![]() | يوجد فقط 14 صفحة متاحة للعرض العام |
المستخلص Time senes observations are sometimes influenced by interruptive events, such as strikes, outbreaks of war, sudden political or economic crises, unexpected heat or cold waves, or even unnoticed errors of typing and recording. The consequences of these interruptive events create spurious observations that are inconsistent with the rest of the senes. Such observations are usually referred to as outliers. When the timing and causes of interruptions are known, their effects can be accounted for by using the intervention analysis approach. In practice, however, the timing of interruptive events are usually unknown. The4 existence of outliers, in time series data may cause serious effects on the usual statistics of identification; the sample autocorrelation function (SACF) and the sample partial autocorrelation function (SPACF). In particular, it may cause serious biases in these sample statistics, thus jeopardising their usefulness as tools for model identification. The biases depend upon the number, type, magnitude and relative position of outliers, and for typically sized data sets, their presence can lead to under or over specification of orders of the autoregressive and moving average processes ¢(B) and () (B) respectively, where ¢ (B) and () (B) are two polynomials of degree p and q respectively It is important, therefore, to have procedures that will detect and remove such outliers effects. The detection of time series outliers was first studied by Fox (1972) , where two statistical models, additive and innovational, were introduced. Other references on this topic include Abraham and Box (1979), Marttin (1980), Chang and Tiao (1983), Tsay (1986), chang Tiao and chen (1988) , Abraham and chuang (1983) , chen and Liu (1993), Sakata and white (1989) and Glendinning (1998). The rationale for outlier detection III time Series data may be briefly stated as : - Better understanding of the data. - Better specification and Estimation. - Better forecasts. |