NIMS - Dissertações de Mestrado em Estatística e Gestão da Informação (Statistics and Information Management)
URI permanente para esta coleção:
Navegar
Entradas recentes
- Probability of Default Estimation in Low-Default Portfolios: Calibration, Uncertainty, and Capital Implications under an IRB-Inspired FrameworkPublication . Santos, Sofia Laura Cardoso dos; Lopes, Samuel José da RochaLow-default portfolios (LDPs) pose a fundamental challenge for probability of default (PD) estimation because the very limited number of observed adverse events constrains both statistical reliability and prudential calibration. This study addresses a gap in the literature: existing research rarely examines, within a single corporate LDP framework, whether models that improve discrimination also deliver stable, calibrated, and capital-relevant PDs. Using Moody’s Orbis data for EU and US non-financial corporates, the final sample contains 766,775 firm-year observations and 317 oneyear-ahead adverse legal events. The empirical framework combines lagged logistic regression, Bayesian regularisation, machine-learning and oversampling benchmarks, calibration diagnostics, external anchoring, macro-financial stress testing, and IRBinspired loss and capital translation. Results show that Bayesian and machine-learning models improve ranking, but none resolves the calibration gap. Oversampling enhances class separation while severely distorting probability levels. The study contributes an integrated framework that treats risk ranking, probability calibration, stress sensitivity, and prudential translation as distinct but complementary stages in corporate low-default modelling.
- Early Warnings Indicators for Credit Risk of Development Banks: Evidence from Portuguese Development BankPublication . Mata, Leonor Vieira de Resende Fogaça da; Lopes, Samuel José da RochaThis thesis examines how different guarantee scheme designs affect borrower selection and development outcomes at a public development bank. The study exploits a natural experiment where a single institution simultaneously offers two structurally distinct modalities to the same commercial banks and SMEs: individual guarantees with development bank risk assessment (Modality 1) versus portfolio guarantees with commercial bank discretion (Modality 2). Drawing on loan-level data (2017–2024), the study combines logistic regression and machine-learning classification techniques to evaluate borrower risk profiles. In addition, a difference-in-differences approach is employed to assess firm-level outcomes over the 2022–2025 post-COVID recovery period. First, portfolio guarantee recipients exhibit significantly lower predicted distress risk (8.3% vs. 11.2%). Second, the study analyzes actual firm outcomes during postCOVID recovery, discriminating between information advantage (better identifying viable credit-constrained firms) and cherry-picking (selecting already-safe firms) mechanisms. Results reveal that portfolio guarantees (Modality 2) targeted distressedbut-viable firms requiring emergency capital redeployment, demonstrating genuine development impact through firm preservation, capital redeployment, and productivity enhancement. Results contribute to understanding guarantee design effects on credit allocation in development banks.
- Ex Ante Regime-Aware Evaluation of VaR-ES Tail Risk Models: A Multidimensional Assessment of Statistical Adequacy, Predictive Performance, and StabilityPublication . Tereso, Mário André Pereira Campos; Ashofteh, AfshinThe regulatory transition in market risk oversight from Value at Risk to Expected Shortfall has transformed how tail-risk forecasts are evaluated. Model validity is now assessed in a complementary evaluation that combines backtesting procedures with joint VaR-ES scoring, not only by violation frequency but also by tail-loss severity. Despite this regulatory adjustment, methodological challenges arise in maintaining statistical and practical consistency in tail risk estimates due to volatility clustering and state-dependent shifts in conditional variance dynamics. While regime dependent volatility structures have become more widely recognized in empirical finance, it is not yet evident whether prospective regime adjustments drive sustained improvements across adequacy dimensions within this scoring-based context. Without incorporating scoring-based evaluation, comparative assessments of VaR-ES forecasts could lead to variability in evaluation results, fundamentally altering how regime-specific findings are interpreted. This analysis compares baseline and regime-aware specifications to assess performance trade offs across statistical, scoring-based, and stability dimensions. The modeling approach evaluates conventional volatility models against macro financial regime dependent refinements activated by observable threshold indicators. This evaluation incorporates joint VaR-ES scoring, rank-stability diagnostics, and conditional coverage backtests within a multidimensional evaluation approach at the 95% and 99% tail quantiles. The empirical findings indicate that incorporating regime awareness produces state-contingent performance gains rather than uniform improvements in statistical adequacy and scoring-based evaluation across evaluation dimensions. Enduring volatility persistence exerts greater influence than regime responsiveness in evaluating performance under extreme market conditions. The results indicate that regime-aware adjustments yield contextdependent marginal improvements but do not consistently enhance overall model adequacy.
- Consumer Satisfaction in Inter-Island Logistics Services: The Roles of Perceived Value and Island Context in Cape VerdePublication . Gomes, Indira Julieta Duarte Lopes; Neves, Maria de Fátima dos Santos TrindadeInter-island logistics services are central to territorial cohesion, market integration, and access to essential goods in archipelagic settings such as Cape Verde. Despite their structural importance, the factors shaping consumer satisfaction in this specific logistical context remain poorly understood, leaving a relevant gap between the centrality of inter-island logistics and the knowledge available to improve it. To address this gap, this study develops and tests a conceptual model grounded in Expectation– Confirmation Theory and a multidimensional view of service quality, analyzing how key service quality dimensions influence Perceived Value and, in turn, Consumer Satisfaction, while also considering the moderating role of Island context. The empirical analysis is based on survey data collected from 214 valid respondents across the Cape Verdean archipelago and is examined using Partial Least Squares Structural Equation Modeling (PLS-SEM). The findings show that Perceived Value is the strongest direct predictor of Consumer Satisfaction. Operational Quality, Information Quality, and Service Quality all positively and significantly influence Perceived Value, with Information Quality emerging as the strongest antecedent. Island context significantly moderates the relationship between Perceived Value and Consumer Satisfaction, indicating that the translation of value into satisfaction is not territorially uniform across the archipelago, with the value-satisfaction relationship being amplified in islands with greater service exposure and connectivity. All three quality dimensions also produce significant indirect effects on Consumer Satisfaction through Perceived Value. These findings contribute to a more nuanced understanding of service evaluation in archipelagic logistics contexts and highlight the importance of integrating territorial heterogeneity into both research models and service improvement strategies.
- The Impact of Social Media on Financial Literacy and the Adoption of Investment Strategies Among Portuguese Young AdultsPublication . Sousa, Tomás Almeida da Silva; Neves, Joana Paisana Pires Costa dasThis study investigates the role of social media, particularly finfluencers, in shaping financial literacy and investment behavior among young adults in Portugal. As social media has become a central source of financial information for younger generations, understanding its impact on financial decision-making is increasingly relevant. Using a quantitative research design, data were collected through an online survey administered to Portuguese young adults and analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM). The results indicate that exposure to finfluencer content is positively associated with engagement in financial topics and with the adoption of simplified investment strategies, such as dollar-cost averaging. However, increased exposure does not necessarily translate into higher objective financial literacy, suggesting the presence of overconfidence and superficial understanding. Trust and perceived credibility play a central mediating role in the relationship between digital financial content and investment behavior. Overall, the study provides empirical evidence from a Portuguese context and highlights both the opportunities and risks of social-media-based financial information.
- Impacto das Inundações e Tempestades Intensas na Rentabilidade do Seguro Multirriscos Habitação em PortugalPublication . Ramos, Luana Fonseca; Costa, Ana Cristina Marinho daAs alterações climáticas têm colocado o setor segurador perante desafios significativos. A crescente ocorrência de eventos meteorológicos extremos força as seguradoras a ajustarem os seus modelos de avaliação de risco e definição de prémios, sob risco de colocarem em causa o seu equilíbrio e viabilidade financeira. Esta dissertação analisa o impacto das inundações e das tempestades intensas na rentabilidade do seguro multirriscos habitação em Portugal, avaliando de que forma a evolução e a variabilidade dos fenómenos meteorológicos extremos se refletem no risco técnico do ramo. O estudo assenta na caracterização conjunta da exposição segurada e do comportamento dos sinistros ao longo do tempo, permitindo analisar a frequência, a severidade e a volatilidade das perdas associadas a eventos hidrometeorológicos. A metodologia incluiu a aplicação de Modelos Lineares Generalizados, Modelos Aditivos Generalizados, e técnicas estatísticas não paramétricas. A análise evidencia uma forte assimetria na distribuição das perdas, com uma elevada concentração do custo total em eventos extremos, bem como uma sazonalidade marcada, compatível com os padrões climáticos nacionais. A distinção entre inundações e tempestades revela perfis de risco distintos, tanto em termos de impacto financeiro como de instabilidade temporal, reforçando a importância de uma abordagem diferenciada na gestão do risco. Para além das métricas estatísticas tradicionais, são utilizados indicadores financeiros proxy, como o custo médio por unidade de exposição e a evolução da volatilidade das perdas, de modo a avaliar a pressão exercida sobre a rentabilidade técnica e a previsibilidade dos resultados. Os resultados sugerem um aumento da instabilidade da sinistralidade associada a eventos extremos, traduzindo-se numa maior incerteza na gestão do portefólio e numa pressão acrescida sobre a sustentabilidade do seguro multirriscos habitação. Embora não seja possível estabelecer uma relação causal direta com as alterações climáticas, a evidência empírica aponta para a necessidade de adaptação das práticas de precificação, gestão de risco e resseguro, num contexto de crescente exposição a fenómenos meteorológicos extremos.
- Comparative Time Series Analysis: The case of Emigration in PortugalPublication . Santos, Rita Piano Alves dos; Baptista, Maria Helena Miranda FloresThis thesis explores emigration dynamics in Portugal through a quantitative time series framework, with a primary focus on comparing the forecasting performance of classical statistical models and modern foundational models. Using annual data, the study examines the relationship between emigration flows and selected socio-economic and demographic variables, namely unemployment rate, education, Gross Domestic Product (GDP) and fertility index. Classical approaches including Autoregressive Integrated Moving Average with eXogenous variables (ARIMAX) models and Generalized Additive Models (GAM), are benchmarked against a deep learning-based foundational model, the Temporal Fusion Transformer (TFT). The models are evaluated in terms of their ability to capture temporal dependencies, non-linear effects, and multivariate interactions: forecasting performance is assessed using standard error metrics such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). The comparative analysis reveals differences in predictive accuracy between classical models and the TFT, highlighting the strengths and limitations of each modelling approach in the context of emigration forecasting. Overall, the findings contribute to a better understanding of the socio-economic drivers of emigration and demonstrate the potential of foundational models for demographic and migration forecasting, offering relevant insights for policy analysis and sustainable development planning.
- Rethinking Child Poverty in the EU: A cluster-based alternative to AROPEPublication . Serra, Maria João Gonçalves Dutra; Baptista, Maria Helena Miranda FloresThis thesis develops a multidimensional, child-specific typology of poverty and social exclusion across European Union Member States using harmonised national-level indicators from Eurostat. While the EU’s official measure, AROPE, plays a central role in social monitoring, it relies on household-level income and employment metrics that overlook key disadvantages experienced by children. To address this limitation, the study constructs a dataset of nine child-focused indicators capturing housing deprivation, economic insecurity, and early education access. After standardisation and targeted imputation, two clustering techniques, Ward’s hierarchical method and k-means, are applied to identify cross-country deprivation profiles. The analysis yields a four-cluster structure that distinguishes low, moderate, and high levels of child deprivation, including two extreme profiles corresponding to Greece and the Bulgaria–Romania group. In parallel, the AROPE components are disaggregated into seven mutually exclusive intersections and subjected to the same analytical pipeline, producing a simpler three-cluster typology characterised by a large low-risk group and a sharply isolated high-risk cluster. PaCMAP embeddings provide additional insight into the geometric structure of the data, revealing that the child-specific indicators form a dispersed and multidimensional landscape, whereas the AROPE components generate a compressed configuration with limited internal variation. Overall, the findings demonstrate that child-focused indicators capture structural patterns of disadvantage that remain invisible under income-based measures alone, underscoring the importance of multidimensional approaches for monitoring child well-being and informing EU social policy.
- Study of patterns in aircraft airframe MRO using a data driven approachPublication . Luz, Maria Helena Abreu; Baptista, Márcia Lourenço; Damásio, Bruno Miguel PintoAir travel is widely recognized as one of the safest and most convenient modes of transport worldwide. The aircraft maintenance industry is an important field of operation, therefore itis paramount to find an optimal approach to estimate aircraft tasks durations. This study aims to examine the available literature on existing methods to process data and to estimate duration in aviation. The research comprises an exploratory analysis, with the goal of understanding the real-world maintenance data set and finding relevant hidden insights. The analysis encountered patterns in task efficiency by skill type, aircraft type, and location. The second part of the research conducts an experimental analysis that compares four predictive machine learning models with traditional method PERT (Program Evaluating and Review Technique), a project management technique used to estimate the duration of tasks by considering optimistic, pessimistic, and most likely time estimates. The results of the study demonstrate the value of data-driven approaches in improving accuracy in maintenance task planning and performance.
- Integrating Climate Covariates into Mortality Forecasting Models: A Feasibility Study for Longevity Risk PricingPublication . Matteucci, Alessandro; Bravo, Jorge Miguel VenturaThis thesis investigates whether climate-related determinants—specifically extreme heat intensity—can be meaningfully incorporated into actuarial mortality modelling and whether such integration is empirically justified within the context of Southern European populations. Motivated by the growing intersection between environmental risk, demographic change, and longevity risk management, the study examines how exogenous climate variables may enter mortality-intensity frameworks traditionally structured around age, period, and cohort effects. At a conceptual level, the thesis formalizes the integration of a standardized heatanomaly indicator into a mortality modelling architecture consistent with the Generalized Age–Period–Cohort (GAPC) paradigm. At an empirical level, rather than estimating a fully stochastic climate-augmented GAPC model, the study implements a reduced-form Poisson Generalized Linear Model (GLM) panel specification. This approach allows the direct identification of the marginal association between regional annual heat anomalies and mortality rates at ages 65–85, while controlling for flexible age effects, region-specific heterogeneity, and common temporal shocks. Using harmonized mortality, exposure, and climate data for Southern European regions, the analysis finds that the estimated heat coefficients are generally small in magnitude and not consistently statistically significant under cluster-robust inference. Even where statistical significance is detected, the implied proportional change in annual mortality associated with a one-standard-deviation increase in heat intensity remains modest. Within the annual fixed-effects framework adopted, extreme heat anomalies do not emerge as dominant drivers of mortality dynamics. The contribution of the thesis lies in establishing a disciplined methodological bridge between climate indicators and actuarial mortality modelling. By linking demographic forecasting, environmental risk, and actuarial finance, the study offers an early step toward climate-aware longevity modelling systems.