Understanding the characteristics of financial time series through neural network and SVM approaches

Exchange rate has been always a focal point for researchers within international scope. Globalisation and the role of exchange rate create a challenging market where short-term prediction is concerned. The ability to predict the exchange rate is a challenging topic for professionals and practitioners. This paper proposes a method to address the current issues of predicting the market changes using characteristics of financial time series. The main idea is that neural network and support vector machine (SVM) approaches are employed to train and test the results in different instances. Findings indicate the superiority of correct sets over incorrect, while criteria sets had been sometimes better results. Furthermore, linear kernel was more likely to encounter convergence problems than other types which oppose to primary dataset. Finally, the accuracy of the proposed prediction methods is analysed and compared with related works.