P50 / P90

What P50 and P90 actually mean

For a grid-scale storage project, the financial model produces not a single revenue number but a distribution of possible annual results. That spread comes from uncertain inputs: wholesale energy and ancillary-service price forecasts, achievable cycles per day, plant availability, auxiliary load, and the capacity-fade trajectory of the battery racks.

Running thousands of Monte Carlo scenarios, or applying analytic exceedance curves, yields a cumulative probability curve for the chosen metric. P50 is the 50th-percentile point on that curve; P90 is the 90th-percentile downside point. Both are read off the same distribution, so they are only comparable when the metric, year, and assumptions match.

The "P" is an exceedance probability, not a probability of occurrence. P90 does not mean "90% likely to happen"; it means there is a 90% probability the actual result equals or beats it, and only a 10% chance of falling short. By symmetry, an upside P10 case, with only a 10% chance of being exceeded, is sometimes shown to bracket the range for investors. P50 is the median, which for a skewed revenue distribution is not the same as the mean: a merchant BESS with fat upside from scarcity pricing can have a mean expectation well above its P50.

The convention is inherited from wind and solar resource assessment, where the distribution reflects interannual weather variability around a measured resource. A stationary BESS has no resource to measure; its distribution is dominated by market outcomes, so the exceedance horizon matters even more.

A one-year P90 captures a bad single year, while a ten-year P90 averages good and bad years across the debt tenor and therefore sits much closer to the P50. Lender term sheets specify both the percentile and the horizon, for example "P90 one-year" for near-term stress and "P50 ten-year" for the base case.

Why it matters in a real grid-scale project

Debt sizing is the headline consequence. Lenders set the maximum loan so that even in the conservative case the project still services debt with margin to spare, applying a minimum Debt Service Coverage Ratio to the P90 (sometimes P95 or P99) cash flow rather than the P50.

A wide gap between P50 and P90 signals high revenue uncertainty, forces lower Gearing / leverage, and pushes more of the capital stack onto sponsor equity. Equity investors meanwhile run their Internal Rate of Return and Net Present Value on the P50 case, so the same Discounted Cash Flow model is read at two different percentiles by two different audiences.

For merchant or revenue-stacked storage, the P50-to-P90 spread is dominated by price-forecast risk, which is far larger than the spread on a contracted offtake such as a tolling agreement or a fixed capacity-market payment. This is why a signed long-term contract tightens the distribution, narrows the P50/P90 gap, and unlocks cheaper, higher-leverage debt for the same physical asset. The effect flows through to the Discount rate as well: a tight, contracted distribution reads as lower risk and supports a lower cost of capital than a wide merchant one.

Typical values and conventions

Lenders commonly require a P90 DSCR on the order of 1.20 to 1.40x for well-contracted storage and materially higher, often 1.5 to 2.0x or more, for merchant revenue, with the exact threshold tied to the contracted share of cash flow. An alternative merchant convention applies a haircut instead, sizing debt on only a portion, commonly around 50 to 80%, of P50 merchant revenue. Some structures test a P90 one-year for near-term cover alongside a P90 ten-year or a P50 average across the tenor; always confirm which exceedance basis and which horizon a given P-number refers to.

On typical spreads: for a largely merchant BESS, independent forecasts commonly place annual P90 revenue roughly 15 to 40% below P50, depending on the market, the revenue mix, and how much of the stack is capacity- or contract-backed; a fully tolled project can see the gap collapse to a few percent, driven mainly by availability and degradation rather than price.

There is no governing standard that defines P50/P90 for storage; it is a modelling convention policed by lenders' independent engineers and market consultants, so two consultants can legitimately produce different P50s for the same plant from different price scenarios.

Keep the statistical P-curve separate from the engineering warranty and the degradation model. Capacity guarantees from the cell or system supplier define a contractual floor: for LFP systems, end-of-warranty usable energy is typically guaranteed near 65 to 70% of beginning-of-life nameplate after a defined cycle and calendar count, with availability guarantees commonly 95 to 98% measured annually, up to about 99%, and round-trip efficiency guarantees in the mid-to-high-80s percent AC-AC.

The financial P90 instead reflects the probabilistic spread around the expected case. Both feed the model, but they answer different questions.

How it shows up in specs, studies and contracts

A working engineer meets P-values in the independent engineer's revenue and energy report, in lender term sheets ("debt sized to a 1.30x DSCR on P90 one-year cash flow"), in market consultants' price forecasts delivered as P50/P90 curves per year, and in tax-equity or investor models.

The first questions to ask of any P-number: which metric (delivered MWh, gross revenue, EBITDA, or cash available for debt service), which year or averaging horizon, what distribution and correlation assumptions produced it, and whether degradation, availability, and auxiliary load sit inside the Monte Carlo or are deterministic overlays.

The engineering inputs are where a BESS engineer can actually move the curve. Usable energy at the point of interconnection, net of PCS, transformer, and auxiliary losses, must match what the revenue model dispatches; the model's cycles per year must fit inside the warranty's cycle allowance and the augmentation plan; and the degradation trajectory must be the same one the capacity guarantee is written against. A mismatch, such as a revenue model cycling at 500 full cycles per year against a warranty priced for 365, quietly makes the P50 unachievable and the P90 a fiction.

Common pitfalls

The most frequent error is comparing P-numbers on different bases. A P90 of delivered energy, a P90 of gross revenue, and a P90 of net cash flow are three different statements, and an annual figure is not a tenor average. Correlations matter too: high prices often coincide with heavy cycling and faster fade, so treating price, throughput, and degradation as independent random draws understates the true downside and flatters the P90.

A subtler basis question catches anyone arriving from cost estimating rather than project finance: which end of the distribution the conservative case sits on. On a quantity you want large — revenue, delivered energy, cash available for debt service — the cautious number is the one you exceed nine times in ten, so the P90 lies below the P50, and that is the sense used throughout this entry.

On a quantity you want small, such as an installed cost or a schedule duration, the cautious number is the one you stay under, so it lies above the median instead, and a P90 quoted against it may well be a non-exceedance percentile rather than an exceedance one. The label alone does not say which. Establish the metric, the horizon and the direction before setting two P-numbers side by side.

Second, a P90 is not a guaranteed floor; roughly one project year in ten is expected to land below it, and the number is only as good as the underlying distribution. Merchant price forecasts from different consultants routinely differ by more than the P50-to-P90 spread itself, so sensitivity-test across forecast vendors, not just across percentiles. Finally, do not let the statistical P90 substitute for the contractual capacity floor, or vice versa: one is a probabilistic modelling output, the other an enforceable warranty term.