Visible and Hidden Risk Factors for Banks

1
Visible and Hidden Risk Factors for Banks

• Til Schuermann, Kevin J. Stiroh
• Research, Federal Reserve Bank of New York
• FDIC-JFSR Bank Research Conference
• Arlington, VA 13-15 September, 2006

Any views expressed represent those of the
authors only and not necessarily those of the
Federal Reserve Bank of New York or the Federal
Reserve System.
2
Banks and Systemic Risk

• Are banks closely tied to the observable risk
  factors?
• Are those residuals highly correlated?
• Are banks more similar to each other than other
  sectors?
• If yes, banks susceptible to systemic risk
• DeBandt and Hartmann (2002) 2 channels
• Narrow contagion
• Broad simultaneous shock
• Rajan (2005) compensation-induced herding

3
Overview

• Estimate a range of standard market models and
  compare
• Explanatory power
• Residual correlations
• Factor loadings
• Principal component analysis (PCA) of residuals
• Explanatory power of 1st PC
• Diffusion of hidden factors
• Homogeneity of PC loadings
• To provide context
• Large vs. small banks
• Large banks vs. large firms in other sectors

4
Market Models
5
Data

• Weekly bank equity returns, 1997 2005,
  year-by-year
• On avg. 488 banks/year
• CRSP
• Conditioning variables from various data sources
• Define large as inclusion in SP 500
• About 34 large banks per year
• About 454 small banks per year

6
Comparing Market Models

• Need a way to compactly analyze ? 16,340
  regressions (about 454?9?4 bank/year/model
  estimates)
• Data is a panel, so one may think of each year as
  a random coefficient model (Swamy 1970)
• Use mean group estimator (MGE) interpretation due
  to Pesaran and Smith (1995)
• Firms may on average have b 1, but with
  variation around that mean (sb)
• Use cross-sectional distribution of estimated
  parameters to make inference on betas in a
  given year t

7
Comparing Market Models Results

• Market factor dominates, followed by Fama-French
  factors
• Rise in explanatory power from 1999-2002, but no
  obvious trend
• Bank factors have relatively little impact
• Change from empirical literature in the 1980s
  (Flannery James 1984)
• Risk management / hedging
• Other factors show considerable heterogeneity
• Reflects differences in banks strategies and
  exposures

8
Comparing Market Models Results
9
Adjusted R2 large banks
10
Adjusted R2 other banks
11
Relative to Large Banks, Small Banks Show

• Lower correlated returns
• Mean pair-wise correlation of 11 vs. 57 (large)
• Smaller link to systematic risk factors
• Lower adj. R2 of 13 vs. 46
• Stronger evidence of conditional independence
• Mean pair-wise correlation of residuals of 3
  vs. 25
• Less systematic market risk
• ?m of 0.5 vs. 1.2
• Tighter link to interest rate and credit spread
  factors
• Less intensive users of interest rate/credit
  derivatives
• Stronger loadings on Fama-French factors

12
Average correlation of returns/residuals
Large Banks
Small Banks
13
Finding those Hidden Factors

• Considerable residual variation remains for large
  banks
• Mean pair-wise correlation of residuals ? 25
• Are hidden factors important?
• Remaining variation is diffuse with 1st PC
  accounting for only ? 27 of residual variance
• But, ? 93 of loadings on 1st PC have the same
  sign
• Systemic implication
• Given a shock to hidden factor, virtually all
  (big) banks will move the same way
• Recent interest in credit risk
• Frailty models of Das, Duffie, Kapadia Saita
  (2006)

14
Are Banks Different?

• Compare large banks to other large firms
• 10 other sectors comprised of SP 500 firms
• Return correlation is highest
• 57 vs. 36 (sector median)
• Returns are relatively easy to explain
• adj. R2, Nine-Factor model 46 vs. 28
• Residuals are typically diffuse
• 1st PC 27 vs. 21
• Residuals are relatively homogeneous and
  correlated
• Factor loading on 1st PC 93 vs. 84
• Mean pair-wise correlation of resids 24 vs. 12

15
Average Adj. R2 across Sectors, 1997-2005
16
Conclusions

• Positive no special risk factor for banks
• Returns can be modeled conventionally
• Residuals typically diffuse
• Negative residuals are relatively correlated and
  homogeneous
• Broad systemic concern?

17
Thank You! http//nyfedeconomists.org/schuermann/