基于生成对抗网络的高频算法及其回测研究  

High frequency algorithm and its back-testing results based on GAN

作　　者：孟徐然 毕秀春 张曙光[1] MENG Xuran;BI Xiuchun;ZHANG Shuguang(School of Management, University of Science and Technology of China, Hefei 230026, China;School of Mathematics and Statistics, Guizhou University of Finance and Economics, Guiyang 550025, China)

机构地区：[1]中国科学技术大学管理学院,安徽合肥230026 [2]贵州财经大学数学与统计学院,贵州贵阳550025

出　　处：《中国科学技术大学学报》2020年第6期801-810,共10页JUSTC

基　　金：国家自然科学基金(11471304);贵州财经大学科研项目(2020YJ021,2020YJ026)资助.

摘　　要：在金融工程的分类任务中,由于金融数据噪音大、信息比率低的特点,传统深度算法的有监督训练模式往往过于依赖数据本身的绝对标签从而进一步放大了噪音对最终结果的影响.生成对抗网络(generative adversarial network,GAN)能够利用深度网络挖取数据特征,增强数据,减少噪音影响,应用于金融序列分析时效果优异.这里将GAN模型用于高频交易,具体做法为:将数据按波动性分为有标签数据与无标签数据两类,利用生成网络G与判别网络D互相对抗训练来深度学习这些数据的内在特性,训练好后的D网络根据有标签数据信息亦可对真实数据进行类别判别,得到涨跌分类模型,进而给出量化交易策略.基于期货主力合约数据进行了实证分析,结果表明,基于GAN训练下的LSTM模型显著优于有监督训练下的LSTM等深度模型和Logistics回归模型.In the financial classification mission,due to the big noise and low information-ratio in financial data,traditional supervised-learning regime may extend the noise influence because of the over dependent on the data label.GAN(generative adversarial network)can learn the data characters and reduce the influence of noise.When it is used to analyze the financial data,it has great results.We apply GAN to the high frequency trading:set the data labeled or unlabeled based on its volatility,then use the adversarial training between generative network G and discriminative network D to learn the intrinsic characters of the data,finally use the well trained D to get the up and down classification model and the quantization strategy.The sample is based on the future data,and the final results show that the LSTM model training by GAN is better than the deep learning models such as LSTM with supervised training and the Logistic regression model.

关 键 词：深度学习 生成对抗网络 涨跌分类模型 量化策略 

分 类 号：F830.9[经济管理—金融学]