Using Consensus Economics survey data on experts' expectations, we aim to model the 3- and 12-month ahead ex-ante risk premia on the Yen/USD and the British Pound/USD exchange markets. For each market and at a given horizon, we show that the risk premium is well determined by the conditional expected variance of the change in the real exchange rate, agents' real net market position in assets and a constant composite risk aversion coefficient, as suggested by a two-country portfolio asset pricing model. The expected variance depends on the past values of the observed variance and the unobservable real net market position is estimated as a state variable using the Kalman filter methodology. We found that the trends of our estimated horizon-specific net market positions are consistent with the ones of the observed short term aggregate net market positions calculated using the U.S. Treasury International Capital System dataset. Moreover, we show that the ex-post premia tend to adjust towards the ex-ante values, suggesting that experts' beliefs provide a relevant information to the market. These results bring new responses to the difficulties reported by the widespread ex-post risk premium literature and enhances the usefulness of survey data in modelling the risk premium.