## Reject Inference in Credit Scoring : Building a robust Credit Scorecard

For statistical model building the key assumption is that the sample used to develop the model is indicative of the overall population. In particular, the sample should be similar to the population on which it will be applied. This is true for behavioural modeling, but this assumption does not hold true for application scorecards. For any application score card the known good bad population (KGB) is only the population which was approved in earlier cases. For the rejected population we do not have the information about the performance hence we can not use the same for modeling in normal cases. Without the build sample being similar to the target application population, the chances of model performing good and reasonable reduces to a great extent as it introduces sampling bias.

## Model Validation Measures

Validation along with disclosures of the same for risk management models in banks is one the important parameters for success of BASEL accord. While there are no clear guidelines around validation measures is provided by the Basel committee, there is given a clear framework which should be adhering to by the bank or financial institute. It says that there should be independent validation should be done on timely basis and it should be well documented and declared. Given this it is of paramount improtance for banks to have clearly defined validation measures and sufficient disclosures of the same.