Machine learning approach for the prediction of stock index price in Saudi Arabia: An investigation

The prediction of stock index price is vital issue in financial market research. Index performance is a subject to different investigation due to its importance to several stakeholders including markets players and policy maker [1, 2]. Market players for example, track performance of stock index due to plan and execute investment and funding decision. Funds managers also utilize index performance as benchmark for their investment performance. Moreover, policy maker is also interested to understand dynamic and movement of stock market index as discovery tool. In Saudi Arabia, Tadawul All Share Index (TASI) is most important index for the Saudi equity market as well as the exchanges in the MENA region. Currently, the index traces the performance of the biggest markets in the Arab region. Traditionally, oil price is considered the most important index determinant [3, 4]. However, the prediction of TASI index might be still questionable in particular after economic reform in the last 5 years which might change dependence of Saudi equity markets to Oil price. In the financial sector, artificial intelligence (AI) including machine learning and deep learning are transformative technologies which disrupts many sectors but offer numerous potential benefits. The forecasts estimate that, by incorporating AI in its operations, the financial services sector can save a billion dollars by 2030 [5]. Prior studies adopted several models to predict the future values of TASI [6]. Almost all applied the traditional econometric models to predict the future index price. Only few studies attempted to test the Saudi market index using the AI, ML, or DL algorithms [7, 8, 9]. For the example, the aim of the study [4] is to forecast the Saudi Stock Market Index by using its previous values and the impact of peoples sentiments on their financial decisions. The Global Data on Events, Location, and Tone (GDELT) dataset has been used to extraction two-time series filtered for Saudi Arabia news. These time series reflect the everyday values of tone and social media attention. The features of the multivariate time series produced were then examined and multiple multivariate models are compared to each other to predict the daily index of the Saudi stock market. Usually, previous studies adopted traditional machine learning methods for TASI prediction. For example, Support Vector Machines, Nave Bayes, K-Nearest Neighbors etc were used for TASI prediction [Using Machine Learning Classifiers to Predict Stock Exchange Index,]. Moreover, model and feature fusion has not be considered in previous studies, up to our knowledge. Furthermore, previous literature did not emphasize importance of the macroeconomic factors and technical analysis indicators as additional features to build better machine learning models to improve the index price prediction. Therefore, the current study intends to fill this gap by conducting an investigation study that is going to adopt several markets, economic and index features and predict the index price of TASI. We intend to combine such information at feature space Moreover, ensemble machine learning methods will be considered and will be compared to the fusion of multiple machine learning methods.