Indicator Traps and Curve Fitting
The danger of a strategy that looks perfect on historical data.
Researched and written by Laurent The Over-Optimization Trap
You've just backtested your strategy on two years of EUR/USD data. It shows a 62% win rate and 2.1 profit factor. Excellent. Now you want to improve it further, so you start tweaking: "What if I change RSI from 14 to 11? Better. What if I use 48 EMA instead of 50? Better still." You keep adjusting until you have a nearly perfect backtest.
This is called curve fitting (or over-optimization), and it destroys strategies. You've created a system so specifically tuned to that exact stretch of EUR/USD history that it will likely fail everywhere else. You've memorized the past rather than found a real edge.
Why It Happens
If you run enough tests with enough parameter variations, you will eventually find a combination that produces good results on your specific historical dataset. This is pure statistics, not evidence of an edge. The strategy is exploiting the specific quirks of that particular data, quirks that won't repeat in the future in the same way.
The Robustness Test
A real edge works across similar conditions, not just the exact data it was designed on. Before trading any strategy live, test it on:
- Different time periods: Does performance in your test window hold in earlier periods too?
- Different but similar pairs: If it works on EUR/USD, does a similar approach work on GBP/USD?
- Different market regimes: Does it work in both trending years and ranging years?
If the strategy collapses outside its original test window, it was curve-fitted. If it holds up with some degradation across different conditions, it has a real edge.
Confirmation Bias in Live Trading
Even after a strategy is validated, confirmation bias will sabotage you. Confirmation bias is the tendency to notice evidence that confirms what you already believe and ignore evidence that contradicts it.
In trading: you decide EUR/USD is going up. Suddenly, every piece of chart evidence you look at seems to confirm it. The 50 EMA "looks like support." The RSI "looks like it's turning up." The pattern "looks like a double bottom." You see what you want to see.
The fix: before looking for confirming evidence, actively look for evidence against your thesis. If EUR/USD looks like it's breaking down, what would make you wrong? If you can't find a good answer, reconsider the trade.
The Checklist as an Antidote
The best protection against both curve fitting and confirmation bias is a rigid checklist. Before every trade:
- Is the trend filter met? (Yes/No, not "almost yes")
- Is the setup condition met? (Yes/No)
- Is the entry signal present? (Yes/No)
- Is the risk-reward at least my minimum? (Yes/No)
If any answer is "No," there is no trade. The checklist removes ambiguity and makes it harder to rationalize exceptions.
Key Takeaways
- • Curve fitting: tweaking indicator settings until they "perfectly" fit past data. This destroys forward performance.
- • A strategy tuned to one year's data often fails on the next year's because markets change.
- • Robustness test: does the strategy work on different pairs, different years, different market conditions?
- • Confirmation bias causes traders to only see setups that confirm what they already want to do.