Modeling Industrial Production in the EU Countries: Autoregressive Models Versus Sentiment-Enhanced Regression

Darja Števančec, Patricija Jaklič, Aleša Lotrič Dolinar, Mojca Bavdaž
Statistika, 106(2): 136–151
https://doi.org/10.54694/stat.2025.29

Abstract

Motivated by rising demand for timely economic insights, we explore the predictive power of the Industrial Confidence Indicator (ICI) for forecasting the Industrial Production Index (IPI) across EU Member States. Our purpose is to assess whether business sentiment data can serve as a real-time leading indicator for industrial production and explore possible structural patterns across EU countries and over time. The study analyses monthly IPI and ICI data for 27 EU Member States from 2008 to 2024 (while also considering only the prepandemic part of the time series) using correlation analysis, ARIMA/ARIMAX forecasting methods (with expanding and rolling window techniques) and clustering. Our analysis results in rather weak linear correlation between ICI and IPI, very limited forecasting dominance of ARIMAX models over ARIMA models (especially during volatile periods), and identification of contextually fairly meaningful clusters. Using all three methods the pre-pandemic data turn out to better reflect expected relationships compared to the whole time series.

Keywords
Business tendencies, time-series, timeliness, real-time, estimates, nowcasting, clustering