Predicting BRICS NIFTY50 returns using XAI and S.A.F.E AI Lens

Indranil Ghosh¹Tamal Datta Chaudhuri²Golnoosh Babaei^3*Paolo Giudici^4*Emanuela Raffinetti³

• ¹IT & Analytics Area, Institute of Management Technology Hyderabad, Hyderabad-501218, Telangana, India
• ²Bengal Economic Association, Kolkata, India
• ³Universita di Pavia, Pavia, Italy
• ⁴University of Pavia, Pavia, Italy

Purpose – Global fund managers, in their effort towards risk diversification and generating higher returns, design portfolios that consist of financial assets of various countries. In the process, they expose their investors not only to the fundamentals of the assets but also to transnational volatility, macroeconomic shocks of different countries, and exchange rate fluctuations. These factors make forecasting returns from such global funds quite difficult and, at the same time, challenging. To aid global fund managers and investors, this paper presents a forecasting framework for predicting returns from Goldman Sachs BRICs Nifty 50 Developed Markets Index (BRICS NIFTY 50), which is a traded and listed financial asset. It is a global portfolio, which not only exposes investors to the fundamentals of different companies but also to country risk. Design/methodology/approach – Gradient Boosting Regression (GBR) and SHAP-based XAI are used to identify the top significant country-specific explanatory variables. Subsequently, with the selected variables, GBR, CatBoost, Light Gradient Boosting Machine (LGBM), Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Extra Tree Regressor (ETR) are applied for forecasting returns from BRICS NIFTY 50. Along with standard evaluation tools, the S.A.F.E AI framework is used for measuring predictive accuracy, sustainability, and contribution of each predictor. To evaluate the relative efficacy of the six predictive models, the underlying research resorts to a multi-criteria decision-making (MCDM) framework. Findings – We find that country-specific market volatility, industrial performance, financial sector development, and exchange rate fluctuations explain global returns significantly. Further, the exercise also reveals that explanatory factors specific to India, China, and Brazil emerge to be relatively important. Research limitations/implications – The paper focuses on a single index. Future work will extend it to other indices and global funds. Practical implications – The proposed methodology will be of practical use to global fund managers and investors. Policymakers may find it useful for identifying factors that make foreign direct investment and portfolio investment attractive. Originality/value – Development of a two-step forecasting framework, identifying effects of country-specific explanatory variables, and applying different evaluation criteria to measure predictive efficiency underscore the novelty of the work.