• Communications for Statistical Applications and Methods
• /
• v.30 no.6
• /
• pp.531-550
• /
• 2023

The estimation of extreme quantiles is one of the main objectives of statistics of extremes (which deals with the estimation of rare events). In this paper, a robust estimator of extreme quantile of a heavy-tailed distribution is considered. The estimator is obtained through the minimum density power divergence criterion on an exponential regression model. The proposed estimator was compared with two estimators of extreme quantiles in the literature in a simulation study. The results show that the proposed estimator is stable to the choice of the number of top order statistics and show lesser bias and mean square error compared to the existing extreme quantile estimators. Practical application of the proposed estimator is illustrated with data from the pedochemical and insurance industries.

• The Korean Journal of Applied Statistics
• /
• v.25 no.1
• /
• pp.101-114
• /
• 2012

In car insurance, the loss ratio is the ratio of total losses paid out in claims divided by the total earned premiums. In order to minimize the loss to the insurance company, estimating extreme quantiles of loss ratio distribution is necessary because the loss ratio has essential prot and loss information. Like other types of insurance related datasets, the distribution of the loss ratio has heavy-tailed distribution. The Peaks over Threshold(POT) and the Hill estimator are commonly used to estimate extreme quantiles for heavy-tailed distribution. This article compares and analyzes the performances of various kinds of parameter estimating methods by using a simulation and the real loss ratio of car insurance data. In addition, we estimate extreme quantiles using the Hill estimator. As a result, the simulation and the loss ratio data applications demonstrate that the POT method estimates quantiles more accurately than the Hill estimation method in most cases. Moreover, MLE, Zhang, NLS-2 methods show the best performances among the methods of the GPD parameters estimation.

• Communications for Statistical Applications and Methods
• /
• v.16 no.5
• /
• pp.803-812
• /
• 2009

The application of extreme value theory to financial data is a fairly recent innovation. The classical annual maximum method is to fit the generalized extreme value distribution to the annual maxima of a data series. An alterative modern method, the so-called threshold method, is to fit the generalized Pareto distribution to the excesses over a high threshold from the data series. A more substantial variant is to take the point-process viewpoint of high-level exceedances. That is, the exceedance times and excess values of a high threshold are viewed as a two-dimensional point process whose limiting form is a non-homogeneous Poisson process. In this paper, we apply the two-dimensional non-homogeneous Poisson process model to daily losses, daily negative log-returns, in the data series of KBW/USD exchange rate, collected from January 4th, 1982 until December 31 st, 2008. The main question is how to estimate extreme quantiles of losses such as the 10-year or 50-year return level.

• Communications for Statistical Applications and Methods
• /
• v.30 no.3
• /
• pp.259-272
• /
• 2023

In this paper, we suggest a new method for the prediction of sharp changes in particulate matter (PM₁₀) using quantile mapping. To predict the current PM₁₀ density in Seoul, we consider PM₁₀ and precipitation in Baengnyeong and Ganghwa monitoring stations observed a few hours before. For the PM₁₀ distribution estimation, we use the extreme value mixture model, which is a combination of conventional probability distributions and the generalized Pareto distribution. Furthermore, we also consider a quantile generalized additive model (QGAM) for the relationship modeling between precipitation and PM₁₀. To prove the validity of our proposed model, we conducted a simulation study and showed that the proposed method gives lower mean absolute differences. Real data analysis shows that the proposed method could give a more accurate prediction when there are sharp changes in PM₁₀ in Seoul.

• Smart Structures and Systems
• /
• v.20 no.1
• /
• pp.11-22
• /
• 2017

The long-term effect of ambient temperature on bridge strain is an important and challenging problem. To investigate this issue, one year data of strain and ambient temperature of a long-span cable-stayed bridge is studied in this paper. The measured strain-time history is decomposed into two parts to obtain the strains due to vehicle load and temperature alone. A linear regression model between the temperature and the strain due to temperature is established. It is shown that for every $1^{\circ}C$ increase in temperature, the stress is increased by 0.148 MPa. Furthmore, the extreme value distributions of the strains due to vehicle load, temperature and the combination effect of them during the remaining service period are estimated by the average conditional exceedance rate approach. This approach avoids the problem of declustering of data to ensure independence. The estimated results demonstrate that the 95% quantile of the extreme strain distribution due to temperature is up to $1.488{\times}10^{-4}$ which is 2.38 times larger than that due to vehicle load. The study also indicates that the estimated extreme strain can reflect the long-term effect of temperature on bridge strain state, which has reference significance for the reliability estimation and safety assessment.

• Communications for Statistical Applications and Methods
• /
• v.24 no.5
• /
• pp.493-505
• /
• 2017

Maximum likelihood estimation (MLE) of the generalized extreme value distribution (GEVD) is known to sometimes over-estimate the positive value of the shape parameter for the small sample size. The maximum penalized likelihood estimation (MPLE) with Beta penalty function was proposed by some researchers to overcome this problem. But the determination of the hyperparameters (HP) in Beta penalty function is still an issue. This paper presents some data adaptive methods to select the HP of Beta penalty function in the MPLE framework. The idea is to let the data tell us what HP to use. For given data, the optimal HP is obtained from the minimum distance between the MLE and MPLE. A bootstrap-based method is also proposed. These methods are compared with existing approaches. The performance evaluation experiments for GEVD by Monte Carlo simulation show that the proposed methods work well for bias and mean squared error. The methods are applied to Blackstone river data and Korean heavy rainfall data to show better performance over MLE, the method of L-moments estimator, and existing MPLEs.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2017.05a
• /
• pp.399-400
• /
• 2017

홍수나 가뭄 등 극치 현상의 통계분석 및 빈도해석에 있어 극치분포형이 널리 사용되고 있으며, 이러한 극치분포형의 특성을 이해하기 위해서는 분포형의 오른쪽 꼬리(right tail) 부분 특성을 자세히 분석할 필요가 있다. 이에 따라 본 연구에서는 Monte Carlo 모의를 통하여 다양한 극치분포형의 오른쪽 꼬리 부분의 통계적 특성 및 그 예측 능력을 연구하였다. 극치분포형으로는 우리나라 확률수문량 산정에 널리 활용되고 있는 generalized extreme value (GEV), Gumbel, generalized logistic 분포를 사용하였으며, 매개변수 산정 방법으로는 확률가중모멘트법을 사용하였다. 모의실험의 모분포로는 수문빈도해석에서 많이 사용되는 GEV 분포를 사용하였고, 30년 이상 자료를 보유한 기상청 지점 자료의 왜곡도를 조사하여 모의실험에 사용되는 모집단의 왜곡도로 가정하여 표본 자료를 발생시켰다. 예측 능력의 평가는 재현기간 10~1000년의 확률수문량을 왜곡도계수를 고려한 GEV 도시위치공식을 이용하여 GEV 확률지에 도시하고, 평균제곱근오차(root mean square error), 편의(bias), 평균상대오차(mean relative difference), 평균절대상대오차(mean absolute relative difference)를 이용하여 최적 분포형을 선정함으로써 이루어진다. 또한 예측 능력 평가결과의 타당성 확인을 위해 극치분포형의 적합정도를 잘 나타낸다고 알려진 modified Anderson-Darling 방법의 검정결과와 비교하여 적절성을 확인하였다.

• Journal of Korea Water Resources Association
• /
• v.45 no.8
• /
• pp.827-837
• /
• 2012

The estimation of the rainfall quantile is of great importance in designing hydrologic structures. Conventionally, the rainfall quantile is estimated by univariate frequency analysis with an appropriate probability distribution. There is a limitation in which duration of rainfall is restrictive. To overcome this limitation, bivariate frequency analysis by using 3 copula models is performed in this study. Annual maximum rainfall events in 5 stations are used for frequency analysis and rainfall depth and duration are used as random variables. Gumbel (GUM), generalized logistic (GLO) distributions are applied for rainfall depth and generalized extreme value (GEV), GUM, GLO distributions are applied for rainfall duration. Copula models used in this study are Frank, Joe, and Gumbel-Hougaard models. Maximum pseudo-likelihood estimation method is used to estimate the parameter of copula, and the method of probability weighted moments is used to estimate the parameters of marginal distributions. Rainfall quantile from this procedure is compared with various marginal distributions and copula models. As a result, in change of marginal distribution, distribution of duration does not significantly affect on rainfall quantile. There are slight differences depending on the distribution of rainfall depth. In the case which the marginal distribution of rainfall depth is GUM, there is more significantly increasing along the return period than GLO. Comparing with rainfall quantiles from each copula model, Joe and Gumbel-Hougaard models show similar trend while Frank model shows rapidly increasing trend with increment of return period.

• The Korean Journal of Applied Statistics
• /
• v.31 no.3
• /
• pp.343-352
• /
• 2018

We can use quantile regression and expectile regression analysis to estimate trends in extreme regions as well as the average trends of response variables in given explanatory variables. In this paper, we compare the performance between the parametric and nonparametric methods for expectile regression. We introduce each estimation method and analyze through various simulations and the application to real data. The nonparametric model showed better results if the model is complex and difficult to deduce the relationship between variables. The use of nonparametric methods can be recommended in terms of the difficulty of assuming a parametric model in expectile regression.

• Communications for Statistical Applications and Methods
• /
• v.31 no.1
• /
• pp.37-53
• /
• 2024

In this study, flood records from 79 sites across Thailand were analyzed to estimate flood indices using the regional frequency analysis based on the L-moments method. Observation sites were grouped into homogeneous regions using k-means and Ward's clustering techniques. Among various distributions evaluated, the generalized extreme value distribution emerged as the most appropriate for certain regions. Regional growth curves were subsequently established for each delineated region. Furthermore, 20- and 100-year return values were derived to illustrate the recurrence intervals of maximum rainfall across Thailand. The predicted return values tend to increase at each site, which is associated with growth curves that could describe an increasing long-term predictive pattern. The findings of this study hold significant implications for water management strategies and the design of flood mitigation structures in the country.