What you will be able to do
Run a sensitivity analysis and a break-even analysis on a business case and read the result.
Build scenarios, weight them and judge the downside next to the expected value.
Explain what a Monte Carlo simulation and a decision tree add, and where each misleads.
Say why raising the discount rate is a blunt answer to uncertainty.
Value the right to wait with a one-step binomial tree and explain why volatility raises its value.
Spot when an option is invoked to excuse a weak project.
1Sensitivity: which assumption does the project depend on?
A business case is a handful of assumptions wearing a spreadsheet. Before judging the risk, find out which assumptions carry the weight.
A project is a set of drivers, and they are not equally dangerous
Start from the drivers, not from the cash flows
A driver is a basic assumption from which the cash flows are built: units sold, price, unit cost, fixed costs, the discount rate. Some drivers lie outside management's control, such as market volume and interest rates. Others are partly controllable, such as unit cost or launch date. The split matters: controllable drivers are managed, outside ones are priced, hedged or accepted.
Change one driver at a time and read the swing
Sensitivity analysis moves one driver by a plausible amount, keeps the others fixed and records the new NPV. The drivers are then ranked by the size of the swing, the gap between the NPV when the driver goes wrong and when it goes right. Drawn as horizontal bars, from longest to shortest, the result is called a tornado chart.
| Driver | Adverse move | NPV | Favourable move | NPV | Swing |
|---|---|---|---|---|---|
| Unit cost | +10% (2,640) | 0.86 | −10% (2,160) | 13.78 | 12.92 |
| Volume | −20% (6,400) | 1.40 | +20% (9,600) | 13.24 | 11.84 |
| Price | −5% (3,325) | 2.61 | +5% (3,675) | 12.03 | 9.42 |
| Fixed costs | +20% (3.6m) | 5.30 | −20% (2.4m) | 9.34 | 4.04 |
| Discount rate | 12% | 5.45 | 6% | 9.46 | 4.01 |
Harmony Cargo line, NPV in € million. Base case +7.32.
Three drivers dominate: unit cost, volume and price. Fixed costs and the discount rate matter much less. A 5% price move has a big effect because each 1% of price is 35 per bike falling straight to profit, while each 1% of unit cost is 24. Choose each range by asking how far the driver has really moved in the past.
Break-even asks how wrong the forecast can be before value is lost
The break-even value of a driver is the value at which the NPV is exactly zero. For the Cargo line, an NPV of zero needs 15.0 / 4.486 = 3.34 million of cash flow a year, against 4.975 in the base case. Solving driver by driver gives three limits.
Volume: 6,023 bikes a year, which is 24.7% below plan.
Price: 3,228 per bike, which is 7.8% below plan.
Unit cost: 2,672 per bike, which is 11.3% above plan.
These are the most useful numbers in the analysis. They turn is the forecast right? into how likely is a price fall of 8%?, which sales managers can debate.
1.1 Two small mistakes at once
Harmony cuts the price by 5% and sells 10% fewer bikes than planned, with every other driver unchanged. Compute the new annual cash flow and the NPV at 9%. Compare with the NPV of each mistake taken alone (2.61 for the price cut, and 4.36 for 10% lower volume). What does the comparison say about reading a tornado chart?
The price falls to 3,325, so each bike contributes 925. Sales of 7,200 bikes give 6.66 million. After 3.0 of fixed costs, EBITDA is 3.66. The cash flow is 3.66 × 0.75 + 0.625 = 3.37 million a year. The NPV is 3.37 × 4.486 − 15.0 = +0.12 million.
Alone, the price cut leaves an NPV of 2.61 and the volume shortfall leaves 4.36. Each looks tolerable on a tornado chart. Together they remove 7.2 million of value, almost all of it. A tornado shows effects one by one and hides how fast small errors in different drivers add up.
sliders for volume, price, unit cost, fixed costs and discount rate, with the NPV and a live tornado chart. A second panel shows the break-even value of each driver. The learner discovers that the ranking of the drivers depends on the range chosen for each, and that two modest moves together can erase the whole NPV.
Your uncle Karim owns a restaurant and plans a second one. His accountant says the plan is solid and gives one profit figure. Explain to him why you would ask for the break-even sales level instead, and what you would do with the answer.
Karim: one profit figure hides how wrong it can be. The break-even sales level tells him where the new restaurant starts to lose value, and he can ask whether that level is likely in his neighbourhood.
Anaïs: the deposit bought the right, not the duty, to buy while she waited for the facts. If the loan failed she lost only the deposit. It is the price of flexibility, worth paying when the outcome is uncertain.
1
What does a tornado chart rank?
Right. For the Cargo line, unit cost, volume and price dominate.
It looks inside one project.
It ranks drivers, not years.
Probabilities belong to scenario analysis.
The Cargo line needs 3.34 million of cash flow a year for a zero NPV. What is the break-even volume?
That is the 10% shortfall of the exercise.
That is the plan, not the break-even.
Right. (3.34 − 0.625) / 0.75 + 3.0 = 6.625 of margin, divided by 1,100 per bike.
That divides the cash flow by the margin and forgets fixed costs and tax.
A 5% price cut alone leaves an NPV of 2.61; 10% fewer bikes alone leaves 4.36. What happens if both occur?
Effects compound, they do not average.
Both moves reduce the cash flow.
Right. A tornado shows effects one at a time.
It is easily valued: +0.12.
2Scenarios, simulation and trees: the shape of the uncertainty
Sensitivity says where the project is fragile. The next tools describe how the pieces move together and what management does when they do.
Scenarios move drivers together and tell a story
A scenario is a coherent world, not a random set of numbers
In scenario analysis, several drivers change at once, in a way that could really happen. A good scenario is a short story first and a column of numbers second. Harmony's finance team writes three.
| Bad | Base | Good | |
|---|---|---|---|
| Story | A rival launches a cheaper cargo bike and battery prices rise | Plan holds | Delivery firms adopt the bike faster than expected |
| Bikes per year | 5,500 | 8,000 | 10,000 |
| Net price | 3,300 | 3,500 | 3,600 |
| Unit cost | 2,500 | 2,400 | 2,400 |
| Cash flow a year (m) | 1.675 | 4.975 | 7.375 |
| NPV (m) | −7.49 | +7.32 | +18.08 |
Harmony Cargo line, three scenarios at 9%.
An expected NPV needs probabilities, and the downside needs its own look
If the finance team gives the three worlds probabilities of 25%, 50% and 25%, the expected NPV is 0.25 × (−7.49) + 0.50 × 7.32 + 0.25 × 18.08 = +6.31 million. The probabilities are judgements, not measurements, so show them and let the committee argue with them.
The average hides the shape. One world in four loses 7.49 million of value, which is 13.9% of Harmony's 54.0 million of equity. Whether that is acceptable depends on what Harmony can absorb, not on the average. Sizing the worst case against equity, covenants and the cash available is a stress test.
Simulation and decision trees extend the same idea
Monte Carlo simulation draws thousands of scenarios from ranges
A Monte Carlo simulation gives each driver a range and a shape, for example a volume that is most likely near 8,000 but could plausibly lie between 5,000 and 11,000. It also states how drivers move together. A computer then draws, say, 10,000 combinations and computes an NPV for each. The histogram of results is the project's risk profile: its average, the share of draws below zero and the size of the worst 5%.
The method has three limits. The output is only as good as the guessed ranges and links between drivers. Correlations are hard to specify and easy to leave out. And the result looks precise, which tempts people to stop thinking. Simple scenarios are far more common than simulation.
A decision tree adds the choices management makes along the way
A decision tree lays out the sequence: decide, learn something, decide again. It is solved from the right, starting with the last decision and working back to today. This captures a fact that scenario tables ignore: managers react.
A higher discount rate is a blunt answer to uncertainty
A common reflex is to add three points to the rate and call it prudence. For the Cargo line, 12% instead of 9% lowers the NPV from 7.32 to 5.45, a loss of 1.86 million. The bad scenario at −7.49 is untouched in character: the project still looks safe when the danger sits in one specific, nameable world.
The padding also buries the argument inside one number. Keep the rate at the project's market risk, as CF25 explains, and put project-specific risks in the cash flows, where they can be challenged.
2.1 A more pessimistic committee
The investment committee thinks the bad world is more likely than the finance team does and sets the probabilities at 40% bad, 40% base and 20% good. Compute the expected NPV and the probability of a loss. Does the project still pass the NPV rule, and does it still pass the committee's comfort test?
Expected NPV = 0.40 × (−7.49) + 0.40 × 7.32 + 0.20 × 18.08 = +3.55 million. The probability of a loss is 40%.
The project still passes the NPV rule, with about half the expected value of the finance team's case. Whether it passes the comfort test depends on the committee's appetite for a 40% chance of losing 7.5 million of value.
three sliders for the probabilities of the bad, base and good worlds (always summing to 100%), a bar for the expected NPV and a bar for the chance of loss. A button runs a small simulation of 1,000 draws and shows the histogram. The learner discovers that the expected NPV can stay positive while the chance of loss rises sharply.
2.2 Two projects have the same expected NPV. One has a narrow range of outcomes and the other a wide one. Why might a company with little spare cash prefer the first?
A company with little spare cash cannot absorb a bad outcome without borrowing dearly, cutting other projects or breaching a covenant. The wide-range project may bring that outcome. The narrow one is safer in practice.
2.3 Give two reasons why a Monte Carlo result can be misleading.
The result rests on guessed ranges and guessed links between drivers. Leave out a correlation and the spread is understated. The output also looks precise, which invites false confidence.
2.4 In the Germany tree, the pilot costs 2.0 for certain and the pilot route still has the higher NPV. How can that be?
The 2.0 buys information. After the pilot, Harmony avoids the 15.0 outlay in the weak world, where the rollout would lose 4.0. The avoided loss outweighs the pilot's cost, so the NPV is 3.05 against 1.97.
2
What makes a good scenario?
That is a sensitivity.
Right. A story first, a column of numbers second.
Scenarios must include the bad world.
That is simulation.
Why is adding three points to the discount rate a blunt answer to the Cargo line's risk?
It is a practice issue, not a rule.
Right. Put project risks in the cash flows, where they can be challenged.
The NPV stays positive at 12%.
It still makes it look safe.
In the German rollout, why does the pilot route win even though it costs 2.0 for certain?
Probabilities stay at 50/50.
The pilot reveals demand; it does not change it.
The same 9% is used.
Right. 3.05 against 1.97.
3Real options: flexibility has a price, and it can be worked out
The NPV rule treats a project as now or never. Real companies can wait, stage, expand or stop, and that freedom is worth money.
Static NPV ignores the right to change course
Three conditions turn a project into an option
A real option is the right, without the obligation, to take a business decision later: to invest, to grow, to shrink or to stop. It has value only when three conditions meet. There is real uncertainty. Something will be learned before the decision. And management can act on what it learns, at a cost it can estimate.
| Option | What management can do | Harmony example |
|---|---|---|
| Defer | Wait before committing | Delay the Nordic launch until demand is clearer |
| Stage | Invest in steps, each one conditional | German pilot before rollout |
| Expand | Add capacity if things go well | Leave room on the Lyon site for a second line |
| Contract | Scale down if things go badly | Run the line on one shift at first |
| Abandon | Stop and sell the assets | Resell Cargo tooling to a contract manufacturer |
| Switch | Change inputs or outputs | Assemble standard bikes on the same line |
Six real options, with the abandon option acting like a put (CF20).
Research and development is an option factory: a small spend buys the right, not the duty, to launch if the lab succeeds. The same logic explains staged venture funding (CF34).
The right to defer needs protection. If rivals can launch first, the option is worth much less. It is strongest when the company holds a licence, patent or site that rivals cannot copy.
A binomial tree values the right to wait
The Nordic launch is worth waiting for, even though it loses money today
The idea is to price the option by copying it. Build a portfolio of the project itself and a loan that pays off exactly like the option in every future state. Whatever the portfolio costs today is the option's value. No guess about probabilities or about the right discount rate is needed.
The pretend probabilities, called risk-neutral, are not forecasts. They are a pricing device that matches the portfolio copy. The method assumes the project's value can be tracked by something that trades, which holds roughly for commodities and badly for a unique product. Treat the answer as an estimate.
More uncertainty makes the option worth more, which reverses the usual instinct
In an NPV, more market risk lowers value because it raises the discount rate. In an option, more uncertainty raises value, because the right not to invest cuts off the bad outcomes while the good ones remain. Practice 3.2 shows this. The advantage falls as the decision date approaches, because less is left to learn, and exercising the option early throws its time value away.
3.1 A dearer launch
The Nordic launch now costs 22.0 million instead of 21.0, with every other number unchanged (value 20.0, up to 28.0 or down to 14.0, risk-free rate 3%). Compute the NPV of investing now, the payoff of waiting in the good state and the value of the option to wait. At what opening cost is the option worth nothing?
Investing now gives 20.0 − 22.0 = −2.0 million. Waiting gives a good-state payoff of 28.0 − 22.0 = 6.0. The option is worth 0.4714 × 6.0 / 1.03 = 2.75 million.
The option is worth nothing at an opening cost of 28.0 or more, because even the good state no longer pays.
3.2 A more volatile market
Back to the 21.0 million cost. Now the launch will be worth either 32.0 (× 1.6) or 12.0 (× 0.6) in a year. Compute the pretend probability, the payoff of waiting in the good state and the value of the option. By how much does it exceed 3.20, and what does this tell you?
The pretend probability is (1.03 − 0.6) / (1.6 − 0.6) = 0.43. The good-state payoff is 32.0 − 21.0 = 11.0. The option is worth 0.43 × 11.0 / 1.03 = 4.59 million, which is 1.39 more than 3.20.
Wider swings raise the value because the right not to invest cuts off the bad outcome while the good one grows.
a one-step tree with sliders for the project value, the opening cost, the up and down moves and the risk-free rate. The page shows the replicating portfolio, the option value and the verdict wait or invest. The learner discovers that widening the up and down moves raises the value of waiting, and finds the cost above which waiting is pointless.
Your neighbour Anaïs paid a small deposit to hold a flat for a month while her mortgage was decided. Her friend says she wasted the money, because she would have bought the flat anyway if the loan came through. Explain to the friend, without finance words, what the deposit bought and why it can be worth more than it cost.
3
Which three conditions turn a project into a real option?
Right. Without all three, flexibility is worth nothing.
Volatility matters only if management can wait and act.
Protection helps, but these are not the conditions.
Those are static measures.
Nordic launch: value 20.0, cost 21.0, up to 28.0 or down to 14.0, risk-free 3%. What is the option to wait worth?
That is investing now.
That uses a 50% probability; the risk-neutral one is 0.4714.
That is the good-state payoff, before weighting and discounting.
Right. 0.4714 × 7.0 / 1.03 = 3.20, against −1.0 for investing now.
Why does more uncertainty raise the value of the option to wait?
Investing now is still −1.0.
Right. 4.59 with ×1.6 / ×0.6 against 3.20.
Options are priced without a risk-adjusted rate.
They stay between 0 and 1.
4Using the tools without fooling yourself
Every tool here can be bent to justify a decision already taken. The last part is about how to use them honestly.
Optionality is an asset and also the oldest excuse in the business case
Three tests before an option goes into a business case
Uncertainty that will resolve. What will be known by the decision date that is unknown today?
A named decision, at a known cost. Who decides, when, and how much will it cost to expand, stop or switch?
Protection. What stops a competitor from taking the opportunity while Harmony waits?
A weak project cannot be rescued by calling it strategic. Option values rise with the volatility assumed, and volatility is easy to choose. The dot-com years and some biotech valuations are the standard warning. The safeguards: value the project without options first, show the option as a separate line, and make every option pass the three tests.
Other safeguards: reverse stress tests, climate scenarios and certainty equivalents
A reverse stress test starts from failure and asks what would have to happen. The break-even of Part 1 is the simplest version. A larger firm applies the same logic to its covenants: what combination of events breaks the leverage limit?
For long-lived assets, public climate scenarios, such as those of the Network for Greening the Financial System, supply carbon-price and demand paths. A project that works in only one of them is fragile.
The certainty-equivalent method is an alternative to adjusting the rate. Each cash flow is scaled down by a factor α between 0 and 1, then discounted at the risk-free rate.
NPV = − I₀ + Σ αₜ × CFₜ / (1 + r_f)ᵗ
Its logic is sound, since later cash flows are more uncertain and α falls with time. It is rarely used, because α is even harder to estimate than a beta.
4.1 A manager defends a weak project by saying it has strategic option value, but names no decision date, no cost and no new information. Which of the three tests does it fail?
It fails all three tests: no uncertainty that will resolve, no decision or cost, and no protection from rivals. It is a label, not an option.
4.2 Under the certainty-equivalent method, should α for year 5 be larger or smaller than α for year 1? Why?
Smaller. Cash flows further away are less certain, so more of each must be given up to turn it into a sure amount.
4.3 A simulation of the Cargo line gives a tight histogram around +7. Does that prove it is safe?
No. A tight histogram only shows that drivers were given narrow ranges or assumed independent. Check the inputs before trusting the picture.
4
A manager claims strategic option value but names no date, no cost and no information to come. Which tests does it fail?
Right. It is a label, not an option.
Volatility is not one of the three tests.
An option must pass the tests to be counted.
It names nothing at all.
What is a reverse stress test?
That is computing the IRR.
It is a reasoning method, not a time reversal.
It focuses on failure.
Right. The break-even is the simplest version.
How should an option be shown in a business case?
That hides the assumption.
That is the blunt tool the module warns against.
Right. Option values depend on an easily chosen volatility.
A real option that passes the tests should count.
Summary and cheat sheet
Les formules du module
Touchez un terme pour voir ce qu’il veut dire et sa valeur chez Harmony.
Flux d’équilibre
= /
D’où 6 023 vélos, un prix de 3 228 ou un coût de 2 672 : les marges de 24,7 %, 7,8 % et 11,3 %.
VAN espérée
= Σ ×
Probabilité neutre au risque
= (1 + − ) / ( − d)
Valeur de l’option d’attendre (un pas)
= [q × + (1 − q) × ] / (1 + r)
Attendre vaut 3,20 contre −1,0 pour investir maintenant : 4,20 de mieux.
Équivalent certain
VAN = − I₀ + Σ × CFₜ / (1 + )ᵗ
| Idea | What to remember |
|---|---|
| Driver | A basic assumption behind the cash flows. Separate controllable from outside ones. |
| Sensitivity analysis | Move one driver at a time and rank the swings. It shows fragility, not probability. |
| Break-even | The driver value at which the NPV is zero. It turns a forecast into a debate about how far it can be wrong. |
| Scenario | A coherent story in which several drivers move together. Weight the worlds, and look at the downside as well as the average. |
| Stress test | Size the worst case against what the firm can absorb: equity, covenants, cash. |
| Monte Carlo | Thousands of draws from ranges. Limited by guessed inputs and links. |
| Decision tree | Decide, learn, decide again, solved from the right. Value comes from avoiding the losing branch. |
| Discount rate padding | A blunt tool. Keep the rate at market risk and show project risk in the cash flows. |
| Real option | The right, not the duty, to decide later. Needs uncertainty, learning and the ability to act. |
| Binomial valuation | Copy the option's payoffs with the project and a loan. The copy's cost is the option's value. |
| Volatility | More uncertainty raises option value, if the holder can really wait. |
The one-sentence version
Test a forecast with sensitivities, break-evens and scenarios, keep risk in the cash flows rather than the discount rate, and count flexibility as value only when management has a real, protected right to wait, grow or stop.
Key terms
| Term | Plain-language meaning |
|---|---|
| Driver | A basic assumption, such as volume or price, behind the cash flows. |
| Sensitivity analysis | Changing one driver at a time to see how much the NPV moves. |
| Break-even | The driver value at which the NPV equals zero. |
| Scenario analysis | Changing several drivers together to describe a coherent possible world. |
| Monte Carlo simulation | Computing the NPV for thousands of random combinations of drivers. |
| Decision tree | A map of decisions and events, solved from the last decision back to today. |
| Real option | The right, without the obligation, to make a business decision later. |
| Risk-neutral probability | A pricing device, not a forecast, that values an option as if everything earned the risk-free rate. |
Going further
Vernimmen, Quiry, Le Fur — Finance d'entreprise (Dalloz), chapter on risk in investment decisions and real options.
Brealey, Myers, Allen — Principles of Corporate Finance, chapters on project analysis and real options.
Dixit and Pindyck (1994) — Investment under Uncertainty, Princeton University Press.
Trigeorgis (1996) — Real Options: Managerial Flexibility and Strategy in Resource Allocation, MIT Press.
Independent educational material. Not investment, legal or tax advice. All companies and figures in the running case are fictional.