基于深度学习与信号分解的金融交易决策支持模型  

A Financial Trading Decision-making Support Model Using Deep Learning and Signal Decomposition

作　　者：刘敏 张凡 王林[3] 朱青 Liu Min;Zhang Fan;Wang Lin;Zhu Qing(School of Economics&Management,Nanchang University,Nanchang 330000;School of Management,Xi'an Jiaotong University,Xi'an 710049;School of Management,Huazhong University of Science&Technology,Wuhan 430074;International Business School,Shaanxi Normal University,Xi'an 710061)

机构地区：[1]南昌大学经济管理学院,南昌330000 [2]西安交通大学管理学院,西安710049 [3]华中科技大学管理学院,武汉430074 [4]陕西师范大学国际商学院,西安710061

出　　处：《管理评论》2022年第9期14-26,共13页Management Review

基　　金：国家社会科学基金一般项目(21BTJ028);国家自然科学基金项目(92146005)。

摘　　要：本文提出一种数据表征方式以反映金融市场状态,并根据该表征提出一个短期交易决策支持模型。通过信号分解技术,该模型将一维非平稳时间序列分解为多维平稳子序列,并进一步将子序列重构为二维图像矩阵以表征每日市场状态。在此基础上,该模型利用神经网络的特征学习能力捕捉五日连续窗口的最优决策点。通过统计性能和财务业绩两种评估方式以及模型对比,结果表明本文所提出的短期交易决策模型有很强的应用性和适应性,可在多变的市场环境中获得可观利润。This paper presents a decision support model for algorithmic trading in the financial market,which utilizes a novel hybrid model to make automatic trading decision.The proposed model hierarchically represents the one-dimensional non-stationary time series into multi-dimensional stationary sub-series,and then restructures them into a two-dimensional matrix to represent the daily market state.Then,the neural network,which has a powerful features learning ability,is used to capture the optimal entry and exit points of the fluctuating stock prices.Finally,the actual predicted results are evaluated by two different ways:statistical performance evaluation and financial performance evaluation.The results show that the proposed decision support model in this paper has strong applicability and adaptability which can bring positive profits in multiple environments.

关 键 词：序列分解 神经网络 决策模型 算法交易 

分 类 号：F832.51[经济管理—金融学] F224