Year: 2025
Type: conference-paper
Abstract: This paper introduces a model called MSC-BGRU for intelligent quantitative financial decision-making and risk control, aiming to address the challenges of complexity and volatility in modern financial markets. The model integrates multi-scale Convolutional Neural Networks (CNNs) and Bidirectional Gated Recurrent Units (BiGRU) to effectively process and extract features from multivariate time series data. Initially, the input data is divided into time windows and embedded, followed by efficient feature extraction using a multi-scale hybrid CNN network, capturing interaction features across different time scales. Next, a bidirectional GRU is employed to simultaneously extract forward and backward sequence features, enabling a better understanding of market dynamics. Finally, a module composed of a Multi-Layer Perceptron (MLP) models the physical relationships between variables, reflecting the interactions of cross-variable features. Experimental results show that, compared to various advanced models including Transformer and LSTM, MSC-BGRU achieves higher accuracy and stronger robustness in five-class and ten-class tasks on two datasets. Furthermore, ablation experiments further validate the importance of each module in improving model performance. This research provides new perspectives and technological means for intelligent quantitative financial decision-making and risk management.
Authors: Qianhao Lu, Minghao Zhang, Pinyu Huang, Mingxuan Jiang
Institution: Sanjiang University
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W4412887437