Baifan Yu
Senior Risk Model Analyst
CORE PROJECTS
Loss Timing Curve
The use of loss timing curves is a common practice in financial modeling and the methodology helps identify two important characteristics associated with loan loss: severity and timing. The severity is the final cumulative loss percent per vintage. This is the amount of the original balance of a particular vintage is assumed to default or be uncollectable. The timing is how much loss has been taken at a certain point in time.
For a given vintage, we can draw the historical cumulative default curve representing the cumulative amount of defaulted loans over time divided by the aggregate original outstanding amount of the loans included in the vintage. For those vintages that have been recently originated and, therefore, do not offer extensive historical data, we extrapolate default rates following the historical pattern observed on older vintages.
Later on, I successfully build a dual matrix approach which is used to get a net loss timing curve by combining Gross timing curve and Recovery curve.
Roll Rate Model
The roll-rate methodology predicts losses based on delinquency. It is also known as migration analysis or flow model which is derived from Markov Train theory. Roll and flow models are the most accurate short-term forecast technique. Dollars outstanding are stratifies by delinquency status, typically current, 1-29 days past due, 30-59 days past due, 60-89 days past due, and so on through charge-off.
The table below describes the mechanics of using roll rate analysis to track the migration of balances over a 120-day charge-off period.
LPU
Main
Drivers
As we know, Loss = Loss Severity * PD. And loss per unit is a main indicator of loss severity. If we understand what really drives our recent LPU, we would be able to better monitor our losses and make policy changes.
A stepwise regression model was constructed to identify the best indicators of LPU and LS. Finally LTV and other three components become the most important factors. And we could further monitor our LPU and LS by looking at these variables; also make policy changes based on those predictors.