Cyclical Construction and Gluts

There are at least two ways to compare housing supply over time. You can make a ratio with new homes in the numerator and you can either put total population level or population growth in the denominator. Since 2008, new homes per population has been well below the historical range, but new homes per population growth has remained closer to historical norms because the combination of declining fertility after the Great Recession, the pandemic, and recent anti-immigration policies have lowered population growth to new lows.

Population growth was surprisingly stable until 2008, so both counts of new home supply followed relatively similar trends until then. Higher rates of new home construction in the mid-20th century were related more to declining household size than to extra population growth.

Figure 1 compares the two ratios over time. On a permits per capita x 1,000 basis, the US averaged about 7.4 permits annually before 2006, and about 4 permits annually since then. Probably something a bit above 6 after 2007 would have accommodated continued post-World War II household formation patterns.

Figure 1

Neither ratio quite captures everything. New units per new population captures the long-term demand from population growth. But, if, over the course of a cycle, there is demand for accelerating or decelerating vacant units, seasonally occupied units, or household formation, those changes operate on a per-capita scale. If those factors lead to an increase in demand for housing of, say, 1%, it doesn’t matter what the population growth rate is. So, as population growth slows, housing construction will become more cyclical.

That isn’t a particularly important observation about the market before 2008. The market before 2008 wasn’t particularly cyclical. The 2008 crisis wasn’t caused by the reversal of an unusually large housing cycle. It was caused by the mass hallucination of an unusually large housing cycle that didn’t really happen.

Figure 2 is an estimate of neutral homebuilding based on 2.3 persons per home (including vacant homes) and a condemnation/replacement rate of 0.2% annually. In the mid-20th century, the housing stock was almost always growing relative to population. It has been more neutral since then. It’s hard to say where neutral should lie. It depends on the replacement rate. The replacement rate is hard to measure. It can change over time due to economic and cultural changes. And, it is affected by housing supply and housing supply growth.

If cyclical construction before 2008 had risen higher, the crisis would have been less likely to happen. It’s not a matter of whether a supply glut was large enough to lead to the collapse in incomes and production associated with the 2008 crisis. There was no glut at all. There isn’t an empirical claim to be adjudicated. The glut hypothesis was a category error.

Figure 2

Figure 3 compares an estimate of new homes in the Midwest, as a percentage of the existing stock of homes, and population growth in the Midwest. Notice that in the decades before 2008, when population growth declined, new home construction only partially reacted. Not only are cyclical changes in relative housing demand unrelated to population growth, but when population growth declines, condemnation/replacement levels tend to increase. In the Midwest, the neutral annual condemnation/replacement rate was somewhere around 0.5%.

Figure 3

The opposite happens in high-growth areas. More homes are constructed where growth is high, but, also, households tend to use the existing stock of homes more intensively. Because of this, the glut story about the causes of the 2008 crisis is especially untenable.

One issue I run into regarding 2008 is that while the glut hallucination was an important cause of the crisis, it frequently lurked in the background. The public discussion was much more focused on prices. So, one of the ways to salvage the conventional explanation for the 2008 crisis is to say, “So what if there wasn’t a glut. This was about financing and home prices. Prices were elevated and had to contract, and that led to the crisis.”

The thing is, you could tell a supply story in the Midwest. Obviously not of a scale that would necessitate a massive financial collapse, but there was a growing gulf between new home construction and population growth in the Midwest from the early 1990s to 2005. One could make a plausible argument that the Midwest was at the tail end of some cyclical building that naturally tends to even out over the business cycle. That came more from declining population growth than from rising construction. And, a deep contraction had already taken place by 2007. And, there hadn’t been a price run-up that would have led to the market optimism and treadmill of subprime refinancing that is routinely described as the kindling for the eventually collapse. There wasn’t a price run-up in the Midwest that would have necessitated a deep price collapse, negative equity, defaults, etc.

At the national level, there does appear to be a cyclical supply boom story. There was a positive cycle in the national numbers. It wasn’t any more of a cycle than was common in earlier decades, but there was one.

But, it doesn’t show up in the boom regions. Figure 4 compares the national estimate of new homes per new resident to the estimate for Arizona, Florida, and Nevada. Construction in those high growth states was mostly tracking with population growth. Units per new resident was generally flat in the growth states.

Basically, the housing boom was necessitating migration. There’s a bit of a Simpson’s Paradox here. The national trend doesn’t show up in the regional trends. The regions where they were supposedly piling up homes in places we didn’t need them were barely keeping up. The national average isn’t reflected in the regional trends because the cause of the rising national average was a compositional shift in housing production.

The areas with lower home construction per new resident trends were becoming a larger part of the national average.

Ironically, if new home construction in the high-growth states had tracked more closely with the national trend, the hallucinations about a supply glut probably wouldn’t have seemed so salient, and the crisis wouldn’t have been initiated.

Also, note that population growth collapsed in all 3 states after 2006. Yet, even though new homes per new resident hadn’t increased when population growth was high, it did decrease along with the rest of the country even though population growth collapsed in these three states at the same time that new home construction collapsed.

Figure 4

And as Figure 5 shows, it was the slow-growing areas where new homes per new residents was high and rising. (The Midwest and Northeast estimates are hit and miss here. They are unstable because the denominator - population growth - sometimes is so low that the measure spikes up or goes negative. It is at those times that they are especially inconsequential regions. Their homes per new resident measures spike and become unstable because they had little or no population growth. I hope that there aren’t many people shameless enough to try to argue that the crisis occurred because New York City and Boston had a glut of homes before 2008 resulting from the mismatch between population growth and new construction.)

Figure 5

(I cut these charts off after 2019 because very low or negative population growth, and decennial Census Bureau adjustments make these measures unreliable and volatile during the Covid years.)

In summary, comparing new home production to population growth is a useful exercise. Lower population growth will likely lead to lower rates of homebuilding in the future, though not necessarily as low as pure population growth trends would suggest. And, if anything estimating new home construction relative to population growth weakens the conventional story about the origins of our current shortage.