Working Paper: NBER ID: w14071
Authors: Kenneth S. Rogoff; Vania Stavrakeva
Abstract: Are structural models getting closer to being able to forecast exchange rates at short horizons? Here we argue that misinterpretation of some new out-of-sample tests for nested models, over-reliance on asymptotic test statistics, and failure to sufficiently check robustness to alternative time windows have led many studies to overstate even the relatively thin positive results that have been found. We find that by allowing for common cross-country shocks in our panel forecasting specification, we are able to generate some improvement, but even that improvement is not entirely robust to the forecast window, and much of the gain appears to come from non-structural rather than structural factors.
Keywords: exchange rates; forecasting; structural models; panel data; out-of-sample tests
JEL Codes: C52; C53; F31; F47
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| structural models (E10) | underperformance relative to the random walk model (G17) |
| misinterpretation of out-of-sample test statistics (C52) | underperformance of structural models (C20) |
| overreliance on asymptotic tests (C52) | underperformance of structural models (C20) |
| common cross-country shocks (F44) | marginal improvement in forecasting accuracy (C53) |
| non-structural factors (P23) | substantial gains in forecasting power (C53) |
| structural models (E10) | lack of robustness in forecasting capabilities (C53) |