Classification Models for Stocks' Performance Prediction: A Case Study in Stock Exchange of Thailand (SET)

Abstract

Within stock market, the objective of both stock buyers and sellers is to make profit on price difference based on their expectation on a company’s current and future value. There is no investing strategy that is considered to be the best by experts, and investing decision criteria remain contingent upon an individual investor’s experiences and bias. To address the challenges, this paper uses financial ratios and company’s industry data to construct forecasting models that quantitatively describe the return on stock investment. Various classification models, including Logistic Regression (LR), Decision Tree (DT), Linear Discriminant Analysis (LDA) and K-nearest neighbor are used in the current study to find the best model with high predictive power. Two types of classification models for predicting whether a stock’s one-year return in SET will outperform or underperform the SET Index and whether the return will be positive or negative are constructed in this study with Area Under the Curve (AUC) of Receiver Operating Characteristic (ROC) curve as measurement for models’ performance. This study primarily focuses on the Stock Exchange of Thailand. The resulting AUCs demonstrates that the usefulness of these models can be rated as “Acceptable” to “Good” with AUC range from 0.7 to above 0.8 using Deloitte’s standard.