Forum

Replies

• 

  Vinodh Kumar March 22, 2011 21:04

  Predictive modeling has an underlying assumption that history repeats. We model based on historical data to predict customer behavior. At the point of acquisition, we use bureau scores or similar risk scores available with agencies (such as FICO in US) to categorize customers into various risk segments. We could also integrate this with a score developed internally  (acquisition risk models) and cherry-pick our leads. The scores could vary over a period of time based on customers' performance. Again, our business goal could either be to increase our market base (sales) or revenue. While trying to improve our acquisition rate, we do reject inferencing as well.

   

  While the underlying statistical techniques to develop such models would remain same, the variation in the model performance is caused due to macroeconomic factors. And this is the prime reason as well for the deterioration of model performance over a period of time; in which case we either revisit and tweak the model or build a new one altogether.

   

  Risky customers could be profitable, however a sincere assessment to expected loss will enable banks for precise reserve allocation. The credit line / loan amount depends on the customers' ability to pay, hence a right strategy for line approval will ensure effective loss mitigation when it arises.
• 

  Tankiso Moloi March 1, 2011 15:38

  In South Africa, I understand that banks use the internal rating based models such as the clients probability of default (PD's), exposure at default (EaD) and the Loss Given Defalt (LGD). I will dwell on this tomorrow. I am currently working on my doctoral thesis in this subject.