Forecasting the Accuracy of Stock Prices of The Financial Holding Companies Through Models

• Main Authors: Lo, Chia-Ching, 羅家景
• Other Authors: Liang, Shih-An
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/63491569622453148286

碩士 === 國立臺北大學 === 國際財務金融碩士在職專班 === 93 === Forecasting the Accuracy of Stock Prices of The Financial Holding Companies Through Models by Lo, Chia-Ching June, 2005 ADVISOR(S): Professor Liang, Shih-An ADVISOR(S): Dr. Lin, Chuang -Yuang DEPARTMENT:EXECUTIVE MASTER OF BUSINESS ADMINISTRATION IN INTERNATIONAL FINANCE MAJOR：INTERNATIONAL FINANCE DEGREE：MASTER OF BUSINESS ADMINISTRATION The purpose of this research is to forecast the accuracy of stock prices of the current 13 financial holding companies through models. Also, the future trends of stock prices to be forecasted by various models had been discussed. The ARIMA model, ARIMA+GARCH model, Moving Average model, Exponential Smooth model and Back Propagation Network model were used in the research. Through various evaluation criteria, the estimated data created by each model were observed and the performances of models were compared. Taiwan market weighting stock price index, the financial insurance index and the daily return rates calculated by natural logarithm of current 13 financial holding companies were collected. The sampling period is from January 2, 2003 to March 31, 2005. 556 daily returns rates are the total number. But the last 57 data will be the forecasted data, so the front 499 data will be the tested data for the five kinds of forecast models. The results of this empirical study are as follows: (1) From the Mean Absolute Average Error(MAPE) point of view, ARIMA+GARCH model is the best, then ARIMA , Exponential Smoothing , Back Propagation Network and Moving Average model follow. Forecast errors (MAPE) from 5 models are all smaller than 1.5%. The forecasting accuracy using by MAPE is extremely good. (2) From Correct sign rate (CSR) point of view, the Back Propagation Network model is the best, then Exponential Smoothing, ARIMA+GARCH Moving Average, ARIMA follow. All the CSR of five models are bigger than 50%, thus the precise directions forecasted by 5 models were shown. (3) From the investment returns rate (Strategy) points of view, Back Propagation Network model ranks first, then Exponential Smoothing, Moving Average, ARIMA+GARCH, ARIMA follow. All the investment average return rates (Strategy) of 5 models are higher than the actual average return rate –2.1%. So the investment return rates forecasted by 5 models were definitely good. (4) From the relationships among MAPE, CSR and Strategy, neither the ARIMA, nor the Moving Average, nor the Exponential Smoothing, nor the Back Propagation Network shows its good relationship, but the unidirectional forecast error value and the forecast direction made by ARIMA+GARCH model is shown. (5) All ARIMA, ARIMA+GARCH, Exponential Smoothing, Back Propagation Network models precisely forecast stock price error values and the estimation error values. (6) The stock price error values and the original sequence coefficient of variation all significantly related made by the 5 models.