Forecasting of Demand for Cars

The main purpose of the work was to determine the way to predict seasonal time series with the best predictive power. We chose two methods for comparison: SARIMA and artificial neural networks. The specifications of these models with exogenous variables were also considered. The forecasting object was a time series based on sales of the automobile Lada Granta. A set of factors influencing the dependent variable was based on existing literature in the field of forecasting and in the process of studying the automobile industry in Russia. The data were collected for the period from March 2012 to March 2018 and divided into two samples for training and testing. Comparison of models was based on the MdAPE indicator. The results of the research showed that for a long-term forecast, it is better to use the model of artificial neural networks, but it is necessary to get rid of the trend and seasonality of the series. For the short-term forecast, the SARIMA method is suitable. In both cases, including of exogenous variables led to positive results. The practical application of the findings is that they can be used in real business when choosing the method of forecasting seasonal demand, thus saving time for self-comparison.