Publication detail
Extreme Value Estimation under the Presence of Short-Time Dependence
HOLEŠOVSKÝ, J. MICHÁLEK, J.
Czech title
Odhady extrémních hodnot za vlivu krátkodobé závislosti
English title
Extreme Value Estimation under the Presence of Short-Time Dependence
Type
abstract
Language
en
Original abstract
Study of extremes becomes crucial in many application areas, partially due to their harmful effects on a system reliability. For the purpose of prediction and frequency estimation of an extreme events, the extreme value theory is usually applied. However, the commonly used estimating methods (e.g. maximum likelihood method) are based on an independent observations. The presence of a serial dependence requires a careful investigation. With increasing dependence the extremes show higher tendency to cluster, which can significantly devaluate the analysis if not taken into account. The treatment with dependence is based either on a declustering techniques, which filter out the presence of dependence, or on its proper identification. The first one usually leads to significant reduction of number of observations and thus to a loss of information. The objective of the contribution is to discuss the latter approach, accompanied with estimation of so-called extremal index [1], a measure of extremes clustering. Hereby, we follow the techiques proposed in [1] and [2] with observation at long lags assuming to be independent. The properties of both approaches are illustrated by extreme value frequency estimation on real data. [1] Ancona-Navarrete, M., Tawn, J. A. (2000). A Comparison of Methods for Estimating the Extremal Index. In Extremes 3(1): 5-38. [2] Fawcett, L., Walshaw, D. (2012). Estimating Return Levels from Serially Dependent Extremes. In Environmetrics 23: 272-283.
Czech abstract
Studium extrémních hodnot se stává důležitou částí mnoha odvětví, zejména pro jejich negativní vliv na spolehlivost systému. Pro účely predikce a odhadu frekvence výskytu těchto extrémních hodnot je obvykle aplikována teorie extrémních hodnot. Zde se ovšem často využívá metod odhadů, které jsou založeny na nezávislých pozorováních (např. metoda maximální věrohodnosti). Přítomnost závislosti tedy vyžaduje důkladné studium. Rostoucí závislost roste také tendence ke shlukování extrémních hodnot. Nejsou-li tyto shluky brány v potaz, mohou významně znehodnotit výsledky následných analýz. Přítomnost závislosti vyžaduje buď vývoj nějakých declustrovacích technik, které danou závislost odstraní, nebo její řádnou identifikaci. První přístup běžně vede k odstranění velké části pozorování a tak k redukci obsažené informace. Cílem tohoto příspěvku je zejména diskuze nad druhým přístupem, doprovázená možnostmi odhadu extrémálního indexu [1], jakési míry shlukování. Navazujeme na techniky popsané v [1] a [2] a předpokládáme dostatečně vzdálená pozorování jako nezávislá. Vlastnosti obou přístupů jsou ilustrovány na odhadech návratových úrovní na reálných datech. [1] Ancona-Navarrete, M., Tawn, J. A. (2000). A Comparison of Methods for Estimating the Extremal Index. In Extremes 3(1): 5-38. [2] Fawcett, L., Walshaw, D. (2012). Estimating Return Levels from Serially Dependent Extremes. In Environmetrics 23: 272-283.
English abstract
Study of extremes becomes crucial in many application areas, partially due to their harmful effects on a system reliability. For the purpose of prediction and frequency estimation of an extreme events, the extreme value theory is usually applied. However, the commonly used estimating methods (e.g. maximum likelihood method) are based on an independent observations. The presence of a serial dependence requires a careful investigation. With increasing dependence the extremes show higher tendency to cluster, which can significantly devaluate the analysis if not taken into account. The treatment with dependence is based either on a declustering techniques, which filter out the presence of dependence, or on its proper identification. The first one usually leads to significant reduction of number of observations and thus to a loss of information. The objective of the contribution is to discuss the latter approach, accompanied with estimation of so-called extremal index [1], a measure of extremes clustering. Hereby, we follow the techiques proposed in [1] and [2] with observation at long lags assuming to be independent. The properties of both approaches are illustrated by extreme value frequency estimation on real data. [1] Ancona-Navarrete, M., Tawn, J. A. (2000). A Comparison of Methods for Estimating the Extremal Index. In Extremes 3(1): 5-38. [2] Fawcett, L., Walshaw, D. (2012). Estimating Return Levels from Serially Dependent Extremes. In Environmetrics 23: 272-283.
Keywords in Czech
Extrémní hodnoty, závislé řady, extremální index, declustering.
Keywords in English
Extreme value, dependent series, extremal index, declustering.
Released
18.08.2014
Location
Brno
ISBN
978-80-7401-090-3
ISSN
NEUVEDENO
Book
Biometric Methods and Models in Current Science and Research - Proceedings
Pages from–to
35–35
Pages count
1