Using Deep Pre-Trained Language Models to Understand Investor Sentiment and Volatility for Cryptocurrencies

This repo contains our report and code for our UC Berkeley independent study project: Using Deep Pre-Trained Language Models to Understand Investor Sentiment and Volatility for Cryptocurrencies. A big thanks to Rongbing Liang and Yupeng Liu for all their help with the project!

The project was primarily compiled on a Datalore workspace, so the code is included in a number of Jupyter Notebooks, all of which are included in the repo. However, we have not uploaded any of the Twitter data. If you are interested in recreating our results, the Twitter data can be re-queried using the Tweet Data notebook (you will need an academic API access token) and links to the Bitcoin and Ether price data are included in the BTC and ETH Price Data notebook.