Essays on high frequency data, jumps, and forecasting

This dissertation comprises two essays on financial economics and econometrics. The first essay reviews methodology associated with the construction of nonparametric estimators of integrated volatility, jump tests, and realized volatility decompositions. In an empirical analysis that draws on this methodology, we separate continuous asset return variation and finite activity jump variation from U.S. excess returns on U.S. market sector exchange traded funds (ETFs) during and around the Great Recession of 2008. Our objective is to characterize the financial contagion that was present during one of the greatest financial crises in U.S. history. In particular, we study how shocks, as measured by jumps, propagate through nine different market sectors. One element of our analysis involves the investigation of causal linkages associated with jumps (via the analysis of vector autoregressions), and another involves the examination of the predictive content of jumps for excess returns. We find that as early as 2006, jump spillover effects became more pronounced in the markets. Another important findings that we see is that jumps have a significant effect on excess returns during 2008 and 2009. Thus, jumps play an important role in asset pricing during volatile episodes. In the second essay, we utilize measures of jumps in the markets in order to construct daily indexes of unexpected jump spillover risk associated with major news announcements and events. The methodology that we implement is based on two novel new tools recently developed in the financial econometric and machine learning literatures. First, we implement the jump decomposition methods detailed in A¨ıt-Sahalia and Jacod (2012) in order to decompose quadratic variation into continuous components and jump components; and we further separate large and small jump variations. We then carry out shrinkage via application of ridge, elastic net (EN), least absolute shrinkage operator (LASSO), and a cross validated convex combination ridge, EN and LASSO methods in order to quantify (Granger) causal jump spillover effects across sectors and markets, and construct risk indexes. In an empirical analysis illustrating the methodology proposed for constructing jump based risk indexes, we analyze equally spaced 5-minute high frequency trading data on nine market sector ETFs as well as the S&P500 and the VIX for the period 2005 - 2010.In summary, we believe the indexes proposed in this paper are usefully condensed indicators of the risk associated with unexpected events in the markets, and should be of interest to market participants interested in hedging such risk.Ph.D.Includes bibliographical referencesby Jessica Schlossber