The thesis evaluates several hundred one-day-ahead VaR forecasting models in the time period between the years 2004 and 2009 on data from six world stock indices - DJI, GSPC, IXIC, FTSE, GDAXI and N225. The models model mean using the AR and MA processes with up to two lags and variance with one of GARCH, EGARCH or TARCH processes with up to two lags. The models are estimated on the data from the in-sample period and their forecasting ac- curacy is evaluated on the out-of-sample data, which are more volatile. The main aim of the thesis is to test whether a model estimated on data with lower volatility can be used in periods with higher volatility. The evaluation is based on the conditional coverage test and is performed on each stock index sepa- rately. Unlike other works in this eld of study, the thesis does not assume the log-returns to be normally distributed and does not explicitly select a partic- ular conditional volatility process. Moreover, the thesis takes advantage of a less known conditional coverage framework for the measurement of forecasting accuracy.

Submitted by Angela Maria de Oliveira (amolivei@uepg.br) on 2018-02-08T13:10:32Z No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) Ronaldo Mendes Evaristo.pdf: 2297085 bytes, checksum: 293f7e08aca7690caa0d317480f9e18e (MD5) / Made available in DSpace on 2018-02-08T13:10:32Z (GMT). No. of bitstreams: 2 license_rdf: 811 bytes, checksum: e39d27027a6cc9cb039ad269a5db8e34 (MD5) Ronaldo Mendes Evaristo.pdf: 2297085 bytes, checksum: 293f7e08aca7690caa0d317480f9e18e (MD5) Previous issue date: 2017-12-08 / Nos ultimos anos, o aumento da incidência de doenças cardiovasculares na população mundial vem motivando a comunidade científica a buscar novas técnicas ou inovações tecnológicas para complementar os métodos existentes para avaliação do desempenho do coração. Dentre elas, destaca-se a análise da Variabilidade da Frequência Cardíarca (VFC) via eletrocardiograma (ECG), método não invasivo importante na detecção de patologias leves e moderadas cada vez mais frequentes nos seres humanos como doenças coronarianas, arritmias, bradicardias e taquicardias, além de distúrbios na relação entre os sistemas nervosos simpático e parassimpático. Neste trabalho é introduzida uma inovação em um modelo matemático baseado em modulações de exponenciais gaussianas utilizado para reproduzir a morfologia do ECG de seres humanos. Trata-se da introdução de tacogramas gerados por um processo estocástico autorregressivo (AR), previamente à integração das equações diferenciais do modelo, capaz de reproduzir com maior fidelidade a VFC quando comparados com dados experimentais de adultos saudaveis e de adultos com doença arterial coronariana (DAC). Para validar o modelo, os resultados simulados são comparados com dados experimentais via espectro de potencia da transformada wavelet discreta (TWD), gráficos de Poincare e pela analise de flutuação sem tendência. Verificamos que a DAC não altera a morfologia do ECG em situação de repouso, mas influencia significativamente na VFC, sendo que o modelo matemático proposto absorve e reproduz esse comportamento. / In recent years, the increase in the incidence of cardiovascular diseases in the world population has motivated the scientific community to seek new techniques or technological innovations to complement existing methods for assessing heart performance. Among them, stands out the analysis of the Heart Rate Variability (HRV) by electrocardiogram (ECG), an important non-invasive method for the detection of mild and moderate pathologies that are increasingly frequent in humans such as coronary diseases, arrhythmias, bradycardia and tachycardias, besides disturbances in the relationship between the sympathetic and parasympathetic nervous systems. This work introduces an innovation in a mathematical model based on Gaussian exponential modulations used to reproduce the ECG morphology of humans. This is the introduction of tachograms generated by an autoregressive stochastic process (AR), prior to the integration of the diferential equations of the model, capable of reproducing with better delity the HRV when compared with experimental data of healthy adults and adults with coronary artery disease (CAD). In order to validate the model, the simulated results are compared with experimental data using the discrete wavelet transform (DWT) power spectrum, Poincare plots and the detrended uctuation analysis (DFA). We verified that CAD does not change the morphology of ECG in resting state, but it has a significant in uence on HRV, and the proposed mathematical model absorbs and reproduces this behavior.

Les processus autorégressifs à bifurcation (BAR) ont été au centre de nombreux travaux de recherche ces dernières années. Ces processus, qui sont l'adaptation à un arbre binaire des processus autorégressifs, sont en effet d'intérêt en biologie puisque la structure de l'arbre binaire permet une analogie aisée avec la division cellulaire. L'objectif de cette thèse est l'estimation les paramètres de variantes de ces processus autorégressifs à bifurcation, à savoir les processus BAR à valeurs entières et les processus BAR à coefficients aléatoires. Dans un premier temps, nous nous intéressons aux processus BAR à valeurs entières. Nous établissons, via une approche martingale, la convergence presque sûre des estimateurs des moindres carrés pondérés considérés, ainsi qu'une vitesse de convergence de ces estimateurs, une loi forte quadratique et leur comportement asymptotiquement normal. Dans un second temps, on étudie les processus BAR à coefficients aléatoires. Cette étude permet d'étendre le concept de processus autorégressifs à bifurcation en généralisant le côté aléatoire de l'évolution. Nous établissons les mêmes résultats asymptotiques que pour la première étude. Enfin, nous concluons cette thèse par une autre approche des processus BAR à coefficients aléatoires où l'on ne pondère plus nos estimateurs des moindres carrés en tirant parti du théorème de Rademacher-Menchov. / Bifurcating autoregressive (BAR) processes have been widely investigated this past few years. Those processes, which are an adjustment of autoregressive processes to a binary tree structure, are indeed of interest concerning biology since the binary tree structure allows an easy analogy with cell division. The aim of this thesis is to estimate the parameters of some variations of those BAR processes, namely the integer-valued BAR processes and the random coefficients BAR processes. First, we will have a look to integer-valued BAR processes. We establish, via a martingale approach, the almost sure convergence of the weighted least squares estimators of interest, together with a rate of convergence, a quadratic strong law and their asymptotic normality. Secondly, we study the random coefficients BAR processes. The study allows to extend the principle of bifurcating autoregressive processes by enlarging the randomness of the evolution. We establish the same asymptotic results as for the first study. Finally, we conclude this thesis with an other approach of random coefficient BAR processes where we do not weight our least squares estimators any more by making good use of the Rademacher-Menchov theorem.

<p>The Swedish National Debt Office (SNDO) is the Swedish Government’s financial administration. It has several tasks and the main one is to manage the central government’s debt in a way that minimizes the cost with due regard to risk. The debt management problem is to choose currency composition and maturity profile - a problem made difficult because of the many stochastic factors involved. </p><p>The SNDO has created a simulation model to quantitatively analyze different aspects of this problem by evaluating a set of static strategies in a great number of simulated futures. This approach has a number of drawbacks, which might be handled by using a financial optimization approach based on Stochastic Programming. </p><p>The objective of this master’s thesis is thus to apply financial optimization on the Swedish government’s strategic debt management problem, using the SNDO’s simulation model to generate scenarios, and to evaluate this approach against a set of static strategies in fictitious future macroeconomic developments. </p><p>In this report we describe how the SNDO’s simulation model is used along with a clustering algorithm to form future scenarios, which are then used by an optimization model to find an optimal decision regarding the debt management problem. </p><p>Results of the evaluations show that our optimization approach is expected to have a lower average annual real cost, but with somewhat higher risk, than a set of static comparison strategies in a simulated future. These evaluation results are based on a risk preference set by ourselves, since the government has not expressed its risk preference quantitatively. We also conclude that financial optimization is applicable on the government debt management problem, although some work remains before the method can be incorporated into the strategic work of the SNDO.</p>

The Swedish National Debt Office (SNDO) is the Swedish Government’s financial administration. It has several tasks and the main one is to manage the central government’s debt in a way that minimizes the cost with due regard to risk. The debt management problem is to choose currency composition and maturity profile - a problem made difficult because of the many stochastic factors involved. The SNDO has created a simulation model to quantitatively analyze different aspects of this problem by evaluating a set of static strategies in a great number of simulated futures. This approach has a number of drawbacks, which might be handled by using a financial optimization approach based on Stochastic Programming. The objective of this master’s thesis is thus to apply financial optimization on the Swedish government’s strategic debt management problem, using the SNDO’s simulation model to generate scenarios, and to evaluate this approach against a set of static strategies in fictitious future macroeconomic developments. In this report we describe how the SNDO’s simulation model is used along with a clustering algorithm to form future scenarios, which are then used by an optimization model to find an optimal decision regarding the debt management problem. Results of the evaluations show that our optimization approach is expected to have a lower average annual real cost, but with somewhat higher risk, than a set of static comparison strategies in a simulated future. These evaluation results are based on a risk preference set by ourselves, since the government has not expressed its risk preference quantitatively. We also conclude that financial optimization is applicable on the government debt management problem, although some work remains before the method can be incorporated into the strategic work of the SNDO.

Cette thèse traite du traitement adaptatif spatio-temporel dans le domaine radar. Pour augmenter les performances en détection, cette approche consiste à maximiser le rapport entre la puissance de la cible et celle des interférences, à savoir le bruit thermique et le fouillis. De nombreuses variantes de cet algorithme existent, une d’entre elles est fondée sur une modélisation autorégressive vectorielle des interférences. Sa principale difficulté réside dans l’estimation des matrices autorégressives à partir des données d’entrainement ; ce point constitue l’axe de notre travail de recherche. En particulier, notre contribution porte sur deux aspects. D’une part, dans le cas où l’on suppose que le bruit thermique est négligeable devant le fouillis non gaussien, les matrices autorégressives sont estimées en utilisant la méthode du point fixe. Ainsi, l’algorithme est robuste à la distribution non gaussienne du fouillis.D’autre part, nous proposons une nouvelle modélisation des interférences différenciant le bruit thermique et le fouillis : le fouillis est considéré comme un processus autorégressif vectoriel, gaussien et perturbé par le bruit blanc thermique. Ainsi, de nouvelles techniques d'estimation des matrices autorégressives sont proposées. La première est une estimation aveugle par bloc reposant sur la technique à erreurs dans les variables. Ainsi, l’estimation des matrices autorégressives reste robuste pour un rapport faible entre la puissance de la cible et celle du fouillis (< 5 dB). Ensuite, des méthodes récursives ont été développées. Elles sont fondées sur des approches du type Kalman : filtrage de Kalman étendu et filtrage par sigma point (UKF et CDKF), ainsi que sur le filtre H∞.Une étude comparative sur des données synthétiques et réelles, avec un fouillis gaussien ou non gaussien, est menée pour révéler la pertinence des différents estimateurs en terme de probabilité de détection. / This dissertation deals with space-time adaptive processing in the radar’s field. To improve the detection’s performances, this approach consists in maximizing the ratio between the target’s power and the interference’s one, i.e. the thermal noise and the clutter. Several variants of its algorithm exist, one of them is based on multichannel autoregressive modelling of interferences. Its main problem lies in the estimation of autoregressive matrices with training data and guides our research’s work. Especially, our contribution is twofold.On the one hand, when thermal noise is considered negligible, autoregressive matrices are estimated with fixed point method. Thus, the algorithm is robust against non-gaussian clutter.On the other hand, a new modelling of interferences is proposed. The clutter and thermal noise are separated : the clutter is considered as a multichannel autoregressive process which is Gaussian and disturbed by the white thermal noise. Thus, new estimation’s algorithms are developed. The first one is a blind estimation based on errors in variable methods. Then, recursive approaches are proposed and used extension of Kalman filter : the extended Kalman filter and the Sigma Point Kalman filter (UKF and CDKF), and the H∞ filter. A comparative study on synthetic and real data with Gausian and non Gaussian clutter is carried out to show the relevance of the different algorithms about detection’s probability.

Ce mémoire porte sur l’étude des modèles autorégressifs avec variables exogènes et sélection d’indices. La littérature classique regorge de textes concernant la sélection d’indices dans les modèles autorégressifs. Ces modèles sont particulièrement utiles pour des données macroéconomiques mesurées sur des périodes de temps modérées à longues. Effectivement, la lourde paramétrisation des modèles complets peut souvent être allégée en utilisant la sélection d’indices aboutissant ainsi à des modèles plus parcimonieux. Les modèles à variables exogènes sont très intéressants dans le contexte où il est connu que les variables à l’étude sont affectées par d’autres variables, jouant le rôle de variables explicatives, que l’analyste ne veut pas forcément modéliser. Ce mémoire se propose donc d’étudier les modèles autorégressifs vectoriels avec variables exogènes et sélection d’indices. Ces modèles ont été explorés, entre autres, par Lütkepohl (2005), qui se contente cependant d’esquisser les développements mathématiques. Nous concentrons notre étude sur l’inférence statistique sous des conditions précises, la modélisation ainsi que les prévisions. Notre objectif est de comparer les modèles avec sélection d’indices aux modèles autorégressifs avec variables exogènes complets classiques. Nous désirons déterminer si l’utilisation des modèles avec sélection d’indices est marquée par une différence favorable au niveau du biais et de l’écart-type des estimateurs ainsi qu’au niveau des prévisions de valeurs futures. Nous souhaitons également comparer l’efficacité de la sélection d’indices dans les modèles autorégressifs ayant des variables exogènes à celle dans les modèles autorégressifs. Il est à noter qu’une motivation première dans ce mémoire est l’estimation dans les modèles autorégressifs avec variables exogènes à sous-ensemble d’indices. Dans le premier chapitre, nous présentons les séries temporelles ainsi que les diverses notions qui y sont rattachées. De plus, nous présentons les modèles linéaires classiques multivariés, les modèles à variables exogènes puis des modèles avec sélection d’indices. Dans le deuxième chapitre, nous exposons le cadre théorique de l’estimation des moindres carrés dans les modèles autorégressifs à sous-ensemble d’indices ainsi que le comportement asymptotique de l’estimateur. Ensuite, nous développons la théorie pour l’estimation des moindres carrés (LS) ainsi que la loi asymptotique des estimateurs pour les modèles autorégressifs avec sélection d’indices (SVAR) puis nous faisons de même pour les modèles autorégressifs avec variables exogènes et tenant compte de la sélection des indices (SVARX). Spécifiquement, nous établissons la convergence ainsi que la distribution asymptotique pour l’estimateur des moindres carrés d’un processus autorégressif vectoriel à sous-ensemble d’indices et avec variables exogènes. Dans le troisième chapitre, nous appliquons la théorie spécifiée précédemment lors de simulations de Monte Carlo. Nous évaluons de manière empirique les biais et les écarts-types des coefficients trouvés lors de l’estimation ainsi que la proportion de fois que le modèle ajusté correspond au vrai modèle pour différents critères de sélection, tailles échantillonnales et processus générateurs des données. Dans le quatrième chapitre, nous appliquons la théorie élaborée aux chapitres 1 et 2 à un vrai jeu de données provenant du système canadien d’information socioéconomique (CANSIM), constitué de la production mensuelle de fromage mozzarella, cheddar et ricotta au Canada, expliquée par les prix mensuels du lait de bovin non transformé dans les provinces de Québec, d’Ontario et de la Colombie-Britannique pour la période allant de janvier 2003 à juillet 2021. Nous ajustons ces données à un modèle autorégressif avec variables exogènes complet puis à un modèle autorégressif avec variables exogènes et sélection d’indices. Nous comparons ensuite les résultats obtenus avec le modèle complet à ceux obtenus avec le modèle restreint. Mots-clés : Processus autorégressif à sous-ensemble d’indices, variables exogènes, esti mation des moindres carrés, sélection de modèle, séries chronologiques multivariées, processus stochastiques, séries chronologiques. / This Master’s Thesis focuses on the study of subset autoregressive models with exoge nous variables. Many texts from the classical literature deal with the selection of indexes in autoregressive models. These models are particularly useful for macroeconomic data measured over moderate to long periods of time. Indeed, the heavy parameterization of full models can often be simplified by using the selection of indexes, thus resulting in more parsimonious models. Models with exogenous variables are very interesting in the context where it is known that the variables under study are affected by other variables, playing the role of explanatory variables, not necessarily modeled by the analyst. This Master’s Thesis therefore proposes to study vector subset autoregressive models with exogenous variables. These models have been explored, among others, by Lütkepohl (2005), who merely sketches proofs of the statistical properties. We focus our study on statistical inference under precise conditions, modeling and forecasting for these models. Our goal is to compare restricted models to full classical autoregressive models with exogenous variables. We want to determine whether the use of restricted models is marked by a favorable difference in the bias and standard deviation properties of the estimators as well as in forecasting future values. We also compare the efficiency of index selection in autoregressive models with exogenous variables to that in autoregressive models. It should be noted that a primary motivation in this Master’s Thesis is the estimation in subset autoregressive models with exogenous variables. In the first chapter, we present time series as well as the various concepts which are attached to them. In addition, we present the classical multivariate linear models, models with exogenous variables and then we present subset models. In the second chapter, we present the theoretical framework for least squares estimation in subset autoregressive models as well as the asymptotic behavior of the estimator. Then, we develop the theory for the estimation of least squares (LS) as well as the asymptotic distribution of the estimators for the subset autoregressive models (SVAR), and we do the same for the subset autoregressive models with exogenous variables (SVARX). Specifically, we establish the convergence as well as the asymptotic distribution for the least squares estimator of a subset autoregressive process with exogenous variables. In the third chapter, we apply the theory specified above in Monte Carlo simulations. We evaluate empirically the biases and the standard deviations of the coefficients found during the estimation as well as the proportion of times that the adjusted model matches the true model for different selection criteria, sample size and data generating processes. In the fourth chapter, we apply the theory developed in chapters 1 and 2 to a real dataset from the Canadian Socio-Economic Information System (CANSIM) consisting of the monthly production of mozzarella, cheddar and ricotta cheese in Canada, explained by the monthly prices of unprocessed bovine milk in the provinces of Quebec, Ontario and British Columbia from January 2003 to July 2021. We fit these data with a full autoregressive model with exogenous variables and then to a subset autoregressive model with exogenous variables. Afterwards, we compare the results obtained with the complete model to those obtained with the subset model. Keywords : Subset autoregressive process, exogenous variables, least squares estimation, model selection, multivariate time series, stochastic process, time series.

Cette étude théorique est pensée en lien étroit avec un champ d'application : il s'agit de modéliser la croissance d'une population de cellules qui se divisent selon un taux de division inconnu, fonction d’une variable dite structurante – l’âge et la taille des cellules étant les deux exemples paradigmatiques étudiés. Le champ mathématique afférent se situe à l'interface de la statistique des processus, de l’estimation non-paramétrique et de l’analyse des équations aux dérivées partielles. Les trois objectifs de ce travail sont les suivants : reconstruire le taux de division (fonction de l’âge ou de la taille) pour différents schémas d’observation (en temps généalogique ou en temps continu) ; étudier la transmission d'un trait biologique général d'une cellule à une autre et étudier le trait d’une cellule typique ; comparer la croissance de différentes populations de cellules à travers le paramètre de Malthus (après introduction de variabilité dans le taux de croissance par exemple). / This work is concerned with growth-fragmentation models, implemented for investigating the growth of a population of cells which divide according to an unknown splitting rate, depending on a structuring variable – age and size being the two paradigmatic examples. The mathematical framework includes statistics of processes, nonparametric estimations and analysis of partial differential equations. The three objectives of this work are the following : get a nonparametric estimate of the division rate (as a function of age or size) for different observation schemes (genealogical or continuous) ; to study the transmission of a biological feature from one cell to an other and study the feature of one typical cell ; to compare different populations of cells through their Malthus parameter, which governs the global growth (when introducing variability in the growth rate among cells for instance).

The financial crisis of 2007/8 has triggered a profound debate about public budget finance sustainability, ever-increasing government expenditures and the efficiency of fiscal policy measures. Given this context, the following dissertation provides four contributions that analyze the long-run growth of government spending throughout economic development, discuss potential effects of fiscal policy measures on output, and provide new insights into the assessment of debt sustainability for a variety of industrialized countries. Since the breakout of the European debt crisis in 2009/2010, there has been a revival of interest in the long-term growth of government expenditures. In this context, the relationship between the size of the public sector and economic growth - often referred to as Wagner's law - has been in the focus of numerous studies, especially with regard to public policy and fiscal sustainability. Using historical data from the mid-19th century, the first chapter analyzes the validity of Wagner's law for five industrialized European countries and links the discussion to different stages of economic development. In line with Wagner's hypothesis, our findings show that the relationship between public spending and economic growth has weakened at an advanced stage of development. Furthermore, all countries under review support the notion that Wagner's law may have lost its economic relevance in recent decades. As a consequence of the 2007/8 financial crisis, there has been an increasing theoretical and empirical debate about the impact of fiscal policy measures on output. Accordingly, the Structural Vector Autoregression (SVAR) approach to estimating the fiscal multipliers developed by Blanchard and Perotti (2002) has been applied widely in the literature in recent years. In the second chapter, we point out that the fiscal multipliers derived from this approach include the predicted future path of the policy instruments as well as their dynamic interaction. We analyze a data set from the US and document that these interactions are economically and statistically significant. In a counterfactual simulation, we report fiscal multipliers that abstract from these dynamic responses. Furthermore, we use our estimates to analyze the recent fiscal stimulus of the American Recovery and Reinvestment Act (ARRA). The third chapter contributes to the existing empirical literature on fiscal multipliers by applying a five-variable SVAR approach to a uniform data set for Belgium, France, Germany, and the United Kingdom. Besides studying the effects of expenditure and tax increases on output, we additionally analyze their dynamic effects on inflation and interest rates as well as the dynamic interaction of both policy instruments. By conducting counterfactual simulations, which abstract from the dynamic response of key macroeconomic variables to the initial fiscal shocks, we study the importance of these channels for the transmission of fiscal policy on output. Overall, the results demonstrate that the effects of fiscal shocks are limited and rather different across countries. Further, it is shown that the inflation and interest rate channel are insignificant for the transmission of fiscal policy. In the field of public finances, governmental budgetary policies are among the most controversial and disputed areas of political and scientific controversy. The sustainability of public debt is often analyzed by testing stationarity conditions of government's budget deficits. The fourth chapter shows that this test can be implemented more effectively by means of an asymmetric unit root test. We argue that this approach increases the power of the test and reduces the likelihood of drawing false inferences. We illustrate this in an application to 14 countries of the European Monetary Union as well as in a Monte Carlo simulation. Distinguishing between positive and negative changes in deficits, we find consistency with the intertemporal budget constraint for more countries, i.e. lower persistence of positive changes in some countries, compared to the earlier literature.

Le contrôle explicite de la convergence des sommes convenablement normalisées de variables aléatoires, ainsi que l'étude du principe de déviations modérées associé à ces sommes constituent les thèmes centraux de cette thèse. Nous étudions principalement deux types de processus. Premièrement, nous nous intéressons aux processus indexés par un arbre binaire, aléatoire ou non. Ces processus ont été introduits dans la littérature afin d'étudier le mécanisme de la division cellulaire. Au chapitre 2, nous étudions les chaînes de Markov bifurcantes. Ces chaînes peuvent être vues comme une adaptation des chaînes de Markov "usuelles'' dans le cas où l'ensemble des indices à une structure binaire. Sous des hypothèses d'ergodicité géométrique uniforme et non-uniforme d'une chaîne de Markov induite, nous fournissons des inégalités de déviations et un principe de déviations modérées pour les chaînes de Markov bifurcantes. Au chapitre 3, nous nous intéressons aux processus bifurcants autorégressifs d'ordre p (). Ces processus sont une adaptation des processus autorégressifs linéaires d'ordre p dans le cas où l'ensemble des indices à une structure binaire. Nous donnons des inégalités de déviations, ainsi qu'un principe de déviations modérées pour les estimateurs des moindres carrés des paramètres "d'autorégression'' de ce modèle. Au chapitre 4, nous traitons des inégalités de déviations pour des chaînes de Markov bifurcantes sur un arbre de Galton-Watson. Ces chaînes sont une généralisation de la notion de chaînes de Markov bifurcantes au cas où l'ensemble des indices est un arbre de Galton-Watson binaire. Elles permettent dans le cas de la division cellulaire de prendre en compte la mort des cellules. Les hypothèses principales que nous faisons dans ce chapitre sont : l'ergodicité géométrique uniforme d'une chaîne de Markov induite et la non-extinction du processus de Galton-Watson associé. Au chapitre 5, nous nous intéressons aux modèles autorégressifs linéaires d'ordre 1 ayant des résidus corrélés. Plus particulièrement, nous nous concentrons sur la statistique de Durbin-Watson. La statistique de Durbin-Watson est à la base des tests de Durbin-Watson, qui permettent de détecter l'autocorrélation résiduelle dans des modèles autorégressifs d'ordre 1. Nous fournissons un principe de déviations modérées pour cette statistique. Les preuves du principe de déviations modérées des chapitres 2, 3 et 4 reposent essentiellement sur le principe de déviations modérées des martingales. Les inégalités de déviations sont établies principalement grâce à l'inégalité d'Azuma-Bennet-Hoeffding et l'utilisation de la structure binaire des processus. Le chapitre 5 est né de l'importance qu'a l'ergodicité explicite des chaînes de Markov au chapitre 3. L'ergodicité géométrique explicite des processus de Markov à temps discret et continu ayant été très bien étudiée dans la littérature, nous nous sommes penchés sur l'ergodicité sous-exponentielle des processus de Markov à temps continu. Nous fournissons alors des taux explicites pour la convergence sous exponentielle d'un processus de Markov à temps continu vers sa mesure de probabilité d'équilibre. Les hypothèses principales que nous utilisons sont : l'existence d'une fonction de Lyapunov et d'une condition de minoration. Les preuves reposent en grande partie sur la construction du couplage et le contrôle explicite de la queue du temps de couplage. / The explicit control of the convergence of properly normalized sums of random variables, as well as the study of moderate deviation principle associated with these sums constitute the main subjects of this thesis. We mostly study two sort of processes. First, we are interested in processes labelled by binary tree, random or not. These processes have been introduced in the literature in order to study mechanism of the cell division. In Chapter 2, we study bifurcating Markov chains. These chains may be seen as an adaptation of "usual'' Markov chains in case the index set has a binary structure. Under uniform and non-uniform geometric ergodicity assumptions of an embedded Markov chain, we provide deviation inequalities and a moderate deviation principle for the bifurcating Markov chains. In chapter 3, we are interested in p-order bifurcating autoregressive processes (). These processes are an adaptation of $p$-order linear autoregressive processes in case the index set has a binary structure. We provide deviation inequalities, as well as an moderate deviation principle for the least squares estimators of autoregressive parameters of this model. In Chapter 4, we dealt with deviation deviation inequalities for bifurcating Markov chains on Galton-Watson tree. These chains are a generalization of the notion of bifurcating Markov chains in case the index set is a binary Galton-Watson tree. They allow, in case of cell division, to take into account cell's death. The main hypothesis that we do in this chapter are : uniform geometric ergodicity of an embedded Markov chain and the non-extinction of the associated Galton-Watson process. In Chapter 5, we are interested in first-order linear autoregressive models with correlated errors. More specifically, we focus on the Durbin-Watson statistic. The Durbin-Watson statistic is at the base of Durbin-Watson tests, which allow to detect serial correlation in the first-order autoregressive models. We provide a moderate deviation principle for this statistic. The proofs of moderate deviation principle of Chapter 2, 3 and 4 are essentially based on moderate deviation for martingales. To establish deviation inequalities, we use most the Azuma-Bennet-Hoeffding inequality and the binary structure of processes. Chapter 6 was born from the importance that explicit ergodicity of Markov chains has in Chapter 2. Since explicit geometric ergodicity of discrete and continuous time Markov processes has been well studied in the literature, we focused on the sub-exponential ergodicity of continuous time Markov Processes. We thus provide explicit rates for the sub-exponential convergence of a continuous time Markov process to its stationary distribution. The main hypothesis that we use are : existence of a Lyapunov fonction and of a minorization condition. The proofs are largely based on the coupling construction and the explicit control of the tail of the coupling time.