Testing price, volume and input cost one variable at a time produces three answers, none of which describes the actual downside. The variables correlate, and the combinations that break a covenant are usually not the ones a single-variable sensitivity finds.
Why single-variable sensitivities understate downside
A single-variable sensitivity moves one input, typically price, volume or an input cost, while holding every other assumption at its base case value, and reports the resulting change in a coverage metric. This produces a clean, easily communicated number, but it also implicitly assumes that when price falls, volume and input costs stay exactly where the base case predicted, which is not how these variables behave in practice.
In most commodity-exposed projects, a fall in output price coincides with conditions that also affect volume and cost, whether through reduced demand, operational curtailment, or currency movements tied to the same macro driver. Testing price alone therefore describes a scenario that has a low probability of occurring in isolation, and it systematically misses the combinations that actually threaten a covenant.
Three separate single-variable results, run and reported side by side, give the appearance of thorough downside testing without ever describing what happens when two or three of those variables move together, which is the condition most likely to produce the coverage failure a lender is actually trying to identify.
Correlation between price, availability and recovery
Price, plant or asset availability, and recovery or yield rates are rarely independent in a commodity project. A price decline driven by oversupply, for example, often coincides with operators across the sector deferring maintenance to preserve cash, which affects fleet-wide availability. A grade or ore quality shortfall that reduces recovery can simultaneously push processing costs up, compounding the effect on margin rather than acting alone.
Building the correlation structure between these variables requires going back to the historical data underlying each assumption, rather than assuming independence for modelling convenience. Even an approximate correlation, established through discussion with technical advisors and calibrated against historical periods of stress, is more informative than treating each variable as if it moves in its own sealed compartment.
Building the matrix and choosing the axes
A scenario matrix tests combinations of two or more variables simultaneously, typically presented as a grid with one variable on each axis and the resulting dscr or another coverage metric populating each cell. The choice of axes matters: they should be the two or three variables the technical and market advisors have identified as most likely to move together and most material to cash flow, not simply the variables that are easiest to model.
For a mining or industrials project this is often price and recovery rate; for an energy project it is more often price and availability, with input cost as a third axis where the project has meaningful exposure to a specific feedstock. Reducing the matrix to the axes that matter, rather than including every variable in the model, keeps the output legible enough for a credit committee to use.
The matrix should span a wide enough range on each axis to include the historical extremes for that variable over a period comparable to the facility tenor, since a matrix built only around modest variation around the base case will not surface the combinations that matter.
Naming the breaking point rather than presenting a range
The purpose of the matrix is to identify the specific combination of inputs at which coverage falls below the lock-up test or default threshold, and to name that combination explicitly rather than leaving the reader to infer it from a grid of numbers. A statement such as 'coverage breaches at an 18% price decline combined with a 6% recovery shortfall' is far more useful to a credit committee than a range of outcomes across the full matrix.
This also allows the borrower to test that specific combination against historical frequency, addressing directly whether the breaking point has occurred before in the relevant market and, if so, how often and for how long, which is the question a credit committee is actually trying to answer when it asks for downside analysis.
How lenders read a matrix differently from a tornado chart
A tornado chart ranks variables by the size of their individual impact on a metric, holding all else constant, and is useful for identifying which assumptions deserve the most diligence attention. It does not, however, show what happens when those variables move together, and a credit committee reading only a tornado chart can reasonably conclude that no single variable is dangerous enough to threaten the facility.
A matrix forces the same committee to confront combined outcomes directly, and in Zenith's experience it changes the questions asked in credit committee meetings: away from 'how bad could price get' and toward 'what combination of conditions would actually cause a breach, and how plausible is that combination'. That is a more precise conversation, and it tends to produce covenant packages that are calibrated to genuine risk rather than to a single variable's isolated worst case.
Last reviewed December 2026
Start with the structure, not the pitch.
Tell us the transaction and the timetable. If it is not something we should run, we will say so.