In their model parameter estimates, the two most important covariates affecting the foreclosure rate were negative equity and whether or not the loan was subprime. The subprime variable had a huge effect on the foreclosure rate, and you have to remember that they estimated these model parameters using data from 1989-2007. How many subprime foreclosures are in that data? Not as many as there will be over the next few years, that's for damn sure--this credit cycle downturn is still too young. So even this large estimate of the effect of being subprime is probably too low.
Also, they determine whether or not a loan was subprime by looking at the lender's name. So as far as I can tell, they make no attempt to model alt-A mortgages.
This is one of the footnotes from the paper, and the table that it refers to (emphasis added).
The proportional hazard assumption can be seen in the last row of the bottom panel in Table 4, which displays the effect on the default hazard due to the combination of a subprime purchase mortgage and a one-standard-deviation fall in negative housing equity. The change in the default hazard is dramatic, increasing from approximately 0.05 percent to almost 1.5 percent. The intuition for this huge effect comes from the proportional hazard assumption. The fall in equity increases the default hazard by approximately 220 percent, from 0.05 percent to 0.16 percent. The difference between a subprime purchase mortgage and a prime purchase mortgage then increases the default hazard by 813 percent, from 0.16 to 1.5 percent.
Default Sale (+/-) std. dev. % change hazard % change hazard equity (−) 0.38 222 -15.1 negative equity indicator . 8.5 -32.9 libor (6-month) (+) 1.85 9.2 -14.5 unemployment rate (+) 2.06 10.3 -13.6 % minority (2000 zip-code) (+) 19.58 16.6 6.7 median income (2000 zip-code) (−) $24,493 48 -4 multi-family indicator . 58 4.3 condo indicator . 43.4 66.3 subprime purchase indicator . 813 50.4
HT: Calculated Risk