Title: A comparison of SVR and NARX in financial time series forecasting

Authors: Engin Tas; Ayca Hatice Atli

Addresses: Department of Statistics, Afyon Kocatepe University, Campus ANS, 03200, Afyonkarahisar, Turkey ' Department of Statistics, Afyon Kocatepe University, Campus ANS, 03200, Afyonkarahisar, Turkey

Abstract: Machine learning techniques have become attractive due to their robustness and superiority in predicting future behaviour in various areas. This paper is aimed to predict future stock prices by applying a nonlinear autoregressive network with exogenous inputs (NARX) and support vector regression (SVR). For this aim, we use the daily trade data, including highest price, lowest price, closing price, and trade volume for the stocks with the highest transaction volumes from Borsa Istanbul (BIST). In order to evaluate the performance of the prediction models, various statistical measures are used. The experimental results indicate that the techniques used are quite capable of predicting the future price of a stock. Moreover, both methods are competitive with each other and have superiorities in different aspects.

Keywords: artificial learning; artificial neural networks; financial time series forecasting; nonlinear autoregressive network with exogenous inputs; NARX; support vector regression; SVR.

DOI: 10.1504/IJCEE.2022.122835

International Journal of Computational Economics and Econometrics, 2022 Vol.12 No.3, pp.303 - 320

Received: 10 Sep 2020
Accepted: 26 Apr 2021

Published online: 13 May 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article