Prediction of the BIST 100 index direction with multiple logistic regression and k-nearest neighbors methods

In this study, we aim to predict daily up/down movement direction in the BIST100 index with the help of data mining techniques based on classification. The variables used for data analysis are historical return, volume, returns of the five previous trading days which are called Lag1 through Lag5 and daily closing return exchange rates. Our problem includes predicting whether on a particular day the BIST100 index will increase, i.e. fall into up bucket, or decrease, i.e. fall into down bucket. To predict the up/down movement direction of stock market, we apply two data mining techniques as binary logistic regression and K-nearest neighbors analysis. We compare the prediction performances of these two classification methods. According to data analysis results on BIST100, the binary logistic regression technique achieves better prediction performance in determining the up/down movement direction as opposed to K-nearest neighbors algorithm.