Complex Network Built From Stock Price Returns and Volumes to Predict Market Volatility and Volume

Abstract

This study investigates if network features from stock return and trading volume correlations can improve one-month-ahead forecasts of Vietnam’s VNIndex volatility and volume (2018–2024). We construct dynamic financial networks using Threshold, Top-k, and minimum spanning tree (MST) filtering methods, calculating metrics like density, centrality, and clustering.

Using these features in linear regression and random forest models, we find that threshold-based networks yield the strongest volatility predictions (R² ≈ 0.56). Volume forecasts achieve very high accuracy (R² ≈ 0.95), reflecting strong underlying correlations. Notably, surges in network density and centrality often precede periods of heightened market volatility.

Our findings demonstrate that incorporating complex network measures derived from mixed return-volume correlations can meaningfully enhance market forecasts in an emerging market context.