[downloadable]

Summary

• I test the exchange rate implications of the term structure model Cochrane and Piazzesi (2008) in its ability to explain key anomalous features of the data.
• I find predictions that are at odds with our understanding of the data, particularly the large predictable component in exchange rates implied by the model. I attribute this to ex-post overfitting; the estimation procedure emphasizes the fit of the conditional mean of the pricing kernel. The estimation procedure also relies on the assumption of conditionally homoskedastic disturbances.
• Moreover, the additional information in the term structure of interest rates used to characterize the predictable component of bond excess returns generates an implausible amount of predictability in exchange rates. I show that this result prohibits linear factor models from jointly accounting for the stochastic properties of bond and currency returns under our current understanding of these financial markets; modifications are necessary.
• Admitting conditionally heteroskedastic disturbances is one way to uncouple the direct link between risk and expected returns in bond and currency markets. This choice, however, reduces the amount of exploitable predictability in these financial markets, not to mention their usefulness for forecasting in real-time.
• Additional attempts to uncouple the direct link between risk and return in bond markets are also outlined. These alternative avenues leaves open the possibility that the size of the statistically measured predictable component in bond and currency excess returns are attributed to systematic forecast errors that arise due to the inability of agents to disentangle transitory from longer-lived disturbances.
• The class of linear models are not specific to finance; (log-linear) economic models with time-invariant parameters are not immune to this critical limitation.

Abstract

• I use the term structure model in Cochrane and Piazzesi (2008) and construct currency market prices. The implied currency market prices are counterfactually volatile and predictable, at least with respect to commonly used predictor variables. Getting the model closer to currency market data means reducing bond risk compensation but doing so nearly eliminates predictability in bond markets. One way to generate sensible time-variation in predictable bond and currency excess returns allows the volatility of returns to be time-varying. Additional avenues that can jointly account for the stochastic properties of bond and currency excess returns are also discussed.

JEL Classification: E43, E47, F31, F37, G12, G11, G15

Keywords: forward premium anomaly; return predictability; term structure models; time-varying volatility