This thesis investigates the factors that are associated with annual changes in the share price of Johannesburg Stock Exchange (JSE) listed companies. In this study, an increase in value of a share is when the share price of a company goes up by the end of the financial year as compared to the previous year. Secondary data that was sourced from McGregor BFA website was used. The data was from 2004 up to 2011. Deciding which share to buy is the biggest challenge faced by both investment companies and individuals when investing on the stock exchange. This thesis uses binary logistic regression to identify the variables that are associated with share price increase. The dependent variable was annual change in share price (ACSP) and the independent variables were assets per capital employed ratio, debt per assets ratio, debt per equity ratio, dividend yield, earnings per share, earnings yield, operating profit margin, price earnings ratio, return on assets, return on equity and return on capital employed. Different variable selection methods were used and it was established that the backward elimination method produced the best model. It was established that the probability of success of a share is higher if the shareholders are anticipating a higher return on capital employed, and high earnings/ share. It was however, noted that the share price is negatively impacted by dividend yield and earnings yield. Since the odds of an increase in share price is higher if there is a higher return on capital employed and high earning per share, investors and investment companies are encouraged to choose companies with high earnings per share and the best returns on capital employed. The final model had a classification rate of 68.3% and the validation sample produced a classification rate of 65.2% / Mathematical Sciences / M.Sc. (Statistics)

Para analisar características de interesse a respeito de um conjunto de dados reais da área de Odontologia apresentado em Hadgu & Koch (1999), ajustaremos um modelo de regressão linear multivariado com erros nas variáveis com intercepto nulo. Este conjunto de dados é caracterizado por medições de placa bacteriana em três grupos de voluntários, antes e após utilizar dois líquidos de bochecho experimentais e um líquido de bochecho controle, com medições (sujeitas a erros de medição) no início do estudo, após três e seis meses de utilização dos líquidos. Neste caso, uma possível estrutura de dependência entre as medições feitas em um mesmo indivíduo deve ser incorporada ao modelo e, além disto, temos duas variáveis resposta para cada indivíduo. Após a apresentação do modelo estatístico, iremos obter estimativas de máxima verossimilhança dos parâmetros utilizando o algoritmo iterativo EM e testaremos as hipóteses de interesse utilizando testes assintóticos de Wald, razão de verossimilhanças e score. Como neste caso não existe um teste ótimo, faremos um estudo de simulação para verificar o comportamento das três estatísticas de teste em relação a diferentes tamanhos amostrais e diferentes valores de parâmetros. Finalmente, faremos um estudo de diagnóstico buscando identificar possíveis pontos influentes no modelo, considerando o enfoque de influência local proposto por Cook (1986) e a medida de curvatura normal conformal desenvolvida por Poon & Poon (1999). / To analyze some characteristics of interest in a real odontological data set presented in Hadgu & Koch (1999), we propose the use of a multivariate null intercept errors-in-variables regression model. This data set is composed by measurements of dental plaque index (with measurement errors), which were measured in volunteers who were randomized to two experimental mouth rinses (A and B) or a control mouth rinse. The measurements were taken in each individual, before and after the use of the respective mouth rinses, in the beginning of the study, after three months from the baseline and after six months from the baseline. In this case, a possible structure of dependency between the measurements taken within the same individual must be incorporated in the model. After presenting the statistical model, we obtain the maximum likelihood estimates of the parameters using the numerical algorithm EM, and we test the hypotheses of interest considering asymptotic tests (Wald, likelihood ratio and score). Also, a simulation study to verify the behavior of these three test statistics is presented, considering diferent sample sizes and diferent values for the parameters. Finally, we make a diagnostic study to identify possible influential observations in the model, considering the local influence approach proposed by Cook (1986) and the conformal normal curvature proposed by Poon & Poon (1999).

[pt] Esta dissertação descreve a aplicação de Modelos Lineares Generalizados (MLGs) à análise de um experimento visando identificar a combinação dos níveis das variáveis independentes: concentração de hidróxido de sódio (A), volume de hipoclorito de sódio (B) e sua interação (AB), que minimiza a variável resposta: proporção de itens com defeitos, em um processo de beneficiamento numa indústria têxtil de pequeno porte. A variável resposta encontra-se na forma de proporção, violando os pressupostos básicos do Modelo Linear Clássico e com isso as estimativas dos coeficientes pelo método de Mínimos Quadrados Ordinários (MQO) é menos confiável. O planejamento utilizado foi o fatorial completo 22 com ponto central e replicado. Após o planejamento, a modelagem pelo MLG é aplicada, só então é possível identificar uma subdispersão dos dados, verificar que o modelo empregado está correto e que o volume de hipoclorito de sódio (B) é o único fator significativo, no processo de alvejamento industrial da empresa. Portanto, como a finalidade é minimizar a resposta, utiliza-se o nível inferior (-1) desta variável. Consequentemente, como o intuito é reduzir os custos com insumos químicos pode-se utilizar o nível mínimo da concentração de hidróxido de sódio (A) e o nível máximo da interação entre os fatores (AB), já que eles não são significativos ao modelo. / [en] This dissertation describes the application of Generalized Linear Models (GLMs) to the analysis of an experiment with the purpose identify the levels combination of independent variables: concentration of sodium hydroxide (A) volume of sodium hypochlorite (B) and their interaction (AB), that minimizes the response variable: proportion of defective items, in a process in a small plant of the textile industry. The response variable takes the form of a proportion, that violates the basic assumptions of the Classic Linear Model and, as a result, the estimates of the coefficients by Ordinary Least Squares method is less reliable. The design employed was a replicated complete 22 factorial design with central point. After doing the planning, the modeling by MLG is applied, and then it is possible to identify a underdispersion data; to verify that the model used is correct and that the volume of sodium hypochlorite (B) is the only significant factor in the industrial process of bleaching the company. Therefore, as the purpose is to minimize the response, it is used the lower level (-1) of this variable. Consequently, as the aim is to reduce costs of chemical inputs can use the minimum level of concentration of hydroxide sodium (A) and the maximum level of interaction between factors (AB), since they are not significant to the model.

Η παρούσα διατριβή ασχολείται με δύο ιδιαιτέρως σημαντικά και διαχρονικά επίκαιρα ζητήματα στην ανάλυση χρονολογικών σειρών, τα οποία εντάσσονται, υπό ευρεία έννοια, στο πεδίο της Μακροοικονομετρίας. Ειδικότερα, μελετώνται θέματα και μεθοδολογίες ή τεχνικές ιδιαίτερα χρήσιμες για εκείνους τους ερευνητές, οι οποίοι επικεντρώνονται στην ανάλυση της συμπεριφοράς των συναθροιστικών (aggregate) μεγεθών της οικονομίας, βασιζόμενοι στη χρήση δεδομένων χρονοσειρών ή πιο απλά χρονοσειρές (time series). Το πρώτο ζήτημα αφορά στη μελέτη της δυναμικής αλληλεξάρτησης ανάμεσα σε μακροοικονομικές μεταβλητές κάτω από την υιοθέτηση ενός πολλαπλού πλαισίου ανάλυσης χρονοσειρών. Το ενδιαφέρον εστιάζεται κυρίως στην γενικευμένη ή εκτεταμένη έννοια της αιτιότητας κατά Granger, δηλαδή στην επέκταση της τυπικής έννοιας της αιτιότητας κατά Granger σε μεγαλύτερους του ενός ή σε πολλαπλούς ορίζοντες πρόβλεψης. Το δεύτερο ζήτημα αφορά στην παρουσία μη-γραμμικών χαρακτηριστικών σε μακροοικονομικές χρονοσειρές, καθώς και την υποδειγματοποίηση της μη-γραμμικότητας με τη χρήση μη-γραμμικών οικονομετρικών μοντέλων. Επικεντρώνεται δε ιδιαίτερα στον έλεγχο μοναδιαίας ρίζας κάτω από την εναλλακτική υπόθεση της στασιμότητας γύρω από μη-γραμμικές τάσεις της μορφής τάσεων ομαλής μετάβασης (smooth transition trends) στις μακροοικονομικές χρονοσειρές. Ουσιαστικά, η διατριβή διακρίνεται σε δύο κεφάλαια. Στο Κεφάλαιο 1 παρουσιάζεται η τυπική έννοια της αιτιότητας κατά Granger, καθώς και η γενικευμένη ή εκτεταμένη έννοια της αιτιότητας ή η αιτιότητα σε πολλαπλούς ορίζοντες (multi-horizon causality), στο πλαίσιο των διανυσματικών αυτοπαλίνδρομων υποδειγμάτων (VAR). Η τυπική έννοια της αιτιότητας κατά Granger περιορίζεται στη βελτίωση της προβλεψιμότητας σε ορίζοντα πρόβλεψης μίας περιόδου (one-step ahead), ενώ λαμβάνει υπ'όψιν μόνο τις άμεσες ροές πληροφόρησης μεταξύ των μεταβλητών ενδιαφέροντος (direct causality). Ωστόσο, σε υποδείγματα VAR με περισσότερες από δύο μεταβλητές η τυπική έννοια της αιτιότητας μπορεί να επεκταθεί με την μελέτη της βελτίωσης της προβλεψιμότητας σε μεγαλύτερους του ενός ορίζοντες πρόβλεψης. Σε μία περίπτωση όπως η τελευταία, πλην της άμεσης αιτιότητας, δύνανται να μελετηθούν και οι έμμεσες σχέσεις αιτιότητας (indirect causality) που ενδέχεται να προκύψουν μέσω των πρόσθετων μεταβλητών του συστήματος. Το θεωρητικό πλαίσιο της γενικευμένης έννοιας της αιτιότητας που παρουσιάζει η παρούσα διατριβή έχει αναπτυχθεί από τους Dufour and Renault (1998). Παράλληλα, δίνεται ιδιαίτερη βαρύτητα σε δύο πρόσφατες μεθόδους στατιστικής επαγωγής αιτιωδών σχέσεων κατά Granger σε πολλαπλούς ορίζοντες, οι οποίες παρέχουν πρόσθετη πληροφόρηση σχετικά με τη δυναμική αλληλεξάρτηση οικονομικών χρονοσειρών, και πιο συγκεκριμένα σχετικά με τον άμεσο ή έμμεσο χαρακτήρα των αιτιωδών σχέσεων, το διαχωρισμό μεταξύ βραχυχρόνιας και μακροχρόνιας (μη)-αιτιότητας, καθώς και τις πιθανές χρονικές υστερήσεις της αιτιότητας. Τέλος, στα πλαίσια του Κεφαλαίου 1, ερευνάται η δυνατότητα εφαρμογής των μεθόδων αυτών μέσω εμπειρικών εφαρμογών πάνω σε δύο διαχρονικά ζητήματα αιτιωδών σχέσεων ανάμεσα σε οικονομικές μεταβλητές. Στο Κεφάλαιο 2 παρουσιάζονται υποδείγματα ομαλής μετάβασης, καθώς και έλεγχοι μοναδιαίας ρίζας οι οποίοι επιτρέπουν την στασιμότητα γύρω από ομαλές ή βαθμιαίες μεταβάσεις κάτω από την εναλλακτική υπόθεση. Κύριο χαρακτηριστικό των υποδειγμάτων ομαλής μετάβασης είναι η παρουσία μη-γραμμικών τάσεων στη διαχρονική εξέλιξη των χρονοσειρών. Κεντρικό ρόλο στα υποδείγματα αυτά κατέχουν οι διαρθρωτικές μεταβολές (structural changes) στην προσδιοριστική τάση, οι οποίες, δεδομένου ότι αντιπροσωπεύουν μεταβολές της συναθροιστικής συμπεριφοράς, υποδειγματοποιούνται με τη χρήση ενός προσδιοριστικού στοιχείου το οποίο επιτρέπει την βαθμιαία αντί της στιγμιαίας προσαρμογής. Οι έλεγχοι μοναδιαίας ρίζας, οι οποίοι επιτρέπουν περισσότερη ευελιξία στην συνάρτηση της τάσης σε σχέση με την γραμμική εξειδίκευση της προσδιοριστικής τάσης που χρησιμοποιούν οι τυπικοί έλεγχοι μοναδιαίας ρίζας, αποτελούν το επίκεντρο μελέτης του Κεφαλαίου 2 της διατριβής. Η αναγκαιότητα υιοθέτησης πρόσθετων ελέγχων μοναδιαίας ρίζας, όπως οι έλεγχοι μοναδιαίας ρίζας οι οποίοι επιτρέπουν στασιμότητα γύρω από ομαλές μεταβάσεις κάτω από την εναλλακτική υπόθεση, ισχυροποιείται από τα αποτελέσματα της εφαρμογής των ελέγχων αυτών σε ένα σύνολο οικονομικών χρονοσειρών. / This thesis discusses two central research topics in applied time series econometrics that generally belong in the field of Macroeconometrics. In particular, we investigate issues and methods which are of interest to those researchers who want to analyze economic problems or economic aggregates by means of time series data. The first topic deals with the dynamic interrelationships between sets of theory related variables in a multiple time series context. Research interest is primarily focused on the generalized or extended notion of Granger causality, that is the extension of the standard Granger causality concept to higher forecast horizons. The second topic deals with nonlinear behavior of macroeconomic time series, as well as the modelling of nonlinearities in economic time series using nonlinear econometric models. Specific attention is paid to unit root tests that allow stationarity around nonlinear trends in the form of smooth transitions under the alternative. The dissertation consists of two chapters. The first chapter presents the standard concept of Granger causality, along with the generalized or extended notion of causality, also known as multiple-horizon causality, in the vector autoregressive (VAR) framework. The standard notion of Granger causality restricts prediction improvement to a forecast horizon of one period, while it considers only direct flows of information between the variables of interest. However, in VAR models with more than two variables, the concept of standard Granger causality can be extended by studying prediction improvement at forecast horizons greater than one. If this is the case, then, except for direct causality, indirect flows of information might be revealed through the additional variables of the system. The theoretical framework of the extended concept of causality which is presented in the present dissertation has been developed by Dufour and Renault (1998). In addition, special attention is paid to two recent methods for testing hypothesis of non-causality at various horizons which can provide further information on the dynamic interaction of time series, and more specifically on the direct or indirect nature of causal effects, the distinction between short-run and long-run (non)-causality, as wells as the possibility of causal delays. Finally, the potential implementation of these methods is examined through empirical applications on causality relations among different sets of economic variables. Chapter 2 presents smooth transition (STR) trend models, as well as unit root tests that allow stationarity around smooth transitions under the alternative. Smooth transition regression models presume the presence of nonlinear trends in the long-run evolution of time series. A key feature of these models is the presence of structural changes in the deterministic trend which, given that they represent changes in aggregate behavior (economic aggregates), are modelled through a deterministic component that permits gradual rather than instantaneous adjustment between regimes. Unit root tests that permit a more versatile trend function in the unit root procedure, rather than the standard linear trends, are the main concern of Chapter 2. The necessity of employing additional unit root tests, such as unit root tests that allow stationarity around smooth transitions under the alternative, becomes evident through the unit root test results that are observed in an application in a set of economic time series.

Rieck & Nedelman (1991) propuseram um modelo de regressão log-linear tendo como base a distribuição Birnbaum-Saunders (Birnbaum & Saunders, 1969a). O modelo proposto pelos autores vem sendo bastante explorado e tem se mostrado uma ótima alternativa a outros modelos propostos na literatura, como por exemplo, os modelos de regressão Weibull, gama e lognormal. No entanto, até o presente momento, não existe nenhum estudo tratando de refinamentos para as estatísticas da razão de verossimilhanças e escore nesta classe de modelos de regressão. Assim, um dos objetivos desta tese é obter um fator de correção de Bartlett para a estatística da razão de verossimilhanças e um fator de correção tipo-Bartlett para a estatística escore nesse modelo. Estes ajustes melhoram a aproximação da distribuição nula destas estatísticas pela distribuição qui-quadrado de referência. Adicionalmente, objetiva-se obter ajustes para a estatística da razão de verossimilhanças sinalizada. Tais ajustes melhoram a aproximação desta estatística pela distribuição normal padrão. Recentemente, uma nova estatística de teste foi proposta por Terrell (2002), a qual o autor denomina estatística gradiente. Esta estatística foi derivada a partir da estatística escore e da estatística de Wald modificada (Hayakawa & Puri, 1985). A combinação daquelas duas estatísticas resulta em uma estatística muito simples de ser calculada, não envolvendo, por exemplo, nenhum cálculo matricial como produto e inversa de matrizes. Esta estatística foi recentemente citada por Rao (2005): \"The suggestion by Terrell is attractive as it is simple to compute. It would be of interest to investigate the performance of the [gradient] statistic.\" Caminhando na direção da sugestão de Rao, outro objetivo da tese é obter uma expansão assintótica para a distribuição da estatística gradiente sob uma sequência de alternativas de Pitman convergindo para a hipótese nula a uma taxa de convergência de n^{-1/2} utilizando a metodologia desenvolvida por Peers (1971) e Hayakawa (1975). Em particular, mostramos que, até ordem n^{-1/2}, a estatística gradiente segue distribuição qui-quadrado central sob a hipótese nula e distribuição qui-quadrado não central sob a hipótese alternativa. Também temos como objetivo comparar o poder local deste teste com o poder local dos testes da razão de verossimilhanças, de Wald e escore. Finalmente, aplicaremos a expansão assintótica derivada na tese em algumas classes particulares de modelos. / The Birnbaum-Saunders regression model is commonly used in reliability studies.We address the issue of performing inference in this class of models when the number of observations is small. Our simulation results suggest that the likelihood ratio and score tests tend to be liberal when the sample size is small. We derive Bartlett and Bartlett-type correction factors which reduce the size distortion of the tests. Additionally, we also consider modified signed log-likelihood ratio statistics in this class of models. Finally, the asymptotic expansion of the distribution of the gradient test statistic is derived for a composite hypothesis under a sequence of Pitman alternative hypotheses converging to the null hypothesis at rate n^{-1/2}, n being the sample size. Comparisons of the local powers of the gradient, likelihood ratio, Wald and score tests reveal no uniform superiority property.

Rieck & Nedelman (1991) propuseram um modelo de regressão log-linear tendo como base a distribuição Birnbaum-Saunders (Birnbaum & Saunders, 1969a). O modelo proposto pelos autores vem sendo bastante explorado e tem se mostrado uma ótima alternativa a outros modelos propostos na literatura, como por exemplo, os modelos de regressão Weibull, gama e lognormal. No entanto, até o presente momento, não existe nenhum estudo tratando de refinamentos para as estatísticas da razão de verossimilhanças e escore nesta classe de modelos de regressão. Assim, um dos objetivos desta tese é obter um fator de correção de Bartlett para a estatística da razão de verossimilhanças e um fator de correção tipo-Bartlett para a estatística escore nesse modelo. Estes ajustes melhoram a aproximação da distribuição nula destas estatísticas pela distribuição qui-quadrado de referência. Adicionalmente, objetiva-se obter ajustes para a estatística da razão de verossimilhanças sinalizada. Tais ajustes melhoram a aproximação desta estatística pela distribuição normal padrão. Recentemente, uma nova estatística de teste foi proposta por Terrell (2002), a qual o autor denomina estatística gradiente. Esta estatística foi derivada a partir da estatística escore e da estatística de Wald modificada (Hayakawa & Puri, 1985). A combinação daquelas duas estatísticas resulta em uma estatística muito simples de ser calculada, não envolvendo, por exemplo, nenhum cálculo matricial como produto e inversa de matrizes. Esta estatística foi recentemente citada por Rao (2005): \"The suggestion by Terrell is attractive as it is simple to compute. It would be of interest to investigate the performance of the [gradient] statistic.\" Caminhando na direção da sugestão de Rao, outro objetivo da tese é obter uma expansão assintótica para a distribuição da estatística gradiente sob uma sequência de alternativas de Pitman convergindo para a hipótese nula a uma taxa de convergência de n^{-1/2} utilizando a metodologia desenvolvida por Peers (1971) e Hayakawa (1975). Em particular, mostramos que, até ordem n^{-1/2}, a estatística gradiente segue distribuição qui-quadrado central sob a hipótese nula e distribuição qui-quadrado não central sob a hipótese alternativa. Também temos como objetivo comparar o poder local deste teste com o poder local dos testes da razão de verossimilhanças, de Wald e escore. Finalmente, aplicaremos a expansão assintótica derivada na tese em algumas classes particulares de modelos. / The Birnbaum-Saunders regression model is commonly used in reliability studies.We address the issue of performing inference in this class of models when the number of observations is small. Our simulation results suggest that the likelihood ratio and score tests tend to be liberal when the sample size is small. We derive Bartlett and Bartlett-type correction factors which reduce the size distortion of the tests. Additionally, we also consider modified signed log-likelihood ratio statistics in this class of models. Finally, the asymptotic expansion of the distribution of the gradient test statistic is derived for a composite hypothesis under a sequence of Pitman alternative hypotheses converging to the null hypothesis at rate n^{-1/2}, n being the sample size. Comparisons of the local powers of the gradient, likelihood ratio, Wald and score tests reveal no uniform superiority property.