Working Paper: CEPR ID: DP14112
Authors: Ritong Qu; Allan Timmermann; Yinchu Zhu
Abstract: To answer this question, we develop new testing methods for identifying superior forecasting skills in settings with arbitrarily many forecasters, outcome variables, and time periods. Our methods allow us to address if any economists had superior forecasting skills for any variables or at any point in time while carefully controlling for the role of “luck” which can give rise to false discoveries when large numbers of forecasts are evaluated. We propose new hypotheses and test statistics that can be used to identify specialist, generalist, and event-specific skills in forecasting performance. We apply our new methods to a large set of Bloomberg survey forecasts of US economic data show that, overall, there is very little evidence that any individual forecasters can beat a simple equal-weighted average of peer forecasts.
Keywords: economic forecasting; superior predictive skills; multiple testing; Bloomberg survey
JEL Codes: No JEL codes provided
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| individual forecasters' skills (C53) | forecasting performance (C53) |
| individual forecasters' skills (C53) | outperforming equal-weighted average of peer forecasts (G17) |
| outperforming autoregressive benchmarks (C22) | evidence of superior predictive skills (C52) |
| multiple hypothesis testing (C12) | evidence of any superior predictive skills (C52) |
| specialist skills (Y80) | overall ability to outperform average (D29) |