Forecasting the Turkish Manufacturing Industrial Production Index: An Empirical Comparison of Time Series and Machine Learning Models

Ufuk Bingöl; Hasan Murat Ertuğrul; Necmettin Alpay Koçak

Makale detayları

Forecasting the Turkish Manufacturing Industrial Production Index: An Empirical Comparison of Time Series and Machine Learning Models

Yazarlar:

Ufuk Bingöl | Hasan Murat Ertuğrul | Necmettin Alpay Koçak

Yayıncı:

Istanbul University

Ana Başlıklar:

Ekonomi Teorisi (Diğer)

Dergi:

İstanbul İktisat Dergisi Vol.75 Issue 1, pp.1-16

Belge Tipi:

Makale

Yayın Tarihi:

7/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001355184

pISSN:

2602-4152

Anahtar kelimeler:

Özet:

 Macroeconomic variables are important in both following cyclical economic developments and answering the questions of decision-makers and investors about the future. In this context, investigating the industrial production index dynamics over time provides rapid and important signals about the general economic prospects. Therefore, the effects of the COVID-19 outbreak on the forecasting performance of economic variables have been increasingly investigated in the literature. This study examines the forecasting performance differences between time series and machine learning models for the Turkish Manufacturing Industrial Production Index across the pre- and post-COVID-19 periods. Using econometric and machine learning methods, we identified that the time series models performed better before COVID-19, while the machine learning models excelled post-COVID-19. According to the results for the preCOVID-19 period, the ARDL model, which is a member of the time series model family, produces the best results in terms of forecast performance criteria, however the Principal Component Analysis model, which is a member of the machine learning model family, is found to be the best performing model for the post-COVID-19 period. This finding implies that the forecast performance of the time series and machine learning models is different depending on the COVID-19 outbreak. Time series models produce robust forecast performance before the COVID-19 period, whereas machine learning family member models produce robust results after the COVID-19 period for the Turkish Manufacturing Industrial Production Index variable. These results highlight the shifting utility of model families under economic disruption, offering insights for policymakers and forecasters. JEL Classification : C4 , C45.

 Macroeconomic variables are important in both following cyclical economic developments and answering the questions of decision-makers and investors about the future. In this context, investigating the industrial production index dynamics over time provides rapid and important signals about the general economic prospects. Therefore, the effects of the COVID-19 outbreak on the forecasting performance of economic variables have been increasingly investigated in the literature. This study examines the forecasting performance differences between time series and machine learning models for the Turkish Manufacturing Industrial Production Index across the pre- and post-COVID-19 periods. Using econometric and machine learning methods, we identified that the time series models performed better before COVID-19, while the machine learning models excelled post-COVID-19. According to the results for the preCOVID-19 period, the ARDL model, which is a member of the time series model family, produces the best results in terms of forecast performance criteria, however the Principal Component Analysis model, which is a member of the machine learning model family, is found to be the best performing model for the post-COVID-19 period. This finding implies that the forecast performance of the time series and machine learning models is different depending on the COVID-19 outbreak. Time series models produce robust forecast performance before the COVID-19 period, whereas machine learning family member models produce robust results after the COVID-19 period for the Turkish Manufacturing Industrial Production Index variable. These results highlight the shifting utility of model families under economic disruption, offering insights for policymakers and forecasters. JEL Classification : C4 , C45.

 Macroeconomic variables are important in both following cyclical economic developments and answering the questions of decision-makers and investors about the future. In this context, investigating the industrial production index dynamics over time provides rapid and important signals about the general economic prospects. Therefore, the effects of the COVID-19 outbreak on the forecasting performance of economic variables have been increasingly investigated in the literature. This study examines the forecasting performance differences between time series and machine learning models for the Turkish Manufacturing Industrial Production Index across the pre- and post-COVID-19 periods. Using econometric and machine learning methods, we identified that the time series models performed better before COVID-19, while the machine learning models excelled post-COVID-19. According to the results for the preCOVID-19 period, the ARDL model, which is a member of the time series model family, produces the best results in terms of forecast performance criteria, however the Principal Component Analysis model, which is a member of the machine learning model family, is found to be the best performing model for the post-COVID-19 period. This finding implies that the forecast performance of the time series and machine learning models is different depending on the COVID-19 outbreak. Time series models produce robust forecast performance before the COVID-19 period, whereas machine learning family member models produce robust results after the COVID-19 period for the Turkish Manufacturing Industrial Production Index variable. These results highlight the shifting utility of model families under economic disruption, offering insights for policymakers and forecasters. JEL Classification : C4 , C45.

 Macroeconomic variables are important in both following cyclical economic developments and answering the questions of decision-makers and investors about the future. In this context, investigating the industrial production index dynamics over time provides rapid and important signals about the general economic prospects. Therefore, the effects of the COVID-19 outbreak on the forecasting performance of economic variables have been increasingly investigated in the literature. This study examines the forecasting performance differences between time series and machine learning models for the Turkish Manufacturing Industrial Production Index across the pre- and post-COVID-19 periods. Using econometric and machine learning methods, we identified that the time series models performed better before COVID-19, while the machine learning models excelled post-COVID-19. According to the results for the preCOVID-19 period, the ARDL model, which is a member of the time series model family, produces the best results in terms of forecast performance criteria, however the Principal Component Analysis model, which is a member of the machine learning model family, is found to be the best performing model for the post-COVID-19 period. This finding implies that the forecast performance of the time series and machine learning models is different depending on the COVID-19 outbreak. Time series models produce robust forecast performance before the COVID-19 period, whereas machine learning family member models produce robust results after the COVID-19 period for the Turkish Manufacturing Industrial Production Index variable. These results highlight the shifting utility of model families under economic disruption, offering insights for policymakers and forecasters. JEL Classification : C4 , C45.

Makale detayları

Forecasting the Turkish Manufacturing Industrial Production Index: An Empirical Comparison of Time Series and Machine Learning Models

Yazarlar:

Ufuk Bingöl | Hasan Murat Ertuğrul | Necmettin Alpay Koçak

Yayıncı:

Istanbul University

Ana Başlıklar:

Ekonomi Teorisi (Diğer)

Dergi:

İstanbul İktisat Dergisi Vol.75 Issue 1, pp.1-16

Belge Tipi:

Makale

Yayın Tarihi:

7/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001355184

pISSN:

2602-4152

Forecasting

Industrial production index

Time series models

Machine learning models

Sayı

Makale detayları

Forecasting the Turkish Manufacturing Industrial Production Index: An Empirical Comparison of Time Series and Machine Learning Models

Yazarlar: Ufuk Bingöl | Hasan Murat Ertuğrul | Necmettin Alpay Koçak

Yayıncı: Istanbul University

Ana Başlıklar: Ekonomi Teorisi (Diğer)

Dergi: İstanbul İktisat Dergisi Vol.75 Issue 1, pp.1-16

Belge Tipi: Makale

Yayın Tarihi: 7/2025

Belge Dili: İngilizce

Document ID (Org No.): 2001355184

pISSN: 2602-4152

Forecasting

Industrial production index

Time series models

Machine learning models

Sayı

Yazarlar:

Ufuk Bingöl | Hasan Murat Ertuğrul | Necmettin Alpay Koçak

Yayıncı:

Istanbul University

Ana Başlıklar:

Ekonomi Teorisi (Diğer)

Dergi:

İstanbul İktisat Dergisi Vol.75 Issue 1, pp.1-16

Belge Tipi:

Makale

Yayın Tarihi:

7/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001355184

pISSN:

2602-4152