Abstract
This paper develops a dynamic financial Ising model on heterogeneous investor networks to connect social-media information, investor interactions, and stock returns. We provide explicit economic interpretations for the core physical variables of the Ising model and, based on utility maximization and random utility theory, derive the Bounded-rationality Logit (BRL) distribution as the financial counterpart of the Boltzmann distribution. The model is implemented on five investor-network structures: a two-dimensional lattice, an ER network, a WS network, a BA network, and a hybrid WSBA network. A time-varying external information field is calibrated using stock-level Douyin Index data for 2,382 Chinese A-share stocks over 1,698 days. The simulated returns reproduce key stylized facts of real financial markets, including volatility clustering, persistent absolute-return dependence, and leptokurtic distributions. Among the five networks, the WSBA network provides the closest overall match to the return characteristics of major Chinese equity indices. The simulations further reveal a positive return response to marginal changes in information, and this effect is stronger for smaller-cap stocks. Lagged Fama–MacBeth regressions, within-day standardization tests, and an instrumental-variable analysis based on non-trading-day information confirm this positive response and its market-capitalization heterogeneity. These findings provide a microfounded econophysics framework linking social-network structure and time-varying information to asset-price dynamics.