Ami-Ami (2018) | Site Updates | СКАЧАТЬ: / подробнее
Credit Supply Dynamics and Economic Activity in Euro Area Countries: A Time-Varying Parameter VAR Analysis Martin Bijsterbosch∗

Matteo Falagiarda†

April 8, 2014 ‡

Abstract This paper aims to shed light on the role of credit supply shocks in euro area countries during the recent pre-crisis, bust, and post-crisis periods. A time-varying parameter vector autoregression (TVP-VAR) with stochastic volatility `a la Primiceri (2005) is estimated for each country, and the structural shocks are identified by imposing sign restrictions on impulse response functions based on the theoretical model by Gerali et al. (2010). The results suggest that credit supply shocks have been a key driver of business cycle fluctuations in euro area countries, and that their effects on the economy have generally increased since the recent crisis. Moreover, we report evidence that credit supply shocks contributed positively to output growth during the pre-crisis period and negatively during the downturn in economic activity in 2008-2009 in all the countries considered. In the post-crisis period, by contrast, we observe a strong rise in cross-country heterogeneity, reflecting financial fragmentation in the euro area. Although this heterogeneity across euro area countries seems to have declined since around 2012, the contribution of credit supply shocks to GDP growth and credit volume growth remains negative in most euro area countries, suggesting that constraints in the supply of credit still represent a bottleneck for the recovery.

KEYWORDS: credit supply shocks, euro area, TVP-VAR, sign restrictions JEL Classification: C11, C32, E32, E51



European Central Bank European Central Bank and University of Bologna. Mail: [email protected] Address: Kaiserstraße 29 - 60311 Frankfurt am Main (Germany). Office Phone: +49 6913446091 ‡ The views expressed in this paper are those of the authors and do not necessarily reflect the views of the European Central Bank or the Eurosystem. We are grateful to Carlo Altavilla, Agostino Consolo, Roberto De Santis, Hashmat Khan, Laurent Maurin, Alberto Musso, Heinrich Kick and Giulia Rivolta for valuable suggestions. We also thank participants at seminars at the European Central Bank. †

1

Introduction

Credit markets play a key role in the business cycle of advanced economies. The weakness of bank lending in many economies in the wake of the global financial crisis has led to an intensive debate about its economic implications. A key question in this regard is to what extent the weakness in bank lending is due to tight credit supply conditions or weak demand for credit. Understanding the relative role of credit supply and demand is important as they have different implications for macroeconomic conditions. To the extent that weak bank lending reflects tight supply of credit rather than weak demand, then weak lending is more likely to dampen economic activity. In countries facing weakness in lending, the correct identification of credit supply dynamics is thus crucial for policy makers. Moreover, it is also important to understand how credit markets contribute to the propagation of macroeconomic disturbances arising in other sectors of the economy, and how they can be a source of disturbance by themselves. Inspired by recent events, a fast growing literature has attempted to identify credit supply shocks through vector autoregressions (VAR) by imposing sign restrictions on impulse responses (Halvorsen and Jacobsen, 2009; Busch et al., 2010; De Nicol`o and Lucchetta, 2011; Eickmeier and Ng, 2011; Tam´ asi and Vil´ agi, 2011; Gambetti and Musso, 2012; Hristov et al., 2012; Barnett and Thomas, 2013; Darracq Paries and De Santis, 2013; Houssa et al., 2013; Darracq Paries et al., 2014; Kick, 2014), or by using other identification schemes (Ciccarelli et al., 2010; Abildgren, 2012; Darracq Paries and De Santis, 2013). However, the only paper adopting a time-varying parameter VAR (TVP-VAR) approach with stochastic volatility to study credit supply shocks is due to Gambetti and Musso (2012), who present a systematic comparison across the euro area, the UK and the US. Analyses at the euro area country-level are scarce (see, for example, Hristov et al. (2012)) and rely on estimations of models with fixed parameters and constant volatility. Time-variation in the coefficients and stochastic volatility are necessary ingredients to control for the non-linearities associated with the structural economic changes and heteroscedastic macroeconomic shocks usually occurring over long time spans. This paper attempts to shed light on the role of credit supply shocks in euro area countries during the past decade, focusing on developments in output and credit volumes. To this purpose, a TVP-VAR with stochastic volatility a` la Primiceri (2005) is estimated for each country in our sample (Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, the Netherlands, Portugal and Spain). Quarterly data covering the period 1980Q1-2013Q2 are used, with the exception of Greece (1985Q1-2012Q4), Ireland (1990Q1-2013Q2) and Italy (1995Q1-2013Q2). To tackle the high dimensionality of the parameter space, the estimation is carried out using Bayesian methods. The structural shocks are then identified by imposing sign restrictions on impulse response functions based on the DSGE model proposed by Gerali et al. (2010). The contribution of this paper to the existing literature is twofold. First of all, to the best of our knowledge this is the first paper that applies time-varying parameter VAR methods at the euro area country-level to identify credit supply shocks, allowing us to compare the recent experiences of the countries considered. Second, the paper gives a special emphasis on the role of credit supply shocks over the boom-bust-recovery phases characterizing the past decade. The main results suggest that credit supply shocks have played a key role in business cycle fluctuations in most euro area countries, and that their effects on the economy have generally

2

increased since the recent crisis. A counterfactual exercise conducted through historical decomposition analysis indicates that in all countries credit supply shocks contributed positively to output growth in the pre-crisis phase and negatively during the downturn in economic activity in 2008-2009. In the post-crisis period, by contrast, we find a strong rise in cross-country heterogeneity. In the aftermath of the crisis, credit supply shocks contributed to the divergence in real GDP growth across countries, reflecting financial fragmentation in the euro area. More specifically, in Greece, Ireland, Italy and Spain credit supply shocks exacerbated the downturn, whereas in Austria, Belgium, Germany and the Netherlands credit supply shocks contributed positively to output growth. Although this heterogeneity across euro area countries seems to have declined during the most recent period, the contribution of credit supply shocks to GDP growth remains negative in most countries, suggesting that constraints in the supply of credit still represent a bottleneck for the recovery. In addition, we report evidence that credit supply shocks have also been a major driver of fluctuations in loan growth during the past decade. In line with our findings for GDP growth, a high degree of cross-country heterogeneity in the contribution of credit supply shocks to credit volume movements is observable during the immediate post-crisis period. Our estimates also show that, while bottlenecks in the supply of credit have progressively become a less important factor constraining output in most stressed euro area countries, in particular since mid-2012, this pattern cannot be observed for loan volume growth, which remains persistently subdued partly because of negative credit supply developments. Lastly, we show that the main findings of the paper are robust to an alternative identification scheme used in the literature. The remainder of the paper is organized as follows. Section 2 describes the econometric model. Section 3 presents the main findings. In Section 4 we perform a robustness check. Sections 5 concludes the paper.

2

The Empirical Methodology

This section describes our econometric approach. Our model follows closely that used by Gambetti and Musso (2012), to which the reader may refer to for the technical details on the estimation.

2.1

The Econometric Model

The analysis is performed by estimating the time-varying parameters VAR model with stochastic volatility employed by Gambetti and Musso (2012). Pioneered by Cogley and Sargent (2005) and Primiceri (2005), TVP-VARs with stochastic volatility have recently become of increasing interest for economists (Baumeister et al., 2008; Benati, 2008; Benati and Surico, 2008; Canova and Gambetti, 2009; Clark and Terry, 2010; Fern´andez-Villaverde and Rubio-Ram´ırez, 2010; Franta, 2011; Mumtaz et al., 2011; Nakajima, 2011; Nakajima et al., 2011; Baumeister and Benati, 2013; Prieto et al., 2013). The ability to capture the potential time-varying nature of the underlying structure of the economy and the volatility of macroeconomic shocks in a flexible and robust way is what makes this method particularly appealing to macroeconomists.1 In what 1

For a comprehensive overview of the TVP-VAR technique, with both methodological and empirical applications, see Nakajima (2011).

3

follows, a formal description of the model is provided. Consider the following reduced-form VAR model: Yt = B0,t + B1,t Yt−1 + · · · + Bp,t Yt−p + t ≡ Xt0 θt + t

(1)

where Yt is an n × 1 vector containing our endogenous variables, Yt = [yt , πt , lt , clrt , strt ]0 , i.e. the (year-on-year) real GDP growth rate, the annual rate of inflation, the (year-on-year) growth rate of the stock of credit, a composite lending rate and the short-term market interest rate, respectively; B0,t is an n × 1 vector of time-varying coefficients that multiply constant terms; Bi,t (i = 1, . . . , p) are n × n matrices of time-varying coefficients; t are the VAR’s reduced-form innovations with zero mean and time-varying covariance matrix Σt , which is factorized in a standard way: −1 Σt = A−1 t Ht At

0

(2)

where At is the lower triangular matrix of simultaneous relations     At =    

1 α21,t α31,t α41,t α51,t

0 1 α32,t α42,t α52,t

0 0 1 α43,t α53,t

0 0 0 1 α54,t

0 0 0 0 1



0 0 0 σ4,t 0

0 0 0 0 σ5,t



      

(3)

and Ht is the diagonal matrix     Ht =    

σ1,t 0 0 0 0

0 σ2,t 0 0 0

0 0 σ3,t 0 0

      

(4)

Let us collect in a vector αt the non-zero and non-one elements of matrix At , and in another vector σt the diagonal elements of matrix Ht . Following Primiceri (2005), the model’s timevarying parameters are assumed to evolve as random walks: αt = αt−1 + ξt

(5)

θt = θt−1 + ηt

(6)

and geometric random walks: log σt = log σt−1 + τt

(7)

where ξt , ηt , and τt are white Gaussian noises with zero mean and covariance matrix S, Q, and

4

W , respectively. As in Primiceri (2005), we assume the vector of the innovations to be jointly normally distributed with the following assumptions on the covariance matrix:     Var   

ut ξt τt ηt





     =     

I5 0 0 0

0 S 0 0

0 0 Q 0

0 0 0 W

     

(8)

1

2 where ut is such that t = A−1 t Ht ut . Lastly, to simplify the estimation we assume S to be block diagonal, i.e. the coefficients of the contemporaneous relations among variables are assumed to evolve independently in each equation (Primiceri, 2005). To summarize the most meaningful findings of the model the following functions of the VAR coefficients are reported: the impulse response functions (IRFs), the forecast error variance decomposition and the historical decomposition. The impulse response functions trace out the MA representation of the system, and are derived as follows (Canova and Gambetti, 2009; Gambetti and Musso, 2012):

Yt = µt +

∞ X

Ck,t t−k

(9)

k=1

where C0,t = I, µt =

P∞

k=0 Ck,t B0,t ,

Ck,t = Sn,n (Bkt ), Bt =

Bt

!

, and In(p−1) 0n(p−1),n Sn,n (X) is a function that selects the first n rows and n columns of matrix X. The variance decomposition describes the contribution of each shock to the variance of the forecast error in Yt+τ (τ = 1, 2, . . . ). The historical decomposition measures the contribution of each shock to the deviations of Yt+τ from its baseline forecasted path (τ = 1, 2, . . . ). Estimation is conducted using Bayesian methods, which are particularly efficient in treating the high dimensionality of the parameter space and the non-linearities of the model. More specifically, we use a variant of Markov Chain Monte Carlo (MCMC) methods, the Gibbs sampling, to draw from conditional distributions to approximate joint and marginal distributions. To evaluate posteriors we specify the prior distributions consistently with Primiceri (2005). The priors for the initial states of the time-varying coefficients B, simultaneous relations A and log volatilities H are assumed to be normally distributed. The priors for the hyperparameters S, Q and W are assumed to be distributed as independent inverse Wishart. Ordinary least square estimates of a time invariant model, based on a small initial subsample, are used for the specification of the prior distributions.2 Gibbs sampling is performed in different steps, drawing in turn volatilities (H), simultaneous relations (A), time-varying coefficients (B), and hyperparameters (S, Q, W ). Technical details on the Gibbs sampling algorithm used can be found in Gambetti and Musso (2012). A sample of 15000 iterations of the Gibbs sampler is employed, discarding the first 10000 and collecting one out of five draws. As reported in Appendix A, the convergence diagnostics are satisfactory, as they reveal the convergence of the algorithm. The model is estimated for each country in our sample using quarterly data. The countries considered in the analysis are the initial euro area countries (excluding Luxembourg), i.e. Aus2

For more details on the calibration of the prior distributions refer to Gambetti and Musso (2012).

5

tria, Belgium, Finland, France, Germany, Ireland, Italy, the Netherlands, Portugal and Spain, as well as Greece. The sample period differs among countries, and its length depends on data availability. We use quarterly data covering the period 1980:1-2013:2 for all countries, with the exception of Greece (1985:1-2012:4), Ireland (1990:1-2013:2) and Italy (1995:1-2013:2). Details on the variables used are reported in Appendix B. For computational reasons, and consistently with other papers in the literature (Primiceri, 2005; Franta, 2011; Gambetti and Musso, 2012; Baumeister and Benati, 2013), we limit the number of lags to 2 (p = 2).

2.2

The Identification Strategy

The scheme to identify structural shocks is based on sign restrictions, which allows us to avoid the usual recursive assumptions on the contemporaneous effects between endogenous variables. Since it is problematic to impose sign restrictions directly on the coefficients of the VAR, identification of the structural shocks is achieved ex-post by imposing sign restrictions on impulse response functions. More specifically, we check whether the impulse responses satisfy our sign restrictions. Sign restrictions on impulse responses have been frequently used in the literature to identify VAR structural shocks (Faust, 1998; Uhlig, 2005) and, in particular, credit supply shocks (Busch et al., 2010; De Nicol` o and Lucchetta, 2011; Eickmeier and Ng, 2011; Gambetti and Musso, 2012; Hristov et al., 2012; Barnett and Thomas, 2013; Houssa et al., 2013). As shown by Paustian (2007), sign restrictions can be a useful tool for recovering structural shocks from VAR residuals as long as the imposed restrictions are sufficiently numerous. Moreover, the identification of additional shocks contributes to a better identification of the shocks under scrutiny, as the orthogonality between the structural shocks represents itself an additional restriction (Uhlig, 2005). Lastly, the imposed sign restrictions must be mutually exclusive to uniquely identify the structural shocks. In light of these considerations, we set our sign restrictions as follows. We identify four structural shocks, leaving one shock unidentified in order to capture the effects of any further remaining disturbance. The strategy to identify aggregate supply shocks, aggregate demand shocks and monetary policy shocks is the same as in Gambetti and Musso (2012), who set their restrictions on the basis of standard DSGE models. Table 1 summarizes the set of sign restrictions we impose on the impulse responses. In particular, aggregate supply shocks drive real GDP and inflation in opposite directions. Expansionary aggregate demand shocks are assumed to move real GDP, inflation, the short-term interest rate and the lending rate up. Finally, we define an expansionary monetary policy shock as a shock that has a positive impact on real GDP and inflation and a negative impact on the short-term interest rate. [Table 1 about here] Our strategy for identifying credit supply shocks relies on the impulse response functions generated by the theoretical model proposed by Gerali et al. (2010), reported in Figure 1, and employed by Busch et al. (2010), Gambetti and Musso (2012). As an aggregate demand shock, an expansionary credit supply shocks leads to an increase in output, inflation and the short-term interest rate. What distinguishes these two shocks is that credit volume changes are assumed to be driven by credit supply shocks if the lending rate moves in the opposite direction in comparison with the rate of credit growth. We will stick to this intuitive identification structure 6

for credit supply shocks throughout the whole paper. Some restrictions on the other shocks will be changed when performing sensitivity analysis in Section 4. [Figure 1 about here] From a practical point of view, to obtain impulse response functions that satisfy our sign restrictions, we assume that Pt is the unique lower triangular Cholesky matrix such that Pt Pt0 = Σ. For any Ht such that Ht Ht0 = I, we have that Σ = Pt Pt0 = Pt Ht Ht0 Pt0 . Therefore, we can construct a new decomposition and orthogonalize the shocks by using Pt Ht and check whether the generated impulse response functions, given by IRF (t, k) = Ck,t Pt Ht , satisfy simultaneously the restrictions imposed. Ht is chosen by means of rotation matrices, as described by Canova and De Nicol´ o (2002) and Canova (2007). If a particular rotation matrix generates impulse responses compatible with our sign restrictions, the impulse responses are stored. Otherwise, we discard them. Restrictions on impulse responses are imposed only upon impact, as in Gambetti and Musso (2012). We impose them only upon impact as we express our variables in growth rates instead of in levels (with the exception of interest rates), thereby generating impulse responses characterized by very low persistence. This allows us to adopt a more parsimonious approach for the choice of the period for identification restrictions.

3

Results

Developments in credit and economic activity in the euro area over the past decade have been dominated by the impact of the global financial crisis. After growing strongly in the years preceding the financial crisis, credit growth declined abruptly and real GDP contracted significantly in 2008-2009. In the aftermath of the crisis, economic activity started to pick up again gradually although credit growth remained subdued. As credit and output dynamics were different before, during and in the aftermath of the crisis, we distinguish three sub-periods in our analysis: the pre-crisis, the bust and the post-crisis period (Table 2). The terms bust and crisis are used interchangeably in this paper and refer to the downturn in real GDP growth around 2008-2009 (see Table 2 for the precise dates). Although these three periods broadly coincide for all countries, their precise timing is country-specific. The bust period starts with the first quarter-on-quarter decline in real GDP in 2007-2008 and the start of the post-crisis period is defined by the first subsequent increase in quarterly GDP. [Table 2 about here]

3.1

Evidence of Time-Varying Coefficients and Variance

Before we present our results on the effect of credit supply shocks, we first provide evidence of the appropriateness of using time-varying coefficients and stochastic volatility in order to validate the choice for our econometric approach. Figure 2 plots the time-varying variance of the residuals for the five variables in our model and for each country in the sample. Substantial time variation is evident for all the countries, especially during the 1980s and the recent global financial crisis, reflecting structural economic changes and the fact that these economies were hit by extraordinary shocks. 7

[Figure 2 about here] In order to investigate the presence of time variation in coefficients and volatility formally, we perform three statistical tests. Table 3 reports three tests to check the presence of time-varying coefficients in the matrices A, B and H (see the notes under the table for a description of the tests). The trace test, the Kolmogorov-Smirnov test and the t-test on matrices A and B all reveal strong time-variation in the coefficients, supporting our approach of using time-varying coefficients. In addition, the Kolmogorov-Smirnov test and the t-test on matrix H further confirm our preference for using a specification that allows for stochastic volatility.3 [Table 3 about here]

3.2

Credit Supply Shocks over the Entire Sample Period

Before moving to the main results for the three sub-periods considered, we provide a brief overview of the average effect of credit supply shocks in our countries of interest. The impulse response functions to an expansionary credit supply shock over the entire sample period look intuitive. Figure 3 reports the median of the responses (blue line) and the associated confidence interval represented by the shaded area (showing the 16th and 84th percentiles of the distribution). The confidence intervals are generally narrow, confirming the significance of the responses. The impulse responses show a similar pattern across the euro area countries in our sample. An expansionary credit supply shock has a significant but short-lived positive impact on real GDP growth, lasting for around three quarters in most countries. The positive effect on inflation is more persistent than on output, lasting up to five years in some countries (although it is substantially shorter in most countries). There is more variation across countries for inflation than for GDP growth, with a relatively persistent impact on inflation observed for Belgium, Finland, Greece and Spain. Regarding loan volume growth, credit supply shocks seem to have on average a longer-lasting impact in France, Ireland, Italy, Portugal and Spain, amounting to up to two to three years in these countries while being substantially shorter in the rest of the euro area. The decline in the lending rate seems to be short-lived in all euro area countries, with a significant negative effect lasting for only two to three quarters. The positive responses in the short-term interest rate reflect the increase in inflation resulting from an expansionary credit supply shock and are similar across countries (with Greece standing out to some extent). [Figure 3 about here] Appendix C reports the impulse response functions of aggregate supply shocks, aggregate demand shocks and monetary policy shocks (Figures 13-15). The impact of the other structural shocks exhibits a plausible pattern, suggesting that we correctly identify them as well as credit supply shocks. Our impulse responses are broadly in line with the findings in related studies, in particular with those in Gambetti and Musso (2012). Comparing the results for individual countries with those for the euro area in Gambetti and Musso (2012), we can observe that, 3

A comparison between our TVP-VAR with stochastic volatility and two alternative VARs with time-invariant coefficients (a country-specific VAR and a panel VAR a ` la Hristov et al. (2012)) shows that our model generates indeed much more plausible results. These findings are available upon request from the authors.

8

perhaps unsurprisingly, the impulse responses for Germany are relatively similar to those for the euro area average. Particularly for Greece and Portugal, by contrast, the impulse responses look relatively different from the euro area average, although the differences are not large.

3.3 3.3.1

Credit Supply Shocks before, during and in the aftermath of the Crisis Impulse Response Analysis

One of the main advantages of modeling time-variation is the possibility to track the effects of structural shocks over time. The evolution of the short-term impact of credit supply shocks during the last decade shows a substantial time variation in all the countries of interest. Credit supply shocks had a stronger impact on GDP growth during the crisis than before and afterwards, in particular in Austria, Belgium, Germany, Ireland, Italy and Spain (Figure 4 shows the impulse responses effect after one quarter). As credit supply shocks were relatively large during the crisis, this is an indication that the short-term impact of these shocks may become larger with the size of the credit supply shock. A similar pattern can be observed for inflation. The impact of credit supply shocks peaked during the crisis, particularly in Ireland and Spain. The evolution of the effect of credit supply shocks on loan volumes exhibits a less clear pattern during the three sub-periods. In some countries (such as Belgium, Finland, France and the Netherlands), the impact of credit supply shocks on credit seems to have gradually declined over time. In others, by contrast, the impact of these shocks has increased (such as in Germany, Ireland, Italy and Spain). Regarding interest rates, there is an increase over time in the absolute value of the short-term impact of credit supply shocks on both lending rates as well as the money market rate in most countries. [Figure 4 about here] Statistical tests provide further evidence of the changing impact of credit supply shocks over the three periods considered (Appendix D, Table 7). Both a t-test and a Wilcoxon rank sum test, testing for equal means and medians, respectively, show that for most countries and variables in our model the impact of credit supply shocks has significantly increased over time (the green cells in Table 7). Only in a small number of cases, the impact has not changed significantly (white cells) or has declined (red cells). These results are broadly in line with those in Gambetti and Musso (2012) for the euro area, who also observe a larger short-term impact of credit supply shocks around the global financial crisis, particularly for real GDP and inflation. 3.3.2

Forecast Error Variance Decomposition

The forecast error variance decomposition measures the contribution of a specific shock to the variability of the forecast error for the variables in our model. The evolution of the variance decomposition (at a 20-quarter horizon) suggests that the contribution of credit supply shocks to the variability of our variables declined during the downturn of 2008-2009, before picking up in the aftermath of the crisis (Figure 5). The decrease in the relative importance of credit supply shocks during the recession reflects the greater role of demand shocks during that period. A more heterogeneous picture emerges in the post-crisis period, with the share of the variance explained by credit supply shocks increasing in some countries while decreasing in others. More 9

specifically, after the crisis the fraction of the variance of our variables that can be explained by credit supply shocks increased especially for Greece, Ireland and Italy, whereas a decline occurs particularly in Finland and France. The t-test and Wilcoxon rank sum test in Table 8 in Appendix D show that most of these changes are indeed statistically significant (a significant larger impact of credit supply shocks is marked in green, whereas the red cells represent a significant lower impact). Looking at the variance decomposition for the five variables in our model in more detail, it is worth highlighting the increase in the variance of credit volume growth due to credit supply shocks from the pre-crisis to the post-crisis period in several countries (green cells in Table 8, Appendix D). The same applies to the composite lending rate, suggesting that credit supply shocks had an impact on lending conditions and the evolution of credit volumes in particular after 2009, rather than during the downturn in 2008-2009. [Figure 5 about here] 3.3.3

Historical Decomposition

The historical decomposition decomposes actual data into a trend and the accumulated effects of the structural shocks. Therefore, it allows economists to perform counterfactual exercises by providing a picture of the estimated impact of structural shocks during the period of interest. Table 4 shows the average rate of real GDP growth in euro area countries during the three periods considered, with and without the effect of credit supply shocks. The counterfactual series indicates how a variable would have evolved in the absence of credit supply shocks. Focusing on real GDP, it appears that credit supply shocks made a positive contribution to output growth in the pre-crisis phase and a negative contribution during the crisis in all our countries, whereas in the aftermath of the crisis there was a strong rise in cross-country heterogeneity. Before the crisis, credit supply shocks pushed up real GDP growth in all countries in our sample (green cells represent a positive contribution), particularly in Austria, Greece and Portugal (dark green cells). Once the crisis started, this impact turned negative in all countries (red cells). The negative contribution of credit supply shocks was particularly large (in relative terms) in Austria, Ireland, Italy, Portugal and Spain (dark red cells). Hence, there is some evidence that the larger the positive contribution of credit supply shocks was before the crisis, the more sizeable the subsequent negative impact of these shocks was during the bust. In the aftermath of the crisis, credit supply shocks contributed to the divergence in real GDP growth across the countries in our sample, reflecting financial fragmentation in the euro area. Especially in Greece, Ireland, Italy and Spain, credit supply shocks added to the downturn, whereas in Austria, Belgium, Germany and the Netherlands credit supply contributed positively to output growth. These findings contrast with those documented in Hristov et al. (2012). Also in that study there is a rise in cross-country heterogeneity after the crisis, but it involves different groups of countries. The first one, which comprises Austria, Finland, Ireland, Italy and Portugal, is characterized by negative contributions of credit supply shocks during the first half of the crisis (2007Q3-2008Q4) and positive contributions in the second period (2009Q1-2010Q2). By contrast, in the second group of countries, composed of Belgium, France, Germany, Greece, the Netherlands and Spain, the contribution of credit supply shocks to GDP growth was positive

10

in the first period and negative in the second one. A possible explanation for these conflicting findings may be related to the different empirical approach that these authors follow (i.e. a panel VAR with constant coefficients). [Table 4 about here] Looking at the contribution of credit supply shocks to real GDP growth during the aftermath of the crisis in more detail, differences across euro area countries seem to have declined recently. Figure 6 shows the evolution of actual GDP growth and its counterfactual (i.e. without the effect of credit supply shocks) on a quarterly basis. The most pronounced differences across countries emerged in the immediate aftermath of the crisis (during 2010-2012), with credit supply shocks reducing on average real GDP growth in most euro area countries with Austria, Belgium, Germany and the Netherlands as the main exceptions. The downward impact of these shocks on output growth in several countries, including France, Greece, Italy, Portugal and Spain has waned in the most recent period (in some countries since mid-2012). In almost all countries, however, the contribution of credit supply shocks to GDP growth remains negative, suggesting that constraints in the supply of credit still represent a bottleneck for the recovery. [Figure 6 about here] In order to obtain a clearer picture of the relative importance of credit supply shocks, Figure 7 plots the contributions to real GDP growth of all the structural shocks identified in the model. Not surprisingly, aggregate demand shocks have been a major driver of fluctuations in output growth during the crisis. Moreover, monetary policy shocks have played an important role in the evolution of GDP growth in Belgium, Finland, France, Ireland, Italy and Portugal, whereas aggregate supply shocks have only played a minor role during the past decade. Compared with the other structural innovations, credit supply shocks have also played a leading role, reducing real GDP during the crisis particularly in Austria, Ireland, Italy, Portugal and Spain. But also in the aftermath of the crisis (i.e. since 2010), credit supply shocks have continued to be a key factor dragging down output growth in Ireland, Italy and Spain. [Figure 7 about here] Credit supply shocks have also played a key role in credit volume movements during the past decade. Figure 8 shows the evolution of actual loan growth and its counterfactual, in which credit supply shocks are set to zero. In line with our earlier findings for real GDP growth, the expansionary credit supply shocks before the crisis added to new lending in all countries until 2008, although the degree to which these shocks contributed positively to credit growth differed across countries. Our estimates also show that in the aftermath of the crisis bottlenecks in the supply of credit weakened new lending especially in Ireland, Italy, Portugal and Spain. As in the case of GDP growth, we observe a relative high degree of cross-country heterogeneity during the period immediately after the downturn in economic activity (i.e. during 2009 and 2010). This heterogeneity seems to have declined during the more recent period with a gradual increase in the dampening effect of credit supply shocks on new lending in several euro area countries. Compared with the picture for real GDP in Figure 6, however, the improvement for loan volumes seems to be negligible or even absent in the most recent period, suggesting that credit supply bottlenecks are more severe for new lending than for output growth. 11

[Figure 8 about here] The historical decomposition of the structural shocks identified confirms that credit supply was a major constraining factor for credit growth in the last decade in most of our countries (Figure 9). Our model suggests that constraints in the supply of credit have been a key factor holding back credit growth in France, Ireland, Italy, Portugal and Spain during the past few years. In most countries (e.g. in Austria, Finland, Belgium, France, Germany, Greece, Ireland, Portugal and Spain) also demand-related factors seem to have played an important role in the evolution of credit growth, especially since the crisis. Although this suggests that subdued credit demand may also explain part of the weakness in credit growth, our framework does not enable us to disentangle pure credit demand shocks from other demand-related forces. [Figure 9 about here]

3.4

Are the Identified Credit Supply Shocks Plausible? A Comparison with Survey Data

A successful identification of the structural shocks is one of the key challenges when using the VAR approach. Recent papers quantifying the impact of credit supply shocks have used two types of identification strategies (see Section 1): those based on sign restrictions and those based on survey data such as Ciccarelli et al. (2010). The latter study uses survey data from the ECBs Bank Lending Survey (BLS) as a proxy for credit supply and demand. In this approach, credit supply is based on survey replies to the BLS questions on changes in lending standards applied by banks. In order to investigate the plausibility of the structural shocks identified by our model, we follow Ciccarelli et al. (2010) and construct a credit supply index based on changes in bank lending standards associated with the ability of banks to lend in relation to their balance sheet constraints and competitive pressure. Survey data are available for all countries in our sample except Finland and Greece. Appendix B reports in detail how we calculate the credit supply index. The evolution of our credit supply index based on survey data is very similar to the structural credit supply shocks identified by our model (Figure 10). The correlation between both measures is positive and high for all countries, in particular for Austria, Italy, Portugal and Spain. With the exception of Ireland, both measures indicate a sharp tightening of credit supply standards during the crisis in 2008 and 2009. Both measures also show improved credit supply conditions in 2012 and 2013 in many countries. This exercise thus confirms the plausibility of the structural credit supply shocks identified in our model. However, it should be stressed that the advantage of a model-based approach is twofold. First of all, it is rooted in economic theory, and it allows us to calculate credit supply shocks also for periods for which the BLS is not available. Second, survey indicators may be partly endogenous, reflecting movements due to changing economic conditions. [Figure 10 about here]

12

4

A Robustness Check: Alternative Sign Restriction Identification

We check the robustness of our results by modifying the sign restriction identification, and, in particular, adopting the identification used by Hristov et al. (2012). According to this alternative identification strategy, summarized in Table 5, the short-term interest rate declines in response to an expansionary aggregate supply shock, and the lending rate decreases in response to an expansionary monetary policy shock. We do not change the imposed restrictions on credit supply shocks, which remain consistent with the DSGE model by Gerali et al. (2010). [Table 5 about here] We evaluate the difference between this specification and our baseline model in terms of impulse response functions and historical contribution of credit supply shocks to GDP growth. The impulse responses of a credit supply shock are almost identical to those obtained using the baseline model (Figure 11). Appendix E compares the impulse responses for the other shocks that we identify. Whereas the results for the aggregate demand shock are very close to those of our baseline model, the main differences arise for the aggregate supply shock and monetary policy shock. This is not surprising, as we have changed our identification strategy for those two shocks. These differences do, however, not have any impact on the conclusions of this paper. [Figure 11 about here] The counterfactual exercise performed through historical decomposition analysis confirms that our results do not change when we use the alternative identification strategy. Both counterfactual scenarios for real GDP growth are virtually the same in both models (Figure 12). To sum up, our results seem to be very robust to the main alternative identification strategy used in the literature. [Figure 12 about here]

5

Concluding Remarks

This paper aims to shed light on the role of credit supply shocks in euro area countries during the recent pre-crisis, bust, and post-crisis phases. We estimate a time-varying parameter vector autoregression (TVP-VAR) with stochastic volatility following Primiceri (2005) and Gambetti and Musso (2012) for each country, and the structural shocks are identified by imposing sign restrictions on impulse response functions based on the theoretical model developed by Gerali et al. (2010). The findings suggest that credit supply shocks have played a key role in business cycle fluctuations in the euro area, and that their effects on the economy have generally increased since the recent crisis. The counterfactual exercises carried out in the paper indicate that in all the countries considered credit supply shocks contributed positively to output growth in the pre-crisis phase and negatively during the downturn in economic activity in 2008-2009. In the post-crisis period, by contrast, we observe a strong rise in cross-country heterogeneity between stressed economies on the one hand and non-stressed economies on the other. More specifically, in the aftermath of the crisis, credit supply shocks contributed to the divergence in real GDP 13

growth across countries, reflecting financial fragmentation in the euro area. In Greece, Ireland, Italy and Spain credit supply shocks exacerbated the downturn, whereas in Austria, Belgium, Germany and the Netherlands credit supply shocks contributed positively to output growth. Although this heterogeneity across euro area countries seems to have declined during the most recent period, the contribution of credit supply shocks to GDP growth remains negative in most euro area countries, suggesting that constraints in the supply of credit still represent a bottleneck for the recovery. In addition, we report evidence that credit supply shocks have also played a major role for fluctuations in loan growth during the last decade in most of our countries. In line with our findings for GDP growth, a high degree of cross-country heterogeneity can be observed for credit volumes during the immediate post-crisis period. Our estimates also show that, whereas bottlenecks in the supply of credit have progressively become a less important factor constraining output in most stressed euro area countries, in particular since mid-2012, tight credit supply conditions continue to restrain loan growth in most euro area countries and in stressed economies in particular. This paper suggests several potential avenues for future research. Two are worth mentioning here. The first one would be to further analyze the determinants of credit supply shocks in order to come to more precise policy recommendations. The second avenue would be to distinguish the effects of credit supply shocks originating in different sectors, such as households and nonfinancial corporations. A disaggregation of credit into loans for mortgages and consumption and loans for investment would enable researchers to gain deeper insights into the mechanisms driving economic activity and credit volumes.

14

References Abildgren, K. (2012). Financial structures and the real effects of credit-supply shocks in Denmark 1922-2011. European Review of Economic History, 16(4):490–510. Barnett, A. and Thomas, R. (2013). Has weak lending and activity in the United Kingdom been driven by credit supply shocks? Bank of England Working Papers 482, Bank of England. Baumeister, C. and Benati, L. (2013). Unconventional monetary policy and the Great Recession: Estimating the macroeconomic effects of a spread compression at the zero lower bound. International Journal of Central Banking, 9(2):165–212. Baumeister, C., Durinck, E., and Peersman, G. (2008). Liquidity, inflation and asset prices in a time-varying framework for the euro area. Working Papers Research 142, National Bank of Belgium. Benati, L. (2008). The “Great Moderation” in the United Kingdom. Journal of Money, Credit and Banking, 40(1):121–147. Benati, L. and Surico, P. (2008). Evolving U.S. monetary policy and the decline of inflation predictability. Journal of the European Economic Association, 6(2-3):634–646. Busch, U., Scharnagl, M., and Scheithauer, J. (2010). Loan supply in Germany during the financial crisis. Discussion Papers 05/2010, Deutsche Bundesbank, Research Centre. Canova, F. (2007). Methods for applied macroeconomic research. Princeton University Press. Canova, F. and De Nicol´ o, G. (2002). Monetary disturbances matter for business fluctuations in the G-7. Journal of Monetary Economics, 49(6):1131–1159. Canova, F. and Gambetti, L. (2009). Structural changes in the US economy: Is there a role for monetary policy? Journal of Economic Dynamics and Control, 33(2):477–490. Chib, S. (2001). Markov chain Monte Carlo methods: Computation and inference. In Heckman, J. J. and Leamer, E. E., editors, Handbook of Econometrics, volume 5, chapter 57, pages 3569–3649. Elsevier. Ciccarelli, M., Peydr´ o, J., and Maddaloni, A. (2010). Trusting the bankers: A new look at the credit channel of monetary policy. Working Paper Series 1228, European Central Bank. Clark, T. E. and Terry, S. J. (2010). Time variation in the inflation passthrough of energy prices. Journal of Money, Credit and Banking, 42(7):1419–1433. Cogley, T. and Sargent, T. J. (2005). Drift and volatilities: Monetary policies and outcomes in the post WWII U.S. Review of Economic Dynamics, 8(2):262–302. Darracq Paries, M. and De Santis, R. (2013). A non-standard monetary policy shock: The ECB’s 3-year LTROs and the shift in credit supply. Working Paper Series 1508, European Central Bank.

15

Darracq Paries, M., Maurin, L., and Moccero, D. (2014). Financial conditions index and credit supply shocks for the euro area. Working Paper Series 1644, European Central Bank. De Nicol`o, G. and Lucchetta, M. (2011). Systemic risks and the macroeconomy. NBER Working Papers 16998, National Bureau of Economic Research. Eickmeier, S. and Ng, T. (2011). How do credit supply shocks propagate internationally? A GVAR approach. CEPR Discussion Papers 8720. Faust, J. (1998). The robustness of identified VAR conclusions about money. Carnegie-Rochester Conference Series on Public Policy, 49:207–244. Fern´andez-Villaverde, J. and Rubio-Ram´ırez, J. (2010). Macroeconomics and volatility: Data, models, and estimation. NBER Working Papers 16618, National Bureau of Economic Research. Franta, M. (2011). Identification of monetary policy shocks in Japan using sign restrictions within the TVP-VAR framework. IMES Discussion Paper Series 11-E-13, Institute for Monetary and Economic Studies, Bank of Japan. Gambetti, L. and Musso, A. (2012). Loan supply shocks and the business cycle. Working Paper Series 1469, European Central Bank. Gerali, A., Neri, S., Sessa, L., and Signoretti, F. M. (2010). Credit and banking in a DSGE model of the euro area. Journal of Money, Credit and Banking, 42(s1):107–141. Geweke, J. (1992). Evaluating the accuracy of sampling-based approaches to the calculation of posterior moments. In Bernardo, J. M., Berger, J., Dawid, A. P., and Smith, A. F. M., editors, Bayesian Statistics, pages 169–193. Oxford University Press. Halvorsen, J. I. and Jacobsen, D. H. (2009). Are bank lending shocks important for economic fluctuations? Working Paper 2009/27, Norges Bank. Houssa, R., Mohimont, J., and Otrok, C. (2013). Credit shocks and macroeconomic fluctuations in emerging markets. CESifo Working Paper Series 4281, CESifo Group Munich. Hristov, N., H¨ ulsewig, O., and 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. Kick, H. (2014). Spillover effects of credit demand and supply shocks in the EU countries: Evidence from a structural GVAR. Unpublished manuscript. Mumtaz, H., Zabczyk, P., and Ellis, C. (2011). What lies beneath? A time-varying FAVAR model for the UK transmission mechanism. Working Paper Series 1320, European Central Bank. Nakajima, J. (2011). Time-varying parameter VAR model with stochastic volatility: An overview of methodology and empirical applications. Monetary and Economic Studies, 29:107–142.

16

Nakajima, J., Kasuya, M., and Watanabe, T. (2011). Bayesian analysis of time-varying parameter vactor autoregressive model for the Japanese economy and monetary policy. Journal of the Japanese and International Economies, 5(3):225–245. Paustian, M. (2007). Assessing sign restrictions. The B.E. Journal of Macroeconomics, 7(1):1– 33. Prieto, E., Eickmeier, S., and Marcellino, M. (2013). Time variation in macro-financial linkages. Discussion Papers 13/2013, Deutsche Bundesbank, Research Centre. Primiceri, G. E. (2005). Time varying structural vector autoregressions and monetary policy. Review of Economic Studies, 72(3):821–852. Raftery, A. E. and Lewis, S. (1992). How many iterations in the Gibbs sampler? In Bernardo, J. M., Berger, J., Dawid, A. P., and Smith, A. F. M., editors, Bayesian Statistics, pages 763–773. Oxford University Press. Tam´asi, B. and Vil´ agi, B. (2011). Identification of credit supply shocks in a Bayesian SVAR model of the Hungarian economy. MNB Working Paper 2011/7, Magyar Nemzeti Bank. Uhlig, H. (2005). What are the effects of monetary policy on output? Results from an agnostic identification procedure. Journal of Monetary Economics, 52(2):381–419.

17

Tables and Figures Table 1: Sign restrictions Responses to an expansionary shock Shock

Real GDP

GDP deflator

Short-term rate

Lending rate

Credit

Aggregate supply

+



unrestricted

unrestricted

unrestricted

Aggregate demand

+

+

+

+

unrestricted

Monetary policy

+

+



unrestricted

unrestricted

Credit supply

+

+

+



+

Notes: Sign restrictions are imposed for one quarter on the impulse responses to expansionary shocks.

Table 2: Pre-crisis, bust, and post-crisis periods Pre-crisis period

Bust period

Post-crisis period

Austria

2005Q1-2008Q1

2008Q2-2009Q2

2009Q3-2013Q2

Belgium

2005Q1-2008Q2

2008Q3-2009Q1

2009Q2-2013Q2

Finland

2005Q1-2007Q4

2008Q1-2009Q2

2009Q3-2013Q2

France

2005Q1-2008Q1

2008Q2-2009Q2

2009Q3-2013Q2

Germany

2005Q1-2008Q1

2008Q2-2009Q1

2009Q2-2013Q2

Greece

2005Q1-2008Q3

2008Q4-2010Q4

2011Q1-2013Q2

Ireland

2005Q1-2007Q1

2007Q2-2009Q4

2010Q1-2013Q2

Italy

2005Q1-2008Q1

2008Q2-2009Q2

2009Q3-2013Q2

Netherlands

2005Q1-2008Q1

2008Q2-2009Q2

2009Q3-2013Q2

Portugal

2005Q1-2007Q4

2008Q1-2009Q1

2009Q2-2013Q2

Spain

2005Q1-2008Q1

2008Q2-2009Q4

2010Q1-2013Q2

Notes: The bust period starts with the first QoQ GDP reduction in 2007/2008; the post-crisis period with the first subsequent QoQ GDP increase. Exception: Ireland 2007Q3.

18

19

0.13

0.39

0.03

0.10

0.17

0.18

0.09

0.06

0.40

0.18

Belgium

Finland

France

Germany

Greece

Ireland

Italy

Netherlands

Portugal

Spain

1.17

2.22

2.82

1.77

2.95

1.69

1.24

0.42

2.88

1.97

0.88

16% perc.

1.54

3.08

4.66

2.83

5.11

2.75

1.72

0.58

3.98

3.92

1.21

50% perc.

2.09

4.41

7.34

4.63

11.54

4.58

2.56

0.83

5.68

8.27

1.67

84% perc.

10/10

10/10

10/10

10/10

10/10

10/10

10/10

10/10

10/10

10/10

10/10

1-2

A

9/10

9/10

8/10

10/10

10/10

9/10

10/10

10/10

10/10

10/10

10/10

2-3

53/55

54/55

55/55

49/55

54/55

55/55

55/55

54/55

55/55

54/55

55/55

1-2

B

46/55

54/55

53/55

43/55

39/55

53/55

54/55

49/55

41/55

52/55

49/55

2-3

5/5

5/5

5/5

4/5

5/5

5/5

4/5

5/5

5/5

5/5

5/5

1-2

Kolmogorov-Smirnov test2 H

5/5

5/5

5/5

5/5

5/5

5/5

5/5

5/5

5/5

5/5

5/5

2-3

10/10

10/10

9/10

10/10

9/10

9/10

10/10

9/10

10/10

9/10

9/10

1-2

A

8/10

8/10

7/10

7/10

9/10

8/10

10/10

10/10

8/10

9/10

10/10

2-3

49/55

53/55

53/55

49/55

50/55

51/55

52/55

53/55

50/55

54/55

52/55

1-2

B

39/55

47/55

48/55

44/55

37/55

49/55

46/55

45/55

37/55

50/55

46/55

2-3

t-test3

5/5

5/5

5/5

5/5

5/5

5/5

4/5

5/5

3/5

5/5

5/5

1-2

H

5/5

4/5

5/5

5/5

4/5

5/5

4/5

5/5

5/5

4/5

5/5

2-3

Notes: 1 Trace test (Cogley and Sargent, 2005). The second column reports the trace of the prior variance-covariance matrix (Q). The third, fourth and fifth columns reports the 16%, 50% and 84% percentiles of the posterior of Q. If the trace of the prior variance-covariance matrix is significantly smaller than the posterior of Q, there is evidence of time-variation in the parameters. 2 Kolmogorov-Smirnov test for equality of two distributions. Columns 6-11 report the proportion of parameters for which the null hypothesis that they are from the same continuous distributions is rejected at the 5% significance level. For each country, the sample period is divided into three sub-periods: period 1, period 2 and period 3. The test is performed between period 1 and 2, and between period 2 and 3. The matrices that contain time-varying parameters, i.e. A, B and H, are considered. 3 t-test for equal means of two distributions. Columns 12-17 report the proportion of parameters for which the null hypothesis that they are from distributions with equal means is rejected at the 5% significance level. For each country, the sample period is divided into three sub-periods: period 1, period 2 and period 3. The test is performed between period 1 and 2, and between period 2 and 3. The matrices that contain time-varying parameters, i.e. A, B and H, are considered.

0.08

Austria

trace

Trace test1

Table 3: Testing time variation in coefficients and volatility

Table 4: Average YoY GDP growth without credit supply shocks (TVP-VAR) Pre-crisis period

Bust period

Post-crisis period

Actual

Counterf.

Actual

Counterf.

Actual

Counterf.

Austria

3.49

3.05

−4.22

−2.75

1.53

1.47

Belgium

2.35

2.32

−5.82

−5.08

1.09

1.07

Finland

4.02

3.90

−7.32

−7.15

1.17

1.22

France

2.05

1.96

−3.57

−3.35

1.05

1.11

Germany

3.90

3.74

−6.22

−5.22

3.73

3.62

Greece

3.11

2.80

−4.64

−4.10

0.81

0.95

Ireland

6.46

6.06

−4.19

−1.88

0.69

2.31

Italy

1.57

1.46

−5.97

−4.33

−0.50

0.07

Netherlands

3.44

3.38

−4.14

−3.71

0.23

0.21

Portugal

1.83

1.56

−3.40

−2.73

−0.83

−0.77

Spain

3.52

3.43

−2.91

−2.24

−0.77

−0.59

Notes: Green (red) cells indicate that the contribution of credit supply shocks to the average YoY GDP growth is positive (negative). , , indicate that the percentage difference between actual and counterfactual GDP growth is, respectively: between 0 and 10 percent; between 10 and 20 percent; larger than 20 percent. , , indicate that the percentage difference between actual and counterfactual GDP growth is, respectively: between 0 and 10 percent; between 10 and 20 percent; larger than 20 percent.

Table 5: Alternative sign restrictions Responses to an expansionary shock Shock

Real GDP

GDP deflator

Short-term rate

Lending rate

Credit

Aggregate supply

+





unrestricted

unrestricted

Aggregate demand

+

+

+

+

unrestricted

Monetary policy

+

+





unrestricted

Credit supply

+

+

+



+

Notes: Sign restrictions are imposed for one quarter on the impulse responses to expansionary shocks.

20

Figure 1: Impulse response functions to a shock to the loan rate of entrepreneurs (Gerali et al., 2010) Loans to entrepreneurs

Rate on loans to entrepreneurs 0

0.09

−3

8

0.08

−0.005

6

0.07

−0.01

4

0.06

−0.015

2

0.05

−0.02

0

0.04

−0.025

−2

0.03 0

10

20

−0.03 0

Output

10

20

−4 0

Consumption

0.04

x 10

10

−5

0.025

4

Policy rate

x 10

20

Inflation

3

0.035

0.02 2

0.03 0.015

1

0.025 0 0.01

0.02 0.015 0

−1 10

20

0.005 0

10

20

−2 0

10

20

Notes: The rates are shown as absolute deviations from the steady-state, expressed in percentage points. The other variables are percentage deviations from the steady-state.

21

Figure 2: Stochastic volatility (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: The red line indicates the median of the VAR residual variances. The shaded area indicates the 16 and 84 percentiles.

Figure 3: Impulse response functions to a credit supply shock (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: The blue line denotes the median of the impulse responses to a credit supply shock. The shaded area indicates the 16 and 84 percentiles. The impulse responses are normalized to an expansionary one-standard deviation shock and are expressed in percent terms.

Figure 4: Impact effect of credit supply shocks over time (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: Evolution of the impulse responses to a credit supply shock at horizon = 1. The blue line indicates the median of the impulse responses. The shaded area indicates the 16 and 84 percentiles. The impulse responses are normalized to an expansionary one-standard deviation shock and are expressed in percent terms.

Figure 5: Variance decomposition of credit supply shocks over time (at horizon = 20) (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: Evolution of the variance decomposition of credit supply shocks at horizon = 20. The green line indicates the median of the variance decomposition. The shaded area indicates the 16 and 84 percentiles.

Actual

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

Actual

(g) Ireland

4

2

-2 0 0

-4 -1

-6

-8

-10

6

4

2

0

-2

-4

-6

6

4

2

0

-2

-4

-6 -2 -6

-3 -8

Counterfactual Actual

(i) Netherlands

Counterfactual

Actual

2013Q1

8

2013Q1

Actual

2012Q3

Counterfactual

2012Q3

2012Q3

2012Q1

-1.2

2011Q3

-0.7

2012Q1

-6

2012Q1

-0.2

2011Q1

0.3

2011Q3

0

2011Q3

0.8

2010Q3

4

2011Q1

(e) Germany

2011Q1

8

2010Q1

Actual

2010Q3

Counterfactual

2010Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

-5

2010Q1

-0.6

2009Q3

-12

2010Q1

-4

2010Q1

-0.5

2009Q3

-10

2009Q1

-3

2009Q3

Actual

2009Q3

-0.4

2009Q1

-0.3

-8

2008Q3

-6

2009Q1

-0.2

2009Q1

(c) Finland

2008Q3

-4

2008Q1

Counterfactual

2008Q3

-0.1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

(a) Austria

2008Q3

0

2008Q1

0

2007Q3

-2

2008Q1

0.1

2008Q1

2

2

2007Q3

3

0.2

2007Q3

0.3

4

2007Q1

6

2007Q1

4

2006Q3

0.4

2006Q3

8

2007Q3

6 -5

2007Q1

-8 -2

2007Q1

-4

-4

2006Q1

-2 -1.5

2006Q3

-1

2006Q3

-0.5

2005Q3

0

2006Q1

0.5

2005Q3

1

2006Q1

2 3

2006Q1

6

2005Q1

-6 1.5

2005Q3

-4

2005Q1

-2

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

0 4

2005Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

2

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

4 2

2005Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

6

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

Actual

2010Q1

Actual

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Actual

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Actual

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Figure 6: Evolution of YoY real GDP growth in the absence of credit supply shocks (b) Belgium 0.6

2

1

0.4

0

0.2

-1

0

-2

-3

-0.2

-0.4

-0.6

-0.8

Counterfactual

(d) France 0.3

1

0.2

0

0.1

-1

0

-2

-0.1

-0.2

-0.3

-0.4

Counterfactual

8

(f) Greece

6

4 0.75

-2 2

0 0.25

-4 -0.25

-6

-8 -0.75

-10 -1.25

Counterfactual

4

(h) Italy

2

1 2

1

0 0.5

-2 0

-4 -0.5

-1

-1.5

-2

-2.5

Counterfactual

0.8 3

(j) Portugal 0.6

0.6 2 0.4

0.4 1 0.2

0.2

0

0

-1

0

-0.2

-0.2

-2

-0.4

-0.4

-3

-0.6

-0.6

-4

-0.8

-0.8

-5

-1

Counterfactual

(k) Spain

1

0.8

0.6

0.4

0.2

0

-0.2

-0.4

-0.6

-0.8

-1

Counterfactual

Notes: Actual and counterfactual evolution of real GDP growth (left axis). The green bars indicate the median of the contribution of credit supply shocks to GDP growth (right axis).

CS

AS

AD

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

AD

0

-4

-10 -8

MP CS

1.5

(i) Netherlands

0.5

1 2

(j) Portugal

1.5

0.5

1

-1

-1

-1.5

-1.5

-2

-2.5

-2

MP

CS

AS

AD

2010Q1

2009Q3

2009Q1

2011Q3

2011Q1

2010Q3

2012Q3

-0.5

MP

2013Q1

0

-0.5

AD

2013Q1

-4

2012Q1

-3

2012Q3

-2

2012Q3

-6

2012Q1

0

2012Q1

-1

2011Q3

(h) Italy

2011Q3

1

2011Q1

0

AS MP

2011Q1

-6

AD

2010Q3

-5

-14

AD

2010Q3

-12

AS

2010Q1

CS

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

AD

2010Q1

-2

AS

2009Q3

(g) Ireland AS

2009Q3

MP

2009Q1

(e) Germany

2008Q3

CS

2009Q1

MP

2008Q1

(c) Finland

2008Q3

2

2007Q3 CS

2008Q3

-4

2007Q1

MP

2008Q1

-3

-5

2007Q3

-4

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

(a) Austria

2008Q1

-3

2007Q3

1

2007Q1

1

2007Q3

2

2007Q1

2

2006Q3

-2.5

2006Q3

-2

-5

2007Q1

-4

2006Q3

1

2006Q1

-4

2006Q1

-3

-4

2006Q1

-2

-3

2006Q3

-2

2006Q1

-1

2005Q3

0

-1

2005Q3

1

2005Q3

0

2006Q1

0

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

1

2005Q3

-3

2005Q1

0

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2009Q3

2009Q1

2

2005Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

2

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

AD

2010Q3

AS AD

2010Q1

AD

2010Q1

AS

2009Q3

AS

2009Q3

AS

2009Q3

2009Q1

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

AD

2010Q1

CS

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

AS

2009Q3

CS

2009Q1

CS

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

CS

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

CS

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Figure 7: The contribution of the identified shocks to GDP growth (b) Belgium

MP

0.5

(d) France

-0.5

0

-1

-2

-1.5

-1

MP

(f) Greece

-1 -1 0

-2

-2

MP

1

(k) Spain

0.5

-0.5

0

-1.5

-1

-2.5

-2

-3.5

-3

MP

Notes: Contribution of the identified shocks to GDP growth (median). CS indicates Credit Supply shocks, AS Aggregate Supply shocks, AD Aggregate Demand shocks, MP Monetary Policy shocks.

Actual

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

-10

-20

-30

-40 -6

-50 -8

Counterfactual Actual

9

(i) Netherlands 1.5

7

5 1

3 0.5

-1

1

0

-3

-0.5

-5

-7

-1

-9

-1.5

Counterfactual

Actual

25

2.5

20

2

15

1.5

10

5

1

-5

0.5

0

-10

0

-1

-15

-1.5

-20

-2

-25

-2.5

2013Q1

0

2012Q3

0

2013Q1

2

2012Q1

10

2012Q3

20

2012Q1

4

2011Q3

(g) Ireland

2011Q3

Actual

2011Q1

Counterfactual

2010Q3

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

-1

2011Q1

-3

2010Q1

-0.8

2010Q1

-2 -0.6

2010Q3

-0.4

2009Q3

-1

2010Q1

-0.2

2009Q1

Actual

2009Q3

0.2

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

Actual

2009Q3

0

2008Q3

1 0.4

2009Q1

0.6

2008Q3

(e) Germany

2009Q1

0.8

2008Q1

(c) Finland

2008Q3

Counterfactual

2008Q1

2

2007Q3

Counterfactual

2008Q1

1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

(a) Austria

2008Q1

3

2007Q1

-1

2007Q3

-20

2007Q1

-0.6

2007Q3

-15

2007Q3

-10

2006Q3

-5 -0.2

2006Q3

0.2

2007Q1

0

2007Q1

5

2006Q3

0.6

2006Q3

10

2006Q1

15

2005Q3

-2

2006Q1

-1

2006Q1

-0.5

2006Q1

-14

2005Q1

-9

2005Q1

-4

2005Q3

1

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

0.5

2005Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

6

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

1.5

2005Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

11

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

Actual

2010Q1

Actual

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Actual

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Actual

2009Q3

Actual

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Figure 8: Evolution of YoY credit volume growth in the absence of credit supply shocks

2

(b) Belgium

8 0.6

1

3

0.4

0

0.2

-2 0

-7 -0.2

-1.5 -12

-0.4

-0.6

Counterfactual

0.8

(d) France

7.5 0.8

0.4

0.6

0

2.5

0.4

-2.5

0.2

0

-0.4

-0.2

-7.5 -0.4

-0.8

-0.6

-12.5

-0.8

-1

Counterfactual

20

(f) Greece 2

15 1.5

10

0 5 1

0.5

0

-5 0

-10 -0.5

-1

-15 -1.5

Counterfactual

(h) Italy

12

7 2

2 1

-2 -3

-8 0

-1

-4 -13

-18 -2

-23

-28 -3

-4

Counterfactual

(j) Portugal

12

7 1.5

2 1

-3

0.5

-8 0

-13

-0.5

-1

-1.5

-18

-2

-23

-2.5

-28

-3

Counterfactual

(k) Spain

-0.5

Counterfactual

Notes: Actual and counterfactual evolution of credit volume growth (left axis). The green bars indicate the median of the contribution of credit supply shocks to credit volume growth (right axis).

CS

AS

AD

2013Q1

2012Q3

2012Q1

AD

2011Q3

AS

2011Q1

MP CS

2

(i) Netherlands

1.5

1

0.5

-0.5

0

-1

-1.5

-2

MP

CS AS

AS AD

AD

1

-1 0

-2

-3

-4

-5

-6

-7

2013Q1

2

2013Q1

(j) Portugal 2012Q3

3

2012Q3

MP

2012Q1

-20

2012Q1

-15

2011Q3

0

2011Q3

(g) Ireland AD

2011Q1

5

AS

2011Q1

CS

-2

0

-4

-6

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2013Q1

2

2012Q3

(f) Greece

2012Q3

MP

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

AD

2010Q3

MP

AD

2010Q3

-10

2009Q3

-8

-2

2010Q1

-1.5

2010Q1

-1

2009Q1

-0.5

AS

2009Q3

0

AS

2009Q3

(e) Germany

2008Q3

CS

2009Q1

CS

2009Q1

MP

2008Q3

(c) Finland

2008Q1

4

2008Q1

MP

2008Q3

1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

(a) Austria

2008Q3

0.5

2007Q3

6

2008Q1

1.5

2008Q1

-6

2007Q3

-5

2007Q1

-4

2007Q3

-3

2007Q3

-2

2007Q1

0

2007Q1

-1

2007Q1

1

2006Q3

2

2006Q3

3

2006Q3

4

2006Q3

-2.5

2005Q3

-2

-5

2006Q1

-4

2006Q1

-3

2006Q1

1

2006Q1

-2

2005Q1

1.5

2005Q3

0

2005Q1

3

2005Q3

-1

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

4

2005Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2009Q3

2

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q1

2

2005Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q3

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

5

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

AD

2010Q3

AD

2010Q3

AS

2010Q1

AD

2010Q1

AS

2009Q3

2009Q1

AS

2009Q3

2009Q1

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

AD

2010Q1

CS

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

AS

2009Q3

CS

2009Q1

CS

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

CS

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

CS

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Figure 9: The contribution of the identified shocks to credit volume growth (b) Belgium

0.5

1

-0.5

0

-1.5

-1

MP

1.5

(d) France

0.5

1

-0.5

0

-1.5

-1

-2.5

-2

MP

4

(h) Italy

2

-5 -2 0

-10 -4

-6

-8

MP

6

(k) Spain

4

2

-2

0

-4

-6

MP

Notes: Contribution of the identified shocks to credit volume growth (median). CS indicates Credit Supply shocks, AS Aggregate Supply shocks, AD Aggregate Demand shocks, MP Monetary Policy shocks.

Figure 10: Evolution of structural credit supply shocks and BLS credit supply index (a) Austria

(b) Belgium 15

1

1

10 0.5

0

0.5

5

0

0

0

-5

-0.5

10

5

-0.5

-5

-10

structual shocks TVP-VAR

BLS credit supply index

structual shocks TVP-VAR

(c) Finland

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

-15

2005Q1

-10

-1.5

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

-20

2005Q1

-1.5

-1

2006Q1

-15

2005Q3

-1

BLS credit supply index

(d) France

1.5

1.5

1

1

0.5

0.5

0

0

-0.5

-0.5

-1

-1

-1.5

-1.5

20 15 10 5 0 -5

-15

structual shocks TVP-VAR

structual shocks TVP-VAR

(e) Germany

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

-20

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

-10

BLS credit supply index

(f) Greece

1

1.5

10

0.5

1

5

0.5 0

0 0

-0.5

-5 -0.5

structual shocks TVP-VAR

-1

BLS credit supply index

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

-1.5

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

-15

2005Q3

-10

2005Q1

-1

-1.5

structual shocks TVP-VAR

(g) Ireland

(h) Italy

1

15

15

1

10

10

0.5

0.5

5

0

-5 -0.5

5

0

0

0 -5

-0.5

-10

-10

-1 -15

-15 -1.5

-20

structual shocks TVP-VAR

-20

BLS credit supply index

structual shocks TVP-VAR

(i) Netherlands

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

-25

2005Q1

-2

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

-25

2005Q1

-1.5

2005Q3

-1

BLS credit supply index

(j) Portugal

1.5

1

40 30

1

30 20

0.5

20 0.5

10

10

0

0

0

0 -10

-0.5

-10

-0.5

-20 -1

-20 -30

-1

-40

-30

structual shocks TVP-VAR

structual shocks TVP-VAR

BLS credit supply index

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

-50

2005Q1

-1.5

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

-40

2005Q1

-1.5

BLS credit supply index

(k) Spain 1 10 0.5

5

0

0 -5

-0.5

-10 -1

-15

structual shocks TVP-VAR

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

-20

2005Q1

-1.5

BLS credit supply index

Notes: Comparison between the identified structural credit supply shocks (left axis, red bars) and the BLS credit supply indicator (right axis, black line).

30

Figure 11: Impulse response functions to a credit supply shock (model with alternative identification) (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: The blue line and blue areas denote, respectively, the median and confidence bands (16 and 84 percentiles) of the impulse responses to a credit supply shock using the alternative identification. The dotted line denotes the median of the baseline model. The impulse responses are normalized to an expansionary one-standard deviation shock and are expressed in percent terms.

Actual

Counterfactual BL

2013Q1

2012Q3

2012Q1

2011Q3

Counterfactual BL

2011Q1

-8

Counterfactual ASR Actual

6

4

2

0

-2

-4

-6

Counterfactual ASR

Actual Counterfactual BL

(i) Netherlands

Counterfactual BL Counterfactual ASR

3

(j) Portugal

2

1

-1

0

-2

-3

-4

-5

2013Q1

-10

2013Q1

-6

2012Q3

-4

2012Q3

0

2012Q1

-2

2012Q1

2

2011Q3

(g) Ireland

2011Q3

4 Counterfactual BL

2011Q1

Actual

2011Q1

6 0

-4

-6

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2013Q1

2

2012Q3

-2

2012Q3

4

2012Q1

(f) Greece

2012Q1

Counterfactual ASR

2011Q3

2011Q1

2010Q3

2010Q1

Counterfactual BL

2010Q3

8 Counterfactual BL

2010Q3

Counterfactual ASR

2009Q3

-10

2010Q1

-8

-8

2010Q1

-6

2009Q1

-4

2009Q3

-2

2009Q3

0

2008Q3

2

2009Q1

6

2009Q1

(e) Germany

2008Q1

4

2008Q1

(c) Finland

2008Q3

6

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

(a) Austria

2008Q3

8

2007Q3

Actual

2008Q1

8

2007Q3

Actual

2008Q1

Counterfactual ASR

2007Q1

-5

2007Q3

-4

-12

2007Q3

-3

-10

2007Q1

-8

2006Q3

-6

2007Q1

-4

2007Q1

-2

2006Q3

2

2006Q1

0

2006Q3

Counterfactual ASR

2006Q3

3

4

2006Q1

6

2005Q3

4

2006Q1

8

2006Q1

2

2005Q1

-6

2005Q3

-4

2005Q1

-2

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

0

2005Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q1

2009Q3

2009Q1

2008Q3

2

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q1

4

2005Q3

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

2010Q3

2010Q1

2009Q3

2009Q1

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

6

2005Q1

2013Q1

2012Q3

2012Q1

2011Q3

2011Q1

Counterfactual BL

2010Q3

Counterfactual BL

2010Q3

2010Q1

2009Q3

2009Q1

2008Q3

2008Q3

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Counterfactual BS

2010Q3

2010Q1

2009Q3

2009Q1

2008Q1

2008Q1

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Counterfactual BL

2010Q1

2009Q3

2009Q1

2008Q3

Actual

2008Q3

2007Q3

Actual

2008Q1

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Actual

2008Q1

Actual

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Actual

2007Q3

2007Q1

2006Q3

2006Q1

2005Q3

2005Q1

Figure 12: Evolution of YoY real GDP growth in the absence of credit supply shocks (model with alternative identification)

4

(b) Belgium

3

2

1

-1

0

-2

-3

-4

-5

Counterfactual ASR

(d) France

-1

1

0

-2

Counterfactual ASR

4

(h) Italy

2

-2 0

-4

-6

-8

Counterfactual ASR

5

(k) Spain

4

3

2

-1

1

-2

0

-3

-4

-5

-6

Counterfactual ASR

Notes: Actual and counterfactual evolution of real GDP growth. BL denotes the baseline model; ASR denotes the model with alternative sign restrictions.

Appendix A: Convergence Diagnostics

The Markov Chain Monte Carlo (MCMC) algorithm estimates 1575 hyperparameters (20 for S, 1540 for Q, 15 for W ) and 70 × t coefficients (5 × t for H, 10 × t for A, and 55 × t for B). In order to assess the convergence of the algorithm, we consider three convergence statistics for the estimated parameters: • Autocorrelation of the chain at lag = 20. Low autocorrelation indicates that the draws are almost independent, and, therefore, the algorithm is efficient. P • Inefficiency factors. The inefficiency factor is computed as 1 + 2 ∞ k=1 ρk , where ρk is the sample autocorrelation at lag k, and quantifies the relative efficiency loss in the computation of the variance from correlated versus independent samples (Chib, 2001). Its inverse is the relative numerical efficiency proposed by Geweke (1992), As in Primiceri (2005), we use a 4% tapered window for the estimation of the spectral density at frequency zero. Values of the inefficiency factors below or around 20 are considered as satisfactory. • Raftery and Lewis (1992) diagnostic. It indicates the total number of runs required to achieve a certain precision. As in Primiceri (2005), we test the precision for the 0.025 and 0.0975 quantiles of marginal posteriors, and set the desired accuracy to 0.025 and the probability of achieving the required accuracy to 0.95. A satisfactory level of accuracy of the sampling algorithm is obtained if the required number of runs is lower than the number of iterations actually used. The results of the three tests are reported in Table 8. They suggest that the convergence of the sample algorithm is achieved for all countries. In fact, for the hyperparameters and coefficients the 20th order sample autocorrelation is generally very low, the inefficiency factors are lower than 20 (with the exception of a few cases), and the Raftery and Lewis (1992) required runs are always far less than the total number of iterations used (15000).

33

34

Greece

Germany

France

Finland

Belgium

Austria

S Q W H A B S Q W H A B S Q W H A B S Q W H A B S Q W H A B S Q W H A B

Median 0.008 0.020 −0.006 −0.002 0.008 0.004 0.002 0.089 −0.029 0.000 0.007 0.060 −0.014 0.054 −0.007 −0.005 −0.006 0.058 −0.023 0.071 0.005 0.003 0.000 0.040 −0.006 0.006 −0.005 0.002 −0.002 0.003 0.006 0.004 0.027 −0.004 0.002 0.018

Mean 0.026 0.022 −0.013 0.000 0.010 0.009 0.010 0.107 −0.020 0.002 0.007 0.084 −0.007 0.067 0.000 −0.004 −0.006 0.072 −0015 0.078 0.012 0.004 0.000 0.059 0.002 0.008 0.004 0.004 −0.003 0.006 −0.007 0.004 0.021 −0.002 0.003 0.019

Min −0.037 −0.121 −0.073 −0.092 −0.101 −0.103 −0.033 −0.120 −0.058 −0.087 −0.097 −0.100 −0.059 −0.142 −0.054 −0.089 −0.098 −0.097 −0.052 −0.138 −0.036 −0.086 −0.092 −0.095 −0.075 −0.152 −0.040 −0.081 −0.094 −0.101 −0.055 −0.124 −0.033 −0.066 −0.083 −0.094

Max 0.302 0.239 0.027 0.142 0.190 0.236 0.076 0.444 0.065 0.100 0.123 0.395 0.055 0.498 0.063 0.094 0.083 0.479 0.077 0.345 0.073 0.102 0.108 0.367 0.154 0.176 0.052 0.122 0.097 0.188 0.028 0.154 0.074 0.098 0.114 0.144

20th order sample autocorrelation Median 5.849 5.610 1.990 1.246 1.394 2.918 8.132 10.233 1.706 1.504 1.574 5.713 7.911 8.705 1.656 1.251 1.302 5.564 7.119 9.011 1.489 1.228 1.294 4.773 4.889 4.756 1.731 1.192 1.312 2.320 3.298 4.209 1.844 1.184 1.194 3.575

Mean 6.812 5.862 2.060 1.307 1.697 3.435 8.453 10.404 1.843 1.610 1.710 6.792 8.096 8.814 1.572 1.344 1.350 6.205 7.425 9.189 1.511 1.278 1.431 5.811 4.593 4.891 1.902 1.236 1.398 2.632 3.171 4.394 2.009 1.233 1.209 3.887

Min 4.074 1.561 1.436 0.494 0.551 0.606 6.612 2.881 1.072 0.802 0.651 0.893 3.484 2.245 0.787 0.665 0.518 0.787 1.658 2.228 1.190 0.531 0.606 0.671 3.073 2.040 0.987 0.607 0.535 0.626 1.536 1.021 0.989 0.518 0.544 0.910

Inefficiency factors

Table 6: Convergence diagnostics

Max 18.687 14.647 2.735 3.646 11.407 13.118 11.945 21.295 2.756 4.362 4.972 20.340 11.472 23.588 2.340 4.309 2.742 21.520 11.251 16.428 1.856 3.459 4.477 17.349 6.237 10.155 3.347 2.342 3.749 8.405 4.161 11.145 3.444 2.946 2.548 11.215

Median 188 283 188 161 160 173 196 309 173 160 160 188 173 309 173 160 160 173 184 309 173 160 160 173 203 309 188 160 160 173 173 239 188 160 160 173

Mean1 192 311 192 170 165 185 199 376 185 170 168 220 178 361 179 163 164 214 182 373 173 162 160 199 263 326 180 167 165 175 179 252 190 166 161 196 Min 148 148 160 143 143 143 160 160 160 143 143 143 148 148 148 143 143 143 148 148 148 143 143 143 173 148 148 143 143 143 148 143 148 143 143 143

RL runs Max 260 2800 239 350 462 1470 260 2544 239 283 1184 4302 239 2318 220 220 924 2660 220 1400 220 239 239 1230 804 1212 220 283 800 816 220 1200 239 239 220 1094

35

S Q W Ireland H A B S Q W Italy H A B S Q W Netherlands H A B S Q W Portugal H A B S Q W Spain H A B Notes: 1 The mean

Median Mean Min −0.011 −0.006 −0.067 0.152 0.179 −0.127 0.000 0.005 −0.040 0.003 0.005 −0.066 0.006 0.007 −0.075 0.104 0.165 −0.088 −0.022 −0.019 −0.067 0.020 0.026 −0.178 −0.011 −0.006 −0.050 0.000 0.000 −0.074 −0.001 0.000 −0.091 0.020 0.024 −0.090 −0.032 −0.008 −0.076 0.060 0.072 −0.153 0.000 0.008 −0.029 0.003 0.005 −0.087 0.000 −0.001 −0.103 0.055 0.074 −0.082 −0.006 −0.004 −0.057 0.014 0.020 −0.126 −0.005 −0.008 −0.080 −0.001 0.000 −0.095 −0.003 −0.004 −0.104 0.025 0.035 −0.117 −0.004 −0.007 −0.065 −0.006 −0.006 −0.110 −0.010 −0.011 −0.071 −0.005 −0.005 −0.078 0.000 0.000 −0.091 −0.007 −0.006 −0.102 of the RL runs is rounded to the closest

Max 0.053 0.641 0.046 0.142 0.099 0.616 0.035 0.380 0.053 0.098 0.081 0.200 0.114 0.401 0.069 0.131 0.095 0.389 0.062 0.255 0.039 0.126 0.083 0.223 0.042 0.086 0.040 0.103 0.079 0.113 integer.

20th order sample autocorrelation Median 5.875 11.929 2.480 1.638 1.574 7.399 2.744 6.230 1.365 1.228 1.370 2.951 6.795 8.927 1.964 1.445 1.542 5.044 2.147 3.810 1.369 1.228 1.202 3.607 1.468 1.870 1.462 0.973 0.979 1.190

Mean 6.206 12.684 2.532 1.797 1.742 9.389 2.962 6.667 1.453 1.366 1.571 3.303 6.943 9.354 2.332 1.550 1.611 5.861 2.654 4.295 1.540 1.268 1.371 4.710 1.544 1.929 1.444 1.014 1.011 1.219

Min 1.936 1.766 1.829 0.702 0.529 0.953 1.086 1.783 1.063 0.644 0.431 1.033 1.769 2.341 1.636 0.675 0.759 0.851 1.229 0.742 0.945 0.604 0.447 0.677 1.073 0.578 0.822 0.500 0.517 0.478

Inefficiency factors

Table 6 (continued): Convergence diagnostics

Max 10.980 28.096 3.599 4.866 5.739 25.651 7.110 19.151 2.349 3.183 3.779 10.902 13.312 19.140 4.034 4.187 4.304 19.695 7.975 14.865 2.502 2.808 3.622 14.693 2.312 4.125 1.987 1.879 1.926 2.833

Median 188 346 188 173 160 188 173 309 173 160 160 173 203 354 173 160 160 173 188 239 173 160 160 173 167 188 173 160 160 160

Mean1 190 466 196 180 164 257 182 348 178 171 165 182 201 419 190 168 164 200 204 249 179 166 163 210 167 203 183 160 160 160 Min 148 148 154 143 143 143 148 144 148 148 148 143 160 148 148 143 143 96 148 148 148 148 143 143 148 143 160 143 143 143

RL runs Max 283 5136 260 494 368 4315 239 2946 260 576 260 933 260 2544 338 692 513 1130 408 838 239 239 548 2325 188 824 239 220 239 306

Appendix B: The Data

To estimate the VAR we use quarterly data covering the period 1980:1-2013:2, with the exception of Greece (1985:1-2012:4), Ireland (1990:1-2013:2) and Italy (1995:1-2013:2). • Real GDP (yt ): Nominal gross domestic product at market prices, seasonally adjusted (OECD Economic Outlook - Datastream mnemonic: (CC)OCFGPNB)4 divided by the GDP deflator (see description below). For Germany the series has been adjusted for the break of 1990. • GDP Deflator (πt ): GDP deflator index, seasonally adjusted (OECD Economic Outlook - (CC)OCFDGDE; For Germany: OECD Main Economic Indicators - Datastream mnemonic BDQNA057E). • Short-term interest rate (strt ): Money market interest rates, 3-month rates (Eurostat Statistics Database - Codes: irt st q, irt h mr3 q). • Credit to the non-financial private sector (lt ): Credit to non-financial private sector, lending sector: All sectors, adjusted for breaks (BIS Long series on credit to private nonfinancial sectors). The series have been seasonally adjusted using a 5-order Henderson filter. • Composite lending rate (clrt ): Calculated as the weighted average of interest rates on lending to households for house purchase, quarterly average of monthly data, new business (ECB MFI Interest Rate Statistics - Code: MIR.M.(CC).B.A2C.A.R.A.2250.EUR.N) and lending to non-financial corporations, quarterly average of monthly data, new business (ECB MFI Interest Rate Statistics - Code: MIR.M.(CC).B.A2A.A.R.A.2240.EUR.N). For Greece these data are complemented with Eurostat data, and the lending rate to nonfinancial corporations is relative to loans up to and including EUR 1 million. The weights are based on outstanding amounts of loans to households for house purchase (ECB Balance Sheet Items - Code: BSI.Q.(CC).N.A.A22.A.1.U2.2250.Z01.E) and loans to non-financial corporations (ECB Balance Sheet Items - Code: BSI.Q.(CC).N.A.A20.A.1.U2.2240.Z01.E).

To construct our BLS credit supply index we collect data from the euro area Bank Lending Survey (BLS). Following Ciccarelli et al. (2010), we define changes in credit supply as factors related to bank balance sheet capacity and competition pressures. In particular, we consider factors A and B in the BLS questions Q2, Q9 and Q11. The index is calculated as the additive inverse of the average of the net percentage of banks answering A and B. For France, the weighted net percentage is used. For Austria and Ireland, the diffusion index is used. For Belgium, BLS data are collected from the National Bank of Belgium, and the index is calculated as the average of the net percentage of banks answering A and B. No data are available for Finland and Greece.

4

CC stands for Country Code.

36

Appendix C The impulse response functions of other shocks (baseline model)

Figure 13: Impulse response functions to an aggregate supply shock (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: The blue line denotes the median of the impulse responses to an aggregate supply shocks. The shaded area indicates the 16 and 84 percentiles. The impulse responses are normalized to an expansionary one-standard deviation shock and are expressed in percent terms.

Figure 14: Impulse response functions to an aggregate demand shock (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: The blue line denotes the median of the impulse responses to an aggregate demand shock. The shaded area indicates the 16 and 84 percentiles. The impulse responses are normalized to an expansionary one-standard deviation shock and are expressed in percent terms.

Figure 15: Impulse response functions to a monetary policy shock (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: The blue line denotes the median of the impulse responses to a monetary policy shock. The shaded area indicates the 16 and 84 percentiles. The impulse responses are normalized to an expansionary one-standard deviation shock and are expressed in percent terms.

Appendix D Statistical tests for IRFs and FEVD

42

6=

6=

=

=

6=

6=

6=

=

6=

6=

6=

6=

6=

6=

6=

6=

6=

6=

6=

6=

=

6=

Belgium

Finland

France

Germany

Greece

Ireland

Italy

Netherlands

Portugal

Spain

π

6=

6=

=

=

=

6=

6=

6=

6=

=

6=

1-3

6=

6=

6=

6=

=

=

6=

=

6=

6=

=

1-2

l

6=

=

6=

6=

6=

=

6=

6=

6=

6=

=

1-3

6=

=

6=

6=

6=

=

=

6=

6=

6=

6=

1-2

clr

6=

6=

=

6=

6=

=

=

=

=

6=

6=

1-3

6=

6=

6=

6=

6=

=

6=

6=

6=

=

6=

1-2

str

6=

6=

6=

6=

6=

=

6=

6=

6=

=

6=

1-3

6=

=

6=

6=

=

6=

6=

6=

6=

6=

6=

1-2

y

=

6=

6=

6=

=

6=

6=

6=

=

=

6=

1-3

6=

6=

=

6=

6=

6=

=

6=

6=

6=

=

1-2

π

6=

6=

=

=

=

6=

6=

6=

=

=

=

1-3

=

6=

6=

6=

=

=

6=

=

6=

6=

=

1-2

l

6=

6=

6=

6=

6=

=

6=

6=

6=

6=

=

1-3

6=

=

=

6=

6=

=

=

=

6=

=

6=

1-2

Wilcoxon rank sum test2 clr

6=

6=

=

6=

6=

=

=

=

=

=

=

1-3

6=

6=

6=

6=

6=

=

6=

6=

6=

=

6=

1-2

str

6=

=

6=

6=

6=

=

6=

6=

=

=

6=

1-3

Notes: 1 t-test for equal means of two distributions. A 6= (=) symbol indicates that the null hypothesis that the vectors are from distributions with equal means is (not) rejected at the 5% significance level. A green (red) colored cell indicate that the absolute value of the median has increased (decreased). For each country, three dates (1, 2 and 3) are considered: 2005Q1, 2009Q1 and 2013Q2 (2012Q4 for Greece). The test is performed between period 1 and 2, and between period 1 and 3. 2 Wilcoxon rank sum test for equal medians of two distributions. A 6= (=) symbol indicates that the null hypothesis that they are from the same continuous distributions is (not) rejected at the 5% significance level. A green (red) colored cell indicate that the absolute value of the median has increased (decreased). For each country, three dates (1, 2 and 3) are considered: 2005Q1, 2009Q1 and 2013Q2 (2012Q4 for Greece). The test is performed between period 1 and 2, and between period 1 and 3.

6=

6=

6=

6=

=

6=

6=

6=

6=

=

6=

Austria

1-2

1-3

1-2

y

t-test1

Table 7: Has the impact effect of credit supply shocks changed over time?

43

=

=

Portugal

Spain

=

=

=

6=

=

6=

6=

6=

6= 6=

6=

6=

6=

=

6=

6=

=

=

6=

1-3

6=

=

=

6=

=

6=

6=

=

6=

1-2

l

6=

6=

6=

6=

=

6=

6=

6=

6=

=

6=

1-2

clr

=

6=

=

6=

6=

6=

6=

=

=

6=

6=

1-3

6=

6=

6=

6=

=

=

6=

6=

6=

=

=

1-2

str

=

=

=

6=

6=

=

6=

6=

=

=

=

1-3

=

=

6=

=

6=

=

6=

6=

6=

=

6=

1-2

y

=

=

=

6=

6=

6=

=

6=

6=

6=

6=

1-3

6=

=

6=

6=

6=

=

=

6=

6=

=

=

1-2

π

=

=

=

6=

=

=

6=

6=

6=

=

=

1-3

6=

6=

6=

=

=

6=

6=

6=

6=

=

=

1-2

l

6=

6=

6=

=

6=

=

6=

6=

=

6=

6=

1-3

6=

6=

6=

6=

=

6=

6=

6=

6=

=

6=

1-2

Wilcoxon rank sum test2 clr

=

6=

=

6=

6=

6=

6=

=

=

=

=

1-3

6=

6=

6=

6=

=

=

6=

6=

6=

=

6=

1-2

str

=

6=

=

6=

6=

6=

6=

6=

=

=

=

1-3

Notes: 1 t-test for equal means of two distributions. A 6= (=) symbol indicates that the null hypothesis that the vectors are from distributions with equal means is (not) rejected at the 5% significance level. A green (red) colored cell indicate that the median has increased (decreased). For each country, three dates (1, 2 and 3) are considered: 2005Q1, 2009Q1 and 2013Q2 (2012Q4 for Greece). The test is performed between period 1 and 2, and between period 1 and 3. 2 Wilcoxon rank sum test for equal medians of two distributions. A 6= (=) symbol indicates that the null hypothesis that they are from the same continuous distributions is (not) rejected at the 5% significance level. A green (red) colored cell indicate that the median has increased (decreased). For each country, three dates (1, 2 and 3) are considered: 2005Q1, 2009Q1 and 2013Q2 (2012Q4 for Greece). The test is performed between period 1 and 2, and between period 1 and 3.

=

=

=

6=

Netherlands

6=

6=

6=

=

Italy

=

6=

6=

=

Ireland

6=

6=

=

Greece

6=

=

=

6=

Germany

6=

6=

6=

6=

6=

France

6=

6=

=

6=

Finland

=

6=

=

=

Belgium

=

=

=

6=

1-3

1-2

π

1-3

Austria

1-2

y

t-test1

Table 8: Has the variance decomposition of credit supply shocks at horizon = 20 changed over time?

Appendix E The impulse response functions of other shocks (alternative identification)

Figure 16: Impulse response functions to an aggregate supply shock (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: The blue line and blue areas denote, respectively, the median and confidence bands (16 and 84 percentiles) of the impulse responses to an aggregate supply shock using the alternative identification. The dotted line denotes the median of the baseline model. The impulse responses are normalized to an expansionary one-standard deviation shock and are expressed in percent terms.

Figure 17: Impulse response functions to an aggregate demand shock (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: The blue line and blue areas denote, respectively, the median and confidence bands (16 and 84 percentiles) of the impulse responses to an aggregate demand shock using the alternative identification. The dotted line denotes the median of the baseline model. The impulse responses are normalized to an expansionary one-standard deviation shock and are expressed in percent terms.

Figure 18: Impulse response functions to a monetary policy shock (a) Austria

(b) Belgium

(c) Finland

(d) France

(e) Germany

(f) Greece

(g) Ireland

(h) Italy

(i) Netherlands

(j) Portugal

(k) Spain

Notes: The blue line and blue areas denote, respectively, the median and confidence bands (16 and 84 percentiles) of the impulse responses to a monetary policy shock using the alternative identification. The dotted line denotes the median of the baseline model. The impulse responses are normalized to an expansionary one-standard deviation shock and are expressed in percent terms.

FP_Falagiarda_Bijsterbosch.pdf

†European Central Bank and University of Bologna. .... and Gambetti, 2009; Clark and Terry, 2010; Fern ́andez-Villaverde and Rubio-Ram ́ırez, 2010;. Franta ...

13MB Sizes 0 Downloads 2 Views

Recommend Documents

No documents