Predicting the Direction of US Stock Prices Using Effective Transfer Entropy and Machine Learning Techniques

Abstract

This study aims to predict the direction of US stock prices by integrating time-varying effective transfer entropy (ETE) and various machine learning algorithms. At first, we explore that the ETE based on 3 and 6 months moving windows can be regarded as the market explanatory variable by analyzing the association between the financial crises and Granger-causal relationships among the stocks. Then, we discover that the prediction performance on the stock price direction can be improved when the ETE driven variable is integrated as a new feature in the logistic regression, multilayer perceptron, random forest, XGBoost, and long short-term memory network. Meanwhile, we suggest utilizing the adjusted accuracy derived from the risk-adjusted return in finance as a prediction performance measure. Lastly, we confirm that the multilayer perceptron and long short-term memory network are more suitable for stock price prediction. This study is the first attempt to predict the stock price direction using ETE, which can be conveniently applied to the practical field. INDEX TERMS Econophysics, Effective transfer entropy, Feature engineering, Information entropy, Machine learning, Prediction algorithms, Stock markets, Time series analysis 19 characteristics of the system using random matrix theory 20 and network analysis [7]-[9]. Since then, the studies have 21 discovered that a linear model such as the Pearson correlation 22 is not sufficient enough to quantify the relationships among 23 the stocks. More importantly, the causal relationship is not 24 directly linked to the presence of correlation. In this context, 25 the Granger-causality [10] has been introduced to define 26 the causal relationship between time series. However, its 27 function is limited to express the existence of information 28 flow based on a linear relationship rather than measuring the 29 amount of information flow. 30 To overcome such limitations of a simple linear model of a 31 Granger-causal relationship, the concept of transfer entropy 32 (TE), proposed by Schreiber [11], has been suggested instead 33 to measure the amount of information flow. TE is a non-34 parametric measure of the amount of information transfer 35 from a variable to a variable based on the Shannon entropy 36

Econophysics and the Entropic Foundations of Economics

“…More recent studies have expanded the forms of entropy used in studying financial market dynamics. Thus, transfer entropy has been used by Jizba et al [131] to study differences in related financial times series focusing on spike events by Dimpli and Peter [132] to study cryptocurrency dynamics and by Kim et al [133] for directional stock market forecasting. In addition, permutation entropy has been used in a variety of financial market econophysics applications [134].…”

Combining Measures of Signal Complexity and Machine Learning for Time Series Analyis: A Review

“…Ref [ 32 ]: Here, different approaches to predict the direction of US stock prices are compared with each other. Several algorithms were employed for this task, i.e., logistic regression, a multilayer perceptron (MLP), random forest, XGBoost, and a long short term memory (LSTM) recurrent neural network.…”

“…In [ 32 ], ETE is used to enhance the prediction of the direction of US stocks using a variety of different algorithms/approaches. Here it is used as an additional feature in the input of the algorithm.…”

The Stock Exchange Prediction using Machine Learning Techniques: A Comprehensive and Systematic Literature Review

“…For example, [3], [4], [5] proposed a algorithm to prediction the stock exchange using the decision tree method in a random forest or ensemble bagging decision tree. [6], [7], [8], [9] studied stock trading through deep learning such as deep neural network, long short-term memory, recurrent natural network, and convolutional neural network.…”