Deep Learning for Financial Time Series: Does LSTM Outperform ARIMA and SVR in International Stock Market Predictions?

Time-series analysis and dynamic modeling are crucial in various fields, including business, economics, and finance. This study is based on the prediction of financial time series, which are known for their volatility, nonlinearity, and sensitivity to macroeconomic and psychological factors. This article examines four international stock market indices, such as MASI, S\&P 500, CAC 40, and Nikkei 225, representing Africa, America, Europe, and Asia, respectively, which are challenging to model accurately. This research aims to compare three forecasting models: the classical Autoregressive Integrated Moving Average (ARIMA), the machine learning (ML) model Support Vector Regression (SVR), and the deep learning (DL) model Long-Short-Term Memory (LSTM). The empirical results reveal that LSTM outperforms both SVR and ARIMA in predicting financial time series; SVR outperforms ARIMA in three indices: S\&P 500, CAC 40, and Nikkei 225. In contrast, ARIMA outperforms SVR in the MASI index, proving the effectiveness of this traditional method in specific contexts.