"Can you data mine a starting point for a strategy?"
By Dave Mabe
Here's a question from Sridhar R. (name used with permission, lightly edited for clarity.)
Sridhar R:
I recently came across Jaffray Woodriff's chapter in Hedge Fund Market Wizards and found it fascinating — particularly what he calls "secondary variables." It seems like this maps pretty closely to what you describe as column variables in RealTest.
What struck me most is that his approach is essentially the inverse of what you've been recommending. Rather than starting with a solid base strategy and then running the MabeKit Strategy Cruncher to isolate the best column variables, he appears to data mine across millions of variable combinations first to surface a workable strategy, and then applies column data. I'm curious what you think of his methodology and whether you see meaningful parallels (or divergences) with your own framework.
Dave:
I remember reading this chapter a few years ago and finding it to be a useful thought exercise.
A big part of successful trading is the ability to scale up a strategy to trade it with meaningful size.
To do that, you need a path to confidence.
Another way of saying that is you need to believe in your strategy - deeply.
This is why you can't simply join a trading room and trade someone else's strategy by looking over their shoulder and be successful in the long run.
Because ultimately, you can't truly believe in someone else's strategy because they say it works.
An important part of believing in a system is understanding it fully.
If a machine or some process randomly creates a list of signals without a fundamental hypothesis, it's hard to imagine being able to become confident in it.
Maybe you can have modest confidence, but I think there's a ceiling to it.
The underlying thesis needs to be sound, otherwise optimizing it doesn't really make any sense.
If you don't understand the strategy's signal, then it's impossible to add a rule that tells a coherent story.
And the fundamental way to avoid curve fitting is lost.
For the record, I can imagine this process working, but I think the path is much harder and would require far more computing resources than simply starting with a workable hypothesis.
Very interesting question, Sridhar - thanks for sharing with the group!
-Dave
P.S. Are you in a drawdown and ready to quit? Join the other traders who reversed their drawdowns and are printing money with MabeKit. Get Instant Access