Credit Shocks and the U.S. Business Cycle. Is This Time Different? Raju Huidrom University of Virginia. Midwest Macro Conference

Transcription

1 Credit Shocks and the U.S. Business Cycle: Is This Time Different? Raju Huidrom University of Virginia May 31, 214 Midwest Macro Conference Raju Huidrom Credit Shocks and the U.S. Business Cycle

2 Background Subprime crisis and the Great Recession of has given a renewed interest in the role of financial frictions. Credit shocks are an independent causal factor for business cycles. Definition: shocks to credit supply conditions unrelated to credit worthiness of borrowers. Interpretation: shocks to risk bearing capacity of the financial sector that change the supply of credit. Empirical Inference: exogenous movements in credit spreads after controlling for expected defaults.

3 Existing Empirical Work Structural vector autoregressive (VAR): Gilchrist & Zakrajsek (212), Meeks (212). Time-invariant VAR: constant coefficients & volatility of shocks. Time-invariant VAR only average relationships. Implication: macro effects of credit shocks are time invariant. Those effects are the same during: booms and recessions. financial crises and tranquility. different Fed chairmanships (e.g. Greenspan vs. Bernanke).

4 Research Questions Are the effects of credit shocks on the U.S. economy time varying? If so, are these time varying effects due to: stochastic volatility of shocks? time varying transmission mechanism? The paper examines these questions during

5 Motivation for Time Variation 1 Empirical nature of credit spreads. 2 Monetary regimes: Greenspan vs. Bernanke. 3 State-dependent amplification mechanism in theoretical DSGE models.

6 Excess Bond Premium Excess Bond Premium (EBP) of Gilchrist & Zakrajsek (212). EBP: credit spreads purged of expected defaults (and other borrower-specific characteristics like liquidity). EBP serves as a proxy for intermediary risk aversion. Adverse credit shocks: an exogenous increase in intermediary risk aversion which then reduces the supply of credit.

7 Empirical Motivation Annual % Excess Bond Premium Output 4 Annual %

8 Empirical Motivation Annual % Excess Bond Premium Output 4 Annual %

9 Monetary Regimes Macro effects of credit shocks depend on monetary policy response. Monetary response = monetary reaction function + monetary shocks. Has U.S. monetary reaction function changed from Greenspan s term to Bernanke s? Two possible aspects of time variation: 1 changing weights on output and inflation stabilization? Primiceri (25), Cogley & Sargent (25). 2 credit spreads (financial variables) in monetary reaction function? Curdio & Woodford (21), Gilchrist & Zakrajsek (212).

10 Theoretical Motivation Recent DSGE models with financial intermediaries: Brunnermeier & Sannikov (21), Adrian & Boyarchenko (213), He & Krishnamurthy (213). Amplification mechanism depends on states: shock size, intermediary capital position, leverage. States vary over time. State-dependent amplification time variation.

11 Summary Findings Stochastic volatility: Yes Time varying transmission mechanism: No

12 Econometric Approach Model: Time-varying VAR with stochastic volatility (Primiceri (25)). Drifts in VAR coefficients and identifying matrix time-varying transmission mechanism. Drifts in volatilities of shocks stochastic volatility. Database: 4 variables in VAR: Output, Inflation, Excess Bond Premium, Federal Funds Rate. Quarterly data during Estimation: Bayesian with Gibbs Sampling.

13 Identification Distinguish credit supply from credit demand: Excess Bond Premium (EBP): credit supply condition. Endogeneity of credit supply condition: Use recursive identification. Ordering: output, inflation, EBP, nominal interest rate. Output & inflation: standard ordering in monetary literature. Interest rate ordered last credit shocks affect: output and inflation with a lag. interest rate contemporaneously. Initial response of interest rate: direct response through the monetary reaction function to credit shocks over and above their indirect effects on output and inflation.

14 Stochastic volatility vs. Time varying transmission

15 Conditional Volatility of Credit Shocks Standard Deviation of Credit Shocks Dot com bust Stock market crash Subprime crisis Conditional volatility of credit shocks is systematically higher during periods of financial crisis in the U.S.

16 Conditional Volatility of Credit Shocks Standard Deviation of Credit Shocks Conditional volatility of credit shocks is higher during 21 and recessions.

17 Conditional Volatility of Credit Shocks Standard Deviation of Credit Shocks Time-invariant VAR fails to account for stochastic volatility of credit shocks. In particular, it understates the volatility of credit shocks during the Great Recession.

18 Time Series of Credit Shocks Realized Credit Shocks Adverse credit shocks lead 21 and recessions. 2 Credit shocks during not bigger than Series of adverse credit shocks during vs. one time shock in 21.

19 Time Varying Transmission Mechanism Drifts in VAR coefficients and identifying matrix time varying transmission mechanism. Compute impulse responses by fixing parameters estimated at each date. Normalized impulse responses: fix shock size and sign. EBP rises by.27% on impact shock size estimated at the start of the Great Recession.

20 Impulse Responses: Output Output.5 199: :2.5 2:4.5 26:4.5 27: No evidence of time varying transmission mechanism for output.

21 Impulse Responses: Inflation Inflation 199:2 1995:2 2:4 26:4 27: No evidence of time varying transmission mechanism for inflation. Adverse credit shocks no significant deflation.

22 Impulse Responses: EBP EBP 199:2 1995:2 2:4 26:4 27: No evidence of time varying transmission mechanism for EBP.

23 Impulse Responses: Interest Rate : :2 Interest Rate : : :3

24 Preemptive monetary easing

25 Impulse Responses: Interest Rate on Impact Interest Rate 2 Basis points Initial interest rate response is insignificant or positive. No evidence of preemptive easing through the monetary reaction function in response to adverse credit shocks.

26 Initial Monetary Response to the Subprime Crisis Fed cut rates by 5 basis points in Sep 27 well before the Recession started in Dec 27 preemptive cut. FOMC statement of Sep 27: rate cut is to...help forestall some of the adverse effects on the broader economy that might otherwise arise from the disruptions in financial markets Recall: no evidence of preemptive cuts through the monetary reaction function. Initial rate cut in response to the subprime crisis likely a monetary surprise.

27 Credit shocks and the Great Recession

28 Credit Shocks and the Great Recession How important are credit shocks for the Great Recession? Financial nature and references of credit crunch. Counterfactuals: shut down credit shocks. Time-varying VAR allows one to take into account the economic environment that prevailed around the Great Recession.

29 Credit Shocks and Output Collapse during the Great Recession Output Annual % :3 28:1 28:3 29:1 29: :3 28:1 28:3 29:1 29:3 Exogenous disruptions in credit supply is an important causal factor for the Great Recession. But, these disruptions do not entirely explain all of the depth and duration of the Great Recession.

30 Credit Supply Conditions during the Great Recession EBP during Great Recession Annual % :3 28:1 28:3 29:1 29: :3 28:1 28:3 29:1 29:3 Dynamics of credit supply condition during the Great Recession is not entirely exogenous. Shocks to other sectors of the economy outside financial markets contribute to the deteriorating credit supply.

31 Credit Supply Conditions during the 21 Recession EBP during 21 Recession Annual % :4 21:2 21:4 22:2 22:4 2:4 21:2 21:4 22:2 22:4 Dynamics of credit supply condition during 21 recession is largely exogenous.

32 Endogeneity of Credit Supply during the Great Recession Variance Decomposition of EBP This time is different.

33 Conclusion 1 Credit shocks exhibit stochastic volatility: conditional volatility of credit shocks is systematically higher during financial crises. 2 No evidence of time variation in the transmission mechanism of credit shocks. 3 Non-trivial role of credit shocks during the Great Recession: output growth would have been better off by as much as 4% in the absence of credit shocks. 4 However, credit shocks do not entirely explain the depth and duration of the Great Recession. 5 Unlike the 21 episode, shocks to other sectors of the economy outside financial markets contribute to the deterioration of credit supply during the Great Recession. 6 This time is different.

34 Econometric Model VAR model: y t = c t +B 1,t y t B p,t y t p +u t ; t = 1,2,...T Baseline Specification: y t = {Output, Inflation, EBP, Interest Rate} Lag length p = 2 Time-varying parameters: c t intercept terms allows interest rate in levels in VAR. B i,t, i = 1,...,p coefficients. u t reduced form errors with variance-covariance matrix Ω t.

35 Econometric Model The variance-covariance matrix Ω t is decomposed as: Ω t = A 1 t H t (A 1 t ) 1. A t = α 21,t α n1,t α n2,t α nn 1,t 1 }{{} Covariance states h 1,t. H t = h.. 2,t h n,t }{{} Variances

36 Econometric Model VAR in structural form: y t = X tb t +A t 1 H t 1 2 ǫt H 1 2 t ǫ t : size adjusted structural shocks (H t is diagonal). H t stochastic volatility. B t and A t time varying transmission mechanism. VAR parameters evolve as: B t = B t 1 +ν t α t = α t 1 +ζ t log(h i,t ) = log(h i,t 1 )+σ i η i,t ν t N(,Q) ζ t N(,S) η i,t N(,1), i = 1,...,n.

37 Bayesian Estimation: Overview Time-varying parameters: B t, α t, h i,t ; t and i = 1,...,n Time-invariant parameters: Q, S, σ i ; i = 1,...,n Gibbs sampling: Break up the parameter space into blocks. Draw each block conditional on the rest. Exploit state space representation. Example: draw B t conditional on {α t, h i,t, Q, S, σ i } Detail

38 Gibbs Sampling: Illustration Drawing B t : Return Condition on {α t, h i,t, Q, S, σ i }. State space representation where B t is unobserved state: y t = X tb t +A 1 t H 1 2 t ǫ t B t = B t 1 +ν t (Measurement) (State Transition) Linear and Gaussian system. Draw B t via the Kalman filter and Carter & Kohn (1994).

39 Theoretical Background External Finance Premium (EFP) credit spreads data (Graeve (28)). Puts credit spreads in a DSGE model (BGG) by linking it to EFP. Finance literature: expected defaults comprise a small component of credit spreads data. Residual: Excess Bond Premium (EBP) risk bearing capacity of intermediaries over and above expected defaults. Shocks to EBP (VAR) shocks to EFP (DSGE).

40 Effects of Credit Shocks from DSGE Model An adverse credit shock loans in equilibrium. Intermediated loans for production and consumption output. Economic slack marginal costs (wages and rental rates) inflation. in output and inflation monetary easing nominal interest rate. Current credit spreads augmented in Taylor-type reaction function direct monetary response to credit spreads.

41 Alternative Specifications: Robustness 1 Model A: stronger prior on time varying transmission mechanism evidence against time varying transmission mechanism is robust. 2 Model B: run on (HP) detrended interest rate robust to alternative specification of data trends. 3 Model C: EBP ordered last robust to alternative ordering. 4 Model D: include stock prices robust to asset price interactions.

42 Conditional Volatility of Credit Shocks: Alternative Specifications Standard Deviation of Credit Shocks Stochastic volatility of credit shocks robust to alternative specifications of the time-varying VAR.

43 Conditional Volatility of Structural Shocks in each Equation Output Inflation EBP.6 Interest Rate

44 Time Series of Structural Shocks in each Equation Output Inflation EBP Interest Rate

45 Time-varying vs. Time-invariant VAR Normalized Impulses of Output Time varying Time invariant Time-invariant VAR less precise inferences regarding the transmission of credit shocks for output.

46 Time-varying vs. Time-invariant VAR Shock Size Adjusted Impulses of Output Time varying Time invariant Time-invariant VAR understates the adverse effects of credit shocks on output around the Great Recession.

47 Impulse Responses: Interest Rate at Longer Horizons Interest Rate 6 Qtrs. Ahead 9 Qtrs. Ahead 12 Qtrs. Ahead All negative monetary easing. At these horizons, credit shocks also affect output and inflation indirect response. Upward trend over time, linked to downward trend of interest rate during sample period. But this aspect of time variation is not statistically significant.

48 Credit Shocks and Asset Price Interactions Output Counterfactual during Baseline Model D Actual 27:3 28:1 28:3 29:1 29:3 21:1 21:3 Including asset prices in VAR does not result in a bigger role of credit shocks during the Great Recession. Similar result when credit volume, leverage included in VAR.

49 Counterfactuals: Recession Inflation Annual % :3 28:1 28:3 29:1 29:3 27:3 28:1 28:3 29:1 29:3 Interest Rate 5 Annual % :3 28:1 28:3 29:1 29:3 27:3 28:1 28:3 29:1 29:3

50 Counterfactuals: 21 Recession Output 3 4 Annual % :4 21:2 21:4 22:2 22: :4 21:2 21:4 22:2 22:4 Inflation Annual % :4 21:2 21:4 22:2 22: :4 21:2 21:4 22:2 22:4

51 Interest Rate Counterfactual: 21 Recession Interest Rate Annual % :4 21:2 21:4 22:2 22:4.5 2:4 21:2 21:4 22:2 22:4 Recall: ouptut and inflation fluctuations almost disappear in the counterfactual. Decline in interest rate unrelated to business cycle fluctuations. Consistent with the narrative that interest rate during is too low for too long.

52 Effects of Credit Shocks: vs. 21 Recessions 2 Output.2 Inflation EBP Interest Rate

53 Impulse Response of Output: Alternative Specifications Output Baseline 199: : : : : Model A Model B Model C Model D

54 Impulse Response of Inflation: Alternative Specifications Inflation Baseline 199: : : : : Model A Model B Model C Model D

55 Impulse Response of Interest Rate: Alternative Specifications Interest Rate Baseline 199: : : : : Model A Model B Model C Model D

56 Variance Decomposition 6 Output 25 Inflation EBP 8 Interest Rate

57 Evolution of Intercept Term in Interest Rate Equation Intercept Term in Interest Rate Equation Evolution of intercept term (dotted) picks up downward trend of the nominal interest rate in data (solid).

58 Inefficiency Factors of VAR Parameters.2 VAR Coefficients B.2 Hyperparameter Q Covariance States ALPHA.5 Hyperparameter S Stochastic Volatilities H.15 SIGMA