Table 3.

Picking attractive securities.

Application purpose	Method/data	Performance criteria	References
Measuring asset price premiums	Boosted RT, RF and NN	The higher gain of ML methods compared with leading regression-based strategies for return prediction is shown	Gu et al., 2020
Risk price estimation and dimensionality reduction	Bayesian approach	Building of a robust stochastic discount factor from a large set of stock characteristics	Kozak et al., 2020
Return estimation	RT	RTs were built to determine which firm characteristics out of 30 attributes are likely to drive future returns	Coqueret and Guida, 2018
Feature extraction	Restricted Boltzmann Machine	Proposes an encoder to extract features from stock prices and pass them to a feedforward NN	Takeuchi and Lee, 2013
Prediction of stock markets	RF	Method designed to predict price trends in the stock market	Kamble, 2017; Zhang et al., 2018
Cross-section prediction of exceed return	RF	Select stocks in S&P500 and STOXX600 with the highest monthly predictions	Kaczmarek and Perez, 2021
Benchmarking of ML techniques	RF, GBT, DL	Ensembles of different ML methods in the context of statistical arbitrage for S&P500	Krauss et al., 2017
Building ML signals for long-short strategies	GBT	Boosted Trees to more than 200 features clustered in six families, building an ML signal that outperforms the benchmarks for long-short strategies	Guida and Coqueret, 2018
Distinguish “good” stocks from “bad” stocks	LR, DNN, RF	Effectiveness of the stock selection strategy is validated in the Chinese stock market in both statistical and practical aspects where stacking outperforms other models	Fu et al., 2018