Optiverse
Could not get more textbook than this
How we built it
Instead of trying to get ahead of the curve through dark magic, we focused on the essentials:
- Fine-tuned a BERT model for financial news, invested a lot early on model accuracy; quantisation!
- Stayed true to the market maker paradigm, constantly quoting based on good predictions of theoretical fair price - taking into account the dual-listing for optimal returns
- Went big on directional trades
- Leveraged (rare) arbitrage opportunities
Challenges we ran into
Our mechanisms to ensure the server limits were not breached took us a while to get right... We had internal discussions, our project versions diverged, converged, eventually resulting in a solid solution. Never throw in the towel!
Accomplishments that we're proud of
By far the trading bot with the most stable returns, very high sharp ratios each cycle - little risk, little variance, steady cash ins.
What we learned
A good predictive engine allows you to be WAY more liberal with risk consideration. A good sharpe ratio & steady HFT returns beats insane individual trades.
Built With
- finbert
- machine-learning
- python
- transformers
Log in or sign up for Devpost to join the conversation.