How to Use Machine Learning to Find Α in Stock Market

Abstract:
Nowadays, the selection range of α factors is gradually narrowing and its excess return α is also declining. Fortunately, for the above problems, machine learning (especially deep learning) shows many advantages. These excellent characteristics make machine learning likely to become the most important tool to obtain excess returns in the future. In this article, three machine learning algorithms, support vector machine, adaptive boosting and neural network, are discussed. The other various algorithms are derived and evolved on the basis of these three algorithms. The three machine learning algorithms are summarized and analyzed according to the derivation law of machine learning algorithms. Based on the original linear simple algorithm, support vector machine forecasts data by adding different dimensions, but its learning ability and sample complexity are complementary, so its accuracy will be reduced in the case of sample complexity. Adaptive boost (adaptive boosting) is evolved from the decision tree, which is more inclined to the improvement of learning ability. That is, it improves the accuracy of prediction, but its ability to solve complex samples is limited. Neural network simulates human neuron construction. It can not only solve a large number of complex sample data, but also has high learning ability, which solves the defects of support vector machine and Adaptive boost. By summarizing the representative algorithms in different evolution stages of machine learning algorithms, neural network can predict the performance of stocks well and find more accurate α factors, which have greater development prospects and application space in the growing stock market.