Decision Tree Analysis in Real Options

A decision tree for real options is an explicit diagram of all the choices a manager might face over time and their possible outcomes. By assigning cash flows and probabilities to each branch, the tree reveals the value of optimal decision-making under uncertainty—and how much extra value comes from the ability to wait, learn, and adapt rather than commit blindly to a single plan.

The anatomy of a real-options decision tree

A decision tree is a visual and computational map of the future. The root node represents today’s choice: invest in Phase I of an R&D program, or skip it? If you invest, the outcome is uncertain: maybe the prototype succeeds, maybe it fails. The tree branches left and right, with probabilities on the arrows. At the next branch point, you face another decision: if Phase I succeeded, should you invest in Phase II? If it failed, should you cut losses or pivot?

Each branch’s endpoint—a terminal leaf—shows a final cash flow (profit or loss). Working backward from the leaves, the tree computes the expected value at each decision node, choosing whichever action maximizes that node’s value. This reveals not just the average payoff, but the optimal strategy at each point.

A multi-stage example: drug development

Imagine a pharmaceutical company evaluates a new molecule. Stage 1 (preclinical): invest $5 million. If the chemistry looks good (60% chance), move to Stage 2. If not (40%), abandon and lose the $5 million.

Stage 2 (Phase I trial): invest $20 million. Success (70% chance) leads to Stage 3. Failure (30%) sinks the $20 million (and the company writes off the entire $25 million).

Stage 3 (Phase II trial): invest $100 million. Success (50% chance) yields peak annual revenues of $300 million per year for 10 years (after which the patent expires). Failure (50%) loses the $100 million and any hope of payoff.

Discounting at 10% cost of capital, the Phase III success payoff—$300 million annually for 10 years—has a present value of roughly $1.8 billion.

Backward induction:

At Stage 3 nodes:

• If success (PV = $1.8B), the net is $1.8B − $100M = $1.7B.
• If failure (PV = $0), the net is $0 − $100M = −$100M.
• Expected Stage 3 value = 0.50 × $1.7B + 0.50 × (−$100M) = $800M.

At Stage 2 nodes:

• If success, you advance to Stage 3, gaining $800M expected value.
• If failure, you gain $0.
• Net Stage 2 value = 0.70 × ($800M − $20M) + 0.30 × (−$20M) = 0.70 × $780M − $6M = $540M.

At Stage 1 nodes:

• If success, you advance to Stage 2, gaining $540M expected value.
• If failure, you gain $0.
• Net Stage 1 value = 0.60 × ($540M − $5M) + 0.40 × (−$5M) = 0.60 × $535M − $2M = $319M.

The decision: Invest in Stage 1 today if you have $5 million and believe the probabilities and payoffs. Expected value = $319 million.

Why trees reveal option value

A naive DCF might assume a single 50%-probability path (preclinical succeeds, Phase I succeeds, Phase II succeeds) and compute $1.8B × 0.6 × 0.7 × 0.5 = $378M, then subtract cumulative costs and discount, yielding perhaps $150M NPV. But this ignores a crucial fact: if Phase I fails, you stop. You don’t lose the full value of the drug; you’ve only lost $25 million, preserving your capital for other bets. The tree captures this sequential learning and optionality; a simple static model does not.

The option value is the difference between the tree’s optimal expected value and the “passive” NPV that assumes you must commit fully or not at all. In this example, the tree’s $319M value reflects the flexibility to abandon at each stage; pure NPV on an all-or-nothing $125M upfront investment would be lower.

Choosing probabilities and cash flows

The tree’s accuracy hinges on reasonable inputs. For cash flows, use discounted cash flow logic at each terminal node: project revenues, costs, and tax effects out to end-of-life, then discount to the branch’s date. For probabilities, draw on historical data (e.g., “industry Phase I success rates are 66%”), management expertise, and comparable projects.

Avoid the trap of using probabilities that already embed management’s risk aversion. The tree should reflect the physical/market chance of success, not a conservative haircut. Risk is accounted for in the discount rate.

Practical extensions and complications

Multiple parallel options: A company might have the choice to expand production or to license technology to a partner, rather than a simple go/no-go. The tree branches further; at each decision node, the manager picks the highest-value option.

Continuous variables: If a project’s cash flows depend on a continuously varying parameter (commodity price, market share), the tree can be refined with many more branches, approximating the distribution of that variable. This mirrors the binomial lattice approach.

Partial abandonment: Instead of all-or-nothing, the manager might scale back. The tree’s branches represent a range of commitment levels (e.g., 25%, 50%, 75%, 100% of capacity), each with a different cost and payoff.

Changing probabilities: If learning in Stage 1 might shift the perceived success rate for Stage 2, the tree’s probabilities can be conditional (e.g., “if Stage 1 outcome A, then Stage 2 success is 80%; if outcome B, it’s 40%”). This is called a conditional probability tree and is standard practice in structured finance and project management.

Communicating trees to stakeholders

A well-drawn decision tree is one of the most persuasive tools in the executive suite. Unlike a Black-Scholes formula or binomial model, every manager can see the branches and trace the logic. “If we invest here and it succeeds, we then face this choice…” is intuitive. Sensitivity analysis—“what if Stage I success is 40% instead of 60%?"—can be shown by redrawing the tree or adjusting the probabilities in a spreadsheet and recomputing.

Conversely, be wary of over-engineering. A tree with 20+ endpoints and three layers of nested decisions becomes hard to manage, and the output precision belies uncertainty in the underlying assumptions. Simpler trees (3–5 decision points) are usually more useful.

See also