THE EFFECT OF KERNEL IN SUPPORT VECTOR MACHINE TO FORECASTING PERFORMANCE OF FINANCIAL TIME SERIES

Authors : Aytaç ALTAN, Seçkin KARASU

Pages : 17-21

View : 16 | Download : 17

Publication Date : 2019-06-30

Article Type : Research Paper

Abstract :Nowadays, one of the most important research topics in economic sciences is the estimation of different financial exchange rates. The reliable and accurate forecasting of the exchange rate in the financial markets is of great importance, particularly after the recent global economic crises. In addition, the high accuracy forecasting of the financial exchange rates causes that investors are less affected by financial bubbles and crashes. In this paper, a financial time series forecasting model is identified by support vector machine insert ignore into journalissuearticles values(SVM);, which is one of the machine learning methods, for estimating the closing price of USD/TRY and EUR/TRY exchange rates. The closing price values and commodity channel index insert ignore into journalissuearticles values(CCI); indicator value are used as inputs in financial time series forecasting model. Various models are obtained with different kernel scale values in SVM and the model that estimates financial time series with the highest accuracy is proposed. The performance of the obtained models is measured by means of Pearson correlation and statistical indicators such as mean absolute error insert ignore into journalissuearticles values(MAE);, mean squared error insert ignore into journalissuearticles values(MSE);, and root mean squared error insert ignore into journalissuearticles values(RMSE);. It is seen that the forecasting performance of the proposed SVM model for the financial time series data set is higher than that the performance of the compared other models.   
Keywords : Financial Time Series, Support Vector Machine, Kernel Scale, Exchange Rate Forecasting