Jeremy Grantham on Zulus, Changing Quants, etc.

GMO’s Jeremy Grantham has out a typically lucid new market commentary, and it contains some quotable musings on quantitative investors. Here’s Jeremy on next-generation quants, Zulu warriors, and outlier events:

The good old days of the domination of the first generation quant models, where you simply show up with three concepts – value, momentum, and discipline – are over. But, even more critically and, perhaps like career and business risk, out at the limits of arbitrage, is this need for judgmental overrides on rare macro events. Quants like to show off their discipline by marching off the cliff in rows (it is said, I hope apocryphally, that Shaka, the great Zulu Chief, marched an impi, or regiment, off a cliff to impress
European observers and I hope it did). Well, in real life it would be nice to stop at the edge and say “I don’t like the look of this, perhaps my model missed something.” The extremely diffi cult objective is to maintain the advantages of quant discipline 95% or so of the time and hand over to a human being when you reach the edge of the cliff.

Nicely put.


  1. Josh Stern says:

    It’s a good read. This is a better link for it: Grantham’s article

  2. Mark - NYC says:

    Hmmm…when I was at a big-global bank leadership conference thingy, a senior banker gave the exact same analogy (except he was driving a car as fast as possible with imperfect vision). The leaders were able to stop short of the cliff and turn around, that was his definition for leadership. Maybe bank to hedge fund correlation is greater than thought. Going to read the whole article now.

  3. Like most, he refers to the HF trading world as if it encompassed all that was quantitative, instead of it being a small corner of the quantitative method. Joel Greenblatt is a quant. Oh, bother.
    Accepting his pigeonholing of HF trading as the only thing “quant” and moving on …
    The thing that the HF guys needed was simple. Measure their results, their last 100 trades or so, against the payout distribution that was expected by their models. When it was statistically different – stop trading.
    No “handing off to the humans” necessary.
    Like most problems associated with computers, the error was with the humans (who designed the models) in the first place, and never with the hardware or software itself.

  4. fike silk says:

    Oh, yeah… Put another way, if you can see the darn cliffs, why are you using a computer model in the first place?