8.3 points behind the nation: 0.4 is which industries, 6.8 is how they did here
data · economy · housingActive
What Made It Worse, and What Would Move It
San Francisco's payroll fell 8.3 points behind the nation's after February 2020, and only 0.4 of those points is the industries the city held; 6.8 is the same industries doing worse here. On the ground: business openings run 26% below 2015-19, asking rents rose 1.9 times as fast as rents paid, and downtown stations stand at 40% of 2019, while reported property crime is 51% below 2019 and remote work, which rose everywhere, does not sort the sixteen metros by their recovery. Of 10 indicators read for direction, 5 are improving and 1 worsening.
Why San Francisco's recovery ran behind every other large metro, what that looks like on the ground, which way the record points now, and what could move it: one page over three projects and two new pulls.
Why San Francisco's recovery ran behind every other large metro, what that looks like on the ground, which way the record is pointing now, and what could move it.
The answer the record supports, in three sentences. Payroll in the San Francisco metro fell 8.3 points behind the nation's after 2020-02, and only 0.4 of those points is the industries the city happened to hold; 6.8 is the same industries doing worse here than they did elsewhere. On the ground that looks like a quarter fewer business openings a year, permitting that is somewhat slower, rents that ask 1.9 times as much more as sitting tenants pay, and downtown stations at 40% of their 2019 exits; it does not look like crime, which in the reported record is half what it was, and it is not explained by remote work alone, which rose in every metro and does not sort the sixteen by their recovery. Of the 10 indicators read for direction, 5 are improving and 1 worsening over their latest window.
1 · why it was worse
A twentieth is which industries the city held; the rest is how they did here
Between 2020-02 and 2026-07 the metro's payroll moved −4.0% while the nation's moved +4.3%, a gap of 8.31 points. Splitting that gap by industry separates two explanations that sound alike: the city held industries that did badly everywhere, or the city's own industries did badly here. The first accounts for 5% of the gap and the second for 82%. Five years before the pandemic the metro grew 13.5% against a panel median of 9.8%, 4th of 16, so this is a break rather than a continuation.
the gap, in points of employment
growth by industry, 2020-02 to 2026-07
Professional and business services is a fifth of the metro's payroll and carries 2.9 points of the gap on its own, falling 9.3% here while growing 4.8% nationally. The nine industries cover 95% of payroll; construction and natural resources have no metro series and are carried as one line. Read from SF and the Recovery's decomposition, seasonally adjusted payroll from the Bureau of Labor Statistics.
2 · the symptoms
What operating here looks like, measured, with the read written before the number
Six things about being here that could account for the city's own industries doing worse than the same industries elsewhere. Each carries the rule it was read by and the verdict as it fell; a symptom that did not clear its bar is reported as such, because ruling a cause out is as much of the answer as ruling one in.
Remote work
neither
Worked-from-home share of the metro's workers, 7.2% in 2019 to 35.1% in 2021 and 18.9% in 2024. The metro led the sixteen in 2021 and 2022 and stands 3rd in 2024, behind Denver; its rise since 2019 ranks 4th.
year
SF
rank of 16
leader
2019
7.2%
—
—
2021
35.1%
1
San Francisco
2022
27.0%
1
San Francisco
2023
20.5%
3
Denver
2024
18.9%
3
Denver
The test that matters runs across metros: if remote work explains the gap, the metros whose remote share rose more since 2019 should stand lower on the employment index today. Across 16 metros the rank correlation is -0.35, against a bar of −0.5 set before it ran, which does not clear the bar. Remote work rose everywhere; it is a national change that the city met with an unusually office-based payroll, not a San Francisco cause. Source: Census ACS one-year summary file, table B08006; there is no release for the year the shift began.
Read as registered: San Francisco has the highest worked-from-home share of the sixteen metros in every year after 2020, and the largest rise since 2019.
speaks to how those industries did here · this page
Business formation
finding
Openings in the city's business registry averaged 16,119 a year over 2015 to 2019 and 11,911 over2022 to 2025, −26%. Closures fell too, from 13,686 to 9,153 once the administrative purge of 2018 is set aside, so net formation held at about 2,433 then 2,758 a year: what fell is churn, the rate at which the city tries new things, rather than its net stock of businesses.
openings by year, 2013 to 2025, the current year hollow
Only 59% of recent openings carry an industry code, so no sector story is told from the registry.
Read as registered: mean openings in 2022-2025 at least 15% below 2015-2019.
speaks to how those industries did here · this page
Permitting
no finding
Of building permits other than the over-the-counter kind, the share issued within 180 days of filing was 24% over2015 to 2019 and 22% over 2022 to 2023, −12%: slower, and short of the bar. The share is measured only on cohorts old enough to have had the full 180 days, because a recent filing that has not been issued yet is not a slow one. From 2024 the department's record is kept differently, so the later bars are drawn hollow and not read, and the effect of any permitting reform cannot yet be seen in it.
share issued within 180 days, by filing year
Office-to-residential filings of 20 units or more since 2023: 3 projects, 485 units in all (30 Van Ness, 348; 3333 California, 92; 988 Market, 45). a permit is an application to convert, not a conversion. Record as of 2026-09-07.
Read as registered: the share of permits issued within 180 days of filing in 2022-2023 at least 20% below 2015-2019.
speaks to how those industries did here · this page
Reported crime
finding
Reported larceny, burglary and vehicle theft together: 54,044 in 2019 and 26,618 in 2025, −51%; assault −16% over the same years. Whatever is keeping firms and shoppers away, the police record does not show more of it than before the pandemic. These are reports rather than victimisation, and the share filed online moved from 23% to 17%, so some of the fall may be fewer reports rather than fewer thefts.
property incidents by year, the current year hollow
Read as registered: property incidents (larceny, burglary, vehicle theft) in 2025 at least 20% below 2019: the symptom has receded in the record.
speaks to how those industries did here · this page
The cost of arriving
finding
Asking rents rose 32.9% since 2019 while rents actually paid rose 17.4%, so asking moved 1.89 times as fast. In a rent-controlled city that spread is the price of arriving: a firm that wants to hire someone from outside pays the asking rent, not the sitting one. the spread between what a sitting tenant pays and what the same unit asks is the cost of arriving here.
Read as registered: asking rents rose at least half again as fast as rents paid.
Weekday exits at the four downtown stations stand at 39.9% of 2019 against 46.7% for the other forty-four in the system. The gap is real and short of its bar, and the long commuter stations lost about as much as the offices they carried people to, which is what a metro-wide change in where work happens looks like rather than a downtown-only one.
Read as registered: against every station in the system, the four downtown stations recovered at least 10 points below the median of the others, and a permutation test returns p below 0.05.
Office vacancy has no public record; brokers' figures are not reproducible and are not quoted here.
3 · where it is heading
Direction, not a forecast: what each indicator did over its latest window
Each series is compared over its own last window with the window before it, and the word beside it says whether that direction is the good one for the city. A partial year is drawn but not counted, and nothing here is extrapolated: the record says where things stood and which way they were moving when it was last written.
Payroll employment, SF metro2025-08 to 2026-07 against 2024-08 to 2025-07+0.1%→ flat
Taxable sales, city2025 Q2 to 2026 Q1 against 2024 Q2 to 2025 Q1+7.7%↑ improving
BART weekday exits, system2025 against 2024+9.2%↑ improving
Active business locations2025 against 2024+0.4%→ flat
Business openings2025 against 2024−4.2%↓ worsening
Net income moving in, tax returns2023 against 2022+57.0%↑ improving
negative means more left than arrived; rising means less is leaving.
Reported property incidents2025 against 2024−26.6%↓ improving
Worked-from-home share, SF metro2024 against 2023−8.2%↓ improving
down is the direction that refills offices; the city has no lever on it.
Asking rent, median2025-08 to 2026-07 against 2024-08 to 2025-07+15.1%↑ rising
rising asking rents are returning demand and a higher cost of being here at once.
Downtown office parcels written down, roll2025 against 2024+92.6%↑ rising
rising means the roll is recognising more of the loss; neither good nor bad for the city's course, but it dates the lag.
last 12 months against the 12 before; flat inside ±1%; last complete year against the one before; flat inside ±2%; a partial year is never counted.
The roll: three observed quantities and no number between them
The property tax rests on a roll that records a loss only when a parcel sells, or when the assessor grants a reduction, and both take years. Three quantities that already exist bracket what downtown's office loss could be worth to the city's own revenue; the page draws them side by side and declines to pick a point between them.
City tax is assessed value at the secured rate of 1.18% and the city's 64% share of the levy. A parcel that sold is one a buyer priced; the unsold may not be like it. Values from SF Budget Block's roll analysis.
The ledger: what the roll already records, and what has turned but not reached it
San Francisco's own revenue rests unusually heavily on a property tax roll that is locked below market, losing value downtown, and tied to a shrinking protected residential stock. If the roll moved with the activity records there would be no lag to describe. That was tested: across the neighborhoods both projects measure, assessed value and active businesses rank-correlate at -0.079, against a bar of 0.2, so they do not move together and the lag is real.
Already on the roll
Downtown commercial assessed value18.4%
the roll's own downtown total on parcels present in both years. lag · an assessment appeal takes one to three years to resolve, and the reduction reaches the roll only when it is granted.
Long-held homes against just-sold0.32
Proposition 13 holds long-held assessments far under what the same street fetches at sale; the gap is locked in until the parcel changes hands. lag · a sale in year Y is reflected in the Y+1 roll.
Downtown office parcels written down without a sale52
the only route by which a capped assessment can fall, so this is the roll admitting the market. lag · an assessment appeal takes one to three years to resolve, and the reduction reaches the roll only when it is granted.
Densest non-compliant soft-story stock132 per 1,000
a claim on the same housing the roll rests on, ahead of any earthquake. lag · a completed building enters the roll at full value in the roll after completion.
Property tax as a share of the city's own revenue47.3%
this is the exposure. The more of the city's own revenue rests on the roll, the more a locked and lagging roll matters. lag · the latest CLOSED roll is a year behind the newest activity data, and the budget in force is built on it.
Turned, but not yet on the roll
Implied write-down when a downtown parcel sells0.68
what the market paid for the parcels it actually priced. The roll marks only what sells; the rest is not marked at all. lag · a sale in year Y is reflected in the Y+1 roll.
Employment against the nation, San Francisco's own part−6.84 pts
the gap that is not explained by which industries the city held. Payroll is not on the roll at all; it is what fills the buildings the roll values. lag · the latest CLOSED roll is a year behind the newest activity data, and the budget in force is built on it.
Share of Bay Area city taxable sales−2.3 pts
sales tax is the city's second own-source line and it moves with activity immediately, unlike the roll. lag · none: sales tax follows the quarter.
Downtown BART exits against 201939.9%
who still comes to the buildings downtown. Weak, though the whole network is weak and the long commuter stations are weaker still. lag · the latest CLOSED roll is a year behind the newest activity data, and the budget in force is built on it.
Asking rents against rents actually paid1.89×
the spread between what a sitting tenant pays and what the same unit asks on the market is the incentive to move a unit from the protected regime to the market one. lag · a sale in year Y is reflected in the Y+1 roll.
Exit notices that become a physical removal6.2%
pressure on the protected stock is mostly not demolition. The unit stays and the protection goes, which is a change of regime rather than a change of supply. lag · a sale in year Y is reflected in the Y+1 roll.
Carried from the source: one permit carries 87% of the matches; without it the rate is 0.8%.
4 · what could move it
The levers, sized by the numbers above, and what the city has already tried
What follows is the author's reading of the evidence above. It is the one place on this site that argues rather than describes, and every size it gives is a number from the sections before it.
Attracting different industries
acts on which industries the city held
The industries the city held explain 0.4 of the 8.3-point gap. A lever aimed at composition is aimed at a twentieth of the problem.
The cost of operating here
acts on how those industries did here
6.8 points is the city's own industries doing worse here than elsewhere, 2.9 of it in professional and business services, which fell 9.3% here while growing 4.8% nationally. Proposition M moved the tax burden from small firms to large ones; the loss sits in the large-firm sectors.
Remote work
acts on how those industries did here
The largest measured change in what being here means: worked-from-home rose from 7.2% to 18.9%. But it rose everywhere, and across sixteen metros the rise ranks against the employment recovery at -0.35, which does not clear the bar: remote work does not by itself sort the recovery, so it is not the San Francisco-specific cause. The city has no lever on it in any case; its lever is what makes being here worth the commute.
Turning offices into housing
acts on the tax roll
Since 2023, 3 projects of 20 units or more have filed to convert office space, 485 units in all, against a downtown office roll of $46 billion. A permit is an application, not a conversion, and the count is what the record shows.
Permitting speed
acts on how those industries did here
The share of permits issued within 180 days went from 24% to 22% (-12%), a fall short of the bar. From 2024 the record is kept differently, so whether PermitSF changed anything cannot yet be read from it.
Public safety
acts on how those industries did here
Reported property incidents in 2025 stand 51% below 2019, assault 16% below. A lever aimed at reported crime is aimed at something already moving; the caveat is that these are reports, and the share filed online has fallen.
What the city has enacted since 2020, and whether the record can see it yet
Each policy is placed against the symptom it targets and the indicator that would show it. Where a full year of that indicator does not yet exist after enactment, the page says so rather than reading a result into noise. Policy facts are the one hand-entered content on this page; each carries its date and the city page it was read from.
Proposition C: transfer-tax waiver for office-to-residential conversions
Enacted 2024-03-05. waives the city's real-estate transfer tax on the first five million square feet of non-residential space converted to housing, for projects approved before 2030.
Aimed at conversions; the indicator that would show it is office-to-residential permits of twenty units or more.
measurable1 project of 20+ units filed since enactment, 92 units, against 3 (485 units) since 2023.
Proposition M: business tax reform
Enacted 2024-11-05, in force from 2025-01-01. raises the gross-receipts exemption from $2.25M to $5M, apportions on 75% sales and 25% payroll, cuts the categories from fourteen to seven; smaller firms pay less, larger firms more.
Aimed at cost; the indicator that would show it is business openings, and the professional-services competitive term.
measurableopenings in 2022-2025 average -26% against 2015-2019; the latest year, 2025, is -4.2% on 2024.
First Year Free
Enacted 2021-11-01. waives registration, licence and first-year permit fees for a new commercial-location business with under $5M in receipts; the city counts about 11,000 enrolled and extended it to mid-2026.
Aimed at attendance; the indicator that would show it is no indicator in this record at the scale of a few dozen storefronts.
not measurable hereno indicator in this record at the scale of a few dozen storefronts.
PermitSF
Enacted 2025-02-01. launched in the administration's first hundred days: removes the school district from permit routing, drops the conditional-use hearing for most retail and professional-service uses, ends permits for sidewalk tables and displays.
Aimed at permitting; the indicator that would show it is share of permits issued within 180 days of filing.
not measurable herenot readable: the permit record changed regime in 2024.
Downtown revitalization financing district
Enacted 2025-06-12. a district from the waterfront to Civic Center whose future property-tax growth is paid back to conversion projects for up to thirty years; the city counts about fifty eligible properties with room for about 4,400 homes, enrolment open through 2032.
Aimed at conversions; the indicator that would show it is office-to-residential permits of twenty units or more.
measurable1 project of 20+ units filed since enactment, 92 units, against 3 (485 units) since 2023.
Read 2026-09-07 from this project's own pulls (Census ACS, DataSF permits, and the city's business registry and police record held locally) and from SF and the Recovery, SF Budget Block and SF Housing Pressure. Every number comes from one of those; none is entered here, and a figure whose project is not current is left off rather than shown.
The gap, 2020-02 to 2026-07, and the part that is the city's own
8.3 pts, 6.8 heremeasured
Bureau of Labor Statistics payroll by supersector via FRED, decomposed in SF and the Recovery: growing at the national rate 4.35, industry mix -0.39, the same industries here -6.84, construction residual -1.08. Professional and business services fell 9% here while growing 5% nationally.
Remote work against the recovery, sixteen metros
ρ -0.35measured
Census ACS one-year table B08006, worked-from-home share 2019 to 2024, rank-correlated with each metro's employment index; the bar of −0.5 was set before it ran and was not cleared. San Francisco's share went from 7.2% to 35.1% and back to 18.9%.
Business openings a year, 2022 to 2025 against 2015 to 2019
-26%measured
The city's business registry, locations by start date, 16,119 a year then 11,911; closures fell 33% once administrative closures are set aside, so net formation held. Churn fell, not the stock.
Reported property incidents, 2025 against 2019
-51%measured
Police incident reports, larceny, burglary and vehicle theft; assault -15%. Reports rather than victimisation, and the share filed online fell, which is carried on the page as the caveat.
About this project
San Francisco is the only large metro still below its February 2020 payroll, and the usual explanations sound alike until they are separated. This page separates them. A decomposition of the employment gap by industry shows that the industries the city held explain a twentieth of it and that the city's own industries doing worse here than elsewhere explain most of the rest, with professional and business services alone carrying a third. Six symptoms of operating here are then measured against rules written before the numbers were seen: remote work across sixteen metros from the Census, business openings and closures from the city's registry, time to permit from the building department's record, reported crime from the police record, the spread between asking and paid rents, and downtown transit exits. Each is reported as it fell, including the ones that did not clear their bar. The third section reads direction rather than forecasting: each indicator over its latest window against the one before, and three observed quantities that bracket what downtown's office loss could be worth to the city's own revenue. The last section is the one place on this site that argues: six levers, each sized by a number above, and a scorecard of what the city has enacted since 2020 against the indicator that would show it, saying too early where a year of data does not yet exist.
›Sources and definitions
–The decomposition is copied from SF and the Recovery rather than recomputed, so the two pages cannot disagree about the gap.
–Two sources are pulled by this project itself: Census ACS one-year summary files for sixteen metros, selected by metro code because the names changed in 2023, and the building department's permit record with the same-day over-the-counter permits left out.
–Every symptom read was written with its no-finding rule before the number was seen; two of six did not clear their bar and are reported that way.
–Permitting is measured as the share of filings issued within a fixed horizon, on cohorts old enough to have had it, because a recent filing not yet issued is not a slow one; the record changed regime in 2024 and later years are shown but not read.
–Direction is the latest window against the one before, classified inside a fixed band; a partial year is drawn hollow and never counted, and nothing is extrapolated.
–The fourth section is labelled as the author's reading. Every size it gives is a number from the sections before it, and each enacted policy is scored only where a full year of its indicator exists.
Started
September 4, 2026
Last updated
September 7, 2026
Delivery
native
Tags
dataeconomyhousingpublic finance
Accuracy
Numbers are computed from public records and can be wrong; nothing here is offered as complete or exact.