ChatGPT Trading Strategy Backtesting Framework Prompt
You are a quantitative analyst who has built backtesting frameworks for hedge funds.
Category
📊 Trading
Difficulty
Advanced
Models
3
Last Updated
2026-06-28
Works with
📄 Example output
⚠️ Common Mistakes
❓ FAQ
⚙️ Fill in your variables
📋 Prompt
You are a quantitative analyst who has built backtesting frameworks for hedge funds.
Strategy: [strategy name]
Entry rules: [entry rules — specific, measurable conditions]
Exit rules: [exit rules — TP/SL/trailing/time-based]
Instrument: [instrument]
Task: Build a complete backtesting framework:
1. STRATEGY DEFINITION: Exact entry/exit conditions in code-friendly language
2. PYTHON BACKTEST CODE: Complete script using pandas + vectorised operations
3. PERFORMANCE METRICS: Total return, CAGR, Sharpe ratio, max drawdown, win rate, profit factor, avg RR
4. OPTIMISATION: Parameters to optimise + walk-forward testing approach
5. OVERFITTING SAFEGUARDS: How to distinguish real edge from curve-fitting
6. INTERPRETING RESULTS: What good vs. bad backtest metrics look like for this strategy type
Strategy: [strategy name]
Entry rules: [entry rules — specific, measurable conditions]
Exit rules: [exit rules — TP/SL/trailing/time-based]
Instrument: [instrument]
Task: Build a complete backtesting framework:
1. STRATEGY DEFINITION: Exact entry/exit conditions in code-friendly language
2. PYTHON BACKTEST CODE: Complete script using pandas + vectorised operations
3. PERFORMANCE METRICS: Total return, CAGR, Sharpe ratio, max drawdown, win rate, profit factor, avg RR
4. OPTIMISATION: Parameters to optimise + walk-forward testing approach
5. OVERFITTING SAFEGUARDS: How to distinguish real edge from curve-fitting
6. INTERPRETING RESULTS: What good vs. bad backtest metrics look like for this strategy type
STRATEGY: SMC + EMA200 Trend Filter
Entry: Price below EMA200 + H4 BOS bearish + H1 FVG forms
Exit: 1:2 RR fixed OR trailing stop below swing lows
```python
import pandas as pd
import numpy as np
def backtest_smc_ema(data: pd.DataFrame, risk_pct: float = 0.01) -> dict:
df = data.copy()
# EMA200 filter
df['ema200'] = df['close'].ewm(span=200, adjust=False).mean()
df['below_ema'] = df['close'] < df['ema200']
# Swing low detection (simplified)
window = 5
df['swing_low'] = df['low'][(df['low'] == df['low'].rolling(window*2+1, center=True).min())]
df['last_swing_low'] = df['swing_low'].ffill()
# BOS signal: close below last swing low + below EMA200
df['bos_bearish'] = df['below_ema'] & (df['close'] < df['last_swing_low'])
# FVG: 3-candle bullish imbalance (gap between candle 1 high and candle 3 low)
df['fvg_bearish'] = df['low'] > df['high'].shift(2)
# Entry signal
df['signal'] = df['bos_bearish'] & df['fvg_bearish'].rolling(5).max().astype(bool)
# Simulate trades
trades = []
for idx in df[df['signal']].index:
entry = df.loc[idx, 'close']
sl_dist = entry - (df.loc[idx, 'last_swing_low'] * 0.995)
trades.append({
'entry': entry, 'sl': entry + sl_dist, 'tp': entry - (sl_dist * 2)
})
return {'total_trades': len(trades), 'signals': df['signal'].sum()}
```
PERFORMANCE TARGETS:
✅ Profit Factor > 1.5 (every £1 risked returns £1.50+)
✅ Sharpe Ratio > 1.0
✅ Max Drawdown < 20%
✅ Win rate > 35% at 1:2 RR (expectancy positive)
❌ Win rate 70%+ with tiny RR = likely overfitted
Entry: Price below EMA200 + H4 BOS bearish + H1 FVG forms
Exit: 1:2 RR fixed OR trailing stop below swing lows
```python
import pandas as pd
import numpy as np
def backtest_smc_ema(data: pd.DataFrame, risk_pct: float = 0.01) -> dict:
df = data.copy()
# EMA200 filter
df['ema200'] = df['close'].ewm(span=200, adjust=False).mean()
df['below_ema'] = df['close'] < df['ema200']
# Swing low detection (simplified)
window = 5
df['swing_low'] = df['low'][(df['low'] == df['low'].rolling(window*2+1, center=True).min())]
df['last_swing_low'] = df['swing_low'].ffill()
# BOS signal: close below last swing low + below EMA200
df['bos_bearish'] = df['below_ema'] & (df['close'] < df['last_swing_low'])
# FVG: 3-candle bullish imbalance (gap between candle 1 high and candle 3 low)
df['fvg_bearish'] = df['low'] > df['high'].shift(2)
# Entry signal
df['signal'] = df['bos_bearish'] & df['fvg_bearish'].rolling(5).max().astype(bool)
# Simulate trades
trades = []
for idx in df[df['signal']].index:
entry = df.loc[idx, 'close']
sl_dist = entry - (df.loc[idx, 'last_swing_low'] * 0.995)
trades.append({
'entry': entry, 'sl': entry + sl_dist, 'tp': entry - (sl_dist * 2)
})
return {'total_trades': len(trades), 'signals': df['signal'].sum()}
```
PERFORMANCE TARGETS:
✅ Profit Factor > 1.5 (every £1 risked returns £1.50+)
✅ Sharpe Ratio > 1.0
✅ Max Drawdown < 20%
✅ Win rate > 35% at 1:2 RR (expectancy positive)
❌ Win rate 70%+ with tiny RR = likely overfitted
🏆
💡 Pro Tips
Best model for this prompt
DeepSeek
DeepSeek V3 / R1
Walk-forward testing is mandatory — split data 70% training / 30% out-of-sample, trust only the out-of-sample metrics
Transaction costs must be included — spreads, commission, and slippage typically reduce backtest returns by 20–50%
Monte Carlo simulation on your trade list shows probability of live results matching backtest results
More parameters = higher curve-fitting risk — every extra optimisable variable needs 50+ more trades to be meaningful
Testing on the same data you optimised on — this always produces impressive results that fail live
Ignoring overnight swap costs for multi-day strategies — can turn a winning strategy into a losing one
Lookahead bias — accidentally using future data in entry conditions
Treating backtest results as a guarantee — they're a lower bound on what you need to see before considering live trading
- What software should I use for backtesting?Python (pandas + backtrader or vectorbt): most flexible, free. TradingView Pine Script: quick visual backtesting, good for prototyping. MetaTrader 5 Strategy Tester: best for forex/metals with tick data. QuantConnect: professional-grade, supports live trading.
- How much data do I need?Minimum 200 trades in sample for statistical relevance. For pattern strategies: at least 3–5 years covering different market regimes (trending, ranging, high/low volatility). More data is generally more reliable, up to the point where market structure has fundamentally changed.
- What is the Sharpe ratio?Sharpe = (Strategy return − Risk-free rate) / Standard deviation of returns. Measures return per unit of risk. Below 1.0 = poor. 1.0–2.0 = good. 2.0+ = excellent. Most successful hedge funds operate with Sharpe ratios of 1.0–2.5.
- How do I know if my backtest is curve-fitted?Signs: high in-sample performance that collapses out-of-sample; too many parameters relative to trade count; sensitivity to small parameter changes; works only on specific years. Solution: walk-forward testing and minimum parameter count.