The weekly answer

Is buying in San Francisco worth it?

For a two-to-four-unit building you would live in, we do not guess. Every week we take every one listed for sale in the city and underwrite it against the alternative: not buying, and putting the same money in an S&P 500 index fund. Cautious rents, 3.5% down, the current mortgage rate, the property's actual tax bill where we have it, held ten years. This page is the running score.

This week in San Francisco

Week of October 6, 2026: 102 buildings for sale were underwritten. 2 beat the index fund at cautious rents. 1 passed all three of our tests.

100 lost to the index fund , and 1 was rejected before the math reached a verdict (a unit count the listing gets wrong, commercial space, an address nothing can verify) . Every row, with the reason on the ones that failed, is on this week's San Francisco board.

What we mean by publishing one

Three tests, in order. A building has to pass all of them.

It wins the cautious case. Buy it, live in one unit, rent the others at rents we can defend from the city's own rent registry, hold ten years. That has to leave you wealthier than renting and putting the same money in the index fund at 10% a year.
It survives a rent shortfall. If the rents we assumed came in 15% lower, it still has to win. A building that beats the index by less than our own uncertainty about its rent is a tie, and we do not publish a tie as a win.
Its facts check out. Unit count, bedrooms, square footage, the tax bill: verified against sources other than the listing, line by line.

That is why "beat renting and index funds" and "published" are different numbers. A building can win the math and still miss the bar, and a building whose facts we could not verify is never published, no matter what its numbers say. The full model is on the methodology page.

The whole Bay Area this week

Edition	Run	Underwritten	Beat the S&P	Published
San Francisco	October 6, 2026	102	2	1
East Bay	October 7, 2026	162	14	4
Peninsula	October 8, 2026	27	0	0
South Bay	October 2, 2026	21	0	0
North Bay	September 28, 2026	82	1	0
All editions		394	17	5

Each edition runs on its own weekday, so these are the latest run of each, from different days of the same week. "Beat the S&P" is the first test above; "published" is all three.

Since August 31, 2026: the running record

San Francisco run	Underwritten	Beat the S&P	Published
October 6, 2026	102	2	1
September 22, 2026	2	1	1
September 15, 2026	33	0	0
September 8, 2026	10	2	1
September 1, 2026	14	2	1

Over 5 weeks: 161 buildings underwritten, 7 beat the index (4%), 4 published (2%).

The count is an output. It moves with the market and we do not smooth it: a week with zero is published as zero. Everything we ever published sits on the public track record, with what happened to it since.

Why most of them lose

For each building we compute one number: the annual return the index fund would have to earn, every year for ten years, for not buying to come out even with buying. Above 10%, the building wins, because the index would have to beat its own long-run history to catch it. Below, the index wins in an ordinary decade.

This week, the middle half of the San Francisco buildings the index beat sit between 3.4% and 7.6% (17 buildings with a measured breakeven this week).

The margins are thin enough that a small change in the rent assumption decides the verdict, which is why we publish the cautious case and say so on every row.

If you are asking about a condo or a house

We underwrite two-to-four-unit buildings only, because the rent from the other units is the whole reason the math can work at San Francisco prices. For a condo or a single-family home, run your own numbers in the house hacking calculator: it is the same model, with the rental income set to whatever you would actually get, including nothing. And before you close on anything in the city, size the first bill you will see with the supplemental tax bill calculator.

Updated with every San Francisco run. Every count on this page is read from the board when the site is built; none of it is typed by hand, so it cannot disagree with the board.