Been tweaking my strategy parameters for weeks and getting amazing backtest results but I’m starting to worry I’m just fitting it to historical data.
Tested it on different timeframes and it performs worse. Starting to think I’ve been chasing perfect backtests instead of building something that actually works in live markets.
Stop tweaking when your strategy works consistently across three different market periods without changing parameters. If you’re adjusting settings for each time period just to get decent results, you’re overfitting. Good strategies handle different market conditions with the same core settings. Test it on fresh data from a completely different year.
If it only works on one timeframe, you probably overfit it. Try forward testing with small position sizes first.
Classic trap - been there tons of times.
I stick to 3 parameters max now. More than that and you’re just memorizing data, not finding real patterns.
Walk forward analysis saved me. Test on 6 months, then run it live for the next month. Keep rolling forward. If it breaks every few months, it’s overfit.
Another red flag: if your equity curve looks too smooth in backtests, you’ve probably tweaked it to death. Real trading has drawdowns and messy patches.
My current strategy looks boring in backtests but it’s been solid for 2 years live. Boring and robust beats flashy and broken every time.
Stick to a few key principles. Random entries can perform better without overfitting. Focus on risk management instead.
Demo trade it for two months without changes.
Take a break from optimization for a month. If you’ve got more than five adjustable parameters, you’re likely curve fitting. Good strategies don’t rely on exact numbers; they work within rough parameter ranges. Test on fresh data not used in optimization. Divide your historical data into three parts: optimize on the first, validate on the second, and test on the last. Robust strategies might look average in backtests as they avoid overfitting to specific periods.