Herding, Momentum, and Reversal in China's A-Share Market: An Agent-Based Network Model with Information Diffusion

Battle Count: 58

The paper is highly relevant to quantitative trading in several ways: (1) It provides a mechanistic understanding of momentum and reversal through herding and information diffusion, which are key factors in momentum strategies. (2) The novel tail-based herding indicator (HerdTail) offers a high-frequency complement to CSAD and LSV for detecting correlated trading. (3) The trading-rule illustration shows potential alpha from combining herding signals with price trends. (4) The network structure insights inform understanding of how information and behavior propagate in retail-dominated markets like China's A-shares. However, the trading rule is presented as illustrative only, lacking rigorous backtesting, and the model is primarily a mechanism-identification tool rather than a direct trading signal generator.

The agent-based model requires implementing: (1) heterogeneous Gaussian belief initialization, (2) logistic buy/sell/hold probability functions, (3) network-based herding with convex combination updates, (4) finite-speed information diffusion with mixture belief updating, (5) market clearing via root-finding of aggregate demand, (6) multiple network topologies (lattice, random, small-world), and (7) empirical Johnson SU transformation with rolling windows. The mathematical framework is well-specified but the numerical implementation of market clearing (finding the closest root to previous price) and the interaction between diffusion and herding require careful coding. The empirical component requires access to tick-level or daily A-share data with buy/sell classification.

The paper provides detailed parameter tables (Table 1) and mathematical formulations for all model components. However, no code repository is mentioned, the empirical data pipeline is not fully documented, and the simulation study references an 'original simulation study' whose details are not fully reproduced here. The trading rule illustration lacks transaction costs, turnover analysis, and out-of-sample validation. The Johnson SU transformation fitting procedure within rolling windows is described but implementation details are sparse.