Uso da análise discriminante regularizada (RDA) no reconhecimento de padrões em imagens digitais hiperespectral de sensoriamento remoto

Em cenas naturais, ocorrem com certa freqüência classes espectralmente muito similares, isto é, os vetores média são muito próximos. Em situações como esta, dados de baixa dimensionalidade (LandSat-TM, Spot) não permitem uma classificação acurada da cena. Por outro lado, sabe-se que dados em alta dimensionalidade [FUK 90] tornam possível a separação destas classes, desde que as matrizes covariância sejam suficientemente distintas. Neste caso, o problema de natureza prática que surge é o da estimação dos parâmetros que caracterizam a distribuição de cada classe. Na medida em que a dimensionalidade dos dados cresce, aumenta o número de parâmetros a serem estimados, especialmente na matriz covariância. Contudo, é sabido que, no mundo real, a quantidade de amostras de treinamento disponíveis, é freqüentemente muito limitada, ocasionando problemas na estimação dos parâmetros necessários ao classificador, degradando portanto a acurácia do processo de classificação, na medida em que a dimensionalidade dos dados aumenta. O Efeito de Hughes, como é chamado este fenômeno, já é bem conhecido no meio científico, e estudos vêm sendo realizados com o objetivo de mitigar este efeito. Entre as alternativas propostas com a finalidade de mitigar o Efeito de Hughes, encontram-se as técnicas de regularização da matriz covariância. Deste modo, técnicas de regularização para a estimação da matriz covariância das classes, tornam-se um tópico interessante de estudo, bem como o comportamento destas técnicas em ambientes de dados de imagens digitais de alta dimensionalidade em sensoriamento remoto, como por exemplo, os dados fornecidos pelo sensor AVIRIS. Neste estudo, é feita uma contextualização em sensoriamento remoto, descrito o sistema sensor AVIRIS, os princípios da análise discriminante linear (LDA), quadrática (QDA) e regularizada (RDA) são apresentados, bem como os experimentos práticos dos métodos, usando dados reais do sensor. Os resultados mostram que, com um número limitado de amostras de treinamento, as técnicas de regularização da matriz covariância foram eficientes em reduzir o Efeito de Hughes. Quanto à acurácia, em alguns casos o modelo quadrático continua sendo o melhor, apesar do Efeito de Hughes, e em outros casos o método de regularização é superior, além de suavizar este efeito. Esta dissertação está organizada da seguinte maneira: No primeiro capítulo é feita uma introdução aos temas: sensoriamento remoto (radiação eletromagnética, espectro eletromagnético, bandas espectrais, assinatura espectral), são também descritos os conceitos, funcionamento do sensor hiperespectral AVIRIS, e os conceitos básicos de reconhecimento de padrões e da abordagem estatística. No segundo capítulo, é feita uma revisão bibliográfica sobre os problemas associados à dimensionalidade dos dados, à descrição das técnicas paramétricas citadas anteriormente, aos métodos de QDA, LDA e RDA, e testes realizados com outros tipos de dados e seus resultados.O terceiro capítulo versa sobre a metodologia que será utilizada nos dados hiperespectrais disponíveis. O quarto capítulo apresenta os testes e experimentos da Análise Discriminante Regularizada (RDA) em imagens hiperespectrais obtidos pelo sensor AVIRIS. No quinto capítulo são apresentados as conclusões e análise final. A contribuição científica deste estudo, relaciona-se à utilização de métodos de regularização da matriz covariância, originalmente propostos por Friedman [FRI 89] para classificação de dados em alta dimensionalidade (dados sintéticos, dados de enologia), para o caso especifico de dados de sensoriamento remoto em alta dimensionalidade (imagens hiperespectrais). A conclusão principal desta dissertação é que o método RDA é útil no processo de classificação de imagens com dados em alta dimensionalidade e classes com características espectrais muito próximas.

Welfare costs of inflation when interest-bearing deposits are disregarded: a calculation of the bias

Most estimates of the welfare costs of in ation are devised considering only noninterest- bearing assets, ignoring that since the 80s technological innovations and new regulations have increased the liquidity of interest-bearing deposits. We investigate the resulting bias. Suscient and necessary conditions on its sign are presented, along with closed-form expressions for its magnitude. Two examples dealing with bidimensional bilogarithmic money demands show that disregarding interest-bearing monies may lead to a non-negligible overestimation of the welfare costs of in ation. An intuitive explanation is that such assets may partially make up for the decreased demand of noninterest-bearing assets due to higher in ation.

Comportamento de ajuda entre os "bóias-frias": um estudo exploratório

Esta pesquisa visou levantar as manifestações do comportamento de ajuda entre "bóias-frias" através de seu próprio reato, com o objetivo de analisar as implicações deste comportamento para sua organização social enquanto um grupo específico. Foram entrevistados 47 sujeitos, de ambos os sexos, no seu local de trabalho . O instrumento utilizado foi uma entrevista estruturada construída pela autora desta pesquisa , composta de 26 questões. Estas questões buscaram levantar frequência, razões e situações de ajudai solicitada ou espontânea, do entrevistado em relação aos colegas e de seus colegas em relação a ele mesmo, no trabalho e fora dele . Observou-se que quando perguntados se prestam e recebem ajuda, houve um grande índice de respostas afirmativas e quando solicitados a relatar as situações ocorridas, o índice de respostas diminuiu consideravelmente . Foram discutidas as possíveis razões para a ocorrência de tal fato. A análise das situações de ajuda narradas serviu para a compreensão de como a estrutura do trabalho volante determina as formas que assumem as relações de ajuda entre os "bóias-frias" e o quanto estas relações I por sua vez, influenciam a estruturação das relações sociais destes trabalhadores.

Estimation of DSGE Models: A Monte Carlo Analysis

Neste trabalho investigamos as propriedades em pequena amostra e a robustez das estimativas dos parâmetros de modelos DSGE. Tomamos o modelo de Smets and Wouters (2007) como base e avaliamos a performance de dois procedimentos de estimação: Método dos Momentos Simulados (MMS) e Máxima Verossimilhança (MV). Examinamos a distribuição empírica das estimativas dos parâmetros e sua implicação para as análises de impulso-resposta e decomposição de variância nos casos de especificação correta e má especificação. Nossos resultados apontam para um desempenho ruim de MMS e alguns padrões de viés nas análises de impulso-resposta e decomposição de variância com estimativas de MV nos casos de má especificação considerados.

Non-monotone insurence contracts and their empirical consequences

The goal of t.his paper is to show the possibility of a non-monot.one relation between coverage and risk which has been considered in the literature of insurance models since the work of Rothschild and Stiglitz (1976). We present an insurance model where the insured agents have heterogeneity in risk aversion and in lenience (a prevention cost parameter). Risk aversion is described by a continuou.'l parameter which is correlated with lenience and, for the sake of simplicity, we assume perfect correlation. In the case of positive correlation, the more risk averse agent has higher cost of prevention leading to a higher demand for coverage. Equivalently, the single crossing property (SCP) is valid and implies a positive correlation between coverage and risk in equilibrium. On the other hand, if the correlation between risk aversion and lenience is negative, not only may the sep be broken, but also the monotonicity of contracts, i.e., the prediction that high (Iow) risk averse types choose full (partial) insurance. In both cases riskiness is monotonic in risk aversion, but in the last case t,here are some coverage leveIs associated with two different risks (low and high), which implies that the ex-ante (with respect to the risk aversion distribution) correlation bet,ween coverage and riskiness may have every sign (even though the ex-post correlation is always positive). Moreover, using another instrument (a proxy for riskiness), we give a testable implication to disentangle single crossing and non single crossing under an ex-post zero correlation result: the monotonicity of coverage as a function of riskiness. Since by controlling for risk aversion (no asymmetric informat, ion), coverage is a monotone function of riskiness, this also gives a test for asymmetric information. Finally, we relate this theoretical results to empirica! tests in the recent literature, specially the Dionne, Gouriéroux and Vanasse (2001) work. In particular, they found an empirical evidence that seems to be compatible with asymmetric information and non single crossing in our framework. More generally, we build a hidden information model showing how omitted variabIes (asymmetric information) can bias the sign of the correlation of equilibrium variabIes conditioning on ali observabIe variabIes. We show that this may be t,he case when the omitted variabIes have a non-monotonic reIation with t,he observable ones. Moreover, because this non-monotonic reIat,ion is deepIy reIated with the failure of the SCP in one-dimensional screening problems, the existing lit.erature on asymmetric information does not capture t,his feature. Hence, our main result is to point Out the importance of t,he SCP in testing predictions of the hidden information models.

Is there a price puzzle in Brazil? an application of bias- corrected bootstrap

Using data from the United States, Japan, Germany , United Kingdom and France, Sims (1992) found that positive innovations to shortterm interest rates led to sharp, persistent increases in the price leveI. The result was confirmed by other authors and, as a consequence of its non-expectable nature, was given the name "price puzzle" by Eichenbaum (1992). In this paper I investigate the existence of a price puzzle in Brazil using the same type of estimation and benchmark identification scheme employed by Christiano et aI. (2000). In a methodological improvement over these studies, I qualify the results with the construction of bias-corrected bootstrap confidence intervals. Even though the data does show the existence of a statistically significant price puzzle in Brazil, it lasts for .only one quarter and is quantitatively immaterial.

Bias and size effects of price-comparison search engines : theory and experimental evidence

We analyze the impact on consumer prices of the size and bias of price comparison search engines. In the context of a model related to Burdett and Judd (1983) and Varian (1980), we develop and test experimentally several theoretical predictions. The experimental results confirm the model’s predictions regarding the impact of the number of firms, and the type of bias of the search engine, but reject the model’s predictions regarding changes in the size of the index. The explanatory power of an econometric model for the price distributions is significantly improved when variables accounting for risk attitudes are introduced.

Overconfidence and confirmation bias: are future managers vulnerable?

Decision makers often use ‘rules of thumb’, or heuristics, to help them handling decision situations (Kahneman and Tversky, 1979b). Those cognitive shortcuts are taken by the brain to cope with complexity and time limitation of decisions, by reducing the burden of information processing (Hodgkinson et al, 1999; Newell and Simon, 1972). Although crucial for decision-making, heuristics come at the cost of occasionally sending us off course, that is, make us fall into judgment traps (Tversky and Kahneman, 1974). Over fifty years of psychological research has shown that heuristics can lead to systematic errors, or biases, in decision-making. This study focuses on two particularly impactful biases to decision-making – the overconfidence and confirmation biases. A specific group – top management school students and recent graduates - were subject to classic experiments to measure their level of susceptibility to those biases. This population is bound to take decision positions at companies, and eventually make decisions that will impact not only their companies but society at large. The results show that this population is strongly biased by overconfidence, but less so to the confirmation bias. No significant relationship between the level of susceptibility to the overconfidence and to the confirmation bias was found.

Bias from ignoring price dispersion in demand estimation

Consumers often pay different prices for the same product bought in the same store at the same time. However, the demand estimation literature has ignored that fact using, instead, aggregate measures such as the “list” or average price. In this paper we show that this will lead to biased price coefficients. Furthermore, we perform simple comparative statics simulation exercises for the logit and random coefficient models. In the “list” price case we find that the bias is larger when discounts are higher, proportion of consumers facing discount prices is higher and when consumers are more unwilling to buy the product so that they almost only do it when facing discount. In the average price case we find that the bias is larger when discounts are higher, proportion of consumers that have access to discount are similar to the ones that do not have access and when consumers willingness to buy is very dependent on idiosyncratic shocks. Also bias is less problematic in the average price case in markets with a lot of bargain deals, so that prices are as good as individual. We conclude by proposing ways that the econometrician can reduce this bias using different information that he may have available.

Bounds on policy relevant parameters with discrete policy variation

When estimating policy parameters, also known as treatment effects, the assignment to treatment mechanism almost always causes endogeneity and thus bias many of these policy parameters estimates. Additionally, heterogeneity in program impacts is more likely to be the norm than the exception for most social programs. In situations where these issues are present, the Marginal Treatment Effect (MTE) parameter estimation makes use of an instrument to avoid assignment bias and simultaneously to account for heterogeneous effects throughout individuals. Although this parameter is point identified in the literature, the assumptions required for identification may be strong. Given that, we use weaker assumptions in order to partially identify the MTE, i.e. to stablish a methodology for MTE bounds estimation, implementing it computationally and showing results from Monte Carlo simulations. The partial identification we perfom requires the MTE to be a monotone function over the propensity score, which is a reasonable assumption on several economics' examples, and the simulation results shows it is possible to get informative even in restricted cases where point identification is lost. Additionally, in situations where estimated bounds are not informative and the traditional point identification is lost, we suggest a more generic method to point estimate MTE using the Moore-Penrose Pseudo-Invese Matrix, achieving better results than traditional methods.

Home bias no mercado de capitais brasileiro : um estudo sobre a custódia de investidores em ações

A presente dissertação apresenta uma análise da concentração do portfólio de a-ções de investidores brasileiros nas próprias empresas onde trabalham com o intuito de observar se o Home Bias se aplica à amostra analisada. Nosso estudo foi reali-zado com uma amostra extraída da custódia de 86 clientes de uma corretora de va-lores mobiliários, sendo estes dados reais de mercado. Restringimos a seleção da amostra de forma que metade fosse de clientes que trabalham em empresas de ca-pital aberto e a outra metade não. Foi feita análise cross section de quanto os inves-tidores alocam em ações das empresas para a qual trabalham e verificou-se qual o percentual desta participação em seus portfólios, em comparação a uma amostra de controle de investidores que não trabalham nesta mesma empresa. Além destas a-nálises, separamos a amostra pelo valor total do portfólio e realizamos os mesmos estudos com estes dois grupos diferentes da amostra. Como uma análise de robus-tez, identificamos empresas listadas que não adotam a remuneração com ações como uma sub-amostra do estudo. Como resultado, encontramos evidências de que os funcionários investem significativamente mais (seja em proporção do portfólio ou em valores) em ações das empresas onde trabalham em relação aos demais inves-tidores, evidenciando um viés de familiaridade na tomada de decisão de investimen-tos.

Caracterização da captura incidental de tartarugas marinhas pela pesca de espinhel pelágico no Nordeste do Brasil

The marine turtles biological characteristics and the impact they have been suffering in consequence of human activities have caused in the last decades the decrease of populations to unsustainable levels. All four of the species described in this paper are registered as endangered in a list by IUCN: Caretta caretta, Lepidochelys olivacea, Chelonia mydas, Dermochelys coriacea. The main causes of such impact include several fishing activities, mostly the surface longline. This paper discusses the monitoring of two foreigner longline fleet along the North East Brazilian coast between October of 2004 and September of 2005. Both operated in the West South Atlantic, one using the Chinese technique and the other the American. The American method s target species is the swordfish (Xiphias gladius), and it is characterized by using squid as bait, J 9/0 offset 5º hook, light sticks and night soaking. It also operates in shallower waters than the Chinese method. The source of information about the efforts and the catches came from onboard observers and were used to calculate the catching rate of turtles over 1000 hooks (CPUE). The American equipment caught more turtles (CPUE = 0,059; N= 103), mainly D. coriacea, while the Chinese longline caught mainly the L. olivacea and presented a CPUE= 0,018 (N= 89). The hooks were most frequently found attached to the mouth of C. caretta, C. mydas, and L. olivacea. The D. coriacea were most frequently caught by hooks externally attached to different parts of their body. There was no significant difference between the hook type catching and most turtles were still alive when released. The results suggest a greater potential of turtle catching by the American method. Besides the statistic tests have showed less interaction between the Chinese equipment and marine turtles, the catches of this fishing technique could have been underestimated due to miscommunication between the onboard observer and the vessel s crew plus the retrieve of the longline during night time

A Learning Function for Parameter Reduction in Spiking Neural Networks with Radial Basis Function

Spiking neural networks - networks that encode information in the timing of spikes - are arising as a new approach in the artificial neural networks paradigm, emergent from cognitive science. One of these new models is the pulsed neural network with radial basis function, a network able to store information in the axonal propagation delay of neurons. Learning algorithms have been proposed to this model looking for mapping input pulses into output pulses. Recently, a new method was proposed to encode constant data into a temporal sequence of spikes, stimulating deeper studies in order to establish abilities and frontiers of this new approach. However, a well known problem of this kind of network is the high number of free parameters - more that 15 - to be properly configured or tuned in order to allow network convergence. This work presents for the first time a new learning function for this network training that allow the automatic configuration of one of the key network parameters: the synaptic weight decreasing factor.

Retrievals of a size parameter for phytoplankton and spectral light absorption by colored detrital matter from water-leaving radiances at SeaWiFS channels in a continental shelf region off Brazil

Many efforts are currently oriented toward extracting more information from ocean color than the chlorophyll a concentration. Among biological parameters potentially accessible from space, estimates of phytoplankton cell size and light absorption by colored detrital matter (CDM) would lead to an indirect assessment of major components of the organic carbon pool in the ocean, which would benefit oceanic carbon budget models. We present here 2 procedures to retrieve simultaneously from ocean color measurements in a limited number of bands, magnitudes, and spectral shapes for both light absorption by CDM and phytoplankton, along with a size parameter for phytoplankton. The performance of the 2 procedures was evaluated using different data sets that correspond to increasing uncertainties: ( 1) measured absorption coefficients of phytoplankton, particulate detritus, and colored dissolved organic matter ( CDOM) and measured chlorophyll a concentrations and ( 2) SeaWiFS upwelling radiance measurements and chlorophyll a concentrations estimated from global algorithms. In situ data were acquired during 3 cruises, differing by their relative proportions in CDM and phytoplankton, over a continental shelf off Brazil. No local information was introduced in either procedure, to make them more generally applicable. Over the study area, the absorption coefficient of CDM at 443 nm was retrieved from SeaWiFS radiances with a relative root mean square error (RMSE) of 33%, and phytoplankton light absorption coefficients in SeaWiFS bands ( from 412 to 510 nm) were retrieved with RMSEs between 28% and 33%. These results are comparable to or better than those obtained by 3 published models. In addition, a size parameter of phytoplankton and the spectral slope of CDM absorption were retrieved with RMSEs of 17% and 22%, respectively. If these methods are applied at a regional scale, the performances could be substantially improved by locally tuning some empirical relationships.

Analysis of the Priestley-Taylor alpha parameter for a bean crop

This work deals with the Priestley-Taylor model for evapotranspiration in different grown stages of a bean crop. Priestley and Taylor derived a practical Formulation for energy partitioning between the sensible and latent heat fluxes through the a parameter. Bowen ratio energy balance (BREB) was carried out for daily sensible and latent heat flux estimations in three different crop stages. Mean daily values of Priestley-Taylor a parameter were determined for eleven days during the crop cycle. Diurnal variation patterns of a are presented for the growing, flowering and graining periods. The mean values of 1.13 +/- 0.33, 1.26 +/- 0.74, 1.22 +/- 0.55 were obtained for a day in the growing, in the flowering and for graining periods, respectively. Eleven days values of a are shown and gave a mean value of 1.23 +/- 0.10 which agree on the reported literature.