Kitchen sink econometrics: advance or d(evolution)?

September 4th, 2014 at 2:44 pm

Over at PostEverything. If you can find another article in a major newspaper about the current state of macro-econometrics, I’d like to see it!

Update: Curse you, Mark Thoma!

Print Friendly, PDF & Email

2 comments in reply to "Kitchen sink econometrics: advance or d(evolution)?"

  1. Kevin Rica says:

    The reason that markets acted “irrationally” in the last decade was that the huge influx of foreign money drove interest rates (real, nominal, risk-adjusted) so low that is a desperate quest for some sort of nominal yield investors ignored hart-to-measure risk and did desperate things. When you get desperate, you do dumb things. If you wanted much more that a zero nominal rate, you had to take risks. You had to pull the goalie in the second period.

    When economists get desperate, (see FT article on An astonishing record – of complete failure) they turn to data mining. This is old-fashioned data mining. As the old song goes, “everything old is new again.”

  2. mitakeet says:

    I have done a lot of reading and a little research on the use of evolutionary algorithms/programs to look for correlations in noisy data. In any finite set of data, whether it is ‘real’ random data or only has the appearance, it is quite trivial to select out data and produce highly correlated trends. Such is the nature of statistics. If the trends provide predictive value for long periods of time then it is likely that the underlying data wasn’t so random after all, but given enough trends detected in enough underlying random data, a few of them will appear to provide prediction for some period of time. When I see things like “randomly draws six variables from a basket of 114, runs a forecast and then repeats that procedure 100,000 times” I have to think that they are doing nothing more than playing statistical tricks to provide meaningless correlations with no predictive value. However, I believe it is quite easy for ‘quants’ to produce a mathematical smoke screen so thick that even quants specializing in other areas are unable to refute, so investors happily pour money into those producing the meaningless correlations.

    Of course, if the underlying data is indeed deterministic, then one can expect that running such massive number of combinations might yield something with predictive value, but chaos is only predictable if you know the starting conditions to a high degree of accuracy and your model contains every element that is capable of influencing events, something that, in my experience and education, is totally impractical for something as large as the global economy.