Storefront Register

Snapshot
Complete through
Source DataSF g8m3-pdis
Grain
Filter
Closure
Dates
City
Geography
Window
Sort

Net formation by area

AreaOpenClosedNet

Select a row to filter every panel. Corridor coverage is 41.7% of the panel — areas without one are absent, and they are not a random sample.

All San Francisco

every area combined
net formation
median registration gap · re-let only
…of those, over 182d
no registered tenant

Monthly openings and filed closures

What "closed" counts, and what it lags Filed closures only — the 46,992 rows marked administratively closed are excluded, and every one of those also carries a real end date, so this is a floor. A closure is a filing, not a shutter coming down, and the filing can trail the event by years.

Categories entering and exiting since 2023

Entering and exiting are not comparable Categories are self-reported and present on ~59% of openings but only ~31% of closures, so a positive net is largely a coverage artefact.

How long a space sits between tenants

Registration gap distribution — genuine changes vs business continuity

A registration gap, not measured vacancy Nothing here observes a shop standing empty. And a third of handovers re-let within 30 days because the business carried on — sold, not re-let.

Three cities

CitySpellsActive

San Francisco and Los Angeles record tenancies. Chicago records licences that expire and renew, so its tenancies are reconstructed and its closures are not directly comparable.

Where space re-lets fastest

San Francisco only — the 41 analysis neighbourhoods, which are the same names the registry publishes, so no crosswalk sits between the polygon and the number. Corridors are not mapped: DataSF publishes no geometry for them, only the name string, so they stay in the tables above.

Measure

Hover a neighbourhood for its counts.

San Francisco

hover a neighbourhood
re-let ≤182d
spaces vacated
still no tenant

Censoring is corrected, and it changes the answer Every share here is Kaplan–Meier with spaces that never re-let entering as censored, not dropped. Uncorrected, the same neighbourhoods read three to eight times faster — Hayes Valley is 117 days naive against 923 corrected — and the error is largest exactly where re-letting is slowest, which is the pattern this map exists to show.
Fill encodes confidence, not just value Opacity scales with how many vacancies an area contributes, so a pale polygon is a thin sample rather than a low number. Stippled areas fall below vacancies and are drawn but not ranked. Flat grey is a neighbourhood with no storefront vacancies at all — parks, the Presidio, Treasure Island — which is an absence of storefronts, not an absence of re-letting.

How fast each city re-lets commercial space

Of the commercial spaces trading in January 2015, this follows every one that later lost its registered tenant and asks how many had a new one by a given point. Spaces still without a registered tenant are counted, as censored observations, rather than dropped — which is what separates this from the medians below.

Share re-let, by time since the tenant left

Bars are 95% intervals. 182 days is the Commercial Vacancy Tax's statutory definition of a vacant space, which makes it the one horizon with a legal meaning rather than a round number.

CityVacanciesRe-letBy 182d

San Francisco re-lets faster than Chicago and Los Angeles at every horizon, and the intervals do not overlap at 182 days. That ordering is the claim; it survives counting the spaces that never re-let.

The medians, and why they sit below

Median registration gap among spaces that did re-let

CityFilterSucc.Median
These medians are not identified Each one is computed only over spaces that did re-let — 41% of San Francisco's vacancies, 29% of Chicago's, 25% of Los Angeles's. The selection rate differs by city, so the ratio between two medians is not a like-for-like comparison. Use the shares above.

Restricting each city to assessor-recorded commercial use. Filtering cuts LA's median sharply and SF's moderately, while raising Chicago's — consistent with the first two registries being diluted by home businesses, and Chicago's licence registry already skewing commercial.

Commercial parcels with no registered tenant

Taking each county assessor's list of commercial parcels and asking the business registry whether anyone is registered there. This is not a vacancy rate — registration is not occupancy — but it is checked against an independent source that fails in a different way.

Parcel state, by city

Does anyone file building permits there?

CityOdds ratiozReads as
The check fails differently from the thing it checks All three permit files carry the parcel's own identifier — SF block+lot, LA apn, Cook pin — while the occupancy test joins by address or by distance. So if an unmatched parcel were really occupied and merely missed, permits would still reach it by key and its rate would match. A null here is evidence of join failure, not an absence of evidence.

Odds ratios hold the assessor's own parcel type fixed (Mantel–Haenszel). The crude comparison overstates the gap: a parking lot is rarely registered and rarely permitted, for one reason rather than two.

Chicago: what replaces what

The question San Francisco cannot answer. Chicago's category is a licence type the city assigns, so it is present on 100% of handovers including closed ones — against 7.8% in SF, where the category is self-reported and decays.

Most common transitions · = same category, > a change

FromTonMedian

Where a space is re-let to the same trade, it re-lets fast

Each point is an exiting licence type. Specialised fitout the next same-type tenant can use is an asset; specialised fitout nobody else wants is a liability.