[pt] volatilidade dos ativos financeiros reflete uma reação prosseguida dos agentes a choques no passado ou alterações nas condições dos mercados determinam mudanças na dinâmica da variável? Enquanto modelos fracionalmente integrados vêm sendo extensamente utilizados como uma descrição adequada do processo gerador de séries de volatilidade, trabalhos teóricos recentes indicaram que mudanças estruturais podem ser uma relevante alternativa empírica para o fato estilizado de memória longa. O presente trabalho investiga o que alterações nos mercados significam nesse contexto, introduzindo variações de preços como uma possível fonte de mudanças no nível da volatilidade durante algum período, com grandes quedas (ascensões) nos preços trazendo regimes persistentes de variância alta (baixa). Uma estratégia de modelagem sistemática e flexível é estabelecida para testar e estimar essa assimetria através da incorporação de retornos acumulados passados num arcabouço não-linear. O principal resultado revela que o efeito é altamente significante - estima-se que níveis de volatilidade 25% e 50% maiores estão associados a quedas nos preços em períodos curtos - e é capaz de explicar altos valores de estimativas do parâmetro de memória longa. Finalmente, mostra-se que a modelagem desse efeito traz ganhos importantes para aplicações fora da amostra em períodos de volatilidade alta. / [en] Does volatility reflect lasting reactions to past shocks or changes in the markets induce shifts in this variable dynamics? In this work, we argue that price variations are an essential source of information about multiple regimes in the realized volatility of stocks, with large falls (rises) in prices bringing persistent regimes of high (low) variance. The study shows that this asymmetric effect is highly significant (we estimate that falls of different magnitudes over less than two months are associated with volatility levels 20% and 60% higher than the average of periods with stable or rising prices) and support large empirical values of long memory parameter estimates. We show that a model based on those findings significantly improves out of sample performance in relation to standard methods {specially in periods of high volatility.

[pt] Esta dissertação tem por objetivo comparar o desempenho de modelos não lineares de previsão de retornos em 10 ativos do mercado acionário brasileiro. Entre os modelos escolhidos, pode-se citar o STAR-Tree, que combina conceitos da metodologia STAR (Smooth Transition AutoRegression) e do algoritmo CART (Classification And Regression Trees), tendo como resultado final uma regressão com transição suave entre múltiplos regimes. A especificação do modelo é feita através de testes de hipótese do tipo Multiplicador de Lagrange que indicam o nó a ser dividido e a variável explicativa correspondente. A estimação dos parâmetros é feita pelo método de Mínimos Quadrados Não Lineares para determinar o valor dos parâmetros lineares e não lineares. Redes Neurais, modelos ARMAX (estes lineares) e ainda o método Naive também foram incluídos na análise. Os resultados das previsões foram avaliados a partir de medidas estatísticas e financeiras e se basearam em um negociador automático que informa o instante correto de assumir uma posição comprada ou vendida em cada ativo. Os melhores desempenhos foram alcançados pelas Redes Neurais, pelos modelos ARMAX e pela forma de previsão ARC (Adaptative Regime Combination) derivada da metodologia STAR-Tree, sendo ambos ainda superiores ao retorno das ações durante o período de teste / [en] The goal of this dissertation is to compare the performance of non linear models to forecast return on 10 equities in the Brazilian Stock Market. Among the chosen ones, it can be cited the STAR-Tree, which matches concepts from the STAR (Smooth Transition AutoRegression) methodology and the CART (Classification And Regression Trees) algorithm, having as the resultant structure a regression with smooth transition among multiple regimes. The model specification is done by Lagrange Multiplier hypothesis tests that indicate the node to be splitted and the corresponding explanatory variable. The parameter estimation is done by the Non Linear Least Squares method that determine the linear and non linear parameters. Neural Netwoks, ARMAX models (these ones linear) and the Naive method were also included in the analysis. The forecasting results were calculated using statistical and financial measures and were based on an automatic negociator that signaled the right instant to take a short or a long position in each stock. The best results were reached by the Neural Networks, ARMAX models and ARC (Adaptative Regime Combination ) forecasting method derived from STAR-Tree, with all of them performing better then the equity return during the test period.

Made available in DSpace on 2014-06-11T19:26:06Z (GMT). No. of bitstreams: 0 Previous issue date: 2010-01-15Bitstream added on 2014-06-13T20:54:07Z : No. of bitstreams: 1 gaviolli_vrn_me_jabo.pdf: 597637 bytes, checksum: 970dfc2e9d73c9379c7a724eab7ac345 (MD5) / Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) / Os objetivos deste trabalho foram identificar as funções matemáticas que melhor se ajustam para descrever o crescimento de fêmeas da raça Canchim e do grupo genético MA (filhos de touros da raça Charolesa e vacas ½ Canchim + ½ Nelore), e estimar a herdabilidade dos parâmetros A e k da curva que melhor se ajustou aos dados e a correlação genética desses com a idade (IPP) e o peso (PPP) ao primeiro parto e idade (ISP) e peso (PSP) ao segundo parto. As funções não lineares Brody, Richards, Von Bertalanffy e duas alternativas de Gompertz e de Logístico foram ajustadas pelo método de Gauss Newton e procedimento NLIN do SAS, a dados peso-idade de 1.923 bovinos da raça Canchim e MA, nascidos de 1972 a 2006, com pesagens no mínimo até os 37 meses de idade e no máximo até os 100 meses de idade e com pelo menos sete pesagens. A comparação entre os modelos foi realizada com base na interpretação biológica dos parâmetros e nos avaliadores de qualidade de ajuste. A estimação dos parâmetros genéticos foi feita pelo método da máxima verossimilhança restrita, utilizando análises uni e bicaracterísticas com modelos que incluíram o efeito fixo de grupo de contemporâneos e o efeito aleatório aditivo direto. Os modelos Brody e Von Bertalanffy ajustaram aos dados peso-idade, enquanto que os demais não convergiram, sendo que o modelo Brody foi escolhido para a obtenção dos parâmetros A e k de cada animal. As estimativas de herdabilidade obtidas foram 0,32 ± 0,05; 0,12 ± 0,04; 0,12 ± 0,04; 0,43 ± 0,06; 0,19 ± 0,06 e 0,44 ± 0,06, para A, k, IPP, PPP, ISP e PSP, respectivamente. Estes resultados sugerem que é possível, mas difícil obter mudanças na curva de crescimento dos animais pela seleção. As estimativas de correlação genética dos parâmetros A e k com as idades e os pesos aos partos sugerem que a seleção para modificar A deve provocar mudanças no... / The objectives in this study were to identify the mathematical function that best fits the growth of Canchim and MA (offspring of Charolais bulls and ½ Canchim + ½ Nelore cows) females, and to estimate the heritability of parameters A and k of the function that best fitted the data and the genetic correlations of these parameters with age (IPP) and weight (PPP) at first calving and age (ISP) and weight (PSP) at second calving. Brody, Richards, Von Bertalanffy and two alternatives of Gompertz and of Logistic non-linear models were fitted by Gauss Newton method and NLIN procedure of SAS. Weight-age data of 1,923 females born from 1972 to 2006, with weighing at least at 37 months of age and at most at 100 months of age, and with at least seven weighing were used. The comparison between the models was based on the biological interpretation of parameters and measures of quality of adjustment. The genetic parameters were estimated by the restricted maximum likelihood method using one and two-trait analyses with animal models that included the fixed effect of contemporary group and the random additive direct effect. The Brody and the Von Bertalanffy models adjusted the weight-age data well, while the others did not converge, and the Brody function was chosen to obtain the parameters A and k of each animal. The heritability estimates were 0.32 ± 0.05, 0.12 ± 0.04, 0.12 ± 0.04, 0.43 ± 0.06, 0.19 ± 0.06 and 0.44 ± 0.06 for A, k, IPP, PPP, ISP and PSP were, respectively. These values suggest that it is possible, but difficult to obtain changes in the pattern of the growth curve of the animals by selection. Estimates of genetic correlation of parameters A and k with the other traits suggest that selection to change A should cause changes in the same direction in PPI, PPP, ISP and PSP, and that changes in k, if they occur, should also be followed by changes in the other traits, but in the opposite direction

[pt] O objetivo principal desta tese introduzir um modelo estruturado por árvores que combina aspectos de duas metodologias: CART (Classification and Regression Tree) e STR (Smooth Transition Regression). O modelo aqui denominado STR-Tree. A idéia especificar um modelo não-linear paramétrico através da estrutura de uma árvore de decisão binária. O modelo resultante pode ser analisado como uma regressão com transição suave entre múltiplos regimes. As decisões sobre as divisões dos nós são inteiramente baseadas em testes do tipo Multiplicadores de Lagrange. Uma especificação alternativa baseada em validação cruzada também utilizada. Um experimento de Monte Carlo utilizado para avaliar o desempenho da metodologia proposta comparando-a com outras técnicas comumente utilizadas. Como resultado verifica-se que o modelo STR- Tree supera o tradicional CART quando seleciona a arquitetura de árvores simuladas. Além do mais, utilizar testes do tipo Multiplicadores de Lagrange gera resultados melhores do que procedimentos de validação cruzada. Quando foram utilizadas bases de dados reais, o modelo STR-Tree demonstrou habilidade preditiva superior ao CART. Através de uma aplicação, extende-se a metodologia para a análise de séries temporais. Neste caso, o modelo denominado STAR- Tree, sendo obtido através de uma árvore de decisão binária que ajusta modelos autoregressivos de primeira ordem nos regimes. A série de retornos da taxa de câmbio Euro/Dólar foi modelada e a capacidade preditiva e o desempenho financeiro do modelo foi comparado com metodologias padrões como previsões ingênuas e modelos ARMA. Como resultado obtido um modelo parcimonioso que apresenta desempenho estatístico equivalente às estratégias convencionais, porém obtendo resultados financeiros superiores. / [en] He main goal of this Thesis is to introduce a tree- structured model that combines aspects from two methodologies: CART (Classification and Regression Trees) and STR (Smooth Transition Regression). The model is called STR-Tree, The idea is to specify a nonlinear parametric model through the structure of a binary decision tree. The resulting modelo can be analyzed as a smooth transition regression model with multiple regimes. The decisions for splitting the nodes of the tree are entirely based on Lagrange Multipliers tests. An alternative specification that uses cross- validation is also tried. A Monte Carlo Experiment is used to evaluate the performance of the proposed methodology and to compare with other techniques that are commonly used. The results showed that the STRTree model outperformed the traditional CART when specifying the architecture of a simulated tree. Moreover, the use of Lagrange Multipliers tests gave better results than a cross-validation procedure. After applying the model to real datasets, it could be seen that STR-Tree showed superior predictive ability when compared to CART. The idea was extended to time series analysis through an application. In this situation, we call the model as STAR- Tree which is obtained through a binary decision tree that fits first-order autoregressive models for different regimes. The model was fitted to the returns of Euro/Dolar exchange rate time series and then evaluated statistically and financially. Comparing with the naive approach and ARMA methodology, the STAR-Tree was parsimonious and presented statistical performance equivalent to others. The financial results were better than the others.

Após o ajuste individual das 55 vacas estudadas pelo modelo Gompertz difá- sico com estrutura de erros autorregressiva de ordem 1 (totalizando 7 parâmetros), notou-se que apenas 6 vacas tinham problemas nas estimativas de seus parâmetros (não convergentes ou não signicativos), dessa forma continuou-se o trabalho proposto com 49 animais. Com as estimativas de cada um dos parâmetros (variáveis nessa etapa) foi realizada a análise de componentes principais e observação do gráco biplot, sendo possível a constatação de que 2 dos parâmetros do modelo continham informações ambíguas com pelo menos um dos demais parâmetros e estes foram retirados da análise, restando 5 parâmetros para o estudo. A análise de componentes principais foi realizada novamente apenas com os 5 parâmetros restantes e os três primeiros componentes principais (escolhidos pelo critério da percentagem de variância original explicada) foram utilizados como variáveis em um processo de agrupamento hierárquico. Após a realização da análise de agrupamentos, observou-se que 5 grupos homogêneos de animais foram formados, cada um com caraterísticas distintas. Desta forma, foi possível identicar animais que se destacavam, positiva ou negativamente, no que tange ao seu peso assintótico e taxa de crescimento. / After individual adjustment of the 55 cows studied using the diphasic Gompertz model with autoregressive structure of errors (totalizing 7 parameters), it was noted that only 6 cows had problems on estimates of the parameters (not converged or not signicant), then the proposed work continued with 49 animals. With each of the parameters estimates (variables at this stage) was performed a principal component analysis and observation of the biplot, and it was possible to nd that two of the model parameters contained ambiguous information with at least one of the other parameters, then these 2 parameters were removed from the analysis, leaving 5 parameters for the study. The principal component analysis was performed again with only ve remaining parameters and the rst three principal components (chosen by the criterion of percentage of original explained variance) were used as variables in a process of hierarchical clustering. After performing the cluster analysis, we found that ve homogeneous groups of animals were formed, each with distinct characteristics. Thus, it was possible to identify animals that stood out, positively or negatively, in terms of their asymptotic weight and growth rate.

Supervised learning of piecewise linear models is a well studied problem in machine learning community. The key idea in piecewise linear modeling is to properly partition the input space and learn a linear model for every partition. Decision trees and regression trees are classic examples of piecewise linear models for classification and regression problems. The existing approaches for learning decision/regression trees can be broadly classified in to two classes, namely, fixed structure approaches and greedy approaches. In the fixed structure approaches, tree structure is fixed before hand by fixing the number of non leaf nodes, height of the tree and paths from root node to every leaf node of the tree. Mixture of experts and hierarchical mixture of experts are examples of fixed structure approaches for learning piecewise linear models. Parameters of the models are found using, e.g., maximum likelihood estimation, for which expectation maximization(EM) algorithm can be used. Fixed structure piecewise linear models can also be learnt using risk minimization under an appropriate loss function. Learning an optimal decision tree using fixed structure approach is a hard problem. Constructing an optimal binary decision tree is known to be NP Complete. On the other hand, greedy approaches do not assume any parametric form or any fixed structure for the decision tree classifier. Most of the greedy approaches learn tree structured piecewise linear models in a top down fashion. These are built by binary or multi-way recursive partitioning of the input space. The main issues in top down decision tree induction is to choose an appropriate objective function to rate the split rules. The objective function should be easy to optimize. Top-down decision trees are easy to implement and understand, but there are no optimality guarantees due to their greedy nature. Regression trees are built in the similar way as decision trees. In regression trees, every leaf node is associated with a linear regression function. All piece wise linear modeling techniques deal with two main tasks, namely, partitioning of the input space and learning a linear model for every partition. However, Partitioning of the input space and learning linear models for different partitions are not independent problems. Simultaneous optimal estimation of partitions and learning linear models for every partition, is a combinatorial problem and hence computationally hard. However, piecewise linear models provide better insights in to the classification or regression problem by giving explicit representation of the structure in the data. The information captured by piecewise linear models can be summarized in terms of simple rules, so that, they can be used to analyze the properties of the domain from which the data originates. These properties make piecewise linear models, like decision trees and regression trees, extremely useful in many data mining applications and place them among top data mining algorithms. In this thesis, we address the problem of supervised learning of piecewise linear models for classification and regression. We propose novel algorithms for learning piecewise linear classifiers and regression functions. We also address the problem of noise tolerant learning of classifiers in presence of label noise. We propose a novel algorithm for learning polyhedral classifiers which are the simplest form of piecewise linear classifiers. Polyhedral classifiers are useful when points of positive class fall inside a convex region and all the negative class points are distributed outside the convex region. Then the region of positive class can be well approximated by a simple polyhedral set. The key challenge in optimally learning a fixed structure polyhedral classifier is to identify sub problems, where each sub problem is a linear classification problem. This is a hard problem and identifying polyhedral separability is known to be NP complete. The goal of any polyhedral learning algorithm is to efficiently handle underlying combinatorial problem while achieving good classification accuracy. Existing methods for learning a fixed structure polyhedral classifier are based on solving non convex constrained optimization problems. These approaches do not efficiently handle the combinatorial aspect of the problem and are computationally expensive. We propose a method of model based estimation of posterior class probability to learn polyhedral classifiers. We solve an unconstrained optimization problem using a simple two step algorithm (similar to EM algorithm) to find the model parameters. To the best of our knowledge, this is the first attempt to form an unconstrained optimization problem for learning polyhedral classifiers. We then modify our algorithm to find the number of required hyperplanes also automatically. We experimentally show that our approach is better than the existing polyhedral learning algorithms in terms of training time, performance and the complexity. Most often, class conditional densities are multimodal. In such cases, each class region may be represented as a union of polyhedral regions and hence a single polyhedral classifier is not sufficient. To handle such situation, a generic decision tree is required. Learning optimal fixed structure decision tree is a computationally hard problem. On the other hand, top-down decision trees have no optimality guarantees due to the greedy nature. However, top-down decision tree approaches are widely used as they are versatile and easy to implement. Most of the existing top-down decision tree algorithms (CART,OC1,C4.5, etc.) use impurity measures to assess the goodness of hyper planes at each node of the tree. These measures do not properly capture the geometric structures in the data. We propose a novel decision tree algorithm that ,at each node, selects hyperplanes based on an objective function which takes into consideration geometric structure of the class regions. The resulting optimization problem turns out to be a generalized eigen value problem and hence is efficiently solved. We show through empirical studies that our approach leads to smaller size trees and better performance compared to other top-down decision tree approaches. We also provide some theoretical justification for the proposed method of learning decision trees. Piecewise linear regression is similar to the corresponding classification problem. For example, in regression trees, each leaf node is associated with a linear regression model. Thus the problem is once again that of (simultaneous) estimation of optimal partitions and learning a linear model for each partition. Regression trees, hinge hyperplane method, mixture of experts are some of the approaches to learn continuous piecewise linear regression models. Many of these algorithms are computationally intensive. We present a method of learning piecewise linear regression model which is computationally simple and is capable of learning discontinuous functions as well. The method is based on the idea of K plane regression that can identify a set of linear models given the training data. K plane regression is a simple algorithm motivated by the philosophy of k means clustering. However this simple algorithm has several problems. It does not give a model function so that we can predict the target value for any given input. Also, it is very sensitive to noise. We propose a modified K plane regression algorithm which can learn continuous as well as discontinuous functions. The proposed algorithm still retains the spirit of k means algorithm and after every iteration it improves the objective function. The proposed method learns a proper Piece wise linear model that can be used for prediction. The algorithm is also more robust to additive noise than K plane regression. While learning classifiers, one normally assumes that the class labels in the training data set are noise free. However, in many applications like Spam filtering, text classification etc., the training data can be mislabeled due to subjective errors. In such cases, the standard learning algorithms (SVM, Adaboost, decision trees etc.) start over fitting on the noisy points and lead to poor test accuracy. Thus analyzing the vulnerabilities of classifiers to label noise has recently attracted growing interest from the machine learning community. The existing noise tolerant learning approaches first try to identify the noisy points and then learn classifier on remaining points. In this thesis, we address the issue of developing learning algorithms which are inherently noise tolerant. An algorithm is inherently noise tolerant if, the classifier it learns with noisy samples would have the same performance on test data as that learnt from noise free samples. Algorithms having such robustness (under suitable assumption on the noise) are attractive for learning with noisy samples. Here, we consider non uniform label noise which is a generic noise model. In non uniform label noise, the probability of the class label for an example being incorrect, is a function of the feature vector of the example.(We assume that this probability is less than 0.5 for all feature vectors.) This can account for most cases of noisy data sets. There is no provably optimal algorithm for learning noise tolerant classifiers in presence of non uniform label noise. We propose a novel characterization of noise tolerance of an algorithm. We analyze noise tolerance properties of risk minimization frame work as risk minimization is a common strategy for classifier learning. We show that risk minimization under 01 loss has the best noise tolerance properties. None of the other convex loss functions have such noise tolerance properties. Empirical risk minimization under 01 loss is a hard problem as 01 loss function is not differentiable. We propose a gradient free stochastic optimization technique to minimize risk under 01 loss function for noise tolerant learning of linear classifiers. We show (under some conditions) that the algorithm converges asymptotically to the global minima of the risk under 01 loss function. We illustrate the noise tolerance of our algorithm through simulations experiments. We demonstrate the noise tolerance of the algorithm through simulations.

Orientadores: Victor Hugo Lachos Dávila, Filidor Edilfonso Vilca Labra / Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Matemática, Estatística e Computação Científica / Made available in DSpace on 2018-08-16T04:06:26Z (GMT). No. of bitstreams: 1 MedinaGaray_AldoWilliam_M.pdf: 1389516 bytes, checksum: 2763869ea52e11ede3c860714ea0e75e (MD5) Previous issue date: 2010 / Resumo: Neste trabalho estudamos alguns aspectos de estimação e diagnóstico de influência global e local de modelos não lineares sob a classe de distribuição misturas da escala skew-normal, baseado na metodologia proposta por Cook (1986) e Poon & Poon (1999). Os modelos não lineares heteroscedásticos também são discutidos. Esta nova classe de modelos constitui uma generalização robusta dos modelos de regressão não linear simétricos, que têm como membros particulares distribuições com caudas pesadas, tais como skew-t, skew-slash, skew-normal contaminada, entre outras. A estimação dos parâmetros será obtida via o algoritmo EM proposto por Dempster et al. (1977). Estudos de testes de hipóteses são considerados utilizando as estatísticas de escore e da razão de verossimilhança, para testar a homogeneidade do parâmetro de escala. Propriedades das estatísticas do teste são investigadas através de simulações de Monte Carlo. Exemplos numéricos considerando dados reais e simulados são apresentados para ilustrar a metodologia desenvolvida / Abstrac: In this work, we studied some aspects of estimation and diagnostics on the global and local influence in nonlinear models under the class of scale mixtures of the skewnormal (SMSN) distribution, based on the methodology proposed by Cook (1986) e Poon & Poon (1999). Heteroscedastic nonlinear models are also discussed. This new class of models are a robust generalization of non-linear regression symmetrical models, which have as members individual distributions with heavy tails, such as skew-t, skew-slash, and skew-contaminated normal, among others. The parameter estimation will be obtained with the EM algorithm proposed by Dempster et al. (1977). Studies testing hypotheses are considered using the score statistics and the likelihood ratio test to test the homogeneity of scale parameter. Properties of test statistics are investigated through Monte Carlo simulations. Numerical examples considering real and simulated data are presented to illustrate the methodology / Mestrado / Métodos Estatísticos / Mestre em Estatística

[pt] Nesta tese apresenta-se o Modelo de Redes Neurais Globais- Locais (RNGL) dentro do contexto de modelos de séries temporais. Esta formulação abrange alguns modelos não- lineares já existentes e admite também o enfoque de Mistura de Especialistas. Dedica-se especial atenção ao caso de especialistas lineares, e são discutidos extensivamente aspectos teóricos do modelo: condições de estacionariedade, identificabilidade do modelo, existência, consistência e normalidade assintótica dos estimadores dos parâmetros. Considera-se também uma estratégia de construção do modelo e são discutidos os procedimentos numéricos de estimação, apresentando uma solução para o cálculo de valores iniciais. Finalmente, ilustra-se a metodologia apresentada em duas séries temporais reais, amplamente utilizada na literatura de modelos não lineares. / [en] In this thesis, the Local Global Neural Networks model is proposed within the context of time series models. This formulation encompasses some already existing nonlinear models and also admits the Mixture of Experts approach. We place emphasis on the linear expert case and extensively discuss the theoretical aspects of the model: stationary conditions, existence, consistency and asymptotic normality of the parameter estimates, and model identifiability. A model building strategy is also considered and the whole procedure is illustrated with two real time-series.

Inicialmente, foram analisados dados peso-idade do nascimento até 100 meses de idade, de vacas de quatro grupos genéticos (G): Nelore (NEL), ½Canchim + ½Nelore (CN), ½Angus + ½Nelore (AN) e ½Simental + ½Nelore (SN), pertencentes a Embrapa Pecuária Sudeste, São Carlos. Os animais considerados neste estudo nasceram de 1998 a 2001 (Ano), na primavera e outono (EP), e foram criados em sistema de produção intensiva, recebendo três níveis de suplementação pós-desmama (M): 0, 1,5; 3,0 kg/animal/dia de concentrado. O objetivo neste estudo foi comparar diferentes modelos não-lineares para estimar o crescimento, e avaliar a influência de efeitos de ambiente e grupo genético sobre os parâmetros estimados. Os modelos não-lineares: Brody, Gompertz, Logístico, Von Bertalanffy e Richards foram ajustados por mínimos quadrados ordinários e ponderados, considerando a variância normal e ponderada pelo inverso dos pesos em diferentes períodos. Foi usado o procedimento NLIN do SAS. Os modelos Brody e Von Bertalanffy convergiram para todos os G, havendo, entretanto, leve superioridade do Brody ponderado. Na comparação do ajuste dos modelos considerando o uso do inverso da variância os modelos mostraram-se mais adequado. As estimativas dos parâmetros peso assintótico (A) e taxa de maturidade (k) do modelo de Brody ponderado foram analisadas por meio de modelo que, além do efeito médio global, incluiu os efeitos de G, M, EP e as interações entre estes efeitos. Houve diferenças significativas das curvas de crescimento médias para os G. Na análise individual dos parâmetros A e k estimados através do modelo Brody ponderado, verificou-se que A foi influenciado (P<0,05) por G e EP e k foi influenciado (P<0,05) por M, fornecidos aos animais durante quatro meses após desmama. Melhorias no manejo alimentar resultaram em menor variação na forma das curvas de crescimentos e em altas taxas de maturidade. Na segunda parte, verificou-se a qualidade do modelo Brody, ponderado pelo inverso das variâncias dos pesos, quanto ao ajuste peso-idade como também a influência das estimativas do peso à maturidade (A) e da taxa de maturidade (k) sobre características produtivas das vacas NEL, CN, AN e SN. Foram organizados 10 grupos contemporâneos (GC), com concatenação dos efeitos Ano-EP-M para cada G. Utilizando-se um modelo misto com efeitos de G e GC, foi incluído, alternadamente covariáveis linear e quadrática de A e k, na análise das características produtivas: peso à desmama dos bezerros (PD); número (ND8) e kg (KD8) de bezerros desmamados em até 8 anos de permanência da vaca no rebanho; relação PD/peso da vaca ao parto (PD_PVP); relação PD/peso da vaca à desmama do bezerro (PD_PVD); relação PD/unidade metabólica da vaca (PV0,75) à desmama do bezerro (PDW). Houve diferença significativa (P<0,05) da curva de crescimento entre os grupos genéticos (G) e também entre os grupos de contemporâneos (GC) dentro de G. Verificou-se que estas características foram, em geral, influenciadas (P<0,01) tanto pelos efeitos linear e quadrático de A quanto pelos efeitos linear e quadrático de k. / Initially, were analyzed weight-age data from birth to 100 months of age from cows of four genetic groups (G): Nellore (NEL), ½Nellore + ½Canchim (CN), ½Angus +½Nellore (AN) and ½Simmental + ½Nellore (SN), of a experiment carried out at Embrapa Southeast Cattle Research Center, State of São Paulo, Brazil. The animals considered in this study were born from 1998 to 2001 (Year) in spring and fall (EP), and were managed in intensive production system and submitted to three of levels of supplementation post-weaning (M): 0, 1.5 and 3.0 kg/animal/day of concentrate. The objective of this study was to compare different nonlinear models to fitted growth curves, of beef cattle females, and to evaluate of environmental and genetic group effects on the estimated parameters. The nonlinear models: Brody, Gompertz, Logistic, Von Bertalanffy and Richards were fitted by ordinary least squares and weighted by the inverse of the variances of the weights in different periods. It was used the NLIN procedure of SAS. The parameters asymptotic weight (A) and maturing rate (k) obtained from model of Brody were analyzed by a mixed linear model that, besides the overall mean effect, included the effects of G, M, EP, and the interactions among these effects. The Brody and Von Bertalanffy models converged for all genetic groups, although slight superiority of the weighted Brody. Comparing the goodness of fit of these models, the use of the inverse of variances showed more efficient than the adjust of the models considering normal variances. Individual analysis of A and k estimated the model weighted Brody, the A parameter was influenced (P <0.05) by genetic group and season of birth and k was influenced (P <0 05) for levels of supplementation to the animals. Improvements in feeding supplementation resulted in less variation in the shape of growth curves and rates of maturity. In the second part of the work, it was evaluated the goodness of the Brody model, weighted by the inverse variance weights, in the adjust of weight-age data, and also analyzed the influence of the maturity weight (A) and maturing rate (k) estimates for traits cows productivity. Were organized 10 contemporary groups (CG) with concatenation of effects Year- EP-M for each G. Considering a mixed model with effects of G and CG (10 contemporaneous groups organized by concatenation Year-EP-M effects), linear and quadratic covariate effects of A and k, were added, alternately, for the analysis of the following traits: weaning weight of calve (WW), number (NW8) and kg (KW8) of calves weaned over 8 years of the cow in the herd; WW/weight of the cow at calving (WW_WC); WW/cow weight at weaning of calf (WW_WWC); and WW/metabolic unit of the cow (PV0,75) at weaning of the calf (MW). There was significant difference (P<0.05) of the growth curve among the genetic groups and also among contemporary groups within G. It was found that the production traits were, in general, influenced (P<0.01) by both linear and quadratic effects of A and k.

[pt] Os mercados de commodities tornaram-se uma nova alternativa para investidores nos últimos quinze anos, em um processo conhecido como financeirização dos mercados de commodities. Vários estudos têm explicado as razões deste fenômeno, porém esta é uma questão ainda pouco estudada na literatura de economia agrícola e energética no Brasil. Como a financeirização do mercado de commodities mudou a dinâmica dos preços ao longo dos anos? Esta tese aplica modelos não lineares para entender se a especulação causou os movimentos de preços nos mercados de commodities agrícolas, bem como para investigar a descoberta de preços no mercado brasileiro ao se testar os mecanismos de transmissão dos preços internacionais de energia e commdities agrícolas aos preços brasileiros de etanol e gasolina. Procuramos investigar com os mesmos modelos não lineares os efeitos de transbordamento dos mercados globais de futuros para os preços à vista locais. Por fim, analisa-se o aumento da liquidez nos mercados de commodities, desenvolvemos para tanto uma nova medida para compreender o grau de ambiguidade dos preços de 12 commodities agrícolas. Apesar dos testes econométricos, os resultados foram inconclusivos sobre o papel da especulação no impacto dos retornos dos preços das commodities. Existe um nexo entre os preços internacionais do petróleo e do etanol brasileiro, e os preços globais das commodities aumentaram os efeitos de contágio nos mercados spot brasileiros. Finalmente, a financeirização dos mercados de commodities aumentou a liquidez do mercado medida pelo grau de ambiguidade. Esta tese contribui para o campo ao aplicar abordagens econométricas robustas e inovadoras, bem como ao evidenciar como o price discovery e o risk-sharing afetam a dinâmica dos preços das commodities. / [en] Commodity markets have become a new investment alternative for portfolio investors over the last fifteen years, in a process known as the financialization of commodity markets. Several studies have explained the reasons for this phenomenon (e.g., speculation and increase in biofuels production), leading to a question largely understudied in agricultural and energy economics literature. How has the financialization of the commodities market changed the price dynamic over the years? This thesis applies nonlinear models to understand whether the speculation caused the price movements in the agricultural commodity markets; investigates the price discovery in the Brazilian market by analyzing the transmission of international energy and feedstocks prices to Brazilian ethanol and gasoline prices; and investigates the spillover effects from global futures markets to local spot prices. In addition, it analyzes the increased liquidity in the commodity markets by developing a new measurement to gauge the degree of ambiguity for 12 agricultural commodities prices. Despite the robust econometric tests performed, the findings were inconclusive on the role of speculation in impacting the price returns of commodities. It also found that there exists a nexus between international oil and Brazilian ethanol prices, and global commodities prices have increased the spillover effects on the Brazilian spot markets. Finally, the financialization of commodity markets has increased the liquidity in the market as measured by the degree of ambiguity. This thesis contributes to the field not only by applying more robust, novel econometric approaches but also by evidencing how information discovery and risk-sharing affect the commodity price dynamics.