wilmott.com

Hi All, I am trying to estimate empirical durations and partial durations (or sens with respect to a single swap rate) for the MBS securities. Are there are any standard approaches to do that? E.g. how to solve the multicollinearity problem accross the swap rates (e.g PCA), what time window is standard to use (e.g. 3M), etc? Is there is any literature on this?Thanks, Michael

Hi Michael,Hayre & Chang, "Effective and empirical durations of mortgage securities", Journal of Fixed Income, March 1997, and Goodman & Ho, "Mortgage hedge ratios: Which ones works best", Journal of Fixed Income, December 1997 can be used as a starting point. Goodman & Ho indeed propose to use 3 months of data in order for the regression to reflect recent price development. A standard approach would be to base the regression analysis on a single treasury bond or swap rate, i.e. the bond/swap commonly used in the market for interest rate hedging. Typically, this would be the 10-year swap rate (or 10-year benchmark treasury bond). In case the same calculations are to be used on a larger array of mortgage bonds with various coupons, additional swaps/bonds may be included to reflect that bonds prone to high prepayments are sensitive to shorter rates and not 10-year rates. In Europe, this would correspond to Schatz, Bobl and Bund. The empirical duration (corresponding to a parallel shift) would then simply be the sum of regression coefficients.Collinearity might pose a problem, but I would probably be more concerned by the fact that 1) negative convexity is not adressed in a standard regression analysis, and 2) empirical durations are often based on extrapolating beyond the rates on which the regression was based, i.e. there is no rate experience below (above) the current rate.