Transfer Learning for Cross-Market Predictions: Applications in Emerging and Volatile Economies
Purna Chander Mashetty, Sirish Gangabathula, Naga Venkatesh Gangabathula, Neeraja Pullalarevu, Koushik Reddy Chaganti, Sathvik Reddy Chaganti
Abstract
The procedure of financial forecasting faces tough challenges in growing unstable markets because these regimes bring a shortage of accessible data and unstable markets held back by multiple of translating systems. Such environments are hard for deep learning and traditional models because they require large amounts of high-quality training data to work correctly. The proposed method creates a sophisticated transfer learning framework that employs fixed models pretrained in well-established markets which it then adapts to developing countries using domain adaptation techniques coupled with the process of fine-tuning the models. The proposed scheme accomplishes data distribution minimizing functions from Maximum Mean Discrepancy, in addition to supervised fine-tuning that targets realizing predictions under the suppressed target information. The model suggested above scored better performance by experiment testing across emerging market indices in terms of all metric metrics including mean absolute error, root mean squared error and direction accuracy of compared models. Transfer learning is found to perform well in solving generalization issues in cross-market and hence can be considered as a great method for financial prediction in data-scarce volatile environments.