Skip to content

BACKTEST-PHASE3.md

Latest commit

 

History

History
47 lines (34 loc) · 2.89 KB

BACKTEST-PHASE3.md

File metadata and controls

47 lines (34 loc) · 2.89 KB

Phase 3: Agent-Triggered Backtesting

The orchestrator exposes POST/GET /backtest so agents (or any client) can run a backtest using qtos-core and use the results to gate decisions—e.g. "only alert if this signal would have been profitable in a backtest."

What’s implemented

  • Backtest API: POST /backtest and GET /backtest with symbol, data source (CSV or data-ingestion-service), initial_cash, quantity, strategy_type (currently buy_and_hold), and optional period for data_service.
  • Runner: orchestrator.backtest_runner loads OHLCV from a CSV path or from the data-ingestion-service (DATA_SERVICE_URL), runs qtos-core’s BacktestEngine with BuyAndHoldStrategy, and returns metrics (PnL, CAGR, Sharpe, max drawdown, etc.) and num_trades.
  • Requirements: qtos-core must be present as a sibling of the orchestrator (same workspace root). Optional: data-ingestion-service for live data.

Agent gating pattern

  1. Agent produces a signal (e.g. sentiment shift, insider activity).
  2. Agent (or orchestrator) calls POST /backtest with the relevant symbol and parameters (e.g. quantity, period).
  3. Client checks metrics in the response: sharpe_ratio, max_drawdown_pct, total_return_pct, etc.
  4. Gate the alert: Only notify or act if metrics pass thresholds (e.g. sharpe_ratio > 0.5 and max_drawdown_pct < 20).

Example (pseudo-code in an agent or wrapper):

response = requests.post(
    "http://localhost:8000/backtest",
    json={"symbol": "AAPL", "data_source": "data_service", "period": "1y", "quantity": 10},
)
data = response.json()
metrics = data.get("metrics", {})
if metrics.get("sharpe_ratio", 0) > 0.5 and metrics.get("max_drawdown_pct", 100) < 20:
    send_alert("Signal passed backtest", data)
else:
    log("Signal did not pass backtest; no alert")

Data sources

  • csv — Use csv_path (or default qtos-core sample at qtos-core/examples/data/sample_ohlcv.csv). Good for demos and CI.
  • data_service — Use data-ingestion-service. Set DATA_SERVICE_URL (e.g. http://localhost:8001). Requires the data service to have ingested prices for the symbol.

Extending

  • More strategies: Extend backtest_runner.run_backtest() to accept other strategy_type values and instantiate the corresponding qtos-core strategy (or wrap VectorBT/Backtrader later).
  • Advisors/validators/observers: The qtos-core BacktestEngine already supports advisors, validators, and observers. The runner can be extended to accept optional agent hooks (e.g. regime filter, guardian validator) and pass them into the engine.
  • Richer metrics: The API already returns standard metrics; add custom metrics in the runner or in the client.

References

  • qtos-core backtesting: qtos-core README and backtesting/ module.
  • Orchestrator API: Swagger at http://localhost:8000/docs when the API is running.