LLM for Automated Trading

The project aims to use the power of Large Language Models (LLMs) to achieve better automated trading.

The main approach is to use LLM to analyse the sentiment of the news for a particular stock or crypto currency and use this a signal to buy or sell.

There are 3 main steps

1. Collect related news, using news API or scraping news website;
2. Analyse the news sentiment using LLM, which can be done through LLM API (such as chatGPT API) or run a LLM locally, such as FinGPT (open-source alternative to BloombergGPT) or Llama2 by Meta;
3. Execute the trade based on the analysis results:
   1. We will probably start with ‘paper trading’ (simulated trading) with platform such as Alpaca;
   2. We can also include other trading strategies such as those based on technical analysis.

Readings

See this paper for the details on sentiment analysis using LLM: Expected Returns and Large Language Models

GPM: A graph convolutional network based reinforcement learning framework for portfolio management

Reinforcement-Learning based Portfolio Management with Augmented Asset Movement Prediction States

iQUANT: Interactive Quantitative Investment Using Sparse Regression Factors (Computer Graphics Forum – CFG 2021)

Sentiment Analysis in the Era of Large Language Models: A Reality Check (arXiv 2023)

Software

Alpaca: https://alpaca.markets/

TradingView: https://www.tradingview.com/