Time Series Analysis of Stock Index Forecasting

This thesis investigates stock market forecasting by comparing traditional time series models (ARIMA-GARCH), deep learning models (LSTM), and a hybrid ARIMA-LSTM model, focusing on the S&P 500, Euro Stoxx 50, and CSI 300 indices. Accuracy is evaluated using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), with findings indicating varying model effectiveness based on specific market data characteristics. The ARIMA-LSTM hybrid model excels in markets with strong autocorrelation, such as the S&P 500, demonstrating its ability to capture both linear and non-linear dynamics. In contrast, the single LSTM model is more effective in markets like the Euro Stoxx 50 and CSI 300, where weaker autocorrelation is present.