La prévision d'événements extrêmes est une question cruciale dans des domaines divers allant de la météorologie à la finance. Trois classes d'universalité (Gumbel, Fréchet et Weibull) ont été identifiées pour des variables aléatoires indépendantes et de distribution identique (i.i.d.).La modélisation par des variables aléatoires i.i.d., notamment avec le modèle d'énergie aléatoire de Derrida, a permis d'améliorer la compréhension des systèmes désordonnés. Cette hypothèse n'est toutefois pas valide pour de nombreux systèmes physiques qui présentent de fortes corrélations. Dans cette thèse, nous étudions trois modèles physiques de variables aléatoires fortement corrélées : des fermions piégés,des matrices aléatoires et des marches aléatoires. Dans la première partie, nous montrons plusieurs correspondances exactes entre l'état fondamental d'un gaz de Fermi piégé et des ensembles de matrices aléatoires. Le gaz Fermi est inhomogène dans le potentiel de piégeage et sa densité présente un bord fini au-delà duquel elle devient essentiellement nulle. Nous développons une description précise des statistiques spatiales à proximité de ce bord, qui va au-delà des approximations semi-classiques standards (telle que l'approximation de la densité locale). Nous appliquons ces résultats afin de calculer les statistiques de la position du fermion le plus éloigné du centre du piège, le nombre de fermions dans un domaine donné (statistiques de comptage) et l'entropie d'intrication correspondante. Notre analyse fournit également des solutions à des problèmes ouverts de valeurs extrêmes dans la théorie des matrices aléatoires. Nous obtenons par exemple une description complète des fluctuations de la plus grande valeur propre de l'ensemble complexe de Ginibre.Dans la deuxième partie de la thèse, nous étudions les questions de valeurs extrêmes pour des marches aléatoires. Nous considérons les statistiques d'écarts entre positions maximales consécutives (gaps), ce qui nécessite de prendre en compte explicitement le caractère discret du processus. Cette question ne peut être résolue en utilisant la convergence du processus avec son pendant continu, le mouvement Brownien. Nous obtenons des résultats analytiques explicites pour ces statistiques de gaps lorsque la distribution de sauts est donnée par la loi de Laplace et réalisons des simulations numériques suggérant l'universalité de ces résultats. / Predicting the occurrence of extreme events is a crucial issue in many contexts, ranging from meteorology to finance. For independent and identically distributed (i.i.d.) random variables, three universality classes were identified (Gumbel, Fréchet and Weibull) for the distribution of the maximum. While modelling disordered systems by i.i.d. random variables has been successful with Derrida's random energy model, this hypothesis fail for many physical systems which display strong correlations. In this thesis, we study three physically relevant models of strongly correlated random variables: trapped fermions, random matrices and random walks.In the first part, we show several exact mappings between the ground state of a trapped Fermi gas and ensembles of random matrix theory. The Fermi gas is inhomogeneous in the trapping potential and in particular there is a finite edge beyond which its density vanishes. Going beyond standard semi-classical techniques (such as local density approximation), we develop a precise description of the spatial statistics close to the edge. This description holds for a large universality class of hard edge potentials. We apply these results to compute the statistics of the position of the fermion the farthest away from the centre of the trap, the number of fermions in a given domain (full counting statistics) and the related bipartite entanglement entropy. Our analysis also provides solutions to open problems of extreme value statistics in random matrix theory. We obtain for instance a complete description of the fluctuations of the largest eigenvalue in the complex Ginibre ensemble.In the second part of the thesis, we study extreme value questions for random walks. We consider the gap statistics, which requires to take explicitly into account the discreteness of the process. This question cannot be solved using the convergence of the process to its continuous counterpart, the Brownian motion. We obtain explicit analytical results for the gap statistics of the walk with a Laplace distribution of jumps and provide numerical evidence suggesting the universality of these results.

The Ph.D. thesis deals with extreme wind gust analysis over the area of the Czech Republic. The first part of the thesis deals with processing of wind measurements, in particular maximum wind gusts measurements. Analysis of high-frequency wind measurement using 3-D sonic anemometer on the Kopisty station is included. Homogenization of the highest daily wind gusts was performed. Descriptive statistical analysis of measured wind gust values was performed. The following part of the thesis describes statistical theory of extreme values and discusses its applicability to wind gust data. Some theoretical findings were obtained. Numerous numerical experiments were performed focused on evaluation of proposed method. In the last part of the thesis station measurements were processed using the proposed methods and a model of dependence between extreme and mean wind climate was derived. The model was applied to the map of mean wind climate calculated earlier on the Institute of Atmospheric Physics and thus a map of extreme wind climate was obtained. The accuracy of this map was estimated. The map was compared with other maps of extreme wind calculated by other authors earlier.

The statistical modeling of extreme natural hazards is becoming increasingly important due to climate change, whose effects have been increasingly visible throughout the last decades. It is thus crucial to understand the dependence structure of rare, high-impact events over space and time for realistic risk assessment. For spatial extremes, max-stable processes have played a central role in modeling block maxima. However, the spatial tail dependence strength is persistent across quantile levels in those models, which is often not realistic in practice. This lack of flexibility implies that max-stable processes cannot capture weakening dependence at increasingly extreme levels, resulting in a drastic overestimation of joint tail risk. To address this, we develop new dependence models in this thesis from the class of max-infinitely divisible (max-id) processes, which contain max-stable processes as a subclass and are flexible enough to capture different types of dependence structures. Furthermore, exact simulation algorithms for general max-id processes are typically not straightforward due to their complex formulations. Both simulation and inference can be computationally prohibitive in high dimensions. Fast and exact simulation algorithms to simulate max-id processes are provided, together with methods to implement our models in high dimensions based on the Vecchia approximation method. These proposed methodologies are illustrated through various environmental datasets, including air temperature data in South-Eastern Europe in an attempt to assess the effect of climate change on heatwave hazards, and sea surface temperature data for the entire Red Sea. In another application focused on assessing how the spatial extent of extreme precipitation has changed over time, we develop new time-varying $r$-Pareto processes, which are the counterparts of max-stable processes for high threshold exceedances.

This thesis gives an example of assessing the risk of a financial portfolio with international assets, where the assets may be of different classes, by the use of Monte Carlo simulation and Extreme Value Theory. The simulation uses univariate modelling, models of the assets’ returns as stochastic processes, as well as vine copulas to create dependency between the variables. For the asset returns a modified version of a discretized Merton jump diffusion model was used. The risk assessment used Extreme Value Theory to calculate Value at Risk and Expected Shortfall of the simulated portfolio. However, the resulting return distribution, and the risk assessment thereof, was not entirely satisfactory due to unreasonably large values ascertained. / Denna uppsats ger ett exempel på hur riskbedömning av finanisella portföljer med internationella tillgångar av olika tillgångsslag genom Monte Carlo simulering och extremvärdesteori. Simuleringen använder univariat modelling, modeller för tillgångarnas avkastningar som stokastiska processer, såväl som vine-copulas för att skapa ett beroende mellan tillgångarna. Tillgångarnas avkastningar modellerades med en modifierad version av en diskretiserad Merton-jump-diffusion model. Riskbedömningen använde extremvärdesteori för att beräkna Value-at-Risk och Expected-Shortfall. Dock blev den resulterande avkastningsfördelningen och riskbedömningen därav inte helt tillfredsällande på grund av att orimligt stora värden erhölls.

The objective of this thesis is to investigate the flow conditions that lead to coarse grain entrainment at near incipient motion conditions. Herein, a new conceptual approach is proposed, which in addition to the magnitude of hydrodynamic force or flow power, takes into account the duration of the flow event. Two criteria for inception of grain entrainment, namely the critical impulse and critical energy concepts, are proposed and compared. These frameworks adopt a force or energy perspective, considering the momentum or energy transfer from each flow event to the particle respectively, to describe the phenomenon. A series of conducted mobile particle experiments, are analyzed to examine the validity of the proposed approaches. First a set of bench-top experiments incorporates an electromagnet which applies pulses of known magnitude and duration to a steel spherical particle in a controlled fashion, so as to identify the critical level for entrainment. The utility of the above criteria is also demonstrated for the case of entrainment by the action of turbulent flow, via analysis of a series of flume experiments, where both the history of hydrodynamic forces exerted on the particle as well as its response are recorded simultaneously. Statistical modeling of the distribution of impulses, as well as conditional excess impulses, is performed using distributions from Extreme Value Theory to effectively model the episodic nature of the occurrence of these events. For the examined uniform and low mobility flow conditions, a power law relationship is proposed for describing the magnitude and frequency of occurrence of the impulse events. The Weibull and exponential distributions provide a good fit for the time between particle entrainments. In addition to these statistical tools, a number of Adaptive Neuro-Fuzzy Inference Systems employing different input representations are used to learn the nonlinear dynamics of the system and perform statistical prediction. The performance of these models is assessed in terms of their broad validity, efficiency and forecast accuracy. Even though the impulse and energy criteria are deeply interrelated, the latter is shown to be advantageous with regard to its performance, applicability and extension ability. The effect of single or multiple highly energetic events carried by certain coherent flow structures (mainly strong sweep events) with regard to the particle response is also investigated. / Ph. D.

Portfolio Management is the process of allocating an investor's wealth to in­ vestment opportunities over a given planning period. Not only should Portfolio Management be treated within a multi-period framework, but one should also take into consideration the stochastic nature of related parameters. After a short review of key concepts from Finance Theory, e.g. utility function, risk attitude, Value-at-rusk estimation methods, a.nd mean-variance efficiency, this work describes a framework for the formulation of the Portfolio Management problem in a Stochastic Programming setting. Classical solution techniques for the resolution of the resulting Stochastic Programs (e.g. L-shaped Decompo­ sition, Approximation of the probability function) are presented. These are discussed within both the two-stage and the multi-stage case with a special em­ phasis on the former. A description of how Importance Sampling and EVPI are used to improve the efficiency of classical methods is presented. Postoptimality Analysis, a sensitivity analysis method, is also described. / Statistics / M. Sc. (Operations Research)

<p>In this thesis are extreme value theory used to estimate the probability that large insuranceclaims are exceeding a certain threshold. The expected claim size, given that the claimhas exceeded a certain limit, are also estimated. Two different models are used for thispurpose. The first model is based on maximum domain of attraction conditions. A Paretodistribution is used in the other model. Different graphical tools are used to check thevalidity for both models. Länsförsäkring Kronoberg has provided us with insurance datato perform the study.Conclusions, which have been drawn, are that both models seem to be valid and theresults from both models are essential equal.</p> / <p>I detta arbete används extremvärdesteori för att uppskatta sannolikheten att stora försäkringsskadoröverträffar en vis nivå. Även den förväntade storleken på skadan, givetatt skadan överstiger ett visst belopp, uppskattas. Två olika modeller används. Den förstamodellen bygger på antagandet att underliggande slumpvariabler tillhör maximat aven extremvärdesfördelning. I den andra modellen används en Pareto fördelning. Olikagrafiska verktyg används för att besluta om modellernas giltighet. För att kunna genomförastudien har Länsförsäkring Kronoberg ställt upp med försäkringsdata.Slutsatser som dras är att båda modellerna verkar vara giltiga och att resultaten ärlikvärdiga.</p>

本文以台灣股價加權指數，使用 AR(3)-GJR-GRACH(1,1) 模型，白噪音假設為 Normal 、 Skew-Normal 、 Student t 、 skew-t 、 EPD 、 SEPD 、與 AEPD 等七種分配。著重於兩個部份，(一) Student t 分配一族與 EPD 分配一族在模型配適與風險值估計的比較；(二) 預測風險值區分為低震盪與高震盪兩個區間，比較不同分配在兩區間預測風險值的差異。 實證分析顯示， t 分配一族與 EPD 分配一族配適的結果，無論是只考慮峰態 ( t 分配與 EPD 分配) ，或者加入影響偏態的參數 ( skew-t 分配與 SEPD 分配) ， t 分配一族的配適程度都較 EPD 分配一族為佳。更進一步考慮分配兩尾厚度不同的 AEPD 分配，配適結果為七種分配中最佳。 風險值的估計在低震盪的區間，常態分配與其他厚尾分配皆能通過回溯測試，採用厚尾分配效果不大；在高震盪的區間，左尾風險值回溯測試結果，常態分配與其他厚尾分配皆無法全數通過，但仍以 AEPD 分配為最佳。最後比較損失函數，左尾風險值估計以 AEPD 分配為最佳，右尾風險值則無一致的結果。因此我們認為 AEPD 分配可作為風險管理有用的工具。

This thesis presents a class of graphical models for directly representing the joint cumulative distribution function (CDF) of many random variables, called cumulative distribution networks (CDNs). Unlike graphical models for probability density and mass functions, in a CDN, the marginal probabilities for any subset of variables are obtained by computing limits of functions in the model. We will show that the conditional independence properties in a CDN are distinct from the conditional independence properties of directed, undirected and factor graph models, but include the conditional independence properties of bidirected graphical models. As a result, CDNs are a parameterization for bidirected models that allows us to represent complex statistical dependence relationships between observable variables. We will provide a method for constructing a factor graph model with additional latent variables for which graph separation of variables in the corresponding CDN implies conditional independence of the separated variables in both the CDN and in the factor graph with the latent variables marginalized out. This will then allow us to construct multivariate extreme value distributions for which both a CDN and a corresponding factor graph representation exist. In order to perform inference in such graphs, we describe the `derivative-sum-product' (DSP) message-passing algorithm where messages correspond to derivatives of the joint cumulative distribution function. We will then apply CDNs to the problem of learning to rank, or estimating parametric models for ranking, where CDNs provide a natural means with which to model multivariate probabilities over ordinal variables such as pairwise preferences. We will show that many previous probability models for rank data, such as the Bradley-Terry and Plackett-Luce models, can be viewed as particular types of CDN. Applications of CDNs will be described for the problems of ranking players in multiplayer team-based games, document retrieval and discovering regulatory sequences in computational biology using the above methods for inference and estimation of CDNs.

In this work we model the historical development of best performances in men's 100, 200, 400 and 800m running events. We suppose that the years best performances are independent random variables with generalized extreme value distribution for minima and that there is a decreasing trend in location. Parameters of the models are estimated by using maximum likelihood techniques. The data of years best performances are missing for some years, we treat them as right censored data that are censored by value of world record valid at that time. Graphic tools used for models diagnostics are adjusted to the censoring. The models we get are used to estimate the ultimate records and to predict new records in next years. At the end of the work we estimate several models describing historical development of years best performances for more events at one time.