Analysis of Contagion in China's Stock Market: A Hawkes Process Perspective

Battle Count: 78

The paper provides valuable insights into trend persistence, oversold rebounds, and sector rotation dynamics in Chinese markets. The identification of self-exciting patterns (trend continuation) and inhibitory effects (sector rotation) could inform momentum and mean-reversion strategies. However, the paper lacks practical trading signal generation, backtesting, and transaction cost analysis. The daily frequency and limited sector coverage reduce immediate applicability for high-frequency trading. The theoretical framework is more suited for risk management and market understanding than direct alpha generation.

The Hawkes process framework is mathematically well-defined with clear parameter estimation procedures. The univariate and bivariate cases are straightforward using existing libraries (Python tick module). However, the 6-dimensional nonlinear Hawkes process with ReLU link functions and Monte Carlo gradient estimation requires careful implementation. The stochastic gradient descent algorithm with backpropagation adds complexity. Grid search for hyperparameters (omega) and proper initialization are important practical considerations. Overall moderate complexity for someone familiar with point processes and optimization.

The paper provides detailed mathematical formulations, algorithm pseudocode (Algorithm 1), parameter estimates, and mentions using Python's tick module. However, no code repository is provided, and the specific data sources (e.g., Wind, Tushare) are not explicitly stated. The experimental setup (85/15 train/validation split, grid search for omega) is described but full implementation details are limited.