bill
d794910697
Two-person trading team setup: Bill (quant/algorithmic) and Colette (market observer). Includes phase-by-phase curriculum, separate sprint tracks, trade journal templates, roadmap to prop firm challenge, and agent handoff documentation. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
3.5 KiB
Bill's Learning Track — Quant / Algorithmic
Background: Data scientist, ML experience, cybersecurity/scripting skills Role: Build and validate — data pipelines, backtests, models, automation, alerts
Phase 1 — Market Data Infrastructure
Goal: Get clean, reliable data into your toolchain
- TradingView Pine Script basics — indicators, alerts, screeners
- Python market data:
yfinance,pandas-ta,ccxt(crypto if relevant) - Free vs paid data sources — what's actually needed at this stage
- Build a basic OHLCV data fetcher for a watchlist of tickers
- Understand data quality issues: survivorship bias, split adjustments, gaps
- Set up a local data store (SQLite or Parquet) for backtesting
Deliverable: Script that pulls daily OHLCV for a watchlist and stores it locally
Phase 2 — Indicator Implementation & Backtesting
Goal: Implement and test common signals before trusting them
- Implement SMA, EMA, RSI, MACD, ATR from scratch (don't just use a library — understand them)
- Build a simple backtesting loop (vectorized, not event-driven to start)
- Understand backtest traps: look-ahead bias, overfitting, survivorship bias
- Walk-forward validation basics
- Metrics: win rate, expectancy, Sharpe ratio, max drawdown, Calmar ratio
- Pine Script: build a simple strategy and view equity curve in TradingView
Deliverable: One complete backtest with proper train/test split and metrics reported
Phase 3 — ML Signal Generation
Goal: Apply ML where it actually adds value (not everywhere)
- Feature engineering for price data (returns, rolling stats, lagged features)
- Classification: predict direction (up/down/flat next N bars)
- Avoid the common traps: data leakage, overfitting short samples
- Evaluate with financial metrics, not just accuracy
- Regime detection: is the market trending, ranging, or volatile?
- Understand when ML helps vs when simple rules beat it
Deliverable: A classifier that outputs trade signals with a documented edge (positive expectancy on out-of-sample data)
Phase 4 — Automation & Alerting
Goal: React to markets without staring at screens
- TradingView alerts → webhook → Python handler
- Build a market watch script (runs on schedule, flags setups from Colette's criteria)
- Notification system (email, SMS, or Telegram bot)
- Paper trading automation: auto-log triggered trades to journal
- Understand execution risk: slippage, partial fills, latency (not HFT, but real)
Deliverable: Script that monitors a watchlist and sends an alert when Colette's setup criteria are met
Phase 5 — Prop Challenge Execution
Goal: Apply the system under challenge conditions
- Simulate challenge rules in paper trading: daily loss limit, max drawdown, profit target
- Risk sizing formula locked in (never manual)
- Dashboard: daily P&L, drawdown remaining, progress toward profit target
- Post-trade review script: auto-generate stats from journal
- Two clean simulated challenge passes before paying
Deliverable: Automated prop challenge tracker with real-time status
Tools & Stack
| Purpose | Tool |
|---|---|
| Charting | TradingView |
| Data | yfinance, pandas, pandas-ta |
| Backtesting | Custom vectorized or backtesting.py |
| ML | scikit-learn, lightgbm |
| Automation | Python scripts + cron or systemd |
| Alerts | TradingView webhooks + custom handler |
| Storage | SQLite or Parquet files |