Sentiment Trader

Team Number: 79 My notebook name: Trader.ipynb

My algorithm attempts to capitalize on the Dual Listing Problem. However, it does so while predicting future movements. I.e., I compare the prices of the same instrument on different exchanges after adjusting these prices for my future predictions to then decide if and in which direction there is an arbitrage opportunity.

My algorithm also integrates sentiment prediction for individual stocks. I first created the most relevant words for each stock via tf-idf scores from given training media data. I then check for a media post if there is enough overlap between these words of a company and the words in the media post. If yes, I consider this post for the current company. I then apply a sentiment classification by assigning classification probabilities to the media post for the words "profit" and "loss" (instead of e.g. "good" and "bad" as this is more specific to the world of trading). The score of the highest probability must be above a threshold (0.8), otherwise it is considered noise and ignored. I then predict a 2 % (based on own experiences) price increase/decrease for a profit/loss post and adjust (multiply) these 2% by the classification score.

Built With

• python