ADU financing statistics

How California ADUs were actually paid for

The survey’s loan shares are nested, and nesting is where roundups go wrong. Sixty-six per cent is not sixty-six per cent of ADU owners — it is sixty-six per cent of the 43% who took a bank loan at all.

Resolving those to shares of all surveyed owners is arithmetic on the survey’s own denominators, and it changes the picture considerably: the products the category writes about most were used by a small minority of the people who built.

Chapple, Ganetsos and Lopez, Implementing the Backyard Revolution, Center for Community Innovation, 22 April 2021, of more than 800 California homeowners with 2018 or 2019 permits or certificates of occupancy.

Product shares exceed 100 per cent within the borrower group because some owners used more than one product, so the resolved column should be read the same way — as overlapping shares, not a partition.

HyreADU analysis. The single most quoted ADU financing statistic is “66 per cent used a HELOC,” and it is true of a subgroup that was itself a minority.

Across the whole survey, roughly 28.4% of owners used a home equity line and roughly 3.0% used a construction loan — the product type that, on a slide, looks purpose-built for this job.

That gap between the theoretical product and the observed behavior is the finding the Terner Center’s 2022 financing paper named operationally: renovation loans were expensive, slow, denied at high rates and disliked by contractors waiting on draws.

The rules, as the three agencies publish them

The public conversation has caught up with one half of the barrier the Terner Center identified: all three agencies now recognize accessory-unit rent as qualifying income.

The half that still decides construction deals is that the income has to come from a unit that exists, or from rent that starts before the first payment on the new mortgage.

Published GSE and FHA treatment of accessory-unit rental income, retrieved 2026-09-05. Every cell is quoted or paraphrased from the current guide, fact sheet or letter. A guide describes what an agency will buy or insure. It is not what a particular lender will originate.

Fannie Mae’s UAD 3.6 expansion, effective 31 March 2026 for UAD 3.6 appraisals, allows up to three ADUs on a one-unit property. That is a property-eligibility change, and it does not lift the “one existing ADU” limitation in the rental-income chapter. Confusing the two is the most common current error in ADU financing coverage.

What the 30 per cent cap actually does

The worked cap arithmetic below is the arithmetic the category does not publish, and it is ours. Two published numbers govern how much of an ADU’s rent reaches a debt-to-income calculation: a haircut of 75 per cent, and a cap of 30 per cent.

The Guide language caps accessory-unit income at “30% of the total qualifying income,” and the base of that percentage is not pinned down unambiguously in the text we hold — so we publish both arithmetically available readings rather than choose one.

Reading A takes the base as the borrower’s other qualifying income; reading B takes it as total qualifying income including the counted ADU amount. Reading A is the tighter of the two and is the conservative planning assumption.

HyreADU calculation on published rules. Reading A: countable income = min(0.75 × rent, 0.30 × other income); the cap binds once rent exceeds 0.40 × other income.

Reading B: the cap resolves to 3 ÷ 7, or about 0.4286, of other income, and binds once rent exceeds about 0.57 × other income.

We do not reproduce any agency’s worked example, because the one in circulation could not be reconciled arithmetically on our retrieval, and repeating an unreconciled figure is exactly the failure this desk exists to avoid.

HyreADU analysis, and it is the point of the whole page. The cap does not bite the borrower with a modest rent and a strong income; for them it is irrelevant and the barrier is equity, rate or the renovation-loan operations.

It bites precisely the borrower whose other income is the reason they need the ADU’s rent in the first place.

“The ADU will pay for itself” is not an underwriting sentence: a unit renting for enough to service a six-figure construction loan will not contribute that figure to a debt-to-income ratio.

It will contribute, at most, 30% of the borrower’s other income, after a haircut, on a transaction the guide allows, from a unit that already exists.

The cap against real California rents

Abstract percentages are easy to nod at, so here is the cap against the only published rent series that covers every California metro on a common definition: HUD’s FY 2026 Fair Market Rent for a one-bedroom unit.

Fair Market Rent is a 40th-percentile policy construct used to set voucher payment standards, not observed ADU rent — but it is a defensible, dated, agency-published figure of roughly the right size for the one-bedroom units that were 61 per cent of the 2021 survey’s sample.

FMR dollars are agency-published (HUD FY 2026 Fair Market Rent documentation, retrieved 2026-09-05; Los Angeles uses the 21 May 2026 revision). The haircut, the countable amount and the income threshold are HyreADU calculations under reading A: countable = min(0.75 × FMR, 0.30 × other income), and the threshold is 0.75 × FMR ÷ 0.30.

Read the last column carefully, because it is counter-intuitive. The higher the local rent, the more other income a borrower needs before the cap stops discarding part of it.

A San José one-bedroom at $2,982 is fully countable only above about $7,455 a month of other qualifying income; a Sacramento one-bedroom at $1,832 clears at about $4,580.

The cap is hardest on exactly the combination — expensive market, modest other income — where the rent would have mattered most.

Fair Market Rent is not an ADU lease, and the rent digest keeps that distinction; the point of this table is the shape of the cap, not a forecast of anyone’s rent.

The public money, sized against what an ADU costs

Named public programs are program fact with a funding status, not offers. The status column is what the administrator’s own page said on the retrieval date. The final column is ours: the maximum award set against a construction median, because a grant’s usefulness is not its headline number but its share of the job.

Program terms and statuses as published by each administrator and retrieved 2026-09-05. Construction medians are the 2018–19 owner survey indexed to August 2026 on the DGS California Construction Cost Index (factor 1.536) — see the cost study. The share column is HyreADU arithmetic and is the honest way to read an award ceiling.

A state department’s funding directory described the CalHFA grant in the present tense on the same retrieval date on which the administering agency’s own pages said it had been fully reserved for two years.

Where a directory and an administrator disagree, the administrator is the source of record, and a present-tense description on a third-party page — even a government one — is not evidence that a round is open.

What the grant program actually delivered

CalHFA filed an outcome report in January 2025 under Health and Safety Code § 51532, and it contains the only published account of where the money went.

The snapshot below is dated 20 November 2024. The left column is the Agency’s; the right is what that figure implies once it is set against another published figure.

CalHFA ADU Grant Program outcome report, January 2025, filed under Health and Safety Code § 51532; snapshot dated 20 November 2024. Retrieved 2026-09-05 for the financing study. The implication column is HyreADU arithmetic on the Agency’s own published figures.

The award-level record is not public, and that limits everything above. A separate CalHFA research note attempted to obtain the grant file at census-tract level — to test whether the subsidy reached the households it was designed for — and could not retrieve a single award record from any state source on its retrieval date.

The figures on this row are program-level aggregates the Agency chose to publish. They cannot answer the distributional question, and we do not pretend otherwise.