Yingying Li (Chair)

Abstract

This talk concerns principal component analysis (PCA) and singular value decomposition (SVD) in high dimensional high frequency data. We show that these analyses can reduce to a simpler form with the help of contiguity, thereby making it possible to conduct a deeper exploration of estimators. As an example of application, we focus on SVD based portfolios of financial data. Such portfolios offer the possibility to take the index concept further. They are also a class of implementable and publicly disclosable financial algorithms, which allow the lean cats (the customers) to keep more of their own funds. We show that contiguity can help to obtain hard-to-reach quantities such as how long to learn from SVD/PCA before unleashing estimated singular vectors into a trading strategy. (With work with Lan Zhang at University of Illinois Chicago)