Backtests: Wizard or liar?

A backtest checks strategy on historical data. Sounds objective. Yet backtests lie so well that traders believe in guaranteed profit. Then reality hits, and it's brutal.

Where false results come from?

1. Overfitting

You tweak parameters until backtest shows 300% profit. You found perfect numbers for the past, not the future.

Analogy: You adjusted glasses to see yesterday's landscape perfectly. Today's landscape is different, and you're blind.

Overfitting signs:

• Results change 10%+ with 1% parameter shift
• Equity curve looks too perfect, no drawdowns
• No crashes beyond 5% (unrealistic)

2. Survivorship bias

Backtests often use coins still alive today. But 90% of altcoins die.

Truth: Testing on BTC, ETH, Cardano, Solana isn't universal. You just picked historical winners by accident.

3. No spreads or commissions

Backtest closes at exact prices. Reality:

• Spread (bid/ask gap): 0.05-0.5%
• Exchange fee: 0.05-0.1%
• Slippage: 0-2% on volatility

Backtest: 15% profit Reality: 12-13% after costs

Over 100 monthly trades, this adds up.

4. Black swans and gaps

Backtests use OHLC bars. Data between bars is missing.

Real scenario: Stop-loss set to $40,000. Overnight, BTC gaps to $37,000. Stop fills at $37,500 (slippage).

Backtest: Exact stop execution.

5. Liquidity mismatch

Backtest assumes you can buy/sell any volume at any price.

Reality: On small pairs, a $100k order moves price 2-5%.

How to backtest honestly?

1. Conservative parameters

• Add 0.2% spread
• Include commissions manually
• Use 1-2% slippage on volume

2. Walk-forward analysis Test on chunks of history:

• Optimize on 01.01-30.03
• Test results on 01.04-30.06
• Repeat for different years

If parameters change yearly—strategy isn't stable.

3. Test different conditions

• Uptrend
• Downtrend
• Sideways
• High volatility

If it only works in uptrends, it's not universal.

4. Start with paper trading

AI Traders offers paper trading with real current prices. Test here before real money. One month beats one year of backtests.

Rule of thumb

If backtest shows 50% yearly profit, halve expectations. 25% is realistic.

If it shows 200%+—this isn't investment, it's casino. Reality will disappoint.

Backtests filter ideas, not guarantee profit. Real validation starts on-market.