A Novel Stochastic–Deep Learning Hybrid: GRU with GBM for Financial Forecasting

Abstract

Financial time series are nonlinear, stochastic, and volatile which makes it a complicated task to accurately predict such data. The paper is a comparative study of the Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models in predicting complex financial time-series data, which has been augmented with the Geometric Brownian Motion (GBM) model. To determine the predictive efficiency and the strength of the new GBM-based LSTM and GRU frameworks, they are tested on several large-scale stock market datasets of India. Standard error and goodness-of-fit is used to test the performance of the models to prevent unreliable validation of models across datasets. The experimental findings indicate that the addition of GBM helps the LSTM and GRU models to work much better in terms of forecasting because of their ability to effectively learn stochastic price dynamics and volatility patterns that are inherent to financial markets. The GRU with GBM architecture also achieves the best performance across all datasets compared to the isolated LSTM and GRU architectures, making it more accurate and stable in its predictions. The fact that the performance of various stocks varies also indicates the flexibility of the hybrid approach in varying market conditions. The results overall demonstrate that the GRU-GBM hybrid model is a highly dependable and effective model of short-term financial forecasting, especially in highly unstable and uncertain market conditions. The given strategy can be considered a potential alternative to informed decision-making support in the use of investment analysis and risk management.