Statistical Methods for Large Interacting Systems and Applications to Mortgage Pools

Stanford University, Stanford CA

Investigators

Abstract

This research will model and analyze systemic default events in the mortgage market. It is motivated by the mortgage meltdown of 2007, which was the central catalyst for the subsequent financial crisis. New dynamic stochastic models of default and prepayment in mortgage pools are designed to capture the influence of idiosyncratic, systematic, and contagion effects on event timing. Statistical estimation for such large systems typically has been intractable. The project will develop and rigorously analyze novel statistical methods for model parameter estimation that are computationally tractable for the large pools of mortgages common in practice. The likelihood methods are based on asymptotic approximations for large pools. They are broadly applicable to large interacting stochastic systems that appear in a wide range of fields, including finance, economics, engineering, and other sciences. These statistical methods will be used to perform an empirical study with the goal of identifying key factors responsible for the mortgage collapse, delivering new risk analysis tools for mortgage-backed securities, and providing insight to policy makers on how to prevent similar events in the future. A significant impediment to empirically studying large systems like the mortgage market is the lack of tractable methods for statistical inference. The project will develop and analyze computationally efficient statistical methods for this and related problems, which could be of great use to regulators, industry practitioners, and rating agencies. Moreover, the empirical portion of the project will yield new, significant insights regarding how systemic risk events develop and spread and how that risk can be managed.