Adrian Reid explains why re-optimizing trading systems on only the most recent year(s) of data is dangerous. He argues that recent data often represent a single market state (e.g., a bull market) and will produce parameters that fail when the regime changes. Reid recommends including a broad historical span (at least through 2005–2008, the COVID period, and ideally the dot-com era) to capture bull, bear, sideways, volatile, and quiet markets, and emphasizes seeking stable, consistently upward equity growth when choosing parameters. He states he rarely re-optimizes except when adding new ideas or seeing clear shifts.
Chapters
One of the most common questions I get about optimization is should I optimize my system on the most recent data, the most recent couple of years of history? Because surely they are the most relevant, right? In fact, this could be a huge, huge mistake which will kill your trading results. Stick around, watch this video, I'll explain exactly why. But before we go any further, make sure you hit the subscribe button and give this video a thumbs up so that you get notified of all of my future videos. Because I've got new trading content coming out all the time and I don't want you to miss a thing. Basically the question is, should I re optimize my system every year or so? Because the most, surely the most recent market conditions are the most relevant. And while it's kind of true that things can evolve over time, in the current or the most recent market conditions are probably somewhat more relevant, the challenge is the recent market conditions are usually only one market state. And so if we have a system and we optimize it for the last year or two years of behavior and then trade forward with that, if the last year or two was a bull market, then the new parameters will probably work really well in a bull market. But as soon as the market turns and goes down or goes sideways or goes volatile, you may end up with some really nasty results. The reason we test on more data and don't optimize that often is that you want to make sure that in your optimization you've got bull markets, bear markets, sideways markets, volatile markets, quiet markets. And to get all of that, believe it or not, you actually need a fair bit of history.
Yeah. So you've got to go back. You need to include at least before at least 2005 to 2008 for that really strong bull run followed by the bear market and the recovery. And then you need the COVID bit and the subsequent recovery that gives you a nice kind of balance of markets. If you can go even further back and include the dot com bubble, the dot com bubble and then the burst, and then the subsequent recovery up to 2008, even better. But when you're doing the optimization, you want to make sure that when, when you're choosing your parameters, you, you don't choose parameters that performed really well in the beginning part of the backtest, but really badly in the later half. That's why you look for the system stability, that the edge has to be constant from bottom left to top right. When you look at the equity curve, that will tell you if the current, if the system performs well in the current marketing conditions or only perform well way back when.
So I don't reoptimize my systems all that often.
If I see something shift or if I have some ideas, then I'll chuck the ideas into the system and then I'll reoptimize. So my systems evolve, but mostly because I have new ideas about how to play with them. Not so much because I think, oh, it's time to do my RE optimization.
