Backtest and bot share one function
Deciding and filling are the same code in backtests and in the worker. Replaying the scheduled run over past trading days on real QQQ data matched the backtest down to the fills.
In progress · Paper trading
Quant Lab
Runs rule-based strategies on US daily bars from Toss Securities' Open API, compares them in backtests, and paper-trades them on a schedule. No real orders, and every decision is explained in plain words.
- What
- Web app + worker · for my own use
- Since
- 2026
- Role
- Solo
- Runs on
- Docker Compose · private, over Tailscale
About
The end goal is automated trading. The first milestone is smaller on purpose: a bot that paper-trades on a schedule for two weeks with nobody touching it, and no code path to a real order at all.
It's built so someone who doesn't know stocks can follow along. Strategies have nicknames (“slow trend rider” is the 50/200 golden cross), settings are choices, and each decision reads like “It's trending up. Buy QQQ when the market opens tomorrow.” It's a private tool, so there's no link; the screenshots use synthetic data, since Toss market data can't be shared.
What it does
- Daily bars
- Pulls each symbol's full history into PostgreSQL and checks for gaps, duplicates, non-trading days and outliers. Re-syncs when adjusted prices change.
- Backtests
- Four well-known strategies — golden cross, Faber's 200-day line, Connors RSI(2), turtle breakout — each set against just holding.
- Scheduled paper trading
- Two hours after the US close, the worker decides on the finished bar and fills at the next day's open, with fees and slippage.
- Telegram alerts
- Signals, fills, warnings, restarts and the day's portfolio value. An outside heartbeat speaks up if the worker or server stops.
How it’s built
- Backend
- Python 3.14 · uv · FastAPI · Pydantic v2 · SQLAlchemy 2 · Alembic · pandas
- Frontend
- React · TypeScript · Vite · TanStack Query · Lightweight Charts
- Data
- PostgreSQL 17 · Toss Securities Open API (read-only) · exchange_calendars
- Infra
- Docker Compose · AWS Lightsail (Terraform) · Tailscale · GitHub Actions
- Quality
- pytest · ruff · mypy (strict) · oxlint
Notes
Today's bar isn't finished yet
Toss starts filling in the day's bar before the US market even opens. Only bars two hours past the close count, and a later change triggers a re-sync.
One order, however many restarts
Unique keys on bot, symbol and trading day, one pending order per bot, and an advisory lock so only one worker runs. Evaluating the same bar again does nothing.
Trend following buys calm, not returns
On QQQ since 2000, the 50/200 cross made 10.5% a year with a −36% worst drop, against 8.1% and −83% for holding. Since 2010, holding wins on return.
Not investment advice. Every order is simulated, and the screenshots use synthetic data.