SME RATING: RISK GLOBALLY, MEASURE LOCALLY

AbstractThe aim of this chapter is to investigate the superiority of local modeling in the SME default risk estimation.Both “Regional” and “national” models are developed on a dataset of 4,134 enterprises allocated into three samples: a regional “in-sample” (3,137 companies), a regional “out-of-sample” (515 companies), and a national “out-of-sample” (482 companies). By comparing the models' accuracy (ROC), our findings demonstrate the superiority of regional models for SME default risk estimation on the similar national-based models.When geographical sampling is applied the accuracy increases in all local industry-specific models as well as in the local “general” model. In addition to higher accuracy results, the regional sampling approach made considerable simplification to the rating calibration due to the capability of the regional models to immediately adjust to the observed default rates in the region.