This may be a repeat for regular readers, but maybe this post can serve as a single-source summary of some of the quantitative points of my Great Recession revised history. Let’s start with Figure 1, metropolitan housing permits per capita. I’m going to look at Los Angeles (a “Closed Access” city), Atlanta (a typical city), and Phoenix (a “Contagion” city) here. Really, on the supply side, there are just two stories here. Los Angeles approves far too few homes. Phoenix and Atlanta approved a lot of homes before the Great Recession. Even today, after substantial recovery from the lows, they have approved new homes at less than half the rate they did before the Great Recession, but that is still more than double the rate that Los Angeles does.
There's no harm in repeating this--some people might be coming to this information for the first time. The myth of overbuilding in the housing market prior to the Great Recession did great harm to the country as whole---although many areas of the Sunbelt did a better job of recovering from that (even despite the vicious lending standards that are in place now).
And, can I issue a bleg for a post on your take on "rent softening" as it's being reported in conventional media?
Thanks, I should have paid closer attention to that post. I suppose what popular media means by "softening" is that rent increases might match inflation this year.
Random thought, have you ever tried logging the y-axis on this price/income by income graphs? I wouldn’t be surprised if you get a better fit and makes a lot of sense as a functional form
Good question. It's been a while, but way back when I was putting the framework together I played around with it. It didn't really add any information, and, as you can see in the charts here, the residuals are pretty regular, so there isn't really a problem that needs to be solved. Maybe at low incomes, the outlier ZIP codes tend to have more variance, but the bulk of the data looks pretty regular. There are some cities with heteroskedasticity, but it is from idiosyncratic factors that a log scale doesn't address.
Also, where these trends affect human behavior, they play out on an arithmetic scale. For instance, the migration pressures of high housing costs are at least as strong in areas where, say, rents rise from 40% to 60% of average incomes as they are in areas where rents rise from 10% to 30%, so I don't think it helps it to be predictive of human behavior to compress the higher values.
In time series data, where I can either use the change in price/income ratios or the log change in price, the results generally have been similar. So, I'm not sure that it makes much difference.
I do have to truncate the data because price/income levels don't remain linear at very high ZIP code incomes, but even there, it doesn't appear to be a log vs. arithmetic issue. It's more like there is a price/income floor that acts as an asymptote.
There's no harm in repeating this--some people might be coming to this information for the first time. The myth of overbuilding in the housing market prior to the Great Recession did great harm to the country as whole---although many areas of the Sunbelt did a better job of recovering from that (even despite the vicious lending standards that are in place now).
And, can I issue a bleg for a post on your take on "rent softening" as it's being reported in conventional media?
I sort of did here. Is there another aspect of the issue you want me to address?
https://kevinerdmann.substack.com/p/rent-inflation-and-housing-supply
Thanks, I should have paid closer attention to that post. I suppose what popular media means by "softening" is that rent increases might match inflation this year.
Yeah. I think so. Combined with a flirtation with anecdotal rent deflation that will turn out to be noise.
Random thought, have you ever tried logging the y-axis on this price/income by income graphs? I wouldn’t be surprised if you get a better fit and makes a lot of sense as a functional form
Good question. It's been a while, but way back when I was putting the framework together I played around with it. It didn't really add any information, and, as you can see in the charts here, the residuals are pretty regular, so there isn't really a problem that needs to be solved. Maybe at low incomes, the outlier ZIP codes tend to have more variance, but the bulk of the data looks pretty regular. There are some cities with heteroskedasticity, but it is from idiosyncratic factors that a log scale doesn't address.
Also, where these trends affect human behavior, they play out on an arithmetic scale. For instance, the migration pressures of high housing costs are at least as strong in areas where, say, rents rise from 40% to 60% of average incomes as they are in areas where rents rise from 10% to 30%, so I don't think it helps it to be predictive of human behavior to compress the higher values.
In time series data, where I can either use the change in price/income ratios or the log change in price, the results generally have been similar. So, I'm not sure that it makes much difference.
I do have to truncate the data because price/income levels don't remain linear at very high ZIP code incomes, but even there, it doesn't appear to be a log vs. arithmetic issue. It's more like there is a price/income floor that acts as an asymptote.
Very interesting - thanks for the response!
An excellent review of how housing-production constraints have savaged America's middle- and lower-income groups.
End property zoning, and pre-approve all housing production.