2.765

2022影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

一种多源数据驱动的自动交易系统决策模型

文丹艳 马超群 王琨

文丹艳, 马超群, 王琨. 一种多源数据驱动的自动交易系统决策模型. 自动化学报, 2018, 44(8): 1505-1517. doi: 10.16383/j.aas.2018.c170563
引用本文: 文丹艳, 马超群, 王琨. 一种多源数据驱动的自动交易系统决策模型. 自动化学报, 2018, 44(8): 1505-1517. doi: 10.16383/j.aas.2018.c170563
WEN Dan-Yan, MA Chao-Qun, WANG Kun. A Multi-source Data Driven Decision Model for Automatic Trading Systems. ACTA AUTOMATICA SINICA, 2018, 44(8): 1505-1517. doi: 10.16383/j.aas.2018.c170563
Citation: WEN Dan-Yan, MA Chao-Qun, WANG Kun. A Multi-source Data Driven Decision Model for Automatic Trading Systems. ACTA AUTOMATICA SINICA, 2018, 44(8): 1505-1517. doi: 10.16383/j.aas.2018.c170563

一种多源数据驱动的自动交易系统决策模型

doi: 10.16383/j.aas.2018.c170563
基金项目: 

国家自然科学基金 71431008

国家自然科学基金 71521061

详细信息
    作者简介:

    马超群  湖南大学工商管理学院教授, 国家杰出青年基金获得者, 湖南省"芙蓉学者计划"特聘教授.主要研究方向为金融工程与风险管理.E-mail:cqma1998@hnu.edu.cn

    王琨  湖南大学工商管理学院硕士研究生, 2015年获得湖南大学学士学位.主要研究方向为金融工程与风险管理.E-mail:wagnkunhnu@163.com

    通讯作者:

    文丹艳  湖南大学工商管理学院博士研究生.主要研究方向为金融市场异象、投资策略.本文通信作者.E-mail:wdy_2008@hnu.edu.cn

A Multi-source Data Driven Decision Model for Automatic Trading Systems

Funds: 

National Natural Science Foundation of China 71431008

National Natural Science Foundation of China 71521061

More Information
    Author Bio:

     Professor in Business School of Hunan University. He is the winner of National Outstanding Youth Funds and the distinguished professor of Fu Rong Scholar Programme in Hunan Province. His research interest covers flnancial engineering and risk management

     Master student in Business School of Hunan University. He received his bachelor degree from Hunan University in 2015. His research interest covers flnancial engineering and risk management

    Corresponding author: WEN Dan-Yan  Ph. D. candidate in Business School of Hunan University. Her research interest covers flnancial market anomalies, trading strategies. Corresponding author of this paper
  • 摘要: 股票自动交易系统属于典型的复杂系统,其成功的关键是如何对股价进行有效的预测与决策.股价受多种信息的影响,但传统的自动交易模型多建立在历史交易数据的基础上.针对上述问题,本文综合利用新闻文本数据与股价技术指标数据,基于人工神经网络(Artificial neural netuorks,ANN)方法设计了一种多源数据驱动的股票自动交易决策模型.本文首先分析了各类财经新闻的特点及其对股价的影响,然后设计了相应模板抽取了中文文本中的财经新闻事件;在此基础上,设计了历史股价和新闻事件数据共同驱动的ANN-News模型,并利用实际数据验证了模型的有效性.实验发现,ANN-News模型比传统的机器学习类模型股价预测准确率提升约4%,收益率提升约7%.
    1)  本文责任编委 赵勇
  • 图  1  基于ANN模型的自动交易框架

    Fig.  1  Framework of automatic trading based on ANN

    图  2  4层的ANN (ANN-News)模型

    Fig.  2  4 layer ANN (ANN-News) model

    图  3  句法树关系

    Fig.  3  Parse tree

    图  4  金融事件统计

    Fig.  4  Count of financial events

    图  5  ANN模型准确率

    Fig.  5  Accuracy of ANN

    图  6  ANN模型准确率与收益率

    Fig.  6  The accuracy and return rate of ANN

    图  7  ANN与ANN-News模型准确率对比

    Fig.  7  Comparison of the accuracy between ANN and ANN-News

    图  8  ANN与ANN-News模型收益率对比

    Fig.  8  Comparison of the return between ANN and ANN-News

    表  1  金融事件抽取模板

    Table  1  Template of the extraction of financial events

    类型事件触发词($T_r$)方面词($A$)依存路径(Path)
    1股价股价上涨上/涨股票$E_s/E_c \xrightarrow{\rm SBV} T_r$
    股价下跌下/跌股价
    2业绩公司业绩上扬上/扬业绩
    公司业绩下跌下/跌
    3声誉公司声誉提升点赞捐赠、公益$E_s/E_c \xrightarrow{\rm ATT} A \xrightarrow{\rm SBV} T_r$
    公司声誉受损下滑口碑
    4利润公司利润上升上/升利润
    公司利润下滑下/滑
    5负债公司负债良好良好负债/债务$E_s/E_c \xrightarrow{\rm ATT} T_r \xrightarrow{\rm ATT} C$
    公司负债堪忧堪忧
    6高层公司高层变动变动/调整董事长、CEO$E_c \xrightarrow{\rm ATT} A \xrightarrow{\rm SBV} T_r$
    7业务公司业务扩张扩张业务
    公司合作合作公司($E_c^{(A)}, E_c^{(B)}$)$\{E_c^{(A)} + E_c^{(B)}\}\xrightarrow{\rm SBV} T_r$
    公司并购并购
    下载: 导出CSV

    表  2  技术指标信号介绍

    Table  2  Introduction of technical indicators

    指标公式买卖信号
    MACDEMA(12)-EMA(26)$+1:MACD_{t-1} \le 0~\&~MACD_t>0$
    $-1:MACD_{t-1} \ge 0\&MACD_t<0$
    ROC$P_t-P_{t-20}$$+1:ROC_{t-1} \le 1 ~\&~ ROC_{t} >1$
    $-1:ROC_{t-1} \ge 1 ~\&~ ROC_{t} < 1$
    TRB$P_t^{\rm Max}={\rm Max}(P_{t-1}, P_{t-2}, \cdots, P_{t-20})$
    $P_t^{\rm Min}={\rm Min}(P_{t-1}, P_{t-2}, \cdots, P_{t-20})$
    $+1:P_t>P_t^{\rm Max}$; $-1:P_t \ge P_t^{\rm Min}$
    A/D$\frac{P^{High}_t-P_{t-1}}{P^{High}_t-P^{Low}_{t-1}}$$+1:A/D_t\ge0;~-1:A/D_t<1$
    Dis$P_t/MA(60)$$+1:Dis_t \ge 1$; $-1:Dis_t<1$
    VMAMA(20)-MA(60)$+1:VMA_{t-1}<0~\&~VMA_t>0$
    $+1:VMA_{t-1}\ge 0~\&~VMA_t<0$}
    EMV$\frac{Mid_t-Mid_{t-1}}{BoxRatio_t}$, $Mid_{t}=\frac{P^{High}_t+P^{Low}_t}{2}$$+1:EMV_{t-1}\le 0 ~\&~ EMV_{t} > 0$
    $-1:EMV_{t-1}\ge 0EMV_t<1$
    CCI$\frac{M_t-SM_t}{0.015D_t}, $其中$M_t=\frac{P_t+P_t^{High}+P_t^{Low}}{3}$$+1: CCI_t <-200~or~CCI_t >CCI_{t-1}$
    $-1:CCI_t >200~or~CCI_t \le CCI_{t-1}$
    S/R$Sup(60)_t=MA(60)_t-2\sigma(60)_t$
    $Res(60)_t=MA(60)_t+2\sigma(60)_t$
    $+1:P_t>Sup(60)_t$; $-1:P_t < Res(60)_t$
    RSI$100-\frac{100}{1+(\sum^{n-1}_{i=0}Up_{t-i}/n)(\sum^{n-1}_{i=0}Dw_{i-1}/n)}$$+1:RSI_{t}<30~or~RSI_{t}>RSI_{t-1}$
    $-1:RSI_{t}>70~or~RSI_{t}\le RSI_{t-1}$
    注:其中, $P_t$表示$t$日的收盘价, $P^{High}_t$表示$t$日的最高价, $P^{Low}_t$表示$t$日的最低价. $MA(n)$为过去$n$天的简单移动平均计算为: $\sum^{n-1}_{i=0}p_{t-i}/n$, $EMA(n)$为过去$n$天($n=12/26$)的指数移动平均计算为: $P_t/n+(1-1/n)EMA(n)_{t-1}$, 其中$EMA(n)_0=P_1$, $\sigma(n)_t$为过去$n$天$(n=60)$收盘价的标准差计算为: $\sqrt{\sum^t_{i=t-n}(P_i-MA(n)_t)^2/n}$, $Up_t$是$t$日较上一期价格上涨的幅度(Upward-price-change), $Dw_t$是$t$日较上一期价格下降的幅度(Downward-price-change). $SM_t=\frac{\sum^n_{i=1}M_{t-i+1}}{n}$, $D_t=\frac{\sum^n_{i=1}|M_{t-i+1}-SM_t|}{n}$, $BoxRatio_t = Vol_t/1\, 000(P_t^{High}-P_t^{Low})$. $\pm1$代表买入/卖出信号, 其他情况表示持有.
    下载: 导出CSV

    表  3  财经事件与股价收益统计分析

    Table  3  Statistic on the relationship between financial events and stock returns

    事件IFreq.R0 (%)$d$$p$R1 (%) $d$$p$R2 (%)$d$$p$R5 (%)$d$$p$R10 (%) $d$$p$
    股价上涨219571.71840.011.32810.011.24750.011.19720.011.43700.00
    股价下跌-22039-0.92820.00-1.21770.01-1.21690.00-0.87670.03-0.70670.00
    公司业绩上扬37020.88650.000.78590.031.23580.021.14550.000.95530.02
    公司业绩下跌-2499-0.59540.04-0.87470.01-0.75510.30-0.87500.56-0.71480.42
    公司声誉提升24150.41630.050.44590.270.38560.370.29570.540.23490.15
    公司声誉受损-290-0.26510.05-0.39510.29-0.15500.05-0.13490.52-0.16470.04
    公司利润上升33440.41510.480.41490.180.59470.690.66510.010.79550.01
    公司利润下滑-3200-1.37660.24-0.88640.10-0.71580.11-0.49530.04-0.44570.00
    公司负债良好11890.13540.150.34510.47-0.06500.050.28470.080.23510.06
    公司负债堪忧-1125-0.15560.01-0.27550.08-0.66570.01-0.23500.11-0.25590.04
    公司高层变动1146-0.13510.190.08600.45-0.11500.040.25510.190.17570.16
    公司合作11370.16490.000.19500.360.05500.110.09560.320.14510.04
    公司业务扩张11470.24510.120.58490.15-0.45480.13-0.16490.240.46490.31
    公司并购3730.18540.020.29540.060.35500.160.11530.040.09540.50
    7 0636058555455
    下载: 导出CSV

    表  4  预定义影响力和收益$R_x$的关系

    Table  4  Relationship between predefined impact and $R_x$

    $R_x$$r$$p$
    $R_0$0.8050.000
    $R_1$0.8610.000
    $R_2$0.8210.000
    $R_5$0.8130.000
    $R_{10}$0.8010.001
    下载: 导出CSV

    表  5  数据集1技术信号与新闻信号收益统计

    Table  5  Statistics of the returns generated by technical and news signals on Dataset 1

    信号买入(%)卖出(%)
    $R_1$$R_2$$R_5$$R_1$$R_2$$R_5$
    MACD0.1460.1230.339-0.443-0.510-0.285
    ROC0.1510.2370.358-0.062-0.056-0.097
    A/D0.1030.1870.350-0.104-0.0240.309
    Dis0.0570.1290.199-0.008-0.0250.015
    VMA0.3300.6281.156-0.420-0.415-0.278
    EMV0.0290.1950.163-0.024-0.168-0.169
    TRB0.3450.4590.925-0.276-0.247-0.177
    CCI0.1900.2170.479-0.0540.0630.118
    SR0.0570.1290.200-0.005-0.0030.015
    RSI0.1350.1370.182-0.0170.1050.029
    新闻信号1.0190.8470.515-1.031-0.614-0.227
    下载: 导出CSV

    表  6  表现最佳的3组ANN参数组合

    Table  6  The best three combinations of ANN model

    $n_1$ $n_2$ $mc$数据集1 (%)数据集2 (%)
    训练集测试集训练集测试集
    150260.371.5370.6569.6567.38
    248300.173.7971.8968.2766.05
    349340.478.6072.0374.1471.77
    下载: 导出CSV

    表  7  ANN与ANN-News模型对比

    Table  7  Comparison between ANN and ANN-News

    $n_1$ $n_2$ $mc$ANN (%)ANN-News (%)
    训练集测试集训练集测试集
    150260.371.5370.6569.6567.38
    248300.173.7971.8968.2766.05
    349340.478.6072.0374.1471.77
    449340.769.1465.1778.0872.58
    下载: 导出CSV

    表  8  ANN模型与经典模型对比

    Table  8  The comparison among ANN and other classical models

    准确率收益率
    ANN0.682 (±0.04)1.306 (±0.06)
    ANN-News0.739(±0.02)1.403 (±0.03)
    SVM0.653 (±0.04)1.024 (±0.09)
    SVM-News0.692 (±0.06)1.067 (±0.12)
    Naïve Bayes0.641 (±0.06)1.102 (±0.05)
    Naïve Bayes-News0.687 (±0.08)1.281 (±0.08)
    下载: 导出CSV
  • [1] Ozturk M, Toroslu I H, Fidan G. Heuristic based trading system on Forex data using technical indicator rules. Applied Soft Computing, 2016, 43:170-186 doi: 10.1016/j.asoc.2016.01.048
    [2] Nuij W, Milea V, Hogenboom F, Frasincar F, Kaymak U. An automated framework for incorporating news into stock trading strategies. IEEE Transactions on Knowledge and Data Engineering, 2014, 26(4):823-835 doi: 10.1109/TKDE.2013.133
    [3] Wu Zhen, Wei Gang. One kind of optimal international security investment portfolio and consumption choice problem. Acta Automatica Sinica, 2003, 29(5):673-680 https://www.researchgate.net/publication/266997049_One_kind_of_optimal_control_problem_of_portfolio_and_consumption_choice_with_power_utility_function
    [4] 袁勇, 王飞跃.区块链技术发展现状与展望.自动化学报, 2016, 42(4):481-494 http://www.aas.net.cn/CN/Y2016/V42/I4/481

    Yuan Yong, Wang Fei-Yue. Blockchain:the state of the art and future trends. Acta Automatica Sinica, 2016, 42(4):481-494 http://www.aas.net.cn/CN/Y2016/V42/I4/481
    [5] 王飞跃.人工社会、计算实验、平行系统-关于复杂社会经济系统计算研究的讨论.复杂系统与复杂性科学, 2004, 1(4):25-35 doi: 10.3969/j.issn.1672-3813.2004.04.002

    Wang Fei-Yue. Artificial societies, computational experiments, and parallel systems:a discussion on computational theory of complex social-economic systems. Complex Systems and Complexity Science, 2004, 1(4):25-35 doi: 10.3969/j.issn.1672-3813.2004.04.002
    [6] 崔凯楠, 郑晓龙, 文丁, 赵学亮.计算实验研究方法及应用.自动化学报, 2013, 39(8):1157-1169 http://www.aas.net.cn/CN/abstract/abstract18148.shtml

    Cui Kai-Nan, Zheng Xiao-Long, Wen Ding, Zhao Xue-Liang. Researches and applications of computational experiments. Acta Automatica Sinica, 2013, 39(8):1157-1169 http://www.aas.net.cn/CN/abstract/abstract18148.shtml
    [7] Merrill A A. Behavior of Prices on Wall Street. New York:Analysis Press, 1966:15-31.
    [8] 郭欣, 王蕾, 宣伯凯, 李彩萍.基于有监督Kohonen神经网络的步态识别.自动化学报, 2017, 43(3):430-438 http://www.aas.net.cn/CN/Y2017/V43/I3/430

    Guo Xin, Wang Lei, Xuan Bo-Kai, Li Cai-Ping. Gait recognition based on supervised Kohonen neural network. Acta Automatica Sinica, 2017, 43(3):430-438 http://www.aas.net.cn/CN/Y2017/V43/I3/430
    [9] 秦楚雄, 张连海.基于DNN的低资源语音识别特征提取技术.自动化学报, 2017, 43(7):1208-1219 http://www.aas.net.cn/CN/Y2017/V43/I7/1208

    Qin Chu-Xiong, Zhang Lian-Hai. Deep neural network based feature extraction for low-resource speech recognition. Acta Automatica Sinica, 2017, 43(7):1208-1219 http://www.aas.net.cn/CN/Y2017/V43/I7/1208
    [10] Dash R, Dash P K. A hybrid stock trading framework integrating technical analysis with machine learning techniques. Journal of Finance and Data Science, 2016, 2(1):42-57 doi: 10.1016/j.jfds.2016.03.002
    [11] Hsu M W, Lessmann S, Sung M C, Ma T J, Hohnson J E V. Bridging the divide in financial market forecasting:machine learners vs. financial economists. Expert Systems with Applications, 2016, 61:215-234 doi: 10.1016/j.eswa.2016.05.033
    [12] Micola A R. Reviewed work:handbook of computational economics, Vol. 2:agent-based computational economics by L. Tesfatsion, K. L. Jud. Interfaces, 2007, 37(3):303-304
    [13] Wang L X. Modeling stock price dynamics with fuzzy opinion networks. IEEE Transactions on Fuzzy Systems, 2017, 25(2):277-301 doi: 10.1109/TFUZZ.2016.2574911
    [14] Wang L X. Dynamical models of stock prices based on technical trading rules part I:the models. IEEE Transactions on Fuzzy Systems, 2015, 23(4):787-801 doi: 10.1109/TFUZZ.2014.2327994
    [15] Wang L X. Dynamical models of stock prices based on technical trading rules-part Ⅱ:analysis of the models. IEEE Transactions on Fuzzy Systems, 2015, 23(4):1127-1141 doi: 10.1109/TFUZZ.2014.2346244
    [16] Wang L X. Dynamical models of stock prices based on technical trading rules-part Ⅲ:application to Hong Kong stocks. IEEE Transactions on Fuzzy Systems, 2015, 23(5):1680-1697 doi: 10.1109/TFUZZ.2014.2374193
    [17] Kimoto T, Asakawa K, Yoda M, Takeoka M. Stock market prediction system with modular neural networks. In: Proceedings of 1990 IJCNN International Joint Conference on Neural Networks. San Diego, CA, USA: IEEE, 1990.
    [18] Saad E W, Prokhorov D V, Wunsch D C. Comparative study of stock trend prediction using time delay, recurrent and probabilistic neural networks. IEEE Transactions on Neural Networks, 1998, 9(6):1456-1470 doi: 10.1109/72.728395
    [19] Dhamija A K, Bhalla V K. Exchange rate forecasting:comparison of various architectures of neural networks. Neural Computing and Applications, 2011, 20(3):355-363 doi: 10.1007/s00521-010-0385-5
    [20] Ballings M, van den Poel D, Hespeels N, Gryp R. Evaluating multiple classifiers for stock price direction prediction. Expert Systems with Applications, 2015, 42(20):7046-7056 doi: 10.1016/j.eswa.2015.05.013
    [21] Rosen R J. Merger momentum and investor sentiment:the stock market reaction to merger announcements. The Journal of Business, 2006, 79(2):987-1017 doi: 10.1086/jb.2006.79.issue-2
    [22] Zhang X F. Information uncertainty and stock returns. The Journal of Finance, 2006, 61(1):105-137 doi: 10.1111/j.1540-6261.2006.00831.x
    [23] Schumaker R P, Zhang Y L, Huang C N, Chen H. Evaluating sentiment in financial news articles. Decision Support Systems, 2012, 53(3):458-464 doi: 10.1016/j.dss.2012.03.001
    [24] Yuan Y. Market-wide attention, trading, and stock returns. Journal of Financial Economics, 2015, 116(3):548-564 doi: 10.1016/j.jfineco.2015.03.006
    [25] Hong K S, Dennis A R, Yuan L Y. Trading on twitter:using social media sentiment to predict stock returns. Decision Sciences, 2017, 48(3):454-488 doi: 10.1111/deci.2017.48.issue-3
    [26] Pagolu V S, Reddy K N, Panda G, Majhi B. Sentiment analysis of twitter data for predicting stock market movements. In: Proceedings of 2016 International Conference on Signal Processing, Communication, Power and Embedded System (SCOPES). Paralakhemundi, India: IEEE, 2016.
    [27] Kara Y, Acar Boyacioglu M, Baykan Ö B. Predicting direction of stock price index movement using artificial neural networks and support vector machines:the sample of the istanbul stock exchange. Expert Systems with Applications, 2011, 38(5):5311-5319 doi: 10.1016/j.eswa.2010.10.027
    [28] Patel J, Shah S, Thakkar P, Kotecha K. Predicting stock and stock price index movement using Trend Deterministic Data Preparation and machine learning techniques. Expert Systems with Applications, 2015, 42(1):259-268 doi: 10.1016/j.eswa.2014.07.040
    [29] 赵妍妍, 秦兵, 车万翔, 刘挺.中文事件抽取技术研究.中文信息学报, 2008, 22(1):3-8 doi: 10.3969/j.issn.1003-0077.2008.01.001

    Zhao Yan-Yan, Qin Bing, Che Wan-Xiang, Liu Ting. Research on Chinese event extraction. Journal of Chinese Information Processing, 2008, 22(1):3-8 doi: 10.3969/j.issn.1003-0077.2008.01.001
    [30] Chen Z, Ji H. Language specific issue and feature exploration in Chinese event extraction. In: Proceedings of Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the Association of Computational Linguistics. Boulder, Colorado, USA: ACM, 2009: 209-212
    [31] 胡博磊, 贺瑞芳, 孙宏, 王文俊.基于条件随机域的中文事件类型识别.模式识别与人工智能, 2012, 25(3):445-449 doi: 10.3969/j.issn.1003-6059.2012.03.012

    Hu Bo-Lei, He Rui-Fang, Sun Hong, Wang Wen-Jun. Chinese event type recognition based on conditional random fields. Pattern Recognition and Artificial Intelligence, 2012, 25(3):445-449 doi: 10.3969/j.issn.1003-6059.2012.03.012
    [32] 廖祥文, 陈兴俊, 魏晶晶, 陈国龙, 程学旗.基于多层关系图模型的中文评价对象与评价词抽取方法.自动化学报, 2017, 43(3):462-471 http://www.aas.net.cn/CN/Y2017/V43/I3/462

    Liao Xiang-Wen, Chen Xing-Jun, Wei Jing-Jing, Chen Guo-Long, Cheng Xue-Qi. A multi-layer relation graph model for extracting opinion targets and opinion words. Acta Automatica Sinica, 2017, 43(3):462-471 http://www.aas.net.cn/CN/Y2017/V43/I3/462
  • 加载中
图(8) / 表(8)
计量
  • 文章访问数:  2560
  • HTML全文浏览量:  542
  • PDF下载量:  754
  • 被引次数: 0
出版历程
  • 收稿日期:  2017-10-05
  • 录用日期:  2018-05-04
  • 刊出日期:  2018-08-20

目录

    /

    返回文章
    返回