This paper explores a new direction for empirical models of exchange rate determination. The motivation arises from two well documented facts, the failure of log-linear empirical exchange rate models of the 1970's and the variability of risk premiums in the forward market. Rational maximizing models of economic behavior imply that changes in the conditional variances of exogenous processes, such as future monetary policies, future government spending, and future rates of income growth, can have a significant effect on risk premiums in the foreign exchange market and can induce conditional volatility of spot exchange rates. I examine theoretically how changes in these exogenous conditional variances affect the level of the current exchange rate, and I attempt to quantify the extent that this channel explains exchange rate volatility using autoregressive conditional heteroscedastic models.