Adrian Reid argues that pursuing a perfectly optimized trading system often leads to overfitting and poor real-time performance. Instead, traders should prefer 'good enough' systems that add diversification to a portfolio—systems that perform differently (uncorrelated) from existing strategies are valuable, while highly correlated systems should be replaced or discarded.
Good enough is way better than perfect.
And the more perfect you try and make it, the worse the performance is going to be in real time. Generally, because a perfect backtest is only the result of curve fitting over optimization. Too many filters, too much complexity, and it falls apart.
So when I'm, when I'm working with my students and when I'm thinking about my own trading, what I want to do is I want to look at the strategy in the context of my portfolio. So If I've got 10 different systems, I want to be thinking about, okay, does this new system in its current form add something to the portfolio? Does it add some diversification? And if it makes money where my other strategies don't, and it has a drawdown when my other strategies are making money, then that's a really good addition. And it's probably good enough in its current form to just put into the portfolio. But if it's highly correlated to everything else in my portfolio, then it's not good enough. It's got to either be better than what's in my portfolio to replace it, or I throw it away because it's not going to add any diversity to the, to the portfolio. So I really want a good system, not an exceptional system, because an exceptional system is probably just curve fit, a good system that adds to the portfolio in a form of diversification.
