AI Portfolio Optimisation

Idea and Motivation

Pharmaceutical companies operating in Ukraine face a strategic problem: portfolio rebalancing decisions must account for demand volatility, import restrictions, supply chain fragility, and incomplete competitive data—conditions where traditional portfolio theory (mean-variance optimization, Markowitz frameworks) provides limited guidance. Western pharmaceutical companies facing similar constraints typically employ heuristic rules or executive judgment, with limited rigorous frameworks accounting for the actual information available at decision time.

The research question is pragmatic: given the data quality, market observability, and environmental stability actually available to a Ukrainian pharma company, what portfolio decisions can be justified? This is not a question of pure optimization (maximize return subject to constraints) but decision readiness: a graduated scale from “insufficient data to decide” through structured heuristics to statistically-grounded multi-objective optimization, with the method selected based on objective, measurable criteria.

Goal

The series aims to build a complete, theoretically grounded, and practically deployable framework for pharmaceutical portfolio optimization under resource and information constraints. This means not only developing optimization algorithms but constructing the decision-readiness infrastructure around them: a measurement framework (DRI) to assess portfolio state, a method selection system (DRL) to prescribe appropriate techniques, and a production-grade platform (HPF-P) implementing these methods in a user-accessible system.

The goal is to provide pharmaceutical companies, regulators, and supply chain stakeholders with a shared, evidence-based language for portfolio decisions: whether a given rebalancing choice is justified by available data, what confidence bounds apply, and what decision method (heuristic, algorithmic, or ML-augmented) is appropriate at each decision point.

Scope

The series covers six articles across five thematic phases, addressing pharmaceutical SKU groups and optimization strategies:

Focus

The technical focus is on operationalizing decision readiness: developing quantitative measures of portfolio information sufficiency (DRI), defining method-selection rules that map DRI to appropriate optimization strategies (DRL), and building machine learning models for SKU-level demand forecasting that feed into multi-objective optimization. The framework accommodates incomplete data, addresses geopolitical uncertainty through environmental entropy scoring, and integrates domain knowledge (pharmacoeconomics, regulation, supply chain) throughout.

A central innovation is the five-level decision-readiness system applied to five SKU groups and five optimization strategies: rather than asking “how do we optimize this portfolio?” the framework asks “can we justify optimization, and which method?” This inversion shifts from pure performance to defensibility: a DRL-2 heuristic decision may yield lower theoretical returns than DRL-5 ML optimization, but it is the correct choice when data is insufficient for higher-confidence methods.

Limitations

Scientific Value

The series makes four contributions to operational research and pharmaceutical economics. First, it formalizes the concept of decision readiness in portfolio optimization, translating an intuitive notion (can we rely on this data?) into a measurable framework applicable across decision contexts. Second, it provides a structured methodology for pharmaceutical portfolio optimization under realistic constraints, filling a gap between academic optimization theory (which assumes high-quality complete data) and industry practice (which relies on heuristics and judgment). Third, it explicitly addresses environmental entropy—geopolitical, regulatory, and supply chain volatility—as a first-class model parameter, not an afterthought. Fourth, it delivers HPF-P, a production-grade platform implementing these methods, available for use and extension by other researchers and practitioners.

The work demonstrates that decision method selection can be systematized: a given portfolio state maps unambiguously to a recommended method family, with confidence bounds and limitations made explicit. This bridges theory and practice in pharmaceutical economics.

Resources

• HPF-P Portfolio Optimizer Platform→
• HPF-P API Gateway & Documentation→
• Zenodo Research Collection→
• Author ORCID Profile→
• Series DOI: 10.5281/zenodo.18845429→

Status

In Progress. 6 articles published (January–March 2026). Research is part of an ongoing PhD dissertation at Odesa National Polytechnic University (ONPU). HPF-P platform deployed in beta with select Ukrainian pharmaceutical companies. Further articles addressing advanced optimization scenarios and cross-market validation are in preparation.

Contribution Opportunities

Researchers and practitioners wishing to build on this work are encouraged to:

• DRI/DRL calibration: Apply the framework to pharmaceutical markets in other emerging economies. Validate whether DRI thresholds and method-selection boundaries remain stable across geographies and product categories.
• Environmental entropy modelling: Develop more sophisticated models of geopolitical and regulatory risk. Integrate real-time disruption signals (sanctions, tariffs, approval delays) to improve environmental entropy scoring.
• Demand forecasting: Build domain-specific forecasting models (ML + domain knowledge) for pharmaceutical demand under supply constraints. Integrate epidemiological data and policy signals.
• Multi-period portfolio dynamics: Extend the framework beyond 12-month horizons; model multi-period rebalancing decisions with transaction costs and regulatory constraints.
• HPF-P extension: Deploy the platform with additional pharmaceutical companies; collect case studies demonstrating decision readiness improvements and portfolio performance gains.
• Cross-sector application: Apply the DRI/DRL framework to other supply-constrained portfolio problems: medical device portfolios, agrochemical distribution, energy commodity trading.