As we round the corner into the Thanksgiving holiday, I thought I would share another piece of my old writing. I recently found this paper as I was searching through old documents; it was my first attempt to really crystallize my pattern testing and research into an academic format. Reading it today, I see it as a bit clumsy in some ways (which is what we’d hope to see in our old work), but basically on-target in many aspects.
The conclusions are presented in academic format: typically understated. If you’re used to reading trading books and not academic papers, it’s easy to miss what’s really being said here. There is plenty I would do differently, and plenty I have done differently since then, but this methodology and approach have shaped much of the past 12 years (or so) of my approach to markets and trading.
What’s good about it:
- The evidence-based approach to pattern recognition
- “No exit” and no complete system tests–only testing the atomic elements of market behavior
- Testing against a baseline drawn from the sample itself
- (If only I knew how much hate my casual dismissal of moving average crossovers would get me from the math-illiterate or math-adverse casual trading community lol. (And I hadn’t figured out how to investigate Fibonacci at that time!))
What’s not so good:
- Obviously, full enumeration of the baseline is better than random entries, and this can be accomplished just by taking average returns. Such silliness, but when you give kids computers and random number generators, silliness often follows…
- Distributions were very wrong for the AR models. We should have bootstrapped actual market data. (We knew much of this, but had to cast it in academic language.)
- Samples should obviously cover futures and currencies, and, ideally, different timeframes.
Anyway, I thought you might enjoy some light, pre-holiday reading, so here’s the paper. (Parts of it were redacted for publication and archive. Since then, I’ve shared these patterns in many formats so I won’t specifically speak to the redacted sections of the paper.)
