Generalized Pareto Model Research Articles

Modeling air quality in main cities of Peninsular Malaysia by using a generalized Pareto model.

The air pollution index (API) is an important figure used for measuring the quality of air in the environment. The API is determined based on the highest average value of individual indices for all the variables which include sulfur dioxide (SO2), nitrogen dioxide (NO2), carbon monoxide (CO), ozone (O3), and suspended particulate matter (PM10) at a particular hour. API values that exceed the limit of 100 units indicate an unhealthy status for the exposed environment. This study investigates the risk of occurrences of API values greater than 100 units for eight urban areas in Peninsular Malaysia for the period of January 2004 to December 2014. An extreme value model, known as the generalized Pareto distribution (GPD), has been fitted to the API values found. Based on the fitted model, return period for describing the occurrences of API exceeding 100 in the different cities has been computed as the indicator of risk. The results obtained indicated that most of the urban areas considered have a very small risk of occurrence of the unhealthy events, except for Kuala Lumpur, Malacca, and Klang. However, among these three cities, it is found that Klang has the highest risk. Based on all the results obtained, the air quality standard in urban areas of Peninsular Malaysia falls within healthy limits to human beings.

Response of Extreme Hydrological Events to Climate Change in the Water Source Area for the Middle Route of South-to-North Water Diversion Project

As the water source area for the middle route of China’s South-to-North Water Diversion Project, the upper Hanjiang basin is of central concern for future management of the country’s water resources. The upper Hanjiang is also one of the most flood-prone rivers in China. This paper explores the process of extreme floods by using multivariate analysis to characterize flood and precipitation event data in combination, for historical data and simulated data from global climate models. The results suggested that the generalized extreme value and Gamma models better simulated the extreme precipitation and flood volume sequence than the generalized Pareto model for the annual maximum series, while the generalized Pareto distribution model was the best-fit model for peaks over threshold series. For the two-dimensional joint distributions of precipitation and flood volume, the Frank Copula was preferred in simulation of the annual maximum flood series whereas the Gumbel Copula was the most appropriate function to simulate the points over threshold flood series. We concluded that, compared with the traditional univariate approach, multivariate statistical analysis produced flood estimates that were more physically based and statistically sound and carried lower risk for flood design purposes.

Stochastic evaluation of simple pairing approaches to reconstruct incomplete rainfall time series

Two-station pairing approaches are routinely used to infill missing information in incomplete rainfall databases. We evaluated the performance of three simple methodologies to reconstruct incomplete time series in presence of variable nonlinear correlation between data pairs. Nonlinearity stems from the statistics describing the marginal peak-over-threshold (POT) values of rainfall events. A Monte Carlo analysis was developed to quantitatively assess expected errors from the use of chronological pairing (CP) with linear and nonlinear regression and frequency pairing (FP). CP is based on a priori selection of regression functions, while FP is based on matching the probability of non-exceedance of an event from one time series with the probability of non-exceedance of a similar event from another time series. We adopted a generalized Pareto (GP) model to describe POT events, and a t-copula algorithm to generate reference nonlinearly correlated pairs of random temporal distributions distributed according with the GP model. The results suggest that the optimal methodology strongly depends on GP statistics. In general, CP seems to provide the lowest errors when GP statistics were similar and correlation became linear; we found that a power-2 function performs well for the selected statistics when the number of missing points is limited. FP outperforms the other methods when POT statistics are different and variables are markedly nonlinearly correlated. Ensemble-based results seem to be supported by the analysis of observed precipitation at two real-world gauge stations.

Modeling the Dependence between Extreme Operational Losses and Economic Factors: A Conditional Semi-Parametric Generalized Pareto Approach

We model the severity distribution of operational loss data, conditionally to some covariates. Indeed, previous studies suggest that this distribution might be influenced by firm-specific factors. We introduce a conditional Generalized Pareto model for the tail of the severity distribution, where the shape parameter is an unknown function of a linear combination of the covariates. More precisely, we rely on a single-index assumption to perform a dimension reduction that enables to use univariate nonparametric techniques. Hence, we suffer neither from too strong parametric assumption nor from the curse of dimensionality. We apply this methodology on a novel database provided by the bank UniCredit. We use firm-specific factors to estimate the conditional shape parameter. Our analysis suggests that the leverage ratio of the company and the proportion of the revenue coming from fees have an important impact on the probability of suffering from large operational losses.

Distributions of return values for ocean wave characteristics in the South China Sea using directional–seasonal extreme value analysis

Estimation of ocean environmental return values is critical to the safety and reliability of marine and coastal structures. For ocean waves and storm severity, return values are typically estimated by extreme value analysis of time series of measured or hindcast sea state significant wave height HS. For a single location, this analysis is complicated by the serial dependence of HS in time and its non‐stationarity with respect to multiple covariates, particularly direction and season.Here, we report a non‐stationary extreme value analysis of storm peak significant wave height , assumed temporally independent given covariates, incorporating directional and seasonal effects using a spline‐based methodology incorporating an ensemble of models for different extreme value thresholds. Quantile regression is used to estimate suitable thresholds. For each threshold, a Poisson process is used to estimate the rate of occurrence of threshold exceedances, and a generalised Pareto model characterises the magnitude of threshold exceedances. Covariate effects are incorporated at each stage using penalised tensor products of B‐splines to give smooth model parameter variation as a function of covariates. Optimal smoothing penalties are selected using cross‐validation, and uncertainty is quantified using bias‐corrected and accelerated bootstrap resampling.We use the model to estimate environmental return values for a location in the Makassar Strait, in the South China Sea. Return values distributions for are estimated by simulation under the threshold ensemble model. Return values for HS are then estimated by simulating intra‐storm trajectories of HS consistent with the characteristics of the simulated storm peak events using a matching procedure. Return values for maximum individual crest elevation C are estimated by marginalisation using a pre‐specified conditional distribution for C given HS and other sea state parameters. Model validation is performed by comparing confidence intervals for cumulative distribution functions of and HS for the period of the data with empirical sample‐based estimates. Copyright © 2015 John Wiley & Sons, Ltd.

Comparison of nonstationary generalized logistic models based on Monte Carlo simulation

Abstract. Recently, the evidences of climate change have been observed in hydrologic data such as rainfall and flow data. The time-dependent characteristics of statistics in hydrologic data are widely defined as nonstationarity. Therefore, various nonstationary GEV and generalized Pareto models have been suggested for frequency analysis of nonstationary annual maximum and POT (peak-over-threshold) data, respectively. However, the alternative models are required for nonstatinoary frequency analysis because of analyzing the complex characteristics of nonstationary data based on climate change. This study proposed the nonstationary generalized logistic model including time-dependent parameters. The parameters of proposed model are estimated using the method of maximum likelihood based on the Newton-Raphson method. In addition, the proposed model is compared by Monte Carlo simulation to investigate the characteristics of models and applicability.

일반화 파레토 모형에서의 베이지안 예측

일반화 파레토 모형에서의 베이지안 예측

Return level estimation from non-stationary spatial data exhibiting multidimensional covariate effects

Careful modelling of non-stationarity is critical to reliable specification of marine and coastal design criteria. We present a spline based methodology to incorporate spatial, directional, temporal and other covariate effects in extreme value models for environmental variables such as storm severity. For storm peak significant wave height events, the approach uses quantile regression to estimate a suitable extremal threshold, a Poisson process model for the rate of occurrence of threshold exceedances, and a generalised Pareto model for size of threshold exceedances. Multidimensional covariate effects are incorporated at each stage using penalised (tensor products of) B-splines to give smooth model parameter variation as a function of multiple covariates. Optimal smoothing penalties are selected using cross-validation, and model uncertainty is quantified using a bootstrap re-sampling procedure. The method is applied to estimate return values for large spatial neighbourhoods of locations, incorporating spatial and directional effects. Extensions to joint modelling of multivariate extremes, incorporating extremal spatial dependence (using max-stable processes) or more general extremal dependence (using the conditional extremes approach) are outlined.

On peaks-over-threshold modeling of floods with zero-inflated Poisson arrivals under stationarity and nonstationarity

The peaks-over-threshold (POT) model with Poisson arrivals and generalized Pareto (GP) distributed exceedances remains a popular and useful tool for modelling hydrologic extremes. The use of the Poisson–GP model for flood frequency analysis requires the validation of the hypothesis that the distribution of the annual number of flood events may be described by a Poisson distribution. Such hypothesis is not always valid in practical applications. The present study concerns the use of an alternative distribution for modelling the annual number of floods—the zero-inflated Poisson (ZIP) distribution with two parameters. A ZIP–GP model for flood frequency analysis is proposed. This model is less restrictive than the Poisson–GP model since it allows for a more accurate description of the occurrence process in a POT framework if the fraction of years with no exceedances is significantly higher than the theoretical mass at zero of the Poisson distribution. Furthermore, a nonstationary model (NSZIP–GP) is presented, in which the parameters of the ZIP are allowed to change in time as a function of a covariate, which, even for stationary peak magnitudes, affects the annual maximum flood quantiles with a given non-exceedance probability. Applications of the ZIP–GP model to flood data from Northern Portugal and the evaluation of its performance relative to the Poisson–GP model, including assessments of quantile uncertainty, are presented. An illustrative application of the NSZIP–GP model, using the North Atlantic Oscillation as a covariate is also presented. The applications of both models include assessment of quantile uncertainty.

Comparing generalized Pareto models fitted to extreme observations: an application to the largest temperatures in Spain

In this paper, a subsampling-based testing procedure for the comparison of the exceedance distributions of stationary time series is introduced. The proposed testing procedure has a number of advantages including the fact that the assumption of stationary can be relaxed for some specific forms of non-stationary and also that the two time series are not required to be independently-generated. For this purpose, a test based on the Kolmogorov–Smirnov and the L 1-Wasserstein distances between generalized Pareto distributions is introduced and studied in some detail. The performance of the testing procedure is illustrated through a simulation study and with an empirical application to a set of data concerning daily maximum temperature in the 17 autonomous communities of Spain for the period 1990–2004. The autonomous communities were clustered according to the similarities of the fitted generalized Pareto models and then mapped. The cluster analysis reveals a clear distinction between the four northeast communities on the shores of the Bay of Biscay (which are the regions exhibiting milder temperatures) and the remaining regions. A second cluster corresponds to the southern Mediterranean area and the central region which corresponds to the communities with highest temperatures.

Optimally robust estimators in generalized Pareto models

In this paper, we study the robustness properties of several procedures for the joint estimation of shape and scale in a generalized Pareto model. The estimators that we primarily focus upon, most bias robust estimator (MBRE) and optimal MSE-robust estimator (OMSE), are one-step estimators distinguished as optimally robust in the shrinking neighbourhood setting; that is, they minimize the maximal bias, respectively, on such a specific neighbourhood, the maximal mean squared error (MSE). For their initialization, we propose a particular location–dispersion estimator, MedkMAD, which matches the population median and kMAD (an asymmetric variant of the median of absolute deviations) against the empirical counterparts. These optimally robust estimators are compared to the maximum-likelihood, skipped maximum-likelihood, Cramér–von-Mises minimum distance, method-of-medians, and Pickands estimators. To quantify their deviation from robust optimality, for each of these suboptimal estimators, we determine the finite-sample breakdown point and the influence function, as well as the statistical accuracy measured by asymptotic bias, variance, and MSE – all evaluated uniformly on shrinking neighbourhoods. These asymptotic findings are complemented by an extensive simulation study to assess the finite-sample behaviour of the considered procedures. The applicability of the procedures and their stability against outliers are illustrated for the Danish fire insurance data set from the package evir.

Flood analysis using negative binomial and Generalized Pareto models in partial duration series (PDS)

Summary Two flood analysis estimation schemes, based on, respectively, partial duration series (PDS) and annual maximum series (AMS), are compared. The PDS model assumes a Generalized Pareto (GP) distribution for modeling the flood exceedances above threshold corresponding to a generalized extreme value (GEV) distribution for annual maxima. As a generalization of the common assumption of the Poisson distribution (PD) to count the occurrences of peaks over threshold in the PDS models, the advantage of negative binomial (NB) distribution is explored in this study. The T-year event estimator for the annual maximum distribution corresponding to the parent PDS model is formulated for producing AMS samples consistent with PDS samples which are used in simulations. The performance of the two models in terms of the uncertainty of the T-year event estimator is evaluated in the cases of estimation with the method of probability weighted moments (PWM). In a similar way, the performance of the derived PDS/NB-GP model is compared with the existing PDS/PD-GP model in terms of uncertainty of T-year event estimator using simulation and field data. The results show the T-year event estimate using PDS/NB-GP model yields lower variance compared to PDS/PD-GP models for most cases. However both the models perform similarly at higher return periods more than 300 years, using the ratios of the variance of T-years estimate as an index, and the ratio decreases with an increase in mean number of annual exceedances above threshold (μ). From the results it is observed that both AMS and PDS models yield the same variance when μ varies from 1.4 to 1.65. However, in case of NB distribution the PDS and AMS models gives the same variance of q(T) when variance (σ2) is 1.5 times the mean number of annual exceedance above threshold. The performance of the PDS models and the corresponding AMS models using the available data of Dee (at Cairnton) shows the PDS/NB-GP model to be marginally better at return periods lower than 50 years.

Extended generalised Pareto models for tail estimation

The most popular approach in extreme value statistics is the modelling of threshold exceedances using the asymptotically motivated generalised Pareto distribution. This approach involves the selection of a high threshold above which the model fits the data well. Sometimes, few observations of a measurement process might be recorded in applications and so selecting a high quantile of the sample as the threshold leads to almost no exceedances. In this paper we propose extensions of the generalised Pareto distribution that incorporate an additional shape parameter while keeping the tail behaviour unaffected. The inclusion of this parameter offers additional structure for the main body of the distribution, improves the stability of the modified scale, tail index and return level estimates to threshold choice and allows a lower threshold to be selected. We illustrate the benefits of the proposed models with a simulation study and two case studies.

Modeling the Seasonality of Extreme Waves in the Gulf of Mexico

Statistics of storm peaks over threshold depend typically on a number of covariates including location, season, and storm direction. Here, a nonhomogeneous Poisson model is adopted to characterize storm peak events with respect to season for two Gulf of Mexico locations. The behavior of storm peak significant wave height over threshold is characterized using a generalized Pareto model, the parameters of which vary smoothly with season using a Fourier form. The rate of occurrence of storm peaks is also modeled using a Poisson model with rate varying with season. A seasonally varying extreme value threshold is estimated independently. The degree of smoothness of extreme value shape and scale and the Poisson rate with season are regulated by roughness-penalized maximum likelihood; the optimal value of roughness is selected by cross validation. Despite the fact that only the peak significant wave height event for each storm is used for modeling, the influence of the whole period of a storm on design extremes for any seasonal interval is modeled using the concept of storm dissipation, providing a consistent means to estimate design criteria for arbitrary seasonal intervals. The characteristics of the 100 year storm peak significant wave height, estimated using the seasonal model, are examined and compared with those estimated ignoring seasonality.

A Spatiodirectional Model for Extreme Waves in the Gulf of Mexico

The characteristics of extreme waves in hurricane dominated regions vary systematically with a number of covariates, including location and storm direction. Reliable estimation of design criteria requires incorporation of covariate effects within extreme value models. We present a spatiodirectional model for extreme waves in the Gulf of Mexico motivated by the nonhomogeneous Poisson model for peaks over threshold. The model is applied to storm peak significant wave height HS for arbitrary geographic areas from the proprietary Gulf of Mexico Oceanographic Study (GOMOS) hindcast for the US region of the Gulf of Mexico for the period of 1900–2005. At each location, directional variability is modeled using a nonparametric directional location and scale; data are standardized (or “whitened”) with respect to local directional location and scale to remove directional effects. For a suitable choice of threshold, the rate of occurrence of threshold exceedences of whitened storm peak HS with direction per location is modeled as a Poisson process. The size of threshold exceedences is modeled using a generalized Pareto form, the parameters of which vary smoothly in space, and are estimated within a roughness-penalized likelihood framework using natural thin plate spline forms in two spatial dimensions. By reparameterizing the generalized Pareto model in terms of asymptotically independent parameters, an efficient back-fitting algorithm to estimate the natural thin plate spline model is achieved. The algorithm is motivated in an appendix. Design criteria, estimated by simulation, are illustrated for a typical neighborhood of 17×17 grid locations. Applications to large areas consisting of more than 2500 grid locations are outlined.

Event risk—Parametrization and estimation in a generalized Pareto model with time-varying thresholds

The EU Capital Requirements Directives, which is the European implementation of the Basel II Accord, became effective in 2007. Since then, credit institutions have had to run their risk measurement...

A discussion on mean excess plots

The mean excess plot is a tool widely used in the study of risk, insurance and extreme values. One use is in validating a generalized Pareto model for the excess distribution. This paper investigates some theoretical and practical aspects of the use of the mean excess plot.

Statistical distributions of extreme dry spell in Peninsular Malaysia

Statistical distributions of annual extreme (AE) series and partial duration (PD) series for dry-spell event are analyzed for a database of daily rainfall records of 50 rain-gauge stations in Peninsular Malaysia, with recording period extending from 1975 to 2004. The three-parameter generalized extreme value (GEV) and generalized Pareto (GP) distributions are considered to model both series. In both cases, the parameters of these two distributions are fitted by means of the L-moments method, which provides a robust estimation of them. The goodness-of-fit (GOF) between empirical data and theoretical distributions are then evaluated by means of the L-moment ratio diagram and several goodness-of-fit tests for each of the 50 stations. It is found that for the majority of stations, the AE and PD series are well fitted by the GEV and GP models, respectively. Based on the models that have been identified, we can reasonably predict the risks associated with extreme dry spells for various return periods.

Comparison at optimal levels of classical tail index estimators: a challenge for reduced-bias estimation?

In this article, we begin with an asymptotic comparison at optimal levels of the so-called “maximum likelihood” (ML) extreme value index estimator, based on the excesses over a high random threshold, denoted PORT-ML, with PORT standing for peaks over random thresholds, with a similar ML estimator, denoted PORT-MP, with MP standing for modified-Pareto. The PORT-MP estimator is based on the same excesses, but with a trial of accommodation of bias on the Generalized Pareto model underlying those excesses. We next compare the behaviour of these ML implicit estimators with the equivalent behaviour of a few explicit tail index estimators, the Hill, the moment, the generalized Hill and the mixed moment. As expected, none of the estimators can always dominate the alternatives, even when we include second-order MVRB tail index estimators, with MVRB standing for minimum-variance reduced-bias. However, the asymptotic performance of the MVRB estimators is quite interesting and provides a challenge for a further study of these MVRB estimators at optimal levels.

Estimation of the angular density in bivariate generalized Pareto models

We investigate a method to estimate the angular density non-parametrically in bivariate generalized Pareto models. The angular density can be used as a visual tool to gain a first insight into the tail-dependence structure of given data. We derive a representation of the angular density by means of the Pickands density and use it to construct our estimator. The estimator is asymptotically normal under certain regularity conditions. We also test it with simulated data and give an application to a real hydrological data set. Finally, we show that our estimator cannot be transferred directly to higher dimensions.