Sentimentanalys för investeringsbeslut : Pressmeddelanden och aktieutveckling - Ett försök att slå börsen med hjälp av textklassificering

Sammanfattning: To invest your money in the stock market and successfully choose the stocks which will generate the biggest returns is desired by most. The amount of public stock market information is close to limitless and crucial to making solid investment decisions. One way for companies to supply the market with important information is through press releases. In this paper, it is examined to what extent stock price movements can be modeled from press releases with the use of machine learning. The machine learning model was trained and tested on historical data using a Support Vector Machine to predict stock price movements up or down. To evaluate the model performance indices, including precision, recall and accuracy, were used. Through these, no significant correlation between the sentiment of press releases and stock price movements could be determined. To examine the model further a trade simulation was performed to see how the model would manage an equity portfolio. To perform the simulation, and limit the impact of external factors, the time between opening and closing a trade was set to one hour. Although positive growth was achieved, the portfolio performed worse than the compared stock market indices. It is hard to, with certainty, establish why it performed worse, but to improve the simulation, different trade durations, the limit of simultaneous trades and the model’s parameters could be tuned.