Close

This page had been redirected to a new URL, please update any bookmarks.

Font Size: A A A

macroblog

« Some meaty minutes | Main | Unemployment rate: Count me surprised »

July 17, 2009

When cycles collide

Yesterday we saw that initial claims for unemployment insurance declined rather sharply again last week, another hint that U.S. labor markets may be beginning to mend. But the improvement came with a word of caution from the folks at the Department of Labor, who note that these numbers are being affected by seasonal adjustments to the data that may present a misleading picture.

Virtually all of the economic information that gets reported by the data agencies has been seasonally adjusted. That is, the data are being reported after the agency has adjusted them for their usual variation for that time of year. The unemployment insurance claims data are a useful example. On an unadjusted basis, the initial claims data showed a fairly large increase last week—up 86,000 workers. But claims for unemployment compensation typically rise in early July as auto plants shut down to retool for the new model year. The jump in claims this July hasn't been as large as in years past since many of the auto plants were waylaid earlier in the year. So on a "seasonally adjusted" basis, the data showed a drop in claims of 47,000 workers.

Here we have that statistician's equivalent of an old existential puzzle: Do seasonal layoffs in the auto industry make a sound if there is no one there to lay off? We invite you to write up your own answer to that one. There is a long literature, perhaps most notably the work of Jeffrey Miron, that documents the interplay between the business cycle and the seasonal cycle. The thrust of this research is to help us better understand the general nature of economic cycles. But there are also more mundane issues we need to wrestle with. For instance, how accurate are seasonal adjustments to the data during times of severe cyclical disruption?

To provide some perspective, let's take a deeper look at the recently released June consumer price index (CPI) report. Last month, prices, as measured by the core CPI, were up 2.4 percent (annualized) from a 1.7 percent rise in May. There are a few bits of data that might cause you to scratch your head a little, given what we've been hearing about the economy lately. For instance, apparel prices jumped an annualized 8.8 percent last month. And new car prices were up 8.2 percent. So are department stores and car dealers having a better time of it than they are letting on? I don't think so. I believe the seasonal adjustments in the data offer a more reasonable explanation.

Apparel prices rose 8.8 percent on a seasonally adjusted basis but fell a whopping 25.5 percent (annualized) on an unadjusted basis. And new car prices? Well, they rose on an unadjusted basis, but not nearly as much as the seasonally adjusted data indicated (5.1 percent versus 8.2 percent). Indeed, this pattern seems to be consistent across many of the core components of the CPI in June. On a seasonally adjusted basis, the core inflation measure rose 2.4 percent. But on an unadjusted basis, the core CPI was largely unchanged for the second month in a row (up a slight 0.9 percent, annualized).

Here's a conjecture on my part. Many of the price declines that ordinarily occur in June didn't occur this year. Why? Perhaps it's because the sharp decline in business activity has resulted in such severe production and price cuts already that usual seasonal price discounts have been disrupted. In other words, in the current economic environment, there may not be much "season" to adjust for.

This isn't a criticism of seasonal adjustment. In fact, seasonal adjustment is an entirely appropriate—and necessary—transformation of the data if you are trying to see emerging trends. But it's certainly important to exercise caution when interpreting seasonally adjusted data during a period of strong cyclical movements. If the business cycle alters the usual behavior of the seasonal cycle in the data, seasonal adjustment could produce a misleading snapshot of the data. And I suspect we saw a bit of that in the June CPI report.

In closing, I want to put in a plug that the second weekly postings of the Atlanta Fed's Economic Highlights and Financial Highlights are now available on the Bank's Web site. These summaries of national economic and financial statistics complement our monthly REIN reports on the Southeastern economy.

By Mike Bryan, vice president and senior economist in the Atlanta Fed's research department

July 17, 2009 in Business Cycles, Data Releases, Inflation, Labor Markets | Permalink

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341c834f53ef0115721243c8970b

Listed below are links to blogs that reference When cycles collide:

Comments

In a low inflation era firms tend to raise prices once a year-- typically at the start of the year or season.

This produces a very strong seasonal pattern to the not seasonally adjusted (NSA)core cpi.

Some 55% of the annual increase in the NSA core cpi occurs in the first quarter and another 25% happens in the third quarter.
The third quarter consist largely of home owners rent, tuition and autos.

Moreover, if the first quarter nsa core cpi is less than( or greater than) in the first quarter of the prior year that the annual change is generally less( or greater) than in the prior year.

This rule has worked in 15 or the last 16 years.

Posted by: spencer | July 17, 2009 at 02:07 PM

It's called seasonally mal-adjusted. One should alwasy ignore seasonal figures. These flucuations originate from the FED's mandate to "SUPPLY AN ELASTIC CURRENCY". I.e., the FED's technical staff follows the fallacious "real bills" doctrine.

Posted by: flow5 | July 18, 2009 at 02:10 PM

Contrary to the economic fraternity monetary lags are uniformly fixed in length. The statistical analysis of these crests and troughs are not random. The rates-of-change in these monetary lags (for real-growth, & for inflation), literally oscillate (along the Y axis), between their maximum and minimum levels (as demonstrates by the clustering on a scatter plot diagram).

These oscillations do however suffer from errant data. Errant data may originate from faulty theoretical interpretations, flaws in the data’s definition, and errors in the computation, collection, and reporting of data.

It is instructive that the FED has never cooperated by supplying continuous, comparable, and timely data. Supporting data is required for the proper investigation, the subsequent proof, and ending conclusion, for any economic research (“History is full of bad jokes”).

Posted by: flow5 | July 18, 2009 at 02:11 PM

This article suggests that seasonal adjusted data can lead to misleading snapshots of the economy. The reason is that in estimating the seasonals, the data producers have not taken into account of lower seasonals due to lower trend or trendbreak. In other words, the key assumption is that the seasonals are more or less constant.

However, this key assumption of near-constant seasonal may not be true.

First, DOL or BLS may be using X12-ARIMA (a seasonal procedure) that allows adjustment for sudden trendbreak. Seasonals will change if trendbreaks are adjusted in the seasonals estimation process.

Second, if multiplicative (or log) model is used to estimate the seasonals, lower trend will lead to lower seasonals.

The seasonal-adjusted data may not be much misleading if possible trendbreaks are adjusted for and multiplicative models are used.

Suggest that Alanta Fed check with the DOL or BLS, and see if they agree with your post. It's better to get views from both parties, the data users and the data producers.

Posted by: low | July 28, 2009 at 10:03 PM

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in