Forecasting with Dynamic Factor Model and Mixed-Frequency Data

Abstract

Macroeconomic forecasting plays a crucial role in shaping monetary and fiscal poli- cies, guiding decision-making in both public and private sectors. Traditional economic indicators, often released at lower frequencies, are insufficient for capturing the rapid shifts in economic conditions, particularly in post-pandemic periods. This research aims to enhance macroeconomic forecasting by integrating sentiment analysis derived from higher-frequency textual data, such as central bank communications, into traditional econometric models. The research evaluates MIDAS, Kalman Filter-based Dynamic Factor Models (DFM), and MF-VAR to determine the effectiveness of incorporating sentiment alongside financial market indicators, such as the S&P/ASX 200 index. The findings indicate that sentiment alone does not consistently outperform financial indicators but adds predictive value when combined with asset prices. In addition, integrating both sentiment and financial variables improves GDP forecasting accuracy, particularly at longer horizons. These findings underscore the value of high-frequency textual analysis in macroeconomic forecasting and highlight the complementary role of sentiment and financial market data in improving macroeconomic forecasting.