Energy: a tax-equity flip, with the date derived

A tax-equity partnership whose flip date is derived from the investor's return rather than stated, reconciled against an external model.

Every number below is checked against an independent reference implementation on every commit — period by period, and on each metric, inside a declared tolerance. See benchmark methodology.

The case

A 100 MW-ac solar project financed through a tax-equity partnership. A tax investor funds 98% of the equity and takes 98% of the cash, the depreciation and the investment credit. When its after-tax return reaches 8%, the structure flips: the investor drops to 5% and the sponsor takes the rest.

The flip is not a date. It is a test, and when it lands depends on how the project performs — so the model derives it rather than stating it.

The reference

A national laboratory's open-source project-finance model, run in its leveraged-partnership-flip configuration. It publishes the flip year alongside both partners' cash, so the derived date is checkable and not only the split.

Not vendored. The tool was run once outside the repository and only its output numbers were carried across.

Provenance. The expectations are the output of an independent published tool run on the case's inputs. Nothing else external asserts them.

What it exercises

The flip sits on the partnership interest rather than the plant: the panels do not change when the structure flips, the claim on their cash does.

The test needs no solver. The criterion is an internal rate of return reaching 8%, and at a fixed hurdle the two statements are one:

IRR through period n >= 8%   <=>   NPV at 8% through period n >= 0

A discounted running sum is a recurrence, so the test is arithmetic evaluated once a period. Nor can it be circular: the test at period t reads flows through t-1, and every one of those periods is still pre-flip by construction, so the sharing percentages it depends on are settled before it is evaluated.

The result

The flip date is derived, and it agrees. The transition fires at period 4 against the reference's stated flip in year 3 — the same instant under the deal's own convention, where the year's books close, the return is tested, and the new sharing applies to the year that follows.

Both partners' cash reproduces across all 25 operating periods, through the flip and through the debt cliff at periods 18 and 19.

Asserted: two columns period by period, plus the transition itself.

The delta

Worst disagreement across all 25 periods and both columns: 1.0e-6 dollars, which is the engine's own publication precision rather than a convention difference.

Period 0 is not asserted. The reference books a sponsor development fee at close that this case does not model, and it has no bearing on the flip.

A variant the reference does not publish

The same deal on a monthly grid flips ten months earlier, in the second month of year 3.

By the end of year 2 the investor is $445,000 short of its hurdle, and two months of operating cash clear it — but an annual grid has no period between month 24 and month 36 in which to notice, so the event cannot fire until the next year end. The investor keeps 98% of the cash for ten months it was no longer entitled to, worth about $3.5m here.

The grid is therefore an economic assumption whenever an event decides who gets paid, not a presentation choice. No external source publishes the monthly answer, so it is carried as a fixture rather than as a benchmark.

Run configuration

{ "deterministic": { "annual_discount_rate": 0.08 } }

Verified results

Checked period by period: 2 series across 25 periods — 50 values in all, each within ±0.01 of the reference.

• partnership.distribution.investor
• partnership.distribution.sponsor