En analyse av prognosemetoder for Flust.no AS

Abstract

The purpose of this thesis is to examine how Flust.no AS should perform their forecasting in the future. Today Flust does not have the capacity to develop and implement forecasting to their desired extent. The research question of this thesis is; "Which forecasting methods are best suited to make short-term forecasts for basic products at Flust.no AS?" To answer this question we have in collaboration with Flust chosen to analyze the subcategory; car seats for children in the category children and infants. The product category car seats for children consists of four different weight categories. In addition to analyze each car seat, we have also analyzed aggregated demand for car seats in total and by weight category. Our analysis bases on historical sales data from January 2013 to February 2015. We have used ARIMA analysis because this method analysis the data through a process that provides objective and theoretical foundation for choosing the best suited forecasting model. We have also chosen to analyze time series with averageand exponential smoothing models, as these are easier to implement. The ARIMA process only resulted in significant parameter values in three of the total 16 time series. In these three cases, the ARIMA forecast were better than the exponential smoothing and average models. Our results show that on a general basis, both aggregated and disaggregated demand fluctuates, and that correlation between the different values in the time series is low. Thus, we cannot recommend a specific quantitative method, but rather recommend using simple exponential smoothing as an aid for decision making in forecasting future demand.