| Abstrakt: |
In the world of finance, every investment is intended to maximize profits and minimize related risks. Financial market predictions have been a challenging task for researchers due to the noise, complex, chaotic, volatile, dynamic, and non-parametric nature of time series data. In recent years, the application of deep learning methods has emerged as an attractive tool for market trend prediction. The development of an intelligent and effective prediction system holds significant potential for assisting investors, traders, and business analysts in making informed decisions and mitigating associated risks. With the objective of constructing a reliable prediction model and harnessing the benefits of multivariate analysis, we propose a novel multistep and multivariate deep learning-based hybrid VAR-LSTM (Vector Autoregressive-Long Short-Term Memory) framework for financial trend forecasting. The proposed approach combines the strengths of two distinct models: VAR, which is well-suited for capturing linear trends, and LSTM, known for effectively modeling nonlinear dynamics. To validate its effectiveness, the model was applied to three prominent companies listed on the American stock exchange, namely Amazon Inc., Apple Inc., and Microsoft Corp. Additionally, we have conducted a thorough technical analysis of the dataset and identified key signals for buy and sell decisions using moving averages. Experimental results show that the hybrid VAR-LSTM model outperforms the individual solo models in terms of minimizing the error. The superior performance of the hybrid model showcases its capability to effectively exploit the multivariate nature of stock market data, enabling more reliable and insightful predictions. |
Full text from SpringerLink