Valuation of a Non-Performing Loan Portfolio

Abstract

This master’s thesis focuses on valuation of a non performing loans portfolio, provided by partner company Dignisia. Two models are developed; a combined classificationregression model and a Markov chain model. Valuation performances are decent but explanatory power, i.e. R2 values, are lower or on par with similar research. The two models are tested in two scenarios with the aim of investigating improvement in model performance with knowledge of prior payment history. No clear relation is found between demographic and errand-specific attributes and debt collection rate. The Markov chain model shows similar performance as the more conventional method static pool, in portfolio valuation. However, advantages of the Markov model are the adaptation to new data and the possibility of model extensions which are further discussed. Data quality and quantity are presumed to be the major limiting factors, which is in line with conclusions in the literature.