Transfer Learning for Cross-Market Inference: Approaches in Emerging and Transitional Economies

Emerging and transitional economies face significant challenges in financial modeling due to limited access to high-quality labeled market data. This paper presents a novel cross-market transfer learning framework designed to address this limitation by leveraging knowledge from data-rich developed markets. The proposed architecture integrates domain-adversarial training and instance re-weighting via Kernel Mean Matching (KMM) to learn domain-invariant yet economically relevant representations. Through extensive experiments across target markets including India and Brazil, the model demonstrates superior performance in forecasting tasks, outperforming baseline and standard fine-tuning approaches in terms of RMSE and R² score. Furthermore, the approach enhances generalization, reduces data dependency, and supports scalable deployment in data-constrained financial environments. This work highlights the viability of transfer learning as a foundation for equitable and scalable AI adoption in global economic systems.