Seasonal Adjustment Factors Distorting Analysis

Get your facts first, then you can distort them as you please. – Mark Twain

All the economic news we read is based on data which has been “adjusted” for seasonal variations.  As an example, US Census based retail sales historically fluctuate significantly month by month.  December retail sales are approximately 1/3 larger than January or February.

Yet when January retail sales data is released, the headlines do not scream retail sales are down 30%.  Quantitative factors are developed based on historical data and applied.   If the methodology is perfect, and the economy is perfectly flat – the seasonally adjusted numbers for December and January are equal.  The analysis of the data would say that sales were flat or unchanged.

Welcome to the New Normal

In 2010, the historical data is being distorted by the terrible economy of the Great Recession.  Seasonal adjustment factors are fine tuned at least once per year when the most recent year’s data is integrated into the historical data.

In viewing the post-Great Recession data,  seasonal characteristics seem to have changed.  Many data sets in manufacturing, transport, retail sales, and employment show this new distortion.  If the 2010 data is seasonally adjusted to 2008 or 2009 characteristics, you end up with one conclusion, and if seasonally adjusted to pre-2008 data – you likely have an different interpretation.

We now have a New Normal distortion in seasonal adjustments – yet the seasonal adjustment methodology being used in 2010 is not adjusting for this anomaly.

Christmas Period Unemployment Data Suspect

Econintersect believes the Department of Labor should be commended for its seasonal adjustment methodology – it is considered one of the best-of-the-best.  Their seasonal factors must accommodate a release schedule of one per week – and consider plant shutdowns, layoff cycles at month end, and other firing nuances.   This week their release showed the lowest seasonally adjusted claims number since July 2008 – this is very good news.

The bad news is that the raw data is showing a difference in movement between pre and post recession.  The following graph includes the 20 November 2010 data point.

The closer the seasonally adjusted data is to a straight line in a flat (unchanging) labor market, the better the seasonal adjustment methodology.  Using the BLS U-3 unemployment rate as reference – the years 2006, 2007 and 2009 seasonally adjusted data should be close to a straight line.  2006 and 2007 are close to straight lines – but 2009 shows an ever declining trend.

So far in 2010 in the period of these graphs, the unemployment rate has been constant – and it mimics the profile of the 2009 data.  Econintersect suspects this 2009 data is the profile of the New Normal – and the seasonally adjusted numbers should be showing little week-over-week change.

However, the current 2010 data is over 50,000 claims per week lower then 2009 and currently in a range where historically real job growth can occur – this is good news.

The Department of Labor publishes its seasonal adjustment factors in advance.  There has been no real change for the New Normal yet.

Except for 2007, the seasonal adjustment factors have been fairly static despite significant losses of jobs in goods manufacturing.  Further, Christmas hiring / firing dynamics may have changed in the New Normal – and, if changed, are not taken into account with the current seasonal adjustment factors.

There is no good solution to the seasonal adjustment factor issue except to eyeball the raw data to draw your own conclusion.  Econintersect analyzes the unadjusted data.  For seasonal adjustment factors to have any degree of accuracy, there needs to be several years of consistent data – and the New Normal is just beginning.  For now, we need to recognize that when you read “sales improved 0.7% in November” that the current seasonal adjustment factors do not allow this degree of accuracy.

For the initial unemployment claims data, Econintersect believes the claims are reducing – just not to the extent of the seasonally adjusted numbers are showing currently.  In other periods (such as the 3Q2010), the adjustment factors likely overstated initial unemployment claims.

Others have also discussed concerns about accuracy of seasonal adjustment factors.  For examples, see Michael Shedlock (BLS employment data) and John Lounsbury (NAR housing data).

Economic News this Week:

Econintersect has delayed the release of the December economic forecast to Monday, 29 November 2010 due the the Thanksgiving holiday.  No data released this week is inconsistent with the November forecast of a slightly expanding economy growing at an ever slowing rate.  However, there will be a sliver of good news coming in the December forecast (due November 29).

Our preferred leading indicator, the WLI from ECRI, was less bad in this week’s release improving from -4.5% to -3.1% –  implying the conditions 6 months from today will be roughly similar with possibly a slight negative bias.

Other economic releases reviewed this week by Econintersect are shown in the following table – For interactive table, click here:

Bankruptcy Filings This Week: None

Bank Failures this Week: None