Doctoral research at PUCPR
PhD in Industrial & Systems Engineering
Pontifical Catholic University of Paraná (PUCPR) — 2022 to 2027 (expected). Advisor: Gilberto Reynoso-Meza.
Research on econometric forecasting and multi-objective decision-making for agricultural commodity portfolios: blending time series forecasting models, Bayesian and regime-switching volatility modelling, and Pareto-front portfolio selection under uncertainty.
Time Series Financial & Bayesian Econometrics Multi-objective Portfolio Optimization Extremal Events Forecasting
Academic background
PhD Candidate, Industrial & Systems Engineering
Applied time series analysis, financial and Bayesian econometrics, multi-objective portfolio optimization and modelling of extremal events. Advisor: Gilberto Reynoso-Meza.
MSc in Economic Development
Emphasis on big data, multivariate analysis and econometrics. Read the thesis
BSc in Economics
Undergraduate thesis on oil price volatility (GARCH-M), honored and continued as a research project at the UFPR Laboratory of Statistics and GeoInformation. View the project
Selected academic output
Conference papers
A commodity price forecast is usually delivered as one number, which hides how uncertain it is. Fuzzy regression carries that imprecision explicitly instead of discarding it, and multi-criteria decision methods then compare whole scenarios rather than forcing a single answer.
When is a machine about to fail, and what does waiting cost? Survival analysis — the statistics of time-to-event, borrowed from medicine — is applied to agricultural equipment so maintenance is scheduled by risk rather than by calendar.
Two evolutionary algorithms are routinely used to trade risk against return in portfolios, and practitioners tend to pick one by habit. This puts them on the same problem and measures both the quality of the solutions and the computational cost of reaching them.
Grain prices do not only get more or less volatile — the whole shape of their distribution moves, including skew and fat tails. GAMLSS tracks those moments over time and MSGARCH captures shifts between calm and turbulent regimes, both put to work on portfolio risk. See the case study
Talks & posters
Doctoral research selected as a finalist at the national meeting of Brazilian graduate programmes in industrial engineering.
The poster version of the grain-portfolio work: combining a time-varying distribution shape with Bayesian regime-switching volatility to forecast commodity returns. See the case study
Economic series often arrive yearly when the decision needs monthly numbers. This shows how temporal disaggregation in R rebuilds the higher-frequency series without inventing movement the data cannot support.
A risk–return frontier estimated on a single window can be an artefact of that window. Building frontiers across several windows shows which trade-offs survive and which were an accident of the sample.
Working papers (2025–2026)
Doctoral research pipeline: sixteen manuscripts in preparation and submission. Titles subject to change; drafts available on request.
Portfolio optimization & decision-making
The thesis synthesis: how forecasting, risk modelling and multi-objective optimisation fit together into one decision process instead of three disconnected steps.
Real portfolio decisions weigh more than two goals at once. A benchmark for algorithms that handle many objectives over multiple periods.
Trading rules are usually tuned for return alone. Here they are optimised against several competing criteria at the same time.
A portfolio built for a calm market is the wrong portfolio for a turbulent one. This conditions the decision on the regime the market is actually in.
Methods that assume a specific probability distribution fail when the assumption breaks. This asks what can still be guaranteed without making that assumption.
A strategy that works on paper can be impossible to execute at scale. This adds real capacity limits and extends the approach across several assets.
Reinforcement learning normally optimises one reward. This asks it to hold the risk–return trade-off open instead of collapsing it into a single score.
Forecasting & volatility modeling
Does regime-switching volatility actually beat the simpler alternatives on grain prices, or only look more sophisticated? A direct comparison. See the case study
Skew and kurtosis in crop prices are not constant, yet most models treat them as if they were. GAMLSS lets them move with time.
Some of what looks like random noise in economic data is deterministic structure. This examines which is which, and what follows for forecasting.
Market structure, breaks & early warnings
A series changes course — but was it caused by the event everyone blames, or did it merely coincide with it? Causal methods applied to break detection.
Spotting a break after it happens is a different problem from anticipating one. This separates the two and asks how much of the second is achievable.
When an international price moves, how much of it reaches the Brazilian producer, and how long does it take? Measured across four commodities.
An early-warning system that mistakes a normal seasonal swing for a crisis is worse than none. This makes anomaly detection aware of the regime, for food-security warnings.
Foundations & applications
A systematic review mapping what machine learning and classical econometrics have each contributed to commodity portfolio work, and where they have not yet met.
The value of being able to wait, expand or abandon an investment, estimated with Bayesian methods that carry the uncertainty through instead of hiding it in a point estimate.
Advising
Peer review & committees
Reviewer — Applied Soft Computing
Peer reviewer for the Applied Soft Computing journal (Elsevier), 2024 – present. Reviewer certificate
Reviewer — PRINCIPIA
Peer reviewer for the PRINCIPIA journal (UFJF, Juiz de Fora), 2024 – present. Journal website
Scientific Advisory Committee
Scientific Advisory Committee member at PUCPR (2023; 2024–2025), supporting the evaluation of research and scientific initiation programs.