股票市场预测的机器学习技术与数据：文献综述

a,∗, Christoph Lohrmanna, Pasi Luukkaa, Jari PorrasbA B S T R A C TIn this literature review, we investigate machine learning techniques that are applied for stock marketprediction. A focus area in this literature review is the stock markets investigated in the literature as well asthe types of variables used as input in the machine learning techniques used for predicting these markets. Weexamined 138 journal articles published between 2000 and 2019. The main contributions of this review are: (1)an extensive examination of the data, in particular, the markets and stock indices covered in the predictions,as well as the 2173 unique variables used for stock market predictions, including technical indicators, macro-economic variables, and fundamental indicators, and (2) an in-depth review of the machine learning techniquesand their variants deployed for the predictions. In addition, we provide a bibliometric analysis of these journalarticles, highlighting the most influential works and articles.market efficiency: weak-form, semi-strong form, and strong-form effi-ciency (Atsalakis & Valavanis, 2009a; Fama, 1970).Weak-form market efficiency assumes that information containedin past prices of a time series is already reflected in the current stockprice and does not help in predicting future price movements (Fama,1970). Therefore, in the weak form of EMH, technical analysis cannotoutperform a buy-and-hold strategy in terms of expected return (Fama,1965; Leigh et al., 2002). The second form of the EMH is the semi-strong market efficiency, which states that stock prices fully reflectall publicly available information (Fama, 1965). All publicly availableinformation also includes information about past prices, which meansthat technical analysis may also not lead to consistently higher expectedreturns. Moreover, all publicly available information encompasses, forinstance, fundamental information about economic conditions, politi-cal events, interest rates, and company-specific information, which isavailable to the public and affects stock prices (Wang et al., 2011).Notwithstanding, in the semi-strong form of market efficiency, publiclyavailable information, including fundamental data, does not enable aninvestor to consistently outperform the market. This implies that activemanagement that uses all publicly available information will not con-sistently yield higher risk-adjusted returns than passive management(e.g., buy-and-hold a stock market index). In contrast to the semi-strong form, the strong form of the EMH states that all information,including insider information, is reflected in stock prices. This precludesany investor, even those with insider information, from consistentlyachieving higher expected returns than the market (Fama, 1965, 1970;Leigh et al., 2002). Therefore, the EMH in its strongest form effectively states that returns in the stock market are unforecastable (Timmermann& Granger, 2004). The strong form of the EMH is rather extreme, andeven Eugene (Fama, 1970) himself stated that one would not expectthat insider information (e.g., of company officers) cannot be used togenerate higher expected profits.Over time, there has been an increasing number of challenges ofthe efficient market hypothesis and the fact that securities are pricedrationally (Borovkova & Tsiamas, 2019; Daniel et al., 1998). Therehave been several market anomalies (Malkiel & Mullainathan, 2005)such as the overreaction of financial markets (Bondt & Thaler, 1985,1990) and their underreaction, the existence of short-term momentum,long-term reversal, and the high volatility of asset prices (Daniel et al.,1998) which represent support against the efficient market hypothesis(especially in its weak-form). Some researchers discussed explanationsfor such anomalies that are in line with the EMH such as that over-and under-reactions happen randomly and are equally frequent (Fama,1998) and the possibility of institutional investors being able to offsetthe anomalies created by less sophisticated investors (Shiller, 2003).However, there remained doubt that a model based on investor ratio-nality can accommodate the observed anomalies (Daniel et al., 1998).This led to a shift towards models incorporating human psychology,leading to the emergence of behavioral finance (Bondt & Thaler, 1990;Shiller, 2003), which questions the perfect rationality of investors dueto behavioral biases such as loss aversion, overreaction, and overre-action (Lo, 2004). One attempt to reconcile the EMH and behavioralfinance was the proposal of the adaptive markets hypothesis (AMH),which acknowledges and explains the existence of anomalies in finan-cial markets (Lo, 2004). For a detailed discussion of the evolution ofthe efficient market hypothesis, see, Lim and Brooks (2011).Because of the fact that market anomalies may exist, it is unsurpris-ing that a large number of market participants use information of pastmarket prices, company-specific information such as past earning