It has been surprisingly difficult to demonstrate non-zero currency exposures for individual stocks, and a recent study even finds that the absolute value of estimated exposure is a better predictor than the value itself. We argue that the value of the international-trade option
should be convex in the exchange rate so that exposure depends on the exchange rate level.
Since spot rates move slowly, exposure could then differ substantially across samples. Many large companies, in addition, must be ambidextrous|positively exposed in some activities, and negatively in others. We derive a class of tractable regression models and find that the standard loglinear regression is invariably beaten by the proposed alternative. However, too often we detect at least a partial concavity.