Outlier Detection Of Time Series Models And Applications In Statistical Monitoring

Various types of outlying observations are produced in time series data because they are subject to some unexpected interventions.The presence of outliers may have a significant impact on model identification,parameter estimation,model diagnostics,and forecasting.This paper mainly introduces the method of detecting outliers in time series and applies that method into actual statistical monitoring data.Considering that much statistical monitoring data can be modeled effectively based on linear time series models,an iterative procedure produced by Chen and Liu about detecting outliers in linear time series models is introduced in this paper.The time series generated by the simulation and actual statistical monitoring data are investigated based on that procedure.The procedure is to obtain joint estimates of model parameters and outlier effects.The key step in the procedure is to examine the maximum absolute value of the standardized statistics of the outlier effects,so the autoregressive moving average(ARMA)model with outlier effects is introduced at first to obtain those statistics,then the foundation and major steps of the procedure are introduced,and the way about how to detect outliers based on that procedure in this paper is introduced too.It is demonstrated that the procedure performs effectively in detecting outliers in simulated data.As for actual statistical monitoring data,five outliers are found in consumer price index data and one outlier is found in time series about the production price index of pig,and all outliers are successfully explained.