---

Author: Ivchenko, Oleh Affiliation: Odessa National Polytechnic University Series: AI Portfolio Optimisation Year: 2025

Abstract

Pharmaceutical portfolio management operates at the intersection of scientific uncertainty, regulatory complexity, and market volatility. Traditional optimization approaches assume a stable, well-characterized information environment — an assumption that routinely fails in practice, particularly in emerging markets subject to geopolitical disruption. This paper introduces the Holistic Portfolio Framework (HPF), a decision-theoretic architecture that conditions optimization strategy selection on the measurable sufficiency of available information. HPF introduces two novel constructs: the Decision Readiness Index (DRI), a composite diagnostic score measuring information quality across five dimensions, and Decision Readiness Levels (DRL), a five-tier action taxonomy that prescribes appropriate optimization strategies calibrated to DRI scores. We demonstrate that HPF reduces decision errors in low-information regimes by preventing premature optimization, while preserving the ability to apply sophisticated AI-driven methods when readiness conditions are met. Empirical motivation is drawn from the Ukrainian pharmaceutical market (2020–2025), where compounding disruptions provide a natural experimental context for evaluating decision-readiness frameworks under stress.

1. Introduction

The pharmaceutical industry allocates billions of dollars annually to portfolio decisions: which products to develop, stock, promote, and discontinue. These decisions are complex by nature — they involve demand forecasting under regulatory constraints, competitive dynamics, supply chain vulnerabilities, and long product lifecycles. Standard operations research approaches (linear programming, multi-criteria optimization, Markowitz-style risk models) exist and are widely applied. However, they share a critical assumption: that the data feeding the model is sufficiently reliable to justify the optimization.

In practice, this assumption is violated frequently. A demand forecast built on six months of disrupted sales data is not equivalent to one built on five years of stable history. A regulatory compliance score derived from a market where enforcement has been suspended is not equivalent to one from a mature, stable jurisdiction. Treating these cases as equivalent — by applying the same optimization algorithm regardless of data quality — leads to what we term overconfident optimization: confident-looking outputs from unreliable inputs.

This paper introduces the Holistic Portfolio Framework (HPF) as a systematic response to this problem. HPF is not a replacement for existing optimization methods; rather, it is a meta-framework that governs when and how those methods should be applied, based on a principled assessment of information sufficiency.

The framework emerged from doctoral research in economic cybernetics at Odessa National Polytechnic University (ONPU), motivated by extensive observation of pharmaceutical portfolio management failures in the Ukrainian market — a market that experienced, in rapid succession, the COVID-19 pandemic, supply chain disruptions, and full-scale war beginning in 2022. These cascading shocks created a natural laboratory for studying how portfolio decision quality degrades as information quality deteriorates.

2. Problem Statement

2.1 The Information Sufficiency Gap

Portfolio optimization algorithms require inputs: demand forecasts, price elasticities, cost structures, risk parameters, regulatory compliance scores. When these inputs are derived from data that is incomplete, stale, or structurally unstable, the optimization output inherits these deficiencies — often invisibly. A portfolio manager who runs a multi-objective optimizer on compromised inputs may receive a precise-looking recommendation that is, in practice, no better than an informed guess.

We define the information sufficiency gap as the difference between the information quality required by a given optimization method and the information quality actually available. When this gap is large, sophisticated optimization methods are counterproductive: they add analytical complexity without adding decision value.

2.2 The Ukrainian Pharmaceutical Market as Motivating Context

Ukraine’s pharmaceutical market presents an extreme case of information sufficiency gap dynamics. With approximately 12,000 registered pharmaceutical products and a domestic industry producing roughly 70% of consumed medicines by volume, the market is large and structurally complex. The period 2020–2025 subjected this market to:

• COVID-19 demand shocks (2020–2021): Unprecedented demand spikes for specific therapeutic categories, supply chain disruptions, and regulatory changes (emergency use authorizations, fast-track approvals).
• Full-scale military conflict (2022–present): Territorial disruption, infrastructure damage, population displacement, and international supply route interruptions. These events simultaneously degraded all five dimensions of decision-relevant information, rendering standard optimization approaches unreliable for significant portions of the portfolio.

A major Ukrainian pharmaceutical company managing a portfolio of 800+ SKUs across multiple therapeutic categories faced exactly this problem: their existing optimization tools, calibrated to pre-war conditions, continued producing recommendations — but those recommendations were increasingly disconnected from operational reality. HPF was developed in part to address this specific failure mode.

3. The Holistic Portfolio Framework

3.1 Core Architecture

HPF consists of three integrated components:

1. DRI Computation Engine: Assesses the quality of available information across five dimensions, producing a composite Decision Readiness Index score for each portfolio segment.

1. DRL Assignment: Maps DRI scores to one of five Decision Readiness Levels, each associated with an appropriate optimization strategy.

1. Strategy Execution Layer: Implements the prescribed optimization strategy for each segment, ranging from portfolio abstention (DRL-1) to full multi-objective AI optimization (DRL-5).

This architecture enforces a discipline that existing approaches lack: before any optimization occurs, the system asks whether optimization is warranted, and if so, which type is appropriate given current information conditions.

3.2 The Decision Readiness Index (DRI)

The DRI is a weighted composite of five sub-indices:

• R1 — Data Completeness: Proportion of required data fields populated with values meeting quality thresholds. Missing values, obvious outliers, and structurally anomalous entries degrade R1.
• R2 — Demand Signal Quality: Strength and stability of the demand signal, measured through signal-to-noise ratio, forecast horizon reliability, and absence of structural breaks.
• R3 — Risk Observability: Degree to which supply, competitive, and regulatory risks can be quantified from available data. Unobservable risks degrade portfolio models silently.
• R4 — Regulatory Clarity: Stability and predictability of the regulatory environment for each portfolio segment. Pending approvals, enforcement suspensions, and regulatory uncertainty reduce R4.
• R5 — Temporal Stability: Consistency of the data-generating process over time. Structural breaks, trend reversals, and environmental shocks degrade R5 and reduce the predictive validity of historical patterns.

The composite DRI is computed as:

DRI = w1·R1 + w2·R2 + w3·R3 + w4·R4 + w5·R5

where weights w1–w5 reflect the relative importance of each dimension for the specific portfolio context. Default weights are equal (0.2 each), with domain-specific calibration available.

3.3 Decision Readiness Levels (DRL)

DRI scores map to five DRL tiers:

DRL	DRI Range	Strategy
DRL-1	0.00–0.20	Abstention: hold current allocations, no optimization
DRL-2	0.20–0.40	Proportional rebalancing: simple rule-based adjustments
DRL-3	0.40–0.60	Linear Programming: constrained optimization on cleaned data
DRL-4	0.60–0.80	CVaR optimization: risk-aware portfolio adjustment
DRL-5	0.80–1.00	Multi-objective AI: full Pareto-front optimization

This taxonomy reflects a principled escalation: as information quality improves, progressively more sophisticated methods become justified.

4. Theoretical Foundations

HPF draws on several theoretical traditions:

Information economics: The value of information in decision-making is well-established (Hirshleifer & Riley, 1992). HPF operationalises this insight at the portfolio level by making information quality an explicit input to strategy selection.

Robust optimization: The robust optimization literature (Ben-Tal et al., 2009) addresses decision-making under uncertainty but typically assumes that uncertainty sets are known. HPF addresses the prior problem: assessing whether available data is sufficient to characterize uncertainty sets reliably.

Satisficing and bounded rationality: Simon’s (1955) satisficing framework suggests that decision-makers should choose strategies appropriate to their information environment. DRL operationalises this principle formally.

Systems entropy: The concept of environmental entropy — the degree of unpredictability in a system’s state — provides a theoretical basis for DRI dimension R5. High-entropy environments degrade decision-relevant information faster than stable environments.

5. Contribution and Scope

HPF makes the following original contributions:

1. DRI as a formal construct: The five-dimensional DRI provides a systematic, reproducible method for assessing information sufficiency in portfolio contexts. Unlike informal data quality assessments, DRI produces quantitative scores that can be tracked, compared, and used programmatically.

1. DRL as action taxonomy: The five-tier DRL framework provides clear, actionable prescriptions for each readiness level. Decision-makers no longer need to exercise personal judgment about when sophisticated optimization is warranted.

1. Integration with existing methods: HPF does not replace LP, CVaR, or multi-objective optimization — it governs their appropriate use. Existing algorithmic infrastructure can be reused within the HPF architecture.

1. Empirical grounding: The framework is developed and validated in the context of the Ukrainian pharmaceutical market, providing empirical grounding in an extreme-case environment.

6. Conclusion

This paper has introduced the Holistic Portfolio Framework (HPF) as a systematic approach to decision-readiness in pharmaceutical portfolio management. By conditioning optimization strategy selection on the measurable quality of available information — through the DRI and DRL constructs — HPF addresses a fundamental failure mode of existing approaches: the application of sophisticated optimization methods to data that cannot support them.

Subsequent papers in this series provide detailed technical specifications for each HPF component: DRI dimension definitions and formulas (Article 2), DRL strategy specifications (Article 3), empirical validation in the Ukrainian market context (Article 4), and technical implementation in the HPF-P platform (Article 5).

References

• Ben-Tal, A., El Ghaoui, L., & Nemirovski, A. (2009). Robust Optimization. Princeton University Press.
• Hirshleifer, J., & Riley, J. G. (1992). The Analytics of Uncertainty and Information. Cambridge University Press.
• Simon, H. A. (1955). A behavioral model of rational choice. Quarterly Journal of Economics, 69(1), 99–118.
• Markowitz, H. (1952). Portfolio selection. Journal of Finance, 7(1), 77–91.
• Rockafellar, R. T., & Uryasev, S. (2000). Optimization of conditional value-at-risk. Journal of Risk, 2(3), 21–41.