Universiẗat, Diss., 2003--Dortmund.

Les hedge funds ont fait leur place dans le paysage financier. Ils se sont fermement imposés au cours de cette décennie. La perspective de rendements décorrélés a trouvé écho auprès des investisseurs, secoués après la crise de la bulle internet et partis à la recherche de rendements nouveaux. Afin de répondre à leur objectif, ils s’exposent à l’ensemble des marchés du globe, mais prennent part également à la vie des entreprises. Ils ont recours à une large palette d’instruments financiers. Les sources de risque sont donc hétérogènes, multiples et parfois interconnectées. Ces risques sont par ailleurs amplifiés du levier. Ainsi dans une situation normale, les hedge funds ont des performances supérieures, puisqu’ils exhibent des rendements bien plus attrayants, que ceux des classes d’actifs traditionnelles. Cependant, les hedge fund sont soumis à des risques de pertes extrêmes lorsque des chocs défavorables se produisent sur les marchés. Il est donc nécessaire de rendre compte de manière plus adéquate du risque des hedge funds. A ce titre, la Value at Risk est une alternative intéressante, lorsque le modèle de volatilité est plus sophistiqué que la mesure standard de la volatilité et le quantile retenu pour son estimation dépasse le cadre de la loi normale. L’analyse dynamique des hedge funds met en évidence l’existence d’un régime extrême vers lequel tendent les hedge funds dans le cas d’un retournement de marché. / Hedge funds are getting more and more importance. Fuelled by the prospect of returns disconnected from global markets, a wide range of investors have sought exposure to hedge funds, especially after the losses caused by the dot com bubble. They invest in a wide range of markets as well as in companies. The underlying risks are heterogeneous, varied and sometimes interconnected. Furthermore, those risks are magnified by leverage hedge funds undertake. When markets are normal, hedge funds are able to generate returns more attractive than those provided by traditional assets. However, they exhibit an extreme losses risk when markets go suddenly down. Thus, it is important to have an idea of those risks and think about a more accurate measure of hedge fund risks. We thus take into account Value at Risk for which volatility is evaluated in a better manner and quantile retained is different from the normal law. The dynamic analysis of hedge funds suggest that their returns are exposed to an extreme regime when markets go down.

Cette thèse présente une contribution à l’analyse de l’hétérogénéité des comportements sur les marchés boursiers. Elle propose, d’une part, une revue de la littérature des modèles d’hétérogénéité qui permettent de reproduire des faits observés sur les marchés réels tels que la volatilité excessive des cours, le volume important des transactions, les volatilités groupées, la queue de distribution épaisse et le retour à la moyenne, ce qui contredit l’efficience des marchés et l’hypothèse d’un agent représentatif. Ces modèles permettent également d’expliquer l’apparition des bulles et les comportements parfois chaotiques des prix. La prise en compte explicite de l’hétérogénéité, dans la modélisation, conduit donc à des représentations plus adéquates avec la réalité. D’autre part, nous proposons des travaux empiriques sur l’hétérogénéité des comportements des investisseurs sur le marché des actions français, au niveau des titres individuels et suivant différentes fréquences d’observations. La première estimation considère un modèle de sélection évolutionnaire des stratégies. Nous avons aussi constaté une certaine persistance de l’écart entre les prix et les valeurs fondamentales. Nous avons aussi remarqué la désorientation des investisseurs pendant les périodes de crise avec un changement brutal entre les stratégies et ce souvent pour la majorité des investisseurs. Nous avons conclu pendant ces périodes qu’il existe des phénomènes d’imitation liés probablement au manque d’information et au climat d’incertitude. Ce résultat s’accorde avec ce qu’on observe sur le marché pendant la formation et à l’éclatement d’une bulle. Ces résultats ont été globalement confirmés par une deuxième estimation d’un modèle de type LSTAR-GARCH qui tient compte explicitement de la variance conditionnelle et pose des hypothèses différentes du premier. / This PhD dissertation presents a contribution to the analysis of the behaviour’s heterogeneity on the stock markets. It proposes, firstly, a review of the literature of heterogeneous agents’ models which allow reproducing stylised facts observed in the real markets such as an excessive volatility of prices, an important transaction’s volume, grouped volatilities, a fat tail distribution and a mean return, which contradicts the markets efficiency and the assumption of a representative agent. These models also, allow explaining the bubbles emergence and prices behaviours sometimes chaotic. The explicit heterogeneity hypothesis, in modelling, leads representations more adequate with reality. In addition, we propose empirical works on investor’s behaviours heterogeneity in the French stock market, for individuals stocks and following different observation frequencies. The first estimation considers a model of evolutionary strategies selection. We noted the persistence of the difference between prices and fundamental values. We also noticed the confusion of investors in crisis periods with a brutal change between strategies and this often for the majority of investors. We concluded for these periods that there are imitation phenomena related on lack of information and uncertainty climate. This result agrees with real market during bubble formation and bursting of a bubble. These results are generally confirmed by the second estimation of the LSTAR-GARCH model which explicitly considers a conditional variance and supposes different assumptions from the first.

Field of examination is theoretical and empirical review of dynamic CAPM models that assume non constant volatility and correlation. In other words time evolution is considered in estimation process. As theoretical basement is recommended to be R. Engle's (Dynamic Conditional Beta) research and other sources.

L’objet de cette thèse est d’étudier la dynamique des prix des matières premières, à travers l’étude du degré d’efficience de ces marchés et la dynamique de la prime de risque. A cette fin, ce travail de recherche a été autour de trois principaux chapitres : le premier étant d’ordre théorique tandis que les deux autres chapitres proposent deux essais empiriques. En particulier, le premier chapitre dresse le cadre conceptuel de cette étude, définit les concepts, rappelle les enjeux des stratégies de d’investissement et de diversification sur les marchés des matières premières. Il discute également l’état de la littérature relatif à nos questions de recherche. Le second chapitre applique des tests économétriques paramétriques et non paramétriques pour tester l’hypothèse d’efficience informationnelle à court et à long terme et montre que le degré d’efficience varie selon l’horizon temporel, la région et la commodité.En outre, il propose des simulations de portefeuilles bivariés pour illustrer plus concrètement les gains de diversification et repérer les stratégies d’investissement optimales. Dans le troisième chapitre, le modèle DCC-GARCH (1,1) est estimé pour étudier la dynamique de la prime de risque et expliquer ainsi les sources d’inefficience des marchés de matières premières. Nos résultats ne rejettent pas l’hypothèse de variabilité de prime de risque dans le temps appuyant l’alternance des marchés énergétiques entre l’inefficience à court terme et l’efficience à long terme. / This thesis aims at studying the dynamics of energy price through the investigation of their efficiency degree and the dynamics of risk premium.To this end, this study has been structured into three chapters : The first one is theoretical while the two others are empirical. In particular, the first chapter develops the conceptual framework for this study, defines the concepts, and recalls the issues related to investment strategies and diversification opportunities on energy markets. It also discusses the related literature review. The second chapter focuses on the informational efficiency hypothesis for commodity markets in the short and long terms using several parametric and non-parametric tests. It shows that the efficiency degree varies with commodity, region and temporal horizon. Further, it carries out bivariate portfolio simulations in order to illustrate diversification opportunities and identify optimal investment strategies. In the third chapter, we look at the dynamics of risk premium in order to explain the inefficient character of commodity markets using a DCC-GARCH (1,1) model. Our findings do not reject the hypothesis of time-varying risk premium, which helps to better understand the fact that commodity markets alternate between inefficiency in the short term and efficiency in the long term.

This master thesis deals with extension of the univariate GARCH model to multivari- ate models. We present individual models and deal with methods of their estimation. Then we describe some statistical tests for diagnosting the models. We have programmed in the statistical software R one of them - the Ling-Li test. Afterwards we apply selected models to real data of stock market index S&P 500, stock market index Russell 2000 and stocks of crude oil. For the GO-GARCH model, we compare all available estimation methods and show their differences. Then we compare the results of all models with each other and also with univariate models in terms of estimates of conditional variances, estimates of conditional correlations and also in terms of computational complexity. 1

Processos lineares não capturam a estrutura dos dados em finanças. Há uma variedade muito grande de modelos não lineares disponíveis na literatura. A classe de modelos ARCH (Autoregressive Conditional Heterokedastic) foi introduzida por Engle (1982) com o objetivo de estimar a variância da inflação. A idéia nesta classe é que os retornos sejam não correlacionados serialmente, mas a volatilidade (variância condicional) dependa de retornos passados. A classe de modelos GARCH (Generalized Autoregressive Conditional Heterokedastic) sugerida por Bollerslev (1986, 1987, 1988) pode ser usada para descrever a volatilidade com menos parâmetros que um modelo ARCH. Modelos da classe GARCH são processos estocásticos não lineares, suas distribuições tem cauda pesada com variância condicional dependente do tempo e modelam agrupamento de volatilidade. Apesar da razoável descrição, a forma como os modelos acima foram construídos apresentaram algumas limitações no que se refere ao peso das caudas em suas distribuições não condicionais. Muitos estudos em dados financeiros apontam para caudas com peso considerável. Modelos R-GARCH (Randomized Generalized Autoregressive Conditional Heterokedastic) foram propostos por Nowicka (1998) e incluem os modelos ARCH e GARCH possibilitando o uso de inovações estáveis além da conhecida distribuição normal. Estas permitem captar melhor a propriedade de cauda pesada. Como a função de autocovariância não existe para tais processos introduz-se novas medida de dependência. Métodos de estimação e análises empíricas da classe R-GARCH, assim como de suas medidas de dependência não estão disponíveis na literatura e são o foco deste trabalho. / Linear processes do not capture the structure of financial data. There is a large variety of nonlinear models available in literature. The class of ARCH models (Autoregressive Conditional Heterokedastic) was introduced by Engle (1982) in order to estimate inflation\'s variance. The idea is that, in this class, returns are serially uncorrelated, but the volatility (conditional variance) depends on past returns. The class of GARCH models (Generalized Autoregressive Conditional Heterokedastic) suggested by Bollerslev (1986, 1987, 1988) can be used to describe the volatility with less parameters than ARCH-type models. GARCH-type models are nonlinear stochastic processes, their distribution are heavy-tailed with time-dependent conditional variance model and they model clustering of volatility. Despite the reasonable description, the way that GARCH models are built imposes limits on the heaviness of the tails of their unconditional distribution. Many studies in financial data point to considerable heaviness of the tails. The class of Randomized Generalized Autoregressive Conditional Heterokedastic (R-GARCH) were proposed by Nowicka (1998) and include the ARCH and GARCH models allowing the use of stable innovations in place of normal distribution. This distribution allows to capture the heaviness tail property. As the autocovariance function does not exist for these processes a new measure of dependence was introduced. Estimation methods and empirical analysis of R-GARCH class, as well as their measures of dependence are not available in literature and are the focus of this work.

Processos lineares não capturam a estrutura dos dados em finanças. Há uma variedade muito grande de modelos não lineares disponíveis na literatura. A classe de modelos ARCH (Autoregressive Conditional Heterokedastic) foi introduzida por Engle (1982) com o objetivo de estimar a variância da inflação. A idéia nesta classe é que os retornos sejam não correlacionados serialmente, mas a volatilidade (variância condicional) dependa de retornos passados. A classe de modelos GARCH (Generalized Autoregressive Conditional Heterokedastic) sugerida por Bollerslev (1986, 1987, 1988) pode ser usada para descrever a volatilidade com menos parâmetros que um modelo ARCH. Modelos da classe GARCH são processos estocásticos não lineares, suas distribuições tem cauda pesada com variância condicional dependente do tempo e modelam agrupamento de volatilidade. Apesar da razoável descrição, a forma como os modelos acima foram construídos apresentaram algumas limitações no que se refere ao peso das caudas em suas distribuições não condicionais. Muitos estudos em dados financeiros apontam para caudas com peso considerável. Modelos R-GARCH (Randomized Generalized Autoregressive Conditional Heterokedastic) foram propostos por Nowicka (1998) e incluem os modelos ARCH e GARCH possibilitando o uso de inovações estáveis além da conhecida distribuição normal. Estas permitem captar melhor a propriedade de cauda pesada. Como a função de autocovariância não existe para tais processos introduz-se novas medida de dependência. Métodos de estimação e análises empíricas da classe R-GARCH, assim como de suas medidas de dependência não estão disponíveis na literatura e são o foco deste trabalho. / Linear processes do not capture the structure of financial data. There is a large variety of nonlinear models available in literature. The class of ARCH models (Autoregressive Conditional Heterokedastic) was introduced by Engle (1982) in order to estimate inflation\'s variance. The idea is that, in this class, returns are serially uncorrelated, but the volatility (conditional variance) depends on past returns. The class of GARCH models (Generalized Autoregressive Conditional Heterokedastic) suggested by Bollerslev (1986, 1987, 1988) can be used to describe the volatility with less parameters than ARCH-type models. GARCH-type models are nonlinear stochastic processes, their distribution are heavy-tailed with time-dependent conditional variance model and they model clustering of volatility. Despite the reasonable description, the way that GARCH models are built imposes limits on the heaviness of the tails of their unconditional distribution. Many studies in financial data point to considerable heaviness of the tails. The class of Randomized Generalized Autoregressive Conditional Heterokedastic (R-GARCH) were proposed by Nowicka (1998) and include the ARCH and GARCH models allowing the use of stable innovations in place of normal distribution. This distribution allows to capture the heaviness tail property. As the autocovariance function does not exist for these processes a new measure of dependence was introduced. Estimation methods and empirical analysis of R-GARCH class, as well as their measures of dependence are not available in literature and are the focus of this work.

Volatility is an important variable in financial markets, risk management and making investment decisions. Different volatility models are beneficial tools to use when predicting future volatility. The purpose of this study is to compare the accuracy of various volatility models, including ARCH, GARCH and extensions of the GARCH framework. The study applies these volatility models to the Swedish and American Fixed Income Market for government bonds. The performance of these models is based on out-of-sample forecasting using different loss functions such as RMSE, MAE and MSE, specifically investigating their ability to forecast future volatility. Daily volatility forecasts from daily bid prices from Swedish and American 2, 5- and 10-year governments bonds will be compared against realized volatility which will act as the proxy for volatility. The result show US government bonds, excluding the US 2 YTM, did not show any significant negative volatility, volatility asymmetry or leverage effects. In overall, the ARCH and GARCH models outperformed E-GARCH and GJR-GARCH except the US 2-year YTM showing negative volatility, asymmetry, and leverage effects and the GJR-GARCH model outperforming the ARCH and GARCH models. / Volatilitet är en viktig variabel på finansmarknaden när det kommer till både riskhantering samt investeringsbeslut. Olika volatilitets modeller är fördelaktiga verktyg när det kommer till att göra prognoser av framtida volatilitet. Syftet med denna studie är att jämföra det olika volatilitetsmodellerna ARCH, GARCH och förlängningar av GARCH-ramverket för att ta reda på vilken av modellerna är den bästa att prognosera framtida volatilitet. Studien kommer tillämpa dessa modeller på den svenska och amerikanska marknaden för statsskuldväxlar. Prestandan för modellerna kommer baseras på out-of-sample prognoser med hjälp av det olika förlustfunktionerna RMSE, MAE och MSE. Förlustfunktionernas används endast till att undersöka deras förmåga till att prognostisera framtida volatilitet. Dagliga volatilitetsprognoser baseras på dagliga budpriser för amerikanska och svenska statsobligationer med 2, 5 och 10 års löptid. Dessa kommer jämföras med verklig volatilitet som agerar som Proxy för volatiliteten. Resultatet tyder på att amerikanska statsobligationer förutom den tvååriga, inte visar signifikant negativ volatilitet, asymmetri i volatilitet samt hävstångseffekt. De tvååriga amerikanska statsobligationerna visar bevis för negativ volatilitet, hävstångseffekt samt asymmetri i volatiliteten. ARCH och GARCH modellerna presterade övergripande sett bäst för både svenska och amerikanska statsobligationer förutom den tvååriga där GJR-GARCH modellen presterade bäst.

Syftet med studien är att undersöka bitcoins kapacitet som hedge gentemot den svenska aktiemarknaden. För att identifiera om korrelation existerar mellan avkastningen i bitcoin och SIX30RX (OMXS30 med utdelning) och således besvara forskningsfrågan studeras associationen. Tidsperioden som studeras är 2012-01-02 till 2016-10-21. Associationen undersöks med hjälp av regressionsmodeller. Resultatet visar att bitcoin inte är korrelerat med avkastningen för SIX30RX under den studerade tidsperioden. Bitcoin kan således klassificeras som en hedge gentemot den svenska aktiemarknaden. / This paper examines bitcoin’s capacity as a hedge towards the Swedish stock market. To identify if correlation exists between the returns of bitcoin and SIX30RX (OMXS 30 including dividends) and thus respond to the research question the association is investigated. The time period considered is 2012-01-02 to 2016-10-21. Association is analysed using regression models. The results demonstrate that bitcoin is uncorrelated with the return for SIX30RX during this time period. Therefore, bitcoin can be classified as a hedge against the Swedish stock market.