With the acceptance of the new Basel II banking regulation (implemented in South Africa in January 2008) the search for improved ways of modeling the most important banking activities has become very topical. Since the notion of Levy-process was introduced, it has emerged as an important tool for modeling economic variables in a Basel II framework. In this study, we investigate the stochastic dynamics of banking items that are driven by such processes. In particular, we discuss bank provisioning for loan losses and deposit withdrawals. The first type of provisioning is related to the earnings that the bank sets aside in order to cover loan defaults. In this case, we apply principles from robustness to a situation where the decision maker is a bank owner and the decision rule determines the optimal provisioning strategy for loan losses. In this regard, we formulate a dynamic banking loan loss model involving a provisioning portfolio consisting of provisions for expected losses and loan loss reserves for unexpected losses. Here, unexpected loan losses and provisioning for expected losses are modeled via a compound Poisson process and an exponential Levy process, respectively. We use historical evidence from OECD (Organization for Economic Corporation and Development) countries to support the fact that the provisions for loan losses-to-total assets ratio is negatively correlated with aggregate asset prices and the private credit-to-GDP ratio. Secondly, we construct models for provisioning for deposit withdrawals. In particular, we build stochastic dynamic models which enable us to analyze the interplay between deposit withdrawals and the provisioning for these withdrawals via Treasuries and reserves. Further insight is gained by considering a numerical problem and a simulation of the trajectory of the stochastic dynamics of the sum of the Treasuries and reserves. Since managing the risk that depositors will exercise their withdrawal option is an important aspect of this thesis, we consider the idea of a hedging provisioning strategy for deposit withdrawals in an incomplete market setting. In this spirit, we discuss an optimal risk management problem for a commercial bank whose main activity is to obtain funds through deposits from the public and use the Treasuries and reserves to cater for the resulting withdrawals. Finally, we provide a brief analysis of some of the issues arising from the dynamic models of the banking items derived. / Thesis (Ph.D. (Computer, Statistical and Mathematical Sciences))--North-West University, Potchefstroom Campus, 2008.

Pour V un processus aléatoire càd-làg, on appelle diffusion dans le milieu aléatoire V la solution formelle de l’équation différentielle stochastique \[ dX_t = - \frac1{2} V'(X_t) dt + dB_t, \] où B est un mouvement brownien indépendant de V . Le temps local au temps t et à la position x dela diffusion, noté LX(t, x), donne une mesure de la quantité de temps passé par la diffusion au point x, avant l’instant t. Dans cette thèse nous considérons le cas où le milieu V est un processus de Lévyspectralement négatif convergeant presque sûrement vers −∞, et nous nous intéressons au comportementasymptotique lorsque t tend vers l’infini de $\mathcal{L}_X^*(t) := \sup_{\mathbb{R}} \mathcal{L}_X(t, .)$ le supremum du temps local de ladiffusion, ainsi qu’à la localisation du point le plus visité par la diffusion. Nous déterminons notammentla convergence en loi et le comportement presque sûr du supremum du temps local. Cette étude révèleque le comportement asymptotique du supremum du temps local est fortement lié aux propriétés desfonctionnelles exponentielles des processus de Lévy conditionnés à rester positifs et cela nous amène àétudier ces dernières. Si V est un processus de Lévy, V ↑ désigne le processus V conditionné à rester positif.La fonctionnelle exponentielle de V ↑ est la variable aléatoire $\int_0^{+ \infty} e^{- V^{\uparrow} (t)}dt$ . Nous étudions en particulier sa finitude, son auto-décomposabilité, l’existence de moments exponentiels, sa queue en 0, l’existence et larégularité de sa densité. / For V a random càd-làg process, we call diffusion in the random medium V the formal solution of thestochastic differential equation \[ dX_t = - \frac1{2} V'(X_t) dt + dB_t, \] where B is a brownian motion independent of V . The local time at time t and at the position x of thediffusion, denoted by LX(t, x), gives a measure of the amount of time spent by the diffusion at point x,before instant t. In this thesis we consider the case where the medium V is a spectrally negative Lévyprocess converging almost surely toward −∞, and we are interested in the asymptotic behavior, whent goes to infinity, of $\mathcal{L}_X^*(t) := \sup_{\mathbb{R}} \mathcal{L}_X(t, .)$ the supremum of the local time of the diffusion. We arealso interested in the localization of the point most visited by the diffusion. We notably establish theconvergence in distribution and the almost sure behavior of the supremum of the local time. This studyreveals that the asymptotic behavior of the supremum of the local time is deeply linked to the propertiesof the exponential functionals of Lévy processes conditioned to stay positive and this brings us to studythem. If V is a Lévy process, V ↑ denotes the process V conditioned to stay positive. The exponentialfunctional of V ↑ is the random variable $\int_0^{+ \infty} e^{- V^{\uparrow} (t)}dt$ . For this object, we study in particular finiteness,

In this thesis, an optimal investment problem is studied for an investor who can only invest in a financial market modelled by an Itô-Lévy process; with one risk free (bond) and one risky (stock) investment possibility. We present the dynamic programming method and the associated Hamilton-Jacobi-Bellman (HJB) equation to explicitly solve this problem. It is shown that with purification and simplification to the standard jump diffusion process, closed form solutions for the optimal investment strategy and for the value function are attainable. It is also shown that, an explicit solution can be obtained via a finite training of a neural network using Stochastic gradient descent (SGD) for a specific case.

Ce travail étudie des modèles financiers pour les prix d'options, les taux d'intérêts et le risque de crédit, avec des processus stochastiques à mémoire et comportant des discontinuités. Ces modèles sont formulés en termes du mouvement Brownien fractionnaire, du processus de Lévy fractionnaire ou filtré (et doublement stochastique) et de leurs approximations par des semimartingales. Leur calcul stochastique est traité au sens de Malliavin, et des formules d'Itô sont déduites. Nous caractérisons les probabilités risque neutre en termes de ces processus pour des modèles d'évaluation d'options de type de Black-Scholes avec sauts. Nous étudions également des modèles de taux d'intérêts, en particulier les modèles de Vasicek, de Cox-Ingersoll-Ross et de Heath-Jarrow-Morton. Finalement nous étudions la modélisation du risque de crédit / This work investigates financial models for option pricing, interest rates and credit risk with stochastic processes that have memory and discontinuities. These models are formulated in terms of the fractional Brownian motion, the fractional or filtered Lévy process (also doubly stochastic) and their approximations by semimartingales. Their stochastic calculus is treated in the sense of Malliavin and Itô formulas are derived. We characterize the risk-neutral probability measures in terms of these processes for options pricing models of Black-Scholes type with jumps. We also study models of interest rates, in particular the models of Vasicek, Cox-Ingersoll-Ross and Heath-Jarrow-Morton. Finally we study credit risk models

本研究利用GARCH動態過程的優點捕捉匯率報酬率之異質變異與波動度叢聚性質，並以GARCH動態過程為基礎，考慮跳躍風險服從Lévy過程，再利用特徵函數與快速傅立葉轉換方法推導出GARCH-Lévy動態過程下的歐式匯率選擇權解析解。以日圓兌換美元(JPY/USD)之歐式匯率選擇權為實證資料，比較基準GARCH選擇權評價模型與GARCH-Lévy選擇權評價模型對市場真實價格的配適效果與預測能力。實證結果顯示，考慮跳躍風險為無限活躍之Lévy過程，即GARCH-VG與GARCH-NIG匯率選擇權評價模型，不論是樣本內的評價誤差或是在樣本外的避險誤差皆勝於考慮跳躍風險為有限活躍Lévy過程的GARCH-MJ匯率選擇權評價模型。整體而言，本研究發現進行匯率選擇權之評價時，GARCH-NIG匯率選擇權評價模型有較小的樣本內及樣本外評價誤差。 / In this thesis, we make use of GARCH dynamic to capture volatility clustering and heteroskedasticity in exchange rate. We consider a jump risk which follows Lévy process based on GARCH model. Furthermore, we use characteristic function and fast fourier transform to derive the currency option pricing formula under GARCH-Lévy process. We collect the JPY/USD exchange rate data for our empirical analysis and then compare the goodness of fit and prediction performance between GARCH benchmark and GARCH-Lévy currency option pricing model. The empirical results show that either in-sample pricing error or out-of-sample hedging performance, the infinite-activity Lévy process, GARCH-VG and GARCH-NIG option pricing model is better than finite-activity Lévy process, GARCH-MJ option pricing model. Overall, we find using GARCH-NIG currency option pricing model can achieve the lower in-sample and out-of sample pricing error.

Title: Stochastic Models in Financial Mathematics Author: Bc. Oliver Waczulík Department: Department of Probability and Mathematical Statistics Supervisor: doc. RNDr. Jan Hurt, CSc., Department of Probability and Mathe- matical Statistics Abstract: This thesis looks into the problems of ordinary stochastic models used in financial mathematics, which are often influenced by unrealistic assumptions of Brownian motion. The thesis deals with and suggests more sophisticated alternatives to Brownian motion models. By applying the fractional Brownian motion we derive a modification of the Black-Scholes pricing formula for a mixed fractional Bro- wnian motion. We use Lévy processes to introduce subordinated stable process of Ornstein-Uhlenbeck type serving for modeling interest rates. We present the calibration procedures for these models along with a simulation study for estima- tion of Hurst parameter. To illustrate the practical use of the models introduced in the paper we have used real financial data and custom procedures program- med in the system Wolfram Mathematica. We have achieved almost 90% decline in the value of Kolmogorov-Smirnov statistics by the application of subordinated stable process of Ornstein-Uhlenbeck type for the historical values of the monthly PRIBOR (Prague Interbank Offered Rate) rates in...

Les limites d’échelle de grands arbres aléatoires jouent un rôle central dans cette thèse.Nous nous intéressons plus spécifiquement au comportement asymptotique de plusieurs fonctions codant des arbres de Galton-Watson conditionnés. Nous envisageons plusieurs types de conditionnements faisant intervenir différentes quantités telles que le nombre total de sommets ou le nombre total de feuilles, avec des lois de reproductions différentes.Lorsque la loi de reproduction est critique et appartient au domaine d’attraction d’uneloi stable, un phénomène d’universalité se produit : ces arbres ressemblent à un même arbre aléatoire continu, l’arbre de Lévy stable. En revanche, lorsque la criticalité est brisée, la communauté de physique théorique a remarqué que des phénomènes de condensation peuvent survenir, ce qui signifie qu’avec grande probabilité, un sommet de l’arbre a un degré macroscopique comparable à la taille totale de l’arbre. Une partie de cette thèse consiste à mieux comprendre ce phénomène de condensation. Finalement, nous étudions des configurations non croisées aléatoires, obtenues à partir d’un polygône régulier en traçant des diagonales qui ne s’intersectent pas intérieurement, et remarquons qu’elles sont étroitement reliées à des arbres de Galton-Watson conditionnés à avoir un nombre de feuilles fixé. En particulier, ce lien jette un nouveau pont entre les dissections uniformes et les arbres de Galton-Watson, ce qui permet d’obtenir d’intéressantes conséquences de nature combinatoire. / Scaling limits of large random trees play an important role in this thesis. We are more precisely interested in the asymptotic behavior of several functions coding conditioned Galton-Watson trees. We consider several types of conditioning, involving different quantities such as the total number of vertices or leaves, as well as several types of offspring distributions. When the offspring distribution is critical and belongs to the domainof attraction of a stable law, a universality phenomenon occurs: these trees look like the samecontinuous random tree, the so-called stable Lévy tree. However, when the offspring distributionis not critical, the theoretical physics community has noticed that condensation phenomenamay occur, meaning that with high probability there exists a unique vertex with macroscopicdegree comparable to the total size of the tree. The goal of one of our contributions is to graspa better understanding of this phenomenon. Last but not least, we study random non-crossingconfigurations consisting of diagonals of regular polygons, and notice that they are intimatelyrelated to Galton-Watson trees conditioned on having a fixed number of leaves. In particular,this link sheds new light on uniform dissections and allows us to obtain some interesting resultsof a combinatorial flavor.

由於違約事件不斷發生以及在財務實證上顯示證券的報酬率有厚尾與高狹峰的現象，本文使用縮減式模型與Lévy過程來評價有信用風險下的可轉換公司債。在Lévy過程中，本研究假設股價服從NIG及VG模型，發現此兩種模型比傳統的GBM模型更符合厚尾現象。此外，在Lévy過程參數估計方面，本文使用最大概似法估計參數，在評價可轉換公司債方面，本研究採用最小平方蒙地卡羅法。本文之實證結果顯示，Lévy模型的績效比傳統GBM模型佳。 / Due to the reason that the default events occurred constantly and still continue taking place, empirical log return distributions exhibit fat tail and excess kurtosis, this paper evaluates convertible bonds under Lévy process with default risk using the reduced-form approach. Under the Lévy process, the underlying stock prices are set to be normal inverse Gaussian (NIG) and variance Gamma (VG) model to capture the jump components. In the empirical analysis, we use the maximum likelihood method to estimate the parameters of Lévy distributions, and apply the least squares Monte Carlo Simulation to price convertible bonds. Five examples are shown in pricing convertible bonds using the traditional model and Lévy model. The empirical results show that the performance of Lévy model is better than the traditional one.

本研究以Lévy過程為模型基礎，考慮Merton Jump及跳躍強度服從Hawkes Process的Merton Jump兩種跳躍風險，利用Particle Filter方法及EM演算法估計出模型參數並計算出對數概似值、AIC及BIC。以S&P500指數為實證資料，比較隨機波動度模型、Variance Gamma模型及兩種不同跳躍風險對市場真實價格的配適效果。實證結果顯示，隨機波動度模型其配適效果勝於Variance Gamma模型，且加入跳躍風險後可使模型配適效果提升，尤其在模型中加入跳躍強度服從Hawkes Process的Merton Jump，其配適效果更勝於Merton Jump。整體而言，本研究發現，以S&P500指數為實證資料時，SVHJ模型有較好的配適效果。 / This paper, based on the Lévy process, considers two kinds of jump risk, Merton Jump and the Merton Jump whose jump intensity follows Hawkes Process. We use Particle Filter method and EM Algorithm to estimate the model parameters and calculate the log-likelihood value, AIC and BIC. We collect the S&P500 index for our empirical analysis and then compare the goodness of fit between the stochastic volatility model, the Variance Gamma model and two different jump risks. The empirical results show that the stochastic volatility model is better than the Variance Gamma model, and it is better to consider the jump risk in the model, especially the Merton Jump whose jump intensity follows Hawkes Process. The goodness of fit is better than Merton Jump. Overall, we find SVHJ model has better goodness of fit when S&P500 index was used as the empirical data.

Das Ziel dieser Arbeit ist die Entwicklung eines effizienten parametrischen Schätzverfahrens für den Drift einer durch einen Lévy-Prozess getriebenen Sprungdiffusion. Zunächst werden zeit-stetige Beobachtungen angenommen und auf dieser Basis eine Likelihoodtheorie entwickelt. Dieser Schritt umfasst die Frage nach lokaler Äquivalenz der zu verschiedenen Parametern auf dem Pfadraum induzierten Maße. Wir diskutieren in dieser Arbeit Schätzer für Prozesse vom Ornstein-Uhlenbeck-Typ, Cox-Ingersoll-Ross Prozesse und Lösungen linearer stochastischer Differentialgleichungen mit Gedächtnis im Detail und zeigen starke Konsistenz, asymptotische Normalität und Effizienz im Sinne von Hájek und Le Cam für den Likelihood-Schätzer. In Sprungdiffusionsmodellen ist die Likelihood-Funktion eine Funktion des stetigen Martingalanteils des beobachteten Prozesses, der im Allgemeinen nicht direkt beobachtet werden kann. Wenn nun nur Beobachtungen an endlich vielen Zeitpunkten gegeben sind, so lässt sich der stetige Anteil der Sprungdiffusion nur approximativ bestimmen. Diese Approximation des stetigen Anteils ist ein zentrales Thema dieser Arbeit und es wird uns auf das Filtern von Sprüngen führen. Der zweite Teil dieser Arbeit untersucht die Schätzung der Drifts, wenn nur diskrete Beobachtungen gegeben sind. Dabei benutzen wir die Likelihood-Schätzer aus dem ersten Teil und approximieren den stetigen Martingalanteil durch einen sogenannten Sprungfilter. Wir untersuchen zuerst den Fall endlicher Aktivität und zeigen, dass die Driftschätzer im Hochfrequenzlimes die effiziente asymptotische Verteilung erreichen. Darauf aufbauend beweisen wir dann im Falle unendlicher Sprungaktivität asymptotische Effizienz für den Driftschätzer im Ornstein-Uhlenbeck Modell. Im letzten Teil werden die theoretischen Ergebnisse für die Schätzer auf endlichen Stichproben aus simulierten Daten geprüft und es zeigt sich, dass das Sprungfiltern zu einem deutlichen Effizienzgewinn führen. / The problem of parametric drift estimation for a a Lévy-driven jump diffusion process is considered in two different settings: time-continuous and high-frequency observations. The goal is to develop explicit maximum likelihood estimators for both observation schemes that are efficient in the Hájek-Le Cam sense. The likelihood function based on time-continuous observations can be derived explicitly for jump diffusion models and leads to explicit maximum likelihood estimators for several popular model classes. We consider Ornstein-Uhlenbeck type, square-root and linear stochastic delay differential equations driven by Lévy processes in detail and prove strong consistency, asymptotic normality and efficiency of the likelihood estimators in these models. The appearance of the continuous martingale part of the observed process under the dominating measure in the likelihood function leads to a jump filtering problem in this context, since the continuous part is usually not directly observable and can only be approximated and the high-frequency limit. In the second part of this thesis the problem of drift estimation for discretely observed processes is considered. The estimators are constructed from discretizations of the time-continuous maximum likelihood estimators from the first part, where the continuous martingale part is approximated via a thresholding technique. We are able to proof that even in the case of infinite activity jumps of the driving Lévy process the estimator is asymptotically normal and efficient under weak assumptions on the jump behavior. Finally, the finite sample behavior of the estimators is investigated on simulated data. We find that the maximum likelihood approach clearly outperforms the least squares estimator when jumps are present and that the efficiency gap between both techniques becomes even more severe with growing jump intensity.