Skip to content

BACKTEST_RESULTS.md

Latest commit

 

History

History
45 lines (33 loc) · 2.12 KB

BACKTEST_RESULTS.md

File metadata and controls

45 lines (33 loc) · 2.12 KB
Raw

Backtests

This document explains how backtests work in AlphaMind and how to read the output. It deliberately does not publish a performance track record, because the default data is seeded and synthetic; any numbers you see are illustrative of the mechanics, not evidence of real-world returns.

Running a backtest

Via the API:

curl -s -X POST http://localhost:8000/api/v1/backtest/ \
  -H 'Content-Type: application/json' \
  -d '{
    "strategyId": "<id from /api/v1/strategies/>",
    "startDate": "2023-01-01",
    "endDate": "2024-01-01",
    "initialCapital": 100000
  }'

Via the UI: open Backtest, pick a strategy, set the date range and initial capital, and run.

Reading the result

The response is a flat object:

Field Meaning
totalReturn Total return over the period.
annualisedReturn Return scaled to a yearly rate.
sharpeRatio Excess return per unit of total volatility.
sortinoRatio Excess return per unit of downside volatility.
maxDrawdown Largest peak-to-trough decline (fraction).
winRate Share of winning periods (fraction).
profitFactor Gross profit divided by gross loss.
finalCapital Ending capital from initialCapital.

Interpreting metrics responsibly

  • A backtest is a hypothesis about the past, not a promise about the future.
  • The default series are synthetic and deterministic, so results are reproducible but not market-realistic.
  • To run against real history, supply real price data through the market-data connectors and extend the strategy and backtest services accordingly.

Where the logic lives

The backtesting entry point is the backtest router and its service; additional research-grade backtesting utilities live in code/backend/market_data/backtesting.py. The reinforcement-learning environments in code/ai_models/environments provide an alternative, simulation-based way to evaluate agents.