Applications of deep learning in stock market prediction: recent progress
AApplications of deep learning in stock marketprediction: recent progress
Weiwei Jiang ∗ Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
Abstract
Stock market prediction has been a classical yet challenging problem, with theattention from both economists and computer scientists. With the purpose ofbuilding an effective prediction model, both linear and machine learning toolshave been explored for the past couple of decades. Lately, deep learning modelshave been introduced as new frontiers for this topic and the rapid developmentis too fast to catch up. Hence, our motivation for this survey is to give a latestreview of recent works on deep learning models for stock market prediction. Wenot only category the different data sources, various neural network structures,and common used evaluation metrics, but also the implementation and repro-ducibility. Our goal is to help the interested researchers to synchronize withthe latest progress and also help them to easily reproduce the previous stud-ies as baselines. Base on the summary, we also highlight some future researchdirections in this topic.
Keywords:
Stock market prediction, deep learning, machine learning,feedforward neural network, convolutional neural network, recurrent neuralnetwork
1. Introduction
Stock market prediction is a classical problem in the intersection of financeand computer science. For this problem, the famous efficient market hypoth- ∗ Corresponding author. E-mail address: [email protected]
Preprint submitted to Elsevier Journal March 5, 2020 a r X i v : . [ q -f i n . S T ] F e b sis (EMH) gives a pessimistic view and implies that financial market is ef-ficient (Fama, 1965), which maintains that technical analysis or fundamentalanalysis (or any analysis) would not yield any consistent over-average profitto investors. However, many researchers disagree with EMH (Malkiel, 2003).Some studies are trying to measure the different efficiency levels for mature andemerging markets, while other studies are trying to build effective predictionmodels for stock markets, which is also the scope of this survey.The effort starts with the stories of fundamental analysis and technical analy-sis. Fundamental analysis evaluates the stock price based on its intrinsic value, i.e. , fair value, while technical analysis only relies on the basis of charts andtrends. The technical indicators from experience can be further used as hand-crafted input features for machine learning and deep learning models. After-wards, linear models are introduced as the solutions for stock market prediction,which include autoregressive integrated moving average (ARIMA) (Hyndman &Athanasopoulos, 2018) and generalized autoregressive conditional heteroskedas-ticity (GARCH) (Bollerslev, 1986). With the development of machine learningmodels, they are also applied for stock market prediction, e.g. , Logistic regres-sion and support vector machine (Alpaydin, 2014).Our focus in this survey would be the latest emerging deep learning, whichis represents by various structures of deep neural networks (Goodfellow et al.,2016). Powered by the collection of big data from the Web, the parallel process-ing ability of graphics processing units (GPUs), and the new convolutional neu-ral network family, deep learning has achieved a tremendous success in the pastfew years, for many different applications including image classification (Rawat& Wang, 2017; Jiang & Zhang, 2020), object detection (Zhao et al., 2019), timeseries prediction (Brownlee, 2018; Jiang & Zhang, 2018), etc. With a strongability of dealing with big data and learning the nonlinear relationship betweeninput features and prediction target, deep learning models have shown a betterperformance than both linear and machine learning models on the tasks thatinclude stock market prediction.In the past few years, both the basic tools for deep learning and the new2rediction models are undergoing a rapid development. With the continuousimproved programming packages, it becomes easier to implement and test anovel deep learning model. Also, the collection of online news or twitter dataprovides new sources of predicting stock market. More recently, graph neuralnetworks using various knowledge graph data appear as new ideas. The studyfor stock market prediction is not limited to the academia. Attracted by thepotential profit by stock trading powered by the latest deep learning models,asset management companies and investment banks are also increasing theirresearch grant for artificial intelligence which is represented by deep learningmodels nowadays.Since there are many new developments in this area, this situation makesit difficult for a novice to catch up with the latest progress. To alleviate thisproblem, we summarize the latest progress of deep learning techniques for stockmarket prediction, especially those which only appear in the past three years.We also present the trend of each step in the prediction workflow in these threeyears, which would help the new-comers to keep on the right track, withoutwasting time on obsolete technologies.We focus on the application of stock market, however, machine learning anddeep learning methods have been applied in many financial problems. It wouldbe beyond the scope of this survey to cover all these problems. However, thefindings presented in this survey would also be insightful for other time seriesprediction problems in the finance area, e.g. , exchange rate or cryptocurrencyprice prediction.We also pay a special attention to the implementation and reproducibilityof previous studies, which is often neglected in similar surveys. The list of opendata and code from published papers would not only help the readers to checkthe validity of their findings, but also implement these models as baselines andmake a fair comparison on the same datasets. Based on our summary of thesurveyed papers, we try to point out some future research directions in thissurvey, which would help the readers to choose their next movement.Our main contribution in this survey are summarized as follows:3. We summarize the latest progress of applying deep learning techniques tostock market prediction, especially those which only appear in the pastthree years.2. We give a general workflow for stock market prediction, based on whichthe previous studies can be easily classified and summarized. And thefuture studies can refer to the previous work in each step of the workflow.3. We pay a special attention to implementation and reproducibility, whichis often neglected in similar surveys.4. We point out several future directions, some of which are on-going andhelp the readers to catch up with the research frontiers.The rest of this survey is organize as follows: Section 2 presents relatedwork; Section 3 gives an overview of the papers we cover; Section 4 describesthe major findings in each step of the prediction workflow; Section 5 gives thediscussion about implementation and reproducibility; Section 6 points up somepossible future research directions; We conclude this survey in Section 7.
2. Related Work
Stock market prediction has been a research topic for a long time, and thereare some review papers accompanied with the development and flourishmentof deep learning methods prior to our work. While their focus could also beapplications of deep learning methods, stock market prediction could only be oneexample of many financial problems in these previous surveys. In this section,we list some of them in a chronological order and discuss our motivation andunique perspectives.Back to 2009, Atsalakis & Valavanis (2009) surveys more than 100 relatedpublished articles that focus on neural and neuro-fuzzy techniques derived andapplied to forecast stock markets, with the discussion of classifications of in-put data, forecasting methodology, performance evaluation and performancemeasures used. Li & Ma (2010) gives a survey on the application of artificialneural networks in forecasting financial market prices, including the forecast of4tock prices, option pricing, exchange rates, banking and financial crisis. Nik-farjam et al. (2010) surveys some primary studies which implement text miningtechniques to extract qualitative information about companies and use this in-formation to predict the future behavior of stock prices based on how good orbad are the news about these companies.Aguilar-Rivera et al. (2015) presents a review of the application of evolution-ary computation methods to solving financial problems, including the techniquesof genetic algorithms, genetic programming, multi-objective evolutionary algo-rithms, learning classifier systems, co-evolutionary approaches, and estimationof distribution algorithms. Cavalcante et al. (2016) gives an overview of the mostimportant primary studies published from 2009 to 2015, which cover techniquesfor preprocessing and clustering of financial data, for forecasting future mar-ket movements, for mining financial text information, among others. Tk´aˇc &Verner (2016) provides a systematic overview of neural network applications inbusiness between 1994 and 2015 and reveals that most of the research has aimedat financial distress and bankruptcy problems, stock price forecasting, and deci-sion support, with special attention to classification tasks. Besides conventionalmultilayer feedforward network with gradient descent backpropagation, varioushybrid networks have been developed in order to improve the performance ofstandard models.More recently, Xing et al. (2018) reviews the application of cutting-edgeNLP techniques for financial forecasting, which would be concerned when textincluding the financial news or twitters is used as input for stock market pre-diction. Rundo et al. (2019) covers a wider topic both in the machine learningtechniques, which include deep learning, but also the field of quantitative financefrom HFT trading systems to financial portfolio allocation and optimization sys-tems. Nti et al. (2019) focuses on the fundamental and technical analysis, andfind that support vector machine and artificial neural network are the most usedmachine learning techniques for stock market prediction. Based on its review ofstock analysis, Shah et al. (2019) points out some challenges and research op-portunities, including issues of live testing, algorithmic trading, self-defeating,5ong-term predictions, and sentiment analysis on company filings.Different from other related works that cover more papers from the computerscience community, Reschenhofer et al. (2019) reviews articles covered by theSocial Sciences Citation Index in the category Business, Finance and gives moreinsight on economic significance. It also points out some problems in the exist-ing literature, including unsuitable benchmarks, short evaluation periods, andnonoperational trading strategies.Some latest reviews are trying to cover a wider range, e.g., Shah et al. (2019)covers machine learning techniques applied to the prediction of financial marketprices, and Sezer et al. (2019) covers more financial instruments. However,our motivation is to catch up with the research trend of applying deep learningtechniques, which have been proved to outperform traditional machine learningtechniques, e.g., support vector machine in most of the publications, with onlya few exceptions, e.g., Ballings et al. (2015) finds that Random Forest is thetop algorithm followed by Support Vector Machines, Kernel Factory, AdaBoost,Neural Networks, K-Nearest Neighbors and Logistic Regression, and Ersanet al. (2019) finds that K-Nearest Neighbor and Artificial Neural Network bothoutperform Support Vector Machines, but there is no obvious pros and consbetween the performances of them. With the accumulation of historical pricesand diverse input data types, e.g., financial news and twitter, we think theadvantages of deep learning techniques would continue and it is necessary tokeep updated with this trend for the future research.Compared with Sezer et al. (2019), whose focus is deep learning for finan-cial time series forecasting and a much longer time period (from 2005 to 2019exactly), we focus on the recent progress in the past three years (2017-2019)and a narrower scope of stock price and market index prediction. For readerswho are also interested in other financial instruments, e.g., commodity price,bond price, cryptocurrency price, etc., we would refer them to this work. Wealso care more about the implementation workflow and result reproducibility ofprevious studies, e.g., dataset and code availability, which is a problem that hasdrawn the attention from the AI researchers (Gundersen & Kjensmo, 2018). We6ould also pay more attention to the uniqueness of stock market prediction (orfinancial time series forecasting) from general time series prediction problems,e.g., the evaluation of profitability besides prediction accuracy.
3. Overview
In this section, we give an overview of the papers we are going to review inthis study. All the works are searched and collected from Google Scholar, withsearching keywords such as deep learning, stock prediction, stock forecasting,etc. Most of the covered papers (115 out of 124) are published in the past threeyears (2017-2019). In total, we cover 56 journal papers, 58 conference papersand 10 preprint papers. These preprint papers are all from arXiv.org, which isa famous website for e-print archive and we cover these papers to keep updatedwith the latest progress. The top source journals & conferences sorted by thenumber of papers we cover in this study are shown in Table 1 and Table 2,respectively.
Table 1: List of top source journals and the number of papers we cover in this study.
Journal Name Paper CountExpert Systems with Applications 12IEEE Access 5Neurocomputing 3Complexity 2Journal of Forecasting 2Knowledge-Based Systems 2Applied Soft Computing 2Mathematical Problems in Engineering 2PLOS ONE 2Others in total 24In this study, the major focus would be the prediction of the close prices ofindividual stocks and market indexes. Some financial instrument whose price is7 able 2: List of top conferences and the number of papers we cover in this study.
Conference Name Paper CountInternational Joint Conferences on Artificial In-telligence (IJCAI) 4International Joint Conference on Neural Net-works (IJCNN) 4Conference on Information and KnowledgeManagement (CIKM) 3International Conference on Neural InformationProcessing (ICONIP) 3Annual Meeting of the Association for Compu-tational Linguistics (ACL) 2IEEE Symposium Series on Computational In-telligence (SSCI) 2Hawaii International Conference on System Sci-ences (HICSS) 2ACM SIGKDD Conference on Knowledge Dis-covery and Data Mining (KDD) 2IEEE International Conference on Tools withArtificial Intelligence (ICTAI) 2Others in total 34bounded to the market index is also covered, e.g., some exchange-traded fund(ETF) or equity index futures that track the underlying market index. Forintraday prediction, we would also cover mid-price prediction for limit orderbooks. Other financial instruments are not mentioned in this study, e.g., bondprice and cryptocurrency price. More specifically, if the target to predict thespecific value of the prices, we classify it as a regression problem, and if thetarget is to predict the price movement direction, e.g., going up or down, weclassify it as a classification problem. Most studies are considering the daily8rediction (105 of 124) and only a few of them are considering the intradayprediction (18 of 124), e.g., 5-minute or hourly prediction. Only one of the124 papers is considering both the daily and intraday situations (Liu & Wang,2019).Based on the target output and frequency, the prediction problems can beclassified into four types: daily classification (52 of 124), daily regression (54 of124), intraday classification (8 of 124) and intraday regression (11 of 124). Adetailed paper count of different prediction problem types is shown in Figure 1.The reason behind this could be partially justified by the difficulty of collectingthe corresponding data. The daily historical prices and news titles are easier tocollect and process for research, while the intraday data is very limited in theacademia. We would further discuss the data availability in Section 5. G D L O \ F O D V V L I L F D W L R Q G D L O \ U H J U H V V L R Q L Q W U D G D \ F O D V V L I L F D W L R Q L Q W U D G D \ U H J U H V V L R Q 3 U R E O H P 7 \ S H 3 D S H U & R X Q W \ H D U Figure 1: The paper count of different problem types.
Surveyed markets as well as the most famous stock market index in these9arkets are shown in the Table 3. The paper count of different surveyed marketsis shown in Figure 2 . Most of the studies would focus on one market, whilesome of them would evaluate their models on multiple markets . Both maturemarkets (e.g., US) and emerging markets (e.g., China) are gaining a lot ofattention from the research community in the past three years.Table 3: List of surveyed markets and stock indexes.Country Index DescriptionUS S&P 500 Index of 505 common stocks issued by 500 large-cap companiesUS Dow JonesIndustrialAverage Index of 30 major companiesUS NASDAQComposite Index of common companies in NASDAQ stockmarketUS NYSE Compos-ite Index of common companies in New York StockExchangeUS RUSSEL 2000 Index of bottom 2,000 stocks in the Russell 3000IndexChina SSE Composite Index of common companies in Shanghai StockExchangeChina CSI 300 Index of top 300 stocks in Shanghai and Shen-zhen stock exchangesHongKong HSI Hang Seng Index of the largest companies inHong Kong Exchange Continued on next page An exception of using Europe for Ballings et al. (2015), in which 5767 publicly listedEuropean companies are covered. Due to the differences of trading rules, we list mainland China, Hong Kong and Taiwanseparately. able 3 – continued from previous page Country Index DescriptionJapan Nikkei 225 Index of 225 large companies in Tokyo StockExchangeKorea Korea Compos-ite Index of common companies in Korea Stock Ex-changeIndia BSE 30 Index of 30 companies exist in Bombay StockExchangeIndia NIFTI 50 Index of 50 companies exist in National StockExchangeEngland FTSE 100 Index of 100 companies in London Stock Ex-changeBrazil IBOV Bovespa Index of 60 stocksFrance CAC 40 Index of 40 stocks most significant stocks in Eu-ronext ParisGermany DAX Index of 30 major German companies in Frank-furt Stock ExchangeTurkey BIST 100 Index of 100 stocks in Borsa Istanbul Stock Ex-changeArgentina MER Merval Index in Buenos Aires Stock ExchangeBahrain BAX Bahrain All Share Index of 42 stocksChile IPSA Ipsa Index of 40 most liquid stocksAustralia All Ordinaries Index of 500 largest companies in Australian Se-curities Exchange
4. Prediction Workflow
Given different combinations of data sources, previous studies explored theuse of deep learning models to predict stock market price/movement. In this11 6 & K L Q D + R Q J . R Q J - D S D Q . R U H D , Q G L D ( Q J O D Q G % U D ] L O ) L Q O D Q G ) U D Q F H 7 D L Z D Q * H U P D Q \ 7 X U N H \ 7 K D L O D Q G ' X E D L $ U J H Q W L Q D % D K U D L Q ( X U R S H 5 R P D Q L D Q % U D V L O & K L O H $ X V W U D O L D &