optiver_020

here is our video, have a look. Just keep in mind, we did NOT sleep for a long time.... a very long time :D --removed--

Here is the link to the c9 env (https://eu-central-1.console.aws.amazon.com/cloud9/ide/e692e55ad8f24db0a4dcb82decae7803?region=eu-central-1#). I hope you can access it, if not just try to contact us :)

Description of our algorythm: Our algorythm is mostly made out of two major parts. First of all our news handler, it checks if there are any new news and runs it through our own selfdeveloped NLP model for classification. To improve latency we extract the stocknames by an algorythm and our custom model predicts if it goes up or down. This way we can achieve as much speed as possible. Ones we know which stock is affected by the news we trade on it accordingly once and then lock it. This means it will not be traded until the price of the stock stabilizes. Secondly, we use a market maker which is a optimized version of the provided one, to make some money on the side and to stay inside of the position limits. Theses to parts are executed one after another. -> Check for news -> trade stock 1 -> check for news -> trade stock 2 -> .... This gives the best average reaction time on any new news. We decided against using dual stock trading because it is very unlike to work reliably with minimal risk involved (Even more when 40 algorythms try the same thing). In addition, we decided to trade only on exchange A. This is because it allows us to increase our speed on the ones we do trade.

Repo Overview: algo = our main algorythm (algo/algo.py) news = custom NLP model useful tools = scrips that helped us along the way backup = past iterations of code or scripts without any usage anymore

Thank you so much for hosting this. It was really fun <3