**For those who enjoy mucking around in the entrails of economic data and procedures, I present for your enjoyment this treatise on the seasonal adjustment process and initial claims for unemployment benefits**.

This was provoked by a blog post at Newsbusters.org titled **“Wires Trumpet ‘5-Year Low’ in Seasonally Adjusted Jobless Claims, Ignore Year-Over-Year Rise in Raw Claims.”** The author, **Tom Blumer**, looked at the seasonal adjustment process for initial claims for unemployment and concluded that the data was rigged. I was intrigued because I’ve had suspicions about BLS data for some time. Unfortunately, **my conclusion in this case is that Mr. Blumer did not look closely enough at the data.**

### The Seasonal Adjustment Process

Seasonal adjustment has a long history with data of all sorts. It’s a statistical process designed to identify and quantify the regular ups and downs that occur during a year. Seasonally adjusting the data allows economists to focus on the underlying causes of fluctuations in economic data without having to worry about seasonal factors getting in their way.

**Examples of seasonality are not hard to find. Ice cream sales rise in the summer and fall in the winter. The demand for ski gear rises in the winter and falls in the summer** (unless you’re fortunate enough to be able to follow the snow into the other hemisphere of the planet). ** Identifying seasonal patterns and putting them into numbers lets us improve our analysis and forecasting. Generally, it is a good thing**.

While I’m tempted to include an example here, I’ll put that off until later. Those who would like to see the example can easily scroll to the bottom of this piece where you’ll find all the numbers you could want. First I want to look at initial claims for unemployment benefits (sometimes called first-time claims).

### First-Time Unemployment Benefit Claims

**These claims are filed by people who have just become unemployed for whatever reason**. Consider workers in the retail ice cream business. We might expect a number of them to file first-time claims beginning in mid-November as the ice cream business begins to melt (sorry). ** In this context, “initial” and “first-time” mean the first claim filed for the current spell of unemployment**. It is not necessarily the first time in their lives or even the first time this year. The Department of Labor reports this data weekly, making it an excellent, if noisy, signal of likely changes in broader employment and unemployment statistics. You can download the DOL data by clicking here, but that will give you a web-style table. If you click here you can download the data since the beginning of 2011 as an Excel file (courtesy of your author).

**When would you expect first-time claims to rise the most? In the U.S. most of us would guess after the end of the Christmas shopping (and returns) season. And you would be correct. That’s the basis for Mr. Blumer’s claim made in the following table:**

The problem is that there are not exactly 52 weeks in a year. For “prior year” Mr. Blumer used the week ending January 14, 2012 when the seasonal factor was 144.2. The question is whether that was the correct week for comparison purposes.

Let’s look at the calendar. In 2011 Christmas and New Year’s days were on Sundays. Most businesses closed on at least one of those Mondays. But that left four workdays in the week of Monday, January 2. The DOL says that week was the first week after the Christmas shopping season. Therefore, the seasonal factor is 165.9, indicating that DOL expects 65.9% more first-time claims than the annual weekly average. For the week of January 9, the seasonal factor drops to 144.2. Quite a few layoffs occurred in the week of January 2, with somewhat fewer in the week of January 9. But notice that all three seasonal factors are above 100, meaning the layoffs begin right after Christmas. (Actually, the first seasonal factors less than 100 are in November, extending back into October. That’s when the holiday hiring is happening.)

December, 2011 – January, 2012 | Seasonal | ||||||

Sunday | Monday | Tuesday | Wednesday | Thursday | Friday | Saturday | Factor |

25 | 26 | 27 | 28 | 29 | 30 | 31 | 141.1 |

1 | 2 | 3 | 4 | 5 | 6 | 7 | 165.9 |

8 | 9 | 10 | 11 | 12 | 13 | 14 | 144.2 |

What about 2012? Christmas fell on a Tuesday, as did New Year’s day. Returns continued for the rest of the week of December 31, meaning the real layoffs didn’t begin until the week of January 7.

December, 2012 – January, 2013 | Seasonal | ||||||

Sunday | Monday | Tuesday | Wednesday | Thursday | Friday | Saturday | Factor |

23 | 24 | 25 | 26 | 27 | 28 | 29 | 133.4 |

30 | 31 | 1 | 2 | 3 | 4 | 5 | 148.6 |

6 | 7 | 8 | 9 | 10 | 11 | 12 | 166.0 |

The seasonal adjustment factors match our crazy calendar. There’s one more way to see this. I downloaded weekly data from the beginning of 2011. Here’s what the seasonal adjustment factors look like over that time:

**Every single year there is a spike to the neighborhood of 165 in one of the first two weeks in January. But you must be careful to pick the correct week based on the day of the week on which Christmas and New Year’s fall.**

### Conclusion

When I first read Mr. Blumer’s piece, I thought he had found the smoking gun that would show the statistics are being cooked. I’m afraid that’s just not the case — at least this time. Keep trying, folks.

### A Seasonal Adjustment Example

Let’s use the actual numbers from the first-time claims data. I’ll use the year 2012 since it’s the most recent year for which we have all the data. The calculation is simple: the seasonally adjusted figure (*SA*) is equal to the non-seasonally adjusted number (*NSA*) divided by the seasonal adjustment factor (*SF*) divided by 100:

Why divide by 100? For reasons best known to the statisticians in Washington, D.C., economists like to express “index numbers” multiplied by 100. To use any of these numbers for actual calculations, you must first divide by 100. The Consumer Price Index, the GDP deflator, and a host of other economic statistics fall into this category. Seasonal adjustment factors are another example.

Thus, for the week of January 2, 2012 (ending on January 7), NSA = 646,219, SF = 165.9, and SA = (646,219/(165.9/100)) = 389,523.21. The Department of Labor knows how much noise there is in this data, so they round the seasonally adjusted figure to the nearest thousand, giving the reported figure of 390,000.

Peter CGood eye. It looks like the BLS has long understood the importance of moving holidays in calculating seasonal adjustment factors (see: http://www.bls.gov/cpi/cpisahoma.htm). And since the adjustment factors are weekly, the 2011 1.44 and 2012 1.66 figures already capture the timing vis à vis new years of the days in the week in question. But this would have been totally opaque without your day by day analysis. You should ask BLS if they need a spokesman!

adminThanks for the kind words. But I’m holding out for the position of Celebrity Spokesmodel. I’m just a number-cruncher who believes in data.