Short-term numbers have shortcomings

Most people assign far too much importance to short-term changes in economic indicators, such as consumer spending, unemployment and prices.

Understandably, a high level of public interest in economic matters prevails right now. We are coming through a slowdown after a sustained period of prosperity. Moreover, we are in the middle of an emotional election in which the economy is a hot issue.

A strong desire for information about trends should not lead us to place undue confidence in any indicator, however. That happens all too often.

Back to fundamentals: A real economy exists with people working, hiring, producing and buying. This real economy is large and complex. Measuring it comprehensively is impossible.

Instead, we statistically measure key but relatively small fractions of the economy. Unemployment data, for example, come from information about one in 2,000 U.S. households. The numbers are estimates — not complete reality. They are useful in guiding decision-making, but are not a perfect record of reality.

Definitions and tabulation procedures for key economic indicators are precise. The economy itself is complex. Interpreting what precisely defined and consistently measured numbers say about a complex real economy is subject to pitfalls.

For example, in mid-July, the New York Times published a story headlined, “Hourly Pay in U.S. Not Keeping Pace With Price Rises.” It asserted, “The Bureau of Labor Statistics reported that hourly earnings of production workers … fell 1.1 percent in June, after accounting for inflation. The June drop came after a 0.8 percent fall in real hourly earnings in May.”

Taken literally, the story said real earnings had dropped by nearly 2 percent in two months — 1.1 percent plus 0.8 percent. Readers reasonably might conclude that if things continue, the buying power of average wage earners might drop by 12 percent in a year.

That didn’t seem right. Examining data from the Bureau of Labor Statistics Web site, real wages for production workers clearly had not fallen by 0.8 percent from April to May and not by another 1.1 percent from May to June. Where did these numbers come from?

The BLS seasonally adjusts many indicators. Many seasonal factors — students getting summer jobs, fall layoffs of construction workers, part-time retail hiring for the December holidays – affect employment. The BLS statistically smoothes out these seasonal jumps to make underlying trends easier to see. But they also publish the non-adjusted numbers.

Had the Times’ reporter compared apples and oranges — seasonally adjusted numbers from one month and non-adjusted ones from another? This didn’t check out either.

Had the reporter inadvertently taken numbers from different categories? In addition to “production workers” in general, the BLS tabulates earnings of 18 sub-categories including construction, financial services and transportation. A search produced no 0.8 percent and 1.1 percent changes in such other series, however.

What if the Times’ reporter had adjusted BLS nominal earnings data by a different price index than the one commonly used? The consumer price index is well known, but it has sub-variants in addition to a “GDP Deflator” with its own “Personal Consumption Expenditures” subset. These price indexes parallel each other over the long term. In any given month or quarter, however, discrepancies can arise.

Deflating current earnings with different price indexes did not replicate the Times’ numbers. Then inspiration struck. Instead of poring over data sets, what about the press release the BLS issues with all new data?

It turns out that the respective 0.8 percent and 1.1 percent declines in production worker earnings had not taken place in May and June. They were, rather, the cumulative declines from the same months in 2003. This distinction may sound arcane to the uninitiated, but a 1.1 percent decline over 12 months is very different from a 1.1 percent decline in one month.

Moreover, because numbers cited were “year-over-year” changes, there was no logic in adding them as one understandably might when told that earnings had declined 0.8 percent in May and another 1.1 percent in June.

This does not mean everything is hunky-dory for production workers. Real earnings are stagnant or slightly worse. Pay per hour is rising slightly, but recent increases in gasoline, milk and other volatile items wiped out any nominal pay increases.

The lesson is that measurements in any single month or quarter suffer inherent small inaccuracies.

Over the long term, minor mismeasurements cancel each other out, but in any specific month, the measured change can depart greatly from long-term trends.

Important nuances such as seasonal adjustments, variations in specific data sub-series, variations between different price indexes and so forth further complicate the situation.

While the economic indicators we tabulate are extremely useful, don’t pay too much attention to apparent changes from one month to the next. They may be real, but they may also be erroneous.

© 2004 Edward Lotterman
Chanarambie Consulting, Inc.