Abstract
We estimate the natural rate of interest (r∗) using a semi-structural model of the U.S. economy that jointly characterizes the trend and cyclical factors of key macroeconomic variables such as output, the unemployment rate, inflation, and short- and long-term interest rates. We specify a monetary policy rule and a 10-year Treasury yield equation to exploit the information provided by both interest rates to infer r∗. However, the use of a monetary policy rule with a sample that spans the Great Recession and its aftermath poses a challenge because of the effective lower bound. We devise a Bayesian estimation technique that incorporates a Tobit-like specification to deal with the censoring problem. We compare and validate our model specifications using pseudo-out-of-sample forecasting exercises. Our results show that the smoothed value of r∗ declined sharply around the Great Recession, eventually falling below zero, and remained negative through early 2020. Our results also indicate that obviating the censoring would imply higher estimates of r∗ than otherwise. We also extend our results to the COVID-19 pandemic period, introducing stochastic volatility in the model and dealing with the massive swings in the data, to find that our estimate of r∗ is slightly below 1 percent in early 2023.
Authors
- Manuel González-Astudillo
- Jean-Philippe Laforte
JEL codes
- C32
- C34
- E32