Projects

Algorithmic trading system.

Five neural networks read SPY options data, then pick, size and log the trade.

Ongoing Machine learning Financial
A phone showing a trading app over a candlestick chart
Neural networks 5
Training history 24months
Minimum reward to risk 3:2
Daily drawdown cap 4%

Overview

Live SPY quotes and option chains feed five networks that predict direction, size, volatility, risk and timing. A trade fires only when they agree, inside hard risk limits, and every step is logged for review.

  • Python
  • TensorFlow / Keras
  • Pandas
  • NumPy
  • Yahoo Finance API
  • Black-Scholes
  • TA-Lib
  • JSONL logging
The ensemble

Five networks, one trade.

  1. 01

    Direction

    Which way the move goes.

  2. 02

    Magnitude

    How far it should run.

  3. 03

    Volatility

    How rough the path gets.

  4. 04

    Adverse excursion

    How far it could go against the trade.

  5. 05

    Timing

    When to get in.

How it was built

5 steps
  1. 1

    Scope

    Short-dated SPY options, five signals, hard risk limits.

  2. 2

    Data

    Live quotes and option chains, saved as JSONL.

  3. 3

    Features

    Technical, volatility and options inputs, no lookahead.

  4. 4

    Ensemble

    Five networks that must agree before a trade.

  5. 5

    Deploy

    30- and 90-day backtests, then live paper trading.

The dashboards

Results

Paper trading, ongoing
Paper trade P&L
Model performance