Hazards in Interpreting Seasonals

Professor Casey Mulligan has an interesting post, in which he observes that while retail sales are about 15-20% higher in December than in the previous three months, retail employment is only about 4% higher in December than October, thus proving that fiscal stimulus cannot be very effective at raising employment.

From the low sales to employment ratio, Professor Mulligan concludes:

Although the holiday spending surge is clearly associated with a high level of employment, it also shows how spending is a rather indirect way of creating jobs. That holiday spending of roughly $90 billion more in December is associated with about 500,000 additional jobs for a month — that amounts to $180,000 per job per month!

Both Christmas and the fiscal-stimulus act increase demand, but the fiscal-stimulus act depresses supply, because many of its major programs — the unemployment-insurance extension, the food-stamp program expansion, the home buyer tax credit and more — are directed at people with low incomes.

Certainly, there is a correlation between sales and employment. Below are plotted the time series.


Figure 1: Retail employment, in thousands (blue, left scale), and real retail and food sales, in millions of 1982-84$ (red, right scale), both not seasonally adjusted. Sales (FRED series RSAFS) deflated by CPI-all. NBER defined recession dates, shaded gray. Source: BLS, FRED II, and NBER.

(Note that the sales figure includes auto and auto parts; sales ex.-auto constitutes about 82% of total, using seasonally adjusted data.)

Professor Mulligan’s calculation is essentially a one observation regression of the change nominal retail sales on change in retail employment. Estimating the same regression over the entire 1992M01-2010M11 period, only for Decembers, with no constant, yields a slope coefficient of $197277 per job (standard error = 21319). In real terms, allowing a consant, the slope coefficient is 2009 1982-84 dollars per job (standard error = 13366).

But I think this figure (as well as Professor Mulligan’s) is irrelevant. First, the employment that is relevant is the total employment associated with Christmas-goods production and distribution (in addition to retail employment). Second, the activity variable that is relevant is not sales, but US related value-added. So not:



Δ(value added)/Δ(employment)

The value of retail sales incorporates the value added from retail services, plus the value imbedded in the goods themselves. Those goods were produced over the entire year (i.e., not all Christmas ornaments are made in December). The counter-objection could be that Christmas ornaments aren’t typically made in the US. But then I know that at least some of the gifts are American made (after all, Wisconsin cheese makes a fine holiday gift! And most of the gifts I received were American made). That relates to the value added component. Thus, the relevant numerator is smaller, and the relevant denominator bigger, implying the relevant ratio is smaller than Professor Mulligan purports.

Professor Mulligan ends:

Both Christmas and the fiscal-stimulus act increase demand, but the fiscal-stimulus act depresses supply, because many of its major programs — the unemployment-insurance extension, the food-stamp program expansion, the home buyer tax credit and more — are directed at people with low incomes.

In other words, the less you work and earn, the larger your entitlement to various components of the act.

By reducing supply as it increases demand, the fiscal-stimulus act could well reduce total employment, rather than increasing it as Christmas does.

In any case, our experience with Christmas shows how large amounts of spending do not necessarily create large numbers of jobs.

There are many valid approaches to critiquing the idea of fiscal stimulus efficacy (e.g., CBO (Nov. 2010). This is not one of them. Now, this is not to deny the possibility that holidays related to the winter months might be important for aggregate output. But then one typically uses metrics related to the entire economy, not just the retail sector. (See [1] [2] [3])

By the way, this article highlights the hazards of over-interpreting seasonal effects. The canonical example occurred forty years ago, when Arthur Laffer interpreted the seasonal correlation of GDP and money as a causal relationship [4] (critique here). (This episode is not written down in any textbook as far as I know, but is passed down by word-of-mouth as a cautionary tale.)

12 thoughts on “Hazards in Interpreting Seasonals

  1. jonathan

    I had trouble following this post because I had trouble understanding Mulligan’s point. I read his post and still had trouble understanding him. Why? Is it because he’s not making any sense at all?
    As I read his words, he’s essentially saying that Christmas spending doesn’t – or rather, didn’t this year – translate into a vast number of additional jobs. He then transfers that point to government directed stimulus. You point out one error in this, that goods are produced and that occurs year round, but others are obvious. Such as:
    – if consumer spending doesn’t create jobs, what does? Is it magic? Assuming it’s some form of demand for goods, that is spending – unless you believe in investment and production without consumption. If you going to make the argument he makes, then you need to at least try to say how jobs are created otherwise.
    – taking his specific case, retail sales have been low. This suggests slack in the retail environment, meaning there is room to sell more without necessarily adding as many employees as you would if sales were already high and your workforce was more fully utilized. To say otherwise would be like complaining that a new factory isn’t built when demand for widgets increases because they choose to use the slack in production rather than build a new plant.
    – how exactly are retail sales a proxy for road contracts? How are they a proxy for social service payments that not only support human services employment but which also reduce demand for other social services? So for example, $x100 of retail sales may be rung up by a formerly underworked clerk while $x100 for road repair may generate jobs for a number of people.

  2. Leroy Dumonde

    Casey Mulligan, like Arthur Laffer, is a very clever quack.
    He is a stain on Chicago’s reputation. I mean at least someone like Greg Mankiw, who has his kooky moments, has actually produced some impressive work besides his whiny conservo-libertarian twaddle.
    But Mulligan, like Steven Levitt, is carrying on in the tradition of Becker. Lots of style and very little substance.

  3. dilbert dogbert

    I wonder if Mr Mulligan is near my age and his brain is turning to mush. Mr Google is my friend so I guess I should go ask him.

  4. Mark A. Sadowski

    dilbert dogbert,
    Have I crossed paths with you recently? Your moniker is very familiar and I know we visit the same sites. I seem to remember we agreed/disagreed (more likely the former) about something recently, but I just can’t put my google on it.
    By the way one thing that comes up upon googling is the following (on Arnold Kling’s site):
    “Comment removed pending confirmation of email address and for rudeness.”
    I don’t remember what I said but I’m sure he deserved it.

  5. kharris

    The point salient feature of Mulligan’s flawed analysis is not that it is flawed, but that it is flawed in support of a particular view. The view existed before the flaw. Not to get to post hoc ergo blah-blah, but that does suggest Mulligan’s desire to say bad things about fiscal stimulus led to the flaw in his analysis.
    And while we are poking holes in Mulligan’s rag of an argument, it might be worth noting that it is a bit of a stretch to say that retail sales in a period with very large seasonal distortions is a good analogy for all of government spending or even government spending legislated in ARRA. Mulligan doesn’t even bother trying to show why it is a good analogy. He just assumes it to be so because – you guessed it – it helps him argue for his pre-determined conclusion.
    I have a simple rule of thumb for reading arguments from people whose positions I already know. If the argument in question comes out very much as I would he predicted before reading it, then it wasn’t worth reading. If thinkers cannot go anywhere new, their thinking isn’t really worth the effort. Why exactly does the NYT give this troll space?

  6. mort_fin

    lady” posting about the July Case-Shiller being up, despite the tax credit expiring. Apparently, Casey doesn’t get the difference between SA and NSA, nor does he understand that it is a 3 month moving average, nor does he get that it’s based on closing prices, and closings have been extended to September, so that July is likely to have a fair number of tax credit deals in it.
    He hasn’t taken it back even though prices have been tanking in the 2nd half of the year (a fact discernible from a vast accumulation of anecdotal facts even at the time of casey’s postings). i doubt he ever will,
    I hate it when macroeconomists try to do housing finance. They are usually so phenomonally bad at it.

  7. mort_fin

    Apparently cut and paste managed to mangle my first sentence. I tried to say that mulligan had, in september, posted a “neener, neener, so’s your old lady” posting about the July Case-Shiller being up despite the expiring tax credit.
    mulligan seemed convinced that housing prices were up because we weren’t in a bubble at any point, instead of the alternate explanation that the tax credits had caused a temporary uptick.

  8. dilbert dogbert

    Mark A.
    I don’t think we hang out on too many of the same blogs. At least I don’t remember seeing your name in comments.
    I am trying to learn a bit of economics in my dotage, just for mental exercise. At first I thought it might be useful for investment purposes but was quickly disillusioned of that idea.
    I love it when the blogger mixes it up with the commenters ala menzie, CR and the late Tanta.
    As CR says best to you all.

  9. endorendil

    It would be easy to take a stick to Mulligan like everyone else, but there is some interesting information in what he’s looking at, namely that increasing the amount of work does not necessarily lead to an increase of workers. Retail around the holidays is just much more efficient (and profitable) than during the rest of the year. This is doable because people are willing to put up with bad service (long lines, often poorer selection and rude treatment at the hands of overworked temporary employees), and because it is only for a short time.
    It’s not like this is news, but of course it does translate to other sectors. More work does not lead to more workers as long as productivity increases, longer working hours or poorer service can pick up the difference. When work translates to hiring, the duration of the expected workload increase determines whether the hires are temps or not.
    That is why smaller and short-term stimuli to the economy have intrinsically different effects than bigger and longer-term ones…

Comments are closed.