The accurate prediction of movements in the stock market over time is of major interest for investors and governments alike. In this research, we aimed to classify the returns of a stock market, meaning the daily changes, into four categories. Instead of simply using up and down movements, we consider also whether these movements are strong or small. Our target is the well-known American stock market index S&P500 and we collected more than 130 inputs such as other related market indices and several so-called technical indicators from finance, to build our stock market classification model. Initially, we selected with feature selection from all features those that help us set up a good classification model.
As our model, we chose a Random forest, which is a popular machine learning algorithm. Subsequently, we derived several trading strategies from our classification results to profit from our predictions. One of our main findings is that the strong movements, positive and negative, contribute on average most to our trading strategies, meaning that they bring larger returns than other predictions. Since most previous models only consider two classes, whereas we use four, this indicates that differentiating among more outcomes can be beneficial for stock market predictions.
Lohrmann,C. & Luukka,P. (2019). Classification of intraday S&P500 returns with a Random Forest.International Journal of Forecasting. Volume 35. Issue. 1. p.390-407.
