Stock price prediction using artificial neural networks on lowest price range data

Subject Areas : Stock Exchange

Bahman Ashrafijoo ¹ , Nasser Fegh-hi Farahmand ² , yagoub Alavi matin ³ , Kamaleddin Rahmani ⁴

1 - Department of Industrial management, Tabriz Branch, Islamic Azad University, Tabriz, Iran
2 - Department of Industrial Management, Tabriz Branch, Islamic Azad University, Tabriz, Iran
3 - Department of Industrial Management, Tabriz Branch, Islamic Azad University, Tabriz,, Iran
4 - Department of Industrial Management, Tabriz Branch, Islamic Azad University, Tabriz, Iran

Received: 2022-09-12 Accepted : 2022-12-28 Published : 2023-09-23

Keywords: Artificial Neural Network, Regression, Levenberg-Marquardt, Scaled Conjugate Gradient, Bayesian Regularization, Lowest price range date,

Abstract :

Today, one of the most important challenges in the capital market is stock price prediction. Stock price data represents a financial time series whose trend is very difficult to predict due to its characteristics and dynamic nature. One of the most recent methods used in predicting financial time series is ANN with back propagation of error. In this article, artificial neural networks based on three different Levenberg-Marquardt learning algorithms, scaled conjugate gradient and Bayesian regularization were used to predict the stock market based on the data of the lowest price range as well as the 30-minute data of the stock market index and compared their results together. We compare. All three algorithms provide a 99.9% estimate using the lowest price range data. But when using 30-minute data, the estimation accuracy decreases to 96.2%, 97.0%, and 98.9% for Levenberg-Marquarat algorithm, scaled conjugate gradient, and Bayesian regularization, respectively, which compares with the results Obtained by using the data of the lowest price range, the accuracy of the prediction is significantly reduced. Finally, the optimal neural network is compared with the regression method to determine that the results of the ANN in complex nonlinear time series are more efficient than linear methods.