|
Abstract:
|
This paper examines publicly available Federal Reserve Board Tealbook forecasts of GDP growth for the United States and several foreign countries, focusing on potential time-varying biases and evaluating the Tealbook forecasts relative to other institutions’ forecasts. Tealbook forecasts perform relatively well at short horizons, but with significant heterogeneity across countries. Also, while standard Mincer-Zarnowitz tests typically fail to detect biases in the Tealbook forecasts, recently developed indicator saturation techniques that employ machine learning are able to detect economically sizable and highly significant time-varying biases. Estimated biases differ not only over time, but by country and across the forecast horizon. These biases point to directions for forecast improvement.
|
