Why Technical Success Changes NPV
A conventional net present value (NPV) calculation discounts a planned sequence of cash flows to a common valuation date. In drug development, however, later costs may be incurred only if the program reaches the relevant milestone, and commercial returns may occur only if technical development succeeds. A forecast that treats every planned cost and return as certain can therefore misrepresent the value of the program.
Risk-adjusted NPV (rNPV) addresses this by calculating the expected cash flow at each time, taking technical-success probabilities into account, and discounting those expected cash flows. It is a valuation under specified assumptions, not a prediction that the program will generate the calculated amount. The result can be useful for comparing development alternatives, provided their probabilities, cash-flow definitions, timing, and discount-rate conventions are consistent.
Representing Technical Success and Cash-Flow Timing
Suppose a program has ordered development milestones. Let \(p_j\) be the probability of passing milestone \(j\), conditional on reaching it. The cumulative probability \(Q_m\) of reaching milestone \(m\) is the product of the conditional probabilities up to that milestone:
This product follows the probability chain rule: each conditional probability is evaluated among programs that reached the preceding milestone. It does not require the milestones to be independent. For example, if transition probabilities are 0.80 and 0.50, the probability of reaching the second milestone is \(0.80(0.50)=0.40\).
Cash-flow timing determines which probability weights apply. A cost paid now is incurred regardless of future success and is not weighted by a future probability. A cost paid only after reaching milestone \(m\) is weighted by \(Q_m\). A return paid only if all required technical milestones succeed is weighted by the cumulative probability of that complete success. If an amount is paid in different outcomes or at different times, calculate its expected value using the probability and amount for each outcome and time.
The expected-value calculation is linear: the expected value of a sum of cash flows is the sum of their expected values. Thus each cash flow can be probability-weighted and discounted separately, then added. This is the basis of the milestone-by-milestone computation below.
Worked Example: A Program With Conditional Costs and a Technical-Success Return
Consider a hypothetical development program, with amounts in millions of monetary units. A cost of 12 is paid at time 0. A cost of 18 is paid in year 1 only if the first transition succeeds, with probability 0.80. A cost of 32 is paid in year 2 only if the program reaches the second milestone; the conditional probability of that transition is 0.50. A cost of 25 is paid in year 3 only if the program reaches the third milestone, with conditional probability 0.60. A return of 500 is received in year 5 only if all three transitions succeed. Use an annual discount rate of 10%.
The cumulative probabilities of reaching the three milestones are \(Q_1=0.80\), \(Q_2=0.80(0.50)=0.40\), and \(Q_3=0.80(0.50)(0.60)=0.24\). The return therefore has a technical-success probability of 0.24. The expected costs at years 1, 2, and 3 are respectively \(0.80(18)=14.4\), \(0.40(32)=12.8\), and \(0.24(25)=6.0\) million.
Discount each expected amount to time 0, using the period in which it is paid:
The costs’ expected present values sum to approximately \(12+13.091+10.579+4.509=40.18\) million, and the return’s expected present value is approximately 74.51 million. Their difference is an rNPV of about 34.33 million under the stated assumptions. This positive expected value does not mean that a realized program will earn 34.33 million: outcomes may include failure and a loss, or technical success and a much larger return.
Justification: Expected Cash Flows Before Discounting
For a cash flow that has amount \(C\) if an event \(A\) occurs and zero otherwise, its expected amount is \(P(A)C+[1-P(A)]0=P(A)C\). If the flow is a cost, its sign is negative. Discounting this expected amount by the deterministic factor \((1+r)^t\) gives the time-0 contribution. For a cash flow with multiple possible amounts \(c_k\), occurring in mutually exclusive outcomes with probabilities \(w_k\), the expected cash flow is \(\sum_k w_kc_k\).
Applying these facts at each time gives the general expression:
This expression assumes the same specified discount rate applies to the cash flows under consideration. It also makes clear why probability weighting and discounting have separate roles: technical-success probabilities determine expected cash flows, while the discount factor accounts for the timing of those flows under the chosen rate convention.
Worked Example: Costs Paid Only After Reaching Their Milestones
In a hypothetical program, a cost of 5 million is paid now. A cost of 10 million is paid in year 1 only if the first transition succeeds, with probability 0.70. A cost of 20 million is paid in year 2 only if a second transition succeeds, with conditional probability 0.50. A return of 100 million is received in year 4 only if both transitions succeed. Use an annual discount rate of 8%.
The probability of reaching the second milestone and receiving the return is \(0.70(0.50)=0.35\). The expected year-1 cost is \(0.70(10)=7\) million, and the expected year-2 cost is \(0.35(20)=7\) million. The return’s expected amount is \(0.35(100)=35\) million. Consequently:
This calculation weights each milestone-dependent cost by the probability of reaching the point at which that cost is paid. Weighting the year-2 cost by 0.70 would be wrong: the cost requires both transitions, so its probability is 0.35.
Required Assumptions and Consistent Conventions
An rNPV is only as meaningful as its assumptions. State the valuation date, cash-flow timing, currency and units, technical-success events, and discount-rate convention. The probability attached to a cost must correspond to the event required to incur it; the probability attached to a return must correspond to the success conditions required to receive it. Conditional transition probabilities should be multiplied only along the specified sequence, or cumulative probabilities should be estimated directly from an appropriate model.
The cash-flow estimates should also be defined consistently. If a projected return is already net of specified costs, those costs must not be subtracted again. If returns depend on commercial uptake, pricing, or other nontechnical uncertainties, a technical-success probability alone does not represent those uncertainties; they need to be included in the cash-flow scenarios or probability model if they are within the valuation’s scope.
The discount rate must match the time unit and monetary convention. An annual rate is used with annual periods; a nominal cash-flow forecast should not be combined uncritically with a real discount rate. A discount rate may reflect time value and specified financial risk, but applying an additional risk premium for the same technical failure already represented by explicit success probabilities can double count that risk. The analyst should document what the rate represents.
Worked Example: A Break-Even Technical-Success Probability
A hypothetical program requires a cost of 20 million now and a further cost of 30 million in year 1, paid regardless of the eventual technical outcome. If technical success occurs, the program receives 200 million in year 3; let \(p\) denote the probability of that success. Use an annual discount rate of 10%.
The two costs are certain under the stated assumptions, so neither is probability-weighted. The return is contingent on technical success, so its expected amount is \(200p\). The rNPV is:
Set this expression to zero to find the break-even probability:
Under these assumptions, a technical-success probability of about 31.46% gives an rNPV of zero. At \(p=0.30\), \(rNPV=-47.273+60/1.331\approx-2.194\) million. At \(p=0.40\), \(rNPV=-47.273+80/1.331\approx12.83\) million. These are threshold calculations for the stated cash flows and rate, not universal go/no-go rules.
Diagnostics, Alternatives, and Limitations
Several checks help identify a calculation that is going wrong. First, verify that each cumulative probability is between zero and one and no larger than the probability of reaching the preceding milestone. Second, check each cash flow’s required event against its probability weight: a cost paid before a trial outcome is not contingent on that outcome, whereas a cost incurred only after a successful transition is. Third, confirm the time exponent matches the payment year, and check that discounting reduces the magnitude of a positive future amount when \(r>0\). Finally, recompute the total as the sum of the separately listed present values and verify that positive and negative signs are consistent.
A conventional NPV using all planned costs and returns as certain answers a different question: the value of the planned cash-flow schedule if it occurs as specified. An rNPV instead takes expected cash flows under stated probabilities. Another approach is to represent the possible outcomes explicitly in a decision tree and calculate probability-weighted values for the branches; for a simple fixed schedule, the milestone-weighted rNPV formula is a compact way to do that expected-value calculation. These methods should agree when they use the same outcomes, probabilities, timing, and discount rate.
The result has limitations. It compresses uncertain outcomes into an expectation, so it does not describe the range of possible values or how likely a loss is. It is sensitive to uncertain probabilities, cash-flow forecasts, and discount rates; apparent precision in the final arithmetic does not imply that the inputs are precise. As in the earlier material on expected value of continuing a program, the estimated value depends on the specified actions, states, and payoffs. If a valuation ignores relevant outcomes or mismatches probabilities to cash-flow conditions, correct arithmetic will not make the valuation sound.
Check Your Understanding
Answer each question using the stated assumptions. Solutions follow the questions.
- Two sequential transitions have conditional success probabilities 0.60 and 0.50. What is the cumulative probability of reaching the second milestone?
- A cost of 8 million is paid in year 2 only if the second milestone is reached with probability 0.30. What is its expected cost before discounting?
- A return of 50 million is received in year 3 with probability 0.40. At a 10% annual discount rate, what is its expected present value?
- Explain why a cost paid at program initiation should not be weighted by the probability of eventual technical success.
Solutions. (1) The cumulative probability is \(0.60(0.50)=0.30\). (2) The expected cost is \(0.30(8)=2.4\) million; its signed cash-flow value is \(-2.4\) million. (3) The expected return is \(0.40(50)=20\) million, and its present value is \(20/(1.10)^3=15.03\) million. (4) The initiation cost is incurred regardless of later success, so its occurrence probability is 1. Later technical failure does not undo a cost already paid.