The Problem with Optimism Bias in Biotech Projections: A Quantitative and Probabilistic Analysis

Optimism bias, the cognitive tendency to overestimate the likelihood of favorable outcomes and underestimate risks, is a pervasive issue in biotech projections. It distorts critical aspects of strategy, such as clinical success rates, revenue forecasts, and market penetration models. In an industry governed by uncertainty and fat-tailed risks, this bias contributes to inflated valuations, misguided capital allocation, and eventual underperformance.

This article examines optimism bias from a quantitative perspective, incorporating probabilistic models and frameworks informed by the work of Mandelbrot and Taleb to propose methods for mitigating this systemic issue.

Optimism Bias in Biotech Projections: Key Areas

Optimism bias manifests across three primary domains in biotech:

1. Overestimated Clinical Success Rates

The likelihood of a drug successfully progressing through all clinical phases to market approval is often inflated in projections. The actual probabilities for each stage, based on historical data, are significantly lower than what is frequently assumed in pitch decks.

The probability of success (POS) can be expressed as:

$$
POS = P_{discovery} \cdot P_{preclinical} \cdot P_{phase\_I} \cdot P_{phase\_II} \cdot P_{phase\_III} \cdot P_{approval} \cdot P_{market}
$$

Where:

• Pdiscovery: Probability of a successful drug candidate discovery.
• Ppreclinical: Probability of preclinical validation.
• Pphase I,Pphase II,Pphase III​: Probabilities of success at each clinical phase.
• Papproval​: Probability of regulatory approval.
• Pmarket​: Probability of achieving market success.

Using average historical data:

• Pphase I≈0.63
• Pphase II≈0.31
• Pphase III≈0.58
• Papproval≈0.85

This results in an aggregate POS often below 10%, far lower than the values often cited in fundraising materials.

2. Revenue Overestimation

Revenue forecasts in biotech are frequently inflated by overestimating market share and adoption rates, while ignoring competitive dynamics and pricing pressures. A typical revenue model follows:

$$
R(t) = TAM \cdot M_s(t) \cdot P_{adopt}(t)
$$

Where:

• TAM: Total addressable market size.
• Ms(t): Market share as a function of time.
• Padopt(t): Probability of product adoption over time.

Optimism bias inflates Ms(t) and Padopt(t), leading to projections that assume unrealistically rapid adoption curves or negligible competition. For example, claiming a 20% market share in a competitive oncology field without accounting for pricing or clinical superiority is statistically improbable.

3. Ignoring Fat-Tailed Risks

Most biotech projections implicitly assume Gaussian (normal) distributions for outcomes, which underestimate the likelihood of extreme events. However, clinical and market outcomes often follow fat-tailed distributions, where outlier events—such as a blockbuster success or catastrophic failure—dominate probabilities.

The revenue distribution, for instance, is better modeled with a Pareto distribution:

$$
P(R > x) = \left( \frac{x_m}{x} \right)^\alpha
$$

Where:

• x(m): Minimum revenue threshold.
• α: Tail exponent, indicating the thickness of the tail.

A lower α (e.g., α<2) reflects a heavy-tailed distribution, highlighting the significant probability of extreme values. Most models, by assuming normality, fail to capture this dynamic.

Correcting Optimism Bias: Quantitative Strategies

1. Probabilistic Ranges for Clinical Success

Instead of single-point estimates, companies should present ranges of probabilities based on Bayesian updates. For instance:

$$
P_{posterior} = \frac{P(D | H) \cdot P_{prior}}{P(D)}
$$

Where:

• Pposterior: Updated probability of success given new data (DDD).
• P(D∣H): Likelihood of observing DDD under hypothesis HHH.
• Pprior​: Initial probability of success.
• P(D): Marginal probability of data DDD.

Updating probabilities as clinical data emerges creates more realistic, dynamic forecasts.

2. Fat-Tailed Revenue Projections

Revenue projections should incorporate fat-tailed distributions to reflect the outsized impact of outliers. For example, rather than a deterministic revenue estimate:

$$
E[R] = \int_{x_m}^{\infty} x \cdot \left( \frac{x_m}{x} \right)^\alpha dx
$$

This accounts for the higher probability of extreme outcomes in revenue generation.

3. Stress-Testing Key Assumptions

Projections should undergo rigorous stress-testing to evaluate robustness under adverse scenarios. For example:

• Market Share Sensitivity

$$
R_{\text{stress}}(t) = TAM \cdot 0.5 \cdot P_{adopt}(t)
$$

• Delays in Approval

$$
NPV_{\text{delayed}} = \sum_{t=1}^T \frac{CF_t}{(1 + r)^{t+\Delta t}}
$$

Where Δt reflects delay in years and rrr is the discount rate.

Conclusion

Optimism bias in biotech projections is not merely a cognitive error; it is a structural flaw with measurable consequences. Incorporating probabilistic ranges, Bayesian updates, and fat-tailed models into forecasting can bring much-needed realism to the field. By embracing uncertainty, biotech companies and investors can make more informed decisions, allocate capital more effectively, and reduce the cascading effects of inflated expectations.

As Taleb reminds us, "Uncertainty is not an obstacle to be avoided but the condition under which we thrive." Biotech, a field steeped in uncertainty, must take this lesson to heart.