Split-Session Cluster GARCH for Overnight and Intraday Returns: The Role of Tail Heterogeneity

Battle Count: 90

Highly relevant for quantitative trading: provides improved dynamic correlation forecasts for portfolio construction (GMV portfolios), captures critical overnight-intraday return decomposition that affects trading strategies, models tail heterogeneity important for risk management and position sizing. The 100-asset application demonstrates scalability for institutional portfolio management. Score-driven updates are computationally efficient for real-time implementation. However, the model requires daily open/close data and sector classifications, and the two-stage estimation adds complexity to live trading systems.

High complexity: requires implementing Coupled EGARCH (asset-by-asset), score-driven correlation dynamics with matrix logarithm parameterization, convolution-t likelihood with cluster-specific degrees of freedom, canonical block representation for dimensionality reduction, and diagonal Fisher information approximation. The closed-form scores and information matrices are provided but involve Kronecker products and commutation matrices. Block structure reduces computational burden but the overall pipeline is non-trivial. No public code repository available.

Data available from CRSP (commercial database). Empirical code available from authors upon reasonable request. Model specifications are fully described with closed-form scores and Fisher information matrices. However, no public repository is linked, and CRSP access requires institutional subscription.