The paper proposes DisFT-GNN, a novel distillation-based framework for stock trend prediction that addresses the limitation of existing GNN methods which only rely on historical spatiotemporal dependencies. The key innovation is a teacher model that learns to capture correlations between historical and future data distributions using an attention-based multi-channel fusion mechanism. This teacher generates future-aware spatiotemporal representations, which are then used as intermediate supervision to train a student model via HSIC-based distillation loss. The student model learns to infer future patterns solely from historical data, enabling more accurate predictions without future information leakage during inference. Extensive experiments on S&P 100 and NASDAQ 100 datasets demonstrate that DisFT-GNN significantly outperforms seven state-of-the-art GNN baselines, achieving up to 5.41% improvement in accuracy and substantial gains in MCC. Back-testing simulations further confirm the practical profitability of the approach. The framework is model-agnostic and can be integrated with any GNN-based stock predictor, making it a versatile enhancement for financial forecasting.