Prediction Stock Value by Time Series Regression Models - A Case Study of ASUS Corporation

• Main Authors: TUNG, MEI-SHEN, 董美伸
• Other Authors: JENG, PO-WEN
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/y339ts

碩士 === 國立屏東大學 === 財務金融學系碩士班 === 105 === The application of predictive analysis on financial research is becoming notable doctrine. The research was based on the economic variable dividend and stock price, aiming to explore the long-term equilibrium relationship between the two. The results showed that stock price change was related to the dividend change testing with the statistical model, which gave a significant predictive ability that there was a long-term equilibrium between the dividend and the stock price. The empirical results of the model were summarized as the followings：(1) Time series data using a single test confirmed the dividend and the stock price were the non-stationary sequence time series, and the two variables were the same integration order. (2) Furthermore, a co-integration test was also used to confirm the existence of co-integration between the two variables based on the existence of the co-integration relationship between the two variables. The error correction model was applied to reflect the long-term equilibrium relationship and short-term dynamic behavior between dividend and stock price. (3) On the basis of the long-term equilibrium relationship between the two variables, the dividend change had an impact on stock price movements; therefore, the binary variable variables was used to predict the future development of stock prices, and linear and non-linear regression model of the linear probability model and probit and logit model for stock price were used to predict results stock price. Because the value of variables on linear probability model would make the range of prediction of probability of the value exceed the range of 0 to 1, the logit model and probit model were thus exercised to predict stock price. For this reason of the cumulative distribution probability of the normal distribution on probit and logit model, the probability of model prediction would be the range of 0 to 1. The study found that the probit model was better than the logit model in the comparison of the accuracy and effectiveness of the probit model and the logit model.