Curve Fitting Is Not The Monster Under Your Bed
StatsEdgeTrading
Curve fitting has been turned into this trading boogeyman. Say “optimization” online and ten people jump in to yell “curve fitted!” and log off feeling smart.
Here’s the thing: if you run backtests at all, your strategy has some curve fitting baked in. That’s not a failure. That’s statistics. The goal is not zero curve fitting. The goal is to keep it on a short leash. Quant beats vibes.
Let’s reframe this with something practical. If I wanted to create a beautifully curve fitted disaster system, here’s exactly how I’d do it:
Use a tiny sample of trades. A few dozen, maybe a couple hundred if I’m feeling “responsible.”
Ignore the equity curve and only stare at one metric. Profit factor five? Ship it.
Blindly follow the data. Add filters because the number gets better, not because the rule tells a coherent story.
Stack on rules. Ten, twenty, more. If there’s a filter, I’m using it.
Go live and never reconcile. No tracking, no comparison to base system, no review of trades I filtered out.
Do the exact opposite of that and it becomes surprisingly hard to truly blow yourself up from curve fitting alone. The horror stories you hear where “one system ended my career” are almost never a curve fitting problem. They’re a position sizing problem.
Here’s a saner process:
Start from something with raw edge. Your unoptimized idea should already make money, even if it’s only a profit factor of one point two. Optimization should be polish, not magic.
Get a real sample size. On stocks, I want hundreds of trades at minimum. If a day trading system has less than five hundred trades over multiple years, my skepticism dial turns up.
Make every rule tell a story. If you can’t explain to your future self, in a drawdown, why a filter should work going forward, it probably doesn’t belong.
Watch the live data. Track both the trades you take and the trades your filters remove. If the rejects start outperforming what you kept, that’s your cue to re-evaluate.
Action plan:
Strip your favorite strategy back to the bare rules and re-test.
Add rules one by one, only if you can explain the logic out loud.
Forward test small size, log every trade, and compare live vs backtest monthly.
If you want systems where the curve fitting is controlled, documented, and battle-tested over decades of data, come hang out at StatsEdgeTrading and check out StatsEdge Pro at www.statsedgetrading.com.