Institute for Economic Forecasting Romanian Journal of Economic Forecasting – 3/2011162 CONSUMER CREDIT SCORING

After presenting the main issues in consumer credit market and introducing the issue of credit scorecards, I have used statistical modeling to predict the default probabilities of applicants in a dataset of consumer loans. I have found evidence for the superior accuracy of complex non-linear estimations. In particular, the bagging model offers better results than the traditional tree and logit estimations. The proposed statistical scorecard offers a 60 percent improvement over the baseline model. Lastly, this paper argues that the management must establish a decisional probability threshold in accordance with its propensity for risk. A higher threshold requires a greater promotional effort, although the increased costs may be compensated by a more efficient communication with clients and by more flexible contractual clauses