Improving Prediction of Bursa Malaysia Stock Index Using Time Series and Deep Learning Hybrid Model
The stock market is an important component of the financial world. Most of the stock market contains uncertainty and volatility leading to difficulty in predicting the future price of stocks and the market’s movement. The computing approach is a widely used technique in stock market forecasting that...
Saved in:
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Book Chapter |
| Language: | en |
| Published: |
Springer Cham
2023
|
| Subjects: | |
| Online Access: | http://ir.unimas.my/id/eprint/45996/2/Advances%20in%20Intelligent%20Computing%20Techniques%20and%20Applications.pdf http://ir.unimas.my/id/eprint/45996/ https://link.springer.com/chapter/10.1007/978-3-031-59711-4_11 https://doi.org/10.1007/978-3-031-59711-4_11 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The stock market is an important component of the financial world. Most of the stock market contains uncertainty and volatility leading to difficulty in predicting the future price of stocks and the market’s movement. The computing approach is a widely used technique in stock market forecasting that can assist the rapid and precise study of massive datasets. Existing studies have shown that such a technique can yield comparable or even better performances than traditional time series models in stock forecasting. Hybridizing both computing and traditional approaches lead to better performance since hybrid models utilized the advantage of each model. In this study, a hybrid model which combines autoregressive integrated moving average (ARIMA), generalized autoregressive conditional heteroskedasticity (GARCH) and long short-term memory (LSTM) model was proposed to forecast the closing price of Bursa Malaysia Kuala Lumpur Composite Index (KLCI). The proposed model operated by capturing the linear and volatility pattern from the time series model while the deep learning model handled the remaining non-linear pattern. The findings indicated an overall improvement of 13.32% for RMSE and 0.97% for MAE as compared to other benchmark models. The hybrid models can also forecast the actual data with a shorter computational time of 0.82% of that taken by the regular LSTM model. |
|---|