Checking the accuracy of learning machines in predicting stock returns using the Rough set model, Nearest neighbor and decision tree.

Subject Areas : Financial engineering

mohammad reza karimi pouya ¹ (Student of Accounting, Faculty of Humanities, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran)
mehrdad ghanbari ² (Assistant Professor, Department of Accounting, Faculty of Humanities, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran)
babak jamshidinavid ³ (Assistant Professor, Department of Accounting, Faculty of Humanities, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran)
mansoor esmaeilpour ⁴ (Assistant Professor, Department of Computer Engineering, Faculty of Engineering, Hamedan Branch, Islamic Azad University, Hamedan, Iran)

Received: 2018-05-05 Accepted : 2018-06-05 Published : 2019-03-21

Keywords: Stock Return, Decision tree, nearest neighbor, Estimation (prediction), rough,

Abstract :

Prediction is an essential component of short and medium term planning in any business. A precise prediction can be effective in generating returns, managing cash flows, and allocating resources, enabling an investor to estimate, within a given time frame, its business revenue and its returns. Researchers have the idea to set aside old methods, which takes expense and time, and implement new methods such as the use of learning machines. This research is of the type of research, analytical-empirical, in terms of research design, post-event, in terms of purpose, applied, in terms of implementation logic, deductive and in terms of time, longitudinal and prospective type. In this research, the algorithm model of the nearest neighbor, the Rough method and the decision tree are used to improve predictive power, cost reduction, and time prediction of stock returns. For this purpose, a sample of 113 listed companies in the Tehran Stock Exchange during a 10-year period (2006-2015) was selected from the companies listed in the Tehran Stock Exchange. The results of the research showed that all the hypotheses of this research are based on a difference in the accuracy of estimating these models in the prediction of the three dependent variables.