P50 and P90

Why a yield is a distribution

A yield report does not deliver an annual quantity so much as a statement about how certain that quantity is. The P50 figure is met or exceeded in every second year on average; the P90 figure in nine years out of ten. For solar the gap between them usually sits in the low double digits in percentage terms.

The spread has two sources worth separating. The first is weather: irradiation in a single year departs from the long-run mean, and that share shrinks the longer the period considered. The second is the model: uncertainty in the irradiation dataset, in the transposition to the module plane, and in the loss assumptions. That share does not shrink with time, it shrinks with better data.

Who uses which figure

A lender does not size debt service on the yield available in half of all years. It computes the debt service coverage ratio at a conservative point of the distribution, typically P90 over one year or P75 across the loan tenor, and the amortisation profile follows that calculation. The equity side works with P50 alongside it, because that is the expected value.

Both calculations are correct and they answer different questions: P90 asks whether the loan is served in a bad year, P50 asks what the project earns on average. A pack showing only one of them leaves the other side's question open.

What makes a report bankable

Three things: the irradiation source with the period it covers, the loss assumptions itemised from soiling through to availability, and an uncertainty calculation the percentiles are derived from. A report without that calculation states a number, but it does not state a P90.

Who wrote it belongs in the assessment too. A report from the contractor building the system is a seller's statement and underwriting treats it as one. An independent report costs a fraction of what a haircut on the assumed volume costs across the tenor.

The most common flaw in a project pack is not an inflated yield, it is a yield with no percentile attached. The moment someone asks which figure it is, everything built on it comes apart: debt service, coverage and distributions all hang on that one label.

In practice that means carrying both figures side by side and running the ratios twice. A project whose coverage only works at P50 is not a financeable project with an optimistic report; it is a project with too much debt in it.

The second point is degradation. Percentiles describe a single year, usually the first. Across twenty years module degradation compounds on top, and a model that rolls forward year-one output overstates exactly the late years in which debt service is still running.