Skewed exchange-rate forecasts

Abstract

We used survey data on exchange-rate forecasts of the dollar/euro exchange rate and the yen/dollar exchange rate to analyze the correlation of the skewness of the distribution of heterogeneous forecasts with movements of the exchange rate. Using various measures of skewness, we found a negative correlation of skewness of 1-month-ahead forecasts with exchange-rate movements. In contrast, the correlation of skewness of 12-months-ahead forecast with exchange-rate movements is positive. The negative correlation arising in the case of 1-month-ahead forecasts is consistent with expected mean reversion in exchange rates. The positive correlation arising in the case of longer term forecasts, in turn, is consistent with longer term bandwagon effects.

Keywords:

• exchange rates
• forecasting
• skewness

JEL Classification:

• F31
• D84
• C33

We thank two anonymous reviewers for helpful suggestions. We acknowledge financial support received through the foundation ‘Geld and Währung’ from the Deutsche Bundesbank (S126/10081/11). The usual disclaimer applies. We also thank the Euro Area Macroeconomic Developments Division of the European Central Bank (ECB) for providing the data, and Jan Rülke for helpful comments.

Notes

1. Strategic forecasting does not rule out that skewness helps to predict exchange-rate changes.

2. As a robustness check, we used the exchange rate measured on the days on which forecasts were made. The results were similar to the results reported in Section 4 (results are not reported, but are available from the authors upon request).

3. Our results do not hinge on our choice of the Epanechniko kernel. Other kernels like, for example, a Gaussian kernel gave similar results (the results are available upon request). A kernel function, K(x)≥0, satisfies , where the Epanechnikov kernel function is given by for |x|≤1, and zero else. For details, see Tsay (2005, Section 4.1.5). The histograms and kernel density functions (and the other empirical results presented in this paper) were computed using the software R (R Development Core Team 2009).

4. For better comparability with our third and fourth skewness measures, Equation (1) gives the formula for computing the first skewness measure from the estimated kernel density functions. Given the limited number of forecasts available in a forecasting cycle for estimation of the kernel density functions, it is important to note that we obtained results (available upon request) similar to those reported in Section 4 when we computed the skewness of the distribution of heterogeneous exchange-rate forecasts from the raw data rather than from the estimated kernel density estimates.

5. We thank an anonymous reviewer for suggesting to include exchange-rate volatility in the regression model. We also considered to estimate exchange-rate volatility by means of a GARCH model. Because of the low frequency of our data, however, exchange-rate returns showed no signs of volatility clustering.

6. The Japanese monetary authorities conducted a large number of heavy actual interventions in the yen/dollar market during the sample period we analyzed in our empirical analysis, supported by frequent oral interventions (Chiu 2003). In contrast to the Japanese monetary authorities, the European Central Bank (ECB) so far only intervened on four occasions in 2000 in the dollar/euro market (Frenkel, Pierdzioch, and Stadtmann 2006). Fratzscher (2006) and Jansen and de Haan (2007), however, identify many instances of oral interventions.