I have a one year PD for a set of mortgages and I’m trying to determine if there is a best practice to extend the formula out to 360 months (30 year terms) without having empirical data. It doesn’t seem correct to just divide the formula by 12 and apply that to each monthly balance for 360 months. Thanks.
AFAIK you really need some type of basis of a timing curve for extrapolation of the current curve. Any methodology for extrapolation without a timing curve would not account for how the curve is loaded (flat, front-ended…etc). Perhaps you could take a similar cohort, compare that timing curve to your current data, and extrapolate. For example, if your comparison cohort has a lifetime pd of 10% but 5% occur in the first year (50%) and your data has 6%, you could extrapolate that 10% increase over the lifetime of the new curve. I believe a large portion of mortgage defaults occur in the first 5 years, especially since the WAL of mortgages is (or was) somewhere ~10yrs due to refis and payoff/purchases of new homes. Thus, simple extrapolation is incorrect.
spierce Wrote: ------------------------------------------------------- > AFAIK you really need some type of basis of a > timing curve for extrapolation of the current > curve. > > Any methodology for extrapolation without a timing > curve would not account for how the curve is > loaded (flat, front-ended…etc). > > Perhaps you could take a similar cohort, compare > that timing curve to your current data, and > extrapolate. > > For example, if your comparison cohort has a > lifetime pd of 10% but 5% occur in the first year > (50%) and your data has 6%, you could extrapolate > that 10% increase over the lifetime of the new > curve. > > I believe a large portion of mortgage defaults > occur in the first 5 years, especially since the > WAL of mortgages is (or was) somewhere ~10yrs due > to refis and payoff/purchases of new homes. Thus, > simple extrapolation is incorrect. Thanks! I can use this.