BotTrade simulation contract
Read the product-specific analysis for the simulation contract and agent interaction cycle.
Backtesting methodology
Methodological principles for backtesting AI trading agents
An autonomous trading agent observes a changing market, reasons with a model, invokes tools, and acts under explicit constraints. A valid backtest evaluates this complete decision process rather than relying exclusively on the terminal equity curve.
BotTrade supports research and evaluation; it does not provide financial advice or live-trade execution.
Experimental requirement
Expose only the information available to the agent at each simulated point in time. Permit the agent to observe, decide, trade, and advance one bar at a time. Evaluate return and risk under identical rules across multiple scenarios.
Experimental protocol
- 1
Define the market task
Specify the symbols, starting capital, date range, trading cadence, leverage limits, and short-selling policy. A predefined task makes each result interpretable.
- 2
Exclude future information
At step T, expose only the bars and inputs available at step T. Indicators must not incorporate subsequent data, and the agent must not receive a completed price series.
- 3
Specify execution rules
Define order fills, transaction costs, position limits, and liquidation behavior. Execution assumptions materially affect strategy outcomes.
- 4
Apply controlled comparisons
Hold the scenario contract constant when changing a model or prompt, and evaluate several market conditions rather than optimizing for a single run.
Evaluation measures
Return alone is insufficient for evaluation.
Assess total return together with risk-adjusted metrics, drawdown, liquidation status, trade count, and scenario constraints.
- Return
- The change in equity over the complete run.
- Sharpe and Sortino ratios
- Return relative to total variability and downside risk.
- Maximum drawdown
- The largest observed peak-to-trough decline.
- Liquidation and trade count
- Capital survival and the frequency of agent interventions.
Related BotTrade research
Controlled agent comparisons
Use the agent evaluation protocol to compare models, prompts, and tools under consistent conditions.
MCP integration
Consult the hosted MCP analysis for tool-mediated agent integration.