What it does
We are going full on on news sentiment analysis. Our custom built two part model consists of a gate, that decides what stocks are affected by a message and the predictor, which gives the actual stock prediction based on the news message.
We are combining this with a nice stop loss, take profit ladder where we stop loss at -5% but move up our profit window when we cross the bar that is 80% of the predicted gain.
On top, we're running it completely in rust, on the same aws server cluster as the exchange server for minimal latency (reverse engineered the location)
How we built it
- reimplemented the optibook library in rust
- trained the ML models in python; the model consists of three components: FinBERT to generate embeddings for the ticker, a Gate module that decides which stocks are affected by the headline, and a Predictor module that predicts the stock price movement.
- built a custom dashboard that shows our strategy in real time to allow for better debugging, fully in the terminal and interactive
- re wrote the optibook library in rust for efficiency gains
- reverse engineered which aws server and cluster the official server runs on, rented the same and we are running our code in the same cluster
Challenges we ran into
- our quoting and arbitrage attempts were widely unsuccessful but also cost a LOT of time
Accomplishments that we're proud of
- custom built and custom trained model
- countless attempts and relentless testing
What we learned
- a lot about the market, how o rder books work
What's next for f'(x) = 0
we will fill this out before sunday
Log in or sign up for Devpost to join the conversation.