OUTLIERS IN TIME SERIES

In the study of economic time series we work often with large amount of data. Some observations take unusual values and are significantly different from other observations. These data are referred to as outliers. The causes of these outliers can be different. It may be a mistake in the data. Most of these errors are relatively easy to remove if you have a careful control of data. On the other hand, there may occur the data that are observed properly and still have the character of outliers. These observations make it difficult to build right models for time series and can misrepresent any predictions. In the article we describe the different types of outliers and their integration into the stochastic time series. We work with transfer function models, which are mainly based on ARIMA models and on linear dynamic models. In our paper we show that each outlier must necessarily leave specific track in residuals. This fact is the base for detection outliers, to describe their type and methods of integration into the model. We describe in our article the theory of outliers including their detection. Described theory will be illustrated on practical data.