Dynamic copulas: applications to finance & economics - PASTEL - Thèses en ligne de ParisTech Accéder directement au contenu
Thèse Année : 2007

Dynamic copulas: applications to finance & economics

Copules dynamiques : applications en finance et en économie

Résumé

In this thesis, we show that with the growth of credit derivatives markets, new products are continually being created and market liquidity is increasing. After reviewing these products starting out from the credit default swap, CDS, and describing their evolution since their inception in the early 90s, we demonstrate that this development has been market driven, with the mathematical models used for pricing lagging behind. As the market developed, the weak points of the models became apparent and improved models had to be developed. In October 2003 when the work on this thesis started, CDOs (Collateralised Debt Obligations) were becoming standard products. A new generation of products which we will refer to as third generation credit derivatives were starting to come on line: these include forward-starting CDS, forward-starting CDOs, options on CDOs, CPDO (in full) and so forth. In contrast to early products, these derivatives require a dynamic model of the evolution of the “correlation” between the names over time, something which base correlation was not designed to do. Our objective was to develop a family of multivariate copula processes with different types of upper and lower tail dependence so as to be able to reproduce the correlation smiles/skews observed in credit derivatives in practice. We chose to work with a dynamic version of Archimedean copulas because unlike many other copulas found in the literature, they are mathematically consistent multivariate models. Chapter 2 presents two different approaches for developing these processes. The first model developed is a non-additive jump process based on a background gamma process; the second approach is based on time changed spectrally positive Levy process. The first approach is very convenient for simulations; the second approach is based on additive building blocks and hence is a more general. Two applications of these models to credit risk derivatives were carried out. The first one on pricing synthetic CDOs at different maturities (Chapter 5) was presented at the 5th Annual Advances in Econometrics Conference in Baton Rouge, Louisiane, November 3-5 2006 and has been submitted for publication. The second one which presents a comparison of the pricing given by these dynamic copulas with five well-known copula models, has been submitted to the Journal of Derivatives (see Chapter 6). Having tested the basic dynamic copula models in a credit derivative context, we went on to combine this framework with matrix migration approach (Chapter 3). In order to market structured credit derivatives, banks have to get them rated by rating agencies such as S&P, Moody's and Fitch. A key question is the evolution of the rating over time (i.e. its migration). As the latest innovations in the credit derivatives markets such as Constant Proportion Debt Obligation (CPDO) require being able to model credit migration and correlation in order to handle substitutions on the index during the roll, we propose a model for the joint dynamics of credit ratings of several firms. We then proposed a mathematical framework were individual credit ratings are modelled by a continuous time Markov chain, and their joint dynamics are modelled using a copula process. Copulas allow us to incorporate our knowledge of single name credit migration processes, into a multivariate framework. This is further extended with the multi-factor and time changed approach. A multifactor approach is developed within the new formulated dynamic copula processes, and a time changed Levy process is used to introduce dependency on spread dynamics.
Les dérivés de crédit ont connu en quelques années un développement fulgurant en finance : les volumes de transactions ont augmenté exponentiellement, de nouveaux produits ont été créés. La récente crise du sub-prime a mis en évidence l'insuffisance des modèles actuels. Le but de cette thèse est de créer de nouveaux modèles mathématiques qui prennent en compte la dynamique de dépendance (« tail dependence ») des marchés. Après une revue de la littérature et des modèles existants, nous nous focalisons sur la modélisation de la « corrélation » (ou plus exactement la dynamique de la structure de dépendance) entre différentes entités dans un portefeuille de crédit (CDO). Dans une première phase, une formulation simple des copules dynamiques est proposée. Ensuite, nous présentons une seconde formulation en utilisant des processus de Lévy à spectre positif (i.e. gamma Ornstein-Uhlenbeck). L'écriture de cette nouvelle famille de copules archimédiennes nous permet d'obtenir une formule asymptotique simple pour la distribution des pertes d'un portefeuille de crédit granulaire. L'une des particularités du modèle proposé est sa capacité de reproduire des dépendances extrêmes comparables aux phénomènes récents de contagion sur les marchés comme la crise du « subprime » aux Etats-Unis. Une application sur l'estimation des prix des tranches de CDOs est aussi présentée. Dans cette thèse, nous proposons également d'utiliser des copules dynamiques pour modéliser des migrations jointes des qualités de crédit afin de prendre en compte les co-migrations extrêmes. En effet, les copules nous permettent d'étendre notre connaissance des processus de migration mono-variable à un cadre multi-variables. Afin de tenir compte de multiples sources de risques systémiques, nous développons des copules dynamiques à plusieurs facteurs. Enfin, nous montrons que la brique élémentaire de structure de dépendance induite par une mesure du temps aléatoire « Time Changed Process » rentre dans le cadre des copules dynamiques.
Fichier principal
Vignette du fichier
TheseDanielTotouom.pdf (3.06 Mo) Télécharger le fichier
Loading...

Dates et versions

pastel-00003260 , version 1 (09-01-2008)

Identifiants

  • HAL Id : pastel-00003260 , version 1

Citer

Daniel Totouom Tangho. Dynamic copulas: applications to finance & economics. Humanities and Social Sciences. École Nationale Supérieure des Mines de Paris, 2007. English. ⟨NNT : 2007ENMP1469⟩. ⟨pastel-00003260⟩
439 Consultations
1483 Téléchargements

Partager

Gmail Facebook X LinkedIn More