Forecasting demand of magazines and modelling seasonality

Abstract

The main aim of this thesis is to find a sensible way to model the seasonality and forecast the demand of magazines automatically. Demand forecasting in magazine industry is very complex and historic delivery and sale data are often short, unstable and particularly perturbed by numerous factors. Generating forecasts from these large numbers of time series requires some degree of automation and simple forecasting models. The first part of the thesis explains basic forecasting notions. Especially, the need for an automatic forecasting system is emphasized and the steps of automatic forecasting study are explained. Also, a statistical analysis is done to decide the suitable smoothing model alternatives. Finally, initialization and parameter optimization procedures are discussed. In the second part, demand estimation and handling of the censored demand in case of sellout is analyzed. In addition, the two main strategies used for planning are mentioned: topdown and bottom-up. The third part presents new forecasting methods based on combining forecasts and grouping similar characteristic endpoints by using real data. The last part explains the data organization and calculation of MAD and lost sales by using R and describes the some important algorithms that are used in magazine forecasting. .