Diese Arbeit beschäftigt sich mit der Theorie der sogenannten stochastischen Vorwärts-Rückwärts-Differentialgleichungen (FBSDE), welche als ein stochastisches Anologon und in gewisser Weise als eine Verallgemeinerung von parabolischen quasi-linearen partiellen Differentialgleichungen betrachtet werden können. Die Dissertation besteht aus zwei Teilen: In dem ersten entwicklen wir die Theorie der sogenannten Entkopplungsfelder für allgemeine mehrdimensionale stark gekoppelte FBSDE. Diese Theorie besteht aus Existenz- sowie Eindeutigkeitsresultaten basierend auf dem Konzept des maximalen Intervalls. Es beinhaltet darüberhinaus Werkzeuge um Regularität von konkreten Problemen zu untersuchen. Insgesamt wird die Theorie für drei Klassen von Problemen entwickelt: In dem ersten Fall werden Lipschitz-Bedingungen an die Parameter des Problems vorausgesetzt, welche zugleich vom Zufall abhängen dürfen. Die Untersuchung der beiden anderen Klassen basiert auf dem ersten. In diesen werden die Parameter als deterministisch vorausgesetzt. Gleichwohl wird die Lipschitz-Stetigkeit durch zwei verschiedene Formen der lokalen Lipschitz-Stetigkeit abgeschwächt. In dem zweiten Teil werden diese abstrakten Resultate auf drei konkrete Probleme angewendet: In der ersten Anwendung wird gezeigt wie globale Lösbarkeit von FBSDE in dem sogenannten nicht-degenerierten Fall untersucht werden kann. In der zweiten Anwendung wird die Lösbarkeit eines gekoppelten Systems gezeigt, welches eine Lösung zu dem Skorokhod''schen Einbettungproblem liefert. Die Lösung wird für den Fall einer allgemeinen nicht-linearen Drift konstruiert. Die dritte Anwendung führt auf Lösbarkeit eines komplexen gekoppelten Vorwärt-Rückwärts-Systems, aus welchem optimale Strategien für das Problem der Nutzenmaximierung in unvollständingen Märkten konstruiert werden. Das System wird in einem verhältnismäßig allgmeinen Rahmen gelöst, d.h. für eine verhältnismäßig allgemeine Klasse von Nutzenfunktion auf den reellen Zahlen. / This thesis deals with the theory of so called forward-backward stochastic differential equations (FBSDE) which can be seen as a stochastic formulation and in some sense generalization of parabolic quasi-linear partial differential equations. The thesis consist of two parts: In the first we develop the theory of so called decoupling fields for general multidimensional fully coupled FBSDE in a Brownian setting. The theory consists of uniqueness and existence results for decoupling fields on the so called the maximal interval. It also provides tools to investigate well-posedness and regularity for particular problems. In total the theory is developed for three different classes of FBSDE: In the first Lipschitz continuity of the parameter functions is required, which at the same time are allowed to be random. The other two classes we investigate are based on the theory developed for the first one. In both of them all parameter functions have to be deterministic. However, two different types of local Lipschitz continuity replace the more restrictive Lipschitz continuity of the first class. In the second part we apply these techniques to three different problems: In the first application we demonstrate how well-posedness of FBSDE in the so called non-degenerate case can be investigated. As a second application we demonstrate the solvability of a system, which provides a solution to the so called Skorokhod embedding problem (SEP) via FBSDE. The solution to the SEP is provided for the case of general non-linear drift. The third application provides solutions to a complex FBSDE from which optimal trading strategies for a problem of utility maximization in incomplete markets are constructed. The FBSDE is solved in a relatively general setting, i.e. for a relatively general class of utility functions on the real line.

É muito comum modelar problemas em finanças com processos estocásticos, dada a incerteza de suas variáveis de análise. Além disso, problemas reais nesse domínio são, em geral, de grande custo computacional, o que sugere a utilização de plataformas de alto desempenho (HPC) em sua implementação. As novas gerações de arquitetura de hardware gráfico (GPU) possibilitam a programação de propósito geral enquanto mantêm alta banda de memória e grande poder computacional. Assim, esse tipo de arquitetura vem se mostrando como uma excelente alternativa em HPC. Com isso, a proposta principal desse trabalho é estudar o ferramental matemático e computacional necessário para modelagem estocástica em finanças com a utilização de GPUs como plataforma de aceleração. Para isso, apresentamos a GPU como uma plataforma de computação de propósito geral. Em seguida, analisamos uma variedade de geradores de números aleatórios, tanto em arquitetura sequencial quanto paralela. Além disso, apresentamos os conceitos fundamentais de Cálculo Estocástico e de método de Monte Carlo para simulação estocástica em finanças. Ao final, apresentamos dois estudos de casos de problemas em finanças: \"Stops Ótimos\" e \"Cálculo de Risco de Mercado\". No primeiro caso, resolvemos o problema de otimização de obtenção do ganho ótimo em uma estratégia de negociação de ações de \"Stop Gain\". A solução proposta é escalável e de paralelização inerente em GPU. Para o segundo caso, propomos um algoritmo paralelo para cálculo de risco de mercado, bem como técnicas para melhorar a solução obtida. Nos nossos experimentos, houve uma melhora de 4 vezes na qualidade da simulação estocástica e uma aceleração de mais de 50 vezes. / Given the uncertainty of their variables, it is common to model financial problems with stochastic processes. Furthermore, real problems in this area have a high computational cost. This suggests the use of High Performance Computing (HPC) to handle them. New generations of graphics hardware (GPU) enable general purpose computing while maintaining high memory bandwidth and large computing power. Therefore, this type of architecture is an excellent alternative in HPC and comptutational finance. The main purpose of this work is to study the computational and mathematical tools needed for stochastic modeling in finance using GPUs. We present GPUs as a platform for general purpose computing. We then analyze a variety of random number generators, both in sequential and parallel architectures, and introduce the fundamental mathematical tools for Stochastic Calculus and Monte Carlo simulation. With this background, we present two case studies in finance: ``Optimal Trading Stops\'\' and ``Market Risk Management\'\'. In the first case, we solve the problem of obtaining the optimal gain on a stock trading strategy of ``Stop Gain\'\'. The proposed solution is scalable and with inherent parallelism on GPU. For the second case, we propose a parallel algorithm to compute market risk, as well as techniques for improving the quality of the solutions. In our experiments, there was a 4 times improvement in the quality of stochastic simulation and an acceleration of over 50 times.

In spite of remarkable progress in molecular biology, our understanding of the dynamics and functions of intra- and inter-cellular biological networks has been hampered by their complexity. Kinetics modelling, an important type of mathematical modelling, provides a rigorous and reliable way to reveal the complexity of biological networks. In this thesis, two genetic regulatory networks have been investigated via kinetic models. In the first part of the study, a model is developed to represent the transcriptional regulatory network essential for the circadian rhythms in Drosophila. The model incorporates the transcriptional feedback loops revealed so far in the network of the circadian clock (PER/TIM and VRI/PDP1 loops). Conventional Hill functions are not used to describe the regulation of genes, instead the explicit reactions of binding and unbinding processes of transcription factors to promoters are modelled. The model is described by a set of ordinary differential equations and the parameters are estimated from the in vitro experimental data of the clocks' components. The simulation results show that the model reproduces sustained circadian oscillations in mRNA and protein concentrations that are in agreement with experimental observations. It also simulates the entrainment by light-dark cycles, the disappearance of the rhythmicity in constant light and the shape of phase response curves resembling that of experimental results. The model is robust over a wide range of parameter variations. In addition, the simulated E-box mutation, perS and perL mutants are similar to that observed in the experiments. The deficiency between the simulated mRNA levels and experimental observations in per01, tim01 and clkJrk mutants suggests some differences in the model from reality. Finally, a possible function of VRI/PDP1 loops is proposed to increase the robustness of the clock. In the second part of the study, the sources of intrinsic noise and the influence of extrinsic noise are investigated on an intracellular viral infection system. The contribution of the intrinsic noise from each reaction is measured by means of a special form of stochastic differential equation, the chemical Langevin equation. The intrinsic noise of the system is the linear sum of the noise in each of the reactions. The intrinsic noise arises mainly from the degradation of mRNA and the transcription processes. Then, the effects of extrinsic noise are studied by means of a general form of stochastic differential equation. It is found that the noise of the viral components grows logarithmically with increasing noise intensities. The system is most susceptible to noise in the virus assembly process. A high level of noise in this process can even inhibit the replication of the viruses. In summary, the success of this thesis demonstrates the usefulness of models for interpreting experimental data, developing hypotheses, as well as for understanding the design principles of genetic regulatory networks.

The transport of cosmic rays in the heliosphere is studied by making use of a newly developed modulation model. This model employes stochastic differential equations to numerically solve the relevant transport equation, making use of this approach’s numerical advantages as well as the opportunity to extract additional information regarding cosmic ray transport and the processes responsible for it. The propagation times and energy losses of galactic electrons and protons are calculated for different drift cycles. It is confirmed that protons and electrons lose the same amount of rigidity when they experience the same transport processes. These particles spend more time in the heliosphere, and also lose more energy, in the drift cycle where they drift towards Earth mainly along the heliospheric current sheet. The propagation times of galactic protons from the heliopause to Earth are calculated for increasing heliospheric tilt angles and it is found that current sheet drift becomes less effective with increasing solar activity. Comparing calculated propagation times of Jovian electrons with observations, the transport parameters are constrained to find that 50% of 6 MeV electrons measured at Earth are of Jovian origin. Charge-sign dependent modulation is modelled by simulating the proton to anti-proton ratio at Earth and comparing the results to recent PAMELA observations. A hybrid cosmic ray modulation model is constructed by coupling the numerical modulation model to the heliospheric environment as simulated by a magneto-hydrodynamic model. Using this model, it is shown that cosmic ray modulation persists beyond the heliopause. The level of modulation in this region is found to exhibit solar cycle related changes and, more importantly, is independent of the magnitude of the individual diffusion coefficients, but is rather determined by the ratio of parallel to perpendicular diffusion. / PhD (Space Physics), North-West University, Potchefstroom Campus, 2013

The transport of cosmic rays in the heliosphere is studied by making use of a newly developed modulation model. This model employes stochastic differential equations to numerically solve the relevant transport equation, making use of this approach’s numerical advantages as well as the opportunity to extract additional information regarding cosmic ray transport and the processes responsible for it. The propagation times and energy losses of galactic electrons and protons are calculated for different drift cycles. It is confirmed that protons and electrons lose the same amount of rigidity when they experience the same transport processes. These particles spend more time in the heliosphere, and also lose more energy, in the drift cycle where they drift towards Earth mainly along the heliospheric current sheet. The propagation times of galactic protons from the heliopause to Earth are calculated for increasing heliospheric tilt angles and it is found that current sheet drift becomes less effective with increasing solar activity. Comparing calculated propagation times of Jovian electrons with observations, the transport parameters are constrained to find that 50% of 6 MeV electrons measured at Earth are of Jovian origin. Charge-sign dependent modulation is modelled by simulating the proton to anti-proton ratio at Earth and comparing the results to recent PAMELA observations. A hybrid cosmic ray modulation model is constructed by coupling the numerical modulation model to the heliospheric environment as simulated by a magneto-hydrodynamic model. Using this model, it is shown that cosmic ray modulation persists beyond the heliopause. The level of modulation in this region is found to exhibit solar cycle related changes and, more importantly, is independent of the magnitude of the individual diffusion coefficients, but is rather determined by the ratio of parallel to perpendicular diffusion. / PhD (Space Physics), North-West University, Potchefstroom Campus, 2013

This thesis essentially deals with the development and numerical explorations of a few improved Monte Carlo filters for nonlinear dynamical systems with a view to estimating the associated states and parameters (i.e. the hidden states appearing in the system or process model) based on the available noisy partial observations. The hidden states are characterized, subject to modelling errors, by the weak solutions of the process model, which is typically in the form of a system of stochastic ordinary differential equations (SDEs). The unknown system parameters, when included as pseudo-states within the process model, are made to evolve as Wiener processes. The observations may also be modelled by a set of measurement SDEs or, when collected at discrete time instants, their temporally discretized maps. The proposed Monte Carlo filters aim at achieving robustness (i.e. insensitivity to variations in the noise parameters) and higher accuracy in the estimates whilst retaining the important feature of applicability to large dimensional nonlinear filtering problems. The thesis begins with a brief review of the literature in Chapter 1. The first development, reported in Chapter 2, is that of a nearly exact, semi-analytical, weak and explicit linearization scheme called Girsanov Corrected Linearization Method (GCLM) for nonlinear mechanical oscillators under additive stochastic excitations. At the heart of the linearization is a temporally localized rejection sampling strategy that, combined with a resampling scheme, enables selecting from and appropriately modifying an ensemble of locally linearized trajectories whilst weakly applying the Girsanov correction (the Radon- Nikodym derivative) for the linearization errors. Through their numeric implementations for a few workhorse nonlinear oscillators, the proposed variants of the scheme are shown to exhibit significantly higher numerical accuracy over a much larger range of the time step size than is possible with the local drift-linearization schemes on their own. The above scheme for linearization correction is exploited and extended in Chapter 3, wherein novel variations within a particle filtering algorithm are proposed to weakly correct for the linearization or integration errors that occur while numerically propagating the process dynamics. Specifically, the correction for linearization, provided by the likelihood or the Radon-Nikodym derivative, is incorporated in two steps. Once the likelihood, an exponential martingale, is split into a product of two factors, correction owing to the first factor is implemented via rejection sampling in the first step. The second factor, being directly computable, is accounted for via two schemes, one employing resampling and the other, a gain-weighted innovation term added to the drift field of the process SDE thereby overcoming excessive sample dispersion by resampling. The proposed strategies, employed as add-ons to existing particle filters, the bootstrap and auxiliary SIR filters in this work, are found to non-trivially improve the convergence and accuracy of the estimates and also yield reduced mean square errors of such estimates visà-vis those obtained through the parent filtering schemes. In Chapter 4, we explore the possibility of unscented transformation on Gaussian random variables, as employed within a scaled Gaussian sum stochastic filter, as a means of applying the nonlinear stochastic filtering theory to higher dimensional system identification problems. As an additional strategy to reconcile the evolving process dynamics with the observation history, the proposed filtering scheme also modifies the process model via the incorporation of gain-weighted innovation terms. The reported numerical work on the identification of dynamic models of dimension up to 100 is indicative of the potential of the proposed filter in realizing the stated aim of successfully treating relatively larger dimensional filtering problems. We propose in Chapter 5 an iterated gain-based particle filter that is consistent with the form of the nonlinear filtering (Kushner-Stratonovich) equation in our attempt to treat larger dimensional filtering problems with enhanced estimation accuracy. A crucial aspect of the proposed filtering set-up is that it retains the simplicity of implementation of the ensemble Kalman filter (EnKF). The numerical results obtained via EnKF-like simulations with or without a reduced-rank unscented transformation also indicate substantively improved filter convergence. The final contribution, reported in Chapter 6, is an iterative, gain-based filter bank incorporating an artificial diffusion parameter and may be viewed as an extension of the iterative filter in Chapter 5. While the filter bank helps in exploring the phase space of the state variables better, the iterative strategy based on the artificial diffusion parameter, which is lowered to zero over successive iterations, helps improve the mixing property of the associated iterative update kernels and these are aspects that gather importance for highly nonlinear filtering problems, including those involving significant initial mismatch of the process states and the measured ones. Numerical evidence of remarkably enhanced filter performance is exemplified by target tracking and structural health assessment applications. The thesis is finally wound up in Chapter 7 by summarizing these developments and briefly outlining the future research directions

The goal of this thesis in finance is to combine the use of advanced mathematical methods with a return to foundational economic issues. In that perspective, I study generalized rational expectations and asset pricing in Chapter 2, and a converse comparison principle for backward stochastic differential equations with jumps in Chapter 3. Since the use of stochastic methods in finance is an interesting and complex issue in itself - if only to clarify the difference between the use of mathematical models in finance and in physics or biology - I also present a philosophical reflection on the interpretation of mathematical models in finance (Chapter 4). In Chapter 5, I conclude the thesis with an essay on the history and interpretation of mathematical probability - to be read while keeping in mind the fundamental role of mathematical probability in financial models. / Doctorat en Sciences économiques et de gestion / info:eu-repo/semantics/nonPublished

Cette thèse est composée de trois parties. Dans la première nous montrons l'existence et l'unicité de la solution continue et à croissance polynomiale, au sensviscosité, du système non linéaire de m équations variationnelles de type intégro-différentiel à obstacles unilatéraux interconnectés. Ce système est lié au problème du switching optimal stochastique lorsque le bruit est dirigé par un processus de Lévy. Un cas particulier du système correspond en effet à l’équation d’Hamilton-Jacobi-Bellman associé au problème du switching et la solution de ce système n’est rien d’autre que la fonction valeur du problème. Ensuite, nous étudions un système d’équations intégro-différentielles à obstacles bilatéraux interconnectés. Nous montrons l’existence et l’unicité des solutions continus à croissance polynomiale, au sens viscosité, des systèmes min-max et max-min. La démarche conjugue les systèmes d’EDSR réfléchies ainsi que la méthode de Perron. Dans la dernière partie nous montrons l’égalité des solutions des systèmes max-min et min-max d’EDP lorsque le bruit est uniquement de type diffusion. Nous montrons que si les coûts de switching sont assez réguliers alors ces solutions coïncident. De plus elles sont caractérisées comme fonction valeur du jeu de switching de somme nulle. / There are three main results in this thesis. The first is existence and uniqueness of the solution in viscosity sense for a system of nonlinear m variational integral-partial differential equations with interconnected obstacles. From the probabilistic point of view, this system is related to optimal stochastic switching problem when the noise is driven by a Lévy process. As a by-product we obtain that the value function of the switching problem is continuous and unique solution of its associated Hamilton-Jacobi-Bellman system of equations. Next, we study a general class of min-max and max-min nonlinear second-order integral-partial variational inequalities with interconnected bilateralobstacles, related to a multiple modes zero-sum switching game with jumps. Using Perron’s method and by the help of systems of penalized unilateral reflected backward SDEs with jumps, we construct a continuous with polynomial growth viscosity solution, and a comparison result yields the uniqueness of the solution. At last, we deal with the solutions of systems of PDEs with bilateral inter-connected obstacles of min-max and max-min types in the Brownian framework. These systems arise naturally in stochastic switching zero-sum game problems. We show that when the switching costs of one side are smooth, the solutions of the min-max and max-min systems coincide. Furthermore, this solution is identified as the value function of the zero-sum switching game.

É muito comum modelar problemas em finanças com processos estocásticos, dada a incerteza de suas variáveis de análise. Além disso, problemas reais nesse domínio são, em geral, de grande custo computacional, o que sugere a utilização de plataformas de alto desempenho (HPC) em sua implementação. As novas gerações de arquitetura de hardware gráfico (GPU) possibilitam a programação de propósito geral enquanto mantêm alta banda de memória e grande poder computacional. Assim, esse tipo de arquitetura vem se mostrando como uma excelente alternativa em HPC. Com isso, a proposta principal desse trabalho é estudar o ferramental matemático e computacional necessário para modelagem estocástica em finanças com a utilização de GPUs como plataforma de aceleração. Para isso, apresentamos a GPU como uma plataforma de computação de propósito geral. Em seguida, analisamos uma variedade de geradores de números aleatórios, tanto em arquitetura sequencial quanto paralela. Além disso, apresentamos os conceitos fundamentais de Cálculo Estocástico e de método de Monte Carlo para simulação estocástica em finanças. Ao final, apresentamos dois estudos de casos de problemas em finanças: \"Stops Ótimos\" e \"Cálculo de Risco de Mercado\". No primeiro caso, resolvemos o problema de otimização de obtenção do ganho ótimo em uma estratégia de negociação de ações de \"Stop Gain\". A solução proposta é escalável e de paralelização inerente em GPU. Para o segundo caso, propomos um algoritmo paralelo para cálculo de risco de mercado, bem como técnicas para melhorar a solução obtida. Nos nossos experimentos, houve uma melhora de 4 vezes na qualidade da simulação estocástica e uma aceleração de mais de 50 vezes. / Given the uncertainty of their variables, it is common to model financial problems with stochastic processes. Furthermore, real problems in this area have a high computational cost. This suggests the use of High Performance Computing (HPC) to handle them. New generations of graphics hardware (GPU) enable general purpose computing while maintaining high memory bandwidth and large computing power. Therefore, this type of architecture is an excellent alternative in HPC and comptutational finance. The main purpose of this work is to study the computational and mathematical tools needed for stochastic modeling in finance using GPUs. We present GPUs as a platform for general purpose computing. We then analyze a variety of random number generators, both in sequential and parallel architectures, and introduce the fundamental mathematical tools for Stochastic Calculus and Monte Carlo simulation. With this background, we present two case studies in finance: ``Optimal Trading Stops\'\' and ``Market Risk Management\'\'. In the first case, we solve the problem of obtaining the optimal gain on a stock trading strategy of ``Stop Gain\'\'. The proposed solution is scalable and with inherent parallelism on GPU. For the second case, we propose a parallel algorithm to compute market risk, as well as techniques for improving the quality of the solutions. In our experiments, there was a 4 times improvement in the quality of stochastic simulation and an acceleration of over 50 times.

Dans le présent document on aborde trois divers thèmes liés au contrôle et au calcul stochastiques, qui s'appuient sur la notion d'équation différentielle stochastique rétrograde (EDSR) dirigée par une mesure aléatoire. Les trois premiers chapitres de la thèse traitent des problèmes de contrôle optimal pour différentes catégories de processus markoviens non-diffusifs, à horizon fini ou infini. Dans chaque cas, la fonction valeur, qui est l'unique solution d'une équation intégro-différentielle de Hamilton-Jacobi-Bellman (HJB), est représentée comme l'unique solution d'une EDSR appropriée. Dans le premier chapitre, nous contrôlons une classe de processus semi-markoviens à horizon fini; le deuxième chapitre est consacré au contrôle optimal de processus markoviens de saut pur, tandis qu'au troisième chapitre, nous examinons le cas de processus markoviens déterministes par morceaux (PDMPs) à horizon infini. Dans les deuxième et troisième chapitres les équations d'HJB associées au contrôle optimal sont complètement non-linéaires. Cette situation survient lorsque les lois des processus contrôlés ne sont pas absolument continues par rapport à la loi d'un processus donné. Etant donné ce caractère complètement non-linéaire, ces équations ne peuvent pas être représentées par des EDSRs classiques. Dans ce cadre, nous avons obtenu des formules de Feynman-Kac non-linéaires en généralisant la méthode de la randomisation du contrôle introduite par Kharroubi et Pham (2015) pour les diffusions. Ces techniques nous permettent de relier la fonction valeur du problème de contrôle à une EDSR dirigée par une mesure aléatoire, dont une composante de la solution subit une contrainte de signe. En plus, on démontre que la fonction valeur du problème de contrôle originel non dominé coïncide avec la fonction valeur d'un problème de contrôle dominé auxiliaire, exprimé en termes de changements de mesures équivalentes de probabilité. Dans le quatrième chapitre, nous étudions une équation différentielle stochastique rétrograde à horizon fini, dirigée par une mesure aléatoire à valeurs entières sur $R_+ times E$, o`u $E$ est un espace lusinien, avec compensateur de la forme $nu(dt, dx) = dA_t phi_t(dx)$. Le générateur de cette équation satisfait une condition de Lipschitz uniforme par rapport aux inconnues. Dans la littérature, l'existence et unicité pour des EDSRs dans ce cadre ont été établies seulement lorsque $A$ est continu ou déterministe. Nous fournissons un théorème d'existence et d'unicité même lorsque $A$ est un processus prévisible, non décroissant, continu à droite. Ce résultat s’applique par exemple, au cas du contrôle lié aux PDMPs. En effet, quand $mu$ est la mesure de saut d'un PDMP sur un domaine borné, $A$ est prévisible et discontinu. Enfin, dans les deux derniers chapitres de la thèse nous traitons le calcul stochastique pour des processus discontinus généraux. Dans le cinquième chapitre, nous développons le calcul stochastique via régularisations des processus à sauts qui ne sont pas nécessairement des semimartingales. En particulier nous poursuivons l'étude des processus dénommés de Dirichlet faibles, dans le cadre discontinu. Un tel processus $X$ est la somme d'une martingale locale et d'un processus adapté $A$ tel que $[N, A] = 0$, pour toute martingale locale continue $N$. Pour une fonction $u: [0, T] times R rightarrow R$ de classe $C^{0,1}$ (ou parfois moins), on exprime un développement de $u(t, X_t)$, dans l'esprit d'une généralisation du lemme d'Itô, lequel vaut lorsque $u$ est de classe $C^{1,2}$. Le calcul est appliqué dans le sixième chapitre à la théorie des EDSRs dirigées par des mesures aléatoires. Dans de nombreuses situations, lorsque le processus sous-jacent $X$ est une semimartingale spéciale, ou plus généralement, un processus de Dirichlet spécial faible, nous identifions les solutions des EDSRs considérées via le processus $X$ et la solution $u$ d’une EDP intégro-différentielle associée. / In the present document we treat three different topics related to stochastic optimal control and stochastic calculus, pivoting on thenotion of backward stochastic differential equation (BSDE) driven by a random measure.After a general introduction, the three first chapters of the thesis deal with optimal control for different classes of non-diffusiveMarkov processes, in finite or infinite horizon. In each case, the value function, which is the unique solution to anintegro-differential Hamilton-Jacobi-Bellman (HJB) equation, is probabilistically represented as the unique solution of asuitable BSDE. In the first chapter we control a class of semi-Markov processes on finite horizon; the second chapter isdevoted to the optimal control of pure jump Markov processes, while in the third chapter we consider the case of controlled piecewisedeterministic Markov processes (PDMPs) on infinite horizon. In the second and third chapters the HJB equations associatedto the optimal control problems are fully nonlinear. Those situations arise when the laws of the controlled processes arenot absolutely continuous with respect to the law of a given, uncontrolled, process. Since the corresponding HJB equationsare fully nonlinear, they cannot be represented by classical BSDEs. In these cases we have obtained nonlinear Feynman-Kacrepresentation formulae by generalizing the control randomization method introduced in Kharroubi and Pham (2015)for classical diffusions. This approach allows us to relate the value function with a BSDE driven by a random measure,whose solution hasa sign constraint on one of its components.Moreover, the value function of the original non-dominated control problem turns out to coincide withthe value function of an auxiliary dominated control problem, expressed in terms of equivalent changes of probability measures.In the fourth chapter we study a backward stochastic differential equation on finite horizon driven by an integer-valued randommeasure $mu$ on $R_+times E$, where $E$ is a Lusin space, with compensator $nu(dt,dx)=dA_t,phi_t(dx)$. The generator of thisequation satisfies a uniform Lipschitz condition with respect to the unknown processes.In the literature, well-posedness results for BSDEs in this general setting have only been established when$A$ is continuous or deterministic. We provide an existence and uniqueness theorem for the general case, i.e.when $A$ is a right-continuous nondecreasing predictable process. Those results are relevant, for example,in the frameworkof control problems related to PDMPs. Indeed, when $mu$ is the jump measure of a PDMP on a bounded domain, then $A$ is predictable and discontinuous.Finally, in the two last chapters of the thesis we deal with stochastic calculus for general discontinuous processes.In the fifth chapter we systematically develop stochastic calculus via regularization in the case of jump processes,and we carry on the investigations of the so-called weak Dirichlet processes in the discontinuous case.Such a process $X$ is the sum of a local martingale and an adapted process $A$ such that $[N,A] = 0$, for any continuouslocal martingale $N$.Given a function $u:[0,T] times R rightarrow R$, which is of class $C^{0,1}$ (or sometimes less), we provide a chain rule typeexpansion for $u(t,X_t)$, which constitutes a generalization of It^o's lemma being valid when $u$ is of class $C^{1,2}$.This calculus is applied in the sixth chapter to the theory of BSDEs driven by random measures.In several situations, when the underlying forward process $X$ is a special semimartingale, or, even more generally,a special weak Dirichlet process,we identify the solutions $(Y,Z,U)$ of the considered BSDEs via the process $X$ and the solution $u$ to an associatedintegro PDE.