A joint experiment of a small team aiming to:
- Analyse public engagement with btc
- Predict inflow/outflow of public interest (measured by trading volume)
Approach:
Stage 1
- Scrape timestamped reddit and twitter posts puling data in a timespan of 1 month on several btc and economy related topics. Such as inflation, employment, taxation
- Train a BERT machine learning model on the post data and analyse the sentiment.
Stage 2 Train a timeseries model on financial data for the same timespan.
Stage 3 Feed the data into a RNN model predicting the following day day's trading volume and public engagement.
Results:
The analysis was deployed temporarily using Streamlit on GCS and was able to predict trading volume with an accuracy of about 80%
Cost of aquiring financial data made this experiment unfeasible to be sustained long term without outside investment.