Forecast of sales of selected food products in retail using Fourier series analysis and non-linear regression

PurposeThe purpose of this paper is to apply and verify Fourier series analysis in combination with non-linear regression as a tool of forecasting and planning of inputs in the logistics process of a retail chain store.Design/methodology/approachFor many popular products, a significant effect of seasonality of sales is expected; therefore, the method of Fourier series was chosen as one of the main forecast calculation techniques. However, the use of this method directly for forecasting sales has a limitation in the form of a complete reconstruction of the shape of the curve from of the given monitored time. Thus, the forecast is based only on the significant harmonic components from the Fourier series analysis that will participate in forecast forming. In addition, to respect the trend of series, the results of Fourier series analysis are combined with the non-linear regression.FindingsThe results showed that the number of significant harmonic components from the Fourier series analysis is suitable to reflect the future behaviour of the sale in standard market conditions. Forecasting of the sale and accurate purchase planning of goods has a positive effect on reducing the waste of unsold products after their shelf and on increasing of a customer satisfaction.Research limitations/implicationsThis study has an application in a certain period of time (relatively calm behaviour of the food market) and only for a certain region. Therefore, it is not possible to generalize these results as the behaviour of consumers, e.g. within the state. It will also be interesting to monitor and forecast sales of other food items.Practical implicationsThis provides a practical and relatively simple tool for implementing or improving the process of forecasting seasonally dependent products in the food industry.Originality/valueThis study shows the possibility of forecast that is based on adding the significant harmonic components from the Fourier series analysis to form forecast with the non-linear regression.