We study how subjects extrapolate simple patterns in financial time series in order to develop a descriptive model of actual agent behavior. The laboratory experiment for this analysis was conducted in Germany and Japan. Statistical analyses indicate considerable similarity in expectations formation across cultures and document that agents' expectations are at variance with the notion of standard trend extrapolation. The paper then proposes a method for computing expectations for any economic time series based on the experimental data. Such pattern-based expectations are shown to explain stock prices and the dynamics of the forward discount on the foreign exchange market.