Inflation Forecasting with Monetary Policy Narratives: Evidence from an Emerging Economy

In recent years, central bank communication has become an important part of modern monetary policy. Beyond setting interest rates, central banks have started to use communication to signal their views on the future economic path.

A growing number of studies suggest that the tone extracted from central bank reports or policy statements can improve forecasts for inflation and other macroeconomic variables. However, most of the evidence comes from developed countries or the euro area while emerging countries remain relatively underexplored. Our paper addresses this gap by examining whether narratives contained in official monetary policy documents improve inflation forecasting accuracy in Romania.

We used quarterly data from the first quarter of 2007 to the fourth quarter of 2024 to construct a sentiment indicator derived from the content of monetary policy minutes and complementary press releases published by the National Bank of Romania. The sentiment measure is determined by using a transformer-based language model trained on financial text and it is then included in a multivariate Bayesian time series model along with inflation. To evaluate its usefulness, we generated forecasts in a recursive out-of-sample framework at horizons of one to four quarters ahead which were then evaluated relative to a univariate autoregressive benchmark.

The results showed that the inclusion of the sentiment index extracted from the monetary policy documents published by the central bank improves forecast accuracy, especially in calm periods. These findings suggest that central bank narratives can serve as a complementary source of information for inflation forecasting in an emerging economy context.