Global Portfolio Optimization & Risk Modeling Framework

Customer

Challenge

Building Stable and Realistic Portfolio Optimization

Standard MVO can be unstable with small input changes and integer constraints yielding materially different allocations. Early covariance construction did not ensure positive-semi-definite properties, affecting solver reliability.

• Stability under perturbations: Standard MVO can be unstable; small input changes and integer constraints yielded materially different allocations
• Covariance integrity: Early covariance construction did not ensure positive-semi-definite properties, affecting solver reliability
• Objective clarity: The risk term in the objective required alignment with the intended volatility specification
• Data realism: Continuous series for futures and corporate-action adjustments for equities were needed; trading-hour mismatches and illiquidity pockets distorted naïve correlations

Solution

Comprehensive Risk-First Optimization Framework

A compact Risk & Research squad—Quant Lead, Optimization/Risk Engineer, Data Engineer, Execution Analyst—(1) hardened data, (2) stabilized optimization, and (3) delivered practical integer-aware workflows with clear runbooks.

Data foundation for risk

Built internal continuous contract series aligned to liquid activity and roll schedules. Implemented corporate-action adjustments with cross-checks from multiple public data sources. Standardized to a common bar frequency and introduced correlation alignment for non-overlapping trading hours; removed persistently illiquid instruments after QA.

Stabilized optimization (risk-first)

Rebuilt covariance inputs to ensure PSD properties; aligned the risk term to the intended volatility definition. Added concentration and transaction-cost penalties to reduce knife-edge portfolios and promote allocation stability. Evaluated mixed-integer routes and, where appropriate, used continuous solve + controlled rounding with trade-gap thresholds for realism.

Practical portfolio mechanics

Supported auto-select most-active versus fixed-expiry configurations; exposure checks aggregate across expiries. Introduced a simple config + database layer for adding/removing instruments and backfilling corporate actions.

Technologies & tools

Process

Kickoff & data plan—continuous series + adjusted equities; handle missing overlaps; define common resampling. Covariance & solver hygiene—ensure PSD; align objective to volatility; add stabilizing penalties. Integer path & fallbacks—trial mixed-integer routes; add thresholded rounding if required. Config & roll controls—config/database layer; auto most-active vs specific expiry; expiry aggregation. Review & sign-off—parity checks; validate stability under small input perturbations.

1. Kickoff & data plan
2. Covariance & solver hygiene
3. Integer path & fallbacks
4. Config & roll controls
5. Review & sign-off

Results

• Stable Allocations: PSD covariance with regularization penalties
• Data Integrity: Continuous series with corporate-action adjustments
• Integer-Aware: Practical deployment with controlled rounding

Stabilized allocations via PSD covariance, transaction-cost and concentration penalties, plus an integer-aware deployment path. Data realism through in-house continuous series and adjusted equities with common-horizon panels and cross-market correlation alignment. Operational simplicity via config-driven onboarding and exposure checks aligned with real-world trading practice.