Bank capital constraints, lending supply and real economy: evidence from a BVAR model. by A.M. Conti A. Nobili, F.M. Signoretti (Banca d Italia)

Transcription

1 Bank capital constraints, lending supply and real economy: evidence from a BVAR model by A.M. Conti A. Nobili, F.M. Signoretti (Banca d Italia) Fifth Research Workshop of the MPC Task Force on Banking Analysis for Monetary Policy Bruxelles, 1-2 February 2018

2 Motivation: the case of Italy Increase in bank capitalization due to change in regulation/supervision What about short-run effects for lending conditions and the real economy? 13 Tier1 ratio (per cent) 200 Components of Tier1 ratio (euro billion) Tier1 ratio Tier1 capital RWAs (right-hand scale) 2

3 This paper Estimates a medium-scale B-VAR model for the Italian economy, which includes a large number of macro and banking variables Two objectives: 1. Develop a unified framework for the analysis of credit and banking variables in Italy Taking into account multiple links between real economy and the banking system and «within» the banking system (e.g.; loan growth, loan rates, credit quality, profitability, capitalization) Improving upon single-equation models for policy analysis in BoI (Albertazzi et al., 2014), Bofondi and Ropele (2011), Nobili and Zollino (2017) 2. Use the model to study the effect of a bank capital shock Important source of lending supply shocks Derive policy implications from policymakers and regulators 3

4 Related literature on bank capital shocks The existing methodologies, so far 1. Basel Committee on Banking Superivisons (BCBS, 2010, 2015), Angelini et al. (2011) using a range of DSGE models. Limited effects during Basel III episode 2. One-off event studies using micro data also from supervisory reports (Jimenez et al. 2016; Mesonniér and Monks; 2015; Aiyar et al., 2014; Gropp et al., 2016; De Jonghe et al., 2016). Effects seem to be much more relevant in some epsiodes 3. VAR macro models with endogenous bank capital a. sign-restrictions (Noss and Toffano (2014); Meeks (2017), Groß et al. (2015) and Kanngiesser et al. (2017). Budnik et al. (2018) Macropru Task Force b. Economic capital two-step approach (Hancock et al., 1995; Berrospide and Edge, 2010; De Nicolò, 2015; Mésonnier and Stevanovic, 2017) We propose a different way to identify bank capital shocks following Giannone, Lenza, Pill and Reichlin (EJ, 2012) for unconventional MP conditional forecasting approach to rule out endogeneity 4

5 The model (1/2) Medium-scale Bayesian VAR model Quarterly data: 1993q1-2015q4 16 endogenous variables 4 Macro: GDP, Inflation, short- & long-term rates 4 Loan quantity and rates: NFC loan growth & new lending rate, HH mortgages & new mortgage rate 2 loan default rates: NFC & HH new bad loan ratio 4 bank profitability: Net interest income, other income, loan-loss and other provisions, operational expenses 2 bank capitalization: Tier1 capital ratio, bank stock prices 5

6 Data transformation: (log)-levels 5 lags The model (2/2) Bayesian techniques allow to estimate large VARs by shrinking the coefficients toward a prior model that is parsimonious and naïve Sims-Zha (1998) prior : Normal-Wishart; overall tightness = 0.1; decay factor = 1,0; sum-of-coefficients = 1.0; cointegration = 1.0 Related literature on BVAR models Methodology: Sims and Zha (1998); Banbura, Giannone and Reichlin (2010); Waggoner and Zha (1999); Aastveit, Carriero, Clark and Marcellino (2016) Usefulness to assess policy issues: Jarocinski and Smets, (2008); Lenza, Pill and Reichlin (EJ, 2010); Giannone, Lenza, Pill and Reichlin (2012); Banbura, Giannone and Lenza (2015); Bobeica and Jarocinski (2017) 6

7 The data: macroeconomic variables 7

8 The data: credit variables 8

9 The data: bank profits & capitalization 9

10 Roadmap: 2 parts 1. How does the model fit banking variables? 2. What are the effects of bank capital shocks on lending conditions and real economic activity? Three relevant forecast windows (no longer than two years to avoid Lucas critique!): 2009q2-2010q4: announcement of new Basel III framework 2011q1-2012q4: EBA stress test capital exercise 2013q1-2015q4: Banking Union/Comprehensive Assessment 10

11 1. How does the model fit bank vars since the crisis? Strategy 1. Out-of-sample forecasts of banking variables conditional to actual path of 4 macroeconomic variables: Real GDP, inflation, 3month-Euribor, Italian 10-year BTP yields Compare conditional forecasts, unconditional f. and actual values Unconditional forecast: dynamic forecast of VAR Y given the model Conditional f.: dynamic forecast of VAR Y given the model and future path of variable(s) X If forecast of Y conditional on actual path of X moves away from unconditional f. => X variables are relevant drivers for/can explain Y If conditional f. (still) far from actual values => missing drivers Later explore the existence of structural breaks: the gap b/w conditional f. and actual values could also be due to changes in the relations among variables 11

12 Out-of-sample forecasts (1) Loans, loan rates and default rates following Basel III Note: Conditioning on real GDP, inflation, short-term (3m-euribor) and long-term interest rate (10yBTP). Out-of-sample conditional forecasts starting in 2009q2. 12

13 Out-of-sample forecasts (2) Loans, loan rates and default rates during the EBA stress exercise Note: Conditioning on real GDP, inflation, short-term (3m-euribor) and long-term interest rate (10yBTP). Out-of-sample conditional forecasts starting in 2011q1. 13

14 Out-of-sample forecasts (3) Loans, loan rates and default rates following the start of SSM Note: Conditioning on real GDP, inflation, short-term (3m-euribor) and long-term interest rate (10yBTP). Out-of-sample conditional forecasts starting in 2014q1. 14

15 2. What are the effects of a bank capital shock? We identify a bank capital shock as: k t = Actual value of Tier1 ratio shock t k = k t k t k t = forecast of Tier1 ratio conditional on a large number of variables Large number of conditioning variables captures main reasons why T1 capital may change: macro vars and default rates -> size and riskiness of assets bank profitability -> retained earnings/losses bank stock prices -> equity issuance conditions / market discipline Interpretation k t is the value of capital consistent with developments in its likely main drivers (business cycle, financial conditions, profits). Similar to economic capital concept shock t k captures missing drivers of changes in capital, i.e. regulatory and supervisory pressure to increase capital ratios Conditional forecasts are invariant to the chosen factorization of the variance/covariance matrix (Waggoner and Zha, 1999) 15

16 2. What are the effects of a bank capital shock? In order to estimate the impact of the shock we proceed as follows: Step 1: simulate the model conditional on k t (actual capital) Step 2: simulate the model conditional on k t (conditional path of capital) Impact on variable Y is the difference between the two simulations We use 2 sets of conditioning variables 1. Only macroeconomic variables (less conservative choice) 2. Full set of macro and banking variables (our preferred choice) 16

17 Effects of bank capital shocks: Basel III episode Large and persistent shock with significant effects on loans to HHs Real GDP declined by 2%; limited effect of inflation 17

18 Effects of bank capital shocks: EBA exercise episode Temporary shock with significant effects on loans to both NFCs and HHs Real GDP temporary declined by 0.4%; limited effect of inflation 18

19 Effects of bank capital shocks: Banking Union episode Large and persistent shock with significant effects on loans to NFCs Real GDP declined by 1%; limited effect of inflation 19

20 Large capital shocks or instability in coeffs? Discrepancy between actual values and conditional forecasts may reflect changes in estimated relationships Forecast windows of two years should limit this issue Alternative options to corroborate main findings: Use a time-varying coefficients model using the approach in Koop and Korobilis (2013) Feasible in the Minnesota prior framework Look at in-sample conditional forecasts Very conservative approach! Estimates are a lower bound of the true effects 20

21 Large capital shocks or instability in coeffs? Time-varying coefficients BVAR vs. fixed-coefficients BVAR 21

22 BVAR model using Koop-Korobilis (2013) prior stochastic volatility in the BVAR model 22

23 BVAR model using Koop-Korobilis prior stochastic volatility in the BVAR model 23

24 Time varying VAR Koop Korobilis forecasts: Basel III episode The magnitude of the shock partly attenuates but some effects are larger Differences arise from both the different prior distribution ( Minnesota vs. Sims&Zha prior ) and from stochastic volatility in the estimated equations (i.e. real GDP and inflation) 24

25 Time varying VAR Koop Korobilis forecasts: EBA exercise episode The shock becomes long-lasting and so are the estimated effects, especially on loan rates Differences mainly reflect the different prior distribution ( Minnesota vs. Sims&Zha prior ) and from stochastic volatility in the estimated equations (i.e. long-term rates) 25

26 Time varying VAR Koop Korobilis forecasts: Start of SSM episode The magnitude and persistence of the shock do not change; the real effects are stronger Differences arise mainly from from stochastic volatility in the Tier1 ratio equation 26

27 Large capital shocks or instability in coeffs? in-sample conditional forecasts vs. out-of-sample conditional forecasts 27

28 In-sample vs out-of-sample forecasts: Basel III episode The magnitude and persistence of the shock attenuate The effects are sizable albeit smaller (one-half) 28

29 In-sample vs out-of-sample forecasts: EBA exercise episode The magnitude and persistence of the shock remain unchanged And so are, the estimated effects 29

30 In-sample vs out-of-sample forecasts: Banking Union episode The magnitude and persistence of the shock largely attenuate The estimated effects remain sizable albeit smaller by one-half 30

31 Take away Bank capital shocks acted as adverse credit supply shocks with sizable effects on lending conditions and economic growth Uncertainty about the magnitude of the effects due to potential break in the estimated variance-covariance matrix, i.e. especially, following the Banking Union Policy implications: although capitalization makes the banking sector more resilient in the long-run to adverse shocks, short-run losses may be painful. Avoid excessive/unexpected recapitalization in a lowgrowth environment (due to prociclicality) Consider interactions with (unconventional) monetary policy 31

32 Thank you 32