The purpose of this thesis is to compare rank-based tests of bivariate stochastic order. Given two bivariate distributions $F$ and $G$, the general problem we are dealing with is to test $H_0: F=G$ against $H_1:F<G$, where $F$ and $G$ are independent continuous distributions on $\Re ^2$. (``$F<G$" means that $F(x)\leq G(x)~\forall x\in \Re^2$, and $\exists x\in \Re^2$ such that $F(x)< G(x)$.). In particular, we will analyze three analogues of the one-dimensional Mann-Whitney-Wilcoxon test in two dimensions. Two of the test statistics are new; we call them the Kendall and Spearman statistics. We will then show the asymptotic distributions and carry out empirical comparisons of the Kendall, Spearman and the third two-dimensional Mann-Whitney-Wilcoxon statistics.

The purpose of this thesis is to compare rank-based tests of bivariate stochastic order. Given two bivariate distributions $F$ and $G$, the general problem we are dealing with is to test $H_0: F=G$ against $H_1:F<G$, where $F$ and $G$ are independent continuous distributions on $\Re ^2$. (``$F<G$" means that $F(x)\leq G(x)~\forall x\in \Re^2$, and $\exists x\in \Re^2$ such that $F(x)< G(x)$.). In particular, we will analyze three analogues of the one-dimensional Mann-Whitney-Wilcoxon test in two dimensions. Two of the test statistics are new; we call them the Kendall and Spearman statistics. We will then show the asymptotic distributions and carry out empirical comparisons of the Kendall, Spearman and the third two-dimensional Mann-Whitney-Wilcoxon statistics.

The purpose of this thesis is to compare rank-based tests of bivariate stochastic order. Given two bivariate distributions $F$ and $G$, the general problem we are dealing with is to test $H_0: F=G$ against $H_1:F<G$, where $F$ and $G$ are independent continuous distributions on $\Re ^2$. (``$F<G$" means that $F(x)\leq G(x)~\forall x\in \Re^2$, and $\exists x\in \Re^2$ such that $F(x)< G(x)$.). In particular, we will analyze three analogues of the one-dimensional Mann-Whitney-Wilcoxon test in two dimensions. Two of the test statistics are new; we call them the Kendall and Spearman statistics. We will then show the asymptotic distributions and carry out empirical comparisons of the Kendall, Spearman and the third two-dimensional Mann-Whitney-Wilcoxon statistics.

abstract: This dissertation introduces stochastic ordering of instantaneous channel powers of fading channels as a general method to compare the performance of a communication system over two different channels, even when a closed-form expression for the metric may not be available. Such a comparison is with respect to a variety of performance metrics such as error rates, outage probability and ergodic capacity, which share common mathematical properties such as monotonicity, convexity or complete monotonicity. Complete monotonicity of a metric, such as the symbol error rate, in conjunction with the stochastic Laplace transform order between two fading channels implies the ordering of the two channels with respect to the metric. While it has been established previously that certain modulation schemes have convex symbol error rates, there is no study of the complete monotonicity of the same, which helps in establishing stronger channel ordering results. Toward this goal, the current research proves for the first time, that all 1-dimensional and 2-dimensional modulations have completely monotone symbol error rates. Furthermore, it is shown that the frequently used parametric fading distributions for modeling line of sight exhibit a monotonicity in the line of sight parameter with respect to the Laplace transform order. While the Laplace transform order can also be used to order fading distributions based on the ergodic capacity, there exist several distributions which are not Laplace transform ordered, although they have ordered ergodic capacities. To address this gap, a new stochastic order called the ergodic capacity order has been proposed herein, which can be used to compare channels based on the ergodic capacity. Using stochastic orders, average performance of systems involving multiple random variables are compared over two different channels. These systems include diversity combining schemes, relay networks, and signal detection over fading channels with non-Gaussian additive noise. This research also addresses the problem of unifying fading distributions. This unification is based on infinite divisibility, which subsumes almost all known fading distributions, and provides simplified expressions for performance metrics, in addition to enabling stochastic ordering. / Dissertation/Thesis / Ph.D. Electrical Engineering 2014

The purpose of this thesis is to compare rank-based tests of bivariate stochastic order. Given two bivariate distributions $F$ and $G$, the general problem we are dealing with is to test $H_0: F=G$ against $H_1:F<G$, where $F$ and $G$ are independent continuous distributions on $\Re ^2$. (``$F<G$" means that $F(x)\leq G(x)~\forall x\in \Re^2$, and $\exists x\in \Re^2$ such that $F(x)< G(x)$.). In particular, we will analyze three analogues of the one-dimensional Mann-Whitney-Wilcoxon test in two dimensions. Two of the test statistics are new; we call them the Kendall and Spearman statistics. We will then show the asymptotic distributions and carry out empirical comparisons of the Kendall, Spearman and the third two-dimensional Mann-Whitney-Wilcoxon statistics.

"Uncertainty" is used broadly to refer to things that are unknown or incompletely understood. In operations management, basic sources of uncertainty may include decision uncertainty, model uncertainty, analytical uncertainty, data uncertainty, and so on. Although uncertainty is unavoidable in decision making, different mechanisms can be designed to mitigate the impact of uncertainty. One commonly used strategy is "decision postponement," wherein the decision maker purposefully delays some of the decisions to a time when uncertainty is reduced or resolved. This type of a recourse action provides the decision maker with increased ability to match supply with demand. In this dissertation, we study the value of decision postponement in the context of different settings, including capacity investment, revenue management, and supply chain coordination. These problems share one characteristic in common: decision postponement, and as such, are all modeled as two-stage stochastic programming problems. In the first stage, a set of decisions are made under uncertainty so as to maximize the expected profit or utility. Then in the second stage, all uncertainty is resolved and a deterministic optimization problem is solved to determine the postponed decisions, constrained by the first stage decisions. In capacity investment, we study the capacity, pricing, and production decisions of a monopolist producing substitutable products with flexible or dedicated resources. While the capacity decision needs to be made ex-ante, under demand uncertainty, pricing and production decisions can be postponed until after uncertainty is resolved. We show how key demand parameters (the nature of uncertainty, market size, market risk, and risk attitude) impact the optimal capacity decision under the linear demand function. In particular, we show that if the demand shock is multiplicative, then in terms of the "invest or not" decision, the firm will be immune to forecast errors in parameters of the underlying demand shock distribution. Furthermore, incorrectly modeling the demand shock as additive, when, in fact, it is multiplicative, may lead to overinvestment. On the other hand, while the concept of a growth in market size leads to similar conclusions under both additive and multiplicative demand shocks, how market risk affects the optimal capacity decision depends critically on the form of the demand shock. In addition, the decision-maker's attitude toward risk significantly affects the optimal capacity level, and its impact highly depends on the structure of the resource network. Our analysis provides insights and principles on the optimal capacity investment decision under various settings. In airline revenue management, a well-studied problem is the optimal allocation of seat inventory among different fare-classes, given a capacity for the flight and a demand distribution for each class. In practice, capacity on a flight does not have to be fixed; airlines can exercise some flexibility on the supply side by swapping aircraft of different capacities between flights as partial booking information is gathered. This provides the airline with the capability to more effectively match their supply and demand. In this dissertation, we study the seat inventory control problem considering the aircraft swapping option. Our analytical results demonstrate that booking limits considering the swapping option can be considerably different from those under fixed capacity. We also show that principles on the relationship between the optimal booking limits and demand characteristics (size and risk) developed for the fixed-capacity problem no longer hold when swapping is an option. We develop new principles and insights on how demand characteristics affect the optimal seat allocation under the swapping possibility. We also perform a numerical study, which indicates that the revenue impact of using the "true" optimal booking limits under the swapping possibility can be significant. In supply chain coordination, we consider the influenza vaccine supply chain, which, due to the biological complexity of the production process, has a unique characteristic in that production yield is highly uncertain. Given the market demand and price, a monopolist supplier must decide how much raw material to input into production in the first stage. However, since the yield is unknown and production is costly, it is not necessarily in the supplier's best interest to ensure that all market demand is met. The supplier's input quantity depends on the trade-off between the costs of overproduction and undersupply. This, in fact, is one of the reasons why the influenza vaccine manufacturers in the United States lack motivation to produce sufficient amounts of vaccine to meet all demand [Williams (2005), Chick et al. (2008)]. In operations management, it is a well-known result that decentralized supply chains, where each player is only interested in optimizing her own objective, often lead to poor overall performance for the supply chain. However, a higher efficiency is achievable through contracting on a set of transfer payments [Cachon (2004)]. A "coordinating" contract is referred to as one in which each player's objective is in accordance with the supply chain's objective. Given the fact that influenza vaccine plays an important role in health care industry, it is important to study how different contracts impact the influenza vaccine supply chain, where the uncertainty is on the supply side. We study a game in which the supplier and the retailer are engaged in certain type of contracts that specify how risk is shared between the players. We study both the pre-ordering and the post-ordering settings, which respectively refer to the cases where the retailer orders the vaccine before or after the vaccine production is completed. We show that pre-ordering wholesale price contracts dominate post-ordering wholesale price contracts in terms of the resulting supply chain efficiency, but neither of them are able to fully coordinate the supply chain. We also find that cost-sharing contracts are able to coordinate the supply chain, while payback and advance-ordering wholesale price contracts fail to do so. Finally, we prove that if the unsold vaccine can be salvaged with some positive value, then the supply chain can be easily coordinated with wholesale price contracts. In studying this type of stochastic programming problems, it is not only important to characterize the optimal solution, but also important to gain an understanding of how the optimal solution will be affected by environmental parameters. Since the most inaccurate part in stochastic programming often lies in the parameters of the distribution functions, it is both interesting and meaningful to investigate how the optimal solution varies with the intrinsic nature of the random variables. Consequently, we make use of stochastic order relationships to study the behavior of the optimal solutions when the underlying random variables become either "larger" or "more risky." / Ph. D.

As we know the most popular inference methods for order restricted model are likelihood inference. In such models, the Maximum Likelihood Estimation (MLE) and Likelihood Ratio Testing (LRT) appear some suspect behaviour and unsatisfactory. In this thesis, I review the articles that focused in the behaviour of the Likelihood methods on Order restricted models. For those situations, an alternative method is preferred. However, likelihood inference is satisfactory for simple order cone restriction. But it is unsatisfactory when the restrictions are of the tree order, umbrella order, star-shaped and stochastic order types.

Cette thèse a pour but le développement de certains aspects de la modélisation de la dépendance dans la gestion des risques en dimension plus grande que un. Le premier chapitre est constitué d'une introduction générale. Le deuxième chapitre est constitué d'un article s'intitulant « Estimating Bivariate Tail : a copula based approach », soumis pour publication. Il concerne la construction d'un estimateur de la queue d'une distribution bivariée. La construction de cet estimateur se fonde sur une méthode de dépassement de seuil (Peaks Over Threshold method) et donc sur une version bivariée du Théorème de Pickands-Balkema-de Haan. La modélisation de la dépendance est obtenue via la Upper Tail Dependence Copula. Nous démontrons des propriétés de convergence pour l'estimateur ainsi construit. Le troisième chapitre repose sur un article: « A multivariate extension of Value-at-Risk and Conditional-Tail-Expectation», soumis pour publication. Nous abordons le problème de l'extension de mesures de risque classiques, comme la Value-at-Risk et la Conditional-Tail-Expectation, dans un cadre multidimensionnel en utilisant la fonction de Kendall multivariée. Enfin, dans le quatrième chapitre de la thèse, nous proposons un estimateur des courbes de niveau d'une fonction de répartition bivariée avec une méthode plug-in. Nous démontrons des propriétés de convergence pour les estimateurs ainsi construits. Ce chapitre de la thèse est lui aussi constitué d'un article, s'intitulant « Plug-in estimation of level sets in a non-compact setting with applications in multivariate risk theory», accepté pour publication dans la revue ESAIM:Probability and Statistics. / In this PhD thesis we consider different aspects of dependence modeling with applications in multivariate risk theory. The first chapter is constituted by a general introduction. The second chapter is essentially constituted by the article “Estimating Bivariate Tail: a copula based approach”, actually submitted for publication. It deals with the problem of estimating the tail of a bivariate distribution function. We develop a general extension of the POT (Peaks-Over-Threshold) method, mainly based on a two-dimensional version of the Pickands-Balkema-de Haan Theorem. The dependence structure between the marginals in the upper tails is described by the Upper Tail Dependence Copula. Then we construct a two-dimensional tail estimator and study its asymptotic properties. The third chapter of this thesis is based on the article “A multivariate extension of Value-at-Risk and Conditional-Tail-Expectation” and submitted for publication. We propose a multivariate generalization of risk measures as Value-at-Risk and Conditional-Tail-Expectation and we analyze the behavior of these measures in terms of classical properties of risk measures. We study the behavior of these measures with respect to different risk scenarios and stochastic ordering of marginals risks. Finally in the fourth chapter we introduce a consistent procedure to estimate level sets of an unknown bivariate distribution function, using a plug-in approach in a non-compact setting. Also this chapter is constituted by the article “Plug-in estimation of level sets in a non-compact setting with applications in multivariate risk theory”, accepted for publication in ESAIM: Probability and Statistics journal.

En finance, le risque de modèle est le risque de pertes financières résultant de l'utilisation de modèles. Il s'agit d'un risque complexe à appréhender qui recouvre plusieurs situations très différentes, et tout particulièrement le risque d'estimation (on utilise en général dans un modèle un paramètre estimé) et le risque d'erreur de spécification de modèle (qui consiste à utiliser un modèle inadéquat). Cette thèse s'intéresse d'une part à la quantification du risque de modèle dans la construction de courbes de taux ou de crédit et d'autre part à l'étude de la compatibilité des indices de Sobol avec la théorie des ordres stochastiques. Elle est divisée en trois chapitres. Le Chapitre 1 s'intéresse à l'étude du risque de modèle dans la construction de courbes de taux ou de crédit. Nous analysons en particulier l'incertitude associée à la construction de courbes de taux ou de crédit. Dans ce contexte, nous avons obtenus des bornes de non-arbitrage associées à des courbes de taux ou de défaut implicite parfaitement compatibles avec les cotations des produits de référence associés. Dans le Chapitre 2 de la thèse, nous faisons le lien entre l'analyse de sensibilité globale et la théorie des ordres stochastiques. Nous analysons en particulier comment les indices de Sobol se transforment suite à une augmentation de l'incertitude d'un paramètre au sens de l'ordre stochastique dispersif ou excess wealth. Le Chapitre 3 de la thèse s'intéresse à l'indice de contraste quantile. Nous faisons d'une part le lien entre cet indice et la mesure de risque CTE puis nous analysons, d'autre part, dans quelles mesures une augmentation de l'incertitude d'un paramètre au sens de l'ordre stochastique dispersif ou excess wealth entraine une augmentation de l'indice de contraste quantile. Nous proposons enfin une méthode d'estimation de cet indice. Nous montrons, sous des hypothèses adéquates, que l'estimateur que nous proposons est consistant et asymptotiquement normal / In finance, model risk is the risk of loss resulting from using models. It is a complex risk which recover many different situations, and especially estimation risk and risk of model misspecification. This thesis focuses: on model risk inherent in yield and credit curve construction methods and the analysis of the consistency of Sobol indices with respect to stochastic ordering of model parameters. it is divided into three chapters. Chapter 1 focuses on model risk embedded in yield and credit curve construction methods. We analyse in particular the uncertainty associated to the construction of yield curves or credit curves. In this context, we derive arbitrage-free bounds for discount factor and survival probability at the most liquid maturities. In Chapter 2 of this thesis, we quantify the impact of parameter risk through global sensitivity analysis and stochastic orders theory. We analyse in particular how Sobol indices are transformed further to an increase of parameter uncertainty with respect to the dispersive or excess wealth orders. Chapter 3 of the thesis focuses on contrast quantile index. We link this latter with the risk measure CTE and then we analyse on the other side, in which circumstances an increase of a parameter uncertainty in the sense of dispersive or excess wealth orders implies and increase of contrast quantile index. We propose finally an estimation procedure for this index. We prove under some conditions that our estimator is consistent and asymptotically normal

A distribuição bivariada de Marshall-Olkin é estendida, relaxando-se a hipótese de choques exponencialmente distribuídos e assumindo-se dependência entre os choques individuais. Abordagem semelhante é considerada para sua versão dual. Representação por meio de cópula, propriedades probabilísticas e de confiabilidade assim como resultados em valores extremos são então obtidos. A propriedade de falta de memória bivariada é estendida assumindo-se uma função de dependência sem memória. Uma nova classe de distribuições caracterizada por essa propriedade estendida é introduzida. Correspondentes interpretações geométricas, procedimentos de construção, representação estocástica, relação com cópula de sobrevivência e propriedades de confiabilidade são derivadas. / Bivariate Marshall-Olkin model, Dual model, Exponential representation, Dependence function, Bivariate aging, Copula, Survival copula, Stochastic order, Bivariate extreme value distribution, Pickands measure, Pickands dependence function, Failure rate, Bivariate hazard gradient, Bivariate lack-of-memory, Residual lifetime vector, Characterization.