Learning from Experience in the Foreign Exchange Market

Description

Several stylized facts have been established regarding currency excess returns. First, the uncovered interest rate parity (UIP) condition is strongly violated, meaning that currencies with higher than average interest rates in the cross-section do not depreciate, but rather appreciate (Hansen and Hodrick 1980; Fama 1984). Second, average excess returns on a basket of currencies display predictably countercyclical variation (Lustig et al. 2014). This proposed project is the first to show that both stylized facts can be explained with a learning model, which also generates novel empirical predictions that can be tested using currency pricing and survey data. In the model, the representative agent faces parameter uncertainty and needs to estimate the unconditional mean of consumption growth rates at the country level in real time. As a result, her beliefs about future exchange rates differ from those of an econometrician who analyzes the data ex post, and who knows the true parameter value. The specific learning process used in the learning model is microfounded by the “learning from experience” behavior documented in Malmendier and Nagel (2016), and further developed in Nagel and Xu (2020). In this model, the agent requires a constant risk premium on currencies, and the UIP holds under her subjective expectations. However, her learning process generates persistent expectation errors, and this leads the econometrician to find both predictable variations in realized excess returns and UIP violations ex post. The model also suggests a new predictor of currency excess returns. That predictor will be the difference between the exponentially weighted average of past consumption growth rates in the foreign country and that in the home country, which can be constructed with the estimated learning parameter from survey data. I will test these predictions using professional forecasts on foreign exchange rates from multiple sources. I will also build a structural asset pricing model that quantitatively matches both the magnitude of predictability in currency returns and the empirical patterns found in survey forecasts of exchange rates. Overall, this project aims to deepen our understanding of why currency returns vary over time by studying the dynamics of subjective expectations implied by survey data.