How We Backtest

Real historical OHLCV data, no lookahead bias, the same rule engine used for live trading — and a plain account of what backtests can't tell you.

Real data, not a simulator

Every backtest runs on actual historical daily price bars — open, high, low, close, and volume — sourced from Yahoo Finance, covering US stocks, ETFs, and crypto pairs going back up to 10 years. Indicators (RSI, moving averages, MACD, Bollinger Bands) are computed directly from that price series using the standard formulas, with enough warm-up history trimmed off the front so an indicator like a 200-day moving average is never evaluated before it actually has 200 days behind it.

No lookahead bias

Lookahead bias — letting a strategy see information it couldn't actually have known on the day it traded — is the most common way a backtest quietly lies. For multi-timeframe rules, our engine only ever lets a daily row see the most recently completed week, never the one still in progress. If a rule needs more history than the selected time range can provide, the backtest refuses to run and tells you to widen the range, instead of evaluating against incomplete data.

The backtest and the live bot are the same code

The exact same entry and exit rule functions that produced your backtest are the ones handed to the paper-trading bot and, if you connect a brokerage, the live one. Each backtest starts from $100,000 in simulated capital, applies your position sizing and stop-loss, and assumes fills at each day's closing price.

Benchmarks, not vibes

Every backtest is plotted against two benchmarks computed on the same ticker, the same time range, and the same starting capital: buying and holding from day one, and dollar-cost averaging in at regular intervals. Where we show a confidence score, it is a 95% Wilson lower bound on the strategy's win rate — not the raw win rate — so small, lucky-looking samples are penalized.

What a backtest can't tell you