Saturday, February 13, 2010

"A Simple Way to Data Mine"

James Altucher has an article in the WSJ titled 'A Simple Way to Beat Hedge Funds' in which he discusses the 4x2 system:

I’ve traded this system for years, and even though it has no fancy chaos theory, it works. Basically, if a stock falls in price for four days in a row (for instance, after a bad earnings report) chances are the selling is done. Within four days, everyone’s seen the news, everyone’s had a chance to sell and most of the sellers have probably already sold.
He reports the following cumulative performance from 1992-2009.

I had no problem with the article about a model until this statement (bold mine):
But a simple system like this: wait for a stock to go down four days in a row, then buy until it goes up two days in a row, has never had a down year, and has beaten nearly every hedge fund.
Has "never" had a down year is a BOLD statement. And quite possibly true... if a 17 year period is "never".

While I won't question the above data (although it is almost impossible for the average investor [me] to verify), the Nasdaq has been around since 1971 and one can easily test the broader index with the 4x2 strategy over that entire period (I figure this would at least show the broad trend, but I would love to see the actual results).

My results?

Buying at the close at every period in which the Nasdaq was down 4 days in a row and selling when it was up 2 days in a row, the performance from 1992-2009 was exceptional (close to what is detailed above) with an average return for each trade over that period of more than 1% (though there was a massive draw down in 2001 [with a period in August - September where the Nasdaq did not have a 2 day win streak over a 31 day trading window] that I am not sure how individual security selection would miss).

But from 1971-1991?

Not so much.... returns per trade were less than -1% (and there were more trades in this period, thus overall Nasdaq since inception 4x2 performance is negative).

My take?

At the end of the day he is trying to do exactly what the hedge fund managers he insults in his article are trying to do:
to impress investors or potential investors
My take? Just like the data mined equity signal I created 'The Pub Power Equity Signal', if it sounds too good to be true... it probably is.

Source: Yahoo


  1. Great post! Too often people are concerned with how much a strategy makes as opposed to asking the question: how did that strategy make its money? The trip traveled is just as important as the end result.

    See this discussion on a holy grail like system and how I qualify the results -- it pays to be realistic.

  2. I appreciate your perspective, but it seems you are doing worse data mining than Altucher. The behavior of an index, especially a broad one like the nasdaq may not correlate well with trading individual stocks within a much more limited index.

    It would be interesting to see an apples to apples comparison on equities in the Dow or Nasdaq 100.

  3. agreed that a full analysis of the 4x2 on individual securities is best, but not sure how showing a full set of a history on a broad index is worse than declaring the strategy has NEVER had a down year when showing 17 years of a 38 year history.

  4. Which software are you using? I did on S&P 500 between 1981 and 1992 and got very good results, using Amibroker. Also, I agree -- looking at just the index is a terrible way to data mine.

  5. i used data from the nasdaq, not the s&p 500 using "uber-complex" microsoft excel software (with public data).

    send me your data and i'll publish the results, but please send me data going back further than 1981:

  6. Good stuff. A couple of comments:
    A) I do agree that the behavior of individual stocks is not exactly the same as the underlying index. To really compare apples with apples you should use stocks and not the index.
    B) whats the big difference post-1991 versus pre-1991. The popularity of mutual funds and 401k plans. The markets are fundamentally different right now than they were in the 1970s.
    -James Altucher

  7. So a technical difference in that there are more retail investors driving returns.

    I think that makes sense, but one can definitely make the case that retail becomes less important going forward [with the rise of passive investing (i.e. ETFs) and baby boomers hitting retirement].

  8. Excellent post! I saw that same article, and I knew immediately that if the data was extrapolated backwards in time, we would see that the strategy falls apart. And you just proved it. Good work!

  9. you're testing based on an index... totally different than what the guy describes...

  10. "the guy" (Altucher) posted in the comments section defending why 1992-2009 is different.

    if 1971-1991 showed promising results, my guess is he wouldn't have stated things were different.