Specification of Periodic Autocovariance Structures in the Presence of Outliers.docVIP

Specification of Periodic Autocovariance Structures in the Presence of Outliers Abstract： This paper focuses on the specification of periodic autocovariance structures in the presence of outliers， we evaluate autocovariance structures using various outliers’generating models. The analytical results indicate that outliers affect the estimates of periodic autocovariance function（PACVF） due to biases and inflated standard errors. Robust autocovariance structures that accommodate the influence of outliers in different periodic processes are proposed. Wefit AR （1） model using both the conventional and Jacknife autocovariance structures； the latter shows high precision in the standard errors of the estimates. We demonstrate our proposed methodology with the precipitation data from Maun Airport in Botswana， and the empirical study supports our theoreticalfindings. Key words： Periodic processes； Outliers models； ARIMA； Jacknife and Autocovariance Structures 1. INTRODUCTION The adjustment of a time series model to a dataset is crucial in time series in modeling； and this should follow some basic steps as identification， estimation， diagnostic checking and model selection. Outliers can affect model identification tools， dependence structures of time series models and estimation of model parameters （Chernick et al.， 1982）. The choice of suitable time series model is important when searching for outlying observations in periodic processes， because， on one hand， a large residual variance caused by overall lack offit would result in underidentification of outliers， while on the other hand， a less adequate model unable to explain the local behavior of the series would yield a single large residuals， resulting in over-identification， Battaglia and Orfei （2005）. It is a fact that identification can be marred by the presence of more than one outlier in time series due to masking effect； to alleviate the aforementioned challenges this paper focuses on evaluation of autocovariance stru