Although central bankers’ inflation forecasts tend to be fairly accurate during normal times, they do not perform as well during downturns and periods of extreme uncertainty, such as the COVID-19 pandemic. To improve this performance gap, researchers over the past 20 years have proposed various innovations to a benchmark class of models known as “time-varying parameter models,” which allow the relationships between forecasting variables to change over time. However, most research on the efficacy of these innovations was conducted prior to the COVID-19 pandemic, leaving the question of how these “improved” models have performed during recent extreme events.
Amaze Lusompa and Sai A. Sattiraju investigate whether innovations in time-varying parameter models led to improved inflation forecasting during the pandemic. They find that despite their promise prior to the pandemic, forecasting innovations did not improve the accuracy of inflation forecasts relative to a baseline time-varying parameter model during the pandemic. Their results suggest that forecasters may need to develop a new class of forecasting models, introduce new forecasting variables, or rethink how they forecast to yield more effective inflation forecasts during extreme events.
Publication information: Vol. 107, no. 3