Modeling Within- and Across-Customer Association in Lifetime Value with Copulas

Recent advances in linking Recency-Frequency-Monetary value (RFM) data to Customer Lifetime Value (CLV) in non-contractual settings rely on the assumption of independence between the transaction and spend processes. We propose to model jointly the inter- and intra-customer dependency between both processes using copulas, hereby accounting for the double correlation within and across customers. Applied to a unique data set of securities' transactions, we nd that modeling both associations enhances the accuracy of CLV predictions, thus improving customer valuation and selection tasks.