Why the government took home prices out of its main inflation index

A 1983 change reduced volatility but probably didn't lower the average inflation rate.
Why the government took home prices out of its main inflation index
Photo by paulbr75.

The official inflation rate in March 2022 was 8.5 percent—the highest level in 40 years. But some people suspect the true figure was much higher.

The example that comes up most often is housing. Critics point back to the early 1980s, when the Bureau of Labor Statistics (BLS) made a fundamental shift in the way it calculates the consumer price index (CPI) for shelter.

Previously, the inflation rate for owner-occupied homes was calculated based on actual spending by homeowners: the purchase price of the home, mortgage interest payments, property taxes, and so forth. In 1983, the BLS switched to a new method called owners’ equivalent rent. The agency started estimating how much the homeowner would have paid if they were renting their home from a hypothetical landlord.

This methodological innovation understandably sets off alarm bells among people who are predisposed to distrust the government.

For example, in an article for National Review last December, the investment banker William Levin criticized the rental-equivalence approach used in the CPI, calling it “absurdly subjective and unscientific.”

Last week, in the newsletter Common Sense, Samuel Gregg suggested that the government excluded home prices from the CPI in order to “massage the inflation number down.”

Critiques like this invariably assume the switch to owners’ equivalent rent has systematically lowered the measured inflation rate. But is that really true? When I (Tim) started working on this story, I thought it would be straightforward to check. But I discovered that even most experts didn’t know the details of the old BLS approach.

So I recruited my co-author Aden to help me figure it out. Aden dug deep into the archives of the Bureau of Labor Statistics to figure out exactly how the agency calculated inflation prior to 1983. Eventually, he found a little-noticed chart in an economics paper that reconstructed the index through 2018. The paper’s authors shared their code with us, allowing us to extend their calculation through February 2022.

After doing this research, we believe the critics of current BLS methodology are partly right and partly wrong. Today’s annual inflation rate likely would be a few percentage points higher if the BLS were still using its pre-1983 methodology for shelter inflation. But it’s not true, as many people claim, that the switch led to systematic understatement of the inflation rate over the last 40 years.

“We didn’t find that inflation was on average higher or lower with the old method,” said economist Jonathan Hazell, one of the paper's authors. “Rather, it was more volatile for reasons that didn’t make much sense to us.”

The old methodology didn’t just include home prices, it also included mortgage interest payments. And while home prices have risen rapidly since 1983, mortgage interest rates have plunged over the same period.

Low mortgage rates have made homeownership more affordable

Almost everyone finances home purchases with long-term borrowing. As a result, the cost of owning a home depends not only on the price of the home, but also on mortgage interest rates at the time of purchase. And this creates a challenge for government statisticians.

Suppose you buy a home for $500,000, financing the purchase with a mortgage at a 6.4 percent interest rate. A few years later, you sell for $600,000. Interest rates have fallen, and the new buyer is able to get a mortgage at 4.7 percent.

The price of the home has risen by 20 percent, so you could just say the cost of housing has gone up 20 percent.  But that would be misleading. Assuming a 20 percent down payment, your original mortgage would have cost $2,502 per month, while the new owner’s mortgage would cost $2,489 per month. The monthly mortgage payment required to own the home actually went down.

This isn’t just a theoretical possibility. Over the last 39 years, inflation-adjusted home prices have almost doubled. But mortgage interest rates plunged from 13 percent in 1983 to around 4 percent in the first quarter of 2022. As a result, a typical mortgage payment in the first quarter of 2022 was 25 percent lower, in inflation-adjusted terms, than the mortgage on the same home would have been in 1983.

To be clear, this chart only shows principal and interest payments. It doesn’t factor in other costs of homeownership, such as down payments, homeowners’ insurance, property taxes, and maintenance costs.

Why the BLS switched to owner-occupied rent

Prior to 1983, the BLS did try to factor in all those costs when it computed the shelter inflation rate. In addition to home prices, the formula included property taxes, homeowners insurance, and home maintenance costs. The agency also projected the first 15 years of mortgage interest payments and counted them in the year a home was purchased.

This approach faced increasing criticism during the 1970s. Critics raised several issues.

  • When inflation gets too high, the Federal Reserve typically raises interest rates to “cool off” the economy. But the inclusion of mortgage payments in the CPI meant that rising interest rates mechanically increased the inflation rate. That created confusing signals for the central bank.
  • The inclusion of mortgage interest payments was inconsistent with how the BLS handled most other categories of consumer spending. If someone buys a washing machine using a credit card, for example, the BLS just counts the purchase price of the washing machine. It doesn’t count interest paid on the resulting credit card debt.
  • Counting both the home purchase price and mortgage payments was double counting. If a consumer buys a home with a mortgage then the consumer doesn’t actually pay the purchase price—the bank does. Including both made housing substantially overweight in the CPI relative to other possible methodologies.
  • The BLS formula assumed that everyone got a 30-year fixed-rate mortgage. But in the 1970s, more lenders started offering mortgages with variable rates or shorter terms. People also started experimenting with seller financing. So the BLS would have needed to overhaul its methodology to accurately reflect changes in the mortgage market.

More fundamentally, critics raised a conceptual objection: a consumer price index is supposed to measure goods and services that people consume. A well-maintained home can last for decades, if not centuries. Many homes rise in value and are sold for a profit. In this respect, they are like stocks and bonds—investment assets that are not included in the CPI.

So in the 1970s, economists argued that it made more sense to think of homes as capital goods that “produce” housing services. It’s these housing services that actually get consumed. This is easy to see for renters, since the capital good—the home—is owned by the landlord. But economists argued the same principle could be applied to homeowners, who can be thought of as renting their homes from themselves.

That’s the approach the BLS has taken since 1983. Instead of collecting data on what homeowners actually spend to buy and maintain their homes, the BLS estimates how much homeowners would have to pay to rent their homes from a hypothetical landlord. This “imputed rent” is used to estimate the inflation rate for owner-occupied housing.

Shelter inflation isn’t based on homeowners estimating rental values

The obvious objection here is that homeowners don’t actually pay themselves rent and may not know how much they would pay in such a theoretical transaction.

There’s a common misconception that the BLS calculates the change in owners’ equivalent rent by simply asking consumers how much their homes would be worth if they rented them out. Samuel Gregg, the Common Sense author I quoted earlier, raises this objection in his piece, writing that “homeowners often don’t know” how much their home would rent for and that “only some of the responses are factored into official inflation estimates.”

This criticism isn’t entirely wrong, but it misses an important nuance.

The BLS does ask homeowners to estimate the rental value of their own homes, but only to determine how much weight to give owner-occupied homes relative to other categories like milk or haircuts. Changes in these weights don’t have much impact on the calculated inflation rate.

And the BLS doesn’t use this survey to compute the actual inflation rate for owner-occupied housing. Instead, the BLS looks at market data for rents paid for nearby properties with similar characteristics.

The switch to OER meant higher inflation in 2022—but probably not over the last 40 years

At this point in the article, we’d love to present a chart showing what the inflation rate would be if the BLS was still using its old methodology for shelter inflation. But that’s hard to do—both because some necessary data is no longer being collected, and because available documentation doesn’t specify every detail of the old formula.

But we can learn a lot by looking at a reconstruction of the old index that a team of four economists—Jonathon Hazell, Juan Herreño, Emi Nakamura, and Jón Steinsson—created for this paper. Using modern data on home prices and mortgage interest rates, they estimated what the CPI would have looked like if the BLS hadn’t changed its housing methodology in 1983.

These economists shared their code with us, and one of us (Aden) extended their calculations forward to 2022. Here’s the result:

The blue line shows the annual inflation rate as measured by the official CPI-U price index. The red line is Hazell et al’s reconstruction of the old BLS methodology. Before 1983, the two lines are similar—a sign that the economists did a decent, if not perfect, job of reconstructing the old methodology. After 1983—the year the BLS switched to owners’ equivalent rent—the red line diverges from the blue line more frequently and substantially.

The volatility of the red line is mainly driven by changes in interest rates. For example, the average 30-year mortgage rate shot up 33 percent—from 3.35 percent In December 2012 to 4.46 percent a year later. The chart above shows the result: under the old methodology the inflation rate would have been roughly two percentage points higher in late 2013 and early 2014.

In other years the inclusion of mortgage interest had the opposite effect. For example, the official CPI registered a very slight deflation—0.1 percent—in the year before April 2015. Under the old methodology, official statistics would have registered as much as 2 percent annual deflation that month.

In their paper, Hazell and his co-authors reconstructed the CPI through 2018. When we extended their calculations we found the annual inflation rate in February 2022 would have been roughly 11.5 percent under the old BLS methodology, compared to an official inflation rate of 7.9 percent that month. This was the result of both rising mortgage rates (up 34 percent between February 2021 and February 2022) and higher home prices (up 20 percent during the same period).

But while today’s inflation rate would likely be higher under the old methodology, the average inflation rate since 1983 might actually have been lower. Our calculations show that the average inflation rate over the last 39 years would have been around 2.2 percent under the old approach, compared with 2.7 percent under the official CPI.

Interpret these numbers cautiously

We want to encourage readers not to take these numbers too literally. Hazell and his colleagues didn’t have access to all of the data used by the BLS before 1983, and they also made some simplifying assumptions. After kicking the tires on their calculations, we’re confident that the February 2022 inflation rate would have been higher under the old methodology. But we don’t know if it would be exactly 11.5 percent. We wouldn’t be surprised if it were actually 10 percent or 13 percent.

And while we think the change in methodology probably increased the average inflation rate over the last 40 years, we don’t feel high confidence about this. It’s possible that a more rigorous reconstruction of the old CPI formula would find that the 1983 change slightly reduced the measured inflation rate rather than slightly increasing it.

But we are sure of two things. First, the methodological change had only a small impact on the average inflation rate over the last 39 years. And second, there is no way that government statisticians in the early 1980s could have predicted how the change would impact the inflation rate over the long run. That depended on the relative trajectories of home prices, rents, interest rates, and other variables—none of which could have been predicted with any accuracy.

We also think it’s a mistake to assume—as many BLS critics do—that the old methodology represented the "real inflation rate.” Many economists today believe that the BLS’s current approach has the strongest theoretical justifications. Several other wealthy nations—including Japan, Germany, and the United Kingdom—use a rental-equivalence model similar to that used by the BLS.

Reasonable people can disagree about whether the rental equivalence method is best, or whether another approach is superior. But the pre-1983 approach was an incoherent mess and needed to be overhauled one way or another.

Actual rents lag behind “spot rent”

Finally, we wanted to repeat a point one of us made in a piece last month: the way the BLS measures shelter inflation causes its measure of rents to lag “spot rents” by about a year.

Private-sector measures of housing costs show significant increases over the last year. The Case-Shiller index of home prices rose by around 19 percent over the last year, while the Zillow Observed Rent Index rose by 16 percent. But the shelter component of the consumer price index only rose by 5.0 percent over the same period.

The CPI tries to measure what people are actually paying for goods and services. Most tenants sign leases that fix their rent for a year at a time. When rents rise, most tenants don’t start paying more immediately. Their rents stay the same until their lease expires, at which point they might face a big jump.

So the fact that the CPI shows a slower pace of shelter inflation than a private-sector measure like Zillow’s rent index doesn’t mean that the CPI is wrong. Rather, the CPI measures something different: what people are actually paying, as opposed to what they would pay if they signed a new lease.

Research shows that the CPI tends to follow spot rents with about a one-year lag. So the big increase in market rents over the last year should translate to a big increase in the shelter component of the CPI over the next year. In the long run, the two measures of housing costs should grow at the same rate.

This post was co-authored by Timothy B. Lee and Aden Barton, an economics student at Harvard. You can follow Aden on Twitter.


Only members can comment.
Please subscribe to a free or paid plan or sign in to join the conversation.