Quantifying risk / reward in biotech investing

September 20, 2023

Biotech investing is risky. This post discusses a technique for better understanding that risk.

The post includes an interactive tool for modeling the "odds" of a given biotech investment.

The tool can help understand which companies the market perceives as outliers, with significantly better (or worse) prospects than other similar companies.

It can also help understand which companies may be overvalued or undervalued, and can help investors better assess a company's risk / reward profile.

Monte Carlo simulation

The tool uses Monte Carlo simulation to create hundreds of simulated future paths for a given company ¹.

In one path, maybe the drug fails Phase 3, and the company ends up going bankrupt.

In another, perhaps the company's drug is the biggest drug of all time.

In yet another, maybe the drug gets approved, but has a modest launch due to stiff competition.

Each future path for the company is randomly generated based on historical data from public biotech companies, so the simulations are grounded in reality. This preliminary version of the model does not take into account any company-specific info except for the development stage of the pipeline assets. Thus if you have a cancer company, the distribution of outcomes is for all public companies, not just for cancer companies.

More on how it works below. But first, we'll walk through how you can use this analysis.

How to use this

Add a ticker for a publicly traded, pre-approval therapeutics company above. If the company is in our database, we'll show the company's latest enterprise value.

For example, set the ticker to AKRO and click "Check for ticker". AKRO (Akero Therapeutics) has one product that is currently entering Phase 3 for NASH, so select "Phase 3" for stage (putting in the product name is optional).

Then click "Calculate valuation range".

After a few seconds of thinking, the tool will display a chart with a distribution of potential valuations. The x-axis shows the range of enterprise values, and the height of each bar indicates how likely it is for the company to be worth the corresponding enterprise value (note that most of these distributions will be skewed, as most biotech companies lose money, and a very small number make a large amount of money).

AKRO's chart will show that AKRO is valued higher than 95%+ of simulated values (the exact number will be slightly different for you, as the simulations are randomized; also, AKRO’s stock may move which will also impact this number – the 95%+ figure is correct as of early September 2023).

In other words, the market expects AKRO to be a near-singular success.

Then the question becomes (assuming we trust the model -- see below for limitations): do we agree with this? Do we agree that AKRO will be better than 95% of other Phase 3 companies?

And further, what would drive 20%+ upside to the current valuation? Or 20% downside?

Reverse engineering the market

To answer these questions, we can use the tool to reverse engineer what assumptions the market is using to value the company. We can essentially backsolve for the set of DCF / rNPV models that can 1) justify the company's current valuation and 2) be defensible -- ie, the models do not make any outlandish assumptions that are not supported by data.

The model suggests that AKRO's current valuation implies Phase 3 success, that approval is all but certain, and that the product generates multiple billions a year in peak sales.

We can examine how likely it is for each of those assumptions to be correct. To illustrate, we just focus briefly one assumption in this post -- peak sales.

The second chart above shows the range of peak sales assumptions for each simulated future path the company could take (in blue). The red bars show the peak sales in the scenarios where the company is valued higher than its current valuation.

In blue, we can see that, based on historical data, most drug launches don't generate more than a few hundred million in peak sales. If Akero is just a typical company with a typical drug, it won't generate enough revenue to justify its valuation.

Thus, AKRO's drug must be a blockbuster -- generating $3B+ in peak sales -- to justify its current valuation. We can ascertain this by looking at the red bars, which show the range of peak sales (in combination with other variables) that can yield a valuation higher than AKRO's current value.

Do you agree with the market?

Knowing this information, the question as an investor (at least a fundamentals-focused investor) becomes "do I agree with the market?". Do you believe this drug has a very high likelihood of approval and that it will, if approved, generate $3B+ in peak sales?

We don't have a particular viewpoint on this. We aren't investors in AKRO and have not researched the company beyond quickly perusing its website while writing this post. This is just an illustrative example, not investment advice.

Based on some quick research, it appears that investors assign such high prospects to AKRO because recent Phase 2b data demonstrated that it has best-in-class potential in NASH. NASH is a serious disease that affects millions of people in the US, and (as of Sept 2023) there are no approved drugs. So a good NASH drug has the potential to be a massive success on the order of GLP-1 agonists (like Ozempic and Mounjaro), PD-1 inhibitors or a disease-modifying Alzheimer’s drug.

Thus it seems fair that AKRO is valued higher than a typical Phase 3 company, and one can make a case that it should be valued higher than 95% of Phase 3 companies. On the other hand, the current valuation prices in a very successful scenario. So you won’t see significant upside unless 1) the market sees data suggesting the peak sales are likely to be meaningfully higher (closer to the $4-5B+ range), 2) there is an uptick in sentiment, or 3) the company is acquired (investors may be pricing in a fairly high chance of acquisition if the drug is successful). And there are risks, such as a Phase 3 failure or tougher-than-expected competition (from other NASH drugs or even GLP-1 agonists).

Again, we have no opinion on whether this company is a good investment and have done no research beyond perusing the website briefly. The point of this post is not to express a viewpoint on a particular stock, but to introduce this method of understanding the odds of an investment a bit more quantitatively.

Limitations

The tool in this post is just a proof of concept and not meant to inform actual investment decisions. For example, the only information this model takes about a particular company are the development stages of its pipeline products. A more refined model could incorporate more company-specific information.

This tool only shows the "fundamental" value of a company -- the potential of the business to generate cash flows in the future. Fundamental value is only one of many factors that determine a company's stock price. Thus this tool is not intended to predict or render an opinion on a company's stock price; rather it is intended to help critically assess whether a given valuation is supported by fundamentals.

Specifically, this model does not account for strategic value or speculative value. Big pharma companies often buy smaller companies for significant premiums to their market value. Companies like Akero -- with Phase 3 or approved assets with best-in-class potential targeting large markets -- are some of the most highly valued acquisition targets. It is quite possible (perhaps likely) that the market values AKRO so highly because it assigns a high probability that the company is acquired in the near-term, and that a big pharma company will take on the commercial risk while still paying a very high price for the company. Further, an increase in risk-tolerance generally (like we saw during COVID) can cause stock prices to meaningfully exceed fundamental value, and this model does not capture this phenomenon.

The tool also does not consider platform value or the value of early preclinical assets. This model is actually well suited for valuing platforms in a more rigorous way than is done currently, but that application is beyond the scope of this post.

Further, this post just shows a snapshot of the model output. Specifically it only shows the range of peak sales estimates that could justify a given valuation. The model can generate a full DCF model for each simulated scenario, and show the robustness of each DCF model (ie, does the model rely on one or a few outlier assumptions, or are all key assumptions within the range of defensible values based on historical data).

There are several other technical limitations relating to scope of the dataset, choice of models for each variable, discount rate, etc., but those are beyond the scope of this post. Feel free to contact us with more questions on methodology.

More details on the model

The model is based on an rNPV model framework. rNPV (risk-adjusted NPV) is a common way to value pharma companies. This technique values a company based on the risk-adjusted, discounted future cash flows the company is expected to generate.

The obvious limitation of the rNPV / DCF technique is that it relies on assumptions about the future -- which are inherently uncertain and inevitably inaccurate. Many rNPV models are highly sensitive to assumptions that are difficult to quantify.

To account for this limitation, most rNPV models include a scenario analysis (modeling a base case, upside case and downside case) and sensitivity analyses (changing a few variables within a range and observing the impact on valuation). However, these techniques provide a very rudimentary understanding of risk and are subject to the same uncertain assumptions about the future.

Rather than rely on subjective assumptions about uncertain variables, the model in this post 1) relies on historical data rather than subjective, human assumptions and 2) analyzes a full range of likely values for each assumption, rather than arbitrarily choosing one or a few values for each assumption in the rNPV analysis. We collect current and historical financial data for each assumption underlying a DCF. We then fit a statistical distribution for each assumption based on this historical data.

Then we run an rNPV model, taking one random sample for each variable from the historical data distribution. We repeat this simulation thousands of times, each time choosing a different value for each variable in the rNPV model (COGS % sales, peak sales, Phase 2 probability of success, etc).

This gives us a range of valuations for the company -- grounded in the reality of actual historical data.

¹ The model can run as many simulations as required, but in order for it to run relatively quickly in a web-based tool, we've limited the number of simulations to 500.