3-Way Gated Recurrent Unit Network Architecture for Stock Price Prediction

Abstract

Stock price prediction has been the aim of stock investors since the beginning, which is important for the investors to make rational decisions about buying and selling stocks. Nowadays deep learning techniques and technical indicators are popular tools among researchers for predicting stock prices. Mainly researchers from the field of computer science, Statistics, and finance are actively involving in this research field. This research paper proposed a 3-way gated recurrent unit (3-GRU) architecture to forecast the next day's close price. The model is a combination of component GRU networks, where each component GRU network predicts the next day's close price using a different set of technical indicators. The proposed 3-GRU model was evaluated by comparing its performance with all component GRU networks. Its performance was also compared with the GRU network that combines different sets of the technical indicators into one feature vector uses it to predict the next day's close price.From the experimental results, we observed that the 3-GRU network architecture is able to predict the next day's close price with lower mean squared error and greater consistency than other models and hence concluded that it is the better approach for predicting next day's stock price.

Performance Evaluation of Stock Prediction Models Using Emagru

“…Several studies were the forerunners in applying machine learning to the challenge of stock prediction. RNN could remember the historical context in stock predictions (Saud & Shakya, 2021). Nabipour et al (2020) examined ten widely used technical indicators with several machine learning methods to determine which was superior for predicting stock prices.…”

“…The GRU variant outperformed LSTM and RNN. Saud and Shakya (2021) suggested that the 3-GRU model predicted the next day's closing price after comparing it with GRU-MACD, GRU-KST, GRU-ADX, and GRU-ALL. Radojičić and Kredatus (2020) used Fourier Transforms to extract new features and offered statistically significant improvements in GRU model performance.…”

Directional movement index based machine learning strategy for predicting stock trading signals

“…Researches related to stock market prediction can be basically divided into three categories: i) prediction of stock trading signals [6]- [13], ii) prediction of future stock price [14]- [31], and iii) prediction of the stock market index [16], [17], [30], [31]. In addition, the input features used by researchers to predict the stock market can be broadly divided into four categories: i) historical trading data that include open, high,  ISSN: 2088-8708 Int J Elec & Comp Eng, Vol.…”

“…Therefore, researchers used RNNs to predict stock price movements [38], [39]. Since RNNs suffer from vanishing/exploding gradient descent problem, nowadays long short-term memory (LSTM) and gated recurrent unit (GRU) networks are among the widely used models for stock forecasting [6], [7], [14], [23], [39]. However, none of the aforementioned strategies have utilized DMI indicators for trading signal prediction.…”