Monetary policy has undergone substantial changes in recent years, with central banks increasingly using communication as a key policy instrument. In South Africa, the South African Reserve Bank (SARB) issues Monetary Policy Committee (MPC) statements to communicate its assessment of inflation, risks, and interest rate direction, serving as a tool for forward guidance in addition to traditional policy measures. This study assesses the predictive value of combining textual features from SARBs MPC statements with macroeconomic indicators. A hybrid framework is used, integrating headline CPI and the repurchase rate with sentiment and topic features extracted using Natural Language Processing (NLP). The results show that textual features from MPC statements provide forward-looking signals that enhance predictive accuracy when combined with macroeconomic variables. The correlation analysis confirmed a systematic relationship between the tone of communication and inflation dynamics. The regression and feature importance analyses consistently identified restrictive monetary sentiment and bond market-related topics as statistically significant predictors of the SARBs monetary policy stance. Limitations of the study include sensitivity to topic modelling thresholds, class imbalance in policy categories, and the narrow scope of macroeconomic variables considered. The study concludes that hybrid approaches improve monetary policy forecasting by incorporating SARB communication into prediction models, thereby contributing to greater transparency, accuracy, and forward-looking insights for policymakers and analysts. The key contribution of this study lies in the comparative evaluation of a fully integrated hybrid machine-learning framework that jointly models macroeconomic indicators and textual features from MPC statements, rather than in the construction of a standalone communication or sentiment indicator, thereby extending existing South African monetary policy research.
- Forecasting monetary policy stance using sentiment analysis and economic data for South Africa
- Precious Makola
- Magdalene Kasyoka Wilson
- University of Johannesburg; Master of Applied Data Science
- Master of Applied Data Science, University of Johannesburg
- 9964201407691
- University of Johannesburg
- University of Johannesburg; College of Business & Economics (CBE); Applied Information Systems
- English
- Thesis