Counting your mobile customers one by one: mobile transaction predictions using buy-till-you-die models

Abstract

This study analyses the complete trading records of 217,614 mobile stock traders in Korea to test how the buy-till-you-die (BTYD) class of probabilistic models work in predicting mobile transaction patterns. We find that BTYD models show satisfactory levels of churn prediction performance. To investigate the impact of irregular mobile trading patterns (e.g., binge trading behaviour), we cross-sectionally divide the data across clumpiness levels and check how a catalogue of BTYD models performs for each level of binge behaviour. The results show that BTYD models tend to operate better in a subsample consisting of those customers with less clumpy trading patterns. We confirm that customers with clumpy transaction patterns exacerbate prediction performance, especially when we try to anticipate a longer period. Thus, this study provides practical guidance for mobile app companies engaging in customer relationship management and sheds new light on the literature regarding binge behaviour and transaction pattern prediction in the context of mobile apps.