Does Lot Size Drive Price on the SF Peninsula?

The Finding, In One Line

Within a Peninsula city, you are paying for the house, not the yard. A log-log regression on 316 single-family sales from June and July 2026, pulled from closed SF Peninsula sales and demeaned by city so an expensive city cannot drive the result, splits price between the two square footages like this:

House size on its own explains about 68 percent of the price variation inside a city. Lot size on its own explains about 33 percent, but nearly all of that is borrowed from the fact that bigger lots tend to hold bigger houses. Once the model already knows the house size, adding the lot raises R² from 0.683 to 0.691, under one percentage point. The lot effect is genuine, the t-stat of 2.9 rules out zero, it is simply about seven times smaller than the house effect.

Translated into money, holding the city constant:

• Grow the house from 1,500 to 2,500 sq ft on the same lot: price about ×1.45.
• Grow the lot from 5,000 to 7,500 sq ft with the same house: price about ×1.045.

Stack those together and you get the trade that falls out of the model: swapping a 2,500 sq ft house on a 5,000 sq ft lot for a 1,500 sq ft house on a 7,500 sq ft lot comes in about 28 percent cheaper with 50 percent more land. That is what the coefficients imply for two otherwise-average houses in the same city, not two specific sales you can point at.

Why Land Is Cheap At The Margin

The likeliest reason is that you are not allowed to use the extra land. Peninsula cities cap what you can build through floor area ratio, lot coverage, and setbacks, so past a point the extra lot square footage is yard you can never turn into house. Buyers may love that yard, and it still does not convert into finished square footage, which is what the price is actually tracking. That mechanism cannot be proven from sale prices alone, so treat it as the explanation that fits rather than a demonstrated fact. It is also the reason to read a parcel’s zoning envelope before believing any lot-size premium: the buildable cap, not the lot line, is what your basis should respect. Our guide to floor area ratio and lot coverage covers how those limits are calculated across San Mateo County.

The Lot Effect Is Not Flat, It Bends

The pooled +0.11 is an average across a curve. Add a squared term on log lot size and it comes in significantly negative, a curvature of about −0.13(standard error 0.03), which lifts R² from 0.691 to about 0.712. So the relationship genuinely bends instead of running flat:

• Below roughly 6,000 to 7,000 sq ft, lot size matters more than the pooled average suggests. On a tight lot, extra land is priced.
• Above that band, the premium flattens and edges slightly negative. Extra land adds almost nothing to price, cheap but not free.

The clean way to state it is as a curve that bends around 6,000 to 7,000 sq ft, not as a tidy descending ladder of quartile numbers. The point estimates inside individual lot-size buckets carry standard errors large enough that they are not reliably distinguishable from one another; what the data supports is the smooth curvature term, which is stable across every specification we tried. House size still dominates in every lot-size band. This bend is the practical heart of the finding: the marginal lot foot is worth the most on the smallest lots and least on the largest, which is the opposite of how a flat dollars-per-lot-square-foot comp treats it.

What This Means For What You Pay Per Lot Square Foot

If you underwrite teardowns and spec builds, the received wisdom is that dollars per square foot of house is a retail number and dollars per square foot of lot is the real one. On these sales, that is backwards for pricing the exit, and it has three concrete consequences for your land basis.

• Do not underwrite land basis as linear in lot square footage. Within a city, resale does not scale with lot size, so your basis should not either. A blended dollars-per-lot-square-foot figure lifted from tight lots will systematically overpay you on a big lot, because it prices the unbuildable yard as if it were buildable. Extra lot area earns roughly the 4.5 percent per 50 percent of added land that the market assigns it, not the full average rate, and zoning caps are the likely reason for the discount. One qualifier from the curve: that 4.5 percent is a pooled average across all lot sizes. Above the roughly 6,000 to 7,000 sq ft bend the market assigns extra land even less than that, and on smaller lots materially more, so place the parcel on the curve before you discount its yard.
• The cheapest route to a given lot is the smallest house on it, with one trap. Because price per foot of house climbs as houses shrink, a tiny cottage can read as expensive dirt on a dollars-per-lot-foot screen even when it is a cheap way to acquire the parcel. Do not shop on dollars per lot foot or you end up buying the 900 sq ft cottage that only looked like cheap land. Price the whole deal.
• Solve for land with a residual model, not a lot-size multiple. Work backward from achievable finished value minus build cost, carry, and margin, then solve for the land price that keeps the deal on the right side. That is the method our residual land value ROI guide walks through, and it is the one that does not get fooled by lot square footage.

The blunt version: within the same city, a bigger lot does not linearly justify a higher land basis. Above the bend, two parcels with the same buildable envelope resell for nearly the same finished value even if one carries fifty percent more yard, because the yard past the envelope barely shows up in the finished home you sell. The extra land is worth something, it is just priced at the market’s small lot-size elasticity, not at the blended dollars-per-lot-foot rate a tight-lot comp implies.

What This Does Not Prove

These caveats come out of a working appraiser’s critique of this analysis, and they matter as much as the headline. Read them before you underwrite anything on this.

• This is an elasticity on improved sales, not a land appraisal. Every sale in the regression is a house on a lot, so the result describes how buyers price a house-plus-lot package. It does notisolate the contributory or residual value of the dirt on its own. If your question is “what is the bare land worth,” this bounds it, it does not answer it. Note too that these are ordinary resales, most of them likely bought as places to live (every illustrative comp we checked carries no teardown signal), so the elasticity is closer to how families price living space than to how a builder prices dirt residually against a buildable envelope. Those are different pricing agents. Treat this as a strong prior on the resale market your finished home sells into, not a rule that sets your land bid.
• House size was tested for secretly carrying condition and vintage, and the controls did not move it. The obvious objection is that bigger houses are nicer or newer, so the house-size effect is really a quality effect. Adding decade-built dummies (joined on 56 percent of the sample) barely moved the house elasticity, +0.747 to +0.755 on the matched subsample. A condition index built from listing remarks, the only condition measure here that is not derived from price, is strongly significant on its own, renovated homes sell about 12 percent higher, but it is orthogonal to house size(correlation −0.06), so square footage is not proxying it. Bigger houses on the Peninsula are not systematically the renovated ones. This tests the confound rather than proving it absent: the condition measure is marketing copy and noisy, and measurement error biases toward finding no effect.
• View and finish quality are still uncaptured. No dataset here measures a view, a kitchen, or a level of finish directly, and those clearly move price. Some of what the house-size term earns is surely correlated quality that no available field pins down.
• Lot value is nonlinear. The pooled +0.11 averages across the bend described above. On small lots the lot premium is meaningfully larger; on large lots it is near zero or slightly negative. Do not apply the average to a specific parcel without asking where it sits on the curve.
• The window is two months, and the split is not a claim that land is worthless. Coverage is concentrated in June and July 2026, with essentially nothing before that, so longer-window checks re-use the same sales and are a consistency test, not independent confirmation. And 50 percent more lot adding about 4.5 percent to price is cheap, not free: the lot effect is small and positive, not negative.

One more that pre-empts the sharpest attack: house and lot size are correlated, so a fair objection is that the regression cannot cleanly separate them. Within a city, log house size and log lot size correlate at 0.61, a variance inflation factor of 1.59, low enough that the split between the two is stable rather than arbitrary.

Frequently Asked Questions