The tornado chart is now standard in investment committee packs. One bar per input, ranked by impact on the headline output, holding everything else at the base case. It is fast to produce, easy to read, and reassuringly visual. It is also, in most consequential decisions, misleading.
The qualification is not that single-variable sensitivity is wrong as a presentational device. It is that committees increasingly treat the tornado chart as the analysis of risk, when in fact it answers a narrower and less interesting question: how would the model behave if one assumption moved while every other assumption stood still. Real-world variables do not move in isolation. Commodity prices and currencies co-move. Interest rates and refinancing risk co-move. Demand assumptions and supplier behaviour co-move. A risk picture built on the fiction of independence understates downside materially, and committees that rely on it make poorer decisions.
This note sets out the case for layered scenario architecture: structured, internally coherent worldviews that committees can interrogate as wholes, sitting alongside (not replacing) the conventional sensitivity work.
Where the convention came from, and why it persists
Single-variable sensitivity has a respectable pedigree. It is the natural output of partial-derivative thinking; it is computationally trivial; it identifies which assumptions the result is most sensitive to, which is useful diagnostic information. For an analyst building a model, the tornado chart answers a real question: where to spend the next hour of effort on assumption work.
The convention has carried over from the analyst's workbench to the committee paper largely because no obvious replacement has imposed itself. Monte Carlo work, the natural alternative for institutional audiences, is often poorly received: committees distrust outputs they cannot reconstruct by hand, and probabilistic outputs without specified joint distributions can be more misleading than the tornado chart they replace. The result is a default that everyone knows is incomplete, kept in place because the alternatives feel either too complicated or too speculative.
What the tornado chart misses
Three categories of co-movement matter in practice, and the tornado chart suppresses all three.
- Macro co-movement. Commodity prices, currencies, interest rates and growth rates are not independent draws. A scenario in which oil rises 30 per cent typically arrives with a stronger dollar, tighter credit and a softer demand environment for non-energy goods. Modelling each variable in isolation produces a confidence band that is not symmetric in reality.
- Operational co-movement. A supplier's payment terms tighten when their input costs rise; customer payment behaviour deteriorates when their working capital is squeezed; staff retention falls when the labour market tightens at the same time as wage budgets do. The model treats these as separate inputs because they appear separately on the input sheet. The world does not honour that distinction.
- Behavioural co-movement. Lenders raise covenant scrutiny precisely when operating performance softens. Refinancing windows close at the moment refinancing risk is highest. The interaction of behaviour and arithmetic is the substance of distress modelling; single-variable sensitivity cannot capture it.
The result, in a tornado chart, is that the largest bar tends to be the variable with the widest plausible range, not the variable most likely to drive the actual loss case. Committees that read the chart literally end up over-hedged on the visible exposure and under-hedged on the structural one.
Layered structured scenarios: the proposal
A layered scenario architecture replaces the tornado chart (or, more accurately, supplements it) with a small number of named, internally consistent worldviews. Each worldview specifies a coherent set of values across all material inputs, derived from a shared causal story rather than from independent moves on each axis.
For a typical commercial model, four to six scenarios is the right cardinality. Fewer and the committee cannot triangulate; more and the discussion becomes a comparison exercise rather than a decision one. The set we tend to recommend, adjusted to the engagement, runs roughly as follows.
- Base. The central case, with every assumption set at the agreed central value and explicit documentation of how each central value was reached.
- Operational downside. A coherent story in which the operating environment deteriorates: demand softens, churn rises, key supplier terms tighten, working capital absorbs cash. Macro variables are held near base.
- Macro downside. A coherent story in which the macro environment deteriorates: rates rise, credit tightens, currency moves against the business, commodity inputs spike. Operational variables are held near base, with explicit second-order links (the supplier whose own input costs rise will, in time, push terms).
- Combined stress. The two downsides together, with explicit treatment of how they interact (refinancing in a softer operating environment costs more; covenants bite earlier; capital expenditure decisions taken in the prior period now look heavy).
- Upside. A coherent story in which a defined catalyst (a contract, a regulatory change, a market shift) lands as planned, with the second-order consequences (capacity strain, hiring, working-capital build) modelled rather than ignored.
- Recovery. Where relevant, a scenario describing the path from the combined stress back to base, which is often where the capital structure question actually sits.
Each scenario is a worldview, not a parameter set. The narrative comes first; the input values follow from the narrative. The committee is asked to interrogate the narratives ("is this story plausible; is the path between the variables coherent; what would have to be true for the recovery scenario to land") rather than the individual numbers.
How the committee conversation changes
A committee presented with a tornado chart asks technical questions: is the range on demand growth reasonable, is the working-capital sensitivity calibrated to recent experience. These are useful questions. They are also the questions an analyst can answer from the desk.
A committee presented with four to six structured worldviews asks different questions: which scenario does the management team consider most likely, which scenario is the capital structure designed to survive, what is the trigger that would move us from the operational downside into the combined stress. These are governance questions, and they are the ones the committee is actually paid to ask.
The chair of one investment committee put it this way in a debrief: "the tornado told us what could move the number; the scenarios told us what we were betting on." That is the right distinction. The tornado is a diagnostic tool. The scenarios are a decision tool. Confusing the two is the error this note is written to address.
Practical construction notes
A few points on how to build the architecture in practice.
First, the scenarios must be driven from a single input panel, with each scenario selectable via a toggle. If switching scenarios requires manual override of multiple cells, the structure has not been built; a set of saved snapshots has been. The model must be capable of producing any of the scenarios from the same logic, instantly, with the changes traceable.
Second, the joint moves must be specified, documented and defensible. A scenario that pairs a 30 per cent commodity rise with an unchanged currency is incoherent in most contexts; the model should not permit such a pairing silently. Where the relationship is uncertain, the documentation should say so.
Third, the second-order operational effects must be modelled, not asserted in the narrative. If the macro downside narrative says supplier terms tighten in response to input cost pressure, the model should propagate that move through the working-capital module, not leave it as a sentence in the appendix.
Fourth, the recovery scenario deserves the same attention as the downside. Most committees spend their time on downside stress and almost no time on the path back. In capital structure work in particular, the recovery scenario is what tells you whether the structure is viable or merely solvent.
The point of scenario work is not to estimate the distribution of outcomes. It is to give the committee a small number of internally coherent worldviews that it can interrogate, compare and choose between. The tornado chart cannot do this; the layered architecture can.
Sensitivity tables retain a role. They are the right tool for assumption diagnostics, for identifying where to spend analytical effort, and for communicating, in compact form, which inputs the model is most exposed to. They are the wrong tool for presenting risk to a committee that is expected to take a view. Layered scenario architecture is the discipline that closes that gap.