Sensitivity Analysis, the One Slide Every Funder Reads

Every serious bankability report contains a sensitivity analysis. Most credit committees skim the summary page and stop; the ones who read further usually skip to the sensitivity slide. It is the one place the analyst tells the committee what they actually believe about the project's risk profile.

A sensitivity analysis done well tells the credit committee three things: which variables the project's headline metric is genuinely sensitive to, how big a shock the project can absorb before it fails a covenant or breaches a hurdle, and where the credit committee should focus its due diligence and covenant discipline. A sensitivity analysis done badly tests the wrong variables, presents them in a format nobody can read, and gives the false comfort that the project is robust when the analyst has actually tested only the assumptions the project is already known to be insensitive to.

This piece sits under the m2 pillar on bankability and viability. It sets out what to test, how to present the results, and the failure modes to avoid.

What sensitivity analysis is for

The purpose of sensitivity analysis is to answer the question: "given that we know some of our assumptions are wrong, how wrong do they have to be before the project stops working?"

A base case shows what happens if the assumptions are right. A sensitivity analysis shows what happens if they are not. The gap between the two is the project's resilience.

Sensitivity is not the same as scenario analysis (which changes multiple assumptions together to create internally-consistent alternative futures) and not the same as Monte Carlo simulation (which samples across probability distributions of the assumptions). Sensitivity is the simplest of the three: one variable moved at a time, all others held constant, measuring the impact on the headline metric.

The simplicity is the point. Sensitivity gives the credit committee a clear, decomposed view of which assumptions matter. Scenario and Monte Carlo add value on top of that base but do not replace it.

The tornado chart

The convention for presenting sensitivity results is the tornado chart. Each variable being tested is shown as a horizontal bar; the width of the bar shows the impact on the headline metric of shocking that variable through its plausible range. The variables are sorted by bar width, largest at the top, tapering downward like a tornado.

The tornado chart is powerful because it makes the sensitivity ranking visible at a glance. A credit committee looking at a tornado chart sees immediately which variables are the project's dominant risks and which are noise. The bar widths carry more information than the numbers alone: the two variables at the top are what the covenant discipline needs to cover; the variables at the bottom can be treated as second-order.

A tornado chart done badly has bars sorted alphabetically or by input order rather than by width. A tornado chart done well has the widest bars at the top, is annotated with the shock range applied to each variable, and shows the base-case value and the two shocked values on each bar.

Brealey, Myers, and Allen (2020) treat the tornado chart as the standard presentation format for corporate finance sensitivity work; the World Bank's public-private partnerships reference guide (World Bank, 2017) uses it as the reference format for infrastructure appraisal.

What to test

The right variables to test are the ones the project is plausibly sensitive to, over ranges that are plausible for the specific project.

The starting list for most infrastructure and project finance projects includes:

• Revenue. Volume, price, or both. Shock ranges depend on the underlying revenue certainty: for a take-or-pay contract with an investment-grade offtaker, five percent may be the outside of plausibility; for a merchant-risk project, twenty percent may be conservative.
• Operating costs. Fixed costs, variable costs, or both. Shock ranges from ten to twenty percent are common; industries with volatile input pricing (fuel, imported components) may need wider ranges.
• Construction cost overrun. For projects with construction phases, ten to twenty percent above budget is a standard stress; some infrastructure sectors have historical average overruns above thirty percent, which should inform the stress range.
• Construction delay. Time overrun affects both the operating start date and the interest capitalised during construction. Six, twelve, or eighteen months are common stress points.
• Discount rate. Two percentage points above and below the base-case rate; wider for projects in more volatile-rate environments.
• Foreign exchange (for cross-currency projects). Historical volatility of the relevant currency pair provides a reasonable shock range.
• Interest rate (for floating-rate debt). Two to three percentage points is a standard stress.

For socio-economic evaluations, the sensitivity list should also include the shadow prices used for the major benefit categories, because the BCR is often exquisitely sensitive to those.

What NOT to test

The temptation with sensitivity analysis is to test many variables to look thorough. This produces cluttered tornado charts that hide the important variables among noise.

Variables that should not appear on the tornado chart:

• Contractually-fixed inputs. Testing the sensitivity of a project's NPV to a fixed-price take-or-pay contract at the contracted price plus five percent is not a meaningful sensitivity if the contract price is genuinely fixed. Test the counterparty credit sensitivity instead.
• Variables with immaterial impact. Sensitivity ranges should be presented for variables where the impact on the headline metric is at least a few percent. Variables that move NPV by 0.1 percent are not decision-useful.
• Variables already stressed in the base case. A conservatively-set base-case assumption should not be further stressed downward as a sensitivity; the base case already reflects the caution.

A tornado chart with 25 bars is a chart nobody reads. A tornado chart with 8 to 10 bars, ranked by impact, is a chart credit committees engage with.

Interpreting the results

The two bars at the top of the tornado chart are what the project's viability rides on. Two questions follow:

Is the base-case assumption on those variables defensible? If the project's IRR is most sensitive to a demand-growth assumption of four percent per annum, that assumption needs to be triangulated against at least two independent sources. If it cannot be, the base case is fragile and the sensitivity result confirms the fragility.

Can the covenant or hurdle be met under the plausible-adverse shock? If the project's IRR under a fifteen percent revenue shock falls to nine percent against a ten percent hurdle rate, the project fails on a plausible adverse case. The credit committee needs to decide whether the fifteen percent shock is genuinely plausible; if it is, the transaction is not bankable at the current structure.

Answering these two questions honestly is what sensitivity analysis is for. Sensitivity analysis that raises the questions but does not force the answers is decorative.

Beyond one-variable sensitivity

For projects where the interactions between variables matter (revenue and cost inflation moving together, foreign exchange and interest rate correlated), one-at-a-time sensitivity understates the risk. Scenario analysis, which changes multiple variables together in an internally-consistent way, is the natural next step.

A scenario might combine "adverse revenue" and "adverse operating costs" and "adverse foreign exchange" into a coherent downside scenario. If the project remains viable under a scenario that combines all three at plausible-adverse levels, it is genuinely robust. If it fails under such a scenario, the individual sensitivities may have understated the risk of correlated adverse movements.

Monte Carlo simulation adds another layer, sampling across probability distributions of the input variables to produce a distribution of the output metric. Monte Carlo adds real value on projects where the underlying distributions are well-characterised (typically projects with historical data for the key input variables) and adds noise on projects where the distributions are assumed. Analysts should use Monte Carlo where the data supports it and stick to sensitivity and scenario where the data does not.

Cross-border considerations

Cross-border projects add complexity to the sensitivity analysis because currency, interest rate, and country-risk sensitivities interact in ways that domestic projects do not experience. A cross-border project's tornado chart typically includes currency and country-risk premium as significant variables in addition to the domestic variables.

Sponsors submitting cross-border project finance to a lender consortium should present the sensitivity in a way that separates the country-specific risks (which some lenders take differently than others) from the project-specific risks (which are common to all lenders). This helps the lender consortium calibrate its participation and risk sharing.

What good looks like

A sensitivity analysis presented well shows: a tornado chart with 8 to 10 bars ranked by impact; the specific shock range applied to each variable, annotated on the bar; the base case and the shocked outcomes labelled clearly; the discount rate or hurdle used, so the credit committee can interpret the results against the right benchmark.

Alongside the tornado chart, a defensible presentation includes a small number of scenario cases (base, downside, adverse downside) with the full metric outputs for each, and a discussion of the assumptions driving the differences.

Credit committees who see this presentation engage with the project. Credit committees who see a single-number NPV and a boilerplate sensitivity paragraph either ask for the missing analysis or reject the project on incompleteness. The sensitivity slide is the one that either establishes credibility or forfeits it.

The CentraSolve Bankability and Viability module builds the sensitivity and scenario analyses as first-class artefacts of the workbook, with the tornado chart generated from the underlying model rather than pasted in as a picture. Changes to inputs flow through automatically; the analyst does not have to rebuild the sensitivity when the base case changes.

References

Textbook and framework sources

Brealey, R. A., Myers, S. C., and Allen, F. (2020). Principles of corporate finance (13th ed.). New York: McGraw-Hill Education.

World Bank Group. (2017). Public-private partnerships reference guide (Version 3.0). Washington, DC: World Bank.

Yescombe, E. R., and Farquharson, E. (2018). Public-private partnerships for infrastructure: principles of policy and finance (2nd ed.). Oxford: Butterworth-Heinemann.

South African framework sources

National Treasury, Republic of South Africa. (2019). Budget Facility for Infrastructure: guidelines and templates. Pretoria: National Treasury.