Wednesday, July 29, 2009

A Comment on Seasonal Adjustments

by Calculated Risk on 7/29/2009 10:33:00 AM

What if I wrote that U.S. payroll employment increased by 383 thousand jobs in May 2009 following an increase of 259 thousand jobs in April 2009?

Some readers would suspect CR had been captured by aliens or had visited crazytown.

But, in fact, those numbers are exactly what the BLS reported as the actual change in payroll employment in April and May. The economy added 643 thousand jobs over those two months. However no one reports those numbers because there is a strong seasonal pattern to employment.

Even in the best of years, 2.5 to 3.0 million people lose their jobs in January. It happens every year for a number of reasons such as retail cutting back on holiday hires. And just about every July the economy loses over 1 million jobs for seasonal reasons too.

The following graph shows this seasonal pattern:

Payroll Employment Seasonal Adjustment Click on graph for larger image in new window.

The blue line is the seasonally adjusted (SA) change in net jobs as reported by the BLS, and the red columns are the actual not seasonally adjusted (NSA) data.

No one reports the NSA data because the swings are so wild and the pattern very consistent. Unless you follow the data closely, the NSA numbers are meaningless.

The model used by the BLS for seasonal adjustments is very good, and the SA number is the one to use.

New Home Sales Monthly Not Seasonally Adjusted For new home sales there is a strong seasonal pattern too, but in this case I think it is helpful to look at both the NSA and SA data.

Here is how I present monthly new home sales (NSA - Not Seasonally Adjusted).

This shows the seasonal pattern (Spring buying season), and it is easy to compare the pattern for the current year to the previous years.

Of course, for new home sales, I lead with the headline SA data.

And that brings us to the Case-Shiller data.

Yesterday I posted the following graph of the month-to-month change of the Case-Shiller index for both the NSA and SA data (annualized). Note that Case-Shiller uses a three-month moving average to smooth the data.

Case-Shiller House Prices Seasonal Adjustment The Blue line is the NSA data and there is a clear seasonal pattern for house prices.

The red dashed line is the SA data as provided by Case-Shiller.

For this pattern, I'd expect the SA dashed line to run between the peaks and troughs as it did in the '90s.

However look at the last couple of years. The SA dashed line is very close to the NSA line, even with the wild NSA swings. This suggests to me that the seasonal adjustment is currently insufficient and I expect that the index will show steeper declines, especially starting in October and November.

Even with the wild NSA swings, most media reports used the NSA data and not the SA data (Streitfeld at the NY Times used both).