Initial Unemployment Claims: More Seasonal Adjustment Distortions

Beware The Initial Claims Data – It Might be A Statistical Mirage

by Lee Adler, The Wall Street Examiner

The Labor Department reported that seasonally adjusted (SA) first time claims for unemployment rose by 46,000 to 388,000 from a revised 342,000 (previously 339,000) in the advance report for the week ended October 13, 2012. The consensus estimate of 360,000 was a huge miss, as usual. For a change, the seasonally adjusted data did not totally misrepresent what was going on with the trend of the actual, not seasonally adjusted, data.  But there were calendar issues which probably caused both the SA data and the actual NSA data to be misleading.

Along with the headline seasonally adjusted data, which is the only data the media reports, the Department of Labor (DOL) reports the not seasonally adjusted data. It said in today’s press release, “The advance number of actual initial claims under state programs, unadjusted, totaled 359,048 in the week ending October 13, an increase of 29,129 from the previous week. There were 357,562 initial claims in the comparable week in 2011.” [Added emphasis mine] There’s the blockbuster. This was the first year to year increase in 18 months. That was a one week flash in the pan, and I suspect that this will be as well.

Here’s why. The problem is with the comparable week. The DOL is, not incorrectly, comparing this week with 52 weeks ago. But due to leap year, this week ended on October 13, and 52 weeks ago ended on October 15. The two day drift this year matters because last year was the third Friday in October. This year was only the second Friday. In every week of October going back to 2002 the second Friday in October had a big increase in claims. At the same time, going back to 2002 the third Friday in October had a huge drop in claims on 8 of 10 occasions.

This year results in the anomaly of comparing the second Friday of the month, when there’s normally a big increase, with the third Friday last year, when there’s normally a big decrease. It wasn’t a fair fight. The deck was inadvertently stacked to make the number look bad. The seasonally adjusted number suffered from the same anomaly. If this week’s NSA number was compared with the second Friday of October last year, the difference would have been a year to year decline of 42,000 or 10.4%, which is right in the middle of the range of the rate of change of the past couple of years.

There will be those who accuse me of mental masturbation in that exercise, but I’m confident about the explanation. If I’m correct the numbers should course correct over the next week or two, and the recent trend should be reaffirmed. If that does not happen, then I’m wrong, and we should start looking for confirmation of a change in the trend.

Note: The DOL specifically warns that this is an advance number and states that not seasonally adjusted numbers are the actual number of claimants from summed state claims data. The advance number is virtually always adjusted upward the following week because interstate claims from many states are not included in the advance number. The final number is usually 2,000 to 4,000 higher than the advance estimate. I adjust for this in analyzing the data.

Lately the increase between the advance number and the final number the following week has been around 2,500-3,000. Last week it was closer to 3,000 which is what I used this week as the adjustment. The adjusted number that I used in the data calculations is 362,000, rounded.

Note: To avoid the confusion inherent in the fictitious SA data, I analyze the actual numbers of claims (NSA). It is a simple matter to extract the trend from the actual data and compare the latest week’s actual performance to the trend, to last year, and to the average performance for the week over the prior 10 years. It’s easy to see graphically whether the trend is accelerating, decelerating, or about the same.

While this week’s data is above the range of the improving trend of the past two years we saw a bit of the opposite the prior week. If I’m correct about this being a calendar effect, then the rate of change should shortly settle back within the range of weekly fluctuations in the rate of change from -3% to -20% that have occurred since mid 2010. Since mid 2011 the annual rate of change has been within a couple of percent of -10% in most weeks. The trend has been remarkably consistent.

This week’s data was the second time that the rate of change has been above the range since 2010. The last time was the week ended April 30, 2011 when there was an increase of 5.4%. That instance was immediately reversed, with the rate of change falling back to -3% the following week and -12% in the week after that. If something similar happens here, then nothing has changed. Only if the rate stays positive would that signal a potentially significant change in the trend.

Initial Unemployment Claims - Click to enlargeInitial Unemployment Claims – Click to enlarge

With this week’s calendar issue, it appears that the seasonally adjusted data has again departed from trend. Adjusting the calendar comparison date by one day shows the downtrend in the NSA data continuing on its trend path. But the SA data has departed from that path, showing a less favorable trend of improvement. I had been pointing out the that the seasonally adjusted data was misrepresenting the trend for the past several months. It finally got back in line in the week ended October 6, but now it’s apparently veering off track again. It is normal for the SA trend to be wrong. Regardless of how the NSA actual data goes in the weeks ahead, the seasonally adjusted data will go off track sooner or later. It’s just the nature of the beast.

Initial Claims Seasonal Adjustment Off TrackInitial Claims Seasonal Adjustment Off Track – Click to enlarge

Plotted on an inverse scale, the correlation of the trend of claims with the trend of stock prices over the longer term is strong, while allowing for wide intermediate term swings in stock prices. Both trends are largely driven by the Fed’s operations with Primary Dealers (covered weekly in the Professional Edition Fed Report; See also The Conomy Game, a free report). This chart suggests that as long as this trend in claims is intact, the S&P would be overbought at approximately 1450, and oversold at roughly 1200. On that basis it became overbought in mid September

I wrote earlier in September that “barring a much stronger improvement in the claims data which would suggest accelerating growth, the market will be at greater risk of a correction in the next few months.” But the Fed may have changed the rules on September 13, by adding more cash to the markets every month and promising to continue to so so until the unemployment rate showed sustained improvement. That will take until a new bubble emerges, spawning new fake jobs, or conversely, until rises in food, energy, and materials costs force the Fed to reverse course. [I cover the technical side of the market in the Professional Edition Daily Market Updates.]

Initial Unemployment Claims and Stock Prices - Click to enlargeInitial Unemployment Claims and Stock Prices- Click to enlarge

For more charts and additional analysis, visit the permanent Employment Charts page, updated whenever new data becomes available. You can bookmark that page for future reference.

More Economic Charts

Follow my comments on the markets and economy in real time @Lee_Adler on Twitter!

Stay up to date with the machinations of the Fed, Treasury, Primary Dealers and foreign central banks in the US market, along with regular updates of the US housing market, in the Fed Report in the Professional Edition, Money Liquidity, and Real Estate Package. Try it risk free for 30 days. Don’t miss another day. Get the research and analysis you need to understand these critical forces. Be prepared. Stay ahead of the herd. Click this link and begin your risk free trial NOW!