Analysis of financial time series with model hybridization

Abstract

Purpose- The aim of this study is to obtain better estimation results by hybridizing models that reveal linear and nonlinear relationships used in the intended financial time series. Methodology- ARIMA and Artificial Neural Networks (ANN) models were used in estimating NASDAQ stock market index values between 03.01.2012 and 30.06.2017 comparison of hybrid model results with different ways of error determination in literature. Findings- ARIMA residues have been tested different models where only residues are used with basic indications, only residues and basic. The calculation of residues was done separately with the addition and multiplication function. These residues were modeled with ANN, and the obtained results are collected and established hybrid model with ARIMA forecasts. When the results obtained at the end of the operations are compared, it is seen that the product function of some of the addition functions gives better results in some models. Conclusion- The hybridization of the ANN and NASDAQ index estimates with the ARIMA method resulted in processing for both addition and multiplication functions. Residues calculated with the addition model showed better results in ANN hybrid. What variables are used to calculate residuals is that the hybrid model gives better estimation results than single models.