Forecasting newspaper demand with censored regression

Newspaper circulation has to be determined at the level of the individual retail outlets for each of the editions to be sold through such outlets. Traditional forecasting methods provide no insight into the impact of the service level defined as the probability that no out-of-stock will occur. The service level results in out-of stock situations, causing missed sales and oversupply or returns. In our application management sets a policy aiming at a 97 percent service level. The forecasting system developed provides estimates for excess deliveries and for the expected shortages. The results compare favorably to the traditional moving average approach previously employed by the publisher. Censored regression is a natural approach to the newspaper problem. It provides information on key policy variables and it is relatively simple to integrate into the distribution policy, with only small adaptations to the existing forecasting and distribution policy.