Öngörü Tekniklerinin Doğruluk Kıyaslaması: Basit Ekonometrik, ARMA ve ARMAX Teknikleri

Bu çalışmada Basit Regresyon (X), Otoregresif Hareketli Ortalamalar Ekonometrik Sebep-Sonuç (ARMAX) ve Otoregresif Hareketli Ortalamalar (ARMA) tipi tekniklerin öngörü doğruluk dereceleri Ortalama Mutlak Yüzde Hata (MAPE) ve model seçimi istatistik kriterleri (RMSE, AIC, SBC) açısından karşılaştırılmıştır. Ayrıca bu kriterlerin teknikler arası uyumlulukları araştırılmıştır. Basit econometrik modellerden ARMAX modellerine geçişlerde RMSE, AIC ve SBC değerleri karşılaştırılan model çiftlerinde % 88’in üzerinde, MAPE’ de ise dönemler itibariyle % 63-%79 arasında azalmıştır. ARMA’dan ARMAX’a geçişlerde AIC, SBC ve RMSE’da görülen % 71’lik azalışlar MAPE’de dönemler itibariyle % 24-% 35 arasındadır. Bunlar modellerin genelinde sırasıyle % 86, % 86, % 81 olup MAPE’de dönemler itibariyle sırasıyle % 50, % 50, % 60, % 55’dir. Basit ekonometrik (X) ve ARMA modellerinden komplike ARMAX tipi modellere geçişlerde, RMSE, AIC ve SBC arasında % 95’lere varan bir uyumluluk gözlenirken, bu kriterlerin MAPE ile olan uyumlulukları % 52-62 arasında düşük bulunmuştur. Bu sonuç özetleyici istatistiklerle MAPE arasındaki tutarsızlığı işaret eder. Bu uyumsuzluk RMSE, AIC ve SBC kriterlerine göre ARMAX’ın ARMA’ya % 71 olasılıkla tercih edilmesine karşın MAPE kriterine göre ARMA tekniğinin ARMAX’a ilk dönem tahmininde % 60, iki dönem tahmininde % 70, üç dönem tahmininde % 60 ve dört dönem tahmininde % 65 olasılıkla tercih edilmesi çelişkisini sonuçlandırmıştır. Fakat, MAPE kriterine göre ARMAX tekniği basit ekonometrik sebep-sonuç tekniğine bir dönemlik tahmininde % 64, iki dönemlik tahminde % 73, üç dönemlik tahminde % 82 ve dört dönemlik tahminde % 77’lik bir üstünlük sağlamıştır.Tahmin edilen modeller Dickey-Fuller ko-entegrasyon testine göre uzun dönem ilişki göstermiştir.

Anahtar Kelimeler:

ekonometri, doğruluk

Accuracy Comparisions of Forecasting Tecniques: Simple Regressionn, ARMA and ARMAX tekniques

In this study, each one of Autoregressive Moving Average Cause-Effect (ARMAX), Simple Regression (X) and autoregressive moving average (ARMA) techniques is compared with each other in terms of MAPE and in terms of another three summary statistics of model selection criterions (RMSE, AIC, SBC). And the consistency of these criterions is examined among these techniques. In passing from simple econometric models to ARMAX models more than eighty-eight percent of compared couples indicated reductions in values of RMSE, AIC and SBC statistics. However, the reductions in MAPE values range between sixty-three and seventy-nine percent along prediction periods. In passing from ARMA model to ARMAX model, the reduction in summary statistics is about seventy-one percent but the reductions in MAPE range along twenty-four and thirtyfive percent. In overall comparisons, RMSE declined eight-six percent, AIC declined eighty-nine percent, and SBC declined eighty-one percent in entirely sample predictions. On the other hand, the reduction in MAPE is about fifty percent in one and two periods advance predictions, sixty percent in three periods advance and fifty-five percent in four periods advance sample predictions. In passing from a simple econometric (X) and ARMA model to the complicated ARMAX models, a ninety-five percent consistency is observed among RMSE, AIC and SBC criterions in values, however; their consistency with MAPE ranges between fifty-two and sixty-two percent along the sample prediction period. This result implies inconsistency between summary statistics and MAPE criterion. As a result, ARMA technique outperformed ARMAX technique about seventy-one percent in terms of summary statistics; in contrast, ARMAX technique outperformed ARA technique about sixty percent in one and three periods advanced predictions, seventy percent in two periods advanced and sixty-five percent in four periods advanced the sample predictions in terms of MAPE criterion. However, according to MAPE criterion the ARMAX technique outperformed the simple regression around sixty-four percent in one period, seventy-three percent in two periods, eighty-two percent in three periods and seventy-seven percent in four periods advanced sample predictions. And the estimated models exhibited long run relationships based on the Dickey-Fuller test.

Keywords:

econometrics, accuracy,

PDF

___

Akal, Mustafa (2002), Accuracy Comparison of Forecasting Techniques with Variables on Exchange Rate Series: Turkish Liras Versus United States Dollar, Sakarya University Press, Adapazarı.

Akaike, Hirotugu (1981), “Likelihood of A Model and Information Criteria”, Journal of Econometrics, 16, p.3-4.

Box, G.E and Jenkins, G.M. (1970), Time Series Analysis: Forecasting and Control, Holden-Day, San Francisco.

Dickey, D. and Fuller, W. (1979), “Distribution of the Estimators for Autoregressive Time Series with a Unit Root”, Journal of the American Statistical Association, 74, p. 427-431.

Enders, Walter (1995), Applied Econometrics Time Series, John Wiley & Sons Inc, USA.

Fildes, R., Hibon, M., Makridakis, S., and Meade, N. (1998), “Generalizing About Univariate Forecasting Methods: Further Empirical Evidence”, International Journal of Forecasting 14:3, 339-358.

Geurts, M. D. and Ibrahim, I. B. (1975), “Comparing The Box-Jenkins Approach With The Exponentially Smoothed Forecasting Model Application To Hawaii Tourists”, Journal of Marketing Research 12, 182-188.

Groff, G. K. (1973), “Empirical Comparison of Models for Short-range Forecasting”, Management Science 20:1, 22-31.

Huss, W. R. (1985), “Comparative Analysis of Company Forecasts and Advanced Time Series Techniques in The Electric Utility Industry” International Journal of Forecasting 1, 217-239.

IMF (2001), International Financial Statistics Yearbook, Washington, D.C.

Ljung, G.M. and Box, E.P. (1978), “On Measure of Lack of Fit in Time Series Models”, Biometrica, 65, 2, p. 297-303.

Maddala, G. S. (1992), Introduction to Econometrics, Macmillan Publishing Company, New York.

Mahmoud, E. (1984), “Accuracy in Forecasting: A Survey”, Journal of Forecasting 3,2, p. 139-159.

Makridakis, Spyros (1997), “ARMA Models and Box-Jenkins Methodology”, Journal of Forecasting 16, p. 147-160.

_______ (1986), “The Art and Science of Forecasting”, International Journal of Forecasting 2, p. 15-39.

_______ (1993), “Accuracy Measures: Theoretical and Practical Concerns”, International Journal of Forecasting 9, p. 527-529.

Makridakis, S. and Hibon, M. (1997), “ARMA Models and the Box-Jenkins Methodology”, Journal of Forecasting 16, 147-163.

Makridakis, S. and Winkler, R. L. (1983), “Averages of Forecasts: Some Empirical Results”, Management Science 29:9, 987-995.

McCrae, Michael (2002), “Can Cointegration-Based Forecasting Outperform Univariate Models? An Application to Asian Exchange Rates”, Journal of Forecasting”, 21, 5, p. 355-380

Meade, Nigel (2000), “Evidence for the Selection of Forecasting Methods”, Journal of Forecasting, 19, 6, p. 515-535

Narashimhan, G.V.L. (1975), “A Comparative of Predictive Performance of Alternative Forecasting Techniques: Time Series Models vs. An Econometric Model”, Proceedings of American Statistical Association August, 459-464.

Naylor, T. H. and Seaks, T. G. (1972), “Box-Jenkins Methods: An Alternative To Econometric Models”, International Statistical Review 40:2, 113-137.

Nelson, C. R. (1972), “The Prediction Performance of The FRB-MIT PENN Model of The US Economy”, American Economic Review 62, 121-141.

Schwartz, G. (1978), “Estimating the Dimension of a Model”, Annals of Statistics, 6,p. 461-464.

Web Sayfaları

Fiskobirlik (2002), http://fiskobirlik.org.tr/istatis.htm, Fındık Tarım Satış Kooperatifleri Birliği, FKB.