Decomposition-based AI model enhances Indian stock price forecasting amid macroeconomic shocks
image: It demonstrates the need for conducting research for periods separately, as each has its own macroeconomic characteristics. This data segregation assists in evaluating predictability in normal and turbulent market states. Studies using very long time periods miss out on specific characteristics, as shown in the picture.
Credit: Indranil Ghosh (Institute of Management Technology Hyderabad, India) Tamal Datta Chaudhuri (Bengal Economic Association, India) Sunita Sarkar (Assam University, India) Somnath Mukhopadhyay (Assam University, India) Anol Roy (Assam University, India)
Background and Motivation
Stock markets play a critical role in wealth creation and economic growth, yet their movements are influenced by a complex mix of macroeconomic shocks, market uncertainty, and speculative behaviour. Understanding these dynamics is essential for investors, firms, and policymakers, especially in emerging economies like India, which is highly integrated into global financial systems and sensitive to external and domestic shocks. Traditional forecasting models often fail to disentangle the effects of different factors across varying time horizons. This study introduces a novel decomposition-based framework to predict Indian stock prices by separately analysing short-, medium-, and long-term components, offering a more nuanced approach to market forecasting.
Methodology and Scope
The research employs a hybrid analytical framework combining Ensemble Empirical Mode Decomposition (EEMD) and Fuzzy-C-Means (FCM) clustering to break down stock price series into granular components. The model uses Multiverse Optimisation (MVO) to integrate predictions from three advanced algorithms: Extreme Gradient Boosting Regression (XGBR), Facebook Prophet, and Support Vector Regression (SVR). To ensure interpretability, the study applies Explainable AI (XAI) through SHAP analysis to identify the contribution of each explanatory variable. The framework is tested on nine Indian stocks, spanning large-, mid-, and small-cap categories, across four distinct market phases, including periods of economic stability, the COVID-19 pandemic, and the Russia-Ukraine conflict.
Key Findings and Contributions
- Multi-Scale Predictability: Stock price components are influenced by different factors: long-term trends are driven by macroeconomic indicators and global market sentiment; medium-term movements are linked to volatility measures such as VIX and commodity prices; short-term fluctuations are best explained by technical indicators like RSI and MACD.
- Robust Performance: The proposed framework demonstrates strong predictive accuracy across normal and turbulent market periods, including during the COVID-19 crisis and geopolitical conflicts, with large-cap stocks showing higher predictability.
- Model Superiority: The integrated decomposition-ensemble approach outperforms standalone machine learning and traditional econometric models such as ARIMA and SARIMA.
- Interpretable Insights: SHAP-based analysis clarifies the relative importance of features across time scales, providing actionable insights into what drives stock prices under different market conditions.
Why It Matters
In an era marked by economic uncertainty, geopolitical tensions, and speculative trading, accurate stock price forecasting is more critical than ever. This study not only advances methodological innovation in financial analytics but also provides an adaptable tool that remains reliable during black swan events. By separating the effects of fundamentals, volatility, and speculation, the model enhances transparency and supports more informed decision-making in volatile markets.
Practical Applications
- Investors & Traders: Can use the model to refine entry/exit timing, portfolio allocation, and risk management strategies, especially for large-cap stocks.
- Fund Managers & Advisors: May leverage the multi-scale insights to develop differentiated investment products and improve client reporting.
- Financial Institutions: Could integrate the framework into algorithmic trading systems and robo-advisory platforms.
- Policy Makers: May apply the approach to monitor market stability, assess the impact of macroeconomic policies, and design interventions to curb excessive volatility.
- Academic & Research Bodies: Can extend the methodology to other markets, asset classes, or incorporate alternative data sources such as social media sentiment or ESG metrics.
Discover high-quality academic insights in finance from this article published in China Finance Review International. Click the DOI below to read the full-text! Open access for a limited time!