Building the capital stack and closing the gap — California

What actually happens, and why it is not sequential

Capital stack construction is not a discrete phase. It runs from roughly 60 days after site control through construction loan closing — typically 24 to 48 months — as three interleaved workstreams.

The decisions inside the phase are simultaneous, not ordered. Bond size changes basis, basis changes credit, credit changes equity, equity changes the gap, the gap changes which soft sources you need, and some soft sources cap you as a percentage of total development cost, which loops back into the gap. This is why the work lives in Excel with circular references enabled. You do not solve it once; you iterate to convergence and then re-converge every time a number moves.

The scarce resource in 2026 is state credit, not bond cap

The federal 4% credit is formulaically available to anyone who clears the bond test. Bond authority is not the choke point it was.

Aggregate supply modestly exceeded aggregate demand — but that headline is misleading and should not be used as a planning input.

Compare demand by pool and region against supply by pool and region, never statewide totals.

State credit is where it binds.

CDLAC Regulations Section 5106 opens by voiding the bond allocation outright: a project shall not receive a bond allocation if it requested and is not scheduled to receive an award of State Tax Credits. There is no take-the-bonds-now, get-state-credits-later path. Requesting state credits is a risk position, not a free option. HCD's emergency MHP Gap NOFA of roughly $240,000,000 exists explicitly because state credit competitiveness stranded construction-ready projects.

Bond sizing under the amended test

Two traps. The denominator is building basis plus land, not eligible basis — land is excluded from eligible basis but included here, and getting it wrong over- or under-sizes the bond. And the 25% path is conditional: a deal refunding pre-2026 paper cannot simply drop to 25%.

The effective date is contested inside the corpus. One research pass reads Section 70422(b)(2) as applying to buildings placed in service in taxable years beginning after December 31, 2025; another read the statutory text, did not find that condition, and recommends not implementing until confirmed; two others state the test with no placed-in-service condition at all. Read the act's effective-date section before you size to it.

Model the feasible band between the federal minimum and the state ceiling and flag when the three constraints conflict. Do not render any of it as tax advice — Section 5108 itself says CDLAC may rely on the tax opinion submitted with the application.

Separately, IRC Section 42(h)(3)(I) as amended by Pub. L. 119-21 Section 70422(a)(1) multiplies the per-capita and small-state minimum amounts by 1.12 for calendar years beginning after December 31, 2025, permanently. That is 12%, not the 12.5% much of the trade press reports; 1.125 was the expired 2018–2021 temporary factor this replaced.

One underwriting rulebook, mostly

CTCAC's CCR Title 4 Section 10327 (current version adopted December 10, 2025) is the arithmetic, and HCD incorporates it by reference. Built correctly to Section 10327, one engine serves both the credit application and the state soft-money application.

DSCR is where the two agencies meet.

Whether these collide is disputed. One source in the corpus presents the band between HCD's 1.20 ceiling and CTCAC's 1.15 floor as empty; the reconciliation pass calls that wrong — 1.15 to 1.20 is a non-empty range, the tests do not bind the same year, and DSCR rises under the prescribed trending, so a year-one 1.12 can satisfy HCD and reach 1.15 by year two or three. Treat the interaction as a structuring problem to work through, not an automatic kill.

Developer fee — the number that decides the deal, and the one the sources disagree on

The 4% structure is the one that gets modeled wrong. The fee in basis is uncapped — 15% of unadjusted eligible basis plus 15% non-residential — but the cash portion is capped and everything above it must be deferred or contributed. Three different renderings of that cash-out formula appear across the research corpus.

The reconciliation pass found that this rendering omits a $6,000,000 ceiling and overstates cash fee by $4.67M — 78% — on a $200M-basis deal, an error invisible in the $60M worked example all three renderings used. Do not model cash-out from prose. Reimplement CTCAC's published dfc.xlsx cell for cell and reconcile against a real awarded deal. Note that the $6,000,000 ceiling expires December 31, 2028.

Whether HCD's own limit binds is also unresolved. One source says HCD defers entirely to CTCAC through MHP Guidelines Section 7305(b); another documents an independent $3.5M cap at 25 CCR Section 8312. Read UMR Section 8312 and Section 7305(b) side by side before assuming a single fee engine serves both applications.

Then the squeeze — deferred fee must be repayable from cash flow within roughly 15 years to stay in basis, while Section 10327(g)(6) caps years 1–3 cash flow. Fee deferral needs cash flow; CTCAC caps cash flow. Surface that as a binding constraint, not a footnote.

HCD Administrative Notice 24-03 (June 19, 2024) states flatly that HCD will not allow an increase in developer fees for awarded projects, and that awards may be reduced if the project secures additional funding, to avoid supplantation. Check the vintage of any HCD notice you rely on: Administrative Notice 19-01 still quotes superseded $2.0M/$2.2M caps.

The California soft-money map

The 2026 MHP Gap NOFA is the most diagnostic document in the domain.

The funding-priorities formula is published verbatim: points = 20 − 19.99 × (MHP requested per adjusted restricted unit ÷ 250,000).

The lowest-income scoring is computed on units at 80% AMI or below, explicitly for alignment with TCAC and CDLAC scoring — even though MHP itself assists only units at 60% AMI or below.

AHSC is governed by the Strategic Growth Council with HCD administering, funded from cap-and-trade.

IIG funds infrastructure rather than units, so it does not compete against the same per-unit caps — one of the cleanest stacking partners in the state toolkit.

The calendar is the binding constraint

The dependency chain is the thing to model, because it is where deals actually die. The CTCAC or CDLAC deadline is fixed. It requires enforceable financing commitments — CDLAC wants evidence of commitments for at least 50% of acquisition and construction financing, or at least 50% of permanent financing. Those require soft-money award letters. Those require the HCD MFSN or AHSC round to have closed and awarded. And those rounds are annual or less frequent, and irregular.

The 2026 MHP Gap NOFA is the shape of it: a six-week application window and a roughly three-month adjudication. Miss it and you wait a year, during which construction pricing, your interest rate, and possibly your site control all expire.

Documents move underneath you. The 2026 MHP Gap NOFA was amended twice in six weeks.

Real money, unpublicized, gone in under four weeks. Watching agency publication pages is not administrative overhead in this phase; it is part of the work.

The inputs nobody can source for you

Equity pricing is paywalled. The two canonical series — the CohnReznick Housing Tax Credit Monitor and Novogradac's LIHTC equity pricing page — are subscription products. Secondary reporting places 2025–2026 national pricing broadly in the mid-$0.80s to low-$0.90s per credit dollar with California typically at the high end, but that is not a primary-source figure and should not be treated as one. Carry price as an input with an explicit sensitivity range, never as a constant.

Operating expenses are worse, because a regulatory floor is masquerading as a benchmark. Section 10327(g)(1) makes CTCAC's published minimums binding and UMR Section 8311(c) imports them into HCD underwriting — but CTCAC's 2026 application page links a file named 2022-operating-expense-minimums.pdf, byte-identical to the 2022 original. Insurance, the fastest-growing line, has no public data at all in a market where affordable-housing premiums reportedly rose 50% to 500% between 2022 and 2024. A pro forma built to clear the CTCAC floor passes the agency and fails in year three. Own your expense number; treat the published minimum as a floor to clear, not an estimate.

Senior debt conventions are the same story in reverse: the regulatory constraints are codified (DSCR bands, balloon rules, ceiling-rate treatment for variable debt), but amortization, spread and term for CalHFA, Freddie TEL or FHA 221(d)(4) come from term sheets, which are relationship documents. Take them as inputs.

And several things live entirely in relationships: which bond issuer will actually take the deal and on what timeline; whether the city writes its local commitment letter before the CDLAC deadline; whether HCD exercises the sole and absolute discretion it reserves to exceed the $15M maximum; whether a given syndicator is currently pricing your submarket.

Two long-tail uses belong in the stack from day one because they are routinely omitted from early models.

On the exit side, the qualified-contract exit is legally unavailable in California under H&SC Section 50199.14(f) and 4 CCR Section 10338(h). Any residual value assumption based on year-15 market-rate conversion is not aggressive; it is impossible. Out-of-state capital gets this wrong constantly.