Credit migration models often assume that a withrawn rating would have followed the model if it hadn't withdrawn. This assumption makes no sense to me - if it's your company, are you more likely to withdraw if you expect an upgrade or a downgrade? (I would say a downgrade, obviously). What am I missing?
Not all withdrawals have to do with immenent downgrades; it just may be too expensive for a small company to maintain ratings with AM Best, Moodys, S&P, and Fitch simultaneously for example. A company may move from S&P to AM Best, that would be an example of a "withdrawn" from S&P. The idea is that absent any other information, the model holds. Think of it as pure exposure rating for withdrawns - on average, any withdrawn follows the general model - not that every withdrawn follows it.
There is an interesting policy document from Moody's here that explains when they might withdraw a rating. Lots of technical reasons in addition to the more obvious one of business problems.