The Backtester & Monte Carlo

The Backtester lets you test a strategy on historical data. You choose an asset (BTC, ETH, SOL...), a time period, and the backtester simulates the trades the strategy would have made.

The backtest trap: A perfect backtest guarantees NOTHING for the future. This is the overfitting problem — a strategy over-optimized for the past fails on new data. It's like a student who memorizes answers from a past exam: 100% on that exam, 0% on the next.

Monte Carlo simulation solves this. It runs 1000 simulations by randomly shuffling the order of trades (bootstrap resampling). If your strategy is robust, it will perform well in most of the 1000 parallel universes. If it only works in one specific order, that's a red flag.

The Robustness Score summarizes everything in one number. It combines: percentage of profitable simulations, return stability, and Sharpe Ratio consistency across simulations. A score > 70% indicates a reliable strategy.

Practical exercise: go to the Backtester, test CUDA Evolved on BTC, then run Monte Carlo. Compare with Buy & Hold. The robustness difference will surprise you.

Monte Carlo in Practice

Concretely, what does "1,000 simulations" mean? Imagine your strategy made 50 trades in 2025. Monte Carlo takes these 50 trades and shuffles them randomly 1,000 times. Simulation #1: the winning trade comes first, then 3 losers. Simulation #2: 5 losers in a row, then a winning streak. Simulation #738: trades arrive in a completely different order. Each simulation gives a different final PnL.

If your strategy ends positive in 800 out of 1,000 simulations, your robustness score is 80% — it's robust, the strategy wins regardless of trade order. If it only ends positive in 450 out of 1,000 simulations, your score is 45% — it's fragile. The strategy depends on a specific order to work, which probably won't repeat.

Key thresholds:

- 90-100%: Extremely robust — the strategy almost always wins
- 70-90%: Robust — reliable in most conditions
- 50-70%: Fragile — dependent on specific conditions
- < 50%: Dangerous — the strategy has as much chance of losing as winning

Watch Out for Overfitting

Overfitting is the #1 backtesting trap. It's when a strategy is so optimized for past data that it becomes useless for the future. Signs of overfitting:

- A too perfect backtest (PnL > 100%, zero drawdown) — be suspicious!
- A low Monte Carlo score despite a good backtest (sign that results depend on exact order)
- Ultra-specific parameters (e.g., buy when RSI = 31.7 and volume > 1247) instead of general rules

The golden rule: a perfect backtest on the past guarantees nothing about the future. That's why Monte Carlo exists — to separate real robustness from luck.