Empirical cumulant function based parameter estimation in stable distributions
Failid
Kuupäev
2019-01-07
Autorid
Ajakirja pealkiri
Ajakirja ISSN
Köite pealkiri
Kirjastaja
Abstrakt
Mitmetes eluvaldkondades, näiteks klimatoloogias, füüsikas ning kindlustus- ja finantssektoris, on vajadus ebasümmeetriliste ning ekstreemseid väärtusi sisaldavate andmete modelleerimiseks. Klassikaline normaaljaotus ei sobitu sellistele andmetele ning kasutama peab alternatiivseid võimalusi. Stabiilsed jaotused, mille üheks erijuhuks on ka normaaljaotus, võimaldavad kirjeldada nii sümmeetrilisi kui ebasümmeetrilisi protsesse kui ka arvesse võtta protsessides esineva varieeruvuse dünaamika ja ekstreemsed kõikumised. Viimastel aastakümnetel on stabiilsetest jaotustest esitatud mitmeid uurimusi kuid parameetrite hindamine empiiriliste andmete põhjal on senini väljakutset pakkuv. Olemasolevad meetodid ei ole statistika tarkvaras tihtipeale vabalt kättesaadavad ja see omakorda takistab stabiilsete jaotuste laialdasemat kasutamist. Doktoritöö aluseks on hindamismeetod, mis on arvutuslikult lihtne ja vahetult rakendatav, kuid hinnangud sõltuvad vabalt valitavatest reaalarvulistest argumentidest. Argumentide erinev valik mõjutab tulemusi märgatavalt ja seetõttu ei ole vastav hindamismeetod rakendustes kasutust leidnud. Doktoritöös esitatakse kõnealusest hindamismeetodist parendatud versioon ning antakse põhjendatud soovitused sobivate argumentide valikuks. Doktoritöö tulemusena saab öelda, et stabiilsete jaotuste parameetrite hindamiseks väljapakutud arvutuslikult lihtne protseduur annab samaväärseid või paremaid tulemusi võrreldes keerukamate algoritmiliste meetoditega ning on praktikas hästi kasutatav.
In diverse fields of business, science, and engineering there is a need for modelling asymmetric data with extremes. The classical normal distribution is not suitable in such cases and alternative approaches should be used. Stable distributions, with normal distribution as a special case, can capture the fuzzy dynamics and large fluctuations that result from symmetric and asymmetric stochastic processes. However, a challenging problem in applying stable distributions to practical problems is the estimation of their parameters from the empirical data. To address this problem, in thesis a class of closed-form estimators is studied. The estimation procedure is simple but estimators depend on arbitrary selection of the number of arguments. Different selections of arguments yield different estimates thus making the method not very useful in practice. In thesis an improved version of the estimation method is provided and, based on empirical and theoretical results, suggestions on the selection of arguments are given. As the result of this dissertation it is found that the proposed computationally simple procedure for estimating the parameters of stable laws compares favourably with the more complicated algorithmic methods and can be successfully used in practice.
In diverse fields of business, science, and engineering there is a need for modelling asymmetric data with extremes. The classical normal distribution is not suitable in such cases and alternative approaches should be used. Stable distributions, with normal distribution as a special case, can capture the fuzzy dynamics and large fluctuations that result from symmetric and asymmetric stochastic processes. However, a challenging problem in applying stable distributions to practical problems is the estimation of their parameters from the empirical data. To address this problem, in thesis a class of closed-form estimators is studied. The estimation procedure is simple but estimators depend on arbitrary selection of the number of arguments. Different selections of arguments yield different estimates thus making the method not very useful in practice. In thesis an improved version of the estimation method is provided and, based on empirical and theoretical results, suggestions on the selection of arguments are given. As the result of this dissertation it is found that the proposed computationally simple procedure for estimating the parameters of stable laws compares favourably with the more complicated algorithmic methods and can be successfully used in practice.
Kirjeldus
Märksõnad
probability distributions