This dissertation consists of four papers that investigate long-standing issues in international finance, macroeconomics and international macroeconomics. The common theme of the papers is the fact that each uses novel statistical methods to forecast or evaluate outcomes in international macroeconomics and macroeconomics.

The first two papers focus on the predictability of nominal exchange rates and the profitability of foreign exchange speculation. The first chapter of my dissertation investigates the predictability of nominal exchange rates. Prediction of the exchange rate has long interested economic researchers. A model able to accurately forecast exchange rate movements would have important implications for economic theory and market participants alike. However, it is a stylized fact that empirical models of the exchange rate cannot produce forecasts that are more accurate than those of a random walk—the well-known Meese-Rogoff puzzle. In this chapter, I apply the gradient boosting method due to Hastie, Tibshirani & Friedman (2000) to the problem of forecasting exchange rate movements for nine major currencies relative to the U.S. dollar. The method performs model selection and builds flexible, potentially non-linear models of the exchange rate. I find that the economic fundamentals I consider do contain predictive power at both short and long horizons. The key to successfully forecasting exchange rates is building the 'correct' forecasting model, because I find that model with the greatest explanatory power varies across currencies and across time for any individual currency. At short horizons the method I use is unable to select ex ante a model of the exchange rate that convincingly outperforms a random walk. However, when forecasting 12 months ahead, boosted models produce extremely accurate forecasts of the exchange rate, easily besting the random walk null.

Chapter two, written with Óscar Jordà and Alan M. Taylor, explores the profitability of currency speculation. We document the actions of a hypothetical trader who trades based on forecasts of nominal exchange rates coming from standard forecasting models. Although the efficient markets hypothesis implies that such a trade would yield zero profits on average, these actions would have delivered economically meaningful profits to the trader. That currency carry trades are on average profitable is well-known. The literature accounts for these profits by noting that realized returns exhibit conditional negative skewness, or by presuming that the profits compensate for some other risk. Yet when the trader uses our preferred model of the exchange rate, the ensuing profits exhibit little to no conditional skewness. Intriguingly, we also find that the profits are not compensation for risk, as the excess returns from the carry trade do not covary in a meaningful way with a broad set of conventional risk factors.

The final two papers of my thesis analyze the business cycle. The third chapter of my dissertation, written with Óscar Jordà, investigates the classification of economic activity in the U.S. into expansions and recessions. In the U.S., the Business Cycle Dating Committee (BCDC) of the NBER determines the peaks and troughs of the business cycle. But because there is no universally accepted definition of "recession," the true state of the economy is fundamentally unobservable and the problem of classifying economic data into the two phases of the business cycle is not a trivial undertaking. This chapter evaluates the classification skill of the BCDC relative to several state-of-the-art statistical methodologies designed to describe the unobservable state of the economy. The methods we use are novel to the economics literature and have the advantage of being completely non-parametric. In the final portion of the paper, we introduce a forecasting model to predict future states of the business cycle.

The final chapter of my dissertation, written with Fushing Hsieh, Shu-Chun Chen and Óscar Jordà, concerns business cycles outside of the U.S. The existence of an apolitical, unbiased committee dedicated to determining business cycle peaks and troughs is a luxury that American economists take for granted—only two other economies in the world have such a committee. In most countries, recessions are determined either by the popular press, who can use only rough rules-of-thumb, or by the government, so that such decisions are fraught with political influence. We classify international data into recessions and expansions for 22 OECD economies using a novel classification algorithm. The algorithm is completely non-parametric, a feature that is necessary to complete the analysis because the state-space of the problem makes standard classification methods unfeasible. The algorithm operates by recognizing intervals where economic activity is particularly intense, in the sense that the recurrence time between observations of extreme events (e.g., negative GDP growth) is shorter than would otherwise be expected. We apply the algorithm to monthly country-level data on industrial production, employment and GDP. The resulting binomial series are aggregated in order to produce novel chronologies of country-specific, regional, and global recessions.