I am looking for a way to model the behavior of customers of non-maturing deposits. Basically I want to hedge against the fact that customers of a bank can withdraw their money any time due to personal reasons i.e. immediate need of cash, purchase of a house etc. As you all know when someone deposits money that person gets a floating interest rate. The bank then takes that money and invests it in bonds etc. at the market rate. Obviously the market rate > floating rate that the bank provides the customer, hence the bank makes a profit. There are two main risks in this scenario, namely liquidity risk and interest rate risk. I want to focus on the liquidity risk and not the interest rate risk. I have found many methods that are interest rate risk focused (i.e. replicating portfolio of Maes and Timmermans, Bardenhewer etc.) but I have yet to find a model that is strictly focusing on the liquidity risk aspect. Any suggestions?

Try a compound Poisson -- the chance of a deposit/withdrawal is determined by a Poisson process. Suppose D(t) is the amount on deposit. Once the Poisson counter is triggered,draw positive increments from some distribution on (0,infinity) and negative increments -f D(t-), f a random fraction with support in (0,1]. In other words, the increment in D(t) at time t is drawn from some distribution Q on [-D(t-),infinity]. Your database will tell what Q was historically. Likely a Dirac mass at f = 1 (100% withdrawal), plus a smooth looking density otherwise.

Outrun posted something related on the technical forum. I'll post the link to the thread later. herehttp://www.wilmott.com/messageview.cfm?catid=4&threadid=86408

Thanks Alan, I will look into it. Thanks frenchX, I guess it would be better to continue the conversation over there.