Overview

Count Models look at the deposits accounts as a group, and then model the number of closings that occur, as opposed to looking at individual accounts.

Fitting to Data

• Fitting the Poission Distribution
• Poisson Regression

Complications

While theorectically sound, there are complications when using count models to model account closures. In particular, the model tells you how many acocunts closed but not which ones. Given the differing balances across accounts, this information is often critical to modeling risk.

One way to overcome this issue is to assume that each account is equally likely to be one of the closures. That is, once you know there are {% n %} closures, you can randomly pick {% n %} accounts from the currently open accounts to be closed accounts.

As an alternative, the Survivor Based closure model does not face this limitation.