MOHAMMAD SAID ZAINOL 1 , SITI MERIAM ZAHARI 1 , KAMARUZAMAN IBRAHIM 3, AZAMI ZAHARIM 2 , KAMARUZAMAN SOPIAN 3 1 Center of Studies for Decision Sciences, Faculty Computer Science and Mathematics, Universiti Teknologi MARA 40450 Shah Alam, MALAYSIA. 2 Head Centre for Engineering Education Research, Faculty of Engineering and Built Environment, 43600 UKM Bangi, MALAYSIA 3 Solar Energy Research Institute, Faculty of Engineering and Built Environment, 43600 UKM Bangi, MALAYSIA said@tmsk.uitm.edu.my, smbz_ma@yahoo.com , azami.zaharim@gmail.com, : 8 This paper reports the results of a study of outlier detection in time series data for the outlier of level change (LC) type. The main objective is to derive a test statistic for detecting LC in GARCH(1,1) processes. Subsequently a procedure for testing the presence of outliers using the statistics was developed. In the derivation of the statistics, the method applied was based on an analogy of GARCH(1,1) as being equivalent to ARMA(1,1) for the residuals ε . Because of the difficulty in determining the sampling distributions of the outlier detection statistics, critical regions were estimated through simulations. The developed outlier detection procedure was applied for testing the presence of LC outliers in the daily observations of the Index of Consumer Product Price (ICP) for the period 1990 to 2005. Over the period, the results indicate that LC outlier occurred in year 1998. : LC outlier; GARCH; simulation; least squares method This paper reports the results of a study to detect outliers of the level change (LC) type in GARCH(1,1) processes. Generally an outlier is a data point located “far away” from the rest. Outliers can also be an ‘extreme value’, or an ‘extraordinary’ value relative to the variance of the data set. Issues in outliers have raised a lot of interest in time series literature (see, , Hawkins, 1980; Gelman 1995; Mann, 2001, Pena, 2001; Balke and Fomby, 1994). The presence of these “strange” observations in a data set raises questions regarding its reliability. Further, their presence has been shown to cause estimation problems (Chang ., 1988; Chen and Liu, 1993; Tsay ., 2000; Franses and van Dijk, 2000; Berkoun ., 2003). Serious errors may occur if a time series contaminated by outliers is used to estimate a forecasting model. The estimated parameters in the model may be severely distorted. Since the 1970s, there have been numerous researches on outlier handling procedures, mostly involving outliers in autoregressive (AR) and autoregressive moving average (ARMA) processes. Fox (1972) introduced formal definitions of types of outliers in AR processes. This later led to the growth of outlier studies in other processes, including, ARIMA, bilinear, ARCH, and GARCH. !! ! "# $! % The presence of outliers can have deleterious effects on statistical data analysis. Outlier could affect model identification, estimation, forecasting, diagnosing and testing. Outliers may distort estimates of residual variances (Hogg, 1979 and Martin, 1980), affect estimates of variance (Pena, 1990) and not only increases residuals but also distort model specification and parameter estimates (Abraham and Chuang, 1989). Outlier may also increase error variance and reduce power of statistical tests (Franses 2004), decrease normality and bias or influence estimates (Schwager and Margolin, 1982; RECENT ADVANCES in APPLIED MATHEMATICS ISSN: 1790-2769 480 ISBN: 978-960-474-150-2