Post by Sapphire Capital on Dec 13, 2008 8:36:23 GMT 4
The Failure of Models that Predict Failure: Distance, Incentives and Defaults
Uday Rajan
University of Michigan at Ann Arbor - Stephen M. Ross School of Business
Amit Seru
University of Chicago - Graduate School of Business
Vikrant Vig
London Business School
October 1, 2008
Chicago GSB Research Paper No. 08-19
Abstract:
Why did mortgage default models so severely underestimate defaults in the subprime sector in the 2002-07 period? We analyze data on securitized subprime loans issued in the period 1997-2006 and demonstrate that interest rates on new loans rely increasingly on hard information characteristics -- interest rates become increasingly sensitive to a borrower's FICO score and loan-to-value (LTV) ratio and the distribution of interest rates shrinks over time. A statistical default model fitted in a low securitization period breaks down in the high securitization period in a systematic manner: it underpredicts defaults for borrowers with low documentation, low FICO scores and high LTV ratios. We rationalize these findings in a theoretical model that highlights a reduction in lenders' incentives to collect soft information when securitization becomes common. As a result, among borrowers with similar hard information characteristics, the set that receives loans worsens in a fundamental way with securitization, leading to a systematic failure of default models based on past data. Our results suggest that, although generally ignored, incentive effects leading to a change in the underlying regime are an important dimension of model risk, and caution against regulation with models that ignore the incentives of market participants.
papers.ssrn.com/sol3/Delivery.cfm/SSRN_ID1304724_code753937.pdf?abstractid=1296982&mirid=2
Uday Rajan
University of Michigan at Ann Arbor - Stephen M. Ross School of Business
Amit Seru
University of Chicago - Graduate School of Business
Vikrant Vig
London Business School
October 1, 2008
Chicago GSB Research Paper No. 08-19
Abstract:
Why did mortgage default models so severely underestimate defaults in the subprime sector in the 2002-07 period? We analyze data on securitized subprime loans issued in the period 1997-2006 and demonstrate that interest rates on new loans rely increasingly on hard information characteristics -- interest rates become increasingly sensitive to a borrower's FICO score and loan-to-value (LTV) ratio and the distribution of interest rates shrinks over time. A statistical default model fitted in a low securitization period breaks down in the high securitization period in a systematic manner: it underpredicts defaults for borrowers with low documentation, low FICO scores and high LTV ratios. We rationalize these findings in a theoretical model that highlights a reduction in lenders' incentives to collect soft information when securitization becomes common. As a result, among borrowers with similar hard information characteristics, the set that receives loans worsens in a fundamental way with securitization, leading to a systematic failure of default models based on past data. Our results suggest that, although generally ignored, incentive effects leading to a change in the underlying regime are an important dimension of model risk, and caution against regulation with models that ignore the incentives of market participants.
papers.ssrn.com/sol3/Delivery.cfm/SSRN_ID1304724_code753937.pdf?abstractid=1296982&mirid=2