This research develops a hybrid Time-Varying Vector Autoregressive (TV-VAR) and Neural Network (NN) model to forecast Indonesian stock prices using a two-stage sequential approach. First, the TV-VAR model is applied to capture the primary dynamic relationships and time-varying structural changes between stocks. Second, a Long Short-Term Memory (LSTM) network is employed to model the nonlinear residuals that the TV-VAR model fails to explain. We analyzed four leading stocks (BBCA, BREN, AMMN, and BYAN) using daily closing prices from January to December 2024. The results show that the hybrid TV-VAR–NN model significantly outperforms the standalone TV-VAR model, achieving lower MAPE, MAE, and MSE values. Substantively, the analysis reveals two distinct behavioral clusters. First, BBCA (Financials) and BYAN (Energy) demonstrated price movements primarily driven by internal factors and sector-specific fundamentals. Second, BREN (Infrastructure/Energy) and AMMN (Basic Materials) were found to be significantly more reactive and sensitive to cross-stock dynamics, indicating greater market integration and vulnerability to systemic shocks. This research not only sheds light on the specific market structure of Indonesia’s leading equities but also validates the application of advanced hybrid models for a more robust analysis of financial time-series data.