A0452
Title: Text data insights and machine learning innovations in monetary policy shock identification
Authors: Nickson Cabote - Washington State University (United States) [presenting]
Abstract: A new method is introduced to identify monetary policy shocks in the Philippines and enhance macroeconomic forecasts by integrating textual data from the Philippine Central Bank's policy meeting records. Utilizing machine learning techniques and Natural Language Processing (NLP), such as sentiment analysis and TF-IDF, this approach extends previous frameworks. It analyzes exogenous variations in monetary rates independently of central bank macroeconomic forecasts. The integration of Principal Component Analysis (PCA) of macroeconomic factors with textual data significantly enhances forecast accuracy, improving predictions for GDP growth and inflation by up to 42\% and 8\%, respectively, compared to traditional factor-augmented VAR models. Moreover, the refined models, which incorporate a comprehensive set of textual and economic indicators, explain up to 89\% of the variance in policy rates from 2002 to 2020. Applying these monetary policy shocks with the local projections method reveals a quicker GDP response and addresses the 'price puzzle' commonly seen in standard VARs in developing economies. This approach underscores the potential of advanced word vector models to deepen the understanding of monetary policy dynamics in emerging markets.
