Forecasting restaurant sales using self-selectivity models

Abstract

An independent restaurateur often opens and closes during different times of the week or year. This induces a self-selectivity bias, because sales are observed only when the restaurant is open and with a minimum sales expectation. Moreover, the operator with historical data will have difficulty assesing whether a particular open/close policy is appropriate. These were the problems faced by our client. Three alternative model specifications were considered. It was found that the truncated regression model both fits the data and forecasts best overall. This confirmed its usefulness as a sales forecasting tool and that the operator had been following an appropriate open/close policy.