An Evaluation of Alternative Forecasting Methods to Recreation Visitation

Abstract

This study examines the advantages and disadvantages of basic, intermediate, and advanced methods for visitor use forecasting where seasonality and limited data are characteristics of the estimation problem. The monthly use rates at the Milwaukee County Zoo, Wisconsin are used to illustrate the seasonal time series techniques. Forecasting methods include the Naive 1, Naive 2, single moving average (SMA) with the classical decomposition procedure, single exponential smoothing (SES), double exponential smoothing (DES), Winter's, and the seasonal autoregressive integrated moving average (SARIMA). The variation in visitor rates over the years makes the visitation trend for the Milwaukee County Zoo appealing in this empirical application. The series ranges from January 1981 through December 1999, a total of 228 months. The last 12, 24, or 60 months of those data are excluded from the original analysis, and used to evaluate the various methods. SARIMA and SMA with the classical decomposition procedure are found to be roughly equivalent in performance, as judged by modified mean absolute percentage error (MAPE) and modified root mean square percentage error (RMSPE) values of a longer estimation period with shorter period ahead forecasts. This study also finds that the SMA with classical decomposition method is more accurate than other techniques when a shorter estimation period with longer period ahead forecasts are included. While this study may not speak to all users of leisure related data, it serves as a comparative reference for those who seek guidance in deciding among a set of forecasting tools.