Advances and Applications in Statistics
Volume 20, Issue 1, Pages 25 - 41
(January 2011)
|
|
FOCUSING ON THE LOWER SCORING DATA IN ORDER TO IMPROVE CREDIT SCORING MODEL SELECTION
Arthur L. Dryver
|
Abstract: Under certain circumstances, such as with various credit scoring models, the model�s performance in the tails is especially important. Standard two-sample tests statistics can be improved when the concerned difference is on the tails. A new methodology is developed, which splits the combined samples in half using the overall sample median of the as the cut point in order to place more focus on the tails. This new methodology can be used to aid in model selection for credit scoring models. Simulation results of the new methodology yield higher power when the population mean and standard deviation are the same for the two populations. Thus another application is to test if standardized scores have the same distribution. |
Keywords and phrases: Anderson-Darling test statistic, credit scoring, goodness-of-fit tests, Kolmogorov-Smirnov test statistic, standardized test scores. |
|
Number of Downloads: 372 | Number of Views: 1131 |
|