Robust Transformer-Based One-Step Stock Index Forecasting via Shifted Data Augmentation
Researchers have developed a new Transformer-based architecture for one-step stock index forecasting, addressing challenges like noisy signals and distributional shifts in financial time series. The proposed framework incorporates advanced learning-rate scheduling, specifically cosine annealing with warmup, and a novel Shifted Data Augmentation (SDA) technique. Experiments on the VN30 and S&P 500 datasets showed that SDA significantly reduces forecasting errors and improves robustness, outperforming increased model complexity. AI