Inspiration
We wanted to understand how real trading firms react to fast-moving markets. The Optiver challenge was a perfect chance to combine news analysis, market data, and algorithmic decision-making into one system.
What it does
Our algorithm trades instruments and uses social feed messages to predict which ones to sell or buy. It also takes dual listings and adjusts the price to hedge.
How we built it
We built a news engine to analyze news messages and give an impact score for every stock. The logic regarding dual trading is handled in the function dual_logging.
Challenges we ran into
It was very challenging to understand the topic, as all of us were new to trading. We struggled to understand what the best course of action would be for a market maker in a certain situation. Additionally, due to the fast-paced environment, testing was difficult.
Accomplishments that we're proud of
Our code achieved positive PNL during previous tests, with approximately linear growth over a longer span of time.
What we learned
We learned a lot about algorithmic trading.
What's next for MarketSense
In the future we would like to include embedded representation of the training sets and "real"-learning.
Built With
- python
- training-data
- vectorized-representation
Log in or sign up for Devpost to join the conversation.