Forecasting and Tomorrow’s Jobs Report

July 5th, 2012 at 3:56 pm

I had a chat with a friend the other day—a prominent academic economist whose name I won’t disclose so he doesn’t get shunned in the faculty room—wherein we bemoaned the state of a) micro-theory (predicts implausible elasticities that never show up in the data; marginal product theory—a core premise—looking ever more suspect*) and b) macro-theory (a terrible muddle these days, as Paul K stresses).

But we agreed that econometrics still rules.  Sure, there are those who practice eCONomeTRICKS, but “we regard them with scorn” (extra points for those who can source that quote without Google—even more points for those who can identify why it fits in an econometrics post).

I used to have decent econometrics—statistical analysis of economic data—chops, especially for a former musician/social worker, but alas, no more.  I can still reliably run reduced form regressions and the Kalman Filter using the structural (or “state-space”) model I associate with Andrew Harvey (see previous link).  But I simply haven’t kept up with the cutting edge stuff, though luckily, I know folks who have.

All of which brings me to the fool’s errand of forecasting employment growth for tomorrow’s jobs report.  The consensus is for about 100K.  I run a couple of models.  At this point in the month, I run a regression of the log changes in payrolls on the lagged quarterly payroll growth, the monthly average of 4-wk UI claims, and the ADP (again, all in log changes) and forecast one month ahead (using the actual UI and ADP data for June).

I also try to tease out the longer term trend using the Kalman filter on the NSA data—this is a very good way to get at the underlying recent trend, which right now is running at around 90K, which is actually close to what I get with the standard time series regression noted above.  So that’s about what I expect tomorrow, though given the confidence interval of 100K around these data along with the monthly revisions, the firm birth/death modeling—well, I don’t know anyone who has a great track record on this one.

However, that’s less a critique of econometrics than a warning about realistic expectations when forecasting high-frequency data.

*The great Joe Stiglitz gave a talk recently at the LSE on his new book on inequality (I also interviewed Joe the other day).  Anyway, a bit into the interview, he tells the LSE students, and I’m paraphrasing, “You know, that marginal product theory you’re learning around wage setting—it’s not true…you still have to learn it, but it doesn’t really work.”

Print Friendly

6 comments in reply to "Forecasting and Tomorrow’s Jobs Report"

  1. Mark Thoma says:

    I think it’s important to remember that most econometricians are involved with testing theories — finding out how the world works — rather than forecasting. Most econometric courses don’t teach much about forecasting, it is all about hypothesis testing (e.g. in forecasting, it is often worth it to trade tiny bias for greater precisions, but in hypothesis testing that is avoided, particularly for point estimates).

    An econometric model can, for example, determine whether a particular correlation/prediction is in past data, but still be lousy at predicting out of sample.


    • Jared Bernstein says:

      Makes sense, and you would know, being a tony academic yourself. I do, however, recall a lot of forecasting back in econometrics in grad school, but that was an earlier era…in fact, I remember a really fat book called “Forecasting” by an author with a Greek name–but I may be dreaming all of this up. That where we learned ARIMA, seasonal adjustments, smoothing, and so many other fun things…ahh, to be young again.


  2. A Cassel says:

    Not sure how Tom Lehrer fits your context, but the quote is from “The Folk Song Army”. And boy, are you dating yourself –the album ‘That Was the Year That Was’ was new in 1964, if I recall rightly:

    There are innocuous folksongs, yeah,
    but we regard’em with scorn.
    The folks who sing’em have no social conscience. Why
    they don’t even care if Jimmy cracked corn…


    • Jared Bernstein says:

      That’s right! If you’re still at work, go home early–tell your boss I said it was OK.


  3. Marc says:

    One should not put econometrics on trial, especially considering that there are essentially two distinct “fields” in econometrics, i.e., microeconometrics and macroeconometrics (more commonly known as time series analysis).

    As regards the latter, Mark Thoma is right that forecasting is only a small part of it.

    As regards the former, most of the time we couldn’t care less about forecasting, since we are for the most part interested in disentangling causal relationships from correlations, a task that could not be further removed from forecasting.


  4. Misaki says:

    >marginal product theory

    Remember this chart? http://s122.photobucket.com/albums/o245/Taemojitsu/?action=view&current=theiPodeffect.jpg

    A marginal product which is not sold leads to no increase in revenues. A marginal product which can be sold only be decreasing revenue for other products, through lower prices, can actually lead to a decrease in revenues meaning it would have negative marginal revenue.


Leave a Reply

Your email address will not be published.

Current ye@r *