Jeffrey Russell, Professor of Econometrics and Statistics
Summary: This research proposes a new approach to pricing MBS. First, we propose using micro level, loan specific information such as FICO score and loan‐to‐value, now available through data sources such as Loan Performance or Black Box. Second, we propose modeling the state of a given loan in a security using a Markov Chain where at each point in time, each mortgage is allowed to take one of several states. A simple model would include the states “making payments”, “default”, and “prepayment”. When the loan is in the making payment state, payments are received at their scheduled times. A loan in good standing, however, will have probabilities of transitioning into default or prepayment. Given the time series of probabilities for each mortgage in a pool over the life of the mortgage, distributions and expectations of payouts for any MBS could be constructed. Several models of the dynamics of the transition probabilities are considered. A factorbased version of the dynamics can account for the mass defaults observed during the sub‐prime crisis. Candidates for this time varying factor include measures of housing prices and income, interest rates, and unemployment rates. Combined, the microstructure allows for individual security contracts to have different prices in different mortgage pools while the factors allow for market wide swings in probabilities and the mass defaults observed during the subprime crisis.