Financial Stress and Equilibrium Dynamics in Term Interbank Funding Markets

Transcription

1 Financial Stress and Equilibrium Dynamics in Term Interbank Funding Markets Emre Yoldas a Zeynep Senyuz a a Federal Reserve Board June 17, 2017 North American Summer Meeting of the Econometric Society

2 Disclaimer Motivation and Background The views expressed in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System, its staff or any other person associated with the Federal Reserve System.

3 Interbank Markets Crucial for functioning of the financial system and transmission of monetary policy (e.g. Ivashina et al., 2015) Highly liquid and generally stable until the global financial crisis First signs of stress emerged in August 2007 Stress increased in 2008 and peaked in the aftermath of Lehman Bankruptcy Policy intervention and balance sheet adjustments by banks gradually reduced the stress

4 Related Theoretical Literature Freixas and Jorge (2008): asymmetric information distinct regimes with respect to aggregate liquidity Acharya and Skeie (2011): liquidity hoarding decline in interbank lending and a rise in risk premiums. Heider et al. (2015): endogenous liquidity and counterparty risk that stems from asymmetric information three distinct equilibrium outcomes possible

5 Related Empirical Literature Measuring financial stress: Carlson et al. (2011), Hakkio and Keeton (2009), Oet et al. (2011), Carpenter et al. (2014) Money market spreads as indicators of credit or liquidity risk: Taylor and Williams (2009) and McAndrews et al. (2015) Nonlinear term-structure dynamics: Anderson (1997) and Seo (2003); nonlinear price discovery: Dwyer et al. (1996), Martens et al. (1998), and Theissen (2012)

6 Libor Motivation and Background London Interbank Offered rate (Libor): unsecured funding rates reported by large banks Components of Libor (McAndrews et al., 2015): Libor = Avg. Exp. Overnight Rate + Term Premium + Credit Risk Premium + Funding Liquidity Risk Premium

7 OIS Motivation and Background Overnight index swap (OIS): an interest rate swap where the underlying is the overnight policy rate (effective federal funds rate in the United States) Components of OIS: OIS = Avg. Exp. Overnight Rate + Term Premium

8 Libor-OIS Spread Motivation and Background Libor-OIS is a composite risk premium reflecting both credit and funding liquidity risk Closely-watched indicator of the health of the banking system In short-term funding markets, spreads between term Libor and overnight index swap (OIS) rates soared to record highs,... (Greenbook, October 2008). Libor-OIS remains a barometer of fears of bank insolvency (Alan Greenspan, Feb. 2009) Relative importance of credit and liquidity risk likely changes with market conditions (King and Lewis, 2015).

9 Long-run Relationship of Interest Rates and Spreads Over a relatively long time period, average spreads represent long-run equilibrium risk premia Market frictions or shocks may lead to transitory divergence of rates resulting in potentially large fluctuations in spreads Spreads are mean-reverting due to dynamic repricing of risk and policy intervention during times of stress

10 Linear VEC Model VEC model where the spread serves as an error correction term due to arbitrage and policy intervention. Let y t denote the 2 1 vector of interest rates underlying a given spread. The VEC(p) model is y t = ΨX t + ɛ t, where Ψ = (c, φ, A 1,..., A p ), X t = (1, s t 1, y t 1,..., y t p ), s t = y 2,t y 1,t, and ɛ t is martingale difference vector process with time-varying covariance matrix.

11 Limitations of the Linear VEC Model In practice, the speed of adjustment is more likely to be a function of the magnitude of the spread Changes in asset quality and leverage, policy intervention Distinct states with different dynamics and adjustment toward the long-run equilibrium

12 Threshold VEC Model The n-state threshold VEC (TVEC) model is given by y t = n Ψ j X t 1(γ j 1 < s t 1 γ j ) + ɛ t, j=1 where Ψ j = (c j, φ j, A j 1,..., Aj p), {γ j } n j=0 are the thresholds such that γ 0 = and γ s =, and 1( ) is the standard indicator function. Conditional sequential least squares estimation: Example: For n = 2, conduct grid search over trimmed version of S = {s 1,..., s T 1 } for γ 1 to minimize system SSR

13 Inference in the TVEC Model The asymptotic distribution of the threshold estimate is not nuisance parameter free use subsampling (Politis et al., 1999) to construct asymptotically valid confidence intervals. The threshold estimate converges at rate T, so we treat the threshold as known to conduct inference on {Ψ j } n j=1, which converge at T.

14 Data Motivation and Background Weekly data (avg. as of Wed.) Sample period: January 1, 2002 to March 22, 2017 Tenors: 1-month, 3-month, 6-month Source: Bloomberg

15 Unit-root and Cointegration Tests ERS Test NP Test Panel a: Interest Rates 1-month Libor month Libor month Libor month OIS month OIS month OIS Panel b: Spreads 1-month Libor-OIS month Libor-OIS month Libor-OIS Critical Values 1% % % Note: ERS: Elliott et al. (1996), NP: Ng and Perron (2001).

16 Threshold Nonlinearity Tests Recursive residual based test of Tsay (1998) H 0 : no threshold effect p-values: T 0 1-month 3-month 6-month Note: T 0 denotes size of the initiation sample for computation of recursive residuals.

17 Number of Regimes Tsay s test is not informative on the number of regimes graphical methods and information criteria (Guidolin and Timmermann, 2006) 1-month 3-month 6-month Linear VEC regime TVEC regime TVEC Note: Hannan-Quinn information criterion for the first order VEC and TVEC models.

18 Point and Interval Estimates for Thresholds CI L (γ 1 ) γ 1 CI U (γ 1 ) CI L (γ 2 ) γ 2 CI U (γ 2 ) 1-month month month Note: Point estimates and 90% subsampling confidence intervals expressed in basis points.

19 Economic Interpretation of Regimes A combination of asymmetric information (adverse selection) and endogeneous liquidity in the banking system allows multiple equlibria (Heider et al., 2015): full-participation equilibrium: normal times adverse selection equilibrium with higher spreads: moderate stress liquidity-hoarding (market-breakdown) equilibrium: high-stress

20 Regime Classification for 1-month Libor-OIS Regime 1 Regime 2 Regime Note: Regime classification is based on point estimates.

21 Regime Classification for 6-month Libor-OIS Regime 1 Regime 2 Regime Note: Regime classification is based on point estimates.

22 Regime Classification for 3-month Libor-OIS Regime 1 Regime 2 Regime Note: Regime classification is based on point estimates.

23 TVEC Model Parameters for 3-month Libor-OIS Parameter Regime 1 Regime 2 Regime 3 c (0.01) (0.81) (0.00) c (0.00) (0.55) (0.03) a (0.00) (0.00) (0.11) a (0.40) (0.32) (0.31) a (0.01) (0.00) (0.32) a (0.70) (0.00) (0.00) φ (0.03) (0.82) (0.00) φ (0.01) (0.63) (0.00) Note: 1 indicates OIS; 2 indicates Libor.

24 Response Functions for 3-month Libor-OIS Panel a: Regime 1 OIS Libor Spread Panel b: Regime Panel c: Regime spread

25 Structural Change and Misreporting of Libor Changes in risk assessment, banks funding models, and regulation (Basel III, Dodd-Frank) start the sample in August 2007 Libor misrepresentation use 3x6 FRA-OIS spread, or 2-year Interest Rate Swap (IRS)-OIS spread (Filipovic and Trolle, 2013) Under the risk neutral measure, FRA-OIS (IRS-OIS) is the expected Libor-OIS over the corresponding term

26 Point and Interval Estimates for Thresholds CI L (γ 1 ) γ 1 CI U (γ 1 ) CI L (γ 2 ) γ 2 CI U (γ 2 ) Libor-OIS FRA-OIS IRS-OIS Note: Point estimates and 90% subsampling confidence intervals expressed in basis points.

27 Regime Classification for IRS-OIS Regime 1 Regime 2 Regime

28 TVEC Model for IRS-OIS Parameter Regime 1 Regime 2 Regime 3 c (0.03) (0.01) (0.00) c (0.03) (0.00) (0.00) a (0.67) (0.13) (0.00) a (0.75) (0.01) (0.05) a (0.83) (0.00) (0.00) a (0.38) (0.00) (0.00) φ (0.02) (0.01) (0.00) φ (0.02) (0.00) (0.00) Note: 1 indicates OIS; 2 indicates Libor.

29 Response Functions for IRS-OIS Panel a: Regime 1 OIS Libor Spread Panel b: Regime Panel c: Regime spread

30 Formal evidence for three distinct regimes confirming anecdotal evidence and the theoretical model of Heider et al. (2015) Point and interval estimates for spread thresholds potential benchmarks for policy makers and market participants Strong asymmetry in the equilibrium adjustment mechanism: economically and statistically meaningful adjustments take place in regimes associated with high levels of risk premia Qualitatively similar results with restricted sample and market-based rates