Step 2: Run backtest
Once your strategy is generated and configured, the next step is to see how it would have performed on historical data.
Backtesting helps you verify that the logic makes sense before you deploy an agent.
2.1 Start a Backtest
After Step 1, you’ll see your strategy summary and parameters.
To run a backtest: Click Run Backtest.
Under the hood, the system will:
download historical market data for the selected period,
apply your strategy code and agent metadata,
simulate every signal, order, and position change as if the strategy had been running live.
You don’t need to write any backtest script—everything is driven by the strategy you created
2.2 Review Backtest Results
When the backtest finishes, you’ll see a simple performance view, including:
Overall PnL (profit / loss over the test period)
Drawdown behavior (how deep the worst loss was)
Number of trades and basic hit rate (wins vs losses)
A PnL curve over time
The goal here is not to overwhelm you with metrics, but to answer a few key questions:
Does this strategy roughly behave as I expect?
Is the risk profile acceptable for me?
Is it over-trading or barely trading?
If something looks off (e.g. huge drawdown, too many trades, or flat performance), it’s a signal to adjust.
2.3 Iterate Using Natural Language
You can refine your strategy directly from the backtest view.
Examples of follow-up instructions:
“Why there is no trading?”
"How to improve performance?"
“Trade less frequently.”
“Use a smaller position size.”
“Be more conservative during high volatility.”
“Widen the grid spacing.”
When you give this kind of feedback, the system will:
update the agent metadata (risk settings, parameters),
adjust the strategy code accordingly,
and let you run another backtest with one click.
Repeat this loop until the behavior and risk profile feel right.
Last updated