Regional average rents are kinda noininformative

At the risk of engaging in what Scott Breyer dubbed on Fboo as “tedious overthinking” (😑😑😑😑 like that ever stopped me), I need to say something: regional average rents and average land prices do not really show us all that much of policy significance. People cite them, and it makes me wiggly because measures of central tendency only indicate trends if there are well-behaved, stable distributions. We really don’t know what’s happening to the bottom of the distribution unless we have the standard deviation; without stable distributions, it’s possible for average rents to stay the same, or even go down, while the rents at the bottom go up. It’s an embarrassing empirical mistake, and I’m tired of watching smart people toss out average rents or vacancy rates without caveating their discussions appropriately. The reason people do this is that averages are often all that’s available; unit data are expensive to obtain, and the companies that have these data tend not to let you have them unless you give them gobs of cash. But data constraints are no excuse for poor discussion, bad framing, and overstatement.

First off, usual disclaimer: lowering supply constraints is extremely important, particularly in US coastal cities. So if you plan to ‘splain supply to me again, don’t.

Yesterday I saw somebody claim that the Inland Empire still had higher than the national average rents, so there must be some housing shortage there. I suspect there may a mismatch between the vacant units available and what people are looking for/can afford there, but comparing a given place to the national average…people, look, some data points have to be above the average–it’s a mathematical requirement of the measure. All locations in the US are not made equal in terms of economic productivity, so even if there were absolutely no supply constraints anywhere, and absolutely all units were exactly identical, we would still have a distribution of rents in order to construct an average from.

Unless you are building in Lake Wobegone.

Housing and location markets are segmented, with asymmetries in the ability to move across segments; in markets with supply constraints (and even with unregulated land markets, supply is likely to lag demand with urbanization), those in higher segments of the market generally have greater ability to move down (downward raiding) than those in lower-priced segments have to move up, unless wages are growing sufficiently fast in the lower segments relative to the upper segments. This is not what has happened, in general, over the last few decades: wages at the top have grown in real terms, those at the bottom have stayed stagnant or decreased in real terms.

Market Urbanism also featured a nice modeling exercise illustrating some of these problems here if you want to play with it, including some thoughts about length of adjustment periods. I could parse out my problems with the assumptions in the model, but every model has its assumptions and issues, and this discussion lays out its assumptions fairly. It’s a toy model with representative consumers, and it doesn’t claim to be anything else.

The distribution can change in ways that really disguise important rental differences. Here are two distributions of average rents by zip code in (west and a bit of south) LA versus the lower-end rents recorded in those various zip codes. As you can imagine, there is likely to be quite some difference in building units for the average market rate in some of these locations versus others in the short term, and there is nothing wrong with that–it’s normal market functioning–but there is something wrong with assuming those on the low end of the market will do fine if you change rents in the short term or that movement in average rent, even if it’s the direction we want, is enlightening about what is going on on the bottom.

Plot Zoom

I got these from this map on Zumper. It’s not perfect by a long shot, but’s worth looking at for my point. This turned out to be a nice distribution so that we can see a global average out of the averages.

What we’d really like to know are how many units there are in each of these zones at each price level because equally weighting these zip codes is wrong. We’re just illustrating here, so we have to live with inaccuracy.

We can also show how the rent gradient in LA behaves going west to east, from Santa Monica to downtown:

Gradient ai 100 RGB GPU Preview

One thing I do like about average rents as a metric: if there is sufficient spread in the data to see a credible distribution (this is not true everywhere), I think it could be a good strategy to have a graduated regulatory structure where once rents or land values get more than one standard deviation from the median, it triggers an automatic upzone in the zip code of some kind. I think that would change the incentive structure quite a bit for landowners.

Just as a note: if somebody plans to send me the nice Sightline piece that inspired Scott’s summary above, a) I’ve read it, and it’s quite nice to explore what is going internationally, but b) it’s not a causal analysis and b) please also don’t confuse six decades of good, comprehensive housing and social policy (Germany) with plunking down large new projects in lower income American neighborhoods and expecting anybody to believe we would have the same outcomes. Most of the western liberal democracies that do a better job with land use than us also do a better job of social policy more generally. I’m sure that lower supply constraints is absolutely part of the solution, but it’s probably fighting less of an uphill battle in places where society assumes some financial risks for its members rather than having individuals bear them (and then wonder why individuals are extremely risk averse about home asset values).

One thought on “Regional average rents are kinda noininformative

  1. Yes! On unit-level data: I think we are going to see that become more democratized. As you may already know, folks at Berkeley have been developing scraping of these datasets, and I think as more people do it the price points for the proprietary stuff will come down a bit. I bought a second round of data from a provider after a first hit for some work–and in that 2 year gap the price had dropped substantially ($10K per MSA down to like $5K I think it was).

Comments are closed.