Using big data and statistical machine learning methods in investment decisions in financial markets

This paper aims to forecast the daily maximum prices of the BİST100 index. In this context, the target is to develop investment strategies from models created by using machine learning methods. Forecasting of the BİST100 index, which can be considered as time series analysis, has an important role in making financial decisions and generating returns with investment strategies formed by these decisions. There are many studies in the literature on the prediction of asset prices in capital markets using machine learning methods. The forecast of the daily trend and closing price are generally areas of application. In this study, the investment strategy model is differentiated both by developing the machine learning method to be used and by expanding the data set by using intraday maximum price. In order to forecast the highest intraday price of the BİST100 index, four and five years of data sets were used. Thus, the power of the model was tested with two separate data sets. In predicting the index, historical data of the index were used with delay as a predictor data. In addition, the predictor data set includes economic and external factors such as exchange rate, interest rates, commodity, and CDS. In this study, the content was expanded by using two different machine learning techniques. These are Artificial Neural Networks and XGBoost models that are prominent in the literature. In the comparison of models, many performance criteria were used which are commonly used in literature. Some of those; Mean Absolute Error (MAE), Signal Accuracy (SA), Mean Square Error (MSE) and Root Mean Square Error (RMSE). As a result of the empirical study, it was found that the XGBoost method showed better performance than the Artificial Neural Networks in this situation. The investment strategies created within the scope of the study were formed through with the obtained model performances. The investment strategies created with XGBoost have been found that give better results than passive investment for the same period of time. In the experiments to increase the predictive power of the models, it was observed that adding the opening price of the same day to the predictor data set increased the performance of the models.