基于深度学习的限价订单簿中间价格趋势预测方法

金融市场上的交易系统一般都采用订单驱动的方式来进行，投资者可以提交 以市价或指定价格成交的订单，未成交的限价订单则堆积在订单队列中形成限价 订单簿（Limited Order Book，LOB）。由于互联网上的在线交易模式日益成熟，限 价订单簿中的高频股票交易正在全球市场以非常快的速度发展，加上低频交易数 据分析已不能够充分地反映金融市场上的交易信息，因此对高频限价订单簿价格 趋势预测是一个在近年研究中越来越受关注的热门话题。由于传统时间序列分析 方法在处理高频交易数据时存在参数设置困难和不准确的问题，学者们提出了基 于深度学习的分析方法来有效地应对高频股票数据的敏感性，他们通过人工精巧 的神经网络设计来提取数据特征完成股票预测。本文主要研究的是利用深度学习 预测限价订单簿里面高频股票数据的中间价格移动趋势。首先，本文利用一维卷 积神经网络和 Transformer 模型设计出了新模型 OCET，针对买卖十档价格交易的 趋势预测，同样的推理速度下，OCET 在 FI-2010 数据集五只股票短期预测的不同 时刻的平均预测准确率提升 7%。同时，本文从神经网络模型压缩的角度来探索高 频交易模型在实际部署中的低延迟、小存储的可行性，为投资者和市场监管者提 供一定的价值参考。

[NTAKARIS A, MAGRIS M, KANNIAINEN J, et al. Benchmark dataset for mid-price forecasting of limit order book data with machine learning methods[J]. Journal of Forecasting, 2018,37(8): 852-866.
[2] CAO C, HANSCH O, WANG X. The information content of an open limit-order book[J].Journal of Futures Markets: Futures, Options, and Other Derivative Products, 2009, 29(1): 16-41.
[3] MADHAVAN A. Market microstructure: A survey[J]. Journal of Financial Markets, 2000, 3(3): 205-258.
[4] O’HARA M. Market microstructure theory[M]. John Wiley & Sons, 1998.
[5] BARUCH S. Who benefits from an open limit-order book?[J]. The Journal of Business, 2005,78(4): 1267-1306.
[6] ANAND A, CHAKRAVARTY S, MARTELL T. Empirical evidence on the evolution of liquidity: Choice of market versus limit orders by informed and uninformed traders[J]. Journal ofFinancial Markets, 2005, 8(3): 288-308.
[7] ANAND A, WEAVER D G. Can order exposure be mandated?[J]. Journal of Financial Markets,2004, 7(4): 405-426.
[8] 陶振毅. 量化交易在股市中的应用[J]. 中国外资, 2018: 84-86.
[9] 陈艳, 褚光磊. 股指期货套利交易的风险度量——基于沪深 300 股指期货交易数据的实证分析[J]. 管理现代化, 2014, 34: 86-88.
[10] HOFFMANN P. A dynamic limit order market with fast and slow traders[J]. Journal of FinancialEconomics, 2014, 113(1): 156-169.
[11] LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436-444.
[12] MITCHELL T M, et al. Machine learning: volume 1[M]. McGraw-hill New York, 2007.
[13] HIRSCHBERG J, MANNING C D. Advances in natural language processing[J]. Science, 2015,349(6245): 261-266.
[14] LARSSON M, ZHANG Y, KAHL F. Robust abdominal organ segmentation using regionalconvolutional neural networks[J]. Applied Soft Computing, 2018, 70: 465-471.
[15] TITZE I R, MARTIN D W. Principles of voice production[M]. Acoustical Society of America,1998.
[16] KERCHEVAL A N, ZHANG Y. Modelling high-frequency limit order book dynamics withsupport vector machines[J]. Quantitative Finance, 2015, 15(8): 1315-1329.
[17] ATSALAKIS G S, VALAVANIS K P. Surveying stock market forecasting techniques–Part II:Soft computing methods[J]. Expert Systems with Applications, 2009, 36(3): 5932-5941.52参考文献
[18] CAO Q, LEGGIO K B, SCHNIEDERJANS M J. A comparison between Fama and French’smodel and artificial neural networks in predicting the Chinese stock market[J]. Computers &Operations Research, 2005, 32(10): 2499-2512.
[19] GLEZAKOS M, MYLONAKIS J, KAFOUROS C. The impact of accounting information onstock prices: Evidence from the Athens Stock Exchange[J]. International Journal of Economicsand Finance, 2012, 4(2): 56-68.
[20] TSANTEKIDIS A, PASSALIS N, TEFAS A, et al. Forecasting stock prices from the limitorder book using convolutional neural networks[C]//Conference on Business Informatics. IEEE,2017: 7-12.
[21] O’SHEA K, NASH R. An introduction to convolutional neural networks[A]. 2015.
[22] GREFF K, SRIVASTAVA R K, KOUTNÍK J, et al. LSTM: A search space odyssey[J]. IEEETransactions on Neural Networks and Learning Systems, 2016, 28(10): 2222-2232.
[23] TRAN D T, IOSIFIDIS A, KANNIAINEN J, et al. Temporal attention-augmented bilinearnetwork for financial time-series data analysis[J]. IEEE Transactions on Neural Networks andLearning Systems, 2018, 30(5): 1407-1418.
[24] ZHANG Z, ZOHREN S, ROBERTS S. Deeplob: Deep convolutional neural networks for limitorder books[J]. IEEE Transactions on Signal Processing, 2019, 67(11): 3001-3012.
[25] SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C]//Proceedings of the IEEE conference on Computer Vision and Pattern Recognition. 2016: 2818-2826.
[26] LUO W, YU F. Recurrent highway networks with grouped auxiliary memory[J]. IEEE Access,2019, 7: 182037-182049.
[27] CHUNG J, GULCEHRE C, CHO K, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[A]. 2014.
[28] SANGADIEV A, RIVERA-CASTRO R, STEPANOV K, et al. DeepFolio: Convolutional Neural Networks for Portfolios with Limit Order Book Data[A]. 2020.
[29] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedingsof the IEEE onference on Computer Vision and Pattern Recognition. 2016: 770-778.
[30] LIU G, MAO Y, SUN Q, et al. Multi-scale Two-way Deep Neural Network for Stock Trend Prediction.[C]//Proceedings of the IEEE conference on International Joint Conference on ArtificialIntelligence. 2020: 4555-4561.
[31] ZHANG Z, ZOHREN S. Multi-Horizon Forecasting for Limit Order Books: Novel Deep Learning Approaches and Hardware Acceleration using Intelligent Processing Units[A]. 2021.
[32] BIAIS B, FOUCAULT T, MOINAS S. Equilibrium fast trading[J]. Journal of Financial Economics, 2015, 116(2): 292-313.
[33] CHORDIA T, ROLL R, SUBRAHMANYAM A. Liquidity and market efficiency[J]. Journalof Financial Economics, 2008, 87(2): 249-268.
[34] BARUCH S, PANAYIDES M, VENKATARAMAN K. Informed trading and price discoverybefore corporate events[J]. Journal of Financial Economics, 2017, 125(3): 561-588.53参考文献
[35] KANIEL R, LIU H. So what orders do informed traders use?[J]. The Journal of Business, 2006,79(4): 1867-1913.
[36] SAAR G. Price impact asymmetry of block trades: An institutional trading explanation[J]. TheReview of Financial Studies, 2001, 14(4): 1153-1181.
[37] FOUCAULT T. Order flow composition and trading costs in a dynamic limit order market[J].Journal of Financial markets, 1999, 2(2): 99-134.
[38] GLOSTEN L R. Is the electronic open limit order book inevitable?[J]. The Journal of Finance,1994, 49(4): 1127-1161.
[39] PARLOUR C A. Price dynamics in limit order markets[J]. The Review of Financial Studies,1998, 11(4): 789-816.
[40] PARLOUR C A, SEPPI D J. Liquidity-based competition for order flow[J]. The Review ofFinancial Studies, 2003, 16(2): 301-343.
[41] CHORDIA T, ROLL R, SUBRAHMANYAM A. Order imbalance, liquidity, and market returns[J]. Journal of Financial Economics, 2002, 65(1): 111-130.
[42] FRIEDERICH S, PAYNE R. Dealer liquidity in an auction market: Evidence from the LondonStock Exchange[J]. The Economic Journal, 2007, 117(522): 1168-1191.
[43] HUANG Y C, TSAI P L. Effectiveness of closing call auctions: evidence from the Taiwan StockExchange[J]. Emerging Markets Finance and Trade, 2008, 44(3): 5-20.
[44] BOEHMER E, SAAR G, YU L. Lifting the veil: An analysis of pre-trade transparency at theNYSE[J]. The Journal of Finance, 2005, 60(2): 783-815.
[45] AMAYA D, FILBIEN J Y, OKOU C, et al. Distilling liquidity costs from limit order books[J].Journal of Banking & Finance, 2018, 94: 16-34.
[46] ARMANO G, MARCHESI M, MURRU A. A hybrid genetic-neural architecture for stock indexes forecasting[J]. Information Sciences, 2005, 170(1): 3-33.
[47] KLASSEN M. Investigation of Some Technical Indexes in Stock Forecasting Using NeuralNetworks.[C]//World Engineers Convention. Citeseer, 2005: 75-79.
[48] HASSAN M R, NATH B. Stock market forecasting using hidden Markov model: a new approach[C]//Intelligent Systems Design and Applications. IEEE, 2005: 192-196.
[49] TANAKA-YAMAWAKI M, TOKUOKA S. Adaptive use of technical indicators for the prediction of intra-day stock prices[J]. Physica A: Statistical Mechanics and its Applications, 2007,383(1): 125-133.
[50] MAJHI R, PANDA G, SAHOO G, et al. Stock market prediction of S&P 500 and DJIA usingbacterial foraging optimization technique[C]//Congress on Evolutionary Computation. IEEE,2007: 2569-2575.
[51] CONT R, KUKANOV A, STOIKOV S. The price impact of order book events[J]. Journal ofFinancial Econometrics, 2014, 12(1): 47-88.
[52] QURESHI F. Investigating Limit Order Book Features for Short-Term Price Prediction: AMachine Learning Approach[J]. Available at SSRN 3305277, 2018.
[53] HANDA P, SCHWARTZ R A. Limit order trading[J]. The Journal of Finance, 1996, 51(5):1835-1861.54参考文献
[54] GOLDSTEIN M A, KAVAJECZ K A. Trading strategies during circuit breakers and extrememarket movements[J]. Journal of Financial Markets, 2004, 7(3): 301-333.
[55] PHAM M C, ANDERSON H M, DUONG H N, et al. Dynamics of the limit order book andthe volume-volatility relation[C]//Financial Management Association Conference, San Diego.2018.
[56] BAE K H, JANG H, PARK K S. Traders’choice between limit and market orders: evidencefrom NYSE stocks[J]. Journal of Financial Markets, 2003, 6(4): 517-538.
[57] LO D K, HALL A D. Resiliency of the limit order book[J]. Journal of Economic Dynamicsand Control, 2015, 61: 222-244.
[58] 潘林. 基于小波分析与神经网络的股票市场预测应用研究[D]. 武汉理工大学, 2006.
[59] 马耀兰. 统计数据拟合预测及其在证券分析中的应用[D]. 北方民族大学, 2009.
[60] 肖菁, 潘中亮. 股票价格短期预测的 LM 遗传神经网络算法[J]. 计算机应用, 2012, 32:144-150.
[61] 于卓熙, 秦璐, 赵志文, 等. 基于主成分分析与广义回归神经网络的股票价格预测[J]. 统计与决策, 2018, 34: 168-171.
[62] 武博. 基于 LSTM 模型的股票价格预测[D]. 大连理工大学, 2021.
[63] 刘子诺. 基于秃鹰搜索算法和极限学习机的股票价格预测模型[J]. 中国管理信息化, 2022,25: 157-160.
[64] ANDERSON J A. An introduction to neural networks[M]. Massachusetts Institute of Technology Press, 1995.
[65] RAMACHANDRAN P, ZOPH B, LE Q V. Searching for activation functions[A]. 2017.
[66] RUDER S. An overview of gradient descent optimization algorithms[A]. 2016.
[67] BOTTOU L. Stochastic gradient descent tricks[J]. Neural Networks: Tricks of the Trade:Second Edition, 2012: 421-436.
[68] QIAN N. On the momentum term in gradient descent learning algorithms[J]. Neural Networks,1999, 12(1): 145-151.
[69] DUCHI J, HAZAN E, SINGER Y. Adaptive subgradient methods for online learning andstochastic optimization.[J]. Journal of Machine Learning Research, 2011, 12(7).
[70] KINGMA D P, BA J. Adam: A method for stochastic optimization[A]. 2014.
[71] SVOZIL D, KVASNICKA V, POSPICHAL J. Introduction to multi-layer feed-forward neuralnetworks[J]. Chemometrics and Intelligent Laboratory Systems, 1997, 39(1): 43-62.
[72] GU J, WANG Z, KUEN J, et al. Recent advances in convolutional neural networks[J]. PatternRecognition, 2018, 77: 354-377.
[73] MEDSKER L R, JAIN L. Recurrent neural networks[J]. Design and Applications, 2001, 5:64-67.
[74] BALDI P, SADOWSKI P J. Understanding dropout[J]. Advances in Neural Information Processing Systems, 2013, 26.
[75] MENG X, BRADLEY J, YAVUZ B, et al. Mllib: Machine learning in apache spark[J]. TheJournal of Machine Learning Research, 2016, 17(1): 1235-1241.55参考文献
[76] KNUTSON B, STEVENS P, PATTON M, et al. Consumers’ expectations for service quality ineconomy, mid-price and luxury hotels[J]. Journal of Hospitality & Leisure Marketing, 1993, 1(2): 27-43.
[77] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Advances inneural information processing systems. 2017: 5998-6008.
[78] DEVLIN J, CHANG M W, LEE K, et al. Bert: Pre-training of deep bidirectional transformersfor language understanding[A]. 2018.
[79] HUANG C, GE W, CHOU H, et al. Benchmark Dataset for Short-Term Market Prediction ofLimit Order Book in China Markets[J]. The Journal of Financial Data Science, 2021, 3(4):171-183.
[80] CHENG Y, WANG D, ZHOU P, et al. Model compression and acceleration for deep neuralnetworks: The principles, progress, and challenges[J]. IEEE Signal Processing Magazine, 2018,35(1): 126-136.
[81] CHENG Y, WANG D, ZHOU P, et al. A survey of model compression and acceleration for deepneural networks[A]. 2017.
[82] POLINO A, PASCANU R, ALISTARH D. Model compression via distillation and quantization[A]. 2018.
[83] REED R. Pruning algorithms-a survey[J]. IEEE Transactions on Neural Networks, 1993, 4(5):740-747.
[84] CHOI B, LEE J H, KIM D H. Solving local minima problem with large number of hidden nodeson two-layered feed-forward artificial neural networks[J]. Neurocomputing, 2008, 71(16-18):3640-3643.
[85] FRANKLE J, CARBIN M. The lottery ticket hypothesis: Finding sparse, trainable neural networks[A]. 2018.
[86] ZHU M, GUPTA S. To prune, or not to prune: exploring the efficacy of pruning for modelcompression[A]. 2017.