaviour of banks on their asset side. 6
In the BLS, each bank answers as to its lending standards were tightened, eased or unchanged during the last three months. The individual bank responses (Question 1 in the questionnaire, see appendix) were then aggregated without weighing to the country level before being published by the ECB. (Köhler-Ulbrich, 2016) A positive number corresponds to a net tightening (i.e. more banks tightened than eased their credit standards in the respective country) while a negative number corresponds to a net easing of lending standards. The other two dependent variables are related to the
6 The BLS also contains questions concerning the terms and conditions that banks apply to loans. While they
also have been used as a proxy of bank risk-taking (e.g. in Maddaloni, 2010) they are a less appropriate
11
reasons for which banks changed their lending standards. In question 2 of the BLS, banks are asked to indicate whether different factors led to a tightening or easing of their credit standards. I will use this feature to create two additional measures, CFBSC and RP. The first variable measures the change in lending standards due to bank supply factors while the second measures the change in lending standards due to the risk perception by banks. Of the factors named in the questionnaire, I will follow the argument of Maddaloni (2013) and consider the balance sheet factors, namely the banks’ i) cost related to their capital position, ii) ability to access market finance and iii) liquidity position as being credit supply factors. The responses were again aggregated to net changes before being published by the ECB. I construct my measure as a simple average of these three factors and thus create the net change in lending standards due to bank balance sheet factors (CFBSC). Using this as an alternative measure instead of the overall change in credit standards hence allows me to specifically consider the effect of monetary policy on bank risk-taking through supply side factors. The second measure is constructed equally but with demand factors, i.e. i) the general economic situation and outlook, ii) the industry or firm specific situation and outlook as well as borrower’s creditworthiness and iii) the risk related to the collateral demanded. I will call this measure, which again is constructed as an average, the net change in lending standards due to risk perception (RP). A positive number then indicates that the factor considered contributed to a tightening of credit standards while a negative number indicates that it contributed to an easing of credit standards. Conceptually, these two measures allow me to evaluate whether monetary policy was not only related to the change in overall credit standards but additionally to consider whether the first channel, i.e. the change of banks risk perceptions, was rather active through changes in the banks balance sheets or through changes in the rest of the economy. It thus allows me to empirically investigate hypothesis two.
One issue when it comes to survey responses is their reliability and the truthfulness with which participants respond. However, as this survey is conducted by the ECB, the information gets cross-checked with banking data to assure the consistency of the responses (Maddaloni, 2010). Furthermore, the BLS responses are a good indicator for future bank lending. As shown by de Bondt (2010), the net tightening of credit standards is negatively correlated with different measures for credit availability. Furthermore, they explain significantly loan growth four quarters into the future for companies (de Bondt, 2010).
3.1.2 Independent variables
12
Second, the EONIA reflects both conventional monetary policy measures as well as unconventional ones. (Cicarelli, 2015) The reason is that the spread between the EONIA and the minimum bid rate (or later the fixed rate) is also determined by the liquidity needs of banks (Linzert, 2008). Hence, in particular during the crisis when providing liquidity to banks was a central part of ECB policy, the EONIA might better reflect the measures taken than the policy rate.
A second proxy is the Shadow Short Rate, as calculated by Krippner (2016). In particular after 2012, when the policy rate approached the zero lower bound, EONIA only dipped slightly into negative territory due to it being bound by the deposit facility rate (figure 1). This does not appropriately reflect the enormous expansion of the ECB’s balance sheet which might affect banks through its effect on asset prices. Nominal interest rates, such as the deposit rate, have the problem that they cannot fall much below zero as the cash option will always yield no interest and is only associated with a storage cost. The deposit rate of the ECB is thus necessarily lower bound through the cash option at some negative value close to zero. The shadow rate is then the theoretical rate that would arise if the cash option was not available. Thus, it allows for the incorporation of the effects of unconventional monetary policies that affect the yield curve (McCoy, 2017). In my regression, I will use the shadow short rate of Krippner (2016), as provided on his website.
13
A third issue is the endogeneity of monetary policy. Concerning reverse causality, theoretically the central bank could react to an increase in bank risk-taking by
changing their monetary policy. However, given the unique mandate towards inflation targeting, that is unlikely within the Euro area (Altunbas, 2014). Furthermore, banking supervision is the responsibility of national authorities while monetary policy is
conducted by the ECB (Maddaloni, 2013). Additionally, I follow Maddaloni (2010) and use the variation of Euro area countries in their business cycles to allow for different stances of monetary policy across Euro area countries, even when facing the same interest rate. To exploit this variation, I regress the EONIA rate on GDP and inflation in the different countries. The residuals from this regression are then used as a measure of the stance of monetary policy in the respective country. As in Maddaloni (2010), I will denote these residuals as Taylor rule residuals even though they are not deviations from a Taylor rule and hence do not indicate whether monetary policy was expansive relative to a historical benchmark but whether it was expansive vis-à-vis the other countries in the Euro zone. The use of these residuals then additionally permits me to use time fixed effects to account for common shocks that affect all countries in the Euro area. In the regression using the Taylor rule residuals, the variation is thus mostly cross-sectional, not over time (Maddaloni, 2013).
Furthermore, I lag all explanatory variables by one quarter to account for potential endogeneity, that still might exist.
Additionally, the dataset contains the macro controls GDP and inflation. Both were obtained from Eurostat. Furthermore, I include the Herfindahl index as published by the ECB which was calculated by squaring the market shares of the ten largest credit institutions in the respective country and summing them up. To account for loan demand, I use the responses of banks from the BLS which indicated how loan demand changed in the past three months. Furthermore, I control for changes in macroprudential supervision by using the index by Cerutti et al. (2017) which consists of 12 different macroprudential indicators. The index itself is a simple sum of
dummies that indicate whether a country has adopted the respective policy or not. Additionally, I use the country level index of financial stress which is provided monthly by the ECB and for which I formed quarterly averages. The index was created to identify periods of systemic stress (Duprey, 2015).7 I use it to capture potential effects
of both the financial as well as the sovereign debt crisis.
Summary statistics of all variables are provided in table 0. There is ample variation in both the dependent variables as well as the proxies for monetary policy. There
appears to have been a bias in favour of tightening lending standards over the whole period. This is likely driven by the crisis period.
14
3.2 Econometric methodology
To assess the influence of monetary policy on bank risk-taking, I specify panel data regressions in the general dynamic form:
𝑦𝑖,𝑡= 𝛼𝑖,𝑡+ 𝛿1𝑦𝑖,𝑡−1+ 𝛽1𝑀𝑃𝑖,𝑡−1+ 𝜃1𝑀𝐶𝑖,𝑡−1+ 𝜃2𝑀𝑃𝐼𝑖,𝑡−1+ 𝜃3𝐻10𝑖,𝑡−1+ 𝜃4𝐶𝐿𝐼𝐹𝑆𝑖,𝑡−1+∈𝑖,𝑡
The dependent variable 𝑦𝑖,𝑡 are the three measures of risk-taking as obtained by the
BLS, the net tightening in overall credit standards CS, the net tightening in lending standards due to the banks risk perception, RP and the net tightening in lending standards due to bank balance sheet factors CFBSC. MP is the set of the three measures of monetary policy, EONIA, the shadow short rate SSR and the Taylor rule residuals Taylorres. MC consists of the macroeconomic controls GDP growth and inflation, MPI is the macroprudential index, Comp the Herfindahl index and CLIFS the index of financial stress.
I will specify it in a dynamic model. As visible in figure 2, the means of a static
regression of credit standards on the EONIA rate, including all controls, appear to be serially correlated, in particular before 2013. This is later confirmed in the
regressions, in which the lags of the dependent variable are statistically significant. Based on different regressions, I choose to include two lags for the change in credit standards and one lag for the four other dependent variables.
CLIFS 865 .1188188 .0959773 .0122 .6872333 MPI 865 2.008092 1.430061 0 5 H10 865 .1005488 .0632239 .0173 .2613 inflation 865 1.695877 1.478189 -2.766667 6.433333 GDPgrowth 865 1.649711 3.596073 -10.8 29.2 Taylorres 865 .0498119 1.207056 -2.491668 3.153381 EONIA 865 1.017094 1.39471 -.3582 4.2527 SSR 865 -.6032317 3.123218 -7.336146 4.185488 CFBSC 865 6.507559 17.8734 -41.66667 86.66666 RP 865 16.20385 26.64884 -33.33333 100 CS 865 12.13353 28.34757 -75 100 Variable Obs Mean Std. Dev. Min Max
15
Thus, for the regression not to suffer from omitted variable bias, I will use the Arellano and Bond GMM estimator. All tables will include the p-value of the Sargan test for overidentification which was calculated without robust standard errors.
However, all regression results reported were estimated using robust standard errors and the first and second-order test for autocorrelation is provided to check for serial correlation in the differenced residuals.
16
4. Results
The results in the following section are reported as follows. First, I look at the impact of monetary policy on the change in overall bank lending standards and the channels through which they are changed. Table 1 and 2 report the results for proxying the stance of monetary policy with the EONIA rate and the Shadow short rate,
respectively. Table 3 reports the results using Taylor rule residuals instead. These are the results related to my main research question. In the following tables 4 to 9, I will only report the results for using the shadow short rate and Taylor rule residuals as proxies for monetary policy. As Taylor rule residuals allow for cross-country variation in the business cycle and the inclusion of time fixed effects, the are my preferred specification. I include the shadow rate to explicitly consider effects through unconventional monetary policy.
Thus, in the second section, I evaluate the robustness of the general effects across different time periods. I report results for the period before the global financial crisis (2003q1-2008q2), during the global financial crisis and the government debt crisis (2008q3-2012q4) and after the crisis period (2013q1-2017q4) using the SSR (tables 4 and 5) and Taylor rule residuals (tables 5 and 6) as proxies for the stance of monetary policy.
In the last section, I attempt to identify potential nonlinearities in the effects of
17
4.1 The impact of monetary policy on bank lending standards
In table 1, I report the results of the regression of the net tightening of bank lending standards on the interbank lending rate EONIA. Time fixed effects are not included and all regressions are estimated using robust standard errors. The lagged
dependent variables are both significant, justifying the dynamic specification. As visible in column 1, EONIA has a statistically significant positive effect on the net tightening of overall lending standards.8 This is consistent with the bank risk-taking
channel. A looser stance of monetary policy, proxied by a lower EONIA rate, is associated with a net easing of overall bank lending standards. In column 2 and 3 in which the dependent variables are the net tightening of lending standards due to changes in risk perception and due to changes in bank balance sheets, the effects
8 A quantitative interpretation of the marginal effect is not sensible as the measure only indicates whether
banks tightened their lending standards but not by how much.
Table 1 reports the results of the GMM dynamic panel estimation of the net tightening of lending standards on the interbank lending rate EONIA. The dependent variables are the net tightening of overall lending standards (column 1), the net tightening of lending standards due to changes in risk perception (column 2), the net tightening in lending standards due to bank balance sheet changes (column 3). The proxy for monetary policy is the EONIA rate. Time fixed effects are not included. Standard errors are reported in parentheses. The results were obtained using the Difference GMM estimator of Arellano and Bond with robust standard errors. The results of the Sargan test for overidentification and both AR tests for autocorrelation are reported as p-values.
18
are also positive. This suggests that monetary policy indeed has an effect through both of these channels. A looser stance of monetary policy is associated with a net easing of bank lending standards, both through the banks risk perception (i.e. concerning the general economic outlook, risk related to collateral and the situation of the borrower) and the banks balance sheet (i.e. its capital cost, liquidity position and its access to market finance). These results are consistent with the notion that an easing of monetary policy lowers risk perceptions of banks through changes on their measures of risk which depend on both their own balance sheet as well as the rest of the economy. Hence, this evidence supports the second hypothesis.
Loan demand has a small effect on overall lending standards but not on the other two measures. This suggests that increases in loan demand can induce banks to lower their lending standards albeit not through the perception of risk in the economy. Thus, there might be some unobservable factor not accounted for. One possible explanation might be waves of optimism through the business cycle which are not fully accounted for by GDP and inflation.
GDP growth has the anticipated easing effect on bank lending standards albeit it is not consistently significant. Higher inflation rates are associated with a net tightening of credit standards. Increased bank competition, as measured by the Herfindahl index, does not appear to influence bank risk-taking which is in line with the mixed evidence found on the effect of competition on bank profitability9. A higher
macroprudential index is associated with a tightening of lending standards. An increased number of regulatory policies is thus associated with a net tightening of credit standards. This suggests that such macroprudential policies are indeed effective in restricting the ability of banks to take risks. The financial stress indicator has a large positive effect, likely representing the strong tightening of credit standards that happened during the crisis. The estimates are however only statistically
significant for the balance sheet measure. This indicates that during the crises, it was mostly stress related to factors within the financial system that led banks to tighten their credit standards, rather than factors related to the rest of the economy.
9 See Badarau (2020) for a literature review. Competition might furthermore only be important for the reaction
19
In table 2, I consider the Shadow Rate instead of the EONIA as a proxy of monetary policy. The overall effect on bank risk taking is robust to the use of this alternative measure. The size of the coefficients is smaller but their sign is the same. As the Shadow Rate and EONIA started to diverge only with the adoption of the
unconventional balance sheet expansion of the ECB after 2012 (figure 1), one potential explanation is that these unconventional measures had little effect on bank risk-taking. Thus, there might be different effects of conventional and unconventional policy measures on bank risk-taking, as stated in hypothesis 3. The coefficients of the controls are robust.
Table 2 reports the results of the GMM dynamic panel estimation of the net tightening of lending standards on the Shadow Short Rate. The dependent variables are the net tightening of overall lending standards (column 1), the net tightening of lending standards due to changes in risk perception (column 2), the net tightening in lending standards due to balance sheet changes (column 3). The proxy for monetary policy is the Shadow Short Rate. Time fixed effects are not included. Standard errors are reported in parentheses. The results were obtained using the Difference GMM estimator of Arellano and Bond with robust standard errors. The results of the Sargan test for overidentification and both AR tests for autocorrelation are reported as p-values.
20
Table 3 reports the results of the regressions using the third proxy of monetary policy stance, the Taylor rule residuals. Other than the first two proxy, this measure varies on the country level due to the differences in business cycles across countries. This allows me to include time fixed effects into these regressions and hence to account for common shocks to monetary policy. The main association between the net tightening of lending standards and loose monetary policy is robust to the use of this measure. The estimates are larger, suggesting that cross-country differences
between euro area countries are indeed important. Interestingly, the macroprudential index looses its significance. This might be at least partly explained by the fact that most additional measures of macroprudential policy were adopted as a response to the crisis. Hence, this variation will be mostly absorbed by using time fixed effects. Briefly reviewing the evidence reported in the tables 1 to 3 allows me to conclude that I find robust evidence for my first hypothesis. In each of the three specifications using different proxies for monetary policy, I do find a looser monetary policy to be
associated with a net easing in lending standards by banks. In all of the
Table 3 reports the results of the GMM dynamic panel estimation of the net tightening of lending standards on the Taylor rule residuals. The dependent variables are the net tightening of overall lending standards (column 1), the net tightening of lending standards due to changes in risk perception (column 2), the net tightening in lending standards due to balance sheet changes (column 3). The proxy for monetary policy are country-level Taylor rule residuals. Time fixed effects are included. Standard errors are reported in parentheses. The results were obtained using the Difference GMM estimator of Arellano and Bond with robust standard errors. The results of the Sargan test for overidentification and both AR tests for autocorrelation are reported as p-values.
* p<0.10, ** p<0.05, *** p<0.01
21
specifications, I additionally find evidence for both channels investigated to be important. A change in monetary policy affects bank risk-taking both through their perception of risk as well as their balance sheet.
4.2 The effect of monetary policy on bank risk-taking for different time periods
As a first robustness test, I split the sample into three periods: before the global financial crisis (2003q1-2008q2), the period with both the financial and the sovereign debt crisis (2008q3-2012q4) and the period after the crises (2013q1-2017q4) in Europe. I chose to start the crisis period in the third quarter of 2008 as Lehman collapsed in September 2008. I end the in the last quarter of 2012 when the Greek 10-year government bond yield fell back to around 10% after it had been up to 35% during the crisis. In table 4 and 5, the shadow rate is used as the proxy for monetary policy and in table 6 and 7 monetary policy is proxied by Taylor rate residuals. In table 4 and 6, the dependent variable is the change in overall lending standards while table 5 and 7 contain the changes due to balance sheet factors and risk perception.
Table 4 reports the results of the GMM dynamic panel estimation of the net tightening of lending standards on the Shadow Short Rate for different time periods. The dependent variables is the net tightening of overall lending standards. There are three separate regressions for the period before (column 1), during(column 2), and after (column 3) both the financial and the government debt crisis in Europe. The proxy for monetary policy is the Shadow Short Rate. Time fixed effects are not included. Standard errors are reported in parentheses. The results were obtained using the Difference GMM estimator of Arellano and Bond with robust standard errors. The results of the Sargan test for overidentification and both AR tests for autocorrelation are reported as p-values.
22
Table 4 reports the estimates from a regression of the net tightening of overall credit standards by banks on the shadow short rate. Noticeably, monetary policy only appears to affect the change in lending standards before and during the crises but not afterwards. Additionally, the easing effect was stronger in the pre-crisis period and weaker during the period characterized by both the financial and the sovereign debt crisis. Before the crisis, a net easing in lending standards due to low interest rates can likely be interpreted as increasing overall bank risk. Maddaloni (2010) even argued that this might have been one of the factors contributing to the build-up of the crisis. During the crisis, the easing of monetary policy can rather be interpreted as causing a bounce-back of lending standards that were tightened by banks as an effect of the crisis. This is consistent with the strong tightening effect on lending standards of the financial stress indicator during the crisis period. Thus, while before the crisis, this effect of monetary policy was likely not intentional, during the crisis, the ECB actively attempted to encourage banks to be less prudent and increase lending to kick-start the economy. This is in line with the evidence of Maddaloni (2013) who found that the liquidity measures of the ECB had an easing effect on bank lending standards. In the same way, one would expect the massive balance sheet
expansions after the crisis adopted by the ECB to have an effect. However, this is not the case. After the crisis, loose monetary policy was not associated with a net easing of lending standards, as visible in column 3. While this can be interpreted in a
positive sense, i.e. that there was no increase in general bank risk-taking after the crisis, creating increased risk within the system, it can be also interpreted as a failure of the unconventional monetary policy measures to reach their objective to induce banks to increase lending. In particular in the context of sluggish economic growth over the last years, it appears that unconventional monetary policy failed to stimulate the economy through the bank risk-taking channel. Interestingly, the effect of
23
Table 5 repeats the exercise from table 4 but considers the net tightening of lending standards due to banks’ risk perception (column 1-3) as well as their balance sheet factors (column 4-6) instead of the overall change in credit standards. This indeed stenghtens the finding that before and during the crisis, both channels were active while after the crisis none of them were. After the crisis, a lower shadow short rate was even associated with a small net tightening effect of lending standards due to both the banks’ risk perception and its balance sheet factors. These coefficients are not statistically significant, however.
* p<0.10, ** p<0.05, *** p<0.01 AR(2) 0.670 0.488 0.096 0.537 0.468 0.197 AR(1) 0.005 0.000 0.019 0.019 0.001 0.007 Sargan 0.550 0.511 0.169 0.103 0.430 0.187 Time FE no no no no no no Observations 211 260 340 211 260 340 (17.02) (11.88) (7.03) (22.17) (7.79) (5.09) constant -18.505 12.060 5.534 -42.798* -27.089*** 5.710 (0.08) (0.06) (0.05) Balance sheet i, t-2 -0.060 0.037 0.060 (0.09) (0.09) (0.05) Balance sheet i,t-1 0.547*** 0.401*** 0.511*** (0.09) (0.07) (0.07) Risk perception i,t-2 0.070 0.012 0.199*** (0.08) (0.04) (0.10) Risk perception i,t-1 0.487*** 0.536*** 0.368*** (16.70) (11.95) (9.92) (13.45) (10.56) (5.29) Financial stress i,t-1 17.018 36.260*** 4.555 16.406 46.139*** 1.529 (10.83) (2.29) (0.90) (4.49) (2.96) (0.52) Macroprud. index i,t-1 -8.158 2.352 -1.270 -2.152 2.509 0.278 (149.73) (109.76) (42.30) (272.75) (101.23) (34.35) Herfindahl index i,t-1 181.631 -200.058* -34.331 301.432 143.906 -69.014** (2.29) (0.92) (0.66) (1.08) (0.68) (0.37) Inflation i,t-1 4.304* 3.054*** 1.551** 4.660*** 3.426*** 0.924** (0.58) (0.40) (0.32) (0.66) (0.41) (0.17) GDP growth i,t-1 -2.411*** -1.160*** -0.212 -0.163 -0.162 -0.154 (0.04) (0.03) (0.03) (0.03) (0.03) (0.01) Loan demand i,t-1 0.009 -0.043 -0.066** -0.038 -0.011 0.004 (1.57) (0.98) (0.34) (0.48) (1.01) (0.29) Shadow Short Rate t-1 4.892*** 5.094*** -0.401 4.312*** 3.376*** -0.081 2003q1-2008q2 2003q1-2008q2 2003q1-2008q2 2003q1-2008q2 2003q1-2008q2 2003q1-2008q2 Risk perception Balance sheet (1) (2) (3) (4) (5) (6) Table 5: The effect of monetary policy (Shadow Short Rate) before, during, and after the crises through different channels
24 * p<0.10, ** p<0.05, *** p<0.01 AR(2) 0.330 0.695 0.370 AR(1) 0.008 0.001 0.001 Sargan 0.519 0.405 0.426 Time FE yes yes yes Observations 211 260 340 (42.79) (13.66) (8.45) constant -28.957 -11.180 -2.452 (0.04) (0.08) (0.08) Credit standards i,t-2 0.175*** 0.057 0.073 (0.08) (0.07) (0.08) Credit standards i,t-1 0.367*** 0.364*** 0.220*** (27.48) (17.66) (12.81) Financial stress i,t-1 24.873 9.877 -10.094 (3.53) (3.74) (1.06) Macroprudential index i,t-1 -17.458*** 2.235 -1.047 (357.47) (129.56) (80.65) Herfindahl index i,t-1 111.737 52.698 89.721 (5.13) (1.41) Inflation i,t-1 6.584 6.269*** (0.88) (0.61) (0.53) GDP growth i,t-1 1.183 -0.188 -0.445 (0.05) (0.03) (0.03) Loan demand i,t-1 -0.022 -0.077** -0.132*** (5.25) (1.38) (3.51) Taylor rule residuals i,t-1 21.094*** 8.007*** -5.385 2003q1-2008q2 2003q1-2008q2 2003q1-2008q2 Credit standards (1) (2) (3) Table 6: The effect of monetary policy (Taylor rule residuals) before, during, and after the crises
25
In table 6 and 7, the exercise from table 4 and 5 is repeated, using my preferred measure of monetary policy, Taylor rule residuals. As before, the coefficients are of the same sign but of larger magnitude. Thus, the results are once again robust to the use of different measures of the monetary policy stance.
These findings have several implications for my hypotheses. First, the association between loose monetary policy and a net easing of bank lending standards is not observable for the post-crisis period. It thus appears that the risk-taking channel of monetary policy is not always active but might depend on the form monetary policy takes. Changes in the actual short-term interest rates appear to have a larger effect than the massive expansion of the ECBs’ balance sheet that are better reflected by the shadow short rate. The evidence thus does not support my third and fourth hypothesis. * p<0.10, ** p<0.05, *** p<0.01 AR(2) 0.389 0.691 0.179 0.485 0.703 0.113 AR(1) 0.007 0.000 0.016 0.006 0.002 0.004 Sargan 0.719 0.375 0.222 0.284 0.412 0.196 Time FE yes yes yes yes yes yes Observations 211 260 340 211 260 340 (19.34) (11.57) (7.95) (24.11) (8.32) (5.81) constant 10.936 13.446 3.815 -21.612 -23.662*** 5.519 (0.05) (0.07) (0.05) Balance sheet i, t-2 -0.150*** 0.019 0.085 (0.09) (0.09) (0.04) Balance sheet i,t-1 0.358*** 0.427*** 0.489*** (0.07) (0.07) (0.06) Risk perception i,t-2 0.127* 0.051 0.193*** (0.08) (0.06) (0.09) Risk perception i,t-1 0.466*** 0.462*** 0.339*** (23.35) (12.66) (11.14) (9.75) (11.11) (5.80) Financial stress i,t-1 3.291 28.283** 0.874 -1.277 36.015*** -0.025 (5.23) (2.43) (0.97) (2.58) (2.79) (0.84) Macroprud. index i,t-1 -12.574** 2.279 -0.808 -7.602*** 2.072 -0.131 (202.90) (112.00) (41.23) (262.97) (96.58) (49.69) Herfindahl index i,t-1 -121.154 -188.461* 61.613 242.998 148.434 -33.839 (2.69) (1.10) (1.46) (0.80) Inflation i,t-1 4.606* 5.147*** 0.569 4.127*** (0.60) (0.41) (0.29) (0.78) (0.37) (0.25) GDP growth i,t-1 -0.721 -0.018 -0.205 0.965 0.164 -0.280 (0.04) (0.03) (0.03) (0.02) (0.03) (0.01) Loan demand i,t-1 0.014 -0.033 -0.084*** -0.044** -0.016 -0.005 (3.39) (1.65) (1.88) (2.65) (1.35) (1.81) Taylor rule residuals i~1 11.558*** 5.028*** -2.405 8.148*** 4.826*** -2.096 2003q1-2008q2 2003q1-2008q2 2003q1-2008q2 2003q1-2008q2 2003q1-2008q2 2003q1-2008q2 Risk perception Balance sheet (1) (2) (3) (4) (5) (6) Table 7: The effect of monetary policy (Taylor rule residuals) before, during, and after the crises through different channels
26
4.3 Non-linear effects of monetary policy on bank risk-taking
In table 8 and 9, I restrict the analysis to the change in overall lending standards and examine the relationship between monetary policy and bank risk-taking in different interest rate environments. Different interest rate environments are constructed by separating the sample into two-subsamples, above and below the threshold of the EONIA rate. As potential thresholds, I consider a 2% threshold (column 1-2), the EONIA median at 0.3562% (column 3-4) and a 0% threshold (column 5-6).
Table 8 reports the results using the shadow short rate for different thresholds. For a 2% threshold, the association between loose monetary policy and a net easing of overall lending standards holds both above and below the threshold, although the effect is lower below the threshold. This suggests that indeed there might be a different non-linear effect of changes in monetary policy, according the interest rate environment it is conducted in. For the other two thresholds, the association of loose monetary policy and increased bank risk-taking is no longer statistically significant and even turns negative for an environment which is characterized by a negative EONIA rate. This suggests that the risk-taking channel of monetary policy becomes
* p<0.10, ** p<0.05, *** p<0.01 AR(2) 0.829 0.895 0.355 0.697 0.257 0.684 AR(1) 0.001 0.005 0.002 0.002 0.002 0.001 Sargan 0.352 0.316 0.043 0.172 0.328 0.274 Time FE no no no no no no Observations 576 235 431 380 228 583 (4.85) (18.34) (8.69) (24.27) (15.50) (5.53) constant -13.981*** -58.768*** 6.463 -68.788*** 10.247 -19.089*** (0.04) (0.06) (0.04) (0.06) (0.06) (0.04) Credit standards i,t-2 0.107*** 0.181*** 0.006 0.122** -0.010 0.100*** (0.07) (0.12) (0.06) (0.07) (0.07) (0.08) Credit standards i,t-1 0.456*** 0.406*** 0.197*** 0.424*** 0.152** 0.494*** (7.70) (10.05) (12.62) (12.99) (21.25) (8.42) Financial stress i,t-1 5.903 34.291*** -6.051 44.146*** -15.526 18.040** (0.68) (8.26) (0.94) (4.71) (1.66) (1.18) Macroprud. index i,t-1 2.033*** -13.052 1.065 0.060 -1.995 3.019** (32.30) (256.34) (64.53) (275.34) (130.04) (68.07) Herfindahl index i,t-1 96.838*** 510.055** -72.341 558.243** -56.701 74.394 (0.70) (2.10) (1.03) (1.88) (1.65) (0.69) Inflation i,t-1 3.656*** 4.126** 4.275*** 4.824** 2.639 4.030*** (0.44) (0.89) (0.58) (0.66) (0.38) (0.58) GDP growth i,t-1 -0.257 -1.658* -0.503 -2.346*** -0.204 -0.763 (0.02) (0.06) (0.02) (0.05) (0.04) (0.03) Loan demand i,t-1 -0.032 -0.029 -0.131*** -0.026 -0.087** -0.058** (0.43) (1.18) (0.52) (1.61) (0.56) (0.50) Shadow Short Rate t-1 1.569*** 10.208*** 0.355 7.405*** -0.842 3.144*** <=2.0 >2.0 <=0.3562 >0.3562 <=0 >0 (1) (2) (3) (4) (5) (6) Table 8: Non-linearities in the effect of monetary policy (Shadow Short Rate) on the change in overall lending standards
27
weaker as short-term interest rates approach their lower bound. This evidence directly contradicts hypothesis 5.
One possible explanation is that in a low interest rate environment, it is more
attractive for banks to shift their focus to other sources of income rather than further lowering their lending standards to increase the income from their traditional business model. Another possibility is that banks funds got reduced in such a low interest rate environment as it became less attractive for savers to save through deposits.
Table 9 reports the results using the Taylor rule residuals for different thresholds. The results from this specification are more mixed. While the results are largely similar in their interpretation when it comes to a threshold of 2%, for a threshold of 0.3562% the effect is insignificant below that threshold but significant above it. For a potential 0% threshold, both coefficients are even negative but only the one for the
environment above the threshold is significant. Thus, the finding of non-linearities is not robust to using another proxy for monetary policy.
. * p<0.10, ** p<0.05, *** p<0.01 AR(2) 0.994 0.876 0.655 0.674 0.384 0.989 AR(1) 0.001 0.007 0.002 0.002 0.002 0.001 Sargan 0.407 0.427 0.070 0.104 0.351 0.356 Time FE yes yes yes yes yes yes Observations 576 235 431 380 228 583 (5.99) (25.73) (7.41) (69.40) (15.16) (7.09) constant -6.587 -44.049* 8.182 -102.844 10.671 -7.036 (0.04) (0.05) (0.05) (0.06) (0.07) (0.04) Credit standards i,t-2 0.116*** 0.188*** 0.053 0.110** 0.009 0.134*** (0.07) (0.09) (0.06) (0.06) (0.07) (0.08) Credit standards i,t-1 0.376*** 0.296*** 0.173*** 0.292*** 0.147** 0.398*** (10.63) (23.77) (13.70) (8.92) (23.34) (8.69) Financial stress i,t-1 1.128 16.387 -10.457 8.128 -17.992 9.831 (0.61) (4.32) (1.34) (6.06) (1.63) (1.55) Macroprudential index i~1 0.846 -16.164*** 1.145 -4.867 -3.423** -0.319 (32.61) (268.76) (53.03) (241.65) (121.83) (26.56) Herfindahl index i,t-1 81.831** 348.722 -16.256 430.477* -21.830 37.433 (1.03) (2.69) (1.79) (8.66) Inflation i,t-1 6.079*** 5.415** 11.767*** 15.645* (0.44) (0.83) (0.66) (1.29) (0.51) (0.52) GDP growth i,t-1 -0.036 0.083 0.310 0.971 -0.414 -0.706 (0.02) (0.05) (0.02) (0.04) (0.04) (0.03) Loan demand i,t-1 -0.069*** -0.024 -0.148*** 0.008 -0.101** -0.042 (1.67) (2.72) (4.83) (21.41) (4.19) (1.96) Taylor rule residuals i~1 7.709*** 22.621*** 19.186*** 33.144 -6.444 -6.869*** <=2.0 >2.0 <=0.3562 >0.3562 <=0 >0 (1) (2) (3) (4) (5) (6) Table 9: Non-linearities in the effect of monetary policy (Taylor rate residuals) on the change in overall lending standards
28
5. Conclusion
In this thesis, I provided evidence supportive of the existence of the bank risk-taking channel in the euro zone. In general, a tighter stance of monetary policy is
associated with a net tightening of lending standards by banks. This works both through an effect on banks’ balance sheets as well as an effect on their risk
perception concerning the rest of the economy. Overall, banks appear willing to take on additional risk if monetary policy gets eased.
This relationship however only holds for the period before and during the crisis. This finding is consistent with earlier papers which found that loose monetary policy might have contributed to the build-up of imbalances prior to the crisis. During the crisis, while overall lending standards were strongly tightened by banks, monetary policy likely helped to ease them. For the period after the crisis, no such effect is present, however. There are several possible explanations. First, the macroprudential policies adopted as a reaction to the crisis might lower the ability of banks to lower lending standards as a reaction of expansionary monetary policy. In my regression, the macroprudential policy indicator has however not a consistent effect on bank lending standards. This however might at least in part be due to its construction. As it is only a cumulative indicator of dummy variables indicating whether a country adopted a macroprudential policy, it does not reflect different ways of implementation between countries. Hence, it might omit additional variation between countries and across time.
Second, banks might have been able to lessen the impact on their profitability through the effect of monetary policy on their net interest income by raising non-interest income. As a result, the search-for-yield channel might not be present. Additionally, this might imply that banks shift at least partly their focus away from the traditional business of taking deposits and making loans to other financial activities which are more lucrative. Risk-taking might happen here which implies that overall bank risk still might increase. Furthermore, my analysis only allows me to measure the contemporaneous effects of monetary policy on bank risk-taking. However, the negative effects on bank profitability might take more time to realize. As most bank loans have a fixed rate and are long-term, the effects of a flatter yield curve might only be realized after several years after which most of the old debt has been rolled over into new, lower interest rate loans. In the meantime, a lower interest rate likely has a positive effect on bank profitability as it makes funding which is mainly short-term cheaper.
29
Fourth, as in particular the period after 2012 was characterized by massive
expansions of the ECB balance sheet, this might suggest that such policies had less of an effect on bank risk-taking than traditional interest rate policy. One potential explanation might be that the effect of monetary policy on banks funding side is of higher importance for the risk-taking decision of banks than the flattening of the yield curve and the positive effect on asset prices such a policy might have. This is
consistent with the finding of Maddaloni (2010) that the long-term interest rate played no role in the build-up of the crisis and was not associated with a change in lending standards.
Along these lines, there is room for potential future examinations concerning the use of the BLS to study bank risk-taking in the Euro zone. One potential extension would be the use of an endogenously determined threshold to examine potential non-linearities. Additionally, the use of further measures besides the Krippner shadow rate to account for the influence of unconventional monetary policy might be advisable. In particular the examination of a way to allow for such policies to represent a different stance of monetary policies across Euro area countries, analogously to the Taylor rule residuals used for conventional policy in this thesis, could be a fruitful new path in my opinion. Furthermore, since 2015 the questionnaire additionally included for banks the option to indicate whether they changed their lending standards due to their risk tolerance. This can be used to additionally examine the second channel of monetary policy on risk-taking through incentives. In summary, I cannot confirm the fear that the ultra-low interest rates have led to a significant loosening of banks’ credit standards in the euro zone in the past years. While overall the bank risk-taking channel appears to be active in the euro zone, this is less the case in the past years. On the one hand, fears concerning the
30
6. References
Adrian, T., & Shin, H.S. (2010). Financial Intermediaries and Monetary Economics.
Chapter 12 in Handbook of Monetary Economics, vol. 3A, 601-650. Elsevier B.V.
Adrian, T., Estrella, A., & Shin, H.S. (2019). Risk-taking channel of monetary policy.
Financial Management, 48, 725-738.
Ahrend, R., Cournède, B., & Price, R. (2008). Monetary policy, market excesses and financial turmoil. OECD Economics Department Working Papers No. 597.
Altunbas, Y., Gambacorta, L., & Marques-Ibanez, D. (2014). Does monetary policy affect bank risk?. International Journal of Central Banking, 10(1), 95-135.
Andries, A.M., Cocriş, V., & Pleşcău, I. (2015). Low interest rates and bank risk-taking: Has the crisis changed anything? Evidence from the Eurozone., Review of
Economic & Business Studies, 8(1), 125-148.
Angeloni, I., Faia, E., & Lo Duca, M. (2015). Monetary policy and risk taking. Journal
of Economic Dynamics & Control, 52, 285-307.
Badarau, C., & Lapteacru, I. (2020). Bank risk, competition and bank connectedness with firms: A literature review. Research in International Business and Finance,
51(C).
Bernanke, B.S., & Gertler, M. (1995). Inside the black box: The credit channel of monetary policy transmission. Journal of Economic Perspectives, 9(4), 27-48. Borio, C., & Zhu, H. (2012). Capital regulation, risk-taking and monetary policy: A missing link in the transmission mechanism?. Journal of Financial Stability, 8, 236-251.
Borio, C., Gambacorta, L., & Hofmann, B. (2017). The influence of monetary policy on bank profitability. International Finance, 20, 48-63.
Brana, S., Campmas, A., & Lapteacru, I. (2019). (Un)Conventional monetary policy and bank risk-taking: A nonlinear relationship. Economic Modelling, 81, 576-593. Buch, C.M., Eickmeier S., & Prieto E. (2014). In search for yield? Survey-based evidence on bank risk taking. Journal of Economic Dynamics & Control, 43, 12-30. Busch, R., & Memmel, C. (2015). Banks’ net interest margin and the level of interest rates. Deutsche Bundesbank Discussion Paper No. 16/2015.
Cerutti, E. Claessens, S., & Laeven, L. (2017). The use and effectiveness of
