Econometrics of sentiments- sentometrics and machine learning: The improvement of inflation predictions in Romania using sentiment analysis

Considering the necessity to have accurate inflation forecasts in a pandemic period with hyperinflation in many countries, the aim of this study is to improve the quarterly inflation forecasts provided by the National Bank of Romania using sentiment analysis. The sentiment forecasts based on narratives in the official reports of the central banks outperformed the numerical predictions of the central bank and various combined forecasts on the horizon 2008:Q1–2021:Q4. In addition, more forecasting models based on machine learning, sentiment indices and various forecasts provided by the National Bank of Romania were proposed. The forecasting model that used signals based on Fourier transform as inputs in artificial neural network and support vector machine performed better than all the other models in terms of forecast accuracy.