Automobile spare-parts forecasting: A comparative study of time series methods

Abstract

In Mexico, the automotive industry is considered to be strategic in the industrial and
economic development of the country because it generates production, employment and
foreign exchange. Good demand forecasts are needed for better manufacturing
management. The time series modelling tools applied to the monthly demand forecasting
of automobile spare parts in Mexico are assessed, for the case of a transnational enterprise,
considering affordability. The classic methods of moving averages, final value and
exponential smoothing, the prestigious autoregressive integrated models of moving
averages (ARIMA), the rarely implemented artificial neural networks (ANNs) and the
very little explored ARIMA-ANNs hybrid models are compared. A good performance of
the models involving ANNs is observed, but they were not as steady as the ARIMA
models in the post-sample periods. The mean absolute percentage error (MAPE) was
reduced from an original 57% to 32.65%. The obtained results could help demonstrate
the importance of improving industrial forecasting methodologies for better planning.