Scenarios, sensitivity and Monte Carlo: choosing the right test

Start with the decision the analysis must support

A land bid, a construction commitment and a refinancing decision face different uncertainties. Before running a test, state the choice, the date by which it must be made and the threshold that would change it. That threshold might be a maximum equity contribution, a minimum profit margin or a latest acceptable completion date. An attractive average return does not answer whether the sponsor can fund the worst cash deficit.

Separate a controllable choice from an uncertain assumption. Unit mix may be a design choice; construction duration may be uncertain; a planning condition may be an unresolved fact requiring investigation. Record these distinctions in an assumption register. HM Treasury treats sensitivity, scenarios and simulation as different appraisal techniques, with more complex methods justified by the decision and risk involved.

Sources: HM Treasury: The Green Book (2026); Uncertainty Toolkit for Analysts in Government: Understanding and measuring uncertainty

Use sensitivity to isolate a driver

A one-variable sensitivity changes one independent input while keeping the other independent inputs fixed. Dependent calculations must still update. A later completion should move affected receipts and financing costs according to the model, rather than changing a date label alone. A two-variable table can then test a useful pair, such as achieved price and construction cost, without burying the result in dozens of assumptions.

Choose a range from evidence, a contractual exposure or an explicit stress, and label its basis. In a deliberately simplified example, revenue of 100 and costs of 80 produce profit of 20. A 5% revenue reduction cuts that profit to 15, a 25% reduction, before any secondary effects. This illustrates why a modest input movement can cause a much larger movement in residual profit; it is not a market forecast.

Sources: Microsoft: Introduction to What-If Analysis

Build scenarios that could happen together

A scenario combines assumptions into a coherent account of what happens to the project. A slower sales case might include fewer monthly contracts, revised incentives, later instalment receipts and a longer inventory holding period. An accelerated delivery case might require an additional procurement cost. The relationship between assumptions matters more than whether the scenario has an optimistic or pessimistic name.

Give every case a short explanation, an owner and a list of changed assumptions. Keep the approved base case intact so a reviewer can reproduce the comparison. Do not label three scenarios as probabilities unless those probabilities have a defensible basis. A severe but plausible stress can be useful even when nobody can estimate its likelihood. It reveals a vulnerability, rather than predicting a precise future.

For an original four-quarter example, start with receipts of 0, 10, 35 and 55 million and costs of 20, 30, 20 and 10 million. A slower-selling case reduces receipts to 0, 5, 20 and 70 million, reflecting later collections and lower net proceeds. Add 2 million of non-financing holding and remarketing costs in quarter 3. Everything else stays fixed. This is a defined stress, not a forecast or a probability estimate.

The base case needs 40 million to cover its largest quarter-end cash deficit; the slower-selling case needs 47 million. The extra funding gap is 7 million while final surplus falls from 20 million to 13 million. Keep both consequences in the decision record. Financing charges, tax, restricted balances and intra-quarter cash movements are excluded here and can change the actual capital requirement.

Sources: Uncertainty Toolkit for Analysts in Government: Understanding and measuring uncertainty

Find the point at which the decision changes

A switching value turns an open-ended risk discussion into a specific question: how far can achieved rent fall before the agreed return threshold is missed? How many months of delay exhaust the available liquidity? Solve for the relevant input and report the boundary alongside the current assumption. In Excel, Goal Seek is one tool for a single target and changing input; a sensitivity table can show the surrounding pattern.

Check whether the calculated boundary is physically and commercially possible. A rent that balances the model may be unsupported by occupier demand. A required sales pace may exceed the release programme. If the model contains discontinuities, such as a new financing fee or a construction phase that starts only after a trigger, inspect both sides of the threshold. A single solved number can conceal that change in behaviour.

Sources: HM Treasury: The Green Book (2026); Microsoft: Introduction to What-If Analysis

Use Monte Carlo when its assumptions can be defended

Monte Carlo repeatedly samples uncertain inputs and calculates the model for each sample. The resulting spread describes outcomes conditional on the chosen distributions and relationships. The hard work is specifying those inputs, including dependencies. Cost escalation and programme delay, for example, may share causes. Treating every input as independent can create combinations that are commercially implausible or understate joint exposure.

Document the distributions, limits, correlation approach and evidence. Retain settings needed to reproduce the run, then check whether the reported percentiles remain stable as the number of simulations increases. Thousands of iterations do not validate a guessed distribution. If probability assumptions cannot be supported, present ranges or scenarios without attaching a spurious likelihood. Simulation is a modelling method discussed here, not a statement that any particular software includes it.

Sources: Uncertainty Toolkit for Analysts in Government: Understanding and measuring uncertainty

Review the whole cash path

Compare each case using a consistent set of outputs: development profit, the stated return measure, peak equity requirement, maximum debt and the period of the tightest liquidity. Show which cash flows belong to the project and which belong to the investor. A delay can leave nominal revenue unchanged while moving the funding peak and reducing a dated return. These are different consequences and deserve separate lines in the review.

A retained asset also needs an explicit operating and disposal path. Test occupancy, incentives, operating costs and the exit assumption in a way that avoids counting the same downside twice. For a sold development, distinguish contracts from collections. Explain any restricted receipts or financing conditions separately. A sensitivity grid that shows only the final profit can miss the period in which the project becomes difficult to fund.

Turn the test into an action

An effective review ends with an action linked to evidence: revise the bid, stage the acquisition, obtain a firmer construction quotation, secure a liquidity reserve or postpone a commitment until an approval is clarified. Assign an owner and a date to the uncertainty that matters most. Keep unresolved facts visible rather than burying them inside a wide contingency range.

Quality assurance should ask both whether calculations implement the specification and whether that specification is appropriate for the decision. Reconcile the base case, inspect unusual outcomes and have another reviewer challenge the explanation. Publish the assumptions and limitations with the result. The useful deliverable is a decision that can be revisited when evidence changes, not simply a larger collection of charts.

Sources: UK Government: The AQuA Book