I constructed a model to forecast the prepayment rates for a mortgage loan portfolio (of mortgages in an emerging market) using probit regression on factors such as loan-to-value, PTI, time from settlement, and several characteristics of the borrower.

However when testing with historical data the forecast is not so accurate. What is the best model to make accurate forecasts of prepayment rates for mortgage loans?

  • 2
    $\begingroup$ Can you be more explicit about what you mean by "the forecast is not so accurate"? How many periods are you forecasting? How are you measuring accuracy? Are you modeling cohort-level data or loan-level? $\endgroup$ Feb 2 '12 at 22:24
  • $\begingroup$ What's the purpose of the model? Why build rather than use BlackRock or Yieldbook? $\endgroup$
    – MathAttack
    Feb 3 '12 at 3:30
  • $\begingroup$ @JoshuaUlrich, I am modelling loan level data. the model coluld predict 19% of past prepayments only. $\endgroup$
    – Pasha
    Feb 3 '12 at 8:14

Pre-payment rates are difficult to forecast because of path dependency. The historical interest rate path - not just current market conditions and borrower characteristics - matters because borrowers may have exercised their right to call the mortgage bond and re-finance if rates had previously been at lower levels than the current rate. None on the variables you have identified account for these interest rate paths. For example, if rates are at 4% and have generally been rising there is considerably less prepayment risk then if rates are at 4% and have been declining. In a probit model, you could try adding a variable such as the distance of the 30-yr fixed mortgage rate from the lowest mortgage rate over the last X periods. A more typical and sophisticated approach is to model pre-payments and MBS valuations by monte carlo simulations of many interest rate paths.

A simpler way to go about this would be to fit the terms to a PSA model. There are far fewer parameters here. You could also use option-adjusted spread analysis to determine borrower's optimal refinancing/pre-payment decision. Finally, along the lines of the monte carlo approach indicated above you could apply the two-factor Hull White model to simulate various interest rate paths.

  • $\begingroup$ I do agree what you said about interest rates. But the country where I am trying to forecast prepayment rate has very stable interest rates for 20 years. So in my case interest rate does not have just big impact because simply they don't change. Should Any other factors be considered? $\endgroup$
    – Pasha
    Feb 3 '12 at 8:11
  • $\begingroup$ Really? I find that hard to believe. Developed markets are generally integrated. Also mortgage product markets are cyclical as well. What country market are you referring to? $\endgroup$ Feb 3 '12 at 16:24
  • $\begingroup$ Thank you for your answer and comments. Absolutely, agree with you. it sounds wierd, but it is the reality. the country is not so well known and developed country. Take account its cyclicality, can the economic growth be considered as a factor or may be it will be more appropriate to use other approaches besides logit regressions? $\endgroup$
    – Pasha
    Feb 3 '12 at 21:11
  • $\begingroup$ Updated my answer in response to your comment $\endgroup$ Feb 4 '12 at 18:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.