Horário (ainda sujeito a pequenas alterações)
| Dia 10 – Quinta-feira
10:00h Abertura 10:20h Eduardo Ferioli Gomes 11:00h Patrícia Lusié Velozo da Costa Intervalo para almoço 11:40h-13:20h 13:20h Vinicius Pinheiro Israel Intervalo – 14:00h-14:10h 14:10h Rafael Erbisti 14:50h Guilherme dos Santos café e conversa – 15:30h-16:10h 16:10 Josiane S. Cordeiro Jantar de confraternização – 19:00 |
Dia 11 – Sexta-feira
10:20h Vitor Capdeville 11:00h Rafael Santos Intervalo para almoço 11:40-13:20h 13:20h Jony Arrais Pinto Junior Intervalo – 14:00h-14:10h 14:10h Pamela M. Chiroque-Solano (online) 14:50h Guido Alberti Moreira (online) café e conversa: 15:30-16:10 16:10h Marcus L. Nascimento 16:50h Encerramento |
Resumos das palestras
Palestrante: Eduardo Ferioli Gomes (UFF)
Título: Directional High Frequency Trading in the Kyle-Back Model
Resumo: In traditional Kyle-Back models, the only source of information comes from the insider’s signal. We consider a more realistic version of the Kyle-Back model with a private and a public signal. The insider observes both signals. The private signal, that is only directly observed by the insider, may be static, when the insider knows the value of the asset in advance, or dynamic, when it converges to the true value of the asset at the end of the trading period. The market maker receives a dynamic signal that also converges to the true value of the asset at the end of the trading period.
In the dynamic case, we prove that the insider’s valuation of the asset is given by a linear combination of both the public and private signals and it is a martingale for the insider’s filtration.
Furthermore, we show that the price – which is the market maker’s valuation of the asset – is also given by a linear combination of the public signal and the weighted demand. In addition, it is proven that it converges to the true price of the asset as it is expected in the traditional theory.
An interesting fact that is observed is that it is possible to see an increase in the volatility of the price in the end of the trading period when trading becomes aggressive due to the convergence of both signals to the true price of the asset.
Palestrante: Guido Alberti Moreira (Universidade do Minho, Portugal)
Título: Presence-Only for Marked Point Process Under Preferential Sampling
Resumo: Preferential sampling models have garnered significant attention in recent years. Although the original model was developed for geostatistics, it found applications in other types of data, such as point processes in the form of presence-only data. While this has been recognized in the Statistics literature, there is value in incorporating ideas from both presence-only and preferential sampling literature. In this paper, we propose a novel model that extends existing ideas to handle a continuous variable collected through opportunistic sampling. To demonstrate the potential of our approach, we apply it to sardine biomass data collected during commercial fishing trips. While the data is intuitively understood, it poses challenges due to two types of preferential sampling: fishing events (presence data) are non-random samples of the region, and fishermen tend to set their nets in areas with a high quality and value of catch (i.e., bigger schools of the target species). We discuss theoretical and practical aspects of the problem, and propose a well-defined probabilistic approach. Our approach employs a data augmentation scheme that predicts the number of unobserved fishing locations and corresponding biomass (in kg). This allows for evaluation of the Poisson Process likelihood without the need for numerical approximations. The results of our case study may serve as an incentive to use data collected during commercial fishing trips for decision-making aimed at benefiting both ecological and economic aspects. The proposed methodology has potential applications in a variety of fields, including ecology and epidemiology, where marked point process model are commonly used.
Palestrante: Guilherme dos Santos (UFRJ)
Título: A multivariate approach for correcting reporting delays in infectious disease surveillance
Resumo: Frequently, real-time tracking of epidemics is faced with a concerning issue, the reporting delays of cases and deaths. Delays might occur due to logistical problems, laboratory confirmation, and other reasons. Being able to correct the delay is essential to decision-making with the goal of containing an epidemic. In some cases, the epidemic might be associated with more than one disease, Dengue and Chikungunya are common examples of this phenomenon. We propose a multivariate model to correct reporting delays and accommodate the above-mentioned cases. The model is estimated using the Integrated Nested Laplace Approximation method with the aim of providing faster results. An application for the corrections of reporting delays of Dengue and Chikungunya in the state of Rio de Janeiro during an epidemic in 2019 is provided.
Keywords: Nowcasting; Dengue; Chikungunya; INLA; Bayesian hierarchical model.
Palestrante: Jony Arrais Pinto Junior (UFF)
Título: A Jornada de uma Década após o Doutorado na UFRJ
Resumo: O objetivo desta apresentação é compartilhar minhas experiências profissionais desde a conclusão do doutorado em 2014, com ênfase no meu papel como professor e pesquisador no Departamento de Estatística da Universidade Federal Fluminense (UFF).
Ao longo dessa década, colaborei com diversos centros de pesquisa, como o IPEA, FIOCRUZ e UFBA, fortalecendo minhas contribuições nas mais diversas áreas. Um ponto de grande relevância tem sido a formação de novos talentos da graduação em Estatística na UFF, muitos dos quais estão sendo absorvidos por renomados programas de pós-graduação, como o da UFRJ. Também abordarei as direções atuais da minha pesquisa, que incluem análise de classes latentes espaciais e o estudo de dados longitudinais, destacando as possibilidades e avanços na área.
Palestrante: Josiane S Cordeiro (UFRRJ)
Título: Energy Consumption Forecast in the Brazilian Industrial Sector: Integration of Bottom-Up and Top-Down Methods with Monte Carlo Simulation
Resumo: The Brazilian industrial sector is the largest electricity consumer in the power system. Energy planning in this sector is important mainly due to its economic, social, and environmental impact. In this context, electricity consumption analysis and projections are highly relevant for the decision-making of the industrial sector and organizations operating in the energy system. The electricity consumption data from the Brazilian industrial sector can be organized into a hierarchical structure composed of each geographic region (South, Southeast, Midwest, Northeast, and North) and their respective states. This work proposes a hybrid approach that combines bottom-up and top-down methods using Monte Carlo simulation for electricity consumption forecasting. The exponential smoothing and Box-Jenkins models were used to generate the projections of the individual series. The proposed approach was compared with the bottom-up, top-down, and optimal combination approaches, which are widely used for time series hierarchical forecasting. The performance of the models was evaluated using the mean absolute percentage error (MAPE) and root mean squared error (RMSE) precision measures. The results indicate that the proposed hybrid approach can contribute to the projection and analysis of industrial sector electricity consumption in Brazil.
Palestrante: Marcus L. Nascimento (FGV-EMAp e Fundação José Luis Egydio Setúbal)
Título: An Expectation-Maximization algorithm for noncrossing Bayesian quantile regression.
Resumo: When quantiles are fitted separately, the resultant regression lines may cross, violating the basic probabilistic rule that quantiles are monotonic functions and possibly causing problems for inference in practice. Using location-scale mixture representation of asymmetric Laplace distribution (ALD), we write a joint posterior density function for all quantile levels of interest and develop a constrained Expectation-Maximization algorithm that handles crossing issues.
Palestrante: Pamela M. Chiroque-Solano (University of Regensburg)
Título: Probabilistic Models and Machine Learning Algorithms for Biomass and LAI Prediction Using Multispectral and LiDAR UAV Data.
Resumo: Recent advances in remote sensing have revolutionized environmental monitoring by enabling the integration of multispectral and LiDAR data from UAV platforms. This dual-capability technology provides high-resolution spectral and 3D geometric information critical for precision agriculture and forestry applications. Specifically, the prediction of biomass and leaf area index (LAI)—key indicators for ecosystem health, carbon sequestration, and resource management—has become more accurate and accessible. However, the inherent heterogeneity and multicollinearity in the derived features present significant challenges in model development, requiring sophisticated approaches for optimal predictive performance. This study explores the comparative efficacy of probabilistic models and machine learning algorithms for predicting biomass and LAI from UAV-based multispectral and LiDAR data. By incorporating advanced variable selection methods, our approach mitigates issues of overfitting, reduces bias, and utilizes algorithms capable of handling non-linear outcomes, leading to more reliable predictions in both agricultural and forestry settings. More than thirty models were applied to two distinct datasets.
Our results underscore the critical importance of methodical model selection criteria, where probabilistic approaches and machine learning algorithms can be compared, highlighting their advantages and disadvantages, even under different assumptions. Each approach offers distinct benefits depending on the data characteristics and the specific application, making post-hoc metrics that reconcile these paradigms essential.
Although many methods are compared in this work, others can be considered as well. The outcomes can be used as insights for strategies for biomass and LAI prediction, offering a clear path toward more accurate, scalable solutions in environmental monitoring. Our results highlight not only the practical utility of UAV-derived multispectral and LiDAR data but also the role of rigorous model comparison in advancing remote sensing applications in agriculture and forestry.
Trabalho em conjunto com: Luís Padua e Domingos Manuel Mendes Lopes (University of Trás-os-Montes e Alto Douro).
Palestrante: Patrícia Lusié Velozo da Costa (UFF)
Título: Entre Modelos e Maternidade: Desafios e Experiências Vivenciadas após o Doutorado.
Resumo: Nesta apresentação, compartilharei minha trajetória desde a conclusão do doutorado, abordando as experiências que marcaram minha vida acadêmica ao longo da última década. Refletirei sobre meu papel como professora e pesquisadora no Departamento de Estatística da Universidade Federal Fluminense, onde assumi responsabilidades como Chefia de Departamento e a organização de eventos como a Semana da Estatística e o Hackathon. Além disso, abordarei as direções atuais da minha pesquisa, com foco em dois projetos principais: análise de dados inflacionados de zero e mortalidade materna. Ambos os estudos aplicam modelos lineares generalizados, com a estimação de parâmetros realizada sob a abordagem Bayesiana, utilizando o método de Monte Carlo Hamiltoniano. Em particular, os dados inflacionados de zero são ajustados por meio de modelos de mistura, como o modelo Poisson inflacionado de zero.
Palestrante: Rafael Erbisti (UFF)
Título: ARBOALVO: Bayesian spatiotemporal learning and predictive model for dengue
Resumo: Transmission of urban arboviruses does not occur homogeneously across territories and varies over time. Estimating the risk of dengue through statistical models that consider simultaneous variability in space and time provides more realistic estimates of transmission dynamics, facilitating the identification of priority areas. Additionally, such models enable predictions for timely actions in controlling and surveying urban arboviruses, such as dengue, chikungunya, and Zika. We analyzed the reported cases of dengue by epidemiological week and neighborhood in Natal-RN between 2015 and 2018. Temporal Conditional Autoregressive models are fitted. The predictor comprised a set of entomological, climatic, and sociosanitary indicators with temporal lags, an offset term, and structures of temporal and spatial dependence. Fitting was performed using the Integrated Nested Laplace Approximation method. We forecast dengue case counts for the next four weeks, accounting for both non-occurrences and fluctuations in the time series during periods of non-zero occurrences. Predictive maps of weekly risk dynamics were obtained, allowing timely identification of neighborhoods with high and persistent dengue risk. The best model indicated a significant increase in the probability of dengue occurrence in the observation week, with an increase of one standard deviation in reported cases in the previous week, Aedes egg positivity index from the previous four weeks, and mean daytime temperature in the preceding 6–8 weeks. There was an increase in dengue risk with an increase of one standard deviation in the density of the poor population per occupied area and in the mean Aedes egg density index in the previous 3–5 weeks. The proposed Bayesian space-time analysis can contribute to the operational control of dengue and Aedes Aegypti by detecting priority areas and predicting dengue cases for the next four weeks. Additionally, it identified and quantified the influences of entomological, sociosanitary, climatic, and demographic indicators.
Palestrante: Rafael Souza dos Santos (UFRJ)
Título: Processo de contato sob renovações
Resumo: Na apresentação eu vou falar sobre o processo de contato sob renovações, um assunto no qual trabalhei durante meu pós-doutorado e sigo trabalhando como professor adjunto do Departamento de Métodos Estatísticos da UFRJ. O processo de contato (clássico) busca modelar a evolução de uma doença infecciosa do seguinte modo: Indivíduos infectados transmitem a doença para seus vizinhos após tempos exponenciais de taxa λ > 0 e ficam curados após tempos exponenciais de taxa 1. Já o processo de contato sob renovações flexibiliza esse modelo clássico, permitindo que os tempos até a cura da doença tenham distribuições mais gerais. A ideia é que a apresentação seja a mais inclusiva possível, de modo que não vou entrar em detalhes técnicos sobre o modelo.
Palestrante: Vinicius Pinheiro Israel (UNIRIO)
Título: Punição e liberdades: encarceramento em massa, seletividade penal e população escondida
Resumo: O encarceramento em massa é um problema central das democracias modernas e impacta diretamente nas chances de vida de grande parcela da população. No Brasil, o número de pessoas presas teve um aumento de cerca de 600% desde 1990, o país possui mais de 750 mil presos, colocando-se na terceira posição entre os que mais encarceram no mundo. As explicações sobre o aumento carcerário norteiam-se pela ideia de reorientação liberal, pelo fortalecimento de ideias conservadoras de combate ao crime e pelo aumento da demanda por mais punição. A espinha dorsal dos argumentos no Brasil enfatiza a passagem do regime autoritário ditatorial para a democracia, marcada pela resistência de policiais e funcionários do sistema penitenciário ao novo regime e por uma acentuada instabilidade política na esfera da segurança pública. A partir de um diálogo com a estratificação social, teoria política e metodologia quantitativa, o presente livro busca compreender as causas e consequências desse fenômeno.
Palestrante: Vitor Capdeville (ESSOR Seguros)
Título: Integração de Cálculos Atuariais e Soluções de ETL com Python em Ambientes Corporativos
Resumo: Durante minha atuação na MAG Seguros e Essor, enfrentei desafios relacionados à centralização de cálculos atuariais e à automação de processos de manipulação de dados. Desenvolvi soluções utilizando Python que otimizam tanto a execução de cálculos complexos quanto a gestão de pipelines de dados. O primeiro projeto, um pacote de cálculos atuariais, permitiu que atuários tivessem mais controle sobre a lógica de cálculo de prêmios e reservas, integrando uma API em FastAPI com o sistema da companhia. Já o pacote ExTraLo foi criado para simplificar processos de ETL, oferecendo uma interface flexível que extrai, valida, transforma e carrega dados de fontes diversas.
Além disso, apresento o Psitest, um projeto em desenvolvimento voltado à aplicação de testes psicológicos em ambiente digital. A solução envolve a correção automatizada de testes a partir de imagens e utiliza técnicas de visão computacional para identificar e classificar respostas.