Prediction of Bank Share Prices Using Data Mining: A Case Study of Listed Bank on Ho Chi Minh Stock Exchange

Abstract

This thesis aims at identifying and applying data mining to predict bank share prices listed on HOSE. In this study, the authors divided banks into 03 groups of large banks, medium banks and small banks, so as to test on the performance applications of forecasting programs. For each group of banks, one representative bank share is selected to participate in the test. The representative of big banks listed on HOSE is BIDV (stock code BID), the medium one is Vietnam Technological and Commercial Joint Stock Bank (stock code TCB) and the small one is Tien Phong Commercial Joint Stock Bank (stock code TPB). The input data used in the program are textual data and digital data. The number of textual data are big with 23,879 articles published on the 04 official websites, i.e., cafef.vn, thanhnien.vn, vnexpress.net, and StockBiz.vn. In the meantime, digital data is collected from the historical stock price of BID, TCB, and TPB. Upon completion of downloading these articles from the above-mentioned websites, the articles must be published on the same day, being labelled and classified into categories of good news, moderate news, and bad news based on the next day's stock price movement. Then, the data is undergone pre-processing and is split into 02 parts with the proportion 70%:30% respectively. The first 70% of data is used for building the test for model training and rest of 30% for program performance testing. Based on results from program performance models, the best accuracy level is withdrawn for each case of bank share representing for 03 groups of banks. In addition, the authors selected the website that can provide the most effective information affecting investors emotion to buy or to sell, or hold bank shares. Recognizing that the extracted information, if applications of the 04 websites, will reduce the overall efficiency of the performance program, the authors decided just to take the data of one website that provide the highest accuracy, then applied information from this website to run the programs. In data mining, the four models SMV, Random Forest, Decision Tree, and K-Neighbor are applied, and just one model is selected for each case of bank share. The research results show that SMV is selected for case of BID and TPB share, while Random Forest is the best serve for TCB. The results also show that the highest efficiency in prediction reaches 52.94% of its accuracy, and the website Stockbiz.vn is best served for BID, Cafef.vn is the best serve for TCB, and Vnexpress.net is the best serve for TPB.