Securitization, Screening, and Failure of Default Models to Predict Default
Myron Scholes Global Markets Forum
November 4, 2008
As securitized mortgage lending expanded in recent years, many people made a crucial error that led them to underestimate the risk of these assets, Amit Seru explained. The spread of securitization led to changed incentives in the mortgage industry, which affected the interaction between loan originators and the investors who bought the loans. Because these shifting incentives hinged on what each party knew about borrowers’ riskiness, it altered the relationship between a loan bundle’s outward characteristics and its underlying risks. The changed relationship between available information and risk meant, in turn, that it was no longer possible to forecast default ratios accurately by relying only on the outward characteristics and historical data.
Seru explained the research that he and his coauthors used to uncover this large flaw in historical default models. His research also shows that rating agencies did not appear to account for this problem when evaluating mortgage-backed securities. Furthermore, investors seemed unaware of the shortcomings of those flawed ratings.
Because much of the information that banks could use to screen potential borrowers is difficult for outsiders to see, Seru and his coauthors exploited some rules of thumb in the mortgage lending market involving cutoff points in borrowers’ credit scores. Loans to borrowers with scores above 620, for example, could more easily be repackaged and sold as securities to investors because that was an industry norm. Loans to borrowers with scores below the cutoff, however, were much harder to securitize, so banks had to proceed as though they were much more likely to hold those loans themselves—and bear the risks—if they lent to those borrowers.
Looking at the pattern of lending to borrowers on either side of this cutoff, Seru found that default risks were substantially higher for borrowers just above the threshold. This suggests that these loans were riskier for reasons that could not be detected solely from the credit scores. Compared with the investors who bought these loan portfolios, the banks originating the loans were in a much better position to collect “soft” information on these borrowers as a way to supplement the hard credit scores and better gauge their riskiness. However, the spread of securitization gave banks incentives to shirk on this dimension and concentrate on issuing loans that would meet the observable cutoff. Once Seru analyzed the data in a way that could detect this information-related problem, a clear pattern emerged, showing that the default risks of securitized loans were being systematically underestimated based on historical data. Importantly, this problem was much less pronounced in areas where loan originators faced competition.