|
AUTHOR(S): Sjoert Fleurke
|
TITLE |
 PDF |
KEYWORDS Time Series Analysis, Ensemble Forecasting, Car Registrations, Car Sales, Exponential Smoothing, ARIMA, Artificial Neural Network, Vector Auto Regression, Theta |
ABSTRACT In this paper 5 popular time series forecasting methods are used to predict monthly Dutch new car registrations. The aim is to check whether an ensemble forecast based on averaging would provide better results than single forecasts. Therefore, the performances of these 5 methods are assessed using a data test set of one year and three of them, which had sufficiently independent results, are combined into an ensemble forecast. Using several common performance metrics it is shown that the ensemble performs slightly better than each of these models separately. This is a confirmation of the idea, found in literature, that under certain conditions, a combination of several forecasts leads to better results. |
Cite this paper Sjoert Fleurke. (2017) Ensemble Forecasting the Number of New Car Registrations. International Journal of Transportation Systems, 2, 25-30 |
