The interest rate pass-through in the Euro area during the global financial crisis

Accepted Manuscript The Interest Rate Pass–Through in the Euro Area During the Global Financial Crisis Nikolay Hristov, Oliver Hülsewig, Timo Wollmershäuser PII: DOI: Reference:

S0378-4266(14)00271-4 http://dx.doi.org/10.1016/j.jbankfin.2014.08.004 JBF 4531

To appear in:

Journal of Banking & Finance

Received Date: Accepted Date:

20 December 2013 5 August 2014

Please cite this article as: Hristov, N., Hülsewig, O., Wollmershäuser, T., The Interest Rate Pass–Through in the Euro Area During the Global Financial Crisis, Journal of Banking & Finance (2014), doi: http://dx.doi.org/10.1016/ j.jbankfin.2014.08.004

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

The Interest Rate Pass–Through in the Euro Area During the Global Financial Crisis Nikolay Hristova , Oliver H¨ ulsewiga,c,∗, Timo Wollmersh¨ausera,b a

CESifo and Ifo Institute – Leibniz Institute for Economic Research at the University of Munich, Poschingerstr. 5, 81679 Munich, Germany b Ludwig–Maximilians–University Munich, Ludwigstr. 28, 80539 Munich, Germany c Munich University of Applied Sciences, Am Stadtpark 20, 81243 Munich, Germany

Abstract This paper uses panel vector autoregressive (VAR) models for euro area member countries to explore the widening of retail bank interest rate spreads that emerged in the course of the global financial crisis. We find that the interest rate pass–through was generally complete on impact before the outbreak of the financial crisis, but became significantly distorted in the period thereafter, which hampered the effectiveness of monetary policy. Empirical evidence suggests that the decrease in the interest rate pass–through can be related to a change in the structural parameters characterizing the economies and a substantial increase in the average size of structural shocks. DSGE model simulations show that an increase in the frictions that banks are subject to can explain the decrease in the retail bank interest rate pass–through. Keywords: Euro area, interest rate pass–through, global financial crisis, panel vector autoregressive model, sign restrictions JEL: E40, E43, E52

∗

Corresponding author, Tel.: +49 (0)89 1265 2724; fax: +49 (0)89 1265 2714 Email addresses: [email protected] (Nikolay Hristov), [email protected]> (Oliver H¨ ulsewig), [email protected] (Timo Wollmersh¨ auser) Preprint submitted to Elsevier

August 12, 2014

1. Introduction The European Central Bank (ECB) vigorously cut its policy rates in response to the global financial crisis.1 However, banks in the euro area only partly passed–through the lower refinancing costs to the interest rates charged on loans and offered on deposits. As a result the spreads of retail bank interest rates over money market rates sharply increased from mid–2008 on and remained on this high level since then (see Figure 1). Being concerned by the impediment of the transmission of monetary policy the ECB started to implement several additional unconventional monetary policy measures to restore the effectiveness of monetary policy (see European Central Bank, 2011, for a summary).2 Figure 1: Spreads of euro area retail bank interest rates over money market rates There are at least two possible explanations for the increase in retail bank interest rate spreads. First, a sequence of adverse macroeconomic shocks could have hit the economies during the financial crisis. The recession induced by such shocks, might have pushed retail rate spreads upwards by substantially increasing the riskiness of private borrowers while, at the same time, reducing households’ deposit demand. Second, the financial crisis could have led to fundamental changes in the stochastic properties as well as in the propagation of macroeconomic shocks. In particular, there might have been a systematic structural shift either in the variances of the structural shocks or in the dynamic response of the economies to those shocks, or in both, 1

The ECB lowered the interest rate on its main refinancing operations by 325 basis points between October 2008 and May 2009. 2 Note that in the euro area, retail bank interest rates play an important role in the transmission of monetary policy, since borrowing takes place predominantly through the intermediation of the banking sector, contrary to other economies like the U.S. where securities markets are the main funding source of the real sector (Goddard et al., 2007). Over the period 2004–08 bank financing constituted around 75% of total external financing by non–financial corporations in the euro area and less than 50% in the U.S. (European Central Bank, 2010, p. 62).

2

reflecting, for instance, a permanent decrease in banks’ appetite for risk. Additionally, irrespective of the specific nature of the structural changes caused by the financial crisis, the latter could have been associated with an increase in the relative importance of certain shocks that only play a minor role in normal times. A prominent example would be loan supply shocks, which are typically identified as increasing the interest rate spread and which are found to have played an important role during the global financial crisis (see e.g. Helbling et al., 2011; Hristov et al., 2012). This paper explores whether structural changes in the euro area have contributed to the increase of retail bank interest rate spreads. Thus, it contributes to the ongoing debate on whether the transmission of monetary policy through retail bank rates has been systematically distorted since the onset of the financial crisis in 2008 (see e.g. European Central Bank, 2013; Neri, 2013). We use panel vector autoregressive (VAR) models for the euro area member countries to explore how banks adjusted their retail rates in response to changing money market rates during the period 2003–11.3 We focus on the interest rate pass–through in response to three standard macroeconomic shocks, namely a monetary policy shock, an aggregate supply shock and an aggregate demand shock, which are identified by imposing sign restrictions. We split our sample into a pre–crisis period from 2003–07 and a crisis period from 2008–11 and compare the response of the interest rate spreads to shocks of the same (unit) size. Our main result is that there have been significant structural changes in banks’ behavior as the interest rate pass–through became less complete during the crisis, while it was generally complete before. While this finding 3

It is important to note that monetary policy is transmitted through several channels to the real economy. By investigating the reaction of retail bank interest rates to a policy– induced decline in money market rates our empirical analysis is focussing on the interest rate channel. However, distortions in other transmission channels, such as the asset price channel or the balance sheet channel, might have also contributed to the decrease in the interest rate pass–through. Since our empirical VAR approach is agnostic, any identification of specific transmission channels is difficult. However, in a theoretical analysis presented below we will discuss possible explanations for distortions in the propagation of policy rate changes using a full–fledged macroeconomic model of the euro area that accounts for various monetary transmission channels.

3

holds in particular for the banks’ response to a monetary policy shock, the quantitative reaction of the interest rate spread following the real structural shocks of the same unit size is much smaller and even insignificant in the case of an aggregate supply shock. Our results also hold for different categories of retail bank interest rates, i.e. lending rates and deposit rates with different maturities. Thus, the increase in interest rate spreads in the period 2008–11 can at least be partly explained by banks being structurally more reluctant to lower their loan and deposits rates. In addition to that, we provide evidence in support of a significant increase in the average size of all structural shocks and of a larger role of additional unidentified shocks (e.g. loan supply shocks) during the crisis period. Splitting our panel of euro area member countries into two groups, namely the core countries and the periphery countries reveals additional insights. On the one hand, we find that the decrease in the interest rate pass–through is statistically not different across country groups and hence driven by structural changes in all euro area member countries. On the other hand, the systematic increase in the volatility of the structural shocks as well as the increased importance of the additional unidentified shocks seem to be particularly relevant for explaining the relatively strong increase in interest rate spreads that has been observed in the periphery countries. On basis of our empirical results we turn our focus to recent work on New Keynesian dynamic stochastic general equilibrium (DSGE) models with financial frictions (Curdia and Woodford, 2010; Gerali et al., 2010; Gertler and Karadi, 2011; Kollmann et al., 2011) to provide a theoretical explanation for a structurally weaker retail bank interest rate pass–through. In particular, we resort to the model of Gerali et al. (2010), which incorporates several features of the banking sector that are important for understanding the dynamics of loans and deposits, retail bank interest rates, risk premia and money market rates. In this model banks provide collateralized loans to the private sector and collect deposits from households in an environment of monopolistic competition. Their business is constrained by costs of maintaining an adequate bank capital position and costs related to the adjustment of retail rates.

4

For all these frictions we provide evidence that the financial crisis has increased their severeness. This observation allows us to calibrate two regimes of the model of Gerali et al. (2010): the first corresponding to the time prior to the outbreak of the financial crisis and the second, mimicking important features of loan and deposit markets thereafter. We simulate the response of retail bank interest rate spreads to the same macroeconomic shocks as identified in our empirical analysis for each regime. It turns out that theory replicates our main empirical findings quite well as the model implies a lower degree of policy rate pass–through during the crisis regime. Overall, we find that the structurally distorted transmission of monetary policy via retail bank rates in the period after 2008 might be attributable to lower average loan–to–value ratios, higher costs associated with restoring the soundness of bank capital, weaker competition among banks in loan markets, and a higher degree of interest rate stickiness. In addition, the model simulations also provide some reasoning for our empirical finding of a quantitatively different effect of structural shocks of the same size on retail bank rate spreads. Our work is related to a number of studies (see e.g. Cottarelli and Kourelis, 1994; Toolsema et al., 2001; Sander and Kleimeier, 2004; De Bondt, 2005; Kok Sørensen and Werner, 2006; Kleimeier and Sander, 2006; Kwapil and Scharler, 2010; Van Leuvensteijn et al., 2013; Belke et al., 2013) that largely confirm that retail bank interest rates in Europe react sluggishly to changes in money market rates. Most of the results are derived from error–correction models, which typically comprise different retail bank interest rates next to money market rates. In some papers (see e.g. Kok Sørensen and Werner, 2006; Van Leuvensteijn et al., 2013) additional explanatory variables are considered, such as measures for bank competition, credit risk, bank health, the extent of money market development or the regulatory system. Studies on the impact of the financial crisis on the degree of interest rate pass–through in Europe are relatively scarce. Aristei and Gallo (2014) use a Markov–switching VAR model to explore the change in the speed of adjustment of retail bank interest rates to changing money market rates. Their findings show that since the crisis the pass–through has become less complete. Karagiannis et al. (2010), Blot and Labondance (2011) and Hansen 5

and Welz (2011) provide evidence based on error–correction models that confirms this result. Our work contributes to the literature in several ways. First, we use an identified VAR model to analyze how the impact interest rate pass– through has changed since 2008 conditional on different types of macroeconomic shocks. To our best knowledge such an analysis has not yet been conducted as the existing interest rate pass–through literature does not explicitly distinguish between different sources of macroeconomic fluctuations.4 Our results of both, the empirical VAR analysis and the theoretical DSGE model simulations show that the extent to which the interest rate spread reacts crucially depends on the nature of the structural shocks. Second, we resort to panel techniques. The use of such techniques is rather uncommon in the literature on the interest rate pass–through, but it allows us to cope with the relatively limited length of the available time series and to split our sample in order to evaluate the speed of interest rate pass–through before and during the global financial crisis. Third, we consider several types of structural changes that possibly explain the increase in interest rate spreads. We distinguish between shifts in the stochastic properties of macroeconomic shocks, in their explanatory power and in their propagation to the economy. The remainder of the paper is organized as follows. In Section 2 we introduce the panel VAR model and provide a detailed discussion on the identification of the structural shocks. Section 3 presents the impulse responses to the structural shocks, a decomposition of the forecast error variance and an analysis of structural shock volatility. Section 4 compares our empirical results with those obtained from simulations of the model of Gerali et al. (2010). Section 5 summarizes our main findings and concludes. 4

De Bondt (2005) uses a bivariate VAR model to examine the link between retail bank rates and money market rates. He focuses on the responses of retail bank rates to shocks in money market rates without exploring the origin of the shock.

6

2. Panel VAR models with sign restrictions 2.1. Panel VAR model setup Consider a panel VAR model in reduced form: Yi,t = ci +

p 

Aj Yi,t−j + εi,t ,

(2.1)

j=1

where Yi,t is a vector of endogenous variables for country i, ci is a vector of country–specific intercepts, Aj is a matrix of autoregressive coefficients for lag j, p is the number of lags and εi,t is a vector of reduced–form residuals. The vector Yi,t consists of four variables: Yi,t = [yi,t pi,t st ri,t ] ,

(2.2)

where yi,t denotes real GDP, pi,t is the overall price level, measured by the GDP deflator, st is the nominal short–term interest rate, which serves as the policy instrument of the central bank5 , and ri,t is a retail bank interest rate, which is either a lending rate or a deposit rate. For each variable, we use a pooled set of M · T observations, where M denotes the number of countries and T denotes the number of observations corrected for the number of lags p. The reduced–form residuals εi,t are stacked into a vector εt = [ε1,t . . . εM,t ] , which is normally–distributed with mean zero and variance– covariance matrix Σ. We use quarterly data for the EMU member countries covering the period from 2003Q1 to 2011Q4.6 The set of countries comprises Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Portugal and Spain. The beginning of the sample is determined by the retail bank interest rates, which are available from the ECBs harmonized MFI interest rate statistics only since 2003. The lending rates refer to interest rates on new business loans to non–financial corporations (‘NFC’), new loans to households 5

Since for all member countries of the euro area the nominal short–term interest rate is identical, we set si,t = st for all i. 6 The data is taken from Eurostat and the ECB databases. See Appendix A for a detailed description of the data.

7

for consumption (‘CON’) and new loans to households for house purchases (‘MOR’). The deposit rates are interest rates agreed on new deposits from private households (‘DEP’). The spectrum of maturities comprises an average over all maturities (‘total’) and maturities of up to one year (‘–1y’) as well as over one year (‘1y+’). The policy instrument of the ECB is mirrored by the EONIA, which is the average of overnight rates for unsecured euro area interbank lending. As Ciccarelli et al. (2010) point out, the EONIA rate is a sensible measure of the ECB’s monetary policy especially during the financial crisis. Since the ECB reacted to the crisis by implementing various non–standard measures in its liquidity management, the EONIA reflects much better the monetary policy stance than the official main refinancing rate. We express real GDP and the price level in logs, while the interest rates are expressed in percent. All variables are linearly de–trended. We estimate several panel VAR models, which differ with regard to the respective retail bank interest rates ri,t . The number of countries M varies across specifications depending on the availability of retail bank interest rates.7 The advantage of the panel approach is that it increases the efficiency of the statistical inference. This holds in particular in cases where the sample is short.8 Each panel VAR model is estimated with Bayesian methods using a Normal–inverted Wishart prior, 500 draws and a lag order of p = 2. The matrix of constant terms ci comprises individual country dummies that account for cross–country heterogeneity. Several issues concerning our approach require some discussion. The first issue concerns our choice of using levels of real GDP and prices in our VAR, which are clearly non–stationary variables. Generally speaking, in Bayesian VARs non–stationarity does not matter for inference (Canova, 2007, ch. 4.2.4). But even in classical econometrics it is common practice 7

See Appendix B for information about the country–specific availability of the respective retail bank interest rates. 8 Estimating single models for the individual countries separately would instead suffer from a small number of degrees of freedom if only a limited number of observations is available.

8

to estimate a VAR model in levels instead if using a VECM, as it can be shown that the level VAR model is asymptotically equivalent to the VECM (L¨ utkepohl, 1993, ch. 11.2). Thus, by doing the analysis in levels we allow for implicit cointegration relationships in the data. Second, the assumption of slope homogeneity across countries would yield biased estimates, if the degree of heterogeneity is pronounced (Tillmann, 2013). While the mean–group estimator of Pesaran and Smith (1995) would account for slope heterogeneity across countries, its use requires that the sample is sufficiently long, i.e. “twice as long as usually recommended in the dynamic panel data literature” (Rebucci, 2010, p. 1183), which in our analysis is clearly violated. Mote Carlo simulations by Rebucci (2010) show in fact that the efficiency of the mean–group estimator is very limited in short samples.9 Third, another potential source of bias in the fixed effect estimator of Aj is the correlation between the regressors and the error terms in dynamic panel models. However, it is well known that this so–called Nickell (1981) bias disappears when T becomes sufficiently large. We believe that in our set–up this bias should be small since we have at least T = 14 observations. 2.2. Identification of structural shocks Based on the panel VAR models estimated for the different retail bank interest rates, we generate impulse responses of the variables to structural shocks ηt . As in Canova and De Nicol`o (2002), Peersman (2005) and Uhlig (2005) the shocks are identified by imposing sign restrictions. The reduced– form residuals εt are related to the structural shocks ηt according to ηt = (U Ω1/2 Q)−1 εt , where U Ω1/2 is the Cholesky factor, Σ = U ΩU  , of each draw and Q is an orthogonal matrix, QQ = I, generated from a QR decomposition of some random matrix W , which is drawn from an N (0, 1) density. For each of the 500 Cholesky factors resulting from the Bayesian estimation of the 9

In particular, Rebucci (2010) shows in Monte Carlo simulations that (i) the fixed effect estimator is biased only if slope heterogeneity is sizable, and (ii) in cases where slope heterogeneity is indeed very high, the sample must be remarkably long for the mean–group estimator to outperform the fixed effect estimator.

9

VAR model, the draws of the random matrix W are repeated until a matrix Q is found that generates impulse responses to ηt , which satisfy the sign restrictions. Since each panel VAR model contains four variables, the maximum number of structural shocks that can be identified is four. We identify three of them, a monetary policy shock, an aggregate supply shock and an aggregate demand shock. The sign restrictions we impose are standard (Gal´ı et al., 2003; Peersman, 2005; Straub and Peersman, 2006; Canova and Paustian, 2010) and summarized in Table 1. Table 1: Sign restrictions Shock Monetary policy Aggregate supply Aggregate demand

Real GDP

GDP deflator

↑ ↑ ↓

↑ ↓ ↓

Money market rate ↓ ↓ ↓

Retail bank rate ? ? ?

Notes: Restrictions on variables are imposed for four quarters. ↑ denotes a positive response, ↓ denotes a negative response. Unrestricted variables are marked by ‘?’.

We assume that an expansionary monetary policy shock has a positive effect on output and prices and a negative effect on the money market rate. A favorable aggregate supply shock is assumed to have a positive impact on output and a negative impact on prices and the money market rate. Finally, we assume that an adverse aggregate demand shock has a negative effect on output, prices and the money market rate. We refrain from imposing restrictions on the retail bank interest rates since the adjustment of these interest rates – especially concerning the adjustment of lending rates – is theoretically ambiguous. Therefore, we let the data determine the sign of the responses of the retail bank rates. The fourth shock is interpreted as a residual shock, which captures the remaining variation in the data.10 For all variables we set the time period over which the sign restrictions are binding equal to four quarters. This is in line with Peersman (2005), Farrant and Peersman (2006), Uhlig (2001) and 10

See Eickmeier et al. (2009), among others, for a similar approach.

10

Scholl and Uhlig (2008), who assume that the effects of shocks on economic activity can be quite sustainable.11 All sign restrictions are imposed as ≤ or ≥.12 3. Results Overall, we estimate 12 panel VAR models as we consider four different retail bank interest rates with three different times to maturity. In order to investigate the impact of the financial crisis on the degree of interest rate pass–through we split our sample into two sub–periods, namely the period before the financial crisis 2003–07 and the period of the financial crisis 2008– 11. For every panel VAR model specification the sample split is supported by various tests for a structural break. Table 2 reports statistics of the tests and the 5% critical values.13 Following Stock and Watson (2005) the results of a multi–equation version of the Chow–Wald test for the null hypothesis of equal residual variances across both sub–samples are summarized in the second and third column. For all model specifications the test confirms that the two periods are characterized by a significant degree of heteroscedasticity of the reduced–form residuals, with volatilities as measured by the trace of the residual covariance matrix being on average six times higher during the crisis than before the crisis. The results of alternative tests, i.e. the Chow test and the Wald test, for the null hypothesis of parameter stability across the two sub–samples are shown in the columns four to seven. They also support the view that the financial crisis has caused significant changes in the propagation mechanism of structural shocks. 11

Since the choice of the time period over which the sign restrictions are assumed to hold is arbitrary, we checked the robustness of our results by imposing sign restrictions that are binding only over two quarters. This modification has no impact on our results. 12 The estimation of the Bayesian VAR and the identification of the structural shocks is performed in MATLAB, using the codes bvar.m, bvar chol impulse.m and bvar sign ident.m provided by Fabio Canova (http://www.crei.cat/people/canova/). 13 See Appendix C for details on the tests.

11

Table 2: Tests for a structural break in 2007Q4/2008Q1

NFC total MOR total CON total DEP total NFC -1y MOR -1y CON -1y DEP -1y NFC 1y+ MOR 1y+ CON 1y+ DEP 1y+

Variance statistic 85.1 78.4 112.1 69.8 103.0 76.5 104.9 81.3 84.3 71.7 91.1 63.6

5% critical value 9.47 9.47 9.47 9.47 9.47 9.47 9.47 9.47 9.47 9.47 9.47 9.47

Chow statistic 207 163 192 171 256 170 158 123 174 252 227 100

5% critical value 0.74 0.74 0.74 0.73 0.75 0.74 0.73 0.73 0.74 0.74 0.74 0.73

Wald statistic 803541 228844 2223045 759592 222929 716606 2167637 438507 1670290 394884 1288726 4357290

5% critical value 53 53 53 50 57 53 50 50 53 53 53 47

Notes: Retail bank interest rates include the following categories: lending rates on loans to non–financial cooperations (‘NFC’), lending rates on mortgage loans (‘MOR’), lending rates on consumer credit (‘CON’) and deposit rates on deposits offered to private households (‘DEP’). For each retail bank rate the following maturity categories are considered: total, i.e. the average over all maturities (‘total’), up to 1 year (‘–1y’) and over 1 year (‘1y+’). The variance statistic refers to the multi–equation version of the Chow–Wald test suggested by Stock and Watson (2005).

3.1. Baseline model specifications 3.1.1. Impulse responses We compute impulse responses to the identified structural shocks for both periods 2003–07 and 2008–11. As we are concerned about the adjustment of the retail bank interest rate spreads to the structural shocks, we determine the response of the spreads by calculating the difference between the reaction of the respective retail bank interest rate and the overnight money market rate. The results are summarized in Figure 2, which displays the impact reaction of the spreads to the shocks.14 The structural shocks are normalized to unit shocks to ensure comparability across models and sub–samples. The sign of the shocks is chosen such that the money market rate falls in response to each structural shock. The symbols refer to the retail bank rates used to 14

The entire impulse responses of the variables to the shocks are not reported here, but are available upon request.

12

calculate the spread over the money market rate. If the spread is positive, the decrease in the retail bank interest rates is smaller than the drop in the money market rate and the pass–through is incomplete. If the spread is zero, retail bank interest rates move proportionally with the policy rate and the pass–through is complete. If the spread is negative, the retail bank interest rates are overshooting the drop in the money market rate. Large symbols indicate responses of the spread that are significant at the 10% level. The horizontal axis groups the models according to both, sub–sample periods and maturities of retail bank interest rates. Figure 2: Impact impulse responses of spreads As a general result the pass–through from changes in the money market rate to retail bank rates turns out to be less complete during the crisis. While in the pre–crisis period the pass–through is generally complete as the impact response of the spreads is insignificantly different from zero, the spreads are higher on impact following a decrease in the money market rate in the period 2008–11. Figure 2 also reveals that retail bank interest rates with longer maturities are typically stickier than those with short maturities, in particular during the crisis period. Changes in money markets rates are passed–through much faster to loan and deposit rates with maturities below one year, whereas banks seem to be more reluctant concerning loan contracts and deposits with maturities above one year. A further interesting result is that the source of the shocks seems to matter for the impact response of the spread, which supports our empirical approach to investigate the interest rate pass–through using an identified structural VAR instead of bivariate error–correction models. The largest increase of the spread results from a monetary policy shock (Panel A), which is statistically significant for longer maturities during the crisis period. While an aggregate demand shock of the same unit size (Panel B) also leads to a significant widening of the spread for the period 2008–11, the increase turns out to be much smaller compared to a monetary policy shock. In the 13

case of an aggregate supply shock (Panel C) almost all spread responses are insignificantly different from zero. In Section 4 below we will show that the quantitatively different impact of the structural shocks on the spread is in line with the predictions from a theoretical DSGE model. 3.1.2. Variance decomposition Of course, the extent to which the spread widens depends on the size of the shock hitting the economy. Figure 3 shows that the average size of aggregate demand shocks is much larger than that of monetary policy and aggregate supply shocks. For an analysis of the quantitative importance of the structural shocks we compute the forecast error variance decomposition, which in contrast to the impulse response analysis takes into account the estimated standard deviation of the shocks. Table 3 reports the forecast error variance decomposition of each variable at the 1– to 5–year forecast horizon as an average over all model specifications. Figure 3: Standard deviation of structural shocks Aggregate demand shocks explain most of the variation of real GDP and interest rates both, before and during the crisis, while monetary policy shocks and to a lesser extent also aggregate supply shocks seem to be most relevant for explaining fluctuations in the aggregate price level. The impact of the financial crisis becomes most obvious with regard to the sources of fluctuations of real GDP. Monetary policy and aggregate supply shocks lost most of their explanatory power in the period 2008–11, whereas aggregate demand shocks seemed to be the driving force behind the recession (see De Nicol`o and Lucchetta, 2011, for similar results). The final column, which sums up the contribution of all identified shocks to the variation of the endogenous variables, shows that the role of additional (unidentified) shocks also seems to have increased during the crisis. 3.1.3. Structural shock volatility Figure 3 also supports our results of the test for a structural break in the shock variances between the two sub–samples. On average, structural shocks 14

Table 3: Forecast error variance decomposition

Time period Real GDP 1st 2nd 3rd 4th 5th GDP deflator 1st 2nd 3rd 4th 5th Money market 1st rate 2nd 3rd 4th 5th Retail bank 1st rate 2nd 3rd 4th 5th

Monetary policy shock 03-07 08-11 24 9 23 8 23 8 23 8 23 8 38 21 38 19 37 18 37 18 37 18 13 7 11 6 13 6 13 6 13 6 17 16 13 13 14 13 14 13 14 13

Aggregate supply shock 03-07 08-11 22 5 19 8 19 9 19 9 19 9 23 35 20 30 20 29 20 28 20 28 9 12 10 11 10 11 10 11 10 11 18 19 16 16 16 16 16 16 16 16

Aggregate demand shock 03-07 08-11 31 54 36 52 36 52 36 52 36 51 20 16 23 22 23 24 24 24 24 24 52 48 52 51 51 51 51 51 51 51 38 30 45 37 44 38 44 38 44 38

Sum 03-07 08-11 77 69 78 68 78 69 78 69 78 69 80 72 81 70 81 70 81 70 81 70 74 68 74 68 74 68 74 68 74 68 73 65 74 66 74 67 74 67 74 67

Notes: The Table shows the percentage fraction of each variable’s forecast error variance explained by the structural shocks. It is computed as the mean over all 12 model specifications.

have been much larger during the crisis period. While the standard deviation of monetary policy and aggregate supply shocks was about 1.5 times higher in the years 2008–11, the average size of aggregate demand shocks increased by a factor of 3.5 (see also Table 4 below in Section 3.2 for the average standard deviations over all model specifications). Thus, the large interest rate spreads that have been observed during the years 2008–11 are not only due to distortions in the transmission of monetary policy caused by structural changes, but also a result of the on average larger shocks hitting the euro area economies during the crisis. In addition to this, the role played by the unidentified shocks, which summarize all remaining structural shocks, has 15

increased over time. 3.2. Country groups A crucial assumption underlying our empirical approach is the homogeneity of the slope parameters across countries. Since in the periphery countries (Greece, Ireland, Italy, Portugal and Spain) the interest rate spreads were significantly larger and more persistent than in the core countries of the euro area (Austria, Belgium, Finland, France, Germany and the Netherlands), our main result of the previous Section that the pass–through from money market rates to retail bank interest rates has significantly weakened during the crisis, could be driven by those periphery countries. In order to see whether the impact impulse responses of the spreads are affected by the periphery countries, we split our panel into a core country panel and a periphery country panel. We re–estimate panel VAR models for both country groups separately by using the same setup as in the baseline models and identify the structural shocks by imposing identical sign restrictions.15 For each of the 500 draws we then calculate the difference of the impact impulse responses of the spreads between the sub–panel and the entire euro–area panel of Section 3.1. The results of this exercise are summarized in Figures 4 and 5, which show that in general the differences in the interest rate pass–through in the two sub–panels across the two sub–periods only insignificantly deviate from zero. From this finding we conclude that a change in bank behavior concerning the propagation of structural shocks during the crisis period can be observed in both, the core countries and the periphery countries of the euro area. Thus, the periphery countries are not driving our baseline findings regarding a decrease in the interest–rate pass–through obtained in 3.1. Figure 4: Difference of impact impulse responses of spreads between core countries and the entire euro area

15

Note that the following results should be interpreted cautiously. Since the money market rate equation in VAR models is typically interpreted as the central bank’s reaction function, the separation of countries is somewhat at odds with the notion of a single monetary policy.

16

Figure 5: Difference of impact impulse responses of spreads between periphery countries and the entire euro area However, the larger spreads in the periphery countries that can be observed in the data after the beginning of the financial crisis seem to be attributable to at least some extent to increases in the volatility and/or changes in the relative importance of the structural shocks. Table 4 shows that during the crisis the standard deviation of aggregate demand shocks is much larger in the periphery countries than in the core countries. Moreover, the unidentified fourth shock also plays a larger role in the periphery countries during the years 2008–11. This result is also confirmed by the forecast error variance decomposition for the two sub–panels (see Tables D.1 and D.2 in Appendix D). While for the core countries the three identified shocks explain about the same portion of the variances of the endogenous macroeconomic variables across the two sample periods, their explanatory power decreases for the periphery–countries panel. Thus, our conclusion drawn based on Table 3 that the increase in spreads since the onset of the global financial crisis may be partly due to the increased importance of the fourth unidentified shock seems to be primarily driven by the periphery countries. Table 4: Average standard deviation of structural shocks 2003-07 2008-11 All Core Periphery All Core Periphery countries countries countries countries countries countries Shocks Monetary policy 0.006 0.004 0.007 0.010 0.010 0.010 0.004 0.003 0.006 0.006 0.004 0.008 Aggregate supply 0.105 0.105 0.104 0.367 0.300 0.416 Aggregate demand 0.156 0.135 0.180 0.273 0.219 0.314 Unidentified shock Notes: The average standard deviation of structural shocks is computed as the mean over all 12 model specifications.

3.3. Common monetary policy shock In our panel VAR models with all euro–area member countries included the identification of the monetary policy shock requires some words of caution. Since the models are estimated by assuming slope homogeneity, cross– country differences are captured in the regressions either by the country fixed 17

effects ci or the reduced–form residuals εi,t . From the latter follows that since the identification matrix (U Ω1/2 Q)−1 is common to all member countries, the MP sequence of structural monetary policy shocks ηi,t differs across countries, which is at odds with the idea that all member countries of a monetary union should be hit by the same monetary policy shock. We seek to shed some light on this problem by adopting a two–step procedure. First, for every panel VAR model we generate a series of union–wide monetary policy shocks. The common monetary policy shocks are estimated as the first common factor F1 from the country–specific monetary policy shock series. For each of the 500 draws the common factors are extracted using: MP ηi,t − μi = li1 F1t + . . . + lik Fkt + ζit , (3.1) where i denotes the country, μi is the mean of the country–specific monetary policy shock, F1t , . . . , Fkt are the k unobservable common factors, li1 , . . . , lik are the factor loadings, and ζit is an i.i.d. white–noise error term. The first common factor F1 is standardized to zero mean and unit standard deviation. Second, in every panel VAR model as specified in Section 3.1 we include the respective first common factor as an exogenous variable, such that: Yi,t = ci +

p 

Aj Yi,t−j + BηtM P + εi,t ,

(3.2)

j=1

where ηtM P denotes the respective standardized monetary policy shock that is common to all countries in the euro area. For every model specification the impact of the respective exogenous monetary policy shock on the endogenous variables is finally analyzed by calculating the dynamic multipliers of (3.2). Figure 6 shows the impact responses of the spreads to a common monetary policy shock. Overall, the results are quite similar to those obtained before. During the financial crisis retail bank interest rates responded more sluggishly to changes in money market interest rates following a monetary policy shock of unit size than in the years before the crisis, where the impact reaction of the interest rate spreads was insignificant. Figure 6: Impact impulse responses of spreads to a common monetary 18

policy (unit) shock

4. Discussion Our empirical results suggest that the widening of various retail bank spreads during the financial crisis can to a large extent be explained by structural changes in the transmission of policy rate movements to loan and deposit rates. To interpret these findings and to obtain a deeper economic intuition about the possible structural changes underlying our results we resort to a macroeconomic DSGE model, which was developed by Gerali et al. (2010) and which has a number of advantages. First, in this model the banking sector is attributed a non–trivial role in the propagation of macroeconomic shocks. The banks’ business consists of issuing collateralized loans to entrepreneurs and households, collecting deposits, and accumulating capital out of retained earnings. All these activities are subject to balance–sheet constraints, which establish a link between the business cycle, bank profits, bank capital, and the supply and cost of loans. Accordingly, margins charged on loans and deposits do not only depend on interest rate adjustment costs, but also on banks’ capital–to–assets ratios. Second, this theoretical framework belongs to the class of medium– scale DSGE models, which has a sufficiently rich structure to replicate the cyclical properties of several macroeconomic aggregates. Over the past fifteen years these types of models have become increasingly popular among central banks and other institutions which use them for economic analysis and forecasting purposes. Finally, Gerali et al. (2010) estimated the model for the euro area using data for the period 1998Q1–2009Q1 and showed that their framework fits euro–area data sufficiently well. 4.1. Simulation exercise setup In our simulation exercise we compute the impact reaction of retail bank rate spreads to the same types of shocks as in our empirical analysis for two different steady–state regimes. The first steady state corresponds to the environment prior to the outbreak of the financial crisis, which is assumed to mimic normal conditions. This regime is calibrated by using the 19

parameter values estimated by Gerali et al. (2010). The second steady state corresponds to the environment after the outbreak of the financial crisis, which is calibrated on the basis of empirical observations as well as theoretical suggestions. A comparison between the impact responses of retail bank spreads across regimes reveals changes in the direction and the degree of the pass–through of movements in the policy rate to loan and deposit rates. In particular, we account for possible structural changes between the normal regime and the crisis regime by assuming that the latter is characterized by less favorable values of several parameters which shape agents’ behavior in loan and deposit markets. The financial crisis severely affected banks in the euro area. Large loan losses and an increasing degree of financial distress caused substantial changes in their ability and willingness to lend as well as in their external funding opportunities (see European Central Bank, 2010; Sinn, 2010). This is supported by Figure 7 which displays some results of the ECB Bank Lending Survey along with a measure of concentration in the banking sector and a measure of the volatility of the interbank money market rate. Each pair of bars corresponds to the averages for the periods 2003–07 and 2008–11. In the course of the global financial crisis banks have notably tightened their collateral requirements, which suggests that average loan–to–value (LTV) ratios have significantly increased since 2008. Furthermore, over the same period the fraction of banks pointing to deteriorations in their capital position as a reason for reducing loan supply was systematically higher than before 2008. This evidence suggests that during the crisis commercial banks might have been confronted with more difficulties and higher costs of raising or adjusting their equity capital position. Figure 7: Stylized facts Cottarelli and Kourelis (1994) argue that during episodes of financial distress banks may face a less elastic demand for loans, as customers asking for loans in incomplete financial markets may find it difficult to tap alternative sources of external funds due to higher problems of informational asymmetries. Thus, the degree of competition between banks may have decreased 20

since the beginning of the financial crisis, which is empirically supported by the marked increase of the Herfindahl index (HI) for credit institutions in most euro area countries. Figure 7 shows that in the period 2008–11, the average level of the HI was by 10.2% higher than in the preceding five years (2003–07).16 Additionally, during such a crisis money markets typically become less liquid. As a result, fluctuations in the money market rate tend to become noisier as reflected by the higher volatility in the EONIA since 2008 (see Figure 7). The higher degree of noise in money markets aggravates the uncertainty regarding the precise nature of their fluctuations (Cottarelli and Kourelis, 1994). Accordingly, it becomes more difficult and costly for banks to disentangle policy–induced movements in interbank rates from purely random fluctuations. Simple signal extraction arguments suggest that in an environment characterized by noisier interbank markets a larger fraction of the policy–related movements in the money market rate are not adequately transmitted to retail bank rates. We account for the aforementioned effects of the financial crisis on collateral requirements, bank capital positions, competition among banks and interest rate adjustment costs by assuming that the crisis regime in the theoretical model is characterized by lower LTV ratios, higher costs for adjusting banks’ equity ratio, a lower elasticity of substitution between the loans provided by individual banks, and – to approximate the effects of noisier money market rates – a higher value of the parameter measuring the degree of interest rate stickiness. The parameter values for the crisis regimes, which are specified in Table 5 together with parameters of the normal regime, are chosen arbitrarily and hence only illustrate the qualitative, not the quantitative, implications of the model. In fact, the higher the parameter deviation from the normal regime, the stronger the change in the impact response of the 16

The HI is a measure of concentration in a specific sector of an economy. The higher the index, the higher the concentration in the sector and the lower the degree of competition. The HI for credit institutions is calculated by the European Central Bank. The euro–area aggregate is constructed as weighted average of the national HIs, with the weights being based on total nominal assets of each country’s monetary and financial institutions.

21

loan and deposit rate spreads reported below. Table 5: Setup of simulation exercise Model parameters: Collateral requirements LTV ratio entrepreneurs LTV ratio households Bank capital adjustment Cost parameter Degree of competition Loans to firms Loans to households Interest rate stickiness Loans to firms Loans to households Deposits

Normal regime

Crisis regime

0.35 0.70

0.15 0.50

11.07

31.07

3.12 2.79

2.12 1.79

9.36 10.09 3.50

18.72 20.18 7.00

Notes: Calibrated model parameters for the normal regime and the crisis regime. Parameter values for the normal regime are taken from Gerali et al. (2010).

We simulate the impact reaction of the retail bank spreads to the same exogenous shocks as in our empirical analysis, i.e. an expansionary monetary policy shock, a positive aggregate supply shock and a negative aggregate demand shock. The aggregate supply shock is modeled as a temporary change in technology, and the aggregate demand shock as a temporary change in households’ preferences.17 The sign of each shock is chosen such that the central bank lowers the policy rate on impact. 4.2. Simulation exercise results Figure 8 compares the impact responses of retail bank spreads in the model economy to macroeconomic shocks under the different regimes. To make the impulse responses of the theoretical model comparable with the empirical responses, the shocks are normalized to unit shocks. We denote the regime with normal conditions by (I), while the crisis regimes are summarized by (II–V), where (II) refers to lower loan–to value ratios, (III) to higher bank 17

The results reported below would be the same, if the preference shock was replaced by a government spending shock and the technology shock by a cost push shock, which shifts the goods market Phillips curve upwards.

22

capital adjustment costs, (IV) to a lower degree of competition and, (V) to higher interest rate adjustment costs. Circles refer to loans to households, triangles to loans to firms, and diamonds to deposits. Figure 8: Simulation of impact reaction of retail bank spreads to (unit) shocks under different regimes Overall, our findings indicate a tendency for a widening of the retail bank spreads when the economy moves from the normal regime (I) to one of the crisis regimes (II–V). From this we conclude that changes in certain structural parameters of commercial banks’ behavior might have been responsible for the decrease of the interest rate pass–through since the onset of the global financial crisis. The model simulations also provide some support for our empirical finding of a quantitatively stronger impact of the crisis on retail bank rate setting following a monetary policy shock. In particular, if the crisis regime is characterized by lower LTV ratios (II) or a lower degree of competition in the banking sector (IV), the spread increase is significantly larger than in the case of aggregate demand or aggregate supply shocks of the same unit size. This shock–dependent change in the response of the spread is less pronounced, if either higher bank capital adjustment costs (III) or a higher degree of interest rate stickiness (V) is assumed to simulate the crisis. For these two regimes spreads are significantly larger than in the pre–crisis regime irrespective of the nature of the structural shocks. A possible interpretation of this result would be that in the crisis changes in private agents’ ability to borrow (LTV ratios) or in the degree of competition in the loan market might have been more important than changes in the ability of banks to adjust equity capital or their retail rates. 4.3. Intuition of model simulation results To develop intuition about the impact effects of various types of shocks on retail bank rate spreads in the model of Gerali et al. (2010), it suffices to take a closer look at the equations describing the optimal loan rate setting in the wholesale and in the retail banking sector. Since the structure of the first 23

order conditions for the optimal choice of the deposit rate is analogous, we do not discuss them explicitly.18 The representative wholesale bank operates under perfect competition. The wholesale bank provides loans to retail banks at rate Rtb , which is set according to  Rtb

= rt − κKb

Ktb − νb Bt



Ktb Bt

2 ,

(4.1)

where Bt is the amount of wholesale loans, Ktb is the wholesale bank’s capital position and rt is the policy rate. The dependence on the capital–to–asset ratio Ktb /Bt is due to the assumption that it is costly to deviate from its desired (or required) value ν b . The parameter κKb governs the severity of the corresponding adjustment cost function. A higher κKb implies that it is more costly to change the capital–to–asset ratio. Retail banks operate in a monopolistically competitive environment. They face quadratic adjustment costs of changing individual loan rates. The first order condition for the representative interest rate rtbs charged on loans of type s reads  bs  bs b rt r bs bs Rt 0 = 1 −  +  bs − κbs − 1 bst bs rt rt−1 rt−1  P  bs   bs  s  λt+1 r rt+1 bt+1 , (4.2) +βP Et κbs t+1 −1 P bs bst λt rt rtbs where bst is the loan volume, λPt denotes marginal utility of the bank owners, βP is a discount factor, bs is the elasticity of substitution between loan products of type s and κbs measures the intensity of the loan rate adjustment costs. Linear approximations of equation (4.1) ˆt − B ˆt ), dRtb = drt − κKb (ν b )3 (K 18

(4.3)

See Gerali et al. (2010) for a more detailed description of the theoretical framework.

24

and equation (4.2) drtbs =

κbs βP κbs bs bs dr + Et drt+1 (4.4) t−1 bs − 1 + (1 + βP )κbs bs − 1 + (1 + βP )κbs bs dRtb , + bs  − 1 + (1 + βP )κbs

show that the impact reaction of the spread between the loan rate and the policy rate is indirectly affected by the size of the wholesale bank’s capital– to–asset ratio. dxt denotes the absolute deviation of a variable xt from its steady–state value, while hatted variables refer to the relative deviation from the steady–state value. After inserting equation (4.3) into equation (4.4) we obtain the impact reaction of the retail bank interest rate spread19 bs ˆt + ϕ1 Et drt+1 drtbs − drt ≈ (ϕ2 − 1)drt + ϕ2 κKb (ν b )3 B ,

(4.5)

where: κbs − 1 + (1 + βP )κbs bs = bs .  − 1 + (1 + βP )κbs

ϕ1 = ϕ2

bs

For the calibrations of the model used in this paper a rise in the current policy rate rt reduces the spread, while the spread widens after an increase bs in either the loan volume Bt or the expected future loan rate Et drt+1 . Since |ϕ2 − 1| is much larger than ϕ1 and ϕ2 κKb ν 3 , the impact reaction of the spread is to a large extent driven by the current policy rate. Movements in expected future retail rates only play a role in the case of aggregate demand and aggregate supply shocks, which, unlike monetary policy shocks, induce more persistent and hump–shaped reactions in interest rates, loan volumes  t and the previous period’s Note that on impact both, the response of bank capital K bs loan rate rt−1 are equal to zero, as these variables are determined by decisions in the past. bs  t is the accumulated profit of the wholesale bank from , K While this is obvious for rt−1 previous periods. 19

25

and various demand aggregates. 4.3.1. Lower LTV ratio The impact reaction of the spread indirectly depends on the size of the LTV ratio, as the latter determines the sensitivity of loan demand to changes in interest rates. A sudden decrease in the policy rate induced by a expansionary monetary policy shock causes a rise in aggregate demand associated with an expansion in loan demand and an acceleration of inflation. However, the increase in loan demand is less pronounced when collateral requirements are tighter, i.e. the LTV ratio is lower. In this case, the expansion in aggregate demand and the increase in the inflation rate is weaker, which implies a less pronounced endogenous reaction of the central bank to the economic upturn. Thus, when the LTV ratio is lower, an expansionary monetary policy shock is associated with a stronger decrease in the policy rate, which, in turn, widens the impact spread (see equation 4.5).20 A negative aggregate demand shock reduces the marginal utility of consumer goods relative to housing services and leisure and induces a re–balancing of demand from consumption of non–durables to housing. As a consequence, relative house prices increase which, everything else equal, loosens the collateral constraint, thereby enabling households to expand borrowing and to maintain a higher level of consumption. However, with a lower LTV ratio, the effect of easing the collateral constraint is weaker. Accordingly, in the crisis regime characterized by lower LTV ratios, a negative preference shock implies a larger decrease in aggregate consumption and a stronger deceleration in inflation. Thus, the central bank lowers the policy rate more aggressively and, as a result, the impact spread widens.21 If the economy is hit by a positive technology shock, which is associated 20

Note that the weaker increase in loan demand and the stronger decrease in expected future loan rates associated with a lower LTV ratio tend to dampen the widening of the spread. However, as noted above, the latter two effects are dominated by the one resulting from changes in the policy rate. 21 In the regime characterized by a lower LTV ratio the increase in the loan volume is smaller and the expected future loan rate decreases more strongly. While both effects work towards lowering the spread, they are dominated by the effect of the policy rate on the retail rate spread.

26

with a decline in both, inflation and the policy rate, a lower LTV ratio also weakens the pass–through of policy rate changes to loan rates. Compared to an economy with normal collateral requirements, the increase in loan demand and consumption will be less pronounced and the decline in inflation and the policy rate will be larger. Accordingly, the increase in the impact spread will be stronger, even though the smaller expansion of the loan volume and the more pronounced negative response of expected future rates again put a downward pressure on it. To sum up, the change in the impact response of the spread is substantially larger when the economy is hit by a monetary policy shock. This is due to the financial accelerator effect, which renders the collateral constraints more sensitive to changes in the LTV ratio. By contrast, aggregate supply or aggregate demand shocks induce movements in nominal prices, which partially offset each other and, as a consequence, limit the impact of the tighter collateral constraint. Thus, in the case of real shocks, changes in the LTV ratio are of lesser importance. 4.3.2. Higher bank capital adjustment costs An increase in bank capital adjustment costs, i.e. a higher κKb , makes the loan rate spread in wholesale banking dRtb − drt more sensitive to changes in the loan volume Bt (see equation 4.3). Consequently, the loan rate spread in retail banking drtbs − drt also becomes more exposed to changes in the loan volume, as the wholesale bank rate Rtb is an important determinant of the loan rate charged by retail banks. If the economy is hit by an expansionary monetary policy shock, the loan volume increases and, everything else equal, the wholesale spread widens by t | κKb (ν b )3 |B on impact. The magnitude of this positive reaction is stronger for higher values of κKb . The intuition with respect to a negative aggregate demand and a positive technology shock is similar, as they both lead to an increase

27

in the loan volume Bt .22 The reaction of the loan rate spread is further magnified by a self– reinforcing loop stemming from the direct link between the policy rate and the spread itself. For any given initial reaction of the policy rate, if the spread turns out to be higher (e.g. due to a higher κKb ), then the collateral constraints end–up being relatively tighter, which is associated with a more depressed borrowers’ consumption and a lower inflation rate. The monetary authority sets a relatively lower policy rate, which, in turn, positively affects the loan rate spread (see equation 4.5). The impact of higher bank capital adjustment costs on the interest rate setting behavior of retail banks is similar across shocks, which is due to two opposing mechanisms largely offsetting each other. On the one hand, a monetary policy shock induces an expansion of loan volumes, which is larger than after an aggregate demand or an aggregate supply shock of similar magnitude. Since a higher κKb increases the coefficient of the loan volume in equation (4.5), the impact spreads widens more in the case of a monetary policy shock. On the other hand, if κKb is higher, aggregate demand and aggregate supply shocks induce a more muted decline in future retail bank rates, which positively contributes to the impact response of the spread. 4.3.3. Lower elasticity of substitution between loan varieties When setting loan rates retail banks compare the opportunity cost of deviating from the optimal interest rate (which would prevail under perfectly flexible interest rates) with an increasing marginal costs of interest rate adjustment. For any given shock to marginal costs, the lower the curvature of the profit function, the lower its sensitivity with respect to deviations from the optimal retail bank rate and thus, the smaller the desired adjustment bs . It is straightforward to show that the curvature of the banks’ rtbs − rt−1 profit function is positively related to the degree of substitutability between individual bank products bs . If bs declines, there are two opposing effects on the impact spread (see equation 4.4). On the one hand, the link between the 22

See the discussion in Section 4.3.1 as well as equation (4.5).

28

retail banks’ interest rates rtbs and their marginal costs Rtb becomes weaker, which attenuates the response of retail bank rates to policy rate changes. On the other hand, a lower degree of competition in the retail banking sector strengthens the relationship between current and expected retail bank rates. For standard calibrations the first effect of a lower bs dominates the second. If the economy is hit by a monetary policy shock, the crisis regime with a lower elasticity of substitution between different retail bank loans bs is associated with more pronounced changes in the interest rate spread compared to an economy that is hit by an aggregate demand or an aggregate supply shock. There are two mechanisms responsible for this result. First, a monetary policy shock leads to a substantially stronger drop in the policy rate on impact and a relatively weak negative response of future retail rates. The opposite is the case when the economy is hit by a real shock. Moreover, a lower bs increases the elasticity of the spread with respect to both, movements in the policy rate and in the expected future retail rate (equation 4.5). Thus, even if the impact response of the current policy rate and future retail rates remained unchanged, a lower bs would be associated with a relatively more pronounced change in the spread in the case of monetary policy shock. Second, the impact increase of the spread following an aggregate demand or an aggregate supply shock is dampened, as a smaller value of bs leads to a larger decline in future retail rates. 4.3.4. Higher interest rate adjustment costs The impact of changes in the banks’ costs of adjusting retail bank rates κbs on their interest rate setting behavior is straightforward. Since higher values of κbs directly increase the degree of retail bank rate stickiness, any change in the policy rate is passed–through more reluctantly by banks. In contrast to a shift to a lower elasticity of substitution between individual loan products bs , the change in the impact response of the spread is similar across different types of shocks. The reason is that in the case of supply and demand side shocks the behavior of future retail rates exerts a positive effect on the spread. A switch to a higher degree of interest rate stickiness leads to a much weaker decline in future retail rates while, at the same time, our

29

calibration implies a negligible increase in the coefficient of future retail rates in equation (4.5). 5. Concluding Remarks This paper uses panel VAR models for euro area member countries to explore the widening of retail interest rate spreads that emerged in the course of the global financial crisis. We examine the interest rate pass–through in the periods 2003–07 and 2008–11 by considering three macroeconomic shocks, namely a monetary policy shock, an aggregate demand shock and an aggregate supply shock. Our results can be summarized as follows. First, while the pass–through of policy rate changes in response to macroeconomic shocks was generally complete before the start of the financial crisis, it became significantly less complete thereafter. This finding holds for different interest rate categories of retail bank interest rates, i.e. lending rates and deposit rates with different maturities. In addition, the widening of spreads is also due to a significant increase in the average size of structural shocks and, to some extent, to the larger role of crisis–specific shocks. Thus, the financial crisis has led to structural changes in both, the propagation of macroeconomic shocks and their stochastic properties. Second, splitting our panel into a core country group and a periphery country group reveals that the decrease in the interest rate pass–through is statistically not different across these groups and hence driven by structural changes in all euro area member countries. Moreover, the systematic increase in the volatility of the structural shocks as well as the increased importance of the additional unidentified shocks seem to be particularly relevant for explaining the relatively strong increase in interest rate spreads that has been observed in the periphery countries. Third, the extent to which interest rate spreads react crucially depends on the nature of the structural shocks. Following a monetary policy shock the quantitative reaction of the impact spread turns out to be much larger than after an aggregate demand or an aggregate supply shock of the same unit

30

size. This empirical result is in line with the predictions from a theoretical DSGE model with bank intermediation and financial frictions. Fourth, simulations of such a DSGE model show that the interest rate pass–through becomes more incomplete in response to tighter collateral requirements, higher costs of restoring the bank capital position, weaker competition among banks and higher costs of adjusting interest rates. Empirical evidence shows that these structural parameters have substantially changed in the course of the financial crisis. Thus, the distortions of the transmission of monetary policy during the crisis seem to have been caused by increasing distress in the banking sector, which has led to a fundamental change in the propagation of macroeconomic shocks. Future work might focus on two research questions. First, it would be interesting to identify the nature of additional shocks hitting the euro area during the financial crisis. Besides adverse loan supply shocks such shocks might have been caused by a deterioration of confidence or a lack of trust in the stability of the financial system. Second, the unconventional monetary policy measures taken by the ECB in response to the distortions of the transmission mechanism should be evaluated in the light of our results. Above all, the role played by sovereign bond yields should be addressed more explicitly, as its dispersion across euro area member countries was used by the ECB as one of the main arguments in favor of the Outright Monetary Transactions, i.e. the ECB’s sovereign bond purchasing program, which was announced in 2012. Acknowledgement. We are grateful to two anonymous referees for helpful comments and suggestions. We also thank the participants of the CESifo Area Conference on Macro, Money and International Finance 2012, the 28th Annual Congress of the European Economic Association, 17th Annual International Conference on Macroeconomic Analysis and International Finance and the Economics Research Seminars at the Universities of Innsbruck and Aachen for fruitful discussions. The usual disclaimer applies. This paper is part of the research project: “Transmission und Emission makro¨okonomischer Schocks durch das Bankensystem”. Financial support by the Stiftung Geld

31

und W¨ahrung is gratefully acknowledged.

32

Aristei, D., Gallo, M., 2014. Interest rate pass-through in the euro area during the financial crisis: a multivariate regime-switching approach. Journal of Policy Modeling 36, 273–295. Belke, A., Beckmann, J., Verheyen, F., 2013. Interest rate pass-through in the EMU: New evidence from nonlinear cointegration techniques for fully harmonized data. Journal of International Money and Finance 37, 1–24. Blot, C., Labondance, F., 2011. Bank interest rate pass-through in the eurozone: monetary policy transmission during the boom and since the financial crash. mimeo. Canova, F., 2007. Methods for Applied Macroeconomic Research. Princeton University Press. Canova, F., De Nicol` o, G., 2002. Monetary disturbances matter for business fluctuations in the G-7. Journal of Monetary Economics 49, 1131–1159. Canova, F., Paustian, M., 2010. Measurement with some theory: a new approach to evaluate business cycle models. Economics working papers, Department of Economics and Business, Universitat Pompeu Fabra. Ciccarelli, M., Maddaloni, A., Peydr´ o, J.-L., Jul. 2010. Trusting the bankers: A new look at the credit channel of monetary policy. Working Paper Series 1228, European Central Bank. Cottarelli, C., Kourelis, A., 1994. Financial structure, bank lending rates, and the transmission mechanism of monetary policy. IMF Staff Reports 41 (4), 587–623. Curdia, V., Woodford, M., 2010. Credit spreads and monetary policy. Journal of Money, Credit and Banking 42 (6), 3–35. De Bondt, G. J., 02 2005. Interest rate pass–through: Empirical results for the euro area. German Economic Review 6 (1), 37–78. De Nicol` o, G., Lucchetta, M., April 2011. Systemic risks and the macroeconomy. In: Quantifying Systemic Risk. NBER Chapters. National Bureau of Economic Research, Inc. Eickmeier, S., Hofmann, B., Worms, A., 05 2009. Macroeconomic fluctuations and bank lending: Evidence for Germany and the euro area. German Economic Review 10, 193–223. European Central Bank, 2010. The ECB’s response to the financial crisis. Monthly Bulletin October, 59–74.

33

European Central Bank, 2011. The ECBs non–standard measures – impact and phasing–out. Monthly Bulletin July, 55–69. European Central Bank, 2013. Assessing the retail bank interest rate pass–through in the euro area at times of financial fragmentation. Monthly Bulletin August, 75–91. Farrant, K., Peersman, G., June 2006. Is the exchange rate a shock absorber or a source of shocks? New empirical evidence. Journal of Money, Credit and Banking 38 (4), 939–961. Gal´ı, J., L´ opez-Salido, J. D., Vall´es, J., May 2003. Technology shocks and monetary policy: Assessing the Fed’s performance. Journal of Monetary Economics 50 (4), 723–743. Gerali, A., Neri, S., Sessa, L., Signoretti, F. M., 2010. Credit and banking in a DSGE model of the euro area. Journal of Money, Credit and Banking 42 (6), 108–141. Gertler, M., Karadi, P., January 2011. A model of unconventional monetary policy. Journal of Monetary Economics 58 (1), 17–34. Goddard, J., Molyneux, P., Wilson, J. O., Tavakoli, M., July 2007. European banking: An overview. Journal of Banking & Finance 31 (7), 1911–1935. Hansen, N.-J. H., Welz, P., Apr. 2011. Interest rate pass-through during the global financial crisis: The case of Sweden. OECD Economics Department Working Papers 855, OECD Publishing. Helbling, T., Huidrom, R., Kose, M. A., Otrok, C., April 2011. Do credit shocks matter? A global perspective. European Economic Review 55 (3), 340–353. Hristov, N., H¨ uleswig, O., Wollmersh¨ auser, T., 2012. Loan supply shocks during the financial crisis: Evidence for the euro area. Journal of International Money and Finance 31 (3), 569–592. Karagiannis, S., Panagopoulos, Y., Vlamis, P., May 2010. Interest rate passthrough in Europe and the US: Monetary policy after the financial crisis. Journal of Policy Modeling 32 (3), 323–338. Kleimeier, S., Sander, H., July 2006. Expected versus unexpected monetary policy impulses and interest rate pass–through in euro–zone retail banking markets. Journal of Banking & Finance 30 (7), 1839–1870. Kok Sørensen, C., Werner, T., Jan. 2006. Bank interest rate pass–through in the euro area: A cross country comparison. Working Paper Series 580, European Central Bank.

34

Kollmann, R., Enders, Z., M¨ uller, G. J., April 2011. Global banking and international business cycles. European Economic Review 55 (3), 407–426. Kwapil, C., Scharler, J., March 2010. Interest rate pass–through, monetary policy rules and macroeconomic stability. Journal of International Money and Finance 29 (2), 236–251. L¨ utkepohl, H., 1993. Introduction to Multiple Time Series Analysis, 2nd Edition. Springer. Neri, S., Jun. 2013. The impact of the sovereign debt crisis on bank lending rates in the euro area. Questioni di Economia e Finanza (Occasional Papers) 170, Bank of Italy, Economic Research and International Relations Area. Nickell, S. J., November 1981. Biases in dynamic models with fixed effects. Econometrica 49 (6), 1417–26. Peersman, G., 2005. What caused the early millennium slowdown? Evidence based on vector autoregressions. Journal of Applied Econometrics 20, 185–207. Pesaran, M. H., Smith, R. P., 1995. Estimating long–run relationships from dynamic heterogenous panels. Journal of Econometrics 68, 79–113. Rebucci, A., September 2010. Estimating VARs with long stationary heterogeneous panels: A comparison of the fixed effect and the mean group estimators. Economic Modelling 27 (5), 1183–1198. Sander, H., Kleimeier, S., April 2004. Convergence in euro–zone retail banking? What interest rate pass–through tells us about monetary policy transmission, competition and integration. Journal of International Money and Finance 23 (3), 461–492. Scholl, A., Uhlig, H., September 2008. New evidence on the puzzles: Results from agnostic identification on monetary policy and exchange rates. Journal of International Economics 76 (1), 1–13. Sinn, H.-W., 2010. Casino Capitalism. How the Financial Crisis Came about and What Needs to Be Done Now. Oxford University Press, Oxford. Stock, J. H., Watson, M. W., 09 2005. Understanding changes in international business cycle dynamics. Journal of the European Economic Association 3 (5), 968–1006. Straub, R., Peersman, G., Jun. 2006. Putting the New Keynesian model to a test. IMF Working Papers 06/135, International Monetary Fund.

35

Tillmann, P., 2013. Capital inflows and asset prices: Evidence from emerging Asia. Journal of Banking & Finance 37, 717–729. Toolsema, L. A., Sturm, J.-E., de Haan, J., 2001. Convergence of monetary transmission in EMU new evidence. CESifo Working Papers 465. Uhlig, H., 2001. Did the Fed surprise the markets in 2001? A case study for VARs with sign restrictions. CESifo Working Paper Series 629. Uhlig, H., 2005. What are the effects of monetary policy on output? Results from an agnostic identifcation procedure. Journal of Monetary Economics 52, 381–419. Van Leuvensteijn, M., S¨ orensen, C. K., Bikker, J. A., van Rixtel, A. A., April 2013. Impact of bank competition on the interest rate pass-through in the euro area. Applied Economics 45 (11), 1359–1380.

36

Appendix A. Data We use data for 11 euro area member countries that is taken from the databases of Eurostat and the ECB covering the period from 2003Q1 to 2011Q4. The panel of countries includes Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Portugal and Spain. The data comprises: 1. Real GDP (yt ): Gross domestic product at market prices, calendar and seasonally adjusted, in constant 2000 EUR (Eurostat). 2. GDP deflator (pt ): Price index, 2000=100, gross domestic product at market prices, calendar and seasonally adjusted (Eurostat). 3. Money market rate (st ): Euro Overnight Index Average (EONIA), in percent (ECB). 4. Retail bank rate (rt ): Various retail bank rates with different time to maturity, in percent (ECB, MFI database). The retail bank interest rates include the following categories: lending rates on loans to non–financial cooperations (‘NFC’), lending rates on mortgage loans (‘MOR’), lending rates on consumer credit (‘CON’) and deposit rates on deposits offered to private households (‘DEP’). For all retail bank interest rates we consider three different maturity categories: total, i.e. the average over all maturities (‘total’), up to 1 year (‘–1y’) and over 1 year (‘1y+’).

37

B. VAR model specification with respect to country–data availability For the different countries the availability of data on the respective retail bank interest rates is mixed. Accordingly, in the panel VAR models for every retail bank interest rate category, the number of countries varies. Table B.1 summarizes the countries, which are excluded from the respective estimated models due to the lack of data availability. Table B.1: Data availability of retail bank interest rates

Retail bank rate NFC

CON

MOR

DEP

Maturity total –1y 1y+ total –1y 1y+ total –1y 1y+ total –1y 1y+

Missing countries Greece – Greece Netherlands Belgium, Netherlands Netherlands Greece Belgium Greece Greece, Netherlands Belgium, Ireland Ireland, Netherlands

Notes: Retail bank interest rates include the following categories: lending rates on loans to non–financial cooperations (‘NFC’), lending rates on mortgage loans (‘MOR’), lending rates on consumer credit (‘CON’) and deposit rates on deposits offered to private households (‘DEP’). For each retail bank rate the following maturity categories are considered: total, i.e. the average over all maturities (‘total’), up to 1 year (‘–1y’) and over 1 year (‘1y+’).

38

C. Structural break tests C.1. Tests for equal residual variances We employ a multi–equation version of the Chow–Wald test proposed by Stock and Watson (2005) to test for the null hypothesis of equal residual variances across both subsamples. The test statistic reads    1,t¯ −1 1,t¯ t¯+1,T t¯+1,T diagΣ − diagΣ diagΣ − diagΣ , (C.1) Ω where Σt1 ,t2 denotes the covariance matrix of the residuals of the panel VAR estimated over the subsample [t1 , t2 ], diag returns a column vector formed from the main diagonal of a matrix and Ω = (diag(Ω1,t¯ + Ωt¯+1,T ))1/2 . Ωt1 ,t2 is the heteroscedasticity–autocorrelation–adjusted covariance matrix of ⎛ ⎞ vecεt1 εt1 ⎜ ⎟ .. ⎜ ⎟, (C.2) . ⎝ ⎠ vecεt2 εt2 where vec stacks the elements of a matrix. The test statistic has an asymptotic χ2 –distribution with degrees of freedom equal to the number of equations n = 4 in the panel VAR model. C.2. Tests for parameter instability A common approach to test for structural breaks is to use Chow tests. The test statistic can be computed as   detΣ1,T − det(Σ1,t¯ + Σt¯+1,T ) /ν , (C.3) det(Σ1,t¯ + Σt¯+1,T )/(M · T − 2ν) where Σ1,T is the covariance matrix of the residuals with no breaks, Σ1,t¯ and Σt¯+1,T are the covariance matrices for the two subsamples, det denotes the determinant of a matrix, and ν = (np + M )n is the number of regressors. The Chow statistic has an asymptotic F –distribution with ν and M · T − 2ν degrees of freedom. A major drawback of the Chow test statistic is that it only has an F – distribution if the residual variance in the two subsamples is equal. If this assumption is violated it is more appropriate to compute a Wald statistic for 39

the null hypothesis of no structural change. It can be constructed as ¯

¯

¯

¯

¯

¯

(A1,t − At+1,T ) (V 1,t + V t+1,T )−1 (A1,t − At+1,T ),

(C.4)

where At1 ,t2 is a vector of all parameter estimates of the panel VAR model (2.1) (including the country–fixed effects) estimated over the period [t1 , t2 ], V t1 ,t2 is the related parameter covariance matrix, and t¯ is 2007Q4. The Wald statistic has an asymptotic χ2 –distribution with degrees of freedom equal to the number of estimated parameters ν in the At1 ,t2 vector.23

23

The number of estimated parameters varies across models due to changes in the number of countries for which retail bank rates are available.

40

D. Forecast error variance decomposition for the two sub–panels Table D.1: Forecast error variance decomposition: core countries

Time period Real GDP 1st 2nd 3rd 4th 5th GDP deflator 1st 2nd 3rd 4th 5th Money market 1st rate 2nd 3rd 4th 5th Retail bank 1st rate 2nd 3rd 4th 5th

Monetary policy shock 03-07 08-11 15 5 13 6 13 7 13 8 14 8 38 20 30 18 28 18 28 18 28 18 9 6 10 7 11 9 11 9 11 9 15 19 11 17 12 16 12 16 12 16

Aggregate supply shock 03-07 08-11 24 9 18 7 18 7 18 7 18 7 25 38 24 30 23 28 23 28 23 27 11 9 8 7 10 7 10 7 10 7 17 11 14 9 14 9 14 9 14 9

Aggregate demand shock 03-07 08-11 31 51 37 56 37 57 37 57 38 57 16 18 25 28 26 30 27 31 27 31 45 51 44 59 43 58 43 58 43 58 40 32 44 44 42 46 43 47 43 47

Sum 03-07 08-11 70 64 69 69 69 71 69 72 69 72 79 76 79 76 78 76 77 76 77 76 66 65 62 72 64 74 64 74 64 74 72 62 68 70 68 71 68 71 68 71

Notes: The Table shows the percentage fraction of each variable’s forecast error variance explained by the structural shocks. It is computed as the mean over all 12 model specifications. Core countries comprise Austria, Belgium, Finland, France, Germany and the Netherlands.

41

Table D.2: Forecast error variance decomposition: periphery countries

Time period Real GDP 1st 2nd 3rd 4th 5th GDP deflator 1st 2nd 3rd 4th 5th Money market 1st rate 2nd 3rd 4th 5th Retail bank 1st rate 2nd 3rd 4th 5th

Monetary policy shock 03-07 08-11 15 14 18 15 22 16 23 17 24 17 30 21 35 21 36 22 35 22 35 22 26 14 16 14 21 16 23 17 24 17 26 24 19 23 22 23 25 24 25 24

Aggregate supply shock 03-07 08-11 14 12 16 20 16 20 16 20 15 20 30 34 24 31 23 30 22 30 22 30 12 9 16 15 16 15 16 16 16 16 21 17 20 17 20 18 19 18 19 18

Aggregate demand shock 03-07 08-11 44 53 41 42 38 40 37 39 37 39 23 13 24 16 25 17 25 17 26 18 41 44 48 40 43 39 42 38 41 37 27 29 40 29 38 30 36 29 36 30

Sum 03-07 08-11 73 78 75 76 76 76 76 76 76 76 83 67 83 69 83 69 83 70 83 70 78 67 80 69 80 70 81 70 81 70 74 70 78 70 79 71 80 71 80 71

Notes: The Table shows the percentage fraction of each variable’s forecast error variance explained by the structural shocks. It is computed as the mean over all 12 model specifications. Periphery countries comprise Greece, Ireland, Italy, Portugal and Spain.

42

Figure 1: Spreads of Euro Area retail bank interest rates over money market rates 8

in percentage points

7 6 5 4 3 2 1 0 -1 2003

2004

2005 NFC

2006

2007 CON

2008 MOR

2009

2010

2011

DEP

Notes: Data are taken from the European Central Bank. The money market rate is the EONIA, which is the average of overnight rates for unsecured euro area interbank lending. The retail lending rates refer to interest rates on new business loans to non–financial corporations (‘NFC’), new loans to households for consumption (‘CON’) and new loans to households for house purchases (‘MOR’). The deposit rates are interest rates agreed on new deposits from private households (‘DEP’). The maturities are an average over all maturities.

43

Figure 2: Impact impulse responses of spreads Panel A: Monetary policy (unit) shock 40 NFC MOR CON DEP

35 30

percentage points

25 20 15 10 5 0 −5 −10

03−07

08−11

03−07

−1y

08−11

03−07

total

08−11 1y+

Panel B: Aggregate demand (unit) shock 1.4 NFC MOR CON DEP

1.2

percentage points

1

0.8

0.6

0.4

0.2

0 03−07

08−11 −1y

03−07

08−11 total

44

03−07

08−11 1y+

Panel C: Aggregate supply (unit) shock 40 NFC MOR CON DEP

30

percentage points

20

10

0

−10

−20

−30

03−07

08−11 −1y

03−07

08−11 total

03−07

08−11 1y+

Notes: The markers show the impact impulse responses of the spread of retail bank interest rates over the overnight money market rate. The spread is calculated from the medians of the impulse responses, which are estimated from a Bayesian vector–autoregression with 500 draws. The structural shocks are normalized to unit shocks. The sign of the shock is chosen such that the money market rate decreases in response to the structural shock (see Table 1). Large symbols indicate responses of the spread that are significant at the 10% level. The horizontal axis groups the models according to both, sub–sample periods and maturities of retail bank interest rates.

45

Figure 3: Standard deviation of structural shocks

−3 x 10 Aggregate supply shock

Monetary policy shock 8 6

0.01

4 0.005 2 0

−1y

total

0

1y+

Aggregate demand shock

−1y

total

1y+

Unidentified shock

0.5 0.6 0.4 0.4

0.3 0.2

0.2 0.1 0

−1y

total

0

1y+

−1y

total

1y+

Notes: For each of the estimated 12 VAR models the markers show the median of the standard deviation of structural shocks. The symbols are chosen as in Figure 2, i.e. red triangle: NFC, green circle: MOR, blue square: CON, black diamond: DEP. The black dashed lines show the 90% confidence band of the pre–crisis sample 2003–07; the red dashed lines are those of the crisis sample 2008–11.

46

Figure 4: Difference of impact impulse responses of spreads between core countries and the entire euro area Panel A: Monetary policy shock (core minus entire euro area) 30 NFC MOR CON DEP

25 20

percentage points

15 10 5 0 −5 −10 −15 −20

03−07

08−11

03−07

−1y

08−11

03−07

total

08−11 1y+

Panel B: Aggregate demand shock (core minus entire euro area) 0.2 0

percentage points

−0.2 −0.4 −0.6 −0.8 −1 −1.2 NFC MOR CON DEP

−1.4 −1.6 03−07

08−11 −1y

03−07

08−11 total

47

03−07

08−11 1y+

Panel C: Aggregate supply shock (core minus entire euro area) 60 NFC MOR CON DEP

50

percentage points

40 30 20 10 0 −10 −20

03−07

08−11 −1y

03−07

08−11 total

03−07

08−11 1y+

Notes: The markers show the difference of the impact impulse responses of spreads between core countries and the entire euro area. The difference is calculated as the median of the differences for each of the 500 draws. Large symbols indicate differences of the spread that are significant at the 10% level. The horizontal axis groups the models according to both, sub–sample periods and maturities of retail bank interest rates. Core countries comprise Austria, Belgium, Finland, France, Germany and the Netherlands.

48

Figure 5: Difference of impact impulse responses of spreads between periphery countries and the entire euro area Panel A: Monetary policy shock (periphery minus entire euro area) 20 NFC MOR CON DEP

15 10

percentage points

5 0 −5 −10 −15 −20 −25 −30

03−07

08−11

03−07

−1y

08−11

03−07

total

08−11 1y+

Panel B: Aggregate demand shock (periphery minus entire euro area) 0.6 0.4

percentage points

0.2 0 −0.2 −0.4 −0.6 NFC MOR CON DEP

−0.8 −1

03−07

08−11 −1y

03−07

08−11 total

49

03−07

08−11 1y+

Panel C: Aggregate supply shock (periphery minus entire euro area) 60 NFC MOR CON DEP

50 40

percentage points

30 20 10 0 −10 −20 −30 −40

03−07

08−11 −1y

03−07

08−11 total

03−07

08−11 1y+

Notes: The markers show the difference of the impact impulse responses of spreads between periphery countries and the entire euro area. The difference is calculated as the median of the differences for each of the 500 draws. Large symbols indicate differences of the spread that are significant at the 10% level. The horizontal axis groups the models according to both, sub–sample periods and maturities of retail bank interest rates. Periphery countries comprise Greece, Ireland, Italy, Portugal and Spain.

50

Figure 6: Impact impulse responses of spreads to a common monetary policy (unit) shock

40 NFC MOR CON DEP

35 30

percentage points

25 20 15 10 5 0 −5 −10

03−07

08−11 −1y

03−07

08−11 total

Notes: See Figure 2.

51

03−07

08−11 1y+

Figure 7: Stylized facts 18

Net percentages

1.6

0.072 2003-07

Percent

2008-11

16

1.4

0.07 14

1.2 0.068

12 1 0.066

10

0.8 8

0.064 0.6

6 0.062 0.4

4 0.06

0.2

2

0

0.058 Collateral reqirements

Bank capital position

0 Herfindahl Index

Overnight rate volatility

Notes: Source: European Central Bank. Bank Lending Survey (BLS), net percentages. Collateral requirements: unbalanced average of the answers to questions 3.B.3, 10.B.1, 10.B.2 and 12.B.1. Bank capital position: answers to question 2.A.1. Bank liquidity position: unbalanced average of the answers to questions 2.A.2 and 2.A.3. Overnight rate volatility: standard deviation of EONIA.

52

Figure 8: Simulation of impact reaction of retail bank spreads to (unit) shocks under different regimes

Monetary policy shock

Aggregate demand shock

58

Aggregate supply shock

12

35

11 56 30 10 54

52

basis points

basis points

basis points

25 9

8 20 50 7 15 48 6

46

I

II

III IV Regime

V

5

I

II

III IV Regime

V

10

I

II

III IV Regime

V

Notes: Symbols summarize the impact reaction of retail bank spreads to the (unit) shocks under different regimes. Regime I - normal conditions; Regime II - lower loan–to–value ratios; Regime III - higher costs of bank capital adjustment; Regime IV - lower elasticity of substitution between loan varieties; Regime V - higher costs of interest rate adjustment. Green circles refer to loans to households, red triangles to loans to firms, and black diamonds to deposits.

53

The Interest Rate Pass-Through in the Euro Area during the Global Financial Crisis

-Highlights-

-

Addresses the stability of the euro area monetary transmission mechanism during the global financial crisis

-

Explores how bank retail interest rates adjust to changes in money market rates

-

Shows that interest rate pass-through became significantly distorted since 2008

-

DSGE model simulations suggest that distortions can be related to tighter frictions that banks are subject to