Beruflich Dokumente
Kultur Dokumente
2014 American Economic Association Meetings Philadelphia, PA Levent Guntay, FDIC leguntay@fdic.gov Paul Kupiec, American Enterprise Institute paul.kupiec@aei.org
There are Big Risks in the continued use of some currently popular systemic risk measures
Should government require a mandatory warning label?
We Focus on Two Measures CoVaR and MES (aka SES & SRISK)
CoVaR
Conditional Value at Risk
The value at risk of a conditional stock return distribution Adrian and Brunnermeier, (2011) CoVaR, FRB of New York. Staff Report No. 348.
MES
The expected shortfall of a conditional stock return distribution Acharya, Engle, and Richardson, (2012). Capital Shortfall: A New Approach to Ranking and Regulating Systemic Risks, The American Economic Review 102, 59-64. Acharya, Pedersen, Philippon, and Richardson, (2010). Measuring Systemic Risk, Technical report, Department of Finance, NYU Stern School of Business.
Intuition: Systemic risk will generate tail dependence in financial institutions stock returns
Catastrophic losses in an important financial institution will induce
Large losses in other financial institutions
Defaults, counterparty losses, interconnectedness, contagion
Warning!
CoVaR and MES Confound systemic and systematic risk
Firms with large systematic risk components have large CoVaRs and MESs
How large must tail dependence estimates be before we can reject the null hypothesis of no tail dependence?
Need a proper statistical test
.01,
= 2.32635
1.645 1.645
= 2.32635
= 2.062839
Nonparameteric estimators
MES is estimated as the average stock return on days when the market return is in its 5 percent left-hand tail If null is true, nonparametric estimators are unbiased but not efficient If alternative is true, nonparametric estimators are still unbiased The nonparametric estimators can produce much larger negative CoVaR or MES estimates if there is tail dependence in returns
The 1% quantile of the CRSP equal-weight market portfolio conditional on stock js return equal to its 1 percent quantile
Nonparametric
.375
.2567
Nonparametric
Test Statistics
Our tests evaluate the difference between two estimators Nonparametric estimate-Parametric estimate
CoVaR MES
Quantile regression CoVaR estimate-Gaussian CoVaR estimate Selected sample average MES estimate-Gaussian estimate
Both estimators are unbiased under the null If Null is true, parametric is most efficient estimator
The differencing controls for systematic risk We scale these differences to remove idiosyncratic risk dependence
CoVaR difference is scaled by Gaussian CoVaR estimate MES is scaled by estimate of stock idiosyncratic standard deviation
Correlation remains as a nuisance parameter We calculate critical values for these test statistics using Monte Carlo simulations
Sample size =500 obs .about 2 years of daily data 25,000 Monte Carlo replications
Results
Lots of firm returns reject the null
Many more rejections are nonfinancial than financial
MES and CoVaR often disagree about which firms are potentially systemic
MES test rejects the null much more frequently than CoVaR test
Rejection region
Insurance Industry
Rejection region
Retail Trade
Rejection region
Manufacturing
Rejection region
Can test idea can be extended to systemic risk measures based on CDS-spreads?