search for




 

Tensor regression for short-term mean Korean Stock Price Index (KOSPI) prediction
Journal of the Korean Data & Information Science Society 2022;33:601-14
Published online July 31, 2022;  https://doi.org/10.7465/jkdi.2022.33.4.601
© 2022 Korean Data and Information Science Society.

Jinwon Heo1 · Kwangyee Ko2 · Jangsun Baek3

123Department of Statistics, Chonnam National University
Correspondence to: This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2018R1D1A1B07049729)
1 Ph. D. student, Department of Statistics, Chonnam National University, Gwangju 61186, Korea.
2 Part-time instructor, Department of Statistics, Chonnam National University, Gwangju 61186, Korea.
3 Professor, Department of Statistics, Chonnam National University, Gwangju 61186, Korea. E-mail: jbaek@jnu.ac.kr
Received March 18, 2022; Revised May 9, 2022; Accepted May 23, 2022.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
In modern data analysis, tensor data, which is a large-scale multidimensional array, is often used in statistical models. When using such tensor data, most conventional methods require converting the tensor into the high-dimensional vectors. However, when these methods do not specify the intercorrelated characteristics of multidimensional array variables properly in the statistical model, performance may sometimes be degraded and curse of dimensionality problem is occurred easily. In this paper, we propose a tensor regression model that can reduce insignificant explanatory variables by regularizing regression coefficients without transforming tensor data into vectors. We collect historical stock price index and technical index data over a certain period of time, create tensor data, and apply a tensor regression model to them to predict the short-term future KOSPI. The explanatory data are technical indicators for the top 20 KOSPI stocks and three major international indices from January 2007 to August 2020, and the response data are the mean KOSPI for future 5 and 10 days. Tensor regression techniques and various machine learning analysis techniques (SVM, ANN, Lasso Regression) were applied to predict the mean KOSPI for future 5 and 10 days period, and the MSE and the up and down prediction performance scores (Accuracy, Precision, Recall, F1-Score) were compared. As a result of the experiment, the MSE and the up and down prediction performance scores of the tensor regression showed superior performance than the conventional machine learning methods.
Keywords : KOSPI, machine Learning, technical indicators, tensor decomposition, tensor regression.