Comparison of car demand forecasting models

Automobiles are one of a few products which are bought by some ivestment, credit support etc. They are the second most expensive durable goods after the houses. The importance of automobile industry is not only because of consumer aspect. The producer aspect is also an important side of the industry. Automobile industry's production and demand numbers are one of the indicator of the state of the countries economies. Automobile industry is related with many other sectors. So predicting the future of automobile industry has some positive impact on planning of other sectors related with automobile industry. Automobile demand forecasting studies are not only used for industrial purposes. The output of the automobile demand forecasting can be used to predict the number of automobiles and car ownership density of the countries. Then these results are able to be used as inputs in planning of future infrastructures such as roads, bridge, public transportation infrastructure etc. Automobile demand forecasting can be studied by the support of experts who have important information and experience about the sector. Another option to study with automobile demand forecasting is mathematical modelling of the forecasting problems by finding the relationships between the inputs. The studies in literature have been conducted by many methods and many independent variables. The methods used in the literature are time series, ARIMA, regression, econometric models, artificial neural networks, fuzzy logic, support vector machines, genetic algorithms, gompertz model, ANFIS, Pena Box etc. In our study we used two types of data sets which contains normal and seasonally adjusted monthly sales numbers of automobiles and LCV (Light Commercial Vehicle). Addition to the classical time series methods and linear regression methods, we used data mining methods such as artificial neural networks and support vector regression. The output of the calculations are one period and six period ahead forecast of the data sets. The forecasting power of the models are evaluated by MAPE (Mean Absolute Percentage Error) in the six months and one month periods. The methods which have three least MAPE values for six months prediction interval are given with error indicators of MAPE, NRMSE (Normalized Root Mean Squared Error) andMAD (Mean Absolute Deviation). The results of the applications showed that automobile sales are forecasted best by regression and LCV sales are best forecasted by time series methods.