Resource

Backtesting software: what to check before you commit

Backtesting tools range from spreadsheets to full research platforms. The difference between them is not the equity curve they draw — it is what happens under the hood.

A backtesting tool is a research instrument, and like any instrument it is judged by what it refuses to hide. Before you pay for one, find out how it handles costs, out-of-sample testing, and the boring details of data — those three questions eliminate most of the field.

Execution realism comes before feature count

The fastest way to ruin a backtest is to assume every order fills at the close price with no friction. Spreadsheet tests and entry-level tools quietly do exactly that. Serious software models the spread, the commission, and the slippage you would actually pay, and lets you set those assumptions per instrument rather than hiding them behind a default. Ask a vendor how fills are modeled and whether you can see the cost profile the test used — if the answer is a marketing page, keep looking.

The second question is about time. A strategy that is tuned and scored on the same history is being graded on its own homework. Look for walk-forward testing, a holdout that the tuning process never touches, and statistics that correct for how many variations you tried. These features are what separate a measurement from a story.

A practical test: run the same strategy in two tools. If the returns differ sharply, one of them is lying about fills, costs, or data alignment — and you cannot tell which until you dig into the mechanics.

A short evaluation checklist

Cost modelingCan you set commissions, spread, and slippage per instrument, and see them in the results?
Out-of-sample pathIs there a holdout or walk-forward process that your parameter search never sees?
Data hygieneDo corporate actions, dividends, and delistings get handled, or do you inherit survivorship problems?
From test to tradingCan a validated strategy run in paper mode without being rewritten, or does every idea die in export?

That last row matters more than it looks. Most backtests end as a CSV nobody acts on. Tools where a passing test becomes a paper run in the same system close the gap between research and behavior — which is the entire point of testing in the first place.

The platform

How Trade OS handles backtesting

  • Two engines behind one strategy definition: screen ideas in seconds on the vectorized engine, then re-run the same strategy order-by-order on the event-driven engine with realistic fill modeling and configurable trading-cost profiles.
  • Validation is part of the workflow, not an upgrade: walk-forward validation, deflated Sharpe statistics, an overfitting-risk grade, and seeded Monte Carlo simulation run before anything reaches paper mode.
  • 129 strategy templates across trend-following, mean-reversion, momentum, volatility, and event-driven categories give you working code to test the tooling against, with bounded Optuna parameter sweeps and a separate holdout for out-of-sample checks.
  • Alternative data — news sentiment, factor and macro inputs, earnings, dividends, insider flows — is served through the platform's own market data layer, so a backtest can use more than price and volume without a custom pipeline.

Run your first strategy through both engines on a 14-day free trial.

Start free trial