Comparison of sales forecasting models for an innovative agro-industrial product: Bass model versus logistic function

This paper compares the accuracy of sales forecasting between Bass model (Bass, 1969) and Logistic function (Stoneman, 2010). It uses several ways to estimate the models; least squares with quadratic interpolation, least squares with quasi-Newton, maximum likelihood with quadratic interpolation and maximum likelihood with quasi-Newton. It applies the technique to an innovative agro-industrial product, feta cheese from buffalo milk. Then it compares the performance of the models by Mean Absolute Percentage Error (MAPE) of the out-of-sample test. It matches Bass model and Logistic function which are estimated by the same method and compare their performances. Moreover, it compares the best Bass model with the best Logistic function regardless of the estimation method. The results reveal that, in most pairs, Logistic function is superior than Bass model when the model uses the data between 7 to 24 months which MAPE of Logistic function are improved tremendously. However, the performance of the best Logistic function is insignificantly different to that of Bass model.