Triage a portfolio without turning judgment into a score

Portfolio review becomes theatre when every program is scored independently and the committee negotiates the final list. The central question is collective: which set of programs deserves the next unit of scarce capital?

Establish eligibility first

An asset can be attractive and still be ineligible. A source-backed safety veto, failed mechanism, infeasible study or unaffordable cash requirement belongs in a gate, not in the denominator of a score. ARiDA separates these conditions from ranking.

For eligible assets, the valuation spine calculates stage-gated rNPV and can add correlated Monte Carlo analysis, downside probability, real-option value and value of information. Reference-class probabilities are an outside view, not a replacement for asset evidence. Clinical rates vary by indication, phase and biomarker design (Wong, Siah and Lo), so the chosen cohort must remain visible.

Model shared scientific exposure

Ten programs do not create ten independent bets. Assets can share a target, pathway, modality, biomarker, vendor or chemical series. ARiDA can use molecular similarity as one explicit correlation prior. RDKit’s Morgan fingerprints provide a reproducible representation for molecular comparison (RDKit).

The science branch is conservative about the leap from similarity to finance. It requires source-backed coverage for every asset, aligns the matrix to the portfolio roster, constrains entries to non-negative values and repairs invalid matrices before use. If coverage is partial, the model does nothing. The accepted matrix remains an assumption that a reviewer can replace.

Once correlation enters covariance, the efficient frontier, diversification measures and capital allocation can move. A portfolio that looked balanced on asset count may reveal a concentrated scientific bet.

Allocate under real constraints

ARiDA’s portfolio engine supports efficient-frontier analysis, hierarchical risk parity and integer selection under a budget. The optimization can include concentration limits and resource constraints. It reports the current portfolio, feasible alternatives and the preferred allocation under the stated objective.

This does not automate the committee’s judgment. It makes the trade-off inspectable. Management can see whether an asset was excluded because of cash, capacity, correlation or poor risk-adjusted value.

Put learning into the portfolio

Holding an asset is defensible only when the team knows what it is waiting for. EVSI can estimate the expected value of a proposed study before its results exist (methods guide). A hold decision can therefore carry an experiment, cost, decision threshold and reopening date.

Advance, hold and stop are actions with conditions. The portfolio becomes manageable when each one preserves the evidence state and the reason capital moved.