Due to the rapid development of technology, quality control charts have attracted more attention from manufacturing industries in order to monitor quality characteristics of interest more effectively. Among many control charts, my research work has focused on the multivariate exponentially weighted moving average (MEWMA) and the univariate exponentially weighted moving average (EWMA) control charts by using the Markov chain method. The performance of the chart is measured by the optimal average run length (ARL). My Ph.D. thesis is composed of the following three contributions. My first research work is about differential smoothing. The MEWMA control chart proposed by Lowry et al. (1992) has become one of the most widely used charts to monitor multivariate processes. Its simplicity, combined with its high sensitivity to small and moderate process mean jumps, is at the core of its appeal. Lowry et al. (1992) advocated equal smoothing of each quality variable unless there is an a priori reason to weigh quality characteristics differently. However, one may have situations where differential smoothing may be justified. For instance: (a) departures in process mean may be different across quality variables, (b) some variables may evolve over time at a much different pace than other variables, and (c) the level of correlation between variables could vary substantially. For these reasons, I focus on and assess the performance of the differentially smoothed MEWMA chart. The case of two quality variables (BEWMA) is discussed in detail. A bivariate Markov chain method that uses conditional distributions is developed for average run length (ARL) calculations. The proposed chart is shown to perform at least as well as Lowry et al. (1992)'s chart, and noticeably better in most other mean jump directions. Comparisons with the recently introduced double-smoothed BEWMA chart and the univariate charts for the independent case show that the proposed differentially smoothed BEWMA chart has superior performance. My second research work is about monitoring skewed multivariate processes. Recently, Xie et al. (2011) studied monitoring bivariate exponential quality measurements using the standard MEWMA chart originally developed to monitor multivariate normal quality data. The focus of my work is on situations where, marginally, the quality measurements may follow not only exponential distributions but also other skewed distributions such as Gamma or Weibull, in any combination. The joint distribution is specified using the Gumbel copula function thus allowing for varying degrees of correlation among the quality measurements. In addition to the standard MEWMA chart, charts based on the largest or smallest of the measurements and on the joint cumulative distribution function or the joint survivor function, are studied in detail. The focus is on the case of two quality measurements, i.e., on skewed bivariate processes. The proposed charts avoid an undesirable feature encountered by Xie et al. (2011) for the standard MEWMA chart where in some cases the off-target average run length turns out to be larger than the on-target one. Using the optimal average run length, our extensive numerical results show that the proposed methods provide an overall good detection performance in most directions. Simulations were performed to obtain the optimal ARL results but the Markov chain method using the empirical CDF of the statistics involved verified the accuracy of the ARL results. In addition, an examination of the effect of correlation on chart performance was undertaken numerically. The methods are easily extendable to more than two variables. Final study is about a new ARL criterion for univariate processes studied in detail in this thesis. The traditional ARL is calculated assuming a given fixed process mean jump and a given time point where the jump occurs, usually taken to be from the very beginning in most chart performance studies. However, Ryu et al. (2010) demonstrated that the assumption of a fixed mean shift might lead to poor performance of control charts when the actual size of the mean shift is significantly different and therefore suggested a new ARL-based performance measure, called expected weighted run length (EWRL), by assuming that the size of the mean shift is not specified but rather it follows a probability distribution. The EWRL becomes the expected value of the weighted ordinary ARL with respect to this distribution. My methods generalize this criterion by allowing the time at which the mean shift occurs to also vary according to a probability distribution. This leads to a joint distribution for the size of the mean shift and the time the shift takes place, then the EWRL is calculated as the weighted expected value with respect to this joint distribution. The benefit of the generalized EWRL is that one can assess the performance of control charts more realistically when the process starts on-target and then the mean shift occurs at some later random time. Moreover, I also propose the effective EWRL, which measures the number of additional process runs that on average are needed to detect a jump in the mean after it happens. I evaluate several well-known univariate control charts based on their EWRL and effective EWRL performance. The numerical results show that the choice of control chart depends on the additional information on the transition point of the mean shift. The methods can readily be extended to other control charts, including multivariate charts. / Thesis / Doctor of Philosophy (PhD) / Since the introduction of the standard multivariate exponentially weighted moving average (MEWMA) procedure (Lowry et al. 1992), equal smoothing on all quality variables has been conveniently adopted. In this thesis, a bivariate exponentially weighted moving average (BEWMA) control statistic with unequal smooth- ing parameters is introduced with the aim of improving performance over the standard BEWMA chart. Extensive numerical comparisons reveal that the proposed chart enhances the efficiency and flexibility of the control chart in many mean-shift directions. Recently, Xie et al. (2011) proposed a chart for bivariate Exponential data when the quality measures follow Gumbel’s bivariate Exponential distribution (Gumbel 1960). However, when the process means experience a downward shift (D-D shift), the control charts are shown to break down. In other words, we encounter the strange situation where the out-of-control ARL becomes larger than the in-control ARL. To address this issue, we have proposed two methods, the MAX-MIN and CDF methods and applied them to the univariate EWMA chart. Our numerical results show that not only do our proposed methods prevent the undesirable behaviour from happening, but they also offer substantial improvement in the ARL over the approach proposed by Xie et al. (2011) in many mean shifts. Finally, in general, when it comes to designing a control chart, it is assumed that the size of the mean shift is fixed and known. However, Ryu et al. (2010) proposed a new general performance measure, EWRL, by modelling the size of the mean shift with a probability distribution function. We further generalize the measure by introducing a new random variable, T, which is the transition point of the mean shift. Based on that, we propose several ARL-based criteria to measure the chart performance and try them on several univariate control charts.

Many machine learning (ML) techniques rely on probability, random variables, and stochastic modeling. Although statistics pervades this field, there is a large disconnect between the copula modeling and the machine learning communities. Copulas are stochastic models that capture the full dependence structure between random variables and allow flexible modeling of multivariate joint distributions. Elidan was the first to recognize this disconnect, and introduced copula based models to the ML community that demonstrated magnitudes of order better performance than the non copula-based models Elidan [2013]. However, the limitation of these is that they are only applicable for continuous random variables and real world data is often naturally modeled jointly as continuous and discrete. This report details our work in bridging this gap of modeling and analyzing data that is jointly continuous and discrete using copulas. Our first research contribution details modeling of jointly continuous and discrete random variables using the copula framework with Bayesian networks, termed Hybrid Copula Bayesian Networks (HCBN) [Karra and Mili, 2016], a continuation of Elidan’s work on Copula Bayesian Networks Elidan [2010]. In this work, we extend the theorems proved by Neslehov ˇ a [2007] from bivariate ´ to multivariate copulas with discrete and continuous marginal distributions. Using the multivariate copula with discrete and continuous marginal distributions as a theoretical basis, we construct an HCBN that can model all possible permutations of discrete and continuous random variables for parent and child nodes, unlike the popular conditional linear Gaussian network model. Finally, we demonstrate on numerous synthetic datasets and a real life dataset that our HCBN compares favorably, from a modeling and flexibility viewpoint, to other hybrid models including the conditional linear Gaussian and the mixture of truncated exponentials models. Our second research contribution then deals with the analysis side, and discusses how one may use copulas for exploratory data analysis. To this end, we introduce a nonparametric copulabased index for detecting the strength and monotonicity structure of linear and nonlinear statistical dependence between pairs of random variables or stochastic signals. Our index, termed Copula Index for Detecting Dependence and Monotonicity (CIM), satisfies several desirable properties of measures of association, including Renyi’s properties, the data processing inequality (DPI), and ´ consequently self-equitability. Synthetic data simulations reveal that the statistical power of CIM compares favorably to other state-of-the-art measures of association that are proven to satisfy the DPI. Simulation results with real-world data reveal CIM’s unique ability to detect the monotonicity structure among stochastic signals to find interesting dependencies in large datasets. Additionally, simulations show that CIM shows favorable performance to estimators of mutual information when discovering Markov network structure. Our third research contribution deals with how to assess an estimator’s performance, in the scenario where multiple estimates of the strength of association between random variables need to be rank ordered. More specifically, we introduce a new property of estimators of the strength of statistical association, which helps characterize how well an estimator will perform in scenarios where dependencies between continuous and discrete random variables need to be rank ordered. The new property, termed the estimator response curve, is easily computable and provides a marginal distribution agnostic way to assess an estimator’s performance. It overcomes notable drawbacks of current metrics of assessment, including statistical power, bias, and consistency. We utilize the estimator response curve to test various measures of the strength of association that satisfy the data processing inequality (DPI), and show that the CIM estimator’s performance compares favorably to kNN, vME, AP, and HMI estimators of mutual information. The estimators which were identified to be suboptimal, according to the estimator response curve, perform worse than the more optimal estimators when tested with real-world data from four different areas of science, all with varying dimensionalities and sizes. / Ph. D. / Many machine learning (ML) techniques rely on probability, random variables, and stochastic modeling. Although statistics pervades this field, many of the traditional machine learning techniques rely on linear statistical techniques and models. For example, the correlation coefficient, a widely used construct in modern data analysis, is only a measure of linear dependence and cannot fully capture non-linear interactions. In this dissertation, we aim to address some of these gaps, and how they affect machine learning performance, using the mathematical construct of copulas. Our first contribution deals with accurate probabilistic modeling of real-world data, where the underlying data is both continuous and discrete. We show that even though the copula construct has some limitations with respect to discrete data, it is still amenable to modeling large real-world datasets probabilistically. Our second contribution deals with analysis of non-linear datasets. Here, we develop a new measure of statistical association that can handle discrete, continuous, or combinations of such random variables that are related by any general association pattern. We show that our new metric satisfies several desirable properties and compare it’s performance to other measures of statistical association. Our final contribution attempts to provide a framework for understanding how an estimator of statistical association will affect end-to-end machine learning performance. Here, we develop the estimator response curve, and show a new way to characterize the performance of an estimator of statistical association, termed the estimator response curve. We then show that the estimator response curve can help predict how well an estimator performs in algorithms which require statistical associations to be rank ordered.

This dissertation is composed of two main and independents essays, but complementary. In the first one, we discuss the option price under a bayesian perspective. This essay aims to price and analyze the fair price behavior of the call-on-max (bivariate) option considering marginal heteroscedastic models with dependence structure modeled via copulas. Concerning inference, we adopt a Bayesian perspective and computationally intensive methods based on Monte Carlo simulations via Markov Chain (MCMC). A simulation study examines the bias and the root mean squared errors of the posterior means for the parameters. Real stocks prices of Brazilian banks illustrate the approach. For the proposed method is verified the effects of strike and dependence structure on the fair price of the option. The results show that the prices obtained by our heteroscedastic model approach and copulas differ substantially from the prices obtained by the model derived from Black and Scholes. Empirical results are presented to argue the advantages of our strategy. In the second chapter, we consider the GARCH-in-mean models with asymmetric variance specifications to model the volatility of the assets-objects under the risk-neutral dynamics. Moreover, the copula functions model the joint distribution, with the objective of capturing non-linear, linear and tails associations between the assets. We aim to provide a methodology to realize a more realistic pricing option. To illustrate the methodology, we use stocks from two Brazilian companies, where our the modeling offered a proper fitting. Confronting the results obtained with the classic model, which is an extension of the Black and Scholes model, we note that considering constant volatility over time underpricing the options, especially in-the-money options. / Essa dissertação é composta por dois principais ensaios independentes e complementares. No primeiro discutimos a precificação de opções bivariadas sob uma perspectiva bayesiana. Neste ensaio o principal objetivo foi precificar e analizar o preço justo da opção bivariada call-onmax considerando modelos heterocedásticos para as marginais e a modelagem de dependência realizada por funções cópulas. Para a inferência, adotamos o método computacionalmente intensivo baseado em simulações Monte Carlo via Cadeia de Markov (MCMC). Um estudo de simulação examinou o viés e o erro quadrático médio dos parâmetros a posteriori. Para a ilustração da abordagem, foram utilizados preços de ações de bancos Brasileiros. Além disso, foi verificado o efeito do strike e da estrutura de dependência nos preços das opções. Os resultados mostraram que os preços obtidos pelo método utilizado difere substancialmente dos obtidos pelo modelo clássico derivado de Black e Scholes. No segundo capítulo, consideramos os modelos GARCH-in-mean com especificações assimétricas para a variância com o objetivo de acomodar as características da volatilidade dos ativos-objetos sob uma perspectiva da dinâmica do risco-neutro. Além do mais, as funções cópulas foram utilizadas para capturar as possíveis estruturas de dependência linear, não-linear e caudais entre os ativos. Para ilustrar a metodologia, utilizamos dados de duas companhias Brasileiras. Confrontando os resultados obtidos com o modelo clássico extendido de Black e Scholes, notamos que a premissa de volatilidade constante sub-precifica as opções bivariadas, especialmente dentro-do-dinheiro.

自從1997年東南亞金融風暴，許多跨國企業紛紛倒閉，造成一連串的金融危機連鎖效應，無論是金融機構或投資人皆蒙受巨大的損失，使得金融市場開始正視信用風險的問題，增加對風險管理的重視。除了信用風險的問題外，由於過去幾年利率是一路下滑，使反浮動利率的公司結構債廣受歡迎，如今利率反轉上升，結構債的價格就會下跌虧損，手上握有大筆結構債的基金或是投信公司便因此受到牽連，自2005年起開始發行的CBO，便是為了解決公司結構債的問題而發行。在此環境下，更加速了信用衍生性商品的發展。 資產證券化對金融機構來說，除了有可以將信用風險移轉給投資人的好處之外，也是減低籌資成本的一個管道。另外，還有能增加收入、克服資本限制以及流動性限制等優點。 但在CDO之債權群組中，當債務人間的違約情況具有相關性時，個別債務人發生違約將可能連帶使得整個CDO債務現金流量來源產生嚴重衝擊。因此，如何準確推估CDO與CDO-squared各個分券下合理之信用價差，乃本研究分析商品的重點。 本研究採用Gaussian Copula，並利用蒙地卡羅法以及Probability Bucketing Approach評價擔保債權憑證。雖然Probability Bucketing Approach與蒙地卡羅法所模擬出來的結果很接近，然而Probability Bucketing Approach卻是較有效率的評價方法。在Probability Bucketing Approach中，損失級距的切割將會影響到評價的準確性，切割地越細密，越能準確地計算出損失分配，進而得到精確的信用價差。本文亦發現違約回收率、相關係數、違約率以及債權重複性(Overlap)均會顯著影響分券信用價差的評價，顯示參數正確評估之重要性。

<p>The dissertation consists of four chapters that concern topics on copulas for high dimensions. Chapter 1 proposes a new general model for high dimension joint distributions of asset returns that utilizes high frequency data and copulas. The dependence between returns is decomposed into linear and nonlinear components, which enables the use of high frequency data to accurately measure and forecast linear dependence, and the use of a new class of copulas designed to capture nonlinear dependence among the resulting linearly uncorrelated residuals. Estimation of the new class of copulas is conducted using a composite likelihood, making the model feasible even for hundreds of variables. A realistic simulation study verifies that multistage estimation with composite likelihood results in small loss in efficiency and large gain in computation speed. </p><p>Chapter 2, which is co-authored with Professor Andrew Patton, presents new models for the dependence structure, or copula, of economic variables based on a factor structure. The proposed models are particularly attractive for high dimensional applications, involving fifty or more variables. This class of models generally lacks a closed-form density, but analytical results for the implied tail dependence can be obtained using extreme value theory, and estimation via a simulation-based method using rank statistics is simple and fast. We study the finite-sample properties of the estimation method for applications involving up to 100 variables, and apply the model to daily returns on all 100 constituents of the S\&P 100 index. We find significant evidence of tail dependence, heterogeneous dependence, and asymmetric dependence, with dependence being stronger in crashes than in booms. </p><p>Chapter 3, which is co-authored with Professor Andrew Patton, considers the estimation of the parameters of a copula via a simulated method of moments type approach. This approach is attractive when the likelihood of the copula model is not known in closed form, or when the researcher has a set of dependence measures or other functionals of the copula that are of particular interest. The proposed approach naturally also nests method of moments and generalized method of moments estimators. Drawing on results for simulation based estimation and on recent work in empirical copula process theory, we show the consistency and asymptotic normality of the proposed estimator, and obtain a simple test of over-identifying restrictions as a goodness-of-fit test. The results apply to both $iid$ and time series data. We analyze the finite-sample behavior of these estimators in an extensive simulation study.</p><p>Chapter 4, which is co-authored with Professor Andrew Patton, proposes a new class of copula-based dynamic models for high dimension conditional distributions, facilitating the estimation of a wide variety of measures of systemic risk. Our proposed models draw on successful ideas from the literature on modelling high dimension covariance matrices and on recent work on models for general time-varying distributions. Our use of copula-based models enable the estimation of the joint model in stages, greatly reducing the computational burden. We use the proposed new models to study a collection of daily credit default swap (CDS) spreads on 100 U.S. firms over the period 2006 to 2012. We find that while the probability of distress for individual firms has greatly reduced since the financial crisis of 2008-09, the joint probability of distress (a measure of systemic risk) is substantially higher now than in the pre-crisis period.</p> / Dissertation

In this paper we use the conditional Value at Risk (CoVaR) and CoVaR variation (ΔCoVaR) proposed by Adrian and Brunnermeier (2008, 2011, 2016) to estimate the Peruvian stock market risk (through the IGBVL) conditioned on the international financial market (given that the S&P500) and conditioned on three of the main commodities exported by Peru: copper, silver and gold. Moreover, the CoVaR measures are compared with the VaR of the IGBVL to understand the differences using conditional and unconditional risk measure estimators. The results show that both CoVaR and ΔCoVaR are useful indicators to measure the Peruvian stock market risk. / En este trabajo utilizamos el Valor en Riesgo condicional (CoVaR) y la variación CoVaR (ΔCoVaR) propuestos por Adrian and Brunnermeier (2008, 2011, 2016) para estimar el riesgo bursátil peruano (a través del IGBVL) condicionado en el mercado internacional (dado por el índice S&P500) y condicionado en tres de los principales comodities exportados por el Perú: cobre, plata y oro. Además, las medidas CoVaR son comparadas con el VaR del IGBVL para entender las diferencias al utilizar medidas de riesgo condicionales e incondicionales. Los resultados muestran que ambas medidas CoVaR and ΔCoVaR constituyen indicadores útiles para estimar el riesgo bursátil peruano.

This thesis treats the modelling of a high-dimensional data set of longitudinal binary responses. The data consists of default indicators from different nations around the world as well as some explanatory variables such as exposure to underlying assets. The data used for the modelling is an aggregated term which combines several of the default indicators in the data set into one.  The modelling sets out from a portfolio perspective and seeks to find underlying correlations between the nations in the data set as well as see the extreme values produced by a portfolio with assets in the nations in the data set. The modelling takes a copula approach which uses Gaussian copulas to first formulate several different models mathematically and then optimize the parameters in the models to best fit the data set. Models A and B are optimized using standard stochastic gradient ascent on the likelihood function while model C uses variational inference and stochastic gradient ascent on the evidence lower bound for optimization. Using the different Gaussian copulas obtained from the optimization process a portfolio simulation is then done to examine the extreme values. The results show low correlations in models A and B while model C with it's additional regional correlations show slightly higher correlations in three of the subgroups. The portfolio simulations show similar tail behaviour in all three models, however model C produces more extreme risk measure outcomes in the form of higher VaR and ES. / Denna uppsats behandlar modellering av en datauppsättning bestående av högdimensionell longitudinell binärrespons. Datan består av konkursindikatorer för ett flertal suveräna stater runtom världen samt förklarande variabler så som exponering mot underliggande tillgångar. Datan som används i modelleringen är en aggregerad term som slår samman flera av konkursindikatorerna till en term. Modellerandet tar ett portföljperspektiv och försöker att finna underliggande korrelationer mellan nationerna i datamängden så väl som extremförluster som kan komma från en portfölj med tillgångar i de olika länderna som innefattas av datamängden. Utgångspunkten för modellerandet är ett copula-perspektiv som använder Gaussiska copulas där man först försöker matematiskt formulera flertalet modeller för att sedan optimera parametrarna i dessa modeller för att bäst passa datamängden till hands. För modell A och modell B optimeras log-likelihoodfunktionen med hjälp av stochastic gradient ascent medan i modell C används variational inference och sedan optimeras evidence lower bound med hjälp av stochastic gradient ascent. Med hjälp av de anpassade copula-modellerna simuleras sedan olika portföljer för att se vilka extremvärden som kan antas. Resultaten visar små korrelationer i modell A och B medan i modell C, med dess ytterligare regionala korrelationer, visas något större korrelation i tre av undergrupperna. Portföljsimuleringarna visar liknande svansbeteende i alla tre modeller, men modell C ger upphov till större riskmåttvärden i portföljerna i form av högre VaR och ES.

O presente trabalho foi motivado pela forte demanda por modelos de dependência mais precisos e realistas para aplicações a dados financeiros multivariados. A recente crise financeira de 2007-2009 deixou claro quão importante é uma modelagem precisa da dependência para a avaliação correta do risco financeiro: percepções equivocadas sobre dependências extremas entre diferentes ativos foram um elemento importante da crise do subprime. O famoso teorema dc Sklar (1959) introduziu as cópulas como uma ferramenta para se modelar padrões de dependência mais sofisticados. Ele estabelece que qualquer função de distribuição conjunta ndimensional pode ser decomposta em suas n distribuições marginais e uma cópula, sendo que a última caracteriza completamente a dependência entre as variáveis. Enquanto existe uma variedade de famílias de cópulas bivariadas que podem descrever um amplo conjunto de dependências complexas, o conjunto de cópulas com dimensão mais elevada era bastante restrito até recentemente. Joe (1996) propôs uma construção de distribuições nmltivariadas baseada em pair-copulas (cópulas bivariadas), chamada pair-copula construction ou modelo de vine cópula, que reverteu esse problema. Nesta tese, desenvolvemos três ensaios que exploram a teoria de cópulas para obter modelos de dependência multivariados muito flexíveis para aplicações a dados financeiros. Patton (2006) estendeu o teorema de Sklar para o caso de distribuições condicionais e tornou o parâmetro de dependência da cópula variante no tempo. No primeiro ensaio, introduzimos um novo enfoque para modelar a dependência entre retornos financeiros internacionais ao longo do tempo, combinando cópulas; tempo-variantes e o modelo de mudança Markoviana. Aplicamos esses modelos de cópula e também os modelos propostos por Patton (2006), Jondeau e Rockinger (2006) e Silva Filho et al. (2012a) aos retornos dos índices FTSE 100, CAC 40 e DAX. Comparamos essas metodologias em termos das dinâmicas de dependência resultantes e das habilidades dos modelos em prever Valor em Risco (VaR). Interessantemente, todos os modelos identificam um longo período de alta dependência entre os retornos começando em 2007, quando a crise do subprime teve início oficialmente. Surpreendentemente, as cópulas elípticas mostram melhor desempenho na previsão dos quantis extremos dos retornos dos portfólios. No segundo ensaio, estendemos nosso estudo para o caso de n > 2 variáveis, usando o modelo de vine cópula para investigar a estrutura de dependência dos índices CAC 40, DAX, FTSE 100, S&P 500 e IBOVESPA, e, particularmente, checar a hipótese de dependência assimétrica nesse caso. Com base em nossos resultados empíricos, entretanto, essa hipótese não pode ser verificada. Talvez a dependência assimétrica com caudas inferiores mais fortes ocorra apenas temporariamente, o que sugere que a incorporação de variação temporal ao modelo de vine cópula pode melhorá-lo como ferramenta para modelar dados financeiros internacionais multivariados. Desta forma, no terceiro ensaio, introduzimos dinâmica no modelo de vine cópula permitindo que os parâmetros de dependência das pair-copulas em uma decomposição D-vine sejam potencialmente variantes no tempo, seguindo um processo ARMA(l,m) restrito como em Patton (2006). O modelo proposto é avaliado em simulações e também com respeito à acurácia das previsões de Valor em Risco (VaR) em períodos de crise. Os experimentos de Monte Cailo são bastante favoráveis à cópula D-vine dinâmica em comparação a uma cópula D-vine estática. Adicionalmente, a cópula D-vine dinâmica supera a cópula D-vine estática em termos de acurária preditiva para os nossos conjuntos de dados / This work was motivated by the strong demand for more precise and realistic dependence models for applications to multivariate financial data. The recent financial crisis of 2007-2009 has made it clear how important is a precise modeling of dependence for the accurate assessment of financial risk: misperceptions about extreme dependencies between different financial assets were an important element of the subprime crisis. The famous theorem by Sklar (1959) introduced the copulas as a tool to model more intricate patterns of dependence. It states that any n-dimensional joint distribution function can be decomposed into its n marginal distributions and a copula, where the latter completely characterizes the dependence among the variables. While there is a variety of bivariate copula families, which can match a wide range of complex dependencies, the set of higher-dimensional copulas was quite restricted until recently. Joe (1996) proposed a construction of multivariate distributions based on pair-copulas (bivariate copulas), called pair-copula construction or vine copula model, that has overcome this issue. In this thesis, we develop three papers that explore the copula theory in order to obtain very flexible multivariate dependence rnodels for applications to financial data. Patton (2006) extended Sklar's theorem to the conditional case and rendered the dependence parameter of the copula time-varying. In the first paper, we introduce a new approach to modeling dependence between International financial returns over time, combining time-varying copulas and the Markov switching model. We apply these copula models and also those proposed by Patton (2006), Jondeau and Rockinger (2006) and Silva Filho et al. (2012a) to the return data of FTSE 100, CAC 40 and DAX indexes. We compare these methodologies in terms of the resulting dynamics of dependence and the models' abilities to forecast Value-at-Risk (VaR). Interestingly, ali the models identify a long period of high dependence between the returns beginning in 2007, when the subprime crisis was evolving. Surprisingly, the elhptical copulas perform best in forecasting the extreme quantiles of the portfolios returns. In the second paper, we extend our study to the case of n > 2 variables, using the vine copula model to investigate the dependence structure of the broad stock market indexes CAC 40, DAX, FTSE 100, S&P 500 and IBOVESPA, and, particularly, check the asymmetric dependence hypothesis in this case. Based on our empirical results, however, this hypothesis cannot be verified. Perhaps, asymmetric dependence with stronger lower tails occurs only temporarily, what suggests that incorporating time variation into the vine copula rnodel can improve it as a tool to rnodel multivariate International financial data. So, in the third paper, we introduce dynamics into the vine copula model by allowing the dependence parameters of the pair-copulas in a D-vine decomposition to be potentially timevarying, following a nonlinear restricted ARMA(l,m) process as in Patton (2006). The proposed model is evaluated in simulations and further assessed with respect to the accuracy of Value-at- Risk (VaR) forecasts in crisis periods. The Monte Cario experiments are quite favorable to the dynamic D-vine copula in comparison with a static D-vine copula. Moreover, the dynamic Dvine copula outperforms the static D-vine copula in terms of predictive accuracy for our data sets.

O presente trabalho foi motivado pela forte demanda por modelos de dependência mais precisos e realistas para aplicações a dados financeiros multivariados. A recente crise financeira de 2007-2009 deixou claro quão importante é uma modelagem precisa da dependência para a avaliação correta do risco financeiro: percepções equivocadas sobre dependências extremas entre diferentes ativos foram um elemento importante da crise do subprime. O famoso teorema dc Sklar (1959) introduziu as cópulas como uma ferramenta para se modelar padrões de dependência mais sofisticados. Ele estabelece que qualquer função de distribuição conjunta ndimensional pode ser decomposta em suas n distribuições marginais e uma cópula, sendo que a última caracteriza completamente a dependência entre as variáveis. Enquanto existe uma variedade de famílias de cópulas bivariadas que podem descrever um amplo conjunto de dependências complexas, o conjunto de cópulas com dimensão mais elevada era bastante restrito até recentemente. Joe (1996) propôs uma construção de distribuições nmltivariadas baseada em pair-copulas (cópulas bivariadas), chamada pair-copula construction ou modelo de vine cópula, que reverteu esse problema. Nesta tese, desenvolvemos três ensaios que exploram a teoria de cópulas para obter modelos de dependência multivariados muito flexíveis para aplicações a dados financeiros. Patton (2006) estendeu o teorema de Sklar para o caso de distribuições condicionais e tornou o parâmetro de dependência da cópula variante no tempo. No primeiro ensaio, introduzimos um novo enfoque para modelar a dependência entre retornos financeiros internacionais ao longo do tempo, combinando cópulas; tempo-variantes e o modelo de mudança Markoviana. Aplicamos esses modelos de cópula e também os modelos propostos por Patton (2006), Jondeau e Rockinger (2006) e Silva Filho et al. (2012a) aos retornos dos índices FTSE 100, CAC 40 e DAX. Comparamos essas metodologias em termos das dinâmicas de dependência resultantes e das habilidades dos modelos em prever Valor em Risco (VaR). Interessantemente, todos os modelos identificam um longo período de alta dependência entre os retornos começando em 2007, quando a crise do subprime teve início oficialmente. Surpreendentemente, as cópulas elípticas mostram melhor desempenho na previsão dos quantis extremos dos retornos dos portfólios. No segundo ensaio, estendemos nosso estudo para o caso de n > 2 variáveis, usando o modelo de vine cópula para investigar a estrutura de dependência dos índices CAC 40, DAX, FTSE 100, S&P 500 e IBOVESPA, e, particularmente, checar a hipótese de dependência assimétrica nesse caso. Com base em nossos resultados empíricos, entretanto, essa hipótese não pode ser verificada. Talvez a dependência assimétrica com caudas inferiores mais fortes ocorra apenas temporariamente, o que sugere que a incorporação de variação temporal ao modelo de vine cópula pode melhorá-lo como ferramenta para modelar dados financeiros internacionais multivariados. Desta forma, no terceiro ensaio, introduzimos dinâmica no modelo de vine cópula permitindo que os parâmetros de dependência das pair-copulas em uma decomposição D-vine sejam potencialmente variantes no tempo, seguindo um processo ARMA(l,m) restrito como em Patton (2006). O modelo proposto é avaliado em simulações e também com respeito à acurácia das previsões de Valor em Risco (VaR) em períodos de crise. Os experimentos de Monte Cailo são bastante favoráveis à cópula D-vine dinâmica em comparação a uma cópula D-vine estática. Adicionalmente, a cópula D-vine dinâmica supera a cópula D-vine estática em termos de acurária preditiva para os nossos conjuntos de dados / This work was motivated by the strong demand for more precise and realistic dependence models for applications to multivariate financial data. The recent financial crisis of 2007-2009 has made it clear how important is a precise modeling of dependence for the accurate assessment of financial risk: misperceptions about extreme dependencies between different financial assets were an important element of the subprime crisis. The famous theorem by Sklar (1959) introduced the copulas as a tool to model more intricate patterns of dependence. It states that any n-dimensional joint distribution function can be decomposed into its n marginal distributions and a copula, where the latter completely characterizes the dependence among the variables. While there is a variety of bivariate copula families, which can match a wide range of complex dependencies, the set of higher-dimensional copulas was quite restricted until recently. Joe (1996) proposed a construction of multivariate distributions based on pair-copulas (bivariate copulas), called pair-copula construction or vine copula model, that has overcome this issue. In this thesis, we develop three papers that explore the copula theory in order to obtain very flexible multivariate dependence rnodels for applications to financial data. Patton (2006) extended Sklar's theorem to the conditional case and rendered the dependence parameter of the copula time-varying. In the first paper, we introduce a new approach to modeling dependence between International financial returns over time, combining time-varying copulas and the Markov switching model. We apply these copula models and also those proposed by Patton (2006), Jondeau and Rockinger (2006) and Silva Filho et al. (2012a) to the return data of FTSE 100, CAC 40 and DAX indexes. We compare these methodologies in terms of the resulting dynamics of dependence and the models' abilities to forecast Value-at-Risk (VaR). Interestingly, ali the models identify a long period of high dependence between the returns beginning in 2007, when the subprime crisis was evolving. Surprisingly, the elhptical copulas perform best in forecasting the extreme quantiles of the portfolios returns. In the second paper, we extend our study to the case of n > 2 variables, using the vine copula model to investigate the dependence structure of the broad stock market indexes CAC 40, DAX, FTSE 100, S&P 500 and IBOVESPA, and, particularly, check the asymmetric dependence hypothesis in this case. Based on our empirical results, however, this hypothesis cannot be verified. Perhaps, asymmetric dependence with stronger lower tails occurs only temporarily, what suggests that incorporating time variation into the vine copula rnodel can improve it as a tool to rnodel multivariate International financial data. So, in the third paper, we introduce dynamics into the vine copula model by allowing the dependence parameters of the pair-copulas in a D-vine decomposition to be potentially timevarying, following a nonlinear restricted ARMA(l,m) process as in Patton (2006). The proposed model is evaluated in simulations and further assessed with respect to the accuracy of Value-at- Risk (VaR) forecasts in crisis periods. The Monte Cario experiments are quite favorable to the dynamic D-vine copula in comparison with a static D-vine copula. Moreover, the dynamic Dvine copula outperforms the static D-vine copula in terms of predictive accuracy for our data sets.

O presente trabalho foi motivado pela forte demanda por modelos de dependência mais precisos e realistas para aplicações a dados financeiros multivariados. A recente crise financeira de 2007-2009 deixou claro quão importante é uma modelagem precisa da dependência para a avaliação correta do risco financeiro: percepções equivocadas sobre dependências extremas entre diferentes ativos foram um elemento importante da crise do subprime. O famoso teorema dc Sklar (1959) introduziu as cópulas como uma ferramenta para se modelar padrões de dependência mais sofisticados. Ele estabelece que qualquer função de distribuição conjunta ndimensional pode ser decomposta em suas n distribuições marginais e uma cópula, sendo que a última caracteriza completamente a dependência entre as variáveis. Enquanto existe uma variedade de famílias de cópulas bivariadas que podem descrever um amplo conjunto de dependências complexas, o conjunto de cópulas com dimensão mais elevada era bastante restrito até recentemente. Joe (1996) propôs uma construção de distribuições nmltivariadas baseada em pair-copulas (cópulas bivariadas), chamada pair-copula construction ou modelo de vine cópula, que reverteu esse problema. Nesta tese, desenvolvemos três ensaios que exploram a teoria de cópulas para obter modelos de dependência multivariados muito flexíveis para aplicações a dados financeiros. Patton (2006) estendeu o teorema de Sklar para o caso de distribuições condicionais e tornou o parâmetro de dependência da cópula variante no tempo. No primeiro ensaio, introduzimos um novo enfoque para modelar a dependência entre retornos financeiros internacionais ao longo do tempo, combinando cópulas; tempo-variantes e o modelo de mudança Markoviana. Aplicamos esses modelos de cópula e também os modelos propostos por Patton (2006), Jondeau e Rockinger (2006) e Silva Filho et al. (2012a) aos retornos dos índices FTSE 100, CAC 40 e DAX. Comparamos essas metodologias em termos das dinâmicas de dependência resultantes e das habilidades dos modelos em prever Valor em Risco (VaR). Interessantemente, todos os modelos identificam um longo período de alta dependência entre os retornos começando em 2007, quando a crise do subprime teve início oficialmente. Surpreendentemente, as cópulas elípticas mostram melhor desempenho na previsão dos quantis extremos dos retornos dos portfólios. No segundo ensaio, estendemos nosso estudo para o caso de n > 2 variáveis, usando o modelo de vine cópula para investigar a estrutura de dependência dos índices CAC 40, DAX, FTSE 100, S&P 500 e IBOVESPA, e, particularmente, checar a hipótese de dependência assimétrica nesse caso. Com base em nossos resultados empíricos, entretanto, essa hipótese não pode ser verificada. Talvez a dependência assimétrica com caudas inferiores mais fortes ocorra apenas temporariamente, o que sugere que a incorporação de variação temporal ao modelo de vine cópula pode melhorá-lo como ferramenta para modelar dados financeiros internacionais multivariados. Desta forma, no terceiro ensaio, introduzimos dinâmica no modelo de vine cópula permitindo que os parâmetros de dependência das pair-copulas em uma decomposição D-vine sejam potencialmente variantes no tempo, seguindo um processo ARMA(l,m) restrito como em Patton (2006). O modelo proposto é avaliado em simulações e também com respeito à acurácia das previsões de Valor em Risco (VaR) em períodos de crise. Os experimentos de Monte Cailo são bastante favoráveis à cópula D-vine dinâmica em comparação a uma cópula D-vine estática. Adicionalmente, a cópula D-vine dinâmica supera a cópula D-vine estática em termos de acurária preditiva para os nossos conjuntos de dados / This work was motivated by the strong demand for more precise and realistic dependence models for applications to multivariate financial data. The recent financial crisis of 2007-2009 has made it clear how important is a precise modeling of dependence for the accurate assessment of financial risk: misperceptions about extreme dependencies between different financial assets were an important element of the subprime crisis. The famous theorem by Sklar (1959) introduced the copulas as a tool to model more intricate patterns of dependence. It states that any n-dimensional joint distribution function can be decomposed into its n marginal distributions and a copula, where the latter completely characterizes the dependence among the variables. While there is a variety of bivariate copula families, which can match a wide range of complex dependencies, the set of higher-dimensional copulas was quite restricted until recently. Joe (1996) proposed a construction of multivariate distributions based on pair-copulas (bivariate copulas), called pair-copula construction or vine copula model, that has overcome this issue. In this thesis, we develop three papers that explore the copula theory in order to obtain very flexible multivariate dependence rnodels for applications to financial data. Patton (2006) extended Sklar's theorem to the conditional case and rendered the dependence parameter of the copula time-varying. In the first paper, we introduce a new approach to modeling dependence between International financial returns over time, combining time-varying copulas and the Markov switching model. We apply these copula models and also those proposed by Patton (2006), Jondeau and Rockinger (2006) and Silva Filho et al. (2012a) to the return data of FTSE 100, CAC 40 and DAX indexes. We compare these methodologies in terms of the resulting dynamics of dependence and the models' abilities to forecast Value-at-Risk (VaR). Interestingly, ali the models identify a long period of high dependence between the returns beginning in 2007, when the subprime crisis was evolving. Surprisingly, the elhptical copulas perform best in forecasting the extreme quantiles of the portfolios returns. In the second paper, we extend our study to the case of n > 2 variables, using the vine copula model to investigate the dependence structure of the broad stock market indexes CAC 40, DAX, FTSE 100, S&P 500 and IBOVESPA, and, particularly, check the asymmetric dependence hypothesis in this case. Based on our empirical results, however, this hypothesis cannot be verified. Perhaps, asymmetric dependence with stronger lower tails occurs only temporarily, what suggests that incorporating time variation into the vine copula rnodel can improve it as a tool to rnodel multivariate International financial data. So, in the third paper, we introduce dynamics into the vine copula model by allowing the dependence parameters of the pair-copulas in a D-vine decomposition to be potentially timevarying, following a nonlinear restricted ARMA(l,m) process as in Patton (2006). The proposed model is evaluated in simulations and further assessed with respect to the accuracy of Value-at- Risk (VaR) forecasts in crisis periods. The Monte Cario experiments are quite favorable to the dynamic D-vine copula in comparison with a static D-vine copula. Moreover, the dynamic Dvine copula outperforms the static D-vine copula in terms of predictive accuracy for our data sets.