Insolvency modeling in the cellular telecommunication industry

In this article, we test several models (decision tress and neural networks) to predict customer insolvency at one of the cellular telecommunication operator in Poland. In comparison to previous studies on customer insolvency, our research presents novelty in the following areas. First of all, we deal with cellular telecommunication customers who use services without paying dues (the literature show the studies that dealt mainly with fixed line operators or banking industry). Secondly, we have a large dataset with 205 explanatory variables derived from the database (previous studies dealt with smaller datasets). Thirdly, we test the quality of the models three months after the estimation with lift and ROC curves to assess the models performance over the time. in findings from the research are twofold: (1) artificial neural networks perform well when modeling the customer’s insolvency and the best model can capture significant amount of money owed by insolvent customers; (2) decision trees get old quickly and their performance decrease over the time in the top percentiles of the score.