How do you avoid overfitting your backtests?
MikeAdmin
A backtest that looks amazing is often the most dangerous kind. Here's a quick checklist I use before trusting results:
- Out-of-sample data: tune on one period, test on a period the strategy has never seen
- Realistic costs: commissions, spread and slippage included
- Enough trades: a handful of trades proves very little
- Parameter stability: does it still work if you nudge the settings a little?
- Different market conditions: trending, ranging and high-volatility periods
- Paper trade before live: the market is the final backtest
What would you add? And what tools do you use for backtesting: TradingView, Python (Backtrader, vectorbt…), something else?