Title: Macro forecasting with adaptive learning
Authors: Prajyna Barua Soni - University of Texas at Dallas (United States)
Azharul Islam - University of Texas at Dallas (United States)
Irina Panovska - University of Texas at Dallas (United States) [presenting]
Srikanth Ramamurthy - International Monetary Fund (United States)
Abstract: The aim is to study how incorporating adaptive learning-based inflation expectations can improve the forecasting performance of Unobserved Components (UC) models when it comes to predicting output, inflation, and unemployment. Our model directly integrates the expectations dynamics of the Hybrid New Keynesian Philips Curve while also retaining the appealing statistical features of the UC framework, allowing us to extract information about the output gap. Three interesting sets of results stand out. First, while the perceived persistence of inflation fell during the early stages of the pandemic, it increased sharply and substantially during the period 2021Q2-2022Q2. Second, the estimated output gap started decreasing in mid-2019, decreased sharply during the early stages of the pandemic, and bounced back rapidly. Finally, and most importantly, including information about the output gap and about the inflation expectations process helps improve both inflation forecasts and output growth forecasts relative to benchmark reduced-form models, with the largest improvements in predictive power during recessions and recovery stages.