How to Fix the Housing Component of CPI

Construction on residential apartments in D.C.'s Shaw neighborhood.
Rents here will be higher than those for people who have been living in D.C.’s Shaw neighborhood for years.

Photo by Brendan Smialowski/AFP/Getty Images

For many purposes, including monetary policy, the federal government needs a statistical series to calculate the level of inflation in the economy. Because food and energy prices, though important to consumers, are highly unstable and deeply driven by nonmonetary factors (weather, political upheaval), it’s conventional to strip those out when assessing short-term monetary policy. But when you strip out food and energy prices, then a very large share of the remaining “core” CPI is the price index for housing. And calculating the price index for housing is conceptually complicated because only a nonrandom minority of people rent, so you can’t really just go around and tally up rents.

Adam Ozimek has a paper pointing to what he and I agree is a major practical failing in the way the Bureau of Labor Statistics currently handles this problem, and while I doubt a blog post about his paper is going to go viral on Facebook, it could help prevent future spells of mass unemployment if people read it.

My effort to make this issue a bit more digestable:

How Does the BLS Calculate Inflation for Owner-Occupied Housing Now? It completely ignores issues about the sale price of houses and prevailing mortgage rates and instead “imputes” rental value. Which is to say the question the BLS seeks to answer about your condo is not “how much would it cost someone else to buy your condo” but “how much money are you saving by living in your condo rather than renting an identical one.” Equivalently, it’s asking how much you could earn as a landlord if you started sleeping in your car and renting the house out. So what the BLS does is it surveys the prices that renters are paying for housing, tries to do quality adjustments as it does for all other kinds of goods, and then imputes an equivalent rental value to all the owner-occupied housing.

Doesn’t That Sound Weird? Who Cares About Hypothetical Rents? It does seem weird. It’s worth noting, however, that efforts to create a more “intuitive” price index by looking at what people actually pay have their own problems. In Sweden, for example, they look at actual mortgage payments which are tied to interest rates. That means that raising interest rates to curb inflation is seen as directly causing inflation in the market for owner-occupied houses. At any rate, Ozimek’s point is to raise an issue with the way the rental imputation works not to argue about the merits of rental imputation as an underlying philosophy.

What’s Wrong With the Current Approach to Imputing Rents? The issue is that rents for continuously occupied structures are generally quite sticky. One can debate exactly why this is, but it’s clearly true and you see this very much in gentrifying urban neighborhoods these days. Look at someone in D.C. who’s been renting a house in Shaw in D.C. since 2006, and she is almost certainly getting an outrageously good deal from her landlord compared with what someone trying to move into the neighborhood in 2014 would pay. The landlord fears that if he tried to raise the rent all the way up to the current market price that he would (a) wind up with several months’ worth of vacancy and (b) run the risk of acquiring a problem tenant. Consequently, in an area where rents for new tenants are rising quickly the average rent paid will rise more slowly (and vice versa in a downturn).

What Are You Saying We Should Do About This? The proposal is to ignore average rents for the purposes of imputing rent to owner-occupied housing, and just focus on new rents. In other words, impute rental value to D.C. homeowners on the basis of what someone who just moved to D.C. would have to pay to rent a dwelling rather than on the basis of what existing residents are actually paying. That would create a housing CPI that fluctuates up and down more rapidly in response to changing conditions.

Is There a Thematically Appropriate Song We Could Use To Take a Break? Yep, enjoy:

Obviously in Rent you’re supposed to sympathize with the tenants, but part of the issue here is that when a landlord finds a tenant who does regularly pay the rent on time that’s a valuable relationship and the tenant enjoys a discount relative to what an unknown hypothetical new tenant would pay.

How Does This Relate to Housing Bubbles? Very closely. One of the problems with the existing method is that the housing CPI as currently constructed moves in ways that are unrelated to the ups and downs of housing sale prices. So for example in 2007 and 2008 when house prices were tumbling nationwide, housing CPI rose strongly. That’s because “house prices” are current sales, while average rents embed lots of information about old agreements reached years ago. Focusing on new rents would let the inflation index see the price movements in real time.

Isn’t It Crazy to Totally Ignore What People Are Actually Paying? Not really. Remember that the whole premise here is that we’re imputing rents to homeowners. So however we choose to do the imputation, it’s by definition not based on any rent the homeowner is actually paying. Conceptually we are trying to measure how much money you could earn if you moved out of your house, started sleeping in your car, and became a landlord. Which is to say that we are trying to measure exactly what Ozimek says we should measure—the current going rate for a brand new landlord-tenant match, not a broad average reflecting old agreements.

Does This Make Any Difference in Practice? A huge difference. Throughout the summer of 2008 the Fed was torn between evidence of a weakening economy and evidence of rising inflation. The key driver of inflation was higher food and energy prices, which policymakers knew not to make too big a deal out of. But “core” CPI was above target as well, which policymakers saw as evidence of a general inflationary trend. And, indeed, as measured by CPI the housing component of inflation (the biggest component of core CPI) was more than 2 percent throughout this period. Measured this other way, the housing component of CPI was consistently lower than 2 percent clarifying that the “inflationary” dynamic really was limited to commodities.

So if We Make This Adjustment All Will Be Well With the Housing Element of CPI? Absolutely not. There is a whole very difficult issue around how to account for neighborhood-level quality improvements. If nominal rents are rising and also vacant storefronts are turning into yoga studios and hip coffee shops, does that show that rich new arrivals are driving up prices and also like yoga and fancy coffee, or does it show that rent increases reflect noninflationary improvements in neighborhood quality because yoga and coffee are great? But fixing that would be really hard. Switching from average rents to spot rents is feasible, and we should do it.