How I read the San Francisco market

I publish a lot of San Francisco market data. This page is about the reasoning underneath it, because a number without a method is just a number that agrees with whoever is quoting it. If you are going to rely on my read, you should know how it gets made and where I think it stops being reliable.

1. The citywide number is almost never your number

Whether the median is up or down a few percent across San Francisco tells you very little about what will happen on your block. This is a city of micro-markets, and the spread between them is routinely wider than the year-over-year change everyone reports. In one recent stretch the strongest house district cleared a median far above its list prices while several condo segments sold at or below list in the very same weeks. Same city, same month, opposite markets.

So the first thing I do with any headline is refuse to apply it to you. The relevant question is not what San Francisco did. It is what your property type, in your district, in your price band, did.

2. Property type before neighborhood

People instinctively sort the market by neighborhood. I sort it by property type first, because houses, condos, and TICs have behaved like three different markets here for over a year, and the gap between them has been the widest thing in the local data. A buyer comparing a condo to a house at the same asking price is not comparing like with like, and neither is a seller pricing a condo against a house headline. Only after property type does neighborhood, and then price band, sharpen the picture.

3. Medians, not averages, and I will say which

I report medians almost everywhere. Averages in a city with trophy sales are a way to let a handful of closings speak for a whole district. A single sale at several times the local norm can move an average and tell you nothing true. Medians are duller and more honest. Where I use anything else, I say so.

4. Sale-to-list measures two different things, and I will not pretend otherwise

"Sold over asking" is the most quoted and least understood figure in this market. It mixes real competition with pricing strategy, and in San Francisco the second does a great deal of the work. Listing deliberately below the expected number to attract a crowd is an old, legal, and common tactic here, strongest in mid-priced house districts.

You can see it directly in the data. In one recent year, median list prices rose modestly while median sale prices rose several times faster. The list prices stayed anchored while the market ran, and the space between them is exactly what gets reported as "over asking." The demand is real. The size of the percentage is, in meaningful part, a listing decision. Any read that treats the two as the same thing is overstating what it knows.

5. Data quality describes the sample, not the market

This is the discipline I am most stubborn about. Every figure I publish carries a sense of how much data sits behind it, and I use plain words for it: robust, directional, thin, sparse. Those words describe the size and confidence of the sample. They are not a verdict on whether the market is strong or weak. A thin sample in a hot neighborhood is still thin.

My working rule is that a segment with fewer than roughly ten closed sales in twelve months cannot carry a confident claim. I will still show you what it says. I will not build a strategy on it, and I will tell you that is what I am doing.

6. Season-matched comparisons, or none

Comparing this month to last month in a seasonal market produces confident nonsense. When I compare a period across years I match the window: the same calendar stretch this year against the same stretch last year, and where the question deserves it, against the same stretch in every year back to 2016. A ninety-day spring window runs hotter than a full year by construction, and I say so on the page rather than letting the number imply more than it earned.

7. Fact, interpretation, and assumption are labeled separately

I keep three columns apart in my own head and in what I publish. What the data measures. What I believe it means. What I am assuming to get there.

The clearest example is the idea that buyers priced out of one neighborhood create measurable pressure in the next one over. Agents watch that happen constantly. The MLS cannot yet tell the buyer who always wanted the Sunset from the buyer who gave up on Noe Valley. So when I draw that connection I label it observed, not measured, and I keep it a dotted line. When it graduates to something the closed data can actually show, I say that too, and I show the sales.

If you ever read something of mine and cannot tell which of the three you are looking at, that is my failure and I would like to know.

8. I go deeper where the report stops

National and brokerage-level reports are genuinely good at the top-down view. They are not built to tell you what happened on eleven specific blocks. My work starts where theirs ends: I take a published claim, pull the same closings from the MLS, and look at what is inside the number. Often it holds up and gains texture. Sometimes it turns out to be narrower than the headline suggests, or driven by a composition effect nobody named.

When I disagree with a source, I show my arithmetic and cite theirs. When their figure and mine differ by a hair, I say the difference does not matter rather than manufacture a controversy out of it.

9. The one question that makes the rest easier

Before any analysis, I try to find the single thing that, once clear, makes everything else simpler or unnecessary. For a seller it is usually not "what is my home worth." It is what selling needs to accomplish, and whether it accomplishes that. For a buyer it is rarely "is this a good time to buy." It is what the right home actually has to solve.

Most of the noise in a real estate decision comes from answering a smaller question very precisely while the larger one stays fuzzy. Market data is excellent at the smaller question. It is no help at all with the larger one, and I would rather spend the first conversation there.

10. Help first, including when the answer is do nothing

Sometimes the honest read is hold, rent, wait, or fix the thing that is actually bothering you about the house you already own. I will tell you that, and it costs me a transaction often enough that I know I mean it. A recommendation I would not want you to see the reasoning behind is not a recommendation, it is a pitch.

The filter I use on my own advice is simple: would I be comfortable if you could see my full intention behind it? If not, it does not get said.

Where the numbers come from

Closed sales from the San Francisco MLS, processed through my own data engine and published first in the POTM Blog issues, each stamped with its own data-through date. The interactive version lives at Paulo's Pulse, my research lab, in Spanish at Pulso de Paulo. Data is deemed reliable but not guaranteed, and is subject to correction and revision. Corrections get published, not quietly edited.

What I am is a real estate agent with an engineering background and a long habit of checking the arithmetic. What I am not is a lawyer, lender, inspector, appraiser, or tax advisor. Where a question belongs to one of them, I will say so and help you find the right one.

Related reading

• The topic hubs: big questions answered in full
• The POTM Blog
• Paulo's Pulse, the research lab
• Should I sell or rent?
• About Paulo