Why un-revised jobs numbers matter
Ignore Friday’s completely baseless conspiracy-mongering from Jack Welch about the jobs numbers. The really interesting and pretty simple question is whether you should even pay attention to the Bureau of Labor Statistics’ monthly jobs data and economists’ predictions of them at all. Matt Yglesias argues the revisions are consistently large enough that we should all just chill out and just wait for more reliable data. He’s not alone in scolding the hype surrounding a number that’s going to change anyway. If you’re an ordinary observer, that argument makes complete sense. If you’re a trader, though, there’s plenty of reason to pay attention to the initial numbers: it moves markets, and that’s as much a rationale as a trader needs. There’s also a data-driven explanation of why both economists’ consensus of the initial number and the number itself should move markets.
This month, BLS’ numbers showed that the US economy added 114,000 jobs in September. Based on the Reuters poll of economists, the consensus prediction was an increase of 115,000. In the context of a labor force larger than 150 million people, that’s a pretty accurate prediction. Economists aren’t always that good, of course, but they’re better than most people think.
Over the last ten years, according to data from Thomson Reuters Datastream, the correlation between the economists’ consensus, represented by the Reuters poll, and the initial BLS reported number has been 0.921 — darn strong. And much stronger than the correlation between the the far less labor-intensive “it’ll just be the same this month as it was last month” method: month over month correlation is 0.811. That gap between the two correlations quantifies the skill and expertise that goes into producing economists’ consensus. Consensus is the closest thing we have to tomorrow’s jobs numbers today and it tends to be closer than it’s given credit for.
But what about those revisions? It’s the revised jobs number that’s most accurate, even if it takes a few months to be produced. Those revisions are conducted as the BLS gets more complete data and are more precise than the initial figure. Some of the best news from Friday’s report was that July’s employment number was revised up 40,000 to an increase of 181,000 and August was revised up 46,000 to 142,000.
Those are significant changes, but the initial monthly number moves markets because it’s an early look at what’s going on in the labor market. Beyond being quick, the initial BLS number is also more accurate on its own than it’s often given credit for: using the same data and decade as earlier, initial BLS non-farm payroll number is 0.966 correlated to the final revision. Traders don’t necessarily care about the pin-point accuracy of the final revision, but that correlation between the initial number and the revision buttresses their ability to take the initial print as worth responding to on its own merits.
It makes sense that ever-impatient and probabilistically-thinking traders react to the initial number. It gives them a view of very recent labor market conditions. Even if it’s imperfect, what they want is a window into economic activity, and one that gets delivered weeks, not months. For this purpose, there’s no reason to let a need for omniscience get in the way of seeing un-revised BLS data as an imprecise but meaningfully incremental data point. Asking traders not to react to that is an impossible and needless task.