The bias of forward exchange rates as a predictor of future spot rates is typically explained or decomposed as (1) a risk premium and (2) a convexity term which accounts for the fact that, when there is stochastic inflation, nominal gains from forward currency speculation are higher than real ones and correspondingly losses are smaller. We use Nalebuff's envelope puzzle to explain a third source of bias which involves real profits from foreign exchange speculation. Both the "real profit" bias and stochastic inflation bias arise from the convexity of g(s)=1/s and so derive from Jensen's inequality as observed by Siegel (1972).