The bank lending channel in monetary transmission in the euro area: evidence from Bayesian VAR analysis Matteo Bondesan Graduate student University of Turin (M.Sc. in Economics) Collegio Carlo Alberto (M.A. in Economics) Annual Meeting of the Austrian Economic Association The Future of the European Economic and Monetary Union June 2015, Klagenfurt, Austria M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 1 / 21 Introduction Empirical motivations Introduction The current period of crisis in credit markets has highlighted the crucial role of the behaviour of banks in the transmission mechanism of monetary policy. The crisis has been a reminder that it is impossible to understand monetary policy without understanding financial markets and financial intermediation. Pre-crisis: systematic stability in euro area monetary policy, accompanied by homogeneity in the transmission mechanism to short-term and long rates. Post-crisis: dissimilar behaviour of short-term interest rates, loans and deposits; in particular between long-term interest rates (higher) and long-term loans, deposits (lower). Evidence of structural problems we tried to shed light on the relationship among different categories of loans, their lending rates, interest rates at different maturities and a set of monetary aggregates. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 2 / 21
Introduction Table of Contents 1 Introduction Literature review This Paper 2 Linking representations of reality: VAR-BVAR-DSGE model 3 The Model 4 Results Bayesian VAR evidence DSGE evidence Quantitative Easing: counterfactual experiment 5 Findings and Conclusions M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 3 / 21 Introduction Literature review Literature review Macro and microeconomic approaches Microeconomic emphasis: De Santis, Surico (2013): heterogeneity in banks characteristic leads to diversified monetary transmission. Gambacorta, Illes, Lombardi (2014): bank-specific characteristics have a large impact on credit provision. Apergis, Miller, Alevizopoulou (2015): the reaction of loans growth to the actual short-term interest rate and the interest rate target-rule. Macroeconomic analysis: Bernanke, Gertler, Gilchrist (1999): macroeconomic impact of financial intermediaries behaviour and the financial situation of borrowers. Christiano, Motto, Rostagno (2007): key role of frictions in banking markets. Lenza, Pill, Reichlin (2010): Central banks reactions from normal to crisis scenario. Fahr, Motto, Rostagno, Smets, Tristani (2013): evolution of monetary policy in changing times. Giannone, Lenza, Reichlin (2014): bayesian VAR based findings providing breaks in historical regularities after the crisis. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 4 / 21
Introduction This Paper This Paper Methodological issues Disentangling the bank lending channel: Asset demand channel. Liability demand of money. Asset supply channel. Our focus: the euro area because the financial intermediation is a very feature of European system. 1 Our approach: Heterogeneity in the quantity on credit considering different categories of demanders. Dissimilarity in the cost of borrowing. Monetary aggregates and interest rates. Set of (pre-determined) macroeconomic variables. Our pursuit lies on the reaction of key selected variables to monetary, demand and supply shocks. We provide VAR based evidence, making usage of the bayesian shrinkage. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 5 / 21 Linking representations of reality: VAR-BVAR-DSGE model Linking representations of reality VAR-BVAR-DSGE model VAR model plays the role of a reference model. If DSGE models are indeed misspecified, the VAR will attain the highest posterior probability and the model comparison is based on the question: given a particular loss function, what DSGE model best mimics the dynamics captured by the VAR? VARs typically have many more parameters than DSGE models and the role of prior distributions is mainly to reduce the effective dimensionality of this parameter space to avoid over-fitting. The bayesian paradigm provides a rich framework for inference and decision making with modern macroeconometric models such as DSGE and VARs. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 6 / 21
The Model Bayesian VAR model Model specification How monetary policy is transmitted to the real economy? The bank lending channel Is there a mechanism such that: Expansionary monetary policy Interest rate Asset side value Loans deposits (liability side, along with reserves) aggregate investment aggregate output? We implement a Bayesian VAR model, and we use a data set including 12 quarterly macroeconomic, financial, monetary and credit variables from January 2000 to December 2014 (Data source: ECB Statistical Data Warehouse). We choose the Litterman/Minnesota prior formulation, which assumes a normal prior for θ, and the var-covariance matrix of the error term, Σ ε to be known; thus the latter is replaced with its estimate ˆΣ ε, and it is restricted to be a diagonal matrix. We are left with a prior for the VAR coefficient θ, where θ N(θ 0, V 0 ). We select the values for the hyper-parameters, µ 1, λ 1, λ 2, λ 3 which describe the behaviour of the prior mean and covariance, in order to lessen the risk of over-fitting and to emphasize the fact we want sample information dominates the prior. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 7 / 21 Bayesian VAR model The Model In order to inquire the presence of the so-called bank lending channel, we estimate a VAR model with p = (2) lags: with ε N(0, Σ ε ). y t = α 0 + p A j y t j + ε t j=1 The large dimension, n = 12 and p = 2, of our VAR model makes it preferable the usage of a bayesian shrinkage to mitigate ex-ante the possibility of over-fitting, since the large number of parameters. The likelihood function: l(θ, Σ ε ) ( Σ ε I t ) 1 2 exp{ 1 2 (y (I m X)θ) (Σ ε I T ) 1 (y (I m X)θ)} Assume Σ ε is known, and a multivariate normal prior for θ, θ N(θ 0, V 0 ): Π(θ) ( V 0 ) 1 2 exp{ 1 2 (θ θ 0) V 1 0 (θ θ 0 )} M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 8 / 21
The Model The posterior density can be written as: π(θ y) = exp{ 1 2 (w W θ) (w W θ)} exp{ 1 2 (θ θ) W W (θ θ)+(w W θ) (w W θ)} π(θ y) exp{ 1 2 (θ θ) W W (θ θ)} exp{ 1 2 (θ θ) V 1 (θ θ)} Data set: We begin with 26 variables, but in order to provide clear intuitions, we shrink our BVAR model to a 12 variables specification. Macroeconomic block: GDP, Industrial production, Unemployment, Inflation (HICP) and Households consumption expenditures. Policy instrument: Euribor-six-months. Financial/Credit variables: Money base, Reserves, Loans to households, Loans to non financial corporations, Lending rate of loan to households and Lending rate of loan to non financial corporations. Cholesky Identification scheme: we consider the conduct of the ECB from 2000 to 2014 to be based on a set of structural relationships among innovations in monetary aggregates, credit variables and interest rates. The Euribor six months rate represents the proxy (policy instrument) for the policy target rate structural interpretation to the innovations affecting the status quo. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 9 / 21 Results Bayesian VAR evidence BVAR evidence Monetary shocks.monetary standpoint. Euribor-six-months innovation, ε euribor : Figure : Responses of loans to ε euribor ε euribor Loans. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 10 / 21
Results Bayesian VAR evidence BVAR evidence Monetary shocks.monetary standpoint. Reserves innovation, ε reserves : Figure : Responses of loans to ε reserves ε reserves Asset Loans. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 11 / 21 Results Bayesian VAR evidence BVAR evidence Demand shock.demand standpoint. Consumption innovation, ε exp : Figure : Responses of loans to ε exp ε exp (delaying) increase in Loans to households. ε exp Loans to non-financial corporations (gradually). M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 12 / 21
Results Bayesian VAR evidence BVAR evidence Supply shock.supply standpoint. Gross domestic product innovation, ε gdp : Figure : Responses of loans to ε gdp ε gdp Loans (on impact), then both gradually with heterogeneity. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 13 / 21 Results DSGE evidence DSGE evidence (CMR 10) Theoretical benchmark We calibrate the Christiano Motto Rostagno (2010) DSGE model on euro area data in order to have a simulated theoretical benchmark: Monetary policy restriction: interest rate Value of Asset Net-worth. Monetary policy restriction: interest rate Total Loans (delaying), then they significantly. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 14 / 21
Results Quantitative Easing: counterfactual experiment Quantitative Easing Counterfactual experiment What would happen if the quantitative easing was implemented in 2013, and what will happen in the next two years? Euro area base money: Figure : Sum of Currency in circulation, Overnight deposits and Deposits with agreed maturity The peak between 2012 and 2013 corresponds to (other) non-standard measures adopted by the ECB to cope with crisis times. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 15 / 21 Results Quantitative Easing: counterfactual experiment We rely on our original Bayesian VAR for the scenario without QE, whereas we construct the second fashion by programming a simple code to account for 18 months of QE s action. We proceed according to the following scheme:.1. Since our quarterly data, we transform the 1.1 trillion euro within the 18 periods, into 183.33 millions euro per quarter..2. Creating a new money base variable having the last 6 quarters (18 months) updated by the effect of QE: previous quarter value plus 183.33 millions euro..3. Programming, within the sample, a new bayesian VAR model (perfectly comparable with our), but in the new money base variable..4. Forecasting out-of-sample, from the first quarter of 2015 to the first quarter of 2017 (next two years), with both models. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 16 / 21
Results Quantitative Easing: counterfactual experiment Experimental evidence Table : Results from QE experiment Higher with QE Deposits Loans to households GDP Industrial Production Reserves Lower with QE Euribor-six-month Unemployment Loans to non-financial corporations M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 17 / 21 Results Quantitative Easing: counterfactual experiment QE scenarios Projection of gross domestic product: Figure : Gross domestic product M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 18 / 21
Findings and Conclusions Findings and Conclusions We found evidence for the following general scheme to apply: Expansionary monetary policy interest rate loans (asset side) reserves and deposits (liability side) aggregate output. Our threefold analysis allows to conclude: BVAR evidence: ε euribor Loans. DSGE evidence: Expansionary monetary policy Net-worth, and Total-loans. QE experiment: QE Euribor-six-months and Unemployment, Loans to households and GDP. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 19 / 21 Greetings Findings and Conclusions Thank you for being here. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 20 / 21
Future Improvements Findings and Conclusions 1 Theoretical microfoundation. What Next? 2 Deepening the bayesian prior choice. 3 Modelling the banking sector + Including a stock market index as a control variable. M.Bondesan (CCA, UNITO) Bank lending channel NOeG 2015 21 / 21