This reactionary and insipid bullshit artistry published yesterday by Jacobin contains “five reasons why overall housing supply deserves our attention” and, as usual for ostensibly “socialist” (actually mealymouthed progressive) approaches on the housing problem, frontends supply-sider psychosis while holding social/public and low-income housing in reserve for some later phase. It should be obvious already that the presence of social or low-income housing considerations should be thought of as, more or less, the crocodile tears of a capital accumulation fetishist, so I won’t be discussing those here.
What I did find interesting was the 3rd reason offered by the author: “Rigorous academic research shows that building new housing supply helps and building nonmarket housing helps even more” – a thesis commonly referred to in the literature as '“filtering” or “moving chains”, which insists that the production of market-rate (MR) or even luxury housing units has knock-on effects in the short-term to reduce rents within a metropolitan statistical area. Though a “growing body of research” is pointed to, there are really only two salient papers promulgating a supposedly empirically rigorous methodology to prove the existence of these chains; that said, I would be remiss to not point out that that filtering has existed in the literature as a sort of woo-woo pipe dream for marginalist psychos for years, even decades.1
Anyway. The papers I looked at for this post are Bratu, Harjunen, and Saarimaa’s, “City-Wide Effects of New Housing Supply” and Evan Mast’s “JUE Insight”, “The effect of new market-rate housing construction on the low-income housing market”. The former sees itself as an update of the latter, focusing on Helsinki whereas Mast’s work looked predominantly at Chicago – both broadly share identical methodologies (with a few quirks owing to the availability of granular data) and especially a full-throated belief in the fealty to the law of supply and demand. More on that in a bit.
Both studies begin with rather stunning data that the researchers state indicates new construction of any type within a particular urban area reduces rent prices. Mast: “constructing a new market-rate building that houses 100 people ultimately leads 45 to 70 people to move out of below-median income neighborhoods, with most of the effect occurring within three years”.2 Bratu, Harjunen, and Saarimaa: “for each 100 new, centrally located market-rate units, 29 units get created through vacancy in bottom-quintile income zip codes and 60 units in bottom-half income zip codes”.3
Already, we have to stop here, because what is being demonstrated is explicitly not a price reduction, but “unit availability”, part of a wider conversation about the relationship between vacancies and price reductions (“loosening the housing market”). More on this later. What these papers are at pains to prove at the moment is that the construction of new MR units “creates” availabilities in units that are available to lower-income tenants. If this link cannot be reliably proven, then all further discussions are irrelevant.
The obfuscation on display here is actually pretty complex by the standards of neoclassical dullards. To understand it, we must first dig into the notion of housing searches and utility maximization; secondly, we will delve into how the “moving chain” gets “proven” empirically; and thirdly, we will look a bit deeper at supply and demand the way it is conceived of here.
Utility maximization
The moving chain is predicated on a simple idea: that tenants are constantly seeking to obtain higher-quality “housing services” from their rented residences. Both studies under scrutiny here identify units which suffice as a “flow” of better housing services as newer, more expensive units which are, at study start, occupied by certain higher-income households, which is I guess a fine enough place to start. From here, things get hazier, yet the general assumption is that households enjoy perfect information on how and where to go in order to maximize the utility gained from housing services.
This is, of course, a common neoclassical assumption, and one which is basically lifted wholesale from Alonso-Muth-Mills (AMM) models, themselves based on Justus von Thunen’s monocentric economy models for agricultural rent as a function of the distance traveled to market.4 As it appears in these studies (that is, as an unacknowledged, wholly accepted precept) the simplified AMM provides some basic preconceptions: first, that neighborhoods of higher-income units are known and more desirable to all households; second, that all households are prohibited from accessing these neighborhoods not by economic circumstance (income, etc.) but by lack of unit availability therein; third, that each household deems it imperative to gain access to units in higher-income neighborhoods (in order to experience higher-quality housing services); and fourth, that there are zero costs to this move, which is instantaneous.
That these assumptions are rather insipid is uh. “Beyond the scope of this post”. Suffice it to say that at bottom what these studies are trying to disprove is the existence of “segmented” housing markets in a rather stupid way, by abstracting away actual differences in household incomes, to say nothing of segregation, legal regimes, and enforced structuration of scarcity in specific urban areas that in no way correlate to the idiomatic idealizations generated by the papers here as 'proofs', as we will see.5
This provides the theoretical backing for assuming the existence of the chain; now Mast and Bratu et al. set about to prove it.
Proving the moving chain
Proof of the moving chain first requires some further, and this time methodological, caveats. Both studies assume that
the city in question is closed – “only within the same metro as their new building”, as Mast puts it,6
that units within a particular area are statistically identical,7
and, as such, all households within a particular area are identical as well.
With this in mind, let’s allow Bratu et al. to explain further:
“We identify the individuals that move into the new buildings during the first year the building enters the register. We call the year when this move happens year t. We then follow these individuals back in time and find the units where they used to live the year before the move. We call this year t − 1 and the units they leave origin units. We classify origin units based on the characteristics of the neighborhoods they are located in and based on whether they are located in the HMA [Helsinki Metropolitan Area] or not. In the next step, we identify the individuals that in year t live in the origin units as defined above. We then follow this second set of individuals back in time and find their origin units and classify them in terms of neighborhood characteristics and HMA status. We continue in this manner for a total of six rounds, which corresponds to the analysis by Mast (2021) using US data”.8
To restate, the researchers identify the household-tenant of a new building in year t and then find where they moved from, then claim to repeat the process for 6 rounds, finding that there is a grand, city-wide reshuffling of all households into the next best units.9 Remember that this happens automatically, and household income does not figure in at the moving stage – it is merely assumed that all households are rational actors viewing the securing of ever-better housing services as existentially imperative.
Note that “origin units”, as stated in the quote above, are simple assumptions based on “the characteristics of the neighborhoods”. Crucially, this means that there is actually no knowledge of income for the incoming household, though the illusion that there is an actual chain of advancing moves is precisely what is purportedly being proven. Both researchers are also not tracking actual moves; as Mast notes, “to focus on connectivity rather than data imperfections or chain decay, I proportionally distribute the weight from the untracked building to other similar buildings in the round that can be tracked...units are identified by tract characteristics, which may not match actual unit quality".10 Put simply, when a building cannot be found, statistical counterfactuals jump in to 'rescue' the chain and keep it running, even if there's no actual proof of a move ever taking place.
Both studies move to downplay this fact, but keep in mind that the salient characteristic in order to keep up the chain (and thus 'prove' ever-progressive outcomes) is not even Census-level characteristics but is a median income figure assigned to a specific tract, zip code, or other area. Once assigned, these neighborhood figures are then sorted into deciles for the city region as a whole. Even when, for example, Mast uses more granular data to retrieve data on the previous residence of a moving household, it is only at the building level, and that building is then identified as belonging to a premade income 'bucket' based on what the data company Infutor both recorded and imputed once, in a cross-sectional study from 2018. The researchers only have actual knowledge of the residents of the units at the top of the chain; the 'proof' of the rest of the chain is the imputation of lower-income residents moving 'up', without necessarily any advance in incomes, into areas in which higher income decile units have been stated to be. “While I use the exact set of new buildings in the first round”, Mast writes, “the remainder of the simulation operates at the submarket level, which makes it easier to calibrate counterfactual locations systematically”.11 Bratu et al. try to control for this by narrowing their statistical neighborhoods to a 250 meter grid, but this still does not eliminate the issue of having zero knowledge of every single household and unit on the chain – just mere imputation.
Even with this shaky statistical landscape, there are further issues with the grandiose claims made by the researchers. Though Mast is rather squirrelly about the statistical strength of the chains he identifies, Bratu et al. give us a pretty clear idea.
This is Figure A5, “Chain length”. Though Bratu & Co. purport to further prove Mast’s admittedly ambiguous results with more high-fidelity data, we can see that 27% of chains end after just one round even with the generous conditions given.12 Further, note that this graph says precisely nothing about the graduating income feature of the chain, but indicates trackable chains alone. Looking at the graph, 87% of chains end by the 5th round – which, as they say, is when the supposed benefit to the less fortunate kicks in: "the share of residents originating from the bottom quintile grids is only 10% and the share increases gradually in subsequent rounds reaching 30% by rounds five and six".13
Supply & demand
The grand outcomes ‘proven’ by these studies depend on an even deeper-seated belief than has been discussed thus far – the incontrovertibly of the law of supply and demand. Leaving the fact that the rental market plainly has mitigating factors with respect to supply and demand given the presence of landlords behind, these papers (and all AMM based modeling in general) relies on a basic faith in the law, which Marx called in his Critique of Political Economy a “metaphysical equilibrium”.14 Marx's main point on supply-demand dogmatism throughout Capital and especially Theories of Surplus Value is actually quite salient to the treatment of it above: consider that in both studies it is assumed that there are no moving costs and that households can occupy new units instantly. Obviously, this is a rather hard to swallow assumption, given leases usually exist for longer periods than month-to-month and are unbreakable by tenants.
Consider also that the housing market under study is closed; new entries are ignored and papered over statistically as necessary. What does this mean? It means that already both researchers have foregrounded rental market equilibrium as their central achievement in mind. This is a problem given that, in the real world, there are obviously an innumerable amount of comings and goings, but also, back on the theoretical plane, that there is some confusion about the ‘starting point’ of the analysis. When a new unit is constructed, are we intended to believe that this unit was produced without the market signaling for it directly (since all households were previously housed), or that there was already a lack of housing (which means the new housing ‘loosens’ the market)? Neither study communicates the prior state of the market; we begin with a new building full of the best units coming on-market all at once – thus, the housing stock prior to the commencement of the study was s, and it is now s+1. With no new households h being formed (and thus demanding housing services), we are left with the lopsided s+1=h to define the current state of things, even before any moves have begun.
Let’s take Mast’s proclamation that the construction of 100 new units for the upper decile creates, by weakening demand pressure, 70 units in downstream, lower-decile markets. Assuming this is true (it’s not), this means that the construction of 100 units results in the eventual valorization of 70 units within the closed city, and their removal from the market – with no new population, these units are unfillable. Thus a surplus is maintained at the lower income deciles with no possibility of these properties becoming inhabited, and the lower decile units “loosen” precisely as the upper deciles see their market “stiffen”. But these units are not in demand; they are pure remainder, and have effectively been eliminated from the market altogether.
Also, as one last potshot, I would like to point out the number of citations vs. the number of abstract reads and tweets from Bratu and Friends’ paper. Pretty indicative of uhh something.
I have actually critiqued one such paper on this Substack already; see here. Note that as the paper by Mense focuses to a greater degree on the construction side of things and the empirical studies discussed in this post on housing search with finished units (kinda), there’s not a perfect overlap in terms and state of play. But I don’t want to cite any of the myriad other papers because they’re all, like Mense’s, metaphysical meanderings through the diseased corridors of the neoclassical mind.
Mast, Evan. “JUE Insight: The Effect of New Market-Rate Housing Construction on the Low-Income Housing Market.” Journal of Urban Economics, July 27, 2021, 103383. https://doi.org/10.1016/j.jue.2021.103383, p. 1.
Bratu, Cristina, Oskari Harjunen, and Tuukka Saarimaa. “City-Wide Effects of New Housing Supply: Evidence from Moving Chains.” SSRN Scholarly Paper. Rochester, NY, August 31, 2021. https://doi.org/10.2139/ssrn.3929243, p.14.
See Fujita, Masahisa, and Jacques-François Thisse. “Economics of Agglomeration.” Journal of the Japanese and International Economies 10, no. 4 (December 1, 1996): 339–78. https://doi.org/10.1006/jjie.1996.0021.
Teresa, Benjamin F., and Kathryn L. Howell. “Eviction and Segmented Housing Markets in Richmond, Virginia.” Housing Policy Debate 31, no. 3–5 (September 3, 2021): 627–46. https://doi.org/10.1080/10511482.2020.1839937.
Mast, “JUE Insight”, p. 7.
Both studies make assumptions based on varying “neighborhood” sizes of similarity between the units within a neighborhood (or zip code, census tract, etc.) due to limitations in data availability. Basically, there’s actually no way to know what any given unit in any given area is actually like, but we can make a guess.
Brutu et al., “City-Wide Effects”, p. 7.
As I stated in the Mense post linked to in footnote 1, there’s nothing about what happens to the now (permanently?) vacant “last unit”, which represents a funny contradiction. If it is ‘eliminated’ from the market, wouldn’t this mean a new unit effectively ‘cancels’ itself out? Who knows.
Mast, “JUE Insight”, p. 8.
Ibid., p. 12.
This number is actually hidden in a footnote on page 14 lmao
Brutu et al. “City-Wide Effects”, p. 13.
Marx, Karl. A Contribution to the Critique of Political Economy. Charles H. Kerr, 1904, p. 97.