A thoughtful reader, considering my article earlier where I took up the faulty analysis of subgroups in employment data, asked me about a recent newsletter from John Mauldin.
John Mauldin writes a weekly newsletter with over one million subscribers. From his accounts, he is an influential jet-setter who knows everyone important and provides the inside scoop for his readers. His articles are republished at the most prestigious sources. This one was an Editor's Choice at Seeking Alpha (where I am a contributor), a feature at Barry Ritholtz's acclaimed blog (which I recommend as a top source), and praised and reprinted at The Reformed Broker (which I always read, often quote, and now add to the featured list). At each of these sources the comments were supportive and glowing, praising Mauldin for his wonderful analysis, which (as usual) confirmed the existing beliefs of the readers.
What if he is mistaken?
I take up this subject not just to correct the Mauldin errors. I think his group of advisors is too narrow and too bearish, but I do not think he is intentionally misleading.
But there is a broader question.
Is Economic Blogging Effective in Correcting Errors?
At the Kauffman Foundation's excellent Economic Bloggers Forum, I very much enjoyed the keynote address by David Warsh. He is a pioneer in bringing fine journalism to the Internet. His weekly column (now added to our featured sources) is stimulating and extremely well-written. I recommend watching his entire address here. For those of you without a few minutes, I want to mention one of his themes.
He discusses the differences between print journalism and blogging, and considers a "top down" versus a "bubbling up" approach. It is a theme to which I plan to return. For today, I want to emphasize something that does not seem to be working in the economic blogosphere.
The Internet is supposed to provide a way to correct errors. If there is a blunder in the political realm, the correction "goes viral." Everyone knows about the error in short order. In the economic blogosphere it is quite different.
Errors in fact or analysis persist indefinitely.
This is especially true when the mythical statement reinforces existing perceptions. It is even more powerful when the mythologist adduces a table or chart. The inclusion of a lot of data adds a false sense of authority, since most readers have a poor bullshit detector. I know in advance that I am waging a lost cause, but with Don Quixote in mind, here is a brief attempt.
Mauldin on Job Births and Deaths
Let me focus first on a specific Mauldin contention:
For new readers, the birth / death assessment has nothing to do with people dying, but rather is the BLS's attempt to estimate the number of new businesses that have been created or have "died" within the last month, and they use these numbers to adjust the employment total. They use historical, seasonal numbers to create a model from which they make these estimates. There is nothing conspiratorial about the numbers - they have to make an attempt at such an estimate, otherwise the employment number would be badly off. But the birth/death number can skew the totals a lot more than is typically realized.
Take the last two months. Using the birth/death model, the BLS assumes that 362,000 jobs were created somewhere. That is three times the number of jobs in the headlines we read. Those extra jobs were added into the total because that is what the model told them to do. And over a complete business and employment cycle, those numbers will average out to be pretty close to right. But as I said, they can also be misleading in the short term.
The first paragraph is accurate up until the final sentence. The second paragraph is all wrong. Here are specific errors:
- The BLS does not "assume" anything. Every aspect of the model is based upon the analysis of past data.
- Mauldin makes a rookie blunder, but he is not alone. Nearly everyone does it each month. The birth/death adjustment is not seasonally adjusted. The payroll employment result, the "headlines we read" as John calls it, is seasonally adjusted. The BLS repeatedly warns that you cannot add the two together. As Barry Ritholtz wisely notes here, the non-seasonally adjusted private job creation was 863,000. Any birth/death adjustment is added to this figure, before seasonal adjustments take place.
- The birth/death adjustment is not calibrated to a complete business cycle, as Mauldin asserts. It is geared to annual changes, no matter where we are in the business cycle. This is because the cyclical part of the change is supposed to be picked up by the imputation step -- something that Mauldin does not even mention, but which I have frequently described. These imputations have been tested on an annual basis and worked nicely for many years until early 2009. Those making a different assertion never provide any supporting data.
Now it gets tougher, since you need to understand more about the method to see the error. Mauldin next writes:
The B/D adjustments say that we added 65,000 construction jobs in the last two months, over half the total number of jobs created. Really? US single-family homes set an all-time low sales number this week. Mortgage applications are way down. Home construction is off. Commercial real estate construction is down. Where are those construction jobs?
The problem with this error is that it is more difficult to explain, since no one seems to understand the two-step process by which the BLS estimates job creation.
First, they employ an inference about non-respondents in the survey. The inference, based on past data, is that non-respondents are like those who actually respond. Think about this. Of the 10% who never respond, maybe half of these companies are out of business. The BLS has learned that the offset between dying businesses and new businesses is similar to what is happening in existing firms.
Second, they use the birth/death adjustment to correct for any errors from step one.
To illustrate. Let us suppose that the existing firms showed a huge decline in construction jobs. This is reflected in the "imputation" step. The birth/death adjustment only addresses the residual impact -- what is left over after imputation is finished.
To emphasize, we do not know how many construction jobs were added (or subtracted) by the job creation estimates, but we can guess from the overall report. Construction jobs were down 22,000 on a seasonally adjusted basis. This means that the construction job creation was "punished" proportionally at the imputation step.
The imputation step is much larger, and therefore much more important than the b/d adjustment. It is an error to ignore this as Mauldin (and everyone else) does.
I have been frank about employment problems and more pessimistic than most in my regular monthly estimates. I regularly report that we have disappointing, below trend growth that is not adding enough jobs to solve our problems. We should stick to the facts, not making things worse than they really are.
Meanwhile, the popularity of the Mauldin piece illustrates how economic blogging actually works. It may get better some day, but we are in the early innings of development. We lack a vibrant community where the truth will eventually emerge.