Dynamic Financial Time Series Modeling with Rapid Adaptation via Meta Learning

The Meta learning framework for financial time series forecasting is proposed here, which can adapt to new financial market datasets efficiently with minimal retraining. This methodology includes two phases: the first one is cross-financial time series training on a meta-model and the second adapts this meta-model using recent data to novel market conditions. This approach of dynamic feature augmentation enhances model adaptability across a wide range of financial instruments. The performance results show that the adjusted model has higher accuracy and flexibility compared to the baseline methods. In summary, the contributions of this paper offer a robust flexible model for use in forecasting outcomes for various financial instruments and address several challenges in traditional machine learning models when applied in complex and dynamic financial markets. The proposed framework achieves a 67.33% directional accuracy (DA) in the stock market, significantly outperforming baseline models with a DA of 43.82%.