Skip to main content

Sensitivity & Scenario Analysis

Updated Jun 24, 2026 at 2:35pm

Research Draft High 1,195 words

A DCF produces a single point estimate, but that number rests on a stack of forecasts — revenue growth, margins, reinvestment, discount rate, terminal growth — every one of which is uncertain. Sensitivity and scenario analysis are the two standard techniques for facing that uncertainty honestly: instead of pretending the output is precise, they show how the value moves when the inputs move. The core tension is between rigor and theater. Done well, they reveal which assumptions actually drive value and bound the range of defensible answers; done lazily, they dress up a guess with a grid of numbers that conveys false confidence.

How it's done

The two techniques are mechanically different and frequently confused.

Sensitivity analysis flexes one or two inputs at a time, holding everything else fixed, and tabulates the resulting value. The canonical artifact is the two-way data table: discount rate (WACC) along one axis, terminal growth rate along the other, enterprise or equity value in each cell. In Excel this is built with Data > What-If Analysis > Data Table, which re-runs the model for every input combination. The WACC × terminal-growth pairing is standard because both feed the terminal value via the Gordon growth formula, TV = FCF × (1+g) / (WACC − g) — a ratio so sensitive near WACC ≈ g that small input changes produce large value swings. Step sizes should reflect genuine input uncertainty (the spread in beta, equity risk premium, cost of debt), not arbitrary ±5% bands. A one-variable version, ranked by impact, is the tornado chart.

Scenario analysis instead builds a small number of complete, internally consistent stories — typically base, bull (upside), and bear (downside) cases — in which multiple assumptions shift together. A genuine recession case lowers revenue growth, compresses margins, and delays capex simultaneously, because those move together in reality (the discount rate may also rise if market risk premia and rates genuinely climb — but see the double-counting caveat below before stressing both rate and cash flows for the same risk). Damodaran frames this as one of three "probabilistic approaches" alongside sensitivity analysis and full Monte Carlo simulation, the last replacing point estimates with probability distributions and producing a distribution of values rather than three discrete ones (Damodaran, Probabilistic Approaches in Valuation).

The dividing line, per practitioner guides: sensitivity analysis is mechanical (vary inputs, watch output); scenario analysis is narrative (coherent cases). Flexing WACC alone is not a downside scenario — it's a one-variable sensitivity.

How it's used in practice

In equity research and investment banking these are not optional appendices — they appear in nearly every DCF deliverable. The two-way table is the headline output: a banker shows a grid of per-share values and the client reads off the range around the point estimate. Several such ranges (DCF, comps, precedent transactions, LBO) are stacked into the "football field" chart that frames a valuation negotiation or fairness opinion.

Beyond presentation, the real analytical value is diagnostic. Sensitivity output tells you what the valuation is actually a bet on. If WACC drives most of the movement, the thesis is about risk, durability, and capital structure; if terminal growth drives it, the thesis is about long-run market size and reinvestment efficiency. This matters because in many mature-company DCFs the terminal value commonly represents the majority of enterprise value — figures of roughly 60–80% are widely cited in practitioner material, though the exact share is model-specific. When terminal assumptions dominate, the disciplined move is to cross-check the implied exit EV/EBITDA multiple against comparable companies; if a "reasonable" 3% perpetual growth implies a 25× exit multiple, the growth assumption is the problem.

Adoption, debate & evidence

Adoption is essentially universal — these techniques are codified in the CFA curriculum, every modeling course, and standard banking templates — so the debate is not whether to use them but how much they actually add.

The honest critique, much of it from Damodaran himself, is that garbage in, garbage out applies fully: a sensitivity table is only as good as the model it perturbs, and surrounding a biased base case with a symmetric grid makes the bias look rigorous. He warns specifically against double-counting risk — if you already use a risk-adjusted discount rate, then also lowering cash flows for the same downside, you penalize twice. A second well-documented limitation is correlated inputs: real variables move together (high growth tends to accompany higher margins and reinvestment), so flexing one in isolation produces internally inconsistent and often impossible combinations. Damodaran's two fixes are to vary only the higher-impact input, or to build the correlation explicitly into a simulation — the latter requiring more sophisticated tooling.

There is also a behavioral concern: scenario ranges are usually too narrow, reflecting analyst overconfidence, so the "bear case" still embeds optimism. The empirical takeaway is modest but real — these methods improve transparency and humility about a valuation; there is no evidence they make the central estimate more accurate. They tell you how wrong you could be, not how right you are.

Strengths & limitations

Strengths. They expose the value drivers, force explicit assumptions, communicate a range instead of false precision, and (for scenario analysis) require coherent business stories that surface hidden inconsistencies.

Limitations. They cannot fix a flawed base case; correlated inputs make naive one-at-a-time sensitivity misleading; ranges are usually too tight; and they can double-count risk. The single most common misuse is relabeling a one-variable sensitivity as a "scenario" — e.g. calling a bumped-up WACC the "downside case" without touching revenue, margins, or capex. That produces a number that is internally inconsistent (a stressed discount rate with un-stressed cash flows) and overstates the apparent rigor of the analysis.

Sources

Dispute flag: The "terminal value = 60–80% of enterprise value" figure is widely repeated in practitioner material but is model- and company-specific, not a measured universal constant — treated here as "commonly cited," not exact.