A machine learning framework developed at Parul University demonstrated significantly improved portfolio performance during market stress periods when backtested on historical data spanning 2017 through 2022, including the 2020 COVID-19 market disruption. The research, published in Scientific Reports, introduces a layered approach to portfolio optimization that combines predictive modeling with transparency mechanisms that institutional investors require for regulatory compliance. Why Did Traditional Portfolio Models Fail During the 2020 Crisis? For decades, portfolio management has relied on a fundamental assumption: the relationships between different assets remain relatively stable over time. This works fine in calm markets. But during crises, that assumption collapses entirely. When COVID-19 hit in early 2020, asset correlations spiked simultaneously, volatility exploded, and classical models could not keep pace with the speed and severity of the selloff. Risk parity strategies that major institutional investors depended on failed to protect portfolios because they were designed to react to changes, not anticipate them. Dr. Sanjay Agal, Professor and Head of the Department of Artificial Intelligence and Data Science at Parul University, started with a direct question: Could a system be built that detects early signals of structural market change and repositions the portfolio before the worst damage materializes? Not a system that waits for a crash to happen and then adjusts, but one that adapts to regime shifts before they fully develop. How Does the AI Framework Actually Work? The framework operates through five interconnected layers that work together to monitor markets and make allocation decisions. The system was tested on out-of-sample data spanning 2017 through 2022, a period that included some of the most dramatic market conditions of the past two decades. The results were striking: the framework achieved a Sharpe ratio of 1.38, a metric that measures return per unit of risk. Against traditional risk parity strategies, the framework demonstrated a 55% improvement. What makes this system different from other AI trading models is its emphasis on explainability. Finance is one of the industries most skeptical of artificial intelligence. Regulators, compliance teams, and investment committees need to understand why a model recommends what it recommends. A black box is not acceptable in front of an audit committee. Steps to Implementing Explainable AI in Financial Systems - SHAP-Based Risk Attribution: The framework uses SHAP (SHapley Additive Explanations) analysis to allow risk managers to look inside any decision and see which factors drove the allocation, creating a transparent audit trail for compliance purposes. - Signal Monitoring in Stable Markets: In calm market conditions, the model relies on momentum factors and yield-curve signals, consistent with how experienced human portfolio managers approach steady environments. - Stress Signal Detection: When stress signals appear, the model shifts attention to the VIX (volatility index) and liquidity indicators, exactly what a seasoned risk manager would prioritize during market turbulence. - Real-Time Decision Interrogation: Every decision point can be interrogated and audited, allowing institutional investors to understand the reasoning behind portfolio adjustments before they are executed. The SHAP analysis revealed something genuinely reassuring: the framework was not just producing good numbers. It was learning the logic of financial markets. In stable markets, it relied on momentum factors and yield-curve signals. When stress signals appeared, it shifted attention to the VIX and liquidity indicators. This interpretability is what makes the difference between a research paper and a deployable system that institutions will actually adopt. "The framework was tested on data from 2017 through 2022, a period covering some of the most dramatic market conditions of the past two decades," explained Dr. Sanjay Agal. Dr. Sanjay Agal, Professor and Head of the Department of Artificial Intelligence and Data Science at Parul University What Makes Publication in Scientific Reports Significant? Scientific Reports is published by Springer Nature, the same organization that publishes Nature, widely considered the most prestigious scientific journal in existence. The journal is ranked Q1, the highest possible quartile for academic journals. With over 834,000 citations in 2024, it is the third most-cited journal in the world, and its 2024 impact factor stands at 3.9. Getting past Scientific Reports' peer review process requires work that meets international standards of scientific rigor, novelty, and methodological soundness. The fact that a research paper from Parul University's Department of AI and Data Science cleared this threshold places the department's research output alongside work from institutions that have been producing top-tier research for decades. The paper is open access, meaning anyone in the world can read it. The co-authors are Krishna Raulji and Niyati Dhirubhai Odedra. The DOI is 10.1038/s41598-025-26337-x. What Research Pipeline Is Coming Next? The research pipeline at Parul University is being strengthened by a robust set of manuscripts currently under peer review and editorial evaluation in leading high-impact journals. These ongoing works reflect sustained scholarly momentum and demonstrate active engagement with globally recognized publication platforms. The diversity of topics, coupled with submissions to prestigious IEEE Transactions and Springer Nature journals, highlights both the depth and translational relevance of the research contributions. The 2026 research pipeline is strongly anchored in high-impact Q1 journals, including premier IEEE Transactions and Springer Nature journals. Current submissions under peer review and editorial evaluation span advanced domains including multimodal generative AI, explainable natural language processing (NLP), financial intelligence, privacy-preserving machine learning, and intelligent smart systems. Recent submissions to IEEE Transactions further strengthen the research profile, demonstrating contributions in socially aware AI systems and hybrid NLP architectures. The overall pipeline reflects a strategic focus on scalable, explainable, and real-world AI solutions, with a strong presence in top-tier journals, indicating high potential for impactful publications in the near term. For institutional investors and risk managers, the key takeaway is clear: AI systems that can explain their reasoning are no longer theoretical. They are being published in the world's most rigorous journals and represent a meaningful advance in how financial institutions can manage portfolio risk during market stress.
