Forecasting the development trend of the green finance market based on long short-term memory and transformer models

With the continuous development of the global green finance market, accurately forecasting its trends has become a critical issue in financial research. Traditional time series forecasting methods are commonly used in this domain, but when applied to complex green finance data, they often suffer from low robustness, weak generalization ability, and difficulty in simultaneously capturing global trends and short-term fluctuations. To address these limitations, this paper proposes a task-oriented hybrid deep learning model, TranGRU, which combines Transformer and Gated Recurrent Unit (GRU) architectures for green finance market trend forecasting. The Transformer's global attention mechanism extracts long-term market dependencies, while the GRU's temporal memory capability captures short-term fluctuations, thereby enhancing the model's robustness and generalization in complex data environments. Experimental results indicate that TranGRU achieves comparatively better forecasting performance than several baseline models under the current experimental setting, particularly in terms of prediction error, convergence behavior, and directional consistency. The main contribution of this work lies not in introducing a fundamentally new sequence architecture, but in developing and systematically validating a multimodal forecasting framework tailored to the green finance context, where long-term policy signals and short-term market dynamics need to be modeled jointly. These findings provide empirical evidence that hybrid temporal modeling and multimodal feature fusion may be useful for short-horizon green finance market forecasting. The proposed framework may therefore serve as a supportive analytical tool for policy assessment and investment-related decision-making, while its broader applicability requires further validation across different markets and data conditions.