Title: A new clipping approach for parameter estimation in AR(p)
Authors: Samuel Flimmel - University of Economics in Prague (Czech Republic) [presenting]
Jiri Prochazka - University of Economics, Prague (Czech Republic)
Abstract: With an increasing number of observations the probability of outlier presence also rises. This is a problem we face nowadays in many fields when working with big data. As it is known, standard methods are not able to work correctly with outliers, and, consequently, standard estimates are often biased. Therefore, sufficiently robust methods gain on importance. Autoregressive process AR(p) is well known and frequently used in statistics and economical modelling. One of the requirements for working with AR(p) models is an ability to estimate the parameters of the model correctly. In this poster, authors present a new robust method for parameter estimation in AR(p) models. The method is based on clipping the original time series and working with a binary time series instead. The clipping helps to deal with outliers, and, therefore, the estimation is not affected as much as when using a standard method. The new method is described and compared with existing robust methods using a simulation study performed in the R statistical software.