Average ranks of AFE: Table VI shows average rank statistics obtained over four com-
peting forecasting procedures. Analogously to Table V sector and country specific results are
given only for forecasts of export growth whereas average statistics obtained after pooling
one step ahead forecasting errors are shown for exports and imports. The latter results are
also given whenever the conditioning variable indicates a state of relatively low or high FX
uncertainty. To facilitate the interpretation of the results average ranks being significantly
smaller than 2.5 are given in bold entries. On the level of forecast errors pooled over both
dimensions (country and sector) it turns out that for both, export and import forecasts,
the average ranks obtained for the nonlinear model are between 2.25 and 2.35, whereas
for the linear and ”zero” forecasts the corresponding numbers vary between 2.65 and 2.75.
All statistics differ significantly from 2.5 thereby providing a strong argument in favor of a
nonlinear relationship linking FX uncertainty and trade growth on the pooled level. The
same pattern of relative performance is also obtained when pooling forecast errors for each
sector over the cross section. Pooling along the other dimension delivers the same pattern
numerically, even though a few country specific statistics fail significance. Moreover, the
relative performance of linear and ”zero” forecasts on the one hand and nonlinear forecasts
on the other hand, is, abstracting from a few exceptions, more or less uniform over country
and sector specific forecasts of export growth. Numerous average ranks significantly smaller
than 2.5 are obtained for the threshold model as well as for the semiparametric forecasts
even if ”only” 120 one step ahead forecast errors are evaluated. The frequency of cases where
”zero” or linear predictors perform significantly better on average than nonlinear schemes is
negligible.
5 Conclusions and outlook
In this paper we investigate the impact of FX uncertainty on international trade. Distin-
guishing 10 economic sectors we analyze the dynamics of import and export growth for 15
industrialized economies on a multilateral basis.
Regarding the degree of explanation of partial models it turns out that the causal links
operating from FX uncertainty to trade growth are weak throughout. This finding is in
line with most of the empirical literature on the topic even though predecessors have mostly
analyzed aggregated trade flows. The fact that no uniform pattern of the estimated rela-
tionships can be identified across sectors and countries assigns a prominent role to sector
and country specific characteristics of trade markets. These characteristics may be sector
specific cost and demand functions, preferences or habits of traders that may differ owing to
the particular features of the goods traded.
However, forecasting exercises evaluated by means of average rank statistics for absolute
forecast errors indicate the existence of a relationship between FX uncertainty and trade
growth and, furthermore, support the conclusion that this relationship is nonlinear in nature.
17