Service Retention Forecasting in the Wireless Telecom Industry Using Limited Information

Understanding which timing models best forecast customer survival rates is of paramount importance across a range of marketing-related activities. The author analyzes the postpaid subscriber bases of U.S. wireless providers in order to determine which forecasting models best capture the underlying cohort dynamics across carriers. In doing so, he examines the degree of heterogeneity, duration dependence, seasonality, contractual expirations, and cross-cohort differences in these subscriber bases. The author compares the accuracy of these models both in-sample and out-of-sample. The in-sample parameters are calculated through ordinary least squares, given that—as an external constituency—the author must utilize limited information data structures. The results appear to be more conclusive for certain carriers than others; that is, the best in-sample timing models are not always the best out-of-sample, and often times appear to considerably “overfit” the data. The meager historical data available to test these models is most likely to blame. However, the models developed do provide general framework through which to apply these methods to contractual settings as more data becomes available.