Quant Backtesting Lab

Live app: quant-backtest-lab.onrender.com

A portfolio-oriented quant backtesting project with:

• React + TypeScript frontend
• C# ASP.NET Core backend
• Python strategy execution with user-authored code and dynamic custom params
• Real historical US equities data from free Stooq CSV endpoints
• WebSocket streaming via SignalR for live run progress, charts, and metrics

Features in this MVP

• Backtest controls for symbols/date range/strategy params
• Strategy editor on Overview:
  • single strategy.py Monaco editor
  • dynamic custom parameter builder (string/number/boolean/json)
  • one-click quant template import (momentum, mean reversion, trend + inverse vol)
  • fullscreen editor mode + local autosave in browser
• Momentum long/short engine (event-driven daily loop, periodic rebalance)
• Live metrics cards and live additional statistics streamed from backend
• Equity curve vs SPY benchmark
• Drawdown and returns distribution charts
• Transactions table

Project structure

• src/: frontend
• backend/: C# API, backtest engine, market data service, SignalR hub

Prerequisites

• Bun (or npm/pnpm, but repo is currently using Bun lockfile)
• Docker
• Optional (non-Docker backend dev): .NET SDK 8.0+, Python 3.10+, pandas, numpy

Run frontend

bun install
cp .env.example .env
bun dev

Frontend runs on http://localhost:5173 by default.

Run backend (Docker)

# from repo root
docker build -t quant-backtest-backend ./backend
docker run --rm -p 5055:5055 -e PORT=5055 quant-backtest-backend

Backend will run on http://127.0.0.1:5055.

If you use a different port, update VITE_API_BASE in .env.

Recommended local setup

• Terminal 1 (frontend):

bun dev

• Terminal 2 (backend):

docker build -t quant-backtest-backend ./backend
docker run --rm -p 5055:5055 -e PORT=5055 quant-backtest-backend

API endpoints

• POST /api/runs -> start a run
• GET /api/runs/{runId} -> get finalized result
• GET /api/runs/{runId}/transactions -> get transactions
• WS /hubs/runs -> SignalR progress/events

Live SignalR payloads include run progress, equity points, drawdown points, daily return points, and live metric/stat snapshots.

Python strategy contract

The frontend sends:

• strategyCode: string (single python file contents)
• strategyParams: Record<string, unknown> (user-defined params)

Preferred strategy entrypoint:

def generate_signals(context, params):
    # return one of:
    # 1) DataFrame long-form: columns ['date', 'symbol', 'signal']
    # 2) DataFrame wide-form: index=date, cols=symbol, values=signal
    # 3) Series: symbol -> signal (applied to latest date)
    # 4) dict: either {date: {symbol: signal}} or {symbol: signal}
    ...

Context passed to generate_signals:

• context["dates"]: ordered date list (YYYY-MM-DD)
• context["symbols"]: symbol universe
• context["close"]: pandas DataFrame of closes (date x symbol)
• context["returns"]: pandas DataFrame of daily returns (date x symbol)
• context["datasets"]: helper object with load("market_daily", date_start=None, date_end=None)

Legacy support is kept for class Strategy(...): get_signals(...) and compute_signals(...). If custom Python fails, the run fails by default unless fallbackToBuiltinOnPythonError is set to true.