Backtest trading strategies on real market data

Backtesting software that shows its assumptions

Every backtest is a simulation, and every simulation makes assumptions: how orders fill, what they cost, and what the strategy was allowed to know at each moment. Roboquant shows you those settings on every run and saves them with the result, so you can see what a number is based on and reproduce it later.

Metrics describe a historical simulation, not a forecast. The sections below cover the four things that decide whether a backtest is worth trusting.

1. Tick data vs bar data

A bar records open, high, low and close. It does not record the order in which those prices traded. When a stop and a target both sit inside one bar, a bar-only backtest has to guess which was hit first, and that guess can turn a loser into a winner.

Roboquant offers four fill models:

OHLCV fills are on every plan. Tick fills start on Starter. L2 book fills start on Elite and cover a recent rolling window.

Read more: Tick vs bar backtesting: when intrabar fills change everything

2. Costs you set, applied every time

Commission and slippage are inputs on every backtest, not an afterthought. The engine applies the instrument's contract specification automatically, charges commission on each side (flat, per contract, or percentage), and applies slippage according to the fill model. With L2 fills, market orders walk the visible order book, and the result reports the average levels walked and the slippage cost.

Hypothetical example (invented numbers): a strategy makes 400 round-trip trades with a gross profit of $6,000. At $5 commission per round trip and $12.50 average slippage per trade, costs come to $7,000. The same strategy that looked profitable before costs loses $1,000 after them.

Read more: Why your backtest doesn't match live trading

3. No peeking at the future

Look-ahead bias happens when a backtest uses information that was not available when the decision was made, such as a bar's close or the day's final high. In Roboquant's tick mode, the current bar is the forming bar as known at that tick, and indicators read the last completed bar, so final-bar values do not leak into earlier decisions.

Read more: Look-ahead bias in backtesting: 7 real examples

4. Out-of-sample validation

A parameter set tuned on all of your history will almost always look good on that history. Roboquant's optimizer can hold data back: IS/OOS optimizes on the first part of the date range and checks the result on the untouched remainder, and walk-forward repeats that over rolling windows. IS/OOS starts on Pro; walk-forward starts on Elite.

Also useful: TradingView Strategy Tester limitations, for the settings to check in any backtester.

How strategy backtesting works in Roboquant

1. 1. Describe your idea.

   Write the rules in plain English in chat, or start from a strategy template such as RSI, MACD or mean reversion. The AI writes a Roboquant strategy, compiles it and fixes compile errors before you run anything.

2. 2. Set up the run.

   Choose the symbol, timeframe, date range and fill model. Set starting capital, commission, slippage and leverage, and review the strategy's parameters. Changing a parameter does not require recompiling.

3. 3. Run or Replay.

   Click Run for the fastest result, or Replay to watch the strategy trade across the chart: bars, intrabar tick steps, indicators, trades, stop and target levels, trailing-stop movement, equity and logs. Both produce the same result; Replay speed only changes the display.

4. 4. Read the result.

   Review net P&L, max drawdown, Sharpe and Sortino, win rate, profit factor, average trade, MAE and MFE where available, the full trade list with commissions and slippage, and Monte Carlo trade distributions (Starter and above). Every saved run keeps its configuration, so reopening it restores the exact setup.

Full reference: Backtesting docs

Market data included in your plan

Historical market data comes from CME and is included in every plan. There is nothing extra to buy or connect, and your broker is never used as a data source. If a date range falls outside stored coverage, the run stops with a data-availability error instead of quietly changing your dates.

Availability varies by symbol, timeframe and data type. See pricing for every plan limit.

From backtest to optimization to live trading

A backtest in Roboquant is not a separate prototype. The strategy compiles once into an artifact, and that same artifact runs in the backtest, the optimizer and live trading, so there is no rewrite between testing and trading.

• Optimize. Search parameter ranges with grid search on every plan, and Bayesian, genetic and multi-objective (Pareto) search on Elite and Ultra. Test one parameter set across several symbols and timeframes to look for settings that hold up beyond a single market. The optimizer uses the same data, costs and engine as a normal backtest.
• Validate. Re-run the chosen parameters as a normal backtest, check them out of sample, and review the Monte Carlo distribution before risking money.
• Deploy. Run the same compiled strategy on your connected broker account and follow orders, positions, P&L and logs as it trades. Set maximum order size, maximum position size, an order-rate limit and a daily-loss limit on every deployment. Native live deployment starts on Pro.

FAQ