Backtesting results vs live trading - huge difference why?

algotrader

User: BacktestingDoubt_New
Subject: Backtesting results vs live trading - huge difference why?
I backtested my algo strategy on historical data and got 65% win rate with good returns. But in live trading last 2 months I'm barely breaking even. What am I doing wrong? Is backtesting useless or am I missing something?

algotrader

Reply by: QuantTrader_Experienced
Classic overfitting problem. Your strategy is optimized for historical data but doesnt generalize to new data. Common mistakes: 1) Not accounting for transaction costs and slippage 2) Using future data in backtest (look-ahead bias) 3) Over-optimizing parameters 4) Not testing on out-of-sample data 5) Ignoring market regime changes. Split your data - use 70% for training, 30% for validation. If strategy works on validation set then try paper trading before going live.

algotrader

Reply by: SystematicTrader_Pro
Also backtest period matters. If you backtested only bull market period 2020-2021, strategy wont work in sideways or bear market. Test your algo across different market conditions - trending up, trending down, sideways, high volatility, low volatility. A robust strategy should work decently in all conditions even if not optimal. And yes always include realistic slippage and transaction costs in backtest.

algotrader

Reply by: RegimeAware_Trader
One more thing - market microstructure has changed after new SEBI regulations on F&O in November 2024. Weekly expiry restrictions, increased lot sizes, higher margins etc. If your backtest data is from before these changes, results wont match current reality. Need to be aware of regulatory changes and their impact on strategy performance.