I came across a paper the other day which was being touted as providing proof positive that increases in building supply lowered rents for poor (or “low income”, whatever) people. A conference paper by researcher Andreas Mense, “The Impact of New Housing Supply on the Distribution of Rents” develops an extensive statistical apparatus in order to “prove” that residential construction finishes in Germany correspond with an almost immediate decrease in rents which then taper as the new supply is absorbed into existing. I am not going to impugn his statistical method, a) because I don’t care to, b) it seems mathematically (if not at all conceptually) sound, and c) it really doesn’t matter because it is constructed in order to lend “empirical” heft to some extremely faulty positions. Anyway. Let us begin.
Weather
As a tried-and-true marginalist, Mense begins by embarking on a search for a possible deranging influence – or, ahem, an “exogenous shifter” – on the normally serene equilibrium of the German housing market (since he is not only an emphatic disbeliever in social or public housing policy or construction having any affect on rent prices, but “it is not possible to separate the supply of social housing from the supply of private-market housing in the empirical analysis, he must look elsewhere). He ends up alighting on “local weather shocks during the construction phase” which function as a “temporary shifter of new housing supply at the municipal level in order to identify the effect of new housing supply on the local distribution of private-market rents”.
This claim may seem initially seem relatively inoffensive. Rain really does delay construction, but only up until a point. Once the building is effectively sealed, that is, walls are up and roof on, sealant in place, etc., these delays cease to be an issue and construction may continue more or less unabated even during a downpour. Further, construction firms do their best to plan around/minimize rain delays on new builds, for reasons of lost work days/profits in the production process, as Girmscheid makes abundantly clear elsewhere. Mense notes these facts of the building schedule, writing that early spring is typical for construction starts with the walls coming up mid-summer (though he cannot seem to find data proving this), and notes that “if a wet summer prolongs drying times into October or November, the building cannot be completed before the winter, and construction work can only resume once the winter is over”. Again, this makes sense, but only retains statistical relevance if the projects are expected to be started and completed within the same year, and thus a failure to complete the project by November/December represents a “failure” brought about by the exogenous stressor of weather delays.
Mense also does not make any distinction between building types or sizes, even though these would have obvious effects on the timeline of construction. Schalcher, Stoy, and Dreier, in their article “Construction duration of residential building projects in Germany” provide more rigorous data of construction duration with findings taken from 115 residential buildings and find that a square meter of gross external floor data (GEFA) is produced quicker the larger the project is per month owing to a more efficient use of the factors of production. From Schalcher, Stoy, and Dreier’s data, we find that projects of ~12,000 m^2 GEFA average a completion rate of about 650 m^2 GEFA/month, indicating an average construction time of about 18 and a half months with the fastest reported construction time of 15 months – both well outside the yearly construction cycle Mense assumes. A more average project size of 2,000 m^2 GEFA is found to complete at around 100 m^2 GEFA/month, meaning they take on average 20 months to complete.
This is Mense’s first flaw of the essay – untrue in practical terms, delusional in theoretical ones. It is important to note here, in closing, that he joins a proud tradition of neoclassical economists who, unable to conceive of endogenous non-equilibria causa causans, seek explanations without. The greatest, and most stupendously stupid, of these was none other than the “architect of the marginalist revolution”, William Stanley Jevons, who in 1875 (attempting to explain the panics of 1873) proposed that sunspot activity correlated strongly to disequilibrium in the market. Though Mense is a bit more clever than Jevons here, in both senses, the scapegoat is an irrelevant happenstance.
…Rent
Moving on from the rather insipid origin point, we get to the meat of Mense’s argument: that “an expansion of housing supply by 0.1% of the stock causes a decrease of mean private-market rent per square meter by approximately 3%”. Wow! Further, “the estimates imply that adding one new housing unit to the housing stock for every 100 rental housing units offered on the market per month reduces rents by 0.004—0.007 log points”. So, while real rents (as a portion of CPI) increased by roughly 4.9% year over year from 2011-2018, Mense claims that these would have been stable “if 21 additional new units had been supplied to the market for every 100 new units that were completed over this period”. The calculation estimate is rather disarmingly simple: “consider a municipality with the median number of housing completions in December (0.09% of the stock) and assume that housing expands…The estimate suggests that this reduces mean rents in the subsequent year by about 0.09% x 29.8 ≈ 2.6%. However, this actually does not work at a district level (his original quanta of measurement, as a stand-in for a larger metropolitan housing market) and instead turns to “individual-level regressions” – but these individual regressions do not work in Hamburg and Berlin, or ~10% of the total rents data. Granted, 90% is still not a bad number.
Late in the essay, Mense divulges a bit more his method. He begins with actual nominal rent increases which may both rise year over year as a condition of tenancy (that is, the landlord simply sees fit to raise them in an incremental manner relatively independently of broader market movements) or, of course, as a function of inflation. Mense solves the former away by deflating the nominal rents by consumer price inflation from the year previous to arrive at a “real rent”. However, real rents in real estate are defined as the gross asking rent – rent is not calculated the same way as the real wage may be. However, this figure is crucial. Remember that Mense is out to prove that housing entering the market in the December previous has directly prompted a corresponding decrease in rents within the city/district. This requires some sleight of hand – linking this December with the December previous in order to show a cyclical depression of rents which rise through the summer in order to be depressed in the winter. His entire thesis hinges on this causality. Yet all he has is two numbers – his invented statistic of “real rent” (that is, nominal rent artificially deflated) responding the number of housing finishes from last December. This is used to find, as I mentioned above, that if 21 additional units per 100 new units had been complete in Munich year-over-year from 2011 to 2018, rent would have stayed flat. Again, a spurious claim – as it requires an assumption of zero inflation. But even if we follow along with this, Mense’s insistence that 121 units would have kept rents flat is useless both as a counterfactual woulda coulda shoulda, an aspirational target that exhorts us to just a little bit more. Despite his claims this constitutes a revelation actionable in policy, it ultimately relies on such flimsy premises as to be useless. Remember – even his proof of a rent reduction of a scant fraction of a log point, which hey attributes to new housing entering the market, are only blips in a total rise over the research period of 4.9%.
…& Trickle-Down Theory
Even if we accept Mense’s theoretical and methodological errors, yes, we ultimately end up at trickle-down theory, appearing here in the shape of “filtering” – or, answering the “open question to what extent the tails of the local rent distribution are effected” – that is, the final piece of his argument, where we leave the realm of statistics and make overtures to actual policy. Let me allow him to speak in his own words:
“The main idea behind filtering is that houses, as they depreciate, provide less and less housing services, so that the associated equilibrium rents and prices fall. Households [meaning tenants] can thus sort [move] into newer units of higher quality and older units of lower quality, based on their income. This suggests that the impact of new supply by private markets on rents should not be confined to the upper part of the rent distribution”.
This theory of filtering is attributed to both Muth and Sweeney who claim that private market housing supply “could provide an important source of low-income housing supply”. The way it works is like this: a “chain of sequential moves” distributes units down the chain of households, organized by descending income. “As new units enter the market, some high-income households leave vacant their unit, which can then be occupied by other households or lower income…this process is thought to continue to the bottom of the income (and housing quality) distribution”. That is, “the rents of high-quality housing units…should react first [to a supply shock], while it might take longer until rents decline at the bottom of the quality distribution”. This model also disregards moving costs, though Mense notes households have to weigh the utility benefits of gaining access to newer, nicer housing against moving costs (and, presumably, the typical 3 month deposit in Germany, which is not mentioned"). At bottom, the entire structure depends on an assumption that a household always moves from a worse to a nicer unit. There is not even a theoretical possibility of a lateral or backwards move on the quality ladder. (It is worth noting here that unit quality is almost entirely based on building age, which completely ignores a large number of ‘luxury’ units, especially in older Euro-American cities.)
After stating his basic premise Mense embarks yet again on an elaborate statistical proof, which I don’t care about at all because statistics can be used to prove anything you like. His findings echo his initial premises, and he claims that, again, the “substitutability of housing costs” implied by the ladder of income and quality proves that “new housing supply provided by private developers”, bizarrely assumed here to always be luxury housing by virtue of being new (there is not even a possibility of low- or mixed-income construction entertained here) always nevertheless “effectively lowers rents throughout the rent distribution, shortly after the new units are completed”. Again, similar as his facile proof of weather effects on construction finishes, evidence is arrived at through both statistical obfuscation, the clever isolation of a particular market period in which a certain effect may appear accurate in experimental conditions, and finally the torturing of facts until they fit the model.
In this model, a new unit hits the market. This immediately prompts the household living in the next-best (that is, next-newest) unit to engage in a utility-increasing move to the newest unit. This is assumed to happen locally (within a distinct housing market), regardless of the length of tenancy of that household (which even among renters tends to be quite long in Germany), an extant lease (signing 2 year leases is common) and irrespective of market conditions (such as a general depression, inflation, etc.). This move on the part of the first household then prompts the household living in what is now the third-best unit to jump up to the second, and the fourth to the third, and so on, proceeding in a great chain of utility maximization until the worst-off household vacates the worst-quality (oldest) house and it either falls out of the housing stock in general (causing problems for the supposed need to constantly increase the stock overall) or becoming the domicile of a previously homeless person/household (which it clearly is not). The kick in the pants which causes the initial move that kicks off the entire chain is speculated by Mense to be an “overall quality depreciation during an individual stay” which is “substantial” but this is, of course, due to the fact that the metric of quality is building age – a year passes, quality decreases.
It is never quite made clear exactly how the filtering model proposes that fortunes are improved across the rental market other than households find themselves in a higher quality unit than they previously occupied. Note that at no point is price actually entered as a variable here – simply quality of unit. If it is assumed that price (or at least rent as a portion of the consumer price index) remains stable, then of course this is a net increase. But if that rent increases, it may not – in fact, moving for a household may be impossible on basic terms of income, or so devastating of a portion of household income as to actually minimize their combined utility. However! Were that household to break the chain, they would effectively be depriving the lifestyle improvements in housing quality the households above them enjoyed from both themselves and all those below, given that in this model vacancy does not exist except for frictionally as one household is briefly in transit to another.
Anyway, I have no good way to wrap this up other than saying if this is the best YIMBY-development freak types can do they should uhhh. Try harder.