American investors began 2008 with a simple New Year’s resolution: sell.
The result was the worst start of a year in the S&P 500’s history, with the index falling more than 5 percent over the first five days of trading.
Explaining market moves is usually a mug’s game, but it’s clear that one of the main causes of the sell-off was this month’s labor-market report, which showed job growth in December at a virtual standstill and unemployment jumping to 5 percent.
For many investors, that news seemed to confirm their deepest anxiety: A recession — or at least a stagnation — is at hand.
This may well be so, but the decisiveness of Wall Street’s response to the numbers was still puzzling, since employment statistics are notoriously muddy. To begin with, the two numbers that the government reports each month — one measuring the unemployment rate and the other job growth — are based on very different surveys, and they frequently offer conflicting snapshots of the economy.
The employment, or household, survey looks at 60,000 households, and last month it saw a sharp increase in the number of people without jobs. The payroll report, by contrast, surveys 400,000 business and government establishments, and last month it said that the economy actually added 18,000 new jobs.
Furthermore, both estimates are significantly imprecise: The payroll report has a sampling error of as much as plus or minus 100,000 jobs (which means that, instead of gaining 18,000 jobs last month, we may have lost 82,000), while the household survey’s error margin is even bigger, at plus or minus 400,000.
The payroll numbers are also subject to big revisions: In September, the government reported that the economy had lost 4,000 jobs the previous month, but a later update said that 89,000 jobs had been created.
This uncertainty has made job numbers a favorite target of pundits, who dismiss them as “meaningless” and “irrelevant,” and accuse the Bureau of Labor Statistics of numerical flimflammery.
The payroll report has also become a flash point for political arguments. A few years ago, when the report showed the creation of surprisingly few jobs despite brisk economic growth, Republicans attacked it for missing the boom in self-employment and new-business growth, insisting that the household survey, which showed very low unemployment, was a better indicator.
Flawed as they are, though, the employment numbers represent a dramatic and valuable economic innovation. The idea that the government can and should give the public a reliable picture of the economy is a surprisingly recent one.
It wasn’t until the Great Depression that the government began calculating a national employment rate, and it’s only in the postwar era that employment data have been systematically and rigorously collected.
And if the results are imperfect, that’s because collecting up-to-date, accurate information about the U.S. economy, where millions of jobs are created and lost every year, is remarkably difficult. Imagine that you’re expected to track every job that has been created or lost this month.
The new coffee shop that opened up in Baton Rouge, the guy who just got fired from your local auto-repair shop, and that kid who left his job to go to law school — you need to account for all of them. And you have to do this without much enforcement power or surveillance ability.
Most respondents aren’t obliged to get back to you in a timely fashion — a major reason for the job-number revisions is that only two-thirds of surveyed businesses answer promptly — and there’s no monthly registry for new companies or for businesses that go under. Good luck.
Faced with this task, one option is to confine yourself to the data from the surveys, without trying to make up for what you know you’re missing — new businesses, failing businesses, and so on. Another is to try to fill those gaps using statistical techniques and past data, and to create models that calculate how many new jobs you’ve missed.
The payroll survey does the latter, relying on what’s called the birth/death model to estimate how many new jobs are being created (and how many old ones lost) as companies open and close their doors.
This model is, by definition, an estimate, and at times when the economy is changing — moving from recession to boom, or vice versa — guesses are more likely to be off. That’s why you’ll sometimes read, from job-numbers skeptics, that half the new jobs reported this month were the result of the government’s statistical adjustments, with the clear implication that those jobs don’t really exist.
Yet studies that measure the government estimates against hard data, like unemployment-insurance tax rolls, have found that the birth/death model makes those estimates more accurate. The paradoxical truth about the jobs numbers is that they are much better than their critics say they are but nowhere near as good as investors believe them to be.
As many studies have shown, people don’t have an intuitive understanding of things like margins of error and random sampling; they prefer to focus on a single number, even if it’s falsely precise, and so end up overemphasizing the report’s headline number.
Investors are also subject to the so-called “salience bias” — high-profile information is weighted heavily even if it’s flawed. That’s why market moves in response to government reports are often surprisingly big — especially when, as now, they seem to substantiate investors’ worst fears.
At this point, the market is locked in a hard-to-break feedback loop: The fact that traders act as if the jobs report were definitive makes it so.
A little information can be a dangerous thing.