Hi All,

I am trying to forecast 12 month default rate for a portfolio of accounts. To minimize seasonality or yearly effect, I take several snapshots of the portfolio and generate 12 month's default (my forecast target) for each snapshot. I heard someone mentioned that I need to use proc phreg instead of proc logistic to construct the model. It seems that proc logistic will produce bias and proc phreg will not. Can somebody tell me why?

To illustrate the problem, let's say that I am trying to predict default curve for 2 months in following portfolio: account_# month statement x(my covariate) default (1- default) 1 Jan 08 1 a 0 1 Feb 08 2 b 0 1 Mar 08 3 c 0 1 Apr 08 4 d 0 1 May 08 5 e 1 2 Apr 08 1 f 0 2 May 08 2 g 0 2 Jun 08 3 h 0 2 Jul 08 4 i 0 2 Aug 08 5 j 0 2 Sep 08 6 k 0 2 Oct 08 7 l 0 And I take snapshots at Jan 08, Apr 08 and Jul08, so I end up with the following data(multi-snapshots): account_number x month default 1 a 1 0 1 a 2 0 1 d 1 1 2 f 1 0 2 i 1 0

I wanted to build the model as: proc logistic data=multi-snapshots; model default = x t; run; / *discrete survival model*/ But I was told this is not right. Could somebody let me know why? And how I should solve this problem?

Thanks a lot.