Title: A new approach for ARMA order estimation based on clipping
Authors: Samuel Flimmel - University of Economics in Prague (Czech Republic) [presenting]
Jiri Prochazka - University of Economics, Prague (Czech Republic)
Abstract: Problems related to big data are faced in many fields nowadays. Enormous amounts of information are being stored every second, but processing all these data using standard methods becomes more and more problematic. With an increasing number of observations, the probability of outlier presence also rises, and, therefore, working with sufficiently robust methods gains on importance as well. Since standard methods are not always able to deal with outliers correctly, standard estimates are often biased. ARMA processes are well known and widely used in theoretical and also practical world of statistical modeling. Often, it is necessary to estimate the order of a given ARMA process. Usually, it is the second step in Box-Jenkins method, performed immediately after solving the stationarity and seasonality. In this poster, we present a new robust method for ARMA order estimation. The method is based on clipping the original time series and working with a binary time series instead. This detour provides the required robustness and helps to face outliers. We describe this new method and compare it with existing methods using a simulation study performed in the R statistical software.