Andreea MIRICĂ – Lecturer, PhD. (miricaandreea89@gmail.com)
Bucharest University of Economic Studies, Bucharest, Romania
Octavian CEBAN – PhD. Candidate (octavianceban1995@gmail.com)
Bucharest University of Economic Studies, Bucharest, Romania
Traian-Ovidiu CALOTĂ – Associate Professor, PhD. (traian.calota@infofisc.ro)
Titu Maiorescu University, Faculty of Finances, Banks, Accounting and Busineess Administration, Bucharest, Romania
Roxana-Violeta PARTAS-CIOLAN – PhD. Candidate (roxana.partas@gmail.com)
Bucharest University of Economic Studies, Bucharest, Romania
Liliana CATRINA – PhD. Candidate (liliana.catrina@gmail.com)
Bucharest University of Economic Studies, Bucharest, Romania
Introduction
Economic activities have been severely affected by the COVID-19 outbreak. This causes the presence of outliers at the end point of time series, which may affect the revision of the seasonally adjusted series each time new data becomes available (Eurostat, 2020a). Given the special circumstances, as it is impossible to predict the development of the crisis, Eurostat considers that modelling outliers based solely on statistic criteria using the automatic is an acceptable solution; however, using statistical criteria and economic information is recommended (Eurostat, 2020a). It should also be noted that revisions of the increase rates calculated based on seasonally adjusted data should be kept reasonable (Mirica et al. 2016).). The economy is prone to various shocks, especially on long term, such as war crises, pandemics, and a “good” statistical approach can be compromised by fitting to data ignoring some assumptions or a time-series characteristic (Hendry et al. 2011). Using outliers when modelling implies high risk of inflated errors (Osborne at al. 2004). This, along with the fact that only 8% of the researchers report checks for outliers in their work as outlined by an empirical study (Osborne et al 2001), leads to concerns. The main problem with outliers is that they appear to be generated from a distribution that is different from the one that is modelled and based on which the assumptions are made. An illustration of this situation is presented by Hawkins D.M. who emphasizes that “an outlier is an observation which deviates so much from other observations as to arouse suspicions that it was generated by a different mechanism” (Hawkins, 1980, p.1). Modelling with outliers increase error variance, reduce the power of statistical test, and can decrease the normality (Osborne 2004). Besides the variance, outliers can add bias to the predictions or estimators (Zimmerman 1994). These uncontrolled or unexpected interventions leads to a “moderate to significant impact on the effectiveness of the standard methodology for time series analysis with respect to model identification, estimation and forecasting” (Chung and Lon-Mu 1993).
