Abstract: Using the data from Lossmetric,this paper tries to build forecasting models for non-performing loans of non-extreme recovery. The paper first uses beta-normal transformation and logit transformation to process recovery rates,and gives an individual analysis for each impacting factor on recovery rate. Then the paper proceeds a model-establishing process from simple model to complex model,from single debt obligor's model to multi-debt obligor's model,aiming at a comprehensive investigation on the impacting factors. Empirical results show that both the single debt obligor's model and the multi-debt obligor's model have good predicting power,moreover the single debt obligor's model itself can reflect the relationship between the recovery rate and the impacting factors,and it can be used as the foundation of multi-debt obligor's model.

Key words: non-extreme recovery, non-performing loan, recovery rate, single debtobligor, multi-debt obligor