Monte Carlo Simulation for Biotech Valuation

Every biotech deal valuation carries uncertainty. The question is whether you see it. Risk-adjusted NPV produces a single number that compresses enormous complexity into false precision. Monte Carlo simulation replaces that number with a probability distribution, showing not just the expected outcome but the full range of what could happen and how likely each scenario is. For BD teams negotiating term sheets and boards approving deal commitments, that difference is the difference between a guess and a decision framework.

Across 1,600+ transactions in the Solidus database, actual deal values fall within Monte Carlo 80% confidence intervals 73% of the time, compared to just 41% accuracy for single-point rNPV estimates. This guide explains why, shows how Monte Carlo works in a pharma context, and walks through a practical example you can replicate using our free simulator.

Why Single-Point rNPV Is Not Enough

Risk-adjusted NPV is the industry standard for biotech valuation, and for good reason: it explicitly accounts for clinical attrition risk through probability-of-success adjustments. But rNPV has a fundamental limitation. It takes point estimates for every input -- a single PoS rate, a single peak sales figure, a single discount rate -- and produces a single output. That output implies a precision that does not exist in drug development.

Consider the inputs to a typical Phase 2 oncology asset valuation:

• Probability of success: Your base estimate might be 32%, but reasonable analysts could justify anywhere from 22% to 45% depending on how they weight biomarker data, trial design, and historical comparators. That range alone creates a 2x difference in expected value.
• Peak sales: Consensus might be $1.2B, but the realistic range spans $400M (competitive entry erodes share) to $2.5B (label expansion, pricing power). A 6x range in a single input.
• Time to market: Your model assumes 5.5 years, but clinical holds, enrollment delays, or accelerated approval could shift this by 18-36 months in either direction, materially affecting discounted value.
• Royalty and milestone terms: The structure itself introduces uncertainty -- tiered royalties, anti-stacking offsets, and sales-dependent milestones all create nonlinear payoff profiles.

When you multiply these uncertainties together, the true range of possible deal values spans an order of magnitude or more. A single rNPV number hides that reality. Monte Carlo exposes it.

How Monte Carlo Works in a Pharma Context

Monte Carlo simulation is conceptually simple: instead of calculating value once with fixed inputs, calculate it thousands of times with randomly sampled inputs. Each iteration draws a different combination of values from the probability distributions you define for each uncertain variable, computes the resulting deal value, and records the outcome. After 10,000 iterations, you have a distribution of 10,000 possible deal values that reflects the combined uncertainty of all inputs.

The three core components of a pharma Monte Carlo model are:

1. Input Probability Distributions

Each uncertain input is defined not as a single number but as a distribution reflecting the range of plausible values and their relative likelihood:

• Probability of success (beta distribution): PoS is bounded between 0% and 100%, making the beta distribution a natural fit. For a Phase 2 oncology asset, you might specify a beta distribution with mean 32% and 90% confidence interval of 20-45%. The distribution captures the reality that PoS is uncertain, not the false certainty of saying "32%."
• Peak sales (log-normal distribution): Revenue projections are positively skewed -- there is a floor near zero but no ceiling. A log-normal distribution with median $1.2B and standard deviation $600M captures both the most likely outcome and the long tail of blockbuster scenarios.
• Development timeline (triangular or PERT distribution): Timelines have a minimum (fastest possible), most likely, and maximum duration. A PERT distribution with minimum 4 years, mode 5.5 years, and maximum 8 years reflects the asymmetric risk of delays versus acceleration.
• Market share and pricing (normal or uniform distributions): These inputs typically have less extreme ranges and can be modeled with symmetric distributions, though competitive scenarios may warrant more complex shapes.

2. Correlation Modeling

Real-world variables do not move independently. Monte Carlo models can encode correlations between inputs that a simple sensitivity analysis misses:

• Efficacy and PoS: An asset with stronger-than-expected efficacy data simultaneously has higher probability of success and higher peak sales potential. Modeling these as independent understates both the upside and the correlation between good outcomes.
• Timeline and safety: Clinical holds and safety signals that delay timelines also reduce PoS. Negative correlation between timeline delay and success probability captures this real-world linkage.
• Pricing and competition: Higher competitive entry reduces both market share and pricing power simultaneously. Correlated sampling prevents the unrealistic scenario of low competition paired with low pricing.

3. Output Distribution and Interpretation

The output of 10,000 iterations is a probability distribution of deal values. The key statistics that inform decision-making are:

The shape of the distribution matters as much as the percentiles. A symmetric distribution suggests balanced risk. A right-skewed distribution (common in biotech) means the mean exceeds the median, implying significant upside optionality. A bimodal distribution suggests the asset is likely to be either a significant success or a near-total loss, with few outcomes in between -- typical for binary clinical readouts.

When to Use Monte Carlo vs. rNPV vs. DCF

Each valuation method has a role. The choice depends on the asset, the decision context, and the audience.

The best practice is to run all three and triangulate. Use rNPV as the quick screen, Monte Carlo as the decision tool, and DCF for post-approval commercial projections. Our calculator produces all three perspectives in a single analysis, while the simulator lets you explore Monte Carlo distributions interactively.

Practical Example: Phase 2 Oncology Asset

To make this concrete, here is what a Monte Carlo output looks like for a typical Phase 2 oncology licensing deal. The asset is a bispecific antibody targeting a validated mechanism in non-small cell lung cancer (NSCLC), with early Phase 2 efficacy data showing a 35% objective response rate.

Input Distributions

Output: Total Deal Value Distribution (10,000 Scenarios)

Illustrative example based on Solidus engine output for a Phase 2 bispecific antibody in NSCLC. Actual results vary by specific asset parameters. rNPV comparison uses the same point estimates at distribution means.

Notice the key insight: the rNPV of $920M falls near the Monte Carlo median of $940M, as expected -- they use the same central assumptions. But the Monte Carlo reveals that the P25-P75 range spans $620M to $1.5B, a 2.4x spread. For a licensor, this means the asset could realistically command anywhere from $620M to $1.5B in total deal value. For a licensee, it means a deal at $940M has roughly a 50% chance of outperforming expectations and a 50% chance of underperforming.

This range directly informs deal structure. If you are the licensor, you want the upfront to protect you at the P25 level ($75M) while using milestones and royalties to capture the P75-P90 upside. If you are the licensee, you want the guaranteed payments (upfront) anchored to the P25-P50 range, with higher payments only triggered if the bull case materializes.

What 1,600+ Transactions Reveal About Uncertainty

When we back-test Monte Carlo models against actual deal outcomes in the Solidus database, three patterns emerge consistently:

• Monte Carlo confidence intervals are well-calibrated: Actual deal values fall within the 80% confidence interval (P10-P90) 73% of the time. The slight undershoot from 80% reflects extreme outcomes that even broad distributions underestimate -- acquisitions at massive strategic premiums and deals that collapsed due to unforeseen safety signals.
• Single-point rNPV over-predicts 59% of the time: Because rNPV uses mean inputs and deal outcomes are right-skewed, the mean rNPV exceeds the actual deal value in the majority of cases. Monte Carlo's median (P50) is a better predictor of the actual outcome than rNPV's expected value.
• Phase 2 assets have the widest distributions: The P25-P75 range for Phase 2 assets averages 3.1x, compared to 2.0x for Phase 3 and 1.6x for approved products. This confirms intuition: earlier-stage assets carry more uncertainty, and Monte Carlo is most valuable precisely where that uncertainty is greatest.

These findings are not academic. They directly affect how you should price risk. A deal team using single-point rNPV is systematically overvaluing most assets and undervaluing the information contained in the distribution of possible outcomes.

Common Mistakes in Biotech Monte Carlo Models

Monte Carlo is powerful, but it is not magic. The quality of the output depends entirely on the quality of the inputs and model structure. The most common errors we see:

• Treating all inputs as independent: Ignoring correlations between PoS, peak sales, and timeline produces distributions that are too narrow. In reality, good outcomes tend to cluster (high efficacy drives higher PoS, higher sales, and faster timelines simultaneously), and so do bad outcomes.
• Using normal distributions for inherently skewed variables: Peak sales cannot be negative, and probability of success is bounded between 0% and 100%. Using normal distributions for these variables produces impossible scenarios (negative sales, PoS above 100%) that distort the output.
• Over-fitting distributions to small datasets: If you have Phase 2 data from 30 patients, you do not have enough information to specify a narrow distribution for peak sales. Wider distributions honestly reflect greater uncertainty -- resist the temptation to appear precise.
• Ignoring the zero-value scenario: In biotech, there is always a meaningful probability that the asset fails entirely and the deal value is zero (or near-zero). Models that treat PoS as continuous but never produce a "total failure" scenario understate risk.
• Presenting the mean instead of the median: For right-skewed distributions (most biotech valuations), the mean exceeds the median and is pulled up by blockbuster scenarios. The median is a more honest "expected" outcome for decision-making.

Our Solidus simulator addresses these issues by default: it uses appropriate distribution types for each variable, models key correlations, includes the binary success/failure gate, and reports both mean and median outputs.

Running Your Own Simulation

You can run a Monte Carlo simulation for any biotech asset in under two minutes using the Solidus platform:

1. Open the simulator and select your therapeutic area and indication from 562 options across 12 TAs.
2. Set your asset parameters: clinical phase, modality, deal type, and any relevant designations (breakthrough, orphan, fast track).
3. Review pre-filled distributions: The simulator auto-populates input distributions based on your selections, calibrated against 1,600+ comparable transactions. Override any parameter to match your specific asset.
4. Run the simulation: 10,000 iterations execute in seconds. View the probability distribution, percentile table, and sensitivity tornado.
5. Export results: Download the distribution as a PDF report or share interactive results with your team for deal committee review.

For full deal benchmarking with all 14 engines (including Monte Carlo, rNPV, real options, competitive dynamics, and buyer-specific valuation), use our calculator. The calculator integrates Monte Carlo output with deal term benchmarks and comparable transaction analysis for a complete valuation package.