AUTHORS: Sjoert Fleurke
Download as PDF
ABSTRACT: The aim of this paper is to check in practice to what extent an ensemble forecast based on averaging the outcomes of several forecasting methods provides better results than single forecasts. Therefore, we use data of monthly new car registrations in the Netherlands and car sales in the USA. The performances of seven popular forecasting methods are assessed and the results are combined into Ensemble forecasts. Several common performance metrics are applied on the results of the test data and it is shown that the Ensembles perform slightly better than each of the forecasting models separately. This confirms the idea, found in literature, that under certain conditions, a combination of several forecasts leads to more accurate results.
KEYWORDS: Automobile Registrations, Automobile Sales, ARIMA, Artificial Neural Network, Ensemble Forecasting, Exponential Smoothing, Generalized Linear Model, Theta, Time Series Forecasting, Random Forest, Vector Auto Regression
REFERENCES:
[1] B. Brühl, M. Hülsmann, C.M. Friedrich, D. Reith, A sales forecast model for the German automobile market based on time series analysis and data mining methods, ICDM 2009: Advances in Data Mining. Applications and Theoretical Aspects, pp. 146-160.
[2] M. Hülsmann, D. Borscheid, C.M. Friedrich, D. Reith, General sales forecast models for automobile markets based on time series analysis and data mining techniques, Transactions on Machine Learning and Data Mining, (5)2, 2012, pp. 65-86.
[3] A. Sa-ngasoonsong, Multi-step sales forecasting in automotive industry based on structural relationship identification, Int. J. Production Economics (140), 2012, pp. 875-887.
[4] D. Fantazzini, et al., Forecasting German car sales using Google data and multivariate models, Int. J. Production Economics, 170A, pp. 97-135, 2015.
[5] J. Scott Armstrong, Combining Forecasts Principles of Forecasting: A Handbook for Researchers and Practitioners, Kluwer Academic Publishers, 2001.
[6] C.C. Holt, Forecasting trends and seasonals by exponentially weighted averages, Carnegie Institute of Technology, Pittsburgh ONR memorandum, no. 52. 1957.
[7] V. Assimakopoulos, K. Nikolopoulos, The theta model: a decomposition approach to forecasting, Int. J. of Forecasting, (16), 2000, pp. 521–530.
[8] S. Makridakis, M. Hibon, The M-3 Competition: results, conclusions, and implications, Int. J. of Forecasting, (16), 2000, pp. 451–476.
[9] R.J. Hyndman, B. Billah, Unmasking the Theta method, Int. J. of Forecasting, (19)2, 2003, pp. 287-290.
[10] L. Breiman, Leo, Random Forests, Machine Learning, (45)1, 2001, pp. 5–32.
[11] J. Nelder, R. Wedderburn, Generalized Linear Models, J. of the Royal Stat. Soc., A (135)3, 1972, pp. 370–384.
[12] R.J. Hyndman, Y. Khandakar, Automatic time series forecasting: the forecast package for R, J. of Stat. Software, (26)3, 2008, pp. 1–22.
[13] A. Liaw, M. Wiener, Classification and Regression by randomForest, R News (2)3, 2002, pp. 18-22.
[14] B. Pfaff, Package Vars, 2007. http://cran.rproject.org/web/packages/vars/index.html.
[15] G.E.P. Box, G.M. Jenkins, G.C. Reinsel, Time Series Analysis: Forecasting and Control. 3rd ed. Englewood Cliffs, NJ: Prentice Hall, 1994.