1 Introduction
The impact of exchange rate uncertainty on international trade has been generating a huge
body of controversial theoretical and empirical literature.1-11 Following a seminal argument
risk averse traders will reduce traded quantities when facing costs involved with hedging FX
uncertainty.12 More generally, DeGrauwe13 formalizes a positive (negative) impact of FX rate
uncertainty on trade if the exporters’ revenues are convex (concave) in the exchange rate.
A similar ambiguity is derived by Viane and De Vries14 who formally introduce price deter-
mination on forward markets. The latter contributions underscore the nature of markets,
cost and demand functions, and preferences as major factors when determining the effect of
FX volatility on international trade flows. These factors, however, may vary across different
sectors of the economy, thereby questioning the adequacy of empirical models explaining
international trade flows on an aggregated level. Restricting e.g. the income, price and ex-
change rate risk elasticities of trade to be identical across sectors could involve a presumably
large aggregation bias15 which might explain why the empirical literature is inconclusive
about the dominating impact of FX uncertainty on trade. Klein16 conducts an empirical in-
vestigation for disaggregated US bilateral exports to seven major industrialized economies.
Nine categories of traded goods are considered and the case for a sector specific relationship
is powerfully underscored. Therefore, the analysis herein rests on specific growth rates of
multilateral exports and imports for 15 industrialized economies over 10 economic sectors.
Methodologically, the empirical literature proceeded incorporating major advances in
econometric theory as for instance the concept of cointegration17 or the introduction of
autoregressive conditionally heteroskedastic time series processes ((G)ARCH).18,19 However,
two promising directions of empirical work have not been followed yet. Firstly, almost all
empirical contributions, one exception is Baum et al.9, a-priori postulate a linear relationship
between the variables of interest. Among others, Viane and DeVries14 conjecture, that the
true relation may also be nonlinear. Secondly, there is almost no experience with respect to
the performance of typical regression or dynamic models in terms of ex-ante forecasting. One
reason why forecasting exercises have been constantly ignored yet could be that most existing
empirical models characterizing trade patterns fail to pass simple regression diagnostics as,
for instance, tests against serial error correlation.20 In this paper we will provide a detailed
comparison of linear vs. nonlinear model specifications. Furthermore, competing dynamic
models are compared in both directions, in-sample fitting and ex-ante forecasting.
The remainder of the paper is organized as follows: Starting with a brief motivation,
the next section describes the data, variable construction, and the issue of approximating
FX uncertainty which is latent in nature. Section 3 provides the basic methodology. A
vector error correction model (VECM) is outlined and after isolating the partial impact of
FX uncertainty on trade a semiparametric extension is motivated. Diagnostic and selected
estimation results are given. The forecasting exercises are described and interpreted in