Seasonal Adjustment (Unemployment)

[Just thinking]

Can someone please explain the reasoning behind using numbers for Jobless Claims that are "Seasonally Adjusted" ? ... and just what exactly that means?

For example, the most recent week of seasonally adjusted (SA) jobless claims showed a decrease in claims, but the non-seasonally adjusted (NSA) numbers showed an increase in claims --- not to mention, the NSA numbers were much higher, and have been for a good long time, week after week after week.

Why can't we just use the actual number of claims?

 

[MattC]

Can someone please explain the reasoning behind using numbers for Jobless Claims that are "Seasonally Adjusted" ? ... and just what exactly that means?

Identification of situational factors that have strong relevance to the overall value - in this case, the expected holiday rush necessitated the hiring of more employees.

One of the problems with using gross numbers is that many potentially sharp changes often have very mundane raison d'etres, in this particular case any decrease in the raw amount of jobless claims would probably be at least partially based upon retailers hiring more employees to deal with the holiday "season." Once the season is over, overall revenue will decrease and so some of these employees will be let go as their services become unnecessary or redundant.

Ergo, in this case (and some others: fruit-picking season in the south and southwest, snow-removal season in those places with snow, etc) what "season" it is does impact the overall number of jobs with readily ascertainable and temporary factors. Reporting jobless claims as "seasonally adjusted" is an attempt to reflect the transitory nature of the raw total.

Why can't we just use the actual number of claims?

The reality of a situation is often far too complex to be accurately represented by a single number without additional clarification.

~ Matt

 

[Just thinking]

The reality of a situation is often far too complex to be accurately represented by a single number without additional clarification.

~ Matt

With all due respect ... that sounds like a cop-out of an answer.

Besides, as best I can see, the number of actual claims has been steadily much greater than the SA numbers ... meaning that no amount of averaging (skewing) can bring those numbers below a weekly value of 500,000.

 

[MattC]

With all due respect ... that sounds like a cop-out of an answer.

Besides, as best I can see, the number of actual claims has been steadily much greater than the SA numbers ... meaning that no amount of averaging (skewing) can bring those numbers below a weekly value of 500,000.

In response to this, I would generally ask you what precisely you want out of the jobless figure. You seem to be seeking total (e.g. overall) jobless claims in an overall form, which the web site supplies. Or are you instead questioning the reason behind "seasonal adjustments" in general? I'm confused and would ask for some enlightenment before continuing.

~ Matt

 

[Just thinking]

In response to this, I would generally ask you what precisely you want out of the jobless figure. You seem to be seeking total (e.g. overall) jobless claims in an overall form, which the web site supplies. Or are you instead questioning the reason behind "seasonal adjustments" in general? I'm confused and would ask for some enlightenment before continuing.

~ Matt

Sure.

First, why do it? I'm sure everyone here as well as economists can see trends in data over enough time even given actual numbers. Four-week moving averages are fine too.

Second, I'd like to know the actual math behind the manipulations. How are the numbers changed from NSA to SA? And how can it be that given a year of NSA numbers well over 500,000 each week we get numbers for SA below that figure?

Third, it (to me) seems like too much fudging of data. Just how many times have you heard the phrases "unexpected drop" or "unexpected gain" in jobless claims? ... even given the seasonal adjustments? (Quite often, if you ask me.) Do they really know whether this data becomes more meaningful? How much is just approximated guessing?

 

[Puppycow]

Seasonal adjustments let you see the secular (Long-term) trend by eliminating predictable seasonal factors.

 

[Just thinking]

So, it is guessing.

 

[Puppycow]

What is being guessed at?

Are you saying that seasonal factors are just guesswork?
They are based on past data averaged over many years.
Christmas happens every year. Cold weather happens every year, etc.

 

[Just thinking]

Of course those things happen.

But conditions we are now experiencing (record high layoffs, housing market foreclosures, etc.) are having much greater impacts, and IMHO make the actual numbers more relevant. It's almost as if one can adjust the average temperature of a given month (having non-average temperatures) by looking at previous years' data and chaining the actual recorded temperature accordingly.

And if trends are what you're after, what's wrong with 4-week moving averages? Or looking at last year's actual numbers month-to-month?

 

[Puppycow]

Besides, as best I can see, the number of actual claims has been steadily much greater than the SA numbers ... meaning that no amount of averaging (skewing) can bring those numbers below a weekly value of 500,000.

Since when?

Have you looked at them for a full year? In the summer months the adjusted number of claims will be lower than the SA numbers (farm work and other warm-weather work).

 

[Puppycow]

Here is one from the summer

In the week ending May 30, the advance figure for seasonally adjusted initial claims was 621,000

The advance number of actual initial claims under state programs, unadjusted, totaled 496,822 in the week ending May 30

 

[Just thinking]

Since when?

Have you looked at them for a full year? In the summer months the non-adjusted number of claims will be lower than the SA numbers (farm work and other warm-weather work).

I think that's what you meant.

Yes, that's a pretty consistent trend. But why do the adjusting in the first place? I mean, if it's expected to rise and fall, why not look at the overall actual numbers and compare those ... week to week, month to month, 4-week averages, etc.?

 

[MattC]

I think that's what you meant.

Yes, that's a pretty consistent trend. But why do the adjusting in the first place? I mean, if it's expected to rise and fall, why not look at the overall actual numbers and compare those ... week to week, month to month, 4-week averages, etc.?

Ultimately because the statistic to which you refer, particularly when done in comparatively small intervals (week-to-week, month-to-month) is greatly affected by very predictable trends. To supply an example, many university bookstores hire additional staff during the book buyback periods - these staff are then let go shortly thereafter. The term of this is seldom more than three weeks, so a set of bookend statistics would show a great discrepancy - but because of a very specific (and foreseeable) event.

The important point about seasonal trends, though, isn't the actual employment so much as it is the inevitable termination of that same employment - "seasonal" implies a distinct end point, which may be the end of the university semester (as in my example above), the end of the growing season (in the case of temporary migrant workers), or the end of the holiday shopping rush. These do pose boosts in overall employment (as more people find jobs), but they're temporary boosts and therefore deserve their own special mention, something I'm grateful the web site provides.

~ Matt

 

[Startz]

An introduction to the details of seasonal adjustment is given at http://www.bls.gov/cps/eetech_seas.pdf.

 

[Puppycow]

I think that's what you meant.

Yeah, my bad.

 

[drkitten]

Yes, that's a pretty consistent trend. But why do the adjusting in the first place? I mean, if it's expected to rise and fall, why not look at the overall actual numbers and compare those ... week to week, month to month, 4-week averages, etc.?

Because the overall actual numbers don't tell us as much about the underlying economic situation. If you expect the number of jobs to rise in December, the important question isn't "did the number of jobs actually rise" but "did the number of jobs rise as much as, less than, or more than expected?" If you expect 30,000 department store Santas to be hired, and only 5,000 are hired, then that's a very bad sign even though the actual numbers went up.

And that's why you adjust the numbers, essentially by subtracting off the 30,000 expected Santa's so that if everything happens exactly as the long-term seasonal history suggests, the net result is zero.

 

[Francesca R]

Can someone please explain the reasoning behind using numbers for Jobless Claims that are "Seasonally Adjusted" ? [ . . . ] Why can't we just use the actual number of claims?

Does this explain it well enough?

NSA:

SA:

 

[Just thinking]

Yes ... the lower graph's overall shape comes very close to what one can draw going from peak-to-peak on the upper graph, which pretty much supports my initial claims.

 

[drkitten]

Yes ... the lower graph's overall shape comes very close to what one can draw going from peak-to-peak on the upper graph, which pretty much supports my initial claims.

... if you don't mind losing 11/12 of your initial data. Or perhaps 364/365 of it.

You can certainly compare the raw, unadjusted numbers from one December to the next directly, and figure out if unemployment is worse now than it was a year ago. You can also compare the numbers from May to May and do the same thing.

But how are you supposed to figure out from the raw data whether the economy is better right now than it was last May?

 

[lomiller]

Yes ... the lower graph's overall shape comes very close to what one can draw going from peak-to-peak on the upper graph, which pretty much supports my initial claims.

What you are proposing is a form of seasonal adjustment. Not the best way to do it because you lose so much data but seasonal adjustment nonetheless.