Essays in big data and forecasting methods in financial econometrics

This dissertation comprises two essays on big data and forecasting methods in financial econometrics. Methods for analyzing "big data" have received considerable attention by economists in recent years, given that applied practitioners now have an incredible amount of data available to them, and given that a whole host of new methods have been developed in various disciplines over the last 20 years or so. In the first essay, I discuss some of the latest (and most interesting) methods currently available for analyzing and utilizing big data when the objective is improved prediction. Additionally, I address predictive accuracy testing in the context of big data, and outline new loss function free methods that may be useful for forecast accuracy and model selection assessment. We also provide a brief empirical illustration of big-data in action, in which we show that big data are indeed useful when predicting the term structure of interest rates. This is done in a series of simple prediction experiments where the objective is to predict the term structure of interest rates, and models used include benchmark econometric models, dynamic Nelson Siegel (DNS) models (Diebold and Li, 2007) , diffusion index models (Stock and Watson, 2002), and hybrids of the three. The diffusion indexes in our experiments are estimates of the latent factors from principle component analysis of a macroeconomic dataset including 103 U.S. variables. It is suggested that much remains to be learned regarding the ways in which extant econometric methods can be combined with dimension reduction methods in order to achieve improvements in prediction. In the second essay, an extensive set of forecast experiments is conducted in order to explore the marginal predictive content of latent macroeconomic factors extracted from a so-called "data rich" or real-time dataset in dynamic Nelson-Siegel (DNS) type models. In particular, we assess the following classes of models: DNS type models of the variety, dynamic Nelson Siegel Svensson (NSS) type models (see Svensson (1994)), and various benchmark models, including vector autoregressive (VAR) and autoregressive (AR) models. The macroeconomic factors, or so-called "big data" diffusion indexes that we utilize are extracted using principle component analysis of 130 U.S macro-variables for which McCracken and Ng (2016) have constructed a real-time dataset. Experiments are carried out for various sub-samples between 2001 and 2018, and results are evaluated using a number of benchmark linear models. Additionally, various different dimensions are considered when specifying the yield cross section. Empirical results found are in contrast to the findings of Swanson and Xiong (2017), where including diffusion indexes always yields predictive improvement, although only fully revised macroeconomic data are utilized in that paper. Thus, the usefulness of diffusion indexes appears to hinge on whether or not a data-rich real-time environment is simulated in forecasting experiments or not.Ph.D.Includes bibliographical referencesby Weiqi Xion