Abstract:The time-varying covariance matrix of the financial assets is the key for financial applications such as portfolio allocation and risk management.Previous studies on covariance matrix forecasting use daily or even lower frequency data,causing the problems of parameter estimation difficulty and curse of dimensionality.With intraday high-frequency data,non-parametric estimators of the covariance matrix can be constructed.This turns the covariance matrix from hidden to an observable variable that can be directly modeled,thus reducing the complexity of covariance model estimation and increasing the applicability of covariance models in high-dimension applications.Furthermore,with highfrequency data,cojumps can be identified,which refers to jumps of multiple asset prices in the same intraday sampling interval.Cojumps are often triggered by macroeconomic news announcements and policy releases,and such macro-information will eventu ally be absorbed and reflected in the covariance matrix.Thus,we argue that cojumps may contain information beneficial for covariance forecasting and propose to identify cojumps and use them in the covariance forecasting models.The multivariate heterogeneous autoregressive(MHAR) model is used as the benchmark model for the nonparametric covari ance matrix estimator.The cojump indicators and the cojump intensities estimated by the Hawkes model are included as additional predictors,first separately and then simultaneously.Based on the mean squared error and the mean absolute error criteria,the three extended MHAR models are each compared with the benchmark using the Diebold Mariano test in terms of their out-of-sample forecast performance.The model confidence set test is then used to identify the best models.Besides,the out-of-sample fore casts are used in the global minimum variance portfolio strategy to justify the economic value of the extended models.We consider five high liquidity stocks from different sectors of the SSE 50 index and employ their five-minute prices.Empir ical results show that:l①compared with the cojump indicators,the cojump intensities have more significant contribution to covariance matrix forecasting;②incorporating the cojump intensities can significantly improve the fit and forecast performance of the MHAR model;③the extended MHAR model that uses matrix log transformation to ensure positive definiteness and incorporates both the cojump intensities and the cojump indicators is superior,both statistically and economically.The above results confirm the role of cojumps in forecasting covariance matrices in Chinese stock markets,as well as reveal the contribution of macro-information to covariance forecasting.This study provides practical guidance for financial managers and investors in their financial risk management and asset allocation practices.

Stock co-jump networks

A New Dynamic Community Model For Social Networks

Network Analysis to Uncover Stock Comovement from a Chinese Financial Portal.

Dynamics of Chinese Stock Market from a Complex Network Perspective

Detecting Top-k Active Inter-Community Jumpers in Dynamic Information Networks

SCOREH+: A High-Order Node Proximity Spectral Clustering on Ratios-of-Eigenvectors Algorithm for Community Detection

Count Models of Social Networks in Finance

Jump Risk and Cross Section of Stock Returns: Evidence from China’s Stock Market

Multi-Likelihood Methods for Developing Stock Relationship Networks Using Financial Big Data

Restricted Tweedie Stochastic Block Models

Jump Clustering, Information Flows, and Stock Price Efficiency

Social Network-Based Stock Correlation Analysis and Prediction.

Heterogeneous Dual-Dynamic Attention Network for Modeling Mutual Interplay of Stocks

Stock network inference: A framework for market analysis from topology perspective

A multivariate GARCH–jump mixture model

Detection of Mutual Exciting Structure in Stock Price Trend Dynamics

Inferences on Mixing Probabilities and Ranking in Mixed-Membership Models

The Role of Cojumps in Forecasting Covariance Matrices in Chinese Stock Markets:A Study Based on the Multivariate HAR Model

Analysis of Chinese Stock Market from a Complex Network Perspective: Better to Invest in the Central

Patterns and Portfolios based on Stock Multiplex Networks

Regularized Stochastic Block Model for robust community detection in complex networks