Can the BLS be Replaced?

Private sector substitutes for public sector data

With the firing of the head of the Bureau of Labor Statistics Erika McEntarfer, the present administration has made the world a worse place. The work that they do in estimating the national accounts is incredibly important to firms, policymakers, and to voters. Without the knowledge of what conditions in the economy are like, we cannot evaluate the impact of policies. Firms will have difficulty making plans for the future, and will be less able to know if the shocks they face are idiosyncratic or general. This is not one of those rare instances (like I covered recently) where less information is better.

The accusations manage to be both baseless and logically incoherent. There are two paths to estimating jobs in America, household and establishment, and the reports in question relied upon the latter. They send out a survey to thousands of firms, and have them report their payroll data. Firms are often slow in getting this done, so the BLS essentially reports them as they come in, imputing missing responses, and revises old estimates to match new data. The BLS is doing its job correctly if the direction of the revision is not consistently one way or the other.

It is logically incoherent insofar as Trump has spent several months prevailing upon the Federal Reserve to cut interest rates. The Fed will do so only if employment is weaker than expected and we continue to have low inflation! A low jobs report is bad insofar as he wants employment up, and a high jobs report is bad insofar as he want interest rates low. I found the whole thing appalling in part because it is so clearly a tantrum. The Darth Vader school of personnel management is not a very good one – we should not punish subordinates for accurately reporting adverse information!

The work that they do is not substitutable. Contrary to cynics, it is not being done in secret by hedge funds. The clearest proof is that the market moves after the jobs reports. If firms were already doing what the BLS or the Fed does privately, we would expect their reports to contribute no new information and cause no change in prices. I borrow an example from this blog post; employment was 1% higher than expected. We also saw just a few days ago that, when the revision came in, stocks plunged.

(Separating out the contribution of the Fed is more difficult, because it both reveals information about the true state of the world, and reveals what the Fed’s beliefs are about the world which will then influence monetary policy in the future. In addition, it often comes packaged with concurrent changes in interest rates. There is a considerable literature; I link to Bauer and Swanson (2021) as quite recent work, and you can read the sources in the literature review back if you want.)

It is possible that the statistical work the government does is crowding out private sector work. I think this is highly unlikely, for several reasons. First, information is a classic public good. The marginal cost of reproducing information, after finding it, is approximately zero. Any private company is going to charge an incredibly high markup.

An illustrative example is the closest thing which the private sector does to government data, Nielsen Homescan. They collect two main datasets, the consumer panel and retail scanner data. The former has 40 to 60 thousand people record all of their purchases down to the bar code level, with hundreds of millions of entries a year; the latter is a sample of retailers including all purchases, and is in the tens of billions of observations. It is incredibly powerful, and can answer questions which absolutely nothing else can. (See in particular Jaravel (2019), which measured inequality in inflation experiences which showed up only at the bar code level). It is also extremely expensive – even for academic researchers, who get it heavily discounted, the cost is thousands of dollars.

The government has substantial advantages over private sector firms. Revealing information about your firm is a collective action problem – better if everyone does it, but better still if everyone but you does it. The government can compel responses, and can also collect a census of all people or firms. Back to Nielsen Homescan, for years they couldn’t get Walmart to cooperate with them. What good is it as a measure of the health of the economy as a whole if you don’t know the sales of the largest retailer in America! I do not think there is any way to do what the government does in the Census of Manufactures.

The government has a consistent methodology from year to year, and can guarantee that you are actually getting a sample. This is most relevant for employment figures, where one might think that you could turn to Indeed, BurningGlass, Glassdoor, or other companies to substitute. You don’t know, though, if those companies are a representative sample of the population over time. Maybe things are going well for the company, but not for the economy as a whole.

Any forecaster which specializes in predicting job market data, and does it well, will run into Milgrom-Stokey problems. If someone has a reputation for being right, people won’t naively sell securities to them anymore. Instead, some of the information will leak out. Because they are paying a fixed cost to find information, this will lead to us finding an inefficiently low amount of information.

While we cannot fully substitute for public sector data, there are several things we could try to do to improve private sector data. The first and most important is repealing Sarbanes-Oxley. Passed in 2002, the law substantially increased the cost of becoming a publicly traded company, and is likely responsible for the decline in IPOs since. It has sharp discontinuities based on the size of the stock floated – in particular, companies below $75 million are exempt from hiring an outside auditor, and companies below $700 million receive a number of accommodations on how much they have to test internal controls. Unsurprisingly, companies keep their size below the threshold. You can infer, as Ewens, Xiao, and Xu (2021) do, the value that firms are giving up by keeping small – 4% of market cap! Ahmed, McAnally, Rasmussen, and Weaver (2010) document an average cost of $13 million a year in compliance. A substantial part of the business model of private equity firms is not needing to pay these costs.

Stock price data has advantages which other datasources don’t have. Since it can immediately respond to news, we can infer the costs and benefits of specific events. (So long as, of course, the event is not itself endogenous to the stock market reaction, as I covered before!).

An alternative possibility is to mandate greater data collection by payment processors. Currently, if you have extraordinary access to a private company, you can work with payments. However, these record only what money was sent to who, and not why. This may be impossible, but perhaps we could have what was purchased be automatically integrated into the data going to credit card companies. However, this has enormous privacy concerns, may not consistently represent the whole economy, and will have difficulty linking people together (especially if we want to preserve privacy).

I will be clear. These are inadequate substitutes. The firing is one step closer to banana republic, and is a serious breach of norms. I am confident that the successor to McEntarfer, William Wiatrowski, will execute the duties of the job as he should, but I am less confident that the BLS can withstand continued politicized intrusions. Future administrations should consider setting up the BLS, BEA, and Census Bureau as bodies equivalent to the Federal Reserve, isolated from political vagaries.