How to backtest a trading strategy

Choose a US stock or ETF, a bar size, a date range, and precise entry and exit rules. Describe the setup to Orbit or enter the fields yourself in Builder. An idea such as “buy when momentum looks strong” needs a defined signal and a defined exit before a backtest can evaluate it.

For example, you can specify a long-only SPY test that enters after a closed daily bar meets an RSI threshold and exits at a set target or stop. This is an example of a testable rule, not a recommendation or a claim that it performs well. The AI backtesting page puts the same workflow in one place, including what Stratos will not do.

ATLAS uses historical open, high, low, close, and volume bars. Standard entry signals are evaluated after a bar closes, with entry simulated at the next bar’s open. Stops and targets can exit within a bar, including the entry bar. The result is a simulation under the rules and data you selected, not a record of trades placed in a brokerage account.

Slippage and commission start at zero in Builder. Set values that fit the test you want to examine, then rerun it to see how sensitive the result is to costs. The default comparison is buy and hold; you can also choose one benchmark ticker. A positive strategy return alone does not tell you whether the rules added value over that comparison.

Review the equity curve, drawdown, trade count, and the trades available in your plan. A result that depends on a few trades or one date range deserves more scrutiny. Where available, use the diagnostics and Monte Carlo views to examine sensitivity; they do not turn a historical result into a forecast.

You can save the strategy and monitor it with Stratos paper trading on live market data using virtual capital. Paper results are still simulated. Stratos does not connect to a live brokerage or place real-money orders in this release.

Historical prices cannot establish what will happen next. Results depend on the chosen period, rules, available bars, and cost assumptions. Check the assumptions and compare different periods before trusting a pattern. The Terms of Service explain the limits of hypothetical performance. For the full list of supported markets and tools, see Product facts.