Testing rationality of foreign exchange forecasts under flexible loss: survey evidence from Brazil

Abstract

This study asks whether the Brazilian exchange rate (R$/US$) survey forecasts are rational under flexible loss. For 2001–2011, the forecasts overpredict. The bias in shorter-horizon forecasts is due to an inefficient use of information, while the bias in longer-horizon forecasts seems to reflect asymmetric loss. Further evidence indicates that the shorter-horizon (longer-horizon) forecasts are significantly less accurate than (as accurate as) those of the random walk. These forecasts, however, are not directionally accurate and are thus of no value to a user. The backward-looking nature of the forecasts may be due to reliance on simple forecasting rules (heuristics) since experts have great difficulty understanding the complex market dynamics.

Keywords:

• consensus forecasts
• asymmetric loss
• rationality
• random walk
• directional accuracy

JEL Classification:

• F31
• F47

Notes

¹ Daily consensus (median) forecasts are available at www.bcb.gov.br.

² With the forecasts made in month t, the forecast error follows an h-order moving average process under the null hypothesis of rationality. With the forecast errors generally heteroscedastic, we use the Newey–West procedure to correct for both the inherent h-order serial correlation and the heteroscedasticity. For h = 0, we use the White procedure to correct for heteroscedasticity.

³ See, among others, Sobiechowski (1996) and Baghestani (2010).