• Communications for Statistical Applications and Methods
• /
• 제3권2호
• /
• pp.249-273
• /
• 1996

This paper reviews and develops several statistical models for extreme values, based on threshold methodology. Extreme values of a time series are modeled in terms of tails which are defined as truncated forms of original variables, and Markov property is imposed on the tails. Tails of the generalized extreme value distribution and a multivariate extreme value distributively, of the tails of the series. These models are then applied to real ozone data series collected in the Chicago area. A major concern is given to detecting any possible trend in the extreme values.

• 한국수자원학회:학술대회논문집
• /
• 한국수자원학회 2015년도 학술발표회
• /
• pp.537-537
• /
• 2015

In recent decades, the independence and identical distribution (iid) assumption for extreme events has been shown to be invalid in many cases because long-term climate variability resulting from phenomena such as the Pacific decadal variability and El Nino-Southern Oscillation may induce varying meteorological systems such as persistent wet years and dry years. Therefore, in the current study we propose a new parameter estimation method for probability distribution models to more accurately predict the magnitude of future extreme events when the iid assumption of probability distributions for large-scale climate variability is not adequate. The proposed parameter estimation is based on a metaheuristic approach and is derived from the objective function of the rth power probability-weighted sum of observations in increasing order. The combination of two distributions, gamma and generalized extreme value (GEV), was fitted to the GEV distribution in a simulation study. In addition, a case study examining the annual hourly maximum precipitation of all stations in South Korea was performed to evaluate the performance of the proposed approach. The results of the simulation study and case study indicate that the proposed metaheuristic parameter estimation method is an effective alternative for accurately selecting the rth power when the iid assumption of extreme hydrometeorological events is not valid for large-scale climate variability. The maximum likelihood estimate is more accurate with a low mixing probability, and the probability-weighted moment method is a moderately effective option.

• 상하수도학회지
• /
• 제26권2호
• /
• pp.321-329
• /
• 2012

In this study, statistical analysis under both stationary and non-stationary climate was conducted for rainfall data measured in Seoul. Generalised Extreme Value (GEV) distribution and Gumbel distribution were used for the analysis. Rainfall changes under the non-stationary climate were estimated by applying time variable (t) to location parameter (${\xi}$). Rainfall depths calculated in non-stationary climate increased by 1.1 to 6.2mm and 1.0 to 4.6mm for the GEV distribution and gumbel distribution respectively from those stationary forms. Changes in annual maximum rainfall were estimated with rate of change in the location parameter (${\xi}1{\cdot}t$), and temporal changes of return period were predicted. This was also available for re-evaluating the current sewer design return period. Design criteria of sewer system was newly suggested considering life expectance of the system as well as temporal changes in the return period.

• Structural Engineering and Mechanics
• /
• 제74권1호
• /
• pp.55-67
• /
• 2020

The common practice to predict the characteristic structural load effects (LEs) in long reference periods is to employ the extreme value theory (EVT) for building limit distributions. However, most applications ignore that LEs are driven by multiple loading events and thus do not have the identical distribution, a prerequisite for EVT. In this study, we propose the composite extreme value modeling approach using clustering to (a) cluster initial blended samples into finite identical distributed subsamples using the finite mixture model, expectation-maximization algorithm, and the Akaike information criterion; (b) combine limit distributions of subsamples into a composite prediction equation using the generalized Pareto distribution based on a joint threshold. The proposed approach was validated both through numerical examples with known solutions and engineering applications of bridge traffic LEs on a long-span bridge. The results indicate that a joint threshold largely benefits the composite extreme value modeling, many appropriate tail approaching models can be used, and the equation form is simply the sum of the weighted models. In numerical examples, the proposed approach using clustering generated accurate extrema prediction of any reference period compared with the known solutions, whereas the common practice of employing EVT without clustering on the mixture data showed large deviations. Real-world bridge traffic LEs are driven by multi-events and present multipeak distributions, and the proposed approach is more capable of capturing the tendency of tailed LEs than the conventional approach. The proposed approach is expected to have wide applications to general problems such as samples that are driven by multiple events and that do not have the identical distribution.

• 대한수학회보
• /
• 제53권5호
• /
• pp.1549-1566
• /
• 2016

Denote $M_n$ the maximum of n independent and identically distributed variables from the generalized short-tailed symmetric distribution. This paper shows the pointwise convergence rate of the distribution of $M_n$ to exp($\exp(-e^{-x})$) and the supremum-metric-based convergence rate as well.

• 한국수자원학회논문집
• /
• 제50권3호
• /
• pp.211-221
• /
• 2017

수문통계분야에서는 극치 사상을 해석하기 위해 generalized extreme value (GEV), generalized logistic (GLO), Gumbel (GUM) 모형과 같은 다양한 극치분포들을 사용하여 왔다. 특히 우리나라 강우 사상의 경우 다양한 극치분포 모형 중 GEV 분포와 Gumbel 분포가 비교적 적합한 것으로 알려져 있지만 하나의 형상매개변수를 가지고 있어 각 분포 모형이 나타낼 수 있는 통계적 특성에 한계를 가지고 있다. 이러한 점에서 두 개의 형상매개변수를 가지고 있어 분포 모형이 나타낼 수 있는 통계적 특성의 범위가 넓은 분포의 적용이 필요하다. 이에 본 연구에서는 두 개의 형상매개변수를 가지고 있어 다양한 통계적 특성을 표현할 수 있는 Burr XII 분포와 우리나라 620개 지점의 강우자료의 무차원 L-moment 비를 이용하여 우리나라 강우자료의 수문학적 적용성을 검토하였다. 이를 위해 Burr XII 분포의 L-moment ratio인 L-skewness와 L-kurtosis를 유도하고 그 관계식을 이용하여 L-moment diagram을 작성하고 620개 지점이 해당 영역에 포함되는 정도를 검토하여 그 적용성을 살펴보았다. 그 결과 L-skewness가 L-kurtosis보다 상대적으로 큰 한강 유역에 해당하는 지점들에 대한 Burr XII 분포의 적용성이 우수한 것으로 나타났으며, 이는 일반적으로 많이 사용되는 GEV 또는 Gumbel 분포를 대체할 수 있는 분포가 될 가능성을 보였다고 할 수 있다.

• 한국농공학회논문집
• /
• 제46권3호
• /
• pp.31-42
• /
• 2004

This research seeks to derive the design rainfalls through the L-moment with the test of homogeneity, independence and outlier of data on annual maximum daily rainfall at 38 rainfall stations in Korea. To select the appropriate distribution of annual maximum daily rainfall data by the rainfall stations, Generalized Extreme Value (GEV), Generalized Logistic (GLO), Generalized Pareto (GPA), Generalized Normal (GNO) and Pearson Type 3 (PT3) probability distributions were applied and their aptness were judged using an L-moment ratio diagram and the Kolmogorov-Smirnov (K-S) test. Parameters of appropriate distributions were estimated from the observed and simulated annual maximum daily rainfall using Monte Carlo techniques. Design rainfalls were finally derived by GEV distribution, which was proved to be more appropriate than the other distributions.

• 경영과학
• /
• 제24권2호
• /
• pp.115-126
• /
• 2007

Weekly minima of daily log returns of Korean composite stock price index 200 and its five industry-based business divisions over the period from January 1990 to December 2005 are fitted using two block-based extreme distributions: Generalized Extreme Value(GEV) and Generalized Logistic(GLO). Parameters are estimated using the probability weighted moments. Applicability of two distributions is investigated using the Monte Carlo simulation based empirical p-values of Anderson Darling test. Our empirical results indicate that both the GLO and GEV models seem to be comparably applicable to the weekly minima. These findings are against the evidences in Gettinby et al.[7], who claimed that the GEV model was not valid in many cases, and supported the significant superiority of the GLO model.

• 응용통계연구
• /
• 제28권2호
• /
• pp.137-149
• /
• 2015

본 논문에서는 비정상 극치 강수 자료에 대해 계층적 베이지안 모형을 적용하여 시간에 따른 모수의 변화를 추정하며, 미래 확률 강수량에 대한 극단값 분포를 예측하고 더 나아가 반환기간에 대한 경향과 예측 값을 얻고자 한다. 이전의 고전적 통계 방법을 통한 강수 자료의 모수 추정연구의 경우, 자료의 정상성 가정 하에 고정된 모수를 추정하는 방법으로, 최근 나타난 비정상 강수 사상과 같이 강수량이 가지는 분포의 모수적 변화가 예상되는 경우 해석상 문제가 발생한다. 이러한 문제점을 해결하기 위해 모형의 관심모수에 시간에 따른 자기 상관 선형 회귀 함수를 적합한 계층적 베이지안 모형을 고려한다. 제안된 모형의 효율성을 확인하기 위해서 1973년부터 2011년까지 39년 동안의 우리나라 여러지역의 기상 관측소에서 관측된 일일 강우량 자료가 사용하여 대표적인 극단값 분포인 Generalized Extreme Value(GEV) 분포에 적합시키고, 계층적 베이지안 모형을 이용하여 이들 분포의 모수들에 자기상관 시간모형을 소개한 후 우리나라 여러지역에 대한 반환기간에 대한 시간에 따른 경향을 확인하였다.

• Journal of the Korean Statistical Society
• /
• 제28권1호
• /
• pp.57-72
• /
• 1999

Falk(1994) showed that the asymptotic efficiency of the Pickands estimator of the extreme value index $\beta$ can considerably be improved by a simple convex combination. In this paper we propose an alternative estimator of $\beta$ which is as asymptotically efficient as the optimal convex combination of the Pickands estimators but has a better finite-sample performance. We prove consistency and asymptotic normality of the proposed estimator. Monte Carlo simulations are conducted to compare the finite-sample performances of the proposed estimator and the optimal convex combination estimator.