Rents In San Francisco Are Determined By Demand, New Data Show

Slightly better than the worst possible outcome (detail), by Christine Hou

By putting together 60+ years of housing data, much of it gathered from newspaper microfiche, Eric Fischer came up with a remarkably simple yet powerful model of rents in San Francisco. The model shows that apartment rents in the city come down to just three factors: employment, wages, and the number of housing units. Fischer’s model attracted attention for it’s rare, long-term view of rents and both its elegance and robustness. Soon enough, after appearing on local and special interest blogs, the story was picked up by national outlets like the Washington Post and Gawker. The response has typically been to understand the model as presenting three options, although only one viable.

As Emily Badger of the Washington Post writes:

[I]f San Francisco wants the cost of housing to go down, there are three clear ways to get there. The city could build more of the stuff. Or it could hold out for falling incomes or job losses among the people who compete for housing.

Again and again, the commentators have given us three options and two of them are jokes:

Hamilton Nolan of Gawker:

  1. “… a major earthquake/recession that wipes out demand for living in the Bay Area, by devastating it”
  2. “[no one] who is relatively wealthy should be able to live in the Bay Area,” or,
  3. “build a lot more housing in the Bay Area”

Crosscut’s Drew Atkins:

The basic conclusion of Fisher’s [sic] analysis is that it’s extremely hard to bring rents down in a prosperous city. Only two things can really accomplish it: more housing units, or fewer jobs—particularly high-paying ones. That’s it.

These kind of responses miss the simple fact that of the three variables in Fischer’s model, only employment and wages really matter. (The demand-only model achieves an R squared value of .924 and adding the supply variable, housing units, gets us to .940). And, therefore, any big solution to the affordable housing problem must address housing demand. Since Fischer released his work, others have applied the data to more rigorous statistical testing. One test found that the best model only uses employment and wages. Another found that the best model of San Francisco rents needs just one variable: wages (although wages and housing were a close second).

Fischer's model with demand variables.
Fischer’s model with demand variables. (Eric Fischer)

Fischer's model with supply variable added.
Fischer’s model with supply variable added. (Eric Fischer)

That gets us to an inconvenient question:

Screen capture of Fischer on the Twittersphere. (Twitter)
Screen capture of Fischer on the Twittersphere. (Twitter)

Inconvenient because there are many dedicated intelligent people who believe that increasing the supply of market-rate housing is the best response to skyrocketing rents. And that belief is now challenged by Fischer’s and others’ analyses. That is to say, the opposite conclusion made in the Washington Post, Gawker, Bloomberg, City Observatory, Crosscut, and The Stranger.

With supply apparently sidelined, that leaves the demand variables of employment and wages. Like the authors in the above publications, I don’t see a fair and effective way to reduce employment in order to rein in rent. But we can do something about wages short of recessions and earthquakes: tax the rich. In Fischer’s model, a sustained doubling of housing construction can turn a $3,500 1-bedroom San Francisco apartment into a $2,500 one-bedroom San Francisco apartment (if employment and wages stay the same) by 2035. An income tax of 17% does the same thing by next April without the New Deal-sized building program1.

In Seattle terms, that means a $1,400 Capitol Hill one-bedroom apartment for under a thousand dollars whenever the citizens can pass an income tax or the Supreme Court makes the state legislature do it. And we don’t have to settle for a flat income tax to accomplish this. The bottom end of market rental rates are established by the group earning roughly 80% of area median incomes (about $50,000 in Seattle). The income tax could start small there and ramp up, or start at a higher income with more aggressive rates.

Another demand-based solution ignored by local and national media: rent control. Rent control, which in San Francisco began in 1979 in the middle of Fischer’s 60+ years of data, did not appear to have a significant impact on rent. You can see him, deep in the comments section of his article or on Twitter, repeating himself to incredulous readers, cautiously, “rent control doesn’t seem to have changed things very much” or, as Michael Andersen of BikePortland sums up, “there’s no clear sign in this data that rent control has had this additional anti-new-housing effect on San Francisco. Again: shit was bad before. Shit was bad after.”


  1. The income tax can pay for a New Deal-sized building program. ^

We hope you loved this article. If so, please consider subscribing or donating. The Urbanist is a 501(c)(4) nonprofit that depends on donations from readers like you.

Michael works as a real estate valuation analyst. He graduated from the University of Washington with a Masters in Urban Planning and a specialization in real estate. Before Serial broke the long-form true crime scene wide open, he had a podcast about pet adoption.

Inline Feedbacks
View all comments

“Market-rate” housing in SF is luxury housing. Little ordinary housing is being built and almost no really affordable housing.

The city is corrupt, which is why these garish and overpriced buildings have been granted permits.

This ariticle does not mention the numerous (thousands of?) units that are being kept off the market as well as the massive effect of AirBnB!

Michael Andersen

Michael, thanks belatedly for this. I wrote a second-day story pointing to Fischer’s work and then failed to catch the subsequent analyses showing that wages could fully explain the pattern. This argument is definitely worth considering.

michael goldman

No problemo. I really appreciated your piece because it felt like so many others were not being realistic with the implications of the study.

The effect of the income variable was so surprisingly large. I expected, based on studies of income elasticity of housing (a moving target but I’ve seen it in the 0.3 – 1.0 range; Fischer’s model implies something like 1.6) for the effect to be maybe half of what Fischer found. I suspect what’s going on is a really skewed income distribution in San Francisco where the high income tail is especially thick. That would make an aggregate income variable like Fischer used especially sensitive to changes in that thick tail of the income (because aggregate income is more swayed by the high end of the distribution). So as that tail thickened with the growth of the tech boom and just general rising inequality, the model amplified that distribution change in the aggregate income variable.

Which makes me think the progressive income tax could be have a similar outsized effect — squeezing that fat tail skinny while driving down prices overall.


I’m not sure I’m following the logic here. If we concede that wages and employment have a stronger correlation than supply but acknowledge that supply does have a (smaller but not insignificant) effect, what are we to draw from that? We don’t have a viable public policy tool to surpress wages or employment – you’d be laughed out of city hall if you went in an asked for a municipal wage cap. We do, however, have control over land use.

michael goldman

It provides another good reason, among many others, to work towards a progressive state income tax. We should also advocate for better land use regs, too, but an income tax, according to this research, is much more effective.


According to Dyan Ruiz and Joseph Smoke, “Developers Aren’t Going to Solve the Housing Crisis in San Francisco, Truthout and [people, power, media] News Analysis, Oct 8, 2014, the financial cycle of developments has a lot more to do with this than simple supply and demand.

“Myth 4: As long as you can upzone and deregulate, you can build and build to the point where prices will go down.
Fact: That’s not the way that housing development works. Housing finance has limits.

Even if residents and city politicians agree on how to make housing development
easier and denser, that’s not going to change how banks and housing finance
work. Developers won’t be able to build an unlimited supply of housing, or even
build to a point where prices will drop substantially.

Say San Francisco pushed to flood the housing market with over 100,000 units in just a few short years in an attempt to lower prices. At that point when prices start declining, external pressures are going to kick in. Interest rates might rise to cool the market, making it more expensive for developers to finance new construction. Banks will also stop lending because they don’t want to invest in a sinking market. Investor capital will stop flowing because investors won’t be able to forecast a sufficient profit and will seek investments with higher returns. So, the production of housing will slow or stop before prices start to drop too much.

Similarly,a developer has limits when housing prices are falling. Development is a risky business even in good times. All of the money invested in the development
depends on forecasts of what people might be willing to pay for the apartments
or condos when they’re move-in ready. If there’s any sign of a declining market, developers may still buy sites, but they’ll sit on them until prices start going up again. So whether it’s the banks, investors or developers getting cold feet, increasing supply isn’t going to drop prices to a level working-class people can afford.

Why? That developer still has to pay the bank back their loans based on the forecast from years ago. The developer thought they could charge $4,000 per month for rents. Now that’s not going to happen, so they’ll drop the rents to $3,500. They’ll have less profit, but they can still pay the bank back. If the market
takes a real dive, the developer’s not going to drop the rents to $1,500, or
anywhere near the level that a working-class family can afford.

Banks aren’t going to let developers pay them back less just because the market
tanked and there’s no more demand for $4,000 per month apartments. US housing works on a for-profit model. Building doesn’t happen if developers are
forecasting they’re going to lose money.”

Joe Wolf

The probability of Washington enacting the most progressive income tax in the US – by a huge margin – is less than simultaneous meteor strikes taking out the Amazon and Microsoft campuses.

Did you put your proposal here through one second of critical thinking? Wow.

michael goldman

The 17% tax rate figure is not anything like an actual proposal.

The 30% reduction is a more conservative figure than the ones used by the original author, Eric Fischer. He was trying to see how large and how long a building boom would be required to get a 67% and 50% reduction, according to his model. His figures were illustrative of the difficulty of a supply solution. Mine was illustrative of a political possibility. (17% isn’t that much more than California’s top rate of 13%)

30% is an example of the type of rent reduction that makes eyes pop. A 10% tax is more likely to succeed, giving a less eye-popping but still respectable 18% rent reduction. (This is also not a proposal! but an illustration of the effects achievable in a more likely scenario!)


Advocate lobbying for “repeal and reveal” hidden, antics which making. NAIOP and BOMA.S.F panic article saying “elude” statistical, rendering specifically building more of what? Ask London Breed,Jane Kim,Ed Lee and Scott Wiener my good friends, of oppositions “affordability! Rondey Fong,Christina Davis and Cindy Wu knowing what to do, yeah being illustrative just drawing comparison decadence. Building additional market rate development, going decrease value of residential and commercial irrelevant. Where losing avid rentals why “global trade” forget rationalizations of exclusive deals of NAIOP San Francisco has become. Ideal REIT,SIIQ,LLC and Reality funds. Specialized industries viable has increase (net gain of profits) mention “Transpacifc trade, worth $250 billion for California market is dynamic. Not disclose how many,class A office towers downtown San Francisco without social frown around? 140 and counting whom, international REITS,whom persist to never miss “expected” profits you’ll read but whom. Rondey Fong (Mr.Do No Wrong) explain (“Muni depot’s air rights) “think twice going leased of sold to whom? Global REITS including “3rd. Townsend CalTrain station how many towers for this project anticipated all high rises. Equivalent to (“Hudson Yards”) NYC! Micro housing is resolution to lack affordability in San Francisco. Inaccurate majority sales paid with, cash as “South bay repeal is needed! Global trade ask Cindy Wu willing to pay impact fees, only difference regarding article build more. To reduce price never going up aware firms paying exceptional compensations demographics change. No longer diversity by income San Francisco,public policies of Ed Lee are guilty whom owns the land hello. San Francisco majority of residents including,myself elated that sudden riches has benefitted San Francisco. Where social responsibility everyone City Hall evading concern “rent control” and new housing. Forgotten article didn’t include near completed, 30,000 BMR units how are they equated simply. SRO,S.F Housing Authority which $770 million (RAD loan for only rehabilitation) not increasing the density,In-Law units,Shelters,THO,SRO,Micro housing and dismal amount of new communities!

Compare the REIT market to high yield cities London,Osaka,Hong Kong,Shenzen,Berlin,Manchester,Shanghai and Singapore no more intermittently if planning which approved mayoral influence. Change governmental policies REITS and SIIQ whom eager build make immediate profits majority towers class A some mix used towers. H-REITS disfavor due cost of construction not “tax exemptions” for hospitality for San Francisco. Financial,global cartels,chemical and conglomerates sign up to build quantity of class A ask BOMA.S.F about Turkish commercial development. GYODER adamant building high yield metropolitan regions with support of bureaucracy where, L.A,San Francisco,Long Beach and Seattle informed. Dynamic markets simple circles prominent commercial brokers selling “Bay Area” region of Diamond only problem rental crisis along. South Bay mention “fair rental” policies governor Brown “Prop” 1229 developers excluded from inclusion of “BMR” units furthermore. Ratio of affluent single professionals abundant enjoyed the article aware San Francisco going become “Transpacific” city global banks. Reason class A towers going up, without fuss simply graduates specialized industries lobby. Gentrification when say possible reduction of price, never going up London never be affordable! Or should be, not profound where losing San Francisco I’ll say this those experiencing “gentrification” lobby there is hope Ellis act on ballots! Crisis going loom sudden,realty boom didn’t represent those lesser financial interest recommend. Ed Lee and BOMA S.F hold (forum) upon appropriate policies to retain residency when prices increasing. SmallPro S.F,CAAA and SFAA cease “TIC” units favor reduce rentals tax breaks your laughing! Conclusion “global trade bringing economic stability…increase of population those making ideal salaries. Inflation we need “repeal” if not micro-housing which Scott Wiener and London Breed promotes going average sq ft for renters! Ed Lee only “30,000 units for those facing “evictions” by 2020? As L.A now highest “Ellis Act” evictions going build how many,20,000 before 2019 unity is needed not defeated! Where going retain fair housing allow,
“urban development” attend the planning meetings if not. Renters only be chosen few able to stay in S.F!


Another interesting thing is the 1950–1960 dip in rents. What was happening in SF in those years that made rents dramatically fall? That data as been omitted from the chart, but seeing as it is quite a large exception, and a dramatic fall (isn’t that what people want – a large fall in rents?) then it is worth knowing what happened there rather than adjusting the analysis window to exclude it.

michael goldman

Eric Fischer goes into this a little in his post. He says, “What happened in 1954 to stop prices from dropping? The most obvious explanation is that that was when San Francisco ran out of large tracts of vacant land.”


That explains why rents go up, but it doesn’t explain why they fell. An aerial map of SF is available from 1938, and most of the land area is developed, except for a section near the beach.

Housing markets are regional, we should have seen a drop in rents when BART opened in the 1970s as this is effectively expanding the size of the city’s livable land area, even though it is outside formal gov’t boundaries.

David Rumsey Map Collection


Regarding rent controls, it is worth speculating why there is no impact seen when one would expect it. That finding needs a mechanism to explain it. The rent increases for the past thirty or so years have all been well below the suggested ~ 6% rate. You have to go back to 1984 to see an allowable increase anywhere near 6%.

March 1, 2016 – February 28, 2017 1.6%
March 1, 2015 – February 29, 2016 1.9%
March 1, 2014 – February 28, 2015 1.0%
March 1, 2013 – February 28, 2014 1.9%
March 1, 2012 – February 28, 2013 1.9%
March 1, 2011 – February 29, 2012 0.5%
March 1, 2010 – February 28, 2011 0.1%
March 1, 2009 – February 28, 2010 2.2%
March 1, 2008 – February 28, 2009 2.0%
March 1, 2007 – February 29, 2008 1.5%
March 1, 2006 – February 28, 2007 1.7%
March 1, 2005 – February 28, 2006 1.2%
March 1, 2004 – February 28, 2005 0.6%
March 1, 2003 – February 29, 2004 0.8%
March 1, 2002 – February 28, 2003 2.7%
March 1, 2001 – February 28, 2002 2.8%
March 1, 2000 – February 28, 2001 2.9%
March 1, 1999 – February 29, 2000 1.7%
March 1, 1998 – February 28, 1999 2.2%
March 1, 1997 – February 28, 1998 1.8%
March 1, 1996 – February 28, 1997 1.0%
March 1, 1995 – February 29, 1996 1.1%
March 1, 1994 – February 28, 1995 1.3%
March 1, 1993 – February 28, 1994 1.9%
*December 8, 1992 – February 28, 1993 1.6%*
*March 1, 1992 – December 7, 1992 4%*
March 1, 1991 – February 29, 1992 4%
March 1, 1990 – February 28, 1991 4%
March 1, 1989 – February 28, 1990 4%
March 1, 1988 – February 28, 1989 4%
March 1, 1987 – February 29, 1988 4%
March 1, 1986 – February 28, 1987 4%
March 1, 1985 – February 28, 1986 4%
March 1, 1984 – February 28, 1985 4%
March 1, 1983 – February 29, 1984 7%
April 1, 1982 – February 28, 1983 7%


It is clear that some further step-by-step explanation of how it is possible to get 6% or so increases when recent decades have had increases around 1-2% is possible.

michael goldman

These are the SF Rent Board’s allowable increases, right? I would guess that the difference between before and after 1992 is inflation. The rent board is supposed to consider landlord expenses (like utilities, insurance, repairs, appliances) in the rent increases.


Perhaps the reason why there isn’t much change in price explained by the housing stock variable is because the stock is more or less constant, at least compared to the large changes in income/employment.
It’s a descriptive model, not a predictive model. The original author of the data doesn’t attempt to forecast, for example, next year or 5 years ahead in time.

The conclusions drawn (building more won’t have much of an impact) seems to conflict with simulation model research done by Sonja Trauss. Her simulation model is in the link below. She suggests that new market rate housing takes hi-income competition out of the market.

How does adding expensive housing help the little guy?

I do think that there is a place for a tax, but not an income tax. A progressive land value tax would work well because it would encourage the efficient use of land and promote denser development. Though state laws on property taxation may have to change for that to happen.

Matt the Engineer

Can you either analyze the arguments you’ve linked to or at least plot their predicted curves over the same data set? They read like 1st drafts of PhD theses, and I’m surprised you’re expecting us to take their conclusions as given. Do you understand the math involved well enough to explain it?

michael goldman

The authors explain their statistical methodologies in the linked posts. Both take different approaches to finding the best fit model. I didn’t want to bog down the article in a statistics lesson. But it’s there for those who want to delve deep into that.

I would like to plot all the functions over the same rents. That’s a great idea. Something I can add to the comments section or at the end of the post soon.

Matt the Engineer

I’d appreciate that. This is an important and non-obvious criticism to the idea that more housing drops rents (how can it not?!). But as the math is difficult to reproduce and is unlikely to be peer-reviewed it would be unfortunate to take the conclusions at face value. The original article was digestible (hey, it’s a quadratic equation, and you can see that it matches the market). These *which equation is the best equation* articles are far less so (to me, anyway). They may be right, but I have no way of knowing if they are from their articles without spending a few days’ worth of effort.

michael goldman

I started to write the functions into a freebie graphing software but got frustrated and gave up. Sorry I couldn’t provide the plot overlays. It would have been cool. As a small consolation, this guy put the plotted the data with mathematica and provided the source code (at the bottom of the page) which should (I don’t have the software) let you play with the variables.

Inspector Spacetime

I wish someone would explain to me how is it possible to say with a straight face that building enough net new rental units won’t affect median rents. Am I misunderstanding this here, or is the argument that it doesn’t just based on historical experiences with boom/bust cycles since 1979? How does that hold any water once you take a second to ask yourself what other variables are at play? What other factors have come into play between post-79 and when much of the preexisting housing was constructed?

It seems to me that supply is being artificially constrained through regulation. The booms and busts are economically (independently) triggered, but the form they take are still constrained by the shape of the building codes and the zoning laws; and by the political culture of resistance to new construction that drove these zoning laws and building codes in the first place.

I’m completely open to the idea that this is nonsense thinking, but to me the main problem with assuming rents are solely demand-driven is that you’re being too narrow in one’s consideration. Of course they’re ‘demand-driven’, but supply is still artificially constrained strong enough to resist price pressures. My first-pass response is to just say supply has not been allowed to increase unfettered by demand.

michael goldman

The original regression (and three others I’ve seen) was run on 40 years of rent data. That means 40 years of development regulation changes including rent control. Development wasn’t artificially constrained to the same degree throughout that whole time (cf. rent control and also periodic upzones). If rents were significantly affected by net housing units, we should expect the regressions to show that.

Cameron Newland

I wonder if the reason that additional supply doesn’t always bring prices down is due to the fact that building more housing causes land prices and construction prices to increase, which naturally disincentivizes further housing development, counteracting the tendency toward falling prices in an environment of increasing housing supply.

Housing economists have found a clear correlation between apartment vacancy rates and rental prices. When vacancy rates rise above 5%, landlords and apartment management companies tend to cut rents and/or offer more rent incentives (a free month’s rent, for instance), and consequently, the median market-rate rent tends to fall. When vacancy rates fall below 5%, landlords and apartment management companies tend to raise rents so as to ration the few units that are available and maintain at least a minimal inventory. San Francisco’s apartment vacancy rate has been sitting around 3% for years now, a clear sign that there is more demand for apartments than supply. I know that Eric Fischer didn’t delve into apartment vacancy rates in his price study, however, in doing so, he may have neglected to include some very important data that could’ve given more weight to the build-baby-build argument (I’m not implying that Fischer should’ve more carefully chosen the data so as to advance a personal bias and to influence the outcome of his housing price study, but I do mean to imply that some revealing data might’ve been missing from the study).

michael goldman

I think you are referring to increasing marginal costs in your first part.

Ott Toomet

Matt Tyler at ‘moreuseful’ concludes that:

“We further find that there IS NOT sufficient evidence that total housing unit inventory has a significant effect on the median rent in San Francisco.” You seem to interpret as “sufficient evidence that housing has no impact on rent”. However, the latter does not follow from the former.

I suspect the problem is simply lack of variation, wages have grown more-or-less steadily over time while the housing stock remains roughly constant, and so the data contains too little information about the housing and rents.

If you take the idea literally, then you can also remove a substantial amount of housing while retaining jobs and wages, and rents remain the same.

michael goldman

Good point and it would be nice to get more longevity on the other variables so they can be compared over more variation… but the housing units variable includes quite a bit of variation already. I see at least four boom/bust cycles since 1974. And within those building booms, Fischer and others did not find a significant (or any) relationship with rents.

Ott Toomet

Thanks for the reply 🙂

Maybe I have time to look at the data myself. Boom and bust are not the best ways to establish the relationship as during the boom period both housing stock and wages are growing… You need some sort of force that only changes housing stock but not wages/labor demand.

One can easily imagine a case where housing does indeed have little effect on rents: more housing means more people means more jobs means more demand for new people… I don’t know empirics here but I would imagine people -> new jobs link is rather slow though.

michael goldman

One of the people who re-analyzed the data checked for collinearity and found wages and housing were highly correlated.