Building a Multi-Asset Trading Bot in Python with Async Event Bus

Overview

Most trading-bot tutorials stop at "loop through symbols and call the broker." Real bots need an event bus, multiple broker abstractions, paper-trading parity, a kill switch you can hit fast, and a database schema that survives your first compliance audit. This bot covers all of that — Canadian-market focused (Questrade), with US stocks via Alpaca and crypto via CCXT, all running through one async event loop.

Stack

• Language: Python 3.10+ (tested on 3.10, 3.11, 3.12)
• Async: asyncio, aiohttp, httpx
• Data: pandas, numpy
• Database: SQLAlchemy 2.0 async — SQLite/aiosqlite (dev), PostgreSQL (prod)
• Config: Pydantic Settings (BaseSettings from .env)
• CLI: Click + Rich
• Brokers: alpaca-py, ccxt (crypto)
• Dashboard: FastAPI + uvicorn (optional)
• Testing: pytest, pytest-asyncio, pytest-cov
• Linting: ruff, black, mypy
• Deployment: Docker

Event-Driven Architecture

Components communicate exclusively through an event bus (core/events.py):

• QUOTE_UPDATE — market data updates from a broker
• SIGNAL_GENERATED — a strategy fires
• ORDER_SUBMITTED / ORDER_FILLED — order lifecycle
• POSITION_CHANGED — portfolio update
• KILL_SWITCH_TRIGGERED — emergency stop

Strategies don't know about brokers and brokers don't know about strategies. Everything routes through events.

Broker Abstraction

Every broker implements BaseBroker:

async def connect(self) -> None: ...
async def disconnect(self) -> None: ...
async def get_quote(self, symbol: str) -> Quote: ...
async def submit_order(self, order: Order) -> str: ...
async def cancel_order(self, order_id: str) -> bool: ...
async def get_positions(self) -> list[Position]: ...
async def get_account_balance(self) -> Decimal: ...

Switching from paper to live is one config change.

Strategy Interface

Strategies implement BaseStrategy with initialize(), generate_signals(), and validate_parameters(). The bot auto-discovers strategies via STRATEGY_NAME and STRATEGY_CLASS exports — adding a new one is a single file.

Active Strategies

Database Models

AuditLog, Order, Trade, Position, MarketData, StrategyState, FintracReport. Schema is created directly via init_db() — no migrations layer; the schema is the code.

Risk & Compliance

• Kill switch: any subscriber can publish KILL_SWITCH_TRIGGERED. Brokers cancel open orders and halt new submissions immediately.
• FINTRAC reporting: any single transaction at or above CA$10,000 is auto-flagged into a FintracReport row.
• Position limits: MAX_POSITION_SIZE_PCT (default 5%), MAX_DAILY_LOSS (default $1,000).

Key CLI Commands

trading-bot run --mode paper --strategies sma_crossover
trading-bot backtest -s sma_crossover --start 2024-01-01 --end 2024-12-31 -sym AAPL
trading-bot verify-broker
trading-bot status

Lessons Learned

• float(None) crashes: optional price fields must be guarded — broker APIs return null on illiquid symbols and a single float(quote.bid) deep in a loop will down the bot.
• Questrade 401 retry guard: a naive retry-on-401 will infinite-loop when the refresh token is genuinely dead. Cap retries and surface the error.
• Admin keys belong in request bodies, not query strings: query strings end up in access logs.
• Dashboard auth is critical: a FastAPI dashboard with no auth is a remote shell to your broker. Don't ship without it.

Testing

206 tests (pytest), 43% coverage. CI matrix runs Python 3.10, 3.11, and 3.12 through ruff → black → mypy → pytest with coverage.