This study proposes a novel loss function for neural network models that explores the topological structure of stock realized volatility (RV) data by adding Wasserstein Distance (WD). The study shows that the proposed loss statistically significantly improves the forecast accuracy of neural...