This paper merges two branches of the literature. On one hand we study a heterogeneous agents framework to model exchange rates and stock prices. On the other hand we model the relationship between these two series through a DSGE model. Investors choose one of two rules to form their expectations. One rule is based on an open economy model, which reacts to the information from the financial markets. The second rule follows a backward looking approach. We find that when DSGE agents misinterpret the information coming from the financial markets as exogenous productivity shocks they unknowingly amplify the volatility of these markets. The simulated series replicate the stylized facts of real data. We also estimate the DSGE and chartists expectations, and we find that our DSGE agents make output forecasts that are not qualitatively different than the DSGE forecasts from the recent Bayesian literature.