Esta tese apresenta três ensaios em macroeconomia aplicada e que possuem em comum o uso de técnicas estatísticas e econométricas em problemas macroeconômicos. Dentre os campos de pesquisa da macroeconomia aplicada, a tese faz uso de modelos macroeconômicos microfundamentados, em sua versão DSGE-VAR, e da macroeconomia financeira por meio da avaliação do comportamento da correlação entre os retornos das ações usando modelos Garch multivariados. Além disso, a tese provoca a discussão sobre um novo campo de pesquisa em macroeconomia que surge a partir do advento da tecnologia. No primeiro ensaio, aplicamos a abordagem DSGE-VAR na discussão sobre a reação do Banco Central do Brasil (BCB) as oscilações na taxa de câmbio, especificamente para o caso de uma economia sob metas de inflação. Para tanto, baseando-se no modelo para uma economia aberta desenvolvido por Gali e Monacelli (2005) e modificado por Lubik e Schorfheide (2007), estimamos uma regra de política monetária para o Brasil e examinamos em que medida o BCB responde a mudanças na taxa de câmbio. Além disso, estudamos o grau de má especificação do modelo DSGE proposto. Mais especificamente, comparamos a verossimilhança marginal do modelo DSGE às do modelo DSGE-VAR e examinamos se o Banco Central conseguiu isolar a economia brasileira, em particular a inflação, de choques externos. Nossas conclusões mostram que as respostas aos desvios da taxa de câmbio são diferentes de zero e menores do que as respostas aos desvios da inflação. Finalmente, o ajuste do modelo DSGE é consideravelmente pior do que o ajuste do modelo DSGE-VAR, independentemente do número de defasagens utilizadas no VAR o que indica que de um ponto de vista estatístico existem evidências de que as restrições cruzadas do modelo teórico são violadas nos dados. O segundo ensaio examina empiricamente o comportamento da correlação entre o retorno de ações listadas na BMF&BOVESPA no período de 2000 a 2015. Para tanto, utilizamos modelos GARCH multivariados introduzidos por Bollerslev (1990) para extrair a série temporal das matrizes de correlação condicional dos retornos das ações. Com a série temporal dos maiores autovalores das matrizes de correlação condicional estimadas, aplicamos testes estatísticos (raiz unitária, quebra estrutural e tendência) para verificar a existência de tendência estocástica ou determinística para a intensidade da correlação entre os retornos das ações representadas pelos autovalores. Nossas conclusões confirmam que tanto em períodos de crises nacionais como turbulências internacionais, há intensificação da correlação entre as ações. Contudo, não encontramos qualquer tendência de longo prazo na série temporal dos maiores autovalores das matrizes de correlação condicional. Isso sugere que apesar das conclusões de Costa, Mazzeu e Jr (2016) sobre a tendência de queda do risco idiossincrático no mercado acionário brasileiro, a correlação dos retornos não apresentou tendência de alta, conforme esperado pela teoria de finanças. No terceiro ensaio, apresentamos pesquisas que utilizaram Big Data, Machine Learning e Text Mining em problemas macroeconômicos e discutimos as principais técnicas e tecnologias adotadas bem como aplicamos elas na análise de sentimento do BCB sobre a economia. Por meio de técnicas de Web Scraping e Text Mining, acessamos e extraímos as palavras usadas na escrita das atas divulgadas pelo Comitê de Política Monetária (Copom) no site do BCB. Após isso, comparando tais palavras com um dicionário de sentimentos (Inquider) mantido pela Universidade de Harvard e originalmente apresentado por Stone, Dunphy e Smith (1966), foi possível criar um índice de sentimento para a autoridade monetária. Nossos resultados confirmam que tal abordagem pode contribuir para a avaliação econômica dado que a série temporal do índice proposto está relacionada com variáveis macroeconômicas importantes para as decisões do BCB. / This thesis presents three essays in applied macroeconomics and who have in common the use of statistical and econometric techniques in macroeconomic problems. Among the search fields of applied macroeconomics, the thesis makes use of microfounded macroeconomic models, in tis DSGE-VAR version, and financial macroeconomics through the evaluation of the behavior of correlation between stock returns using multivariate Garch models. In addition, leads a discussion on a new field of research in macroeconomics which arises from the advent of technology. In the first experiment, we applied the approach to dynamic stochastic general equilibrium (DSGE VAR in the discussion about the reaction of the Central Bank of Brazil (CBB) to fluctuations in the exchange rate, specifically for the case of an economy under inflation targeting. To this end, based on the model for an open economy developed by Gali and Monacelli (2005) and modified by Lubik and Schorfheide (2007), we estimate a rule of monetary policy for the United States and examine to what extent the CBC responds to changes in the exchange rate. In addition, we studied the degree of poor specification of the DSGE model proposed. More specifically, we compare the marginal likelihood of the DSGE model to the DSGE-VAR model and examine whether the Central Bank managed to isolate the brazilian economy, in particular the inflation, external shocks. Our findings show that the response to deviations of the exchange rate are different from zero and lower than the response to deviations of inflation. Finally, the adjustment of the DSGE model is considerably worse than the adjustment of the DSGE-VAR model, regardless of the number of lags used in the VAR which indicates that a statistical point of view there is evidence that the restrictions crusades of the theoretical model are violated in the data. The second essay examines empirically the behavior of the correlation between the return of shares listed on the BMF&BOVESPA over the period from 2000 to 2015. To this end, we use models multivariate GARCH introduced by Bollerslev (1990) to remove the temporal series of arrays of conditional correlation of returns of stocks. With the temporal series of the largest eigenvalues of matrices of correlation estimated conditional, we apply statistical tests (unit root, structural breaks and trend) to verify the existence of stochastic trend or deterministic to the intensity of the correlation between the returns of the shares represented by eigenvalues. Our findings confirm that both in times of crises at national and international turbulence, there is greater correlation between the actions. However, we did not find any long-term trend in time series of the largest eigenvalues of matrices of correlation conditional. In the third test, we present research that used Big Data, Machine Learning and Text Mining in macroeconomic problems and discuss the main techniques and technologies adopted and apply them in the analysis of feeling of BCB on the economy. Through techniques of Web Scraping and Text Mining, we accessed and extracted the words used in the writing of the minutes released by the Monetary Policy Committee (Copom) on the site of the BCB. After that, comparing these words with a dictionary of feelings (Inquider) maintained by Harvard University and originally presented by Stone, Dunphy and Smith (1966), it was possible to create an index of sentiment for the monetary authority. Our results confirm that such an approach can contribute to the economic assessment given that the temporal series of the index proposed is related with macroeconomic variables are important for decisions of the BCB.

Esta dissertação analisa a questão da seleção de modelos GARCH multivariados em termos da perfomance de previsão da matriz de covariância condicional. A aplicação empírica é realizada com 7 retornos de índices de ações envolvendo um conjunto de 34 especificações de modelos para os quais computamos as previsões da variância condicional um passo a frente para uma amostra com 60 observações para cada especificação dos modelos GARCH multivariados. A comparação entre os modelos é baseada no procedimento Model Confidence Set (MCS) avaliado através de duas funções perdas robustas a proxies de volatilidade imperfeitas. O MCS é um procedimento que permite comparar vários modelos simultaneamente em termos de sua habilidade preditiva e determinar um conjunto de modelos estatisticamente semelhantes em termos de previsão, dado um nível de confiança. / This paper considers the question of the selection of multivariate GARCH models in terms of covariance matrix forecasting. In the empirical application we consider 7 series of returns and compare a set of 34 model specifications based on one-step-ahead conditional variance forecasts over a sample with 60 observations. The comparison between models is performed with the Model Confidence Set (MCS) procedure evaluated using two loss functions that are robust against imperfect volatility proxies. The MCS is a procedure that allows both a multiple model comparison in terms of forecasting accuracy and the determination of a model set composed of statistically equivalent models, under a confidence level.

Esta dissertação analisa a questão da seleção de modelos GARCH multivariados em termos da perfomance de previsão da matriz de covariância condicional. A aplicação empírica é realizada com 7 retornos de índices de ações envolvendo um conjunto de 34 especificações de modelos para os quais computamos as previsões da variância condicional um passo a frente para uma amostra com 60 observações para cada especificação dos modelos GARCH multivariados. A comparação entre os modelos é baseada no procedimento Model Confidence Set (MCS) avaliado através de duas funções perdas robustas a proxies de volatilidade imperfeitas. O MCS é um procedimento que permite comparar vários modelos simultaneamente em termos de sua habilidade preditiva e determinar um conjunto de modelos estatisticamente semelhantes em termos de previsão, dado um nível de confiança. / This paper considers the question of the selection of multivariate GARCH models in terms of covariance matrix forecasting. In the empirical application we consider 7 series of returns and compare a set of 34 model specifications based on one-step-ahead conditional variance forecasts over a sample with 60 observations. The comparison between models is performed with the Model Confidence Set (MCS) procedure evaluated using two loss functions that are robust against imperfect volatility proxies. The MCS is a procedure that allows both a multiple model comparison in terms of forecasting accuracy and the determination of a model set composed of statistically equivalent models, under a confidence level.

O objetivo deste trabalho consiste em identificar a existência de contágio financeiro utilizando a metodologia de redes bayesianas. Além da rede bayesiana, a análise da interdependência de mercados internacionais em períodos de crises financeiras, ocorridas entre os anos 1996 e 2009, foi modelada com outras duas técnicas - modelos GARCH multivariados e de Cópulas, envolvendo países nos quais foi possível avaliar seus efeitos e que foram objetos de estudos similares na literatura. Com os períodos de crise bem definidos e metodologia calcada na teoria de grafos e na inferência bayesiana, executou-se uma análise sequencial, em que as realidades que precediam períodos de crise foram consideradas situações a priori para os eventos (verossimilhanças). Desta combinação resulta a nova realidade (a posteriori), que serve como priori para o período subsequente e assim por diante. Os resultados apontaram para grande interligação entre os mercados e diversas evidências de contágio em períodos de crise financeira, com causadores bem definidos e com grande respaldo na literatura. Ademais, os pares de países que apresentaram evidências de contágio financeiro pelas redes bayesianas em mais períodos de crises foram os mesmos que apresentaram os mais altos valores dos parâmetros estimados pelas cópulas e também aqueles cujos parâmetros foram mais fortemente significantes no modelo GARCH multivariado. Assim, os resultados obtidos pelas redes bayesianas tornam-se mais relevantes, o que sugere boa aderência deste modelo ao conjunto de dados utilizados neste estudo. Por fim, verificou-se que, após as diversas crises, os mercados estavam muito mais interligados do que no período inicialmente adotado. / This work aims to identify the existence of financial contagion using a metodology of Bayesian networks. Besides Bayesian networks, the analysis of the international markets\' interdependence in times of financial crises, occurred between 1996 and 2009, was modeled using two other techniques - multivariate GARCH models and Copulas models, involving countries in which its effects were possible to assess and which were subject to similar studies in the literature. With well-defined crisis periods and a metodology based on graph theory and Bayesian inference, a sequential analysis was executed, in which the realities preceding periods of crisis were considered to be prior situations to the events (likelihood). From this combination results the new posterior reality, which serves as a prior to the subsequent period and so on. The results pointed to a large interconnection between markets and several evidences of contagion in times of financial crises, with well-defined responsibles and highly supported by the literature. Moreover, the pairs of countries that show evidence of financial contagion by Bayesian networks in over periods of crises were the same as that presented the highest values of the parameters estimated by copulas and the most strongly significant parameters in the multivariate GARCH model. Thus, the results obtained by Bayesian networks become more relevant, suggesting good adherence of the model to the data set used in this study. Finally, it was found that after the various crises, the markets were much more connected.

In this work we will describe methods for modeling multivariate financial time series. We will concentrate on both modeling expected value by multi- variate Box-Jenkins processes and primarily on modeling conditional corre- lations and volatility. Our main object will be DCC (Dynamic Conditional Correlation) model, estimation of its parameters and some other general- izations. Then we will programme DCC model in statistical software R and apply on real data. In applications we will concentrate on problem of high dimension of financial time series and on modeling conditional correlations data with outliers.

This thesis has treated the subject of DCC-GARCH model’s forecasting ability and Value-at- Risk applications on the Scandinavian foreign exchange market. The estimated models were based on daily opening foreign exchange spot rates in the period of 2004-2013, which captured the information in the financial crisis of 2008 and Eurozone crisis in the early 2010s. The forecasts were performed on a one-day rolling window in 2014. The results show that the DCC-GARCH model accurately predicted the fluctuation in the conditional correlation, although not with the correct magnitude. Furthermore, the DCC-GARCH model shows good Value-at-Risk forecasting performance for different portfolios containing the Scandinavian currencies.

The main purpose of this thesis is to assess the potential of emerging stock markets and commodity markets in attracting the attention of international investors who utilize various portfolio diversification strategies to reduce the cumulative risk of their portfolio. A successful portfolio diversification strategy requires low correlation among financial markets. However, it is now well documented that the correlations among financial markets in developed countries are very high and hence the benefits of international portfolio diversification among these markets have been very limited. This fact suggests that investors should look for alternative markets whose correlations with developed markets are low (or even negative if possible) and which have high growth potentials. In this thesis, two emerging countries&#039 / stock markets and two commodity markets are considered as alternative markets. Among emerging countries, Turkey and China are chosen due to their promising growth performance since the mid-2000s. As commodity markets, agricultural commodity and precious metal markets are selected because of the outstanding performance of the former and the &quot / safe harbor&quot / property of the latter. The structures and properties of dependence between these markets and stock markets in developed countries are examined by modeling the conditional correlation in the dynamic conditional correlation framework. The results reveal that upward trend hypothesis is valid for almost all correlations among market pairs and market volatility plays significant role in time varying structures of correlations.

The purpose of this study is to examine the volatility spillover among a country&rsquo / s foreign exchange, bond and stock markets and the volatility transmission from the global bond, stock and commodity markets to these local financial markets. The sample for the study includes data from both emerging and developed economies in the time period between 2004 and 2011. A multivariate GARCH methodology with the BEKK representation is applied for the local financial markets and global variables are included as exogenous variables into the model. The volatility integration of the financial markets of the emerging economies is stronger compared to the integration of the developed economies. Global variables have a spillover effect on the developed markets only after the global financial crisis, whereas they significantly affect the volatility in emerging markets for both the pre- and post-crisis period. North American countries in the sample, U.S. and Mexico, have low local volatility integration in the pre-crisis era and the integration rises in the post-crisis period. Moreover, they are more open to the internal and global short-term shocks in the post-crisis period. Germany and Turkey are the representatives of the EMEA (Europe, Middle East and Africa) region and they have high local market integration and are open to global shocks for both sub-periods. Far Eastern markets, Japan and Korea, also have high local market integration and their vulnerability to the global effects is large and getting larger for the post-crisis period. The most important limitation of this thesis is the difficulty of reaching sharp generalizations due to the small number of countries analyzed. This limitation can be addressed by the inclusion of a larger number of geographically dispersed countries. The most noteworthy originality of this study is the addition of the exogenous global variables for modeling volatility spillovers. Furthermore, comparison of results for emerging versus developed markets and the pre- versus post-crisis periods is another contribution of this study to the existing literature. The findings of this study can be used by investors interested in assessing the risks of investing internationally.

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

When modelling more that one asset, it is desirable to apply multivariate modeling to capture the co-movements of the underlying assets. The GARCH models has been proven to be successful when it comes to volatility forecast- ing. Hence it is natural to extend from a univariate GARCH model to a multivariate GARCH model when examining portfolio volatility. This study aims to evaluate a specific multivariate GARCH model, the DCC-GARCH model, which was developed by Engle and Sheppard in 2001. In this pa- per different DCC-GARCH models have been implemented, assuming both Gaussian and multivariate Student’s t distribution. These distributions are compared by a set of tests as well as Value at Risk backtesting. / I portföljanalys så är det åtråvärt att applicera flerdimensionella modeller för att kunna fånga hur de olika tillgångarna rör sig tillsammans. GARCH-modeller har visat sig vara framgångsrika när det kommer till prognoser av volatilitet. Det är därför naturligt att gå från endimensionella till flerdimensionella GARCH-modeller när volatiliteten av en portfölj skall utvärderas. Den här studien ämnar att utvärdera tillvägagångssättet för prognoser av en viss typ av flerdimensionell GARCH-modell, DCC-GARCH-modellen, vilken utvecklades av Engle och Sheppard 2001. I den här uppsatsen har olika DCC-GARCH modeller blivit implementerade, som antar innovationer enligt både flerdimensionell normalfördelning samt flerdimensionell student's t-fördelning. Dessa jämförs med hjälp av en handfull tester samt Value-at-Risk backtesting.