Forecasting Performance Comparison With Panel Data Models: Environmental Kuznets Curve Analysis
Mücella Şahin, Turgut ÜnIn this study, forecast analysis was conducted on the basis of different panel data structures and predictors. Panel data models, constructed within the framework of the Environmental Kuznets Curve, were developed using two separate unit groups: the G20 country group as heterogeneous panel data and the G8 country group as homogeneous panel data for the period 1990–2020. Subsequently, out-of-sample forecasts were obtained using a fixed effects predictor, a random effects predictor, and combined forecasting methods, and the performances of these forecasts were compared. Forecast values were estimated for out-of-sample 1 year, 3 years, and a 3-year average. Forecast performances were evaluated using the mean squared error and root mean squared error. As a result, it was found that, in line with the literature, homogeneous predictors exhibited better performance. In addition, it was observed that the forecast obtained with the fixed effects predictor in the homogeneous panel data structure performed better, whereas the forecast obtained with the random effects predictor in the heterogeneous panel data structure performed better. The combined forecast in the homogeneous panel data structure was better than the forecast obtained with the random effects predictor, whereas in the heterogeneous panel data structure, the forecast obtained with the fixed effects predictor performed worse than the combined forecasting method. In this study, the combined forecasting method developed by Huang (2019) was examined, and its performance was compared with other forecasting methods. Another perspective of this study was to examine the performance of the combined forecasting method under different heterogeneity and endogeneity conditions.
Panel Veri Modelleri İle Öngörü Performans Kıyaslaması: Çevresel Kuznets Eğrisi Analizi
Mücella Şahin, Turgut ÜnBu çalışmada farklı panel veri yapılarına ve farklı tahmincilere bağlı olarak öngörü analizi yapılmıştır. Çevresel kuznets eğrisi çerçevesinde oluşturulan panel veri modelleri, heterojen ve homojen panel veri grubu olarak iki ayrı birim grubu üzerinden oluşturulmuştur. Heterojen panel veri olarak G20 ülke grubu ve homojen panel veri olarak G8 ülke grubu ele alınarak 1990 – 2020 dönemleri için oluşturulmuştur. Ardından sabit etkiler tahmin öngörüsü, tesadüfi etkiler tahmin öngörüsü ve birleşik öngörü yöntemleri ile örneklem dışı öngörüleri elde edilerek bu öngörülerin performansları karşılaştırılmıştır. Çalışmada örneklem dışı 1 yıl, 3 yıl ve 3 yıl ortalaması için öngörü değerleri tahmin edilmiştir. Öngörü performansları ortalama hata kare ve kök ortalama hata kare ile değerlendirilmiştir. Sonucunda ise literatürdeki çalışmalarla uyumlu olarak homojen tahmincilerin daha iyi performans gösterdiğine ulaşılmıştır. Ayrıca homojen panel veri yapısında sabit etkiler tahmincisi ile elde edilen öngörünün, heterojen panel veri yapısında ise tesadüfi etkiler tahmincisi ile elde edilen öngörünün daha iyi performans sergilediği görülmüştür. Homojen panel veri yapısında birleşik öngörünün tesadüfi etkiler tahmincisi ile elde edilen öngörüden daha iyi olduğu ve heterojen panel veri yapısında ise sabit etkiler tahmincisi ile elde edilen öngörünün birleşik öngörü yönteminden daha kötü performans sergilediğine ulaşılmıştır. Çalışmada, Huang (2019) tarafından geliştirilen yeni öngörü yöntemlerinden birleşik öngörü yöntemi incelenerek diğer öngörü yöntemleri ile performansları kıyaslanmıştır. Bu çalışmanın bir diğer bakış açısı ise birleşik öngörü yönteminin heterojenlik ve içsellik açısından farklı koşullar altındaki performansı da incelenmiştir.
In this study, forecasting analysis is performed on the basis of different panel data structures and different estimators. The panel data model built within the framework of the environmental Kuznets curve is constructed for the period 1990-2020 by taking the G20 country group as heterogeneous panel data and the G8 country group as homogeneous panel data. Then, fixed-effects forecasting, random-effects forecasting, and combined forecasting methods are used to obtain out-of-sample forecasts their performances are compared. In this study, out-of-sample forecasts are estimated for 1 year, 3 years a average. Forecast performances are evaluated using the mean square error (MSE) and root mean square error (RMSE). The results show that homogeneous forecasters perform better in line with the studies in the literature. It is also observed that the prediction obtained with the fixed effects estimator outperforms the prediction obtained with the random effects estimator in the homogeneous panel data structure and the prediction obtained with the random effects estimator in the heterogeneous panel data structure. In the homogeneous panel data structure, combined forecasting outperforms obtained with the random effects estimator in the heterogeneous panel data structure, obtained with the fixed effects estimator performs worse than combined forecasting. Another perspective of this study is to examine the performance of the combined forecasting method under different conditions in terms of heterogeneity and endogeneity.
In the analysis of the study, stationarity analysis of the variables used for the environmental Kuznets curve panel data models for the G20 and G8 countries was performed. Then, it is determined that the model for the G20 country group is heterogeneous and that the panel data model for the G8 country group is homogeneous. Since it is aimed to compare the forecasts of fixed effects estimator, random effects estimator combined estimator as forecasting estimators, panel data models are estimated with these estimators and then out-of-sample forecasts are estimated by estimating in 1-year and 3-year samples.
In this study, G20 countries (Argentina, Australia, Brazil, Canada, China, European Union, France, Germany, India, Indonesia, Italy, Japan, Mexico, Russia, Saudi Arabia, South Africa, Turkey, United States, United Kingdom) covering the period 1990 - 2020 are considered as the heterogeneous panel data model, while G8 countries (Canada, France, Germany, Italy, Japan, Russia, United States, United Kingdom) with the same time dimension covering the period 1990 - 2020 are considered as the homogeneous panel data model. For both G20 and G8 country group panel data, one-year (2020) and three-year (2018, 2019 and 2020) forecasts are made for the combined forecasts developed with the within-group fixed effects estimator, the generalised least squares random effects estimator then the combination of fixed effects and random effects estimators. These forecasts are compared with the MSE and RMSE performance criteria.
The contribution of this study to the literature can be expressed as the comparison of forecasts by considering homogeneous and heterogeneous panel data structures separately. In addition, this study examines the performance of the combined forecasting method proposed by Huang (2019), which emerged as a different forecasting method, in these different panel data structures and analyzes the performance of the forecasts obtained with fixed and random effects estimators in cases of heterogeneity and endogeneity.
The aim of this study is to compare the performance of the prediction values by estimating the models with different panel data estimators and to analyse the performance of the forecasts of the random effects and fixed effects estimators as well as the combined forecasts in the presence of heterogeneity and endogeneity. The second section of the paper presents the panel data model used, the Environmental Kuznets Curve, and reviews related studies. The third section presents the literature on forecasting methods with panel data and the results of these studies. In the fourth section, panel data estimators and forecasting theory are presented, and in the fifth section, environmental Kuznets curve panel data model estimations and findings obtained from forecasts are explained. In the last section, forecasting performance comparisons and evaluations obtained from the application are presented.