Estimating the Leverage Effect Using High Frequency Data

This paper investigates the dynamics of the leverage effect over time, using high frequency data. By applying Realized Kernel techniques, a more precise estimate of Realized Correlation ¨C compared to standard subsampled estimators of Realized Correlation ¨C is derived. This new measure avoids the so-called Epps effect and permits to observe a level of Realized Correlation significantly different from zero ¨C unlike standard subsampled estimators. Modeling the resulting measure, a deeper insight into the dynamic behavior of Realized Correlation ¨C and hence into the dynamics of leverage effects ¨C is obtained. In particular, this paper studies the behavior of Realized Correlation across the recent financial crisis, to gain a deeper understanding of the factors underlying leverage effects at different stages of the crisis.