Abstract:
Studies on intraday conditional correlation dynamics is limited and existing literature mostly depends on classical methodologies that are prone to errors due to unaddressed issues like non-synchronous trading and market microstructure noise. Trades are not homogenously scattered along the day. Hence, upon close inspection of data in high frequency domain such as one-second-long intervals, one sees intermittent and irregular observations. In contrast to its methodological counterparts, combination of Generalized Autoregressive Score (GAS) framework and State-Space Modelling produces reliable results on such data structures by guaranteeing full data usage. Findings of this study reveal that average intraday conditional correlation rises as trading commences, lingers around certain altitude for some time before an upward trend closes out the trading day, which we attribute to the US market opening. Visual inspection of the findings across different market conditions and days of the week reveals elevated correlation levels in volatile markets as well as a distinguishable path for both ends of a week. A closer inspection of findings via Dynamic Time Warping exhibits that intraday conditional correlation patterns are discernibly different for Mondays, Tuesdays and Fridays. Beyond the scholarly contribution, the methodology and findings are of interest to various parties like high-frequency traders, risk and portfolio managers and regulatory agencies in formulating their high frequency trading practices, margin requirements and portfolio construction schemes.