Valuation of Bulk Debt Portfolios Based on Clustering and Historical Operational Data: Evidence from Poland

Abstract

Bulk debt portfolios constitute a significant asset class in the financial market, yet their acquisition value is often determined under uncertainty regarding future recoveries and collection costs. Existing valuation approaches rarely integrate debtor-level operational data, segmentation logic, and cost-adjusted recovery estimates into one transparent decision-support model. This study addresses this gap by developing an interpretable valuation model for bulk debt portfolios based on clustering and historical operational data from the Polish debt collection market. The empirical analysis uses 120,000 real debt claims and applies two unsupervised machine learning methods: K-means clustering and Ward’s hierarchical clustering. The historical portfolio is divided into homogeneous debtor segments, and each segment is assigned observed financial parameters, including recovery rates and collection costs. The value of a new portfolio is then estimated as the discounted sum of expected recoveries less expected collection costs across the identified segments. The results indicate three stable and economically interpretable portfolio segments. The consistency between K-means and Ward’s method is high, with an adjusted Rand index of 0.956, while the silhouette coefficient of 0.306 confirms a moderate but meaningful segmentation structure. The segments differ in debtor characteristics and financial performance, which supports their separate treatment in valuation. The proposed model improves practical portfolio pricing by combining machine learning segmentation with cost-adjusted cash-flow logic, making it suitable for debt collection companies and securitization funds operating in markets with large-scale receivables.