ABSTRACT
Using FRED data, a machine-learning model outperforms the Survey of Professional Forecasters and other models since 2001 in forecasting the unemployment rate.
Acknowledgments
We thank an anonymous referee for suggestions. Views and errors are those of the authors and are not necessarily those of the Federal Reserve Bank of Dallas or the Federal Reserve System.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 For each data row, we include four observations of each variable from the current date to a year before.
2 For one- and two-quarter ahead forecasts, the RMSE was also minimized using non-rolling forecasts, and a logistic activation function with three hidden layers.
3 The three most informative categories for the full sample were international, production and prices for two-quarter ahead forecasts, and international, prices, and labour for one-quarter ahead forecasts.
4 These were in the top-ten for one- and two-quarter-ahead forecasts, with slightly different rankings from four to ten.