The Effect of Precision in Earning Property On Statistical Earning Forecast Models

• Main Authors: ChaneHau Kuo, 郭娟華
• Other Authors: Yeh, Chin-Chen
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/65696413129817535586

碩士 === 淡江大學 === 會計學系 === 87 === Abstract: Past research in the earning forecast are found the most accurate forecasts overall are provided by financial analysts. However, the statistical models have used more popular. It has been an important study in pursuing the property on statistical models. This paper is emphasized on recognizing the earning property. Using the autogressive integrated moving average (ARIMA), autogressive conditional heteroskedasticity(ARCH) and the generalized autogressive conditional heteroskedasticity(GARCH) models to compares the relative predictive ability of above statistical models with analysts'' forecasts, in order to find the best statistical forecast model. In this paper, first separates the earning property of corporations from the structural changes and the volatility. Second, compares the relative predictive ability of ARIMA ARCH(1) GARCH(1,1) and GARCH(p,q) models under the different earning properties to find the superior statistical forecast model ; and then , compares the relative predictive ability of GARCH(1,1) with GARCH(p,q) model, in order to examine whether the predictive ability of GARCH(p,q) model can be improved by the p and q parameter value. Besides, analyzed whether or not the predictive ability of analysts'' forecasts still more precisely than improved statistical model. Finally, indicates the best statistical models under different earning properties of a company. In this paper, I use the minimum Akaile''s (1974) information criterion (AIC) and the minimum Schwarz''s (1978) information criterion (SIC) as the criterias to recognize the best accurate p and q parameter value. The empirical results can be summarized in the followings: 1. Earning property can be separated into 8 categories from the structural change and the volatility: (1). High volatility (have GARCH effect); (2). Stationary volatility (don''t have GARCH effect); (3). Have the structural change in the earning process; (4). Don''t have the structural change in the earning process; (5). High volatility but don''t have the structural change in the earning process; (6) Stationary volatility but with the structural change in the earning process; (7). High volatility and with the structural change in the earning process; (8) Stationary volatility and don''t have the structural change in the earning process. 2. The predictive ability of GARCH (1,1) can be improved by the p and q parameter value of the GARCH (p,q) model. 3. No matter what the earning property a company is, the forecast accuracy of the best GARCH (p,q) model is similar to analysts'' predictions. 4. Structural change depress the predictive ability of ARIMA model ; but the structural change can be improved by the ARCH(1)、GARCH(1,1) and the best GARCH(p,q) models . 5. The best statistical models can be effected by the different earning properties in the earning process. If a company has the structural change in the earning process, GARCH (p,q) model is the best model ; if a company has high volatility (have GARCH effect) , ARIMA and GARCH models are both the best models.