CMStatistics 2020: Start Registration
View Submission - CFE
Title: Realized principal component analysis: A pre-averaging approach Authors:  Francesco Benvenuti - Aarhus University (Denmark) [presenting]
Kim Christensen - Aarhus University (Denmark)
Bezirgen Veliyev - Aarhus University (Denmark)
Abstract: A Realized Principal Component Analysis (RPCA) theory robust to noise is proposed, using a pre-averaging approach. The RPCA is the high-frequency extension of the classic PCA: in this setting noise always contaminates price observations. Hence, we first obtain a consistent estimator of the spot covariance matrix employing the well-known pre-averaging technique. Then, we derive the realized eigenvalue, eigenvectors and principal component estimators for this case, building on a recent theory about volatility functional estimation. We conclude by providing simulation results in different noise scenarios, and we discuss how those estimators work in practice.