A new stochastic multi source approach to improve the accuracy of the sales forecasts

Abstract

Purpose The aim of this paper is to suggest a new approach to the problem of sales forecasting for improving forecast accuracy. The proposed method is capable of combining, by means of appropriate weights, both the responses supplied by the best-performing conventional algorithms, which base their output on historical data, and the insights of company’s forecasters which should take account future events that are impossible to predict with traditional mathematical methods. Design/methodology/approach The authors propose a six-step methodology using multiple forecasting sources. Each of these forecasts, to consider the uncertainty of the variables involved, is expressed in the form of suitable probability density function. A proper use of the Monte Carlo Simulation allows obtaining the best fit among these different sources and to obtain a value of forecast accompanied by a probability of error known a priori. Findings The proposed approach allows the company’s demand forecasters to provide timely response to market dynamics and make a choice of weights, gradually ever more accurate, triggering a continuous process of forecast improvement. The application on a real business case proves the validity and the practical utilization of the methodology. Originality/value Forecast definition is normally entrusted to the company’s demand forecasters who often may radically modify the information suggested by the conventional prediction algorithms or, contrarily, can be too influenced by their output. This issue is the origin of the methodological approach proposed that aims to improve the forecast accuracy merging, with appropriate weights and taking into account the stochasticity involved, the outputs of sales forecast algorithms with the contributions of the company’s forecasters.