Abstract:

Predicting the direction (increase or decrease) of stock market indexes and stock prices has long attracted the attention of investors and researchers. Establishing a connection between the past and future data makes this prediction difficult. The mentioned connection is established through econometric or neural network models. Neural network models do not require strict assumptions like econometric models and can utilize qualitative and quantitative data. In this study, monthly average BIST 100 index values were taken between January 2002 and September 2022, and a two-group dependent variable was formed as “1” for cases with an increase compared to the previous month and “0” for cases with a decrease. The 1st and 2nd lagged values of BIST 100, S&P 500, CAC40, FTSE10, NIKKEI225DAX, SHANGAICOMP, ONSUSD, USDTRY, VIX and REPO variables were taken as independent variables. A total of nine different machine learning methods which are Logistic Regression Analysis (LR), Linear Discriminant Analysis (LDA), Naive Bayes Algorithm (NB), Random Forests Algorithm (RF), K-Nearest Neighborhood Algorithm (KNN), Classification and Regression Trees Algorithm (CART), Artificial Neural Networks (NNET), Support Vector Machines with Radial Basis Function (SVM-RBF), Support Vector Machines with Polynomial Kernel Function (SVM-POLY) were used for direction prediction for BIST 100 index in practice. In conclusion, it is observed that linear methods produce more successful estimation results.

Keywords: