Build a biotech valuation case that exposes its science

A biotech valuation can be mathematically correct and still mislead. The usual problem is not the discount formula. It is the treatment of scientific uncertainty as a single, unexplained probability.

Identify the object being valued

Start with the exact asset revision, indication, development stage, valuation date and rights. Separate observed facts, reference-class benchmarks and management assumptions. A probability of success without a cohort, source date and adjustment rationale is not an assumption a serious reviewer can challenge.

Clinical transition rates vary by therapeutic area, phase and biomarker strategy (Wong, Siah and Lo). Human genetic evidence also has a measurable relationship with clinical success, with substantial contextual variation (Minikel et al.). The model should preserve those differences rather than produce a universal “industry PoS.”

ARiDA builds an outside-view benchmark first, then permits governed scientific evidence to modify eligible benchmark stages. Molecular druggability adjustments are conservative and source-bound. Analyst-entered stages are not silently rewritten. Each stage retains its base value and the scientific modifier that changed it.

Value the distribution

The governed valuation engine runs a correlated Monte Carlo rNPV using a validated covariance structure. It reports P10, P50, P90 and the probability that rNPV is non-positive. A bimodality warning prevents the mean from masquerading as a likely outcome when the distribution has two distinct regimes.

The commercial layer remains glass-box. Population, penetration, price, duration, launch timing and cost assumptions can be inspected separately. Scenario comparison shows which narrative produces which number. A tornado identifies sensitivity, but sensitivity is not importance: a parameter may move value greatly and still be impossible to learn or influence.

Recognize managerial choices

Development is staged. Management may abandon, pause, partner or expand after new information arrives. ARiDA estimates abandonment-option value through a binomial real-options model alongside ordinary rNPV. It also calculates EVPI, EVPPI and EVSI. EVSI asks what a study of a specified design is expected to add by reducing decision uncertainty (methods guide).

These outputs support practical choices: fund the assay package, wait for a competitor readout, license before the next trial, or stop because even perfect information cannot rescue the economics.

Show the assumptions that make the number true

The released valuation should include its reference class, scientific adjustments, commercial build, correlation assumptions, distribution, options, information value and unresolved gaps. ARiDA binds the result to an assumption set so a reviewer can vary a parameter and recompute through the same engine.

The purpose is not to produce a more impressive valuation. It is to make the terms of belief explicit enough for capital to be allocated responsibly.