<abstract xmlns="http://www.w3.org/1999/xhtml"><p>The paper presents a method of detecting atypical observations in time series with or without seasonal fluctuations. Unlike classical methods of identifying outliers and influential observations, its essence consists in examining the impact of individual observations both on the fitted values of the model and the forecasts. The exemplification of theoretical considerations is the empirical example of modelling and forecasting daily sales of liquid fuels at X gas station in the period 2012-2014. As a predictor, a classic time series model was used, in which 7-day and 12-month cycle seasonality was described using dummy variables. The data for the period from 01.01.2012 to 30.06.2014 were for the estimation period and the second half of 2014 which was the period of empirical verification of forecasts. The obtained results were compared with other classical methods used to identify influential observations and outliers, i.e. standardized residuals, Cook distances and DFFIT. The calculations were carried out in the R environment and the Statistica package.</p></abstract>

The paper presents a method of detecting atypical observations in time series with or without seasonal fluctuations. Unlike classical methods of identifying outliers and influential observations, its essence consists in examining the impact of individual observations both on the fitted values of the model and the forecasts. The exemplification of theoretical considerations is the empirical example of modelling and forecasting daily sales of liquid fuels at X gas station in the period 2012-2014. As a predictor, a classic time series model was used, in which 7-day and 12-month cycle seasonality was described using dummy variables. The data for the period from 01.01.2012 to 30.06.2014 were for the estimation period and the second half of 2014 which was the period of empirical verification of forecasts. The obtained results were compared with other classical methods used to identify influential observations and outliers, i.e. standardized residuals, Cook distances and DFFIT. The calculations were carried out in the R environment and the Statistica package.

On the Method of Identification of Atypical Observations in Time Series

Department of Applied Mathematics in Economics, Faculty of Economics

Econometrics

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

{"article-title":"On the Method of Identification of Atypical Observations in Time Series"}

The paper presents a method of detecting atypical observations in time series with or without seasonal fluctuations. Unlike classical methods of identifying...

On the Method of Identification of Atypical Observations in Time Series

Published Online: Jul 31, 2020

Page range: 1 - 16

DOI: https://doi.org/10.15611/eada.2020.2.01

Keywords
forecasts, identification, multiple regression, time series, outliers

© 2020 Maciej Oesterreich, published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

On the Method of Identification of Atypical Observations in Time Series

Published Online: Jul 31, 2020

Page range: 1 - 16

DOI: https://doi.org/10.15611/eada.2020.2.01

Keywordsforecasts, identification, multiple regression, time series, outliers

© 2020 Maciej Oesterreich, published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.

Keywords
forecasts, identification, multiple regression, time series, outliers