Pavuluri Sushhma Seshu

Model Development & Validation with Top U.S. Banks
Regulatory Frameworks: OCC, BSA/AML, OFAC, Basel III, SR 11-7, CCAR

• Developed classification models using statistical, machine learning, and rule-based techniques to monitor transaction activity, detect model performance degradation, and reduce both false negative and false positive rates.
• Collaborated with stakeholders—including model validators, audit teams, and regulators—to create and execute MRMG finding remediation plans, ensuring compliance and timely issue resolution.
• Conducted backtesting and stress testing on Autoregressive Error Correction Models, using accuracy metrics such as MAPE and MAE to assess performance under varied scenarios.
• Estimated Value at Risk (VaR) for non-linear portfolios using Cornish-Fisher expansions (Delta-Gamma Approximation), optimizing for both accuracy and computational efficiency.
• Represented the model development team in discussions with MRMG, effectively communicating model logic, outcomes, and key insights through intuitive presentations and Excel-based visualizations.
• Designed and implemented a model tuning framework (ATL/BTL analysis) for over 22 transaction and product monitoring models, aligning with BSA/OFAC regulations and OCC model risk guidelines.
• Drafted and enhanced model documentation and Model Implementation Documents (MIDs), conducted User Acceptance Testing (UAT) for production releases, and improved Ongoing Performance Monitoring (OPM) metrics (e.g., AUC) in line with model governance standards.
• Partnered with business analysts and IT teams to implement robust data controls, ensuring clean data pipelines, reconciling segmentation logic, managing missing data, and maintaining data integrity for model inputs.
• Automated BAU and production processes, increasing efficiency, reducing manual errors, and supporting model lifecycle governance.

Developed an end-to-end Python-based framework to construct a dynamic stock portfolio outperforming both the SENSEX and a traditional Markowitz portfolio.

• Built a Minimum Spanning Tree (MST) using Kruskal’s algorithm to visualize and analyze correlation structures among stocks. Selected stocks with the lowest pairwise correlations to enhance diversification.
• Applied reinforcement learning (Actor-Critic algorithm leveraging Bellman Optimality Equations) to dynamically allocate portfolio weights over a one-year investment horizon.
• Estimated historical Value at Risk (VaR) and Conditional Value at Risk (CVaR) using Extreme Value Theory (EVT) by fitting the Generalized Pareto Distribution (GPD) to tail losses, effectively capturing liquidity and tail risks.
• Benchmarked performance against the SENSEX, Markowitz portfolio, and equally weighted portfolios.
• The resulting strategy, incorporating coherent risk measures and dynamic weighting, significantly outperformed traditional approaches in terms of risk-adjusted returns.