ml-day-trading-simulator

This day trading simulator uses a Neural Hierarchical Interpolation for Time Series (NHITS) model to simulate high-ish frequency day trading.

Overview

1. It uses yfinance to poll stock data every 3 seconds.
2. It stores that data in a PostgreSQL database.
3. Before the market opens each day, it pulls the data and trains a new NHITS model.
4. In addition to polling the stock data every 3 seconds and writing to the database, it uses the model to predict the stock price 3 seconds later.
5. Based on the predicted price, it simulates a buy or sell and tracks a simulated account balance and stock holdings.
6. At the end of the day, it record the account's value and trades executed throughout the day.

Setup

1. Setup a PostgreSQL database with the quote_data and account_data table columns set as the same in config.py.
2. Provide valid database connection parameters in db.py.
3. Build a docker image with the provided dockerfile and run as a container OR use notebook.ipynb to run the code manually.
4. Setup a cronjob to run the container before each weekday.

Common Issues

1. Yahoo Finance often rate limits quote info pulls. This is the main blocker for this script and often results in lack of data.