The Predictive Power of Behavioral Economics in the Foreign Exchange Market

This dissertation explores the out-of-sample forecastability of changes in exchange rates using behavioral economics and combination methods and contributes to the literature by introducing three approaches presented in three essays. The first essay explores a new approach through behavioral heuristics to forecast changes in exchange rates. One key aspect of behavioral economics is that people do not use the realized distribution of historical data to predict variables. Instead, agents assign subjective probabilities which depend on heuristics in decision-making. These subjective probabilities are used as weights for the ten years of observations prior to time t in linear models of changes in exchange rates to estimate coefficients and form forecasts. This essay develops forecasting models for exchange rates using monthly data for the US dollar versus 37 (advanced and emerging/developing) currencies. The second essay incorporates behavioral economics by adding investor sentiment variables to macroeconomic models to form forecasts for exchange rates. In addition, I examine the predictive ability of a terms of trade index (in changes) both as a single predictor and as an added predictor in the uncovered interest rate parity model. Finally, I examine changes in commodity and oil prices (nominal and real) as predictors of changes in exchange rates. Some models provide promising results for some currencies using Pesaran-Timmermann (PT) statistics in both essays. In the first essay, models under the assumptions of anchoring-toward and optimism perform well. In the second essay, macroeconomic models augmented by behavioral factors outperform the White Noise (WN) model for several countries. The third essay of the dissertation addresses model uncertainty. It focuses on combining forecasts from various individual models introduced in the second essay. I use standard (e.g., equal weights) and new forecast combination approaches to form combined forecasts. Using a regularization technique (Ridge regression), I apply linear and convex combination estimates to combine forecasts. I propose Directional Prediction as a new weighting approach. The results indicate that for several countries, the linear combination (Ridge regression) weighting approach has proportions of correct direction of changes in exchange rates greater than 0.5, which means it performs better than the WN model.