What a site is worth partway through approval

Land bankers hold entitled lots for public homebuilders at yields they publish each quarter. Developers pay option money for the right to buy a parcel once it clears rezoning. Councils reverse approvals and juries set the damages. Each of those is a price on the same thing: what a site is worth at a given point along the path from raw land to a building permit.

Two systems meet that question and set it aside. A residential appraisal report asks whether the present use is the highest and best use and, in most reports we review, answers yes with a template sentence. Automated valuation models learn from closed sales and tax records, neither of which records an approval stage, a condition of approval or an expiration date, and the rules that route a loan to a model skip the zoning checkpoint that the appraisal path requires.^3,4

This feature builds the missing piece in the open. It walks a composite infill site through four milestones, prices each with a stated formula, and then fills the inputs with evidence: what approval costs, how often it fails, how a path closes after it is won, and how long permits take in the cities that publish usable data. Every illustrative number is labeled as one. The two original data runs, on California's housing progress reports and New York's zoning board records, come with their method and their limits.

USPAP's Standards Rule 1-3(b) requires an opinion of highest and best use when one is needed for credible results, and Standards Rule 2-2(a)(x) requires the report to summarize the support and rationale for that opinion.² The legacy form gave the answer one checkbox. UAD 3.6 replaces it with four fields, one per test, plus commentary, which our SEL-2026-08 feature walks through with a Dallas residual land model.

Fannie Mae's own view is that the analysis is usually easy. Its September 2026 Playbook says that for most residential mortgage appraisals the analysis "is straightforward because the HBU is typically the property's present residential use as improved," and that it "becomes more important when a property's value potential includes an alternative use."⁶ Its FAQ adds that present use and highest and best use are "separate determinations."⁷ The ordinary house on the ordinary lot passes the four tests. The gap sits in the exceptions, where zoning has changed, a lot can be split, an accessory unit is possible or a use is legal nonconforming.

Reviewers have described the checkbox habit for years. Philip Spool wrote in 2014 that many residential appraisers "are content with checking the box 'yes'" and that a single paragraph covers all four tests.¹⁰ Russell Barrows wrote in 2018 that the analysis is "more than just checking the appropriate box on the form," and that he has reviewed reports citing residential zoning in towns that have no zoning at all.¹¹ Illinois disciplined one appraiser in 2002 for a report that "misrepresented the zoning of the subject property" and another who "merely gave the definition of Highest and Best Use without doing the necessary analysis."¹² The Appraiser Qualifications Board's criteria keep "subdivisions for which a development analysis/appraisal is necessary" outside the Licensed and Certified Residential scopes, and the required highest and best use coursework lists general topics with nothing on entitlement.¹ The only training we found that prices a site mid-approval was a November 2025 Appraisal Institute litigation session.⁴⁴

The practitioner method for a development site values it as fully entitled and then deducts what remains: the cost still to be spent, the time still to run and a return on the money tied up. Thomas Jensen's article on appraising development sites lays that out for two end states, entitled and unentitled, and stops there.⁴³ Real options research has priced the timing option in land for decades, with Laura Quigg finding an average option premium of about 6 percent of intrinsic value across 2,700 Seattle land sales,⁴² but it prices a single timing option rather than discrete approval stages. The published US sources we read stop at the two end states, so this model fills in the stages between them, with every assumption shown.

V_i = P_i × (V_E − C_i) × D_i + (1 − P_i) × V_R

At stage i, P is the odds of reaching full entitlement from there, V_E is the entitled value, C is the cost still to spend, V_R is raw land value and D discounts the entitled branch for the years still to run at a required rate. Money already spent stays out of the value. It is equity at risk for the owner or the lender, and the table reports it on its own line. A lapse returns the site to V_R and writes that equity off.

The California-style preset uses an entitled value of $1,100,000, a raw value of $500,000, a 15 percent rate and three years from filing to permit. Discounted, the site is worth $514,307 the day the application is filed, $674,669 once rezoning is approved, $884,130 when the final plat clears and its appeal period ends, and $1,084,100 at permit. Switch discounting off and the filing-stage figure rises to $665,000. Three years of carry at 15 percent eat most of the filing premium, so an application with a long clock ahead of it can sit near raw land value even with decent odds.

The market gives the later stages a price. Millrose Properties, the land bank spun out of Lennar, reports a weighted average yield of 8.5 to 8.6 percent on about $6.4 billion of homesites held under option.²⁸ Walton Global's public land banking fund targets 9 to 11.5 percent, with builder takedowns two to four years out.²⁹ Forestar's arrangement with D.R. Horton lets it sell undeveloped land back within about a year at carrying value plus a 16 percent annual return, which prices a shorter and different risk.³⁰ Builder option deposits run about 6 to 20 percent of the purchase price across the filings we read, with the downside capped at the deposit.³¹ Those yields fit the carrying cost of the last two rungs. They say little about the filing stage, where the odds, the cost and the clock all remain open.

Remaining cost is a stack rather than a number. Impact and development fees are the largest layer in the states that charge them. The Terner Center's study of seven California cities put total development fees at 6 to 18 percent of the median new-home price, with Fremont charging $35,000 per single-family house,³² and the Duncan Associates survey put California's average impact fee on a standard house at $31,014, roughly double Maryland's.³³ A full environmental impact report adds $150,000 to $300,000 in consultant fees on top of a state filing fee of about $4,100, a layer a by-right project skips.

The refund rules sort the layers. Building permit fees are commonly 75 to 90 percent refundable when a permit is canceled before work begins. Rezoning and entitlement application fees are kept on denial in the codes we read, and plan check and environmental review fees stay spent once the work is done. That is why sunk spending sits on the equity line and stays out of the market value.

The best tracked program we found is Massachusetts Chapter 40B. Lynn Fisher's study followed 404 projects in 115 communities filed from 1999 to 2005. Local boards approved 81 percent and denied 19 percent. Developers appealed 90 percent of the denials, and 69 percent of those appeals produced a permit through a state ruling or a settlement. Roughly 90 percent of projects were approved in some form. By mid 2008 only 62 percent of the approved projects held a building permit.¹⁴

New York tells the same story at the rezoning stage. The Citizens Budget Commission tracked 171 private rezoning applications filed from 2014 to 2017 and found 60 percent approved after a median of two and a half years, with the rest withdrawn, pending or denied in proportions the summary does not split.⁴¹ A 2026 working paper that priced a standard 20-unit project across 60 cities found the chance of approval within six months fell from 71 percent in 2000 to 44 percent in 2025.¹⁵ That measures speed inside a window, which is a different thing from the chance of ever finishing.

Three adjustments stand between any of those rates and the P in the formula. Withdrawals have to be counted as failures, because an applicant who quits after two years lost the same money as one who was denied. The rate has to be rescaled for how discretionary the subject jurisdiction is compared with the one it came from. And the window has to match the stage: the odds of finishing from a filed application differ from the odds from an approved rezoning. The presets in the calculator start at 55 percent for a California-style discretionary site and 80 percent for a Texas-style by-right site. Both are assumptions to test, with the evidence here as the range.

California requires every jurisdiction to report each housing project's entitlement, building permit and completion dates to the state, and the state publishes the table.²¹ We downloaded all 927,006 rows for 2018 through 2025, kept rows with a single parcel number, took every entitlement approved from 2018 to 2022, and looked for a building permit on the same parcel in the same city dated on or after the approval. A second, separate implementation of the join gave the same Los Angeles and San Diego results to the decimal.

Los Angeles matched 69.0% of 4,732 entitlement rows, with a median of 183 days from approval to permit. San Jose matched 25.2%. For Los Angeles, San Diego and Oakland the match rate holds roughly flat whether a cohort has had two years of runway or seven, which says the gap is mostly real rather than a matter of time. San Francisco and San Jose fall sharply for recent cohorts, so part of their gap is projects still in motion. Even San Francisco's 2018 cohort, with seven years to show a permit, matched only 54.3%.

A non-match has several explanations. Parcels get renumbered when a map records, multi-parcel projects are excluded by our rule, and data entry varies by city. Fresno matched 89.8% with a median of 15 days, which reads as entitlements and permits entered together, so it is left off the chart. The table shows that entitled and permitted are different statuses, and that the gap between them can be measured.

San Francisco's planning department keeps its own count. Its mid-2026 pipeline report lists 75,853 units, of which 61,749 are entitled and 3,327 are under construction. Adding the department's two "no permit filed" categories gives 47,444 entitled units, 77 percent of the entitled stock, with no building permit application on file.²⁰

The clearest dollar figure on a reversal comes from Alabaster, Alabama. A jury found a parcel worth $3,532,849.19 immediately before the city's December 2011 rezoning and $500,000 after it, an 86 percent drop, and the Eleventh Circuit affirmed the judgment in June 2021. The "before" figure was the developer's documented spending on the project, which the jury treated as the value lost.¹³ Most such suits fail; Alabaster won because the jury found the city acted unlawfully, and a referendum certified before the last discretionary approval usually leaves the developer with nothing to claim.

Litigation mostly adds time. Holland and Knight's review of 498 California appellate CEQA decisions from 2013 to 2023 found agencies winning 75 percent of appeals, 64 percent of challenged projects on infill sites and four to five years as the typical duration.¹⁶ For the model that belongs in the years-to-run and cost terms rather than the odds. Expiration is the third path. California tentative maps lapse 24 months after approval by default, with extensions capped at ten years.⁴⁰ Texas vests a project's rules at its first permit application under Chapter 245, but a project that goes dormant can lose that protection, and many city site plans expire on their own schedule.³⁹ The lapse row in the calculator is there for all three.

Miami-Dade County's auditor tried to benchmark its permit times against other large jurisdictions in 2022. Four responded with usable data, and the office concluded that a comparison "could not be conducted due to the inconsistent permit types and non-homogeneous data provided."¹⁷ We hit the same wall in open data: New York's issuance table shows a median near zero days because the review lives in a separate dataset, and Austin's addition permits inherit application dates from master permits filed decades earlier. So the panels below stay small and carry their method.

The most careful comparison we found is a March 2026 report by San Francisco's Budget and Legislative Analyst, which pulled department data from six cities and warned that each presents its data differently.¹⁸ San Francisco's 280-day median includes entitlement time from before 2024; once the city split the clocks, permits filed and issued from January 2024 through August 2025 ran a median of 114 days, with the planning department's entitlement stage running about four months on its own. Denver and Washington report averages rather than medians, so the chart marks them differently. Austin, at 91 days, and San Diego, at 134, sit at the fast end.

Job type matters as much as city. Los Angeles publishes every permit with its application and issue dates, so we computed 2024 medians for one and two family work: 238 days for a new house, 129 for an addition and 20 for an alteration or repair.²³ Discretion stretches the longer clock. A study of 350 Los Angeles multifamily projects found by-right projects reaching permit in a median of 477 days against 747 for projects that needed discretionary approval, and projects that required a full environmental impact report averaged 43 months to approval.¹⁹

New York City's Board of Standards and Appeals publishes every case with a real outcome field, which most permit systems lack. For the general variance under section 72-21, 1,983 cases since 1998 ended 73.6% granted, 19.0% withdrawn, 4.1% dismissed and 3.3% denied. For section 73-622, the special permit to enlarge a one or two family house that no longer meets the zoning, 631 cases ended 94.5% granted and 1.1% denied.²² Cases filed since 2015 took a median of 497.5 days to a grant for the general variance and 358.5 days for the house enlargement. Withdrawn cases ran a median of 501 days before they were pulled, about the same as the grants, so an applicant who gives up tends to do it late.

Historic review follows the same shape. Richmond, Virginia's Commission of Architectural Review decided 834 cases from 2018 through 2024 and denied 26, a 3.1 percent rate, while approval with conditions was the most common outcome in every year.²⁵ New York's Landmarks Preservation Commission issued only 55 percent of its discretionary certificates within its own ten-day target in 2024, against 88 to 98 percent for its routine permit types.²⁶ Accessory units show the gap between a statutory clock and a measured one. California gives cities 60 business days to act on an ADU application, about twelve weeks. San Diego's ADU permits issued in 2024 ran a median of 258 days from application, and Los Angeles ran 211 days for a new detached unit and 106 for a conversion.^23,24 Lot splits under SB 9 remain thin: Los Angeles had approved none of 28 applications by late 2022, while a few small suburbs approved most of theirs.²⁷

For an appraiser or a reviewer looking at a planned addition, a conversion or a lot split, the evidence points one way. Outright denial is uncommon in every board we measured, between 1 and 5 percent of cases. The cost sits in the six months to two years the process runs, in the conditions attached to the yes, and in the share of applicants who quit late. A statutory clock describes the law and says little about the calendar.

Fannie Mae's Selling Guide requires the appraisal to report the specific zoning class and to say whether the property is a legal conforming use, a legal nonconforming use, an illegal use or sits where there is no zoning, and it states that Fannie Mae "will only purchase or securitize a loan that represents the highest and best use of the site as improved."³ The eligibility rules for value acceptance, the program that replaces the appraisal with a model, list the loans that qualify and the ones that do not, and the list never mentions zoning, nonconforming use, accessory units or highest and best use.⁴ Vacant land and land development loans sit outside both GSEs' purchase rules altogether.^5,8 The checkpoint exists on one path by design.

A model learns from closed sales and property records. Dye and McMillen's study of Chicago-area teardowns found that once a house becomes a teardown candidate, its structural characteristics stop explaining its price, because the price is the land; teardown rates ran above 9 percent of the housing stock in Winnetka against about 3 percent in Chicago.³⁴ A model weighting square footage and bedroom count is modeling the wrong variables for that subset, and the subset clusters by neighborhood. Minneapolis's 2040 plan moved single-family prices about 3 to 5 percent, concentrated in land value and on the policy signal before adoption, in the one peer-reviewed study we found.³⁵ A model that learns from accumulating comparable sales catches that late, and in cities where capitalization was weaker it has nothing to catch.

The industry's own record points the same way. Of five AVM-industry comment letters on the 2024 quality control rule that we read in full, none mentions zoning, highest and best use, land value or teardowns, though HouseCanary asked regulators to require validation by property segment.³⁷ Clear Capital's glossary names "large acreage" among the traits that lower a confidence score.³⁶ The rule itself defines a dwelling as a one to four unit residential structure and says nothing about land.³⁸ Freddie Mac's appraisal guide requires at least one comparable with an accessory unit for a compliant ADU and two for a non-compliant one, and says that "to simply ignore the ADU due to lack of comparable sales is an unacceptable appraisal practice."⁹ Freddie Mac's collateral tool has carried a message since 2015 that fires when present use is not the highest and best use.⁴⁵ Each of those rules is written for the appraisal path. Published measures of model error for teardowns, upzoned parcels or lot-split candidates have yet to appear, so this chapter describes a mechanism rather than a rate.

Two presets load the illustrative inputs: a California-style discretionary track with three years to run and a Texas-style by-right track with nine months. Both use the same arithmetic. Odds must rise or hold from one stage to the next, cost and years must fall or hold, and the entitled value must sit above raw land; the table keeps its last valid numbers and names the problem when an input breaks a rule. Without JavaScript, the table shows the California preset: $514,307 at filing, $674,669 at approval, $884,130 at final plat and $1,084,100 at permit, with raw land at $500,000.

Items marked as requirements come from the guides cited. The rest are our suggestions, drawn from the evidence above.

Our development-site report shows an entitlement adjustment supported by before-and-after approval sales, and our Washington upzoning feature covers what a statewide rezoning does to comparable selection.