ABSTRACT
This chapter presents a method of identifying outliers in time series data. An outlier is an observation that is aberrant or unusually different from the rest of the observations. Often outliers arise in real data from known causes such as a change in pricing policy, a business promotion, or a labor strike. When the researcher can easily identify the reason for an outlier, the outlier itself can be identified. Once it is identified, steps to remove its influence on the time series model used for forecasting can be undertaken. This process of removing the influence of outliers makes the models robust to the effects of outliers. The influence of an outlier is apparent in exponential smoothing forecasting methods. The empirical transform outlier identification technique provides an easily implemented preprocessing step in identifying such outliers. Clearly, one never knows whether a particular observation is a true outlier. These graphical techniques can be used on an ergodic process without removing autocorrelations first.
