Analyzing Market shocks with a Hybrid ARMA-GARCH and Deep Learning Model for Enhanced Forecasting

Abstract

Objective: The primary aim of the present study is to provide a model for forecasting stock market shocks using a deep learning approach in the Tehran Stock Exchange.   Methodology: The current research is applied in terms of its objective, meaning it seeks to solve a specific real-world problem, namely, the prediction of stock market shocks. In terms of nature, this research is descriptive-analytical, as its goal is to examine and analyze market fluctuations over time and it utilizes statistical methods for data analysis. For this purpose, intraday 15-minute data of the overall index in the time interval from March 20, 2018, to December 27, 2018, which includes the opening, closing, highest, and lowest values of the aforementioned index, were used. In the proposed model, statistical information from the time series is first extracted using the ARMA-GARCH model and then, using the deep learning method and one-dimensional convolutional neural networks (CNN), the nonlinear relationships of the time series observations are modeled. Finally, the appropriate models are selected, and based on the criteria of mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE), the prediction accuracies are evaluated.   Findings: The results of this study indicated that the combined ARMA-GARCH-CNN model, leveraging the strengths of both traditional and modern methods, was capable of accurately forecasting market shocks.   Conclusion: This research, focusing on the Iranian stock market and utilizing intraday data from the Tehran Stock Exchange, has specifically and accurately addressed the prediction of Iran's market shocks. This localization of the model for the Iranian market significantly increased the accuracy of the results and enabled the simulation of the specific conditions of this market. Overall, this study presented an innovative hybrid model for analyzing the Iranian stock market, which can be beneficial for future analyses in similar markets.