Decomposing Realized Variance: A Point Process of Relevant Price Changes with Long Memory in Volatility

Abstract

A simple, empirical-based approach to decompose Realized Variance (RV) is proposed, with supportive theoretical argument and empirical evidence. Under the proposed framework, RV is interpreted as a product of the intensity and variance of relevant price changes. Holding the variance aspect constant, statistical inference on the event intensity is conducted, with the spot intensity described by Hawkes process. Empirical analysis on Spyder returns in 2008 confirms strong performance of the model. An attempt to track down the source of long memory is merited by the simple decomposition of RV, and serves as starting point for future research.