The Ludwig Institute is, as far as I can tell a not-particularly prominent think tank mostly centered around the idea that commonly reported government statistics about the state of the economy are, in some way, flawed. In particular, they argue that all the government statistics are under-reporting how bad the US economy truly is.

I haven't seen media outlets pick them up much, but we do get questions about them on askecon from time to time, so I figured it would be nice to be able to link to a post explaining why I think their work is mostly very bad.

I'll be focusing on one particular component -- housing -- of their True Living Cost index, which purports to be a Consumer Price Index (CPI) that's more representative for a typical low-income household. For what it's worth, this isn't actually a bad idea; it'd be nice to have an inflation index that was calibrated for a lower income consumer's consumption bundle since the consumption bundle of a lower-income household might be systematically different than for a higher-income one.

Thankfully, the BLS has done this already, and they find that lower-income households have experienced somewhat higher rates of inflation than higher-income ones; from 2003 to the end of 2021 lower income households experienced a cumulative 3 percentage points more inflation than the representative household and 6 percentage points more than a high-income household.

​

3 percentage points is very different from what the Ludwig Institute reports in their True Living Cost Index, however:

the cost of household minimal needs rose nearly 1.4 times faster than the CPI from 2001-2020, 63.5% compared to the CPI’s 46.2%

​

One of the biggest reasons for this discrepancy, and what the focus of this R1 will be on, is the treatment of housing. The Ludwig Institute writes:

The CPI Housing Index rose 54%; the TLC Index for housing rose 149%

​

This is, obviously, a massive difference, so let's look at the Ludwig Institute's Methodology Report to see what could be driving it.

First, what do they think the BLS is doing when they calculate the shelter component of the CPI?

There are further anomalies that result from the construction of the CPI. One is the failure for the CPI to represent the cost of shelter. Because the CPI measures housing costs as imputed rents (what someone thinks that their current dwelling would rent for) the CPI often does not react to market changes in current rents or housing prices. People are less likely to change their estimation of their house from year to year even if someone looking for rent that year will face different prices.

​

This is completely wrong. The BLS does not do this. The BLS gets their weights for owners' equivalent rents by asking owners what they think their house would rent for, but the values that make up inflation are coming from looking at rental prices for units that are comparable to what the owner lives in.

If an owner lives in a 2 bedroom single family home in Spokane, Washington, their rate of shelter inflation will be calculated by looking at changes in rent prices of similar 2 bedroom single family homes in Spokane, Washington and not by asking the owner of the house every year what they think they could rent the property for.

This isn't a perfect approach, in particular it will have problems when a particular neighborhood has very few rental properties, which can be common in some suburban areas, but it's a very bad look when an institute makes very strong claims and gets basic definitions wrong.

With that out of the way, what does the Ludwig Institute do?

At a really high-level, they take data from Housing and Urban Development's (HUD's) data on fair market rents, which represent the 40th percentile of rent prices* for various kinds of homes (studio, 1 bedroom, 2 bedroom, etc) at various geographies (county, MSA, etc) and use that to construct and index of rental inflation. There are some conceptual issues with this, but the main issue is that their numbers, as far as I can tell, are completely wrong. The numbers are so wrong they don't even agree with themselves.

They claim on their publications that from 2001-2020 the CPI Housing Index rose 54% and the TLC Index for housing rose 149%, but in the data that they say generates this statistic, housing inflation was only 114% over that time period. If you expand it to include 2022 you get up to 146%, which still isn't exact, but is at least closer. This discrepancy, as far as I can tell, comes from the fact that the graph they publish on their website says that the cost of housing increased by about 35% from 2001 to 2002. The fact that they thought there was 35% rental inflation in a year should have been a major red flag for their numbers.

I also have no idea where the claimed 54% is coming from. If you look at the CPI for shelter from 2001:2020 there was about a 64% increase in prices; if you look until 2022 it was about 88%.

So that's not good, but even worse when I try to replicate their numbers -- either the one's on the website or in the spreadsheet -- using HUD's own data I can't. In fact, when I try to replicate them I get something that tracks the CPI for housing very closely! Now, I'm not bothering to exactly replicate their methodology, but it's very noteworthy that me taking an hour to do some really quick and dirty calculations gets me very close to the CPI and they differ from the CPI by almost 60 percentage points (and by about 90 compared to what's on their website).

What's also very strange is that, outside their inability to understand how the CPI for housing works, their methodology for housing is mostly fine. They adjust for some quirks in FMR that I don't adjust for (see the footnote), but then there's all the weirdness of their spreadsheet not matching their publications, and both deviating so much from my numbers and the CPI. I almost wonder if they're making some Excel errors somewhere -- unfortunately they don't release anything detailing the exact steps they used to take the raw data and turn it into their spreadsheet.

To wrap, I actually think some of their ideas are fine, but the execution is so sloppy I'd ignore anything they put out.

• pic of inflation calcs: https://imgur.com/a/nNkQjW2
• there summary: https://assets-global.website-files.com/63ba0d84fe573c7513595d6e/63c1bb0c3b30d0736184ae8c_TLC White Paper Abridged.pdf
• their methodology report: https://assets-global.website-files.com/63ba0d84fe573c7513595d6e/65401d87a95f64045245334e_LISEP TLC Methodology.pdf
• bls inflation for lower income households: https://www.bls.gov/spotlight/2022/inflation-experiences-for-lower-and-higher-income-households/home.htm
• FAQ for owner's equivalent rent: https://www.brookings.edu/articles/how-does-the-consumer-price-index-account-for-the-cost-of-housing/
• link to my code: https://pastebin.com/BwJPXFYt

*40th percentile for the most part -- in a handful of areas, usually representing between 15-30 % of the US population, it will be the 45th or 50th percentile for certain years before 2016. It'd be better to adjust for this in my calculations, but enough counties bounce between categories that it should more or less wash out. For robustness, I redid the calculation using only counties that were always the 40th percentile and it changes nothing.