Data scientist at IPEA · PhD in Mathematics (TUM)

Google Scholar · GitHub · LinkedIn · Lattes · ORCID

I work on statistics for time series, stochastic processes and indirect inference, with a focus on simulation-based estimation of stochastic volatility models. I also have a background in cryptography, mainly lattice-based and post-quantum schemes.

I believe in open source: I publish my research code and enjoy contributing to other people's projects.

• Data Scientist, IPEA — present
• Researcher in Cryptography, CEPESC — previous

• PhD in Mathematics, Technical University of Munich (2019) Thesis: Simulation-based estimation of time series and stochastic volatility processes Advisor: Claudia Klüppelberg · Visiting researcher at the Department of Statistics, Columbia University · CNPq scholar
• MSc in Statistics, University of Brasília (2013) Thesis: Modelos combinados AR-GARCH governados por distribuições estáveis (AR-GARCH models driven by stable distributions)
• BSc in Mathematics and BSc in Statistics, University of Brasília (2010)

All repositories: github.com/thiagodoregosousa

**R packages **

• GEVStableGarch (on CRAN) — ARMA-GARCH and ARMA-APARCH models with stable, GEV and GAt innovations: estimation, simulation, forecasting and Value-at-Risk. docs · source
• bgev (on CRAN) — bimodal GEV distribution: density, distribution function, quantiles, random variates and MLE. docs · source
• mbev — multimodal bivariate extreme value (MBEV) distribution (R).
• cran_journey — a guide to making your first R package and getting it onto CRAN.

SageMath code for cryptography papers

• matrix-ntru — matrix NTRU (Nayak et al., 2008) and its LLL attack (do Rego Sousa and Carneiro, 2024).
• ETRU — the ETRU cryptosystem (Jarvis and Nevins, 2015) and May's dimension-reduction idea for the LLL attack.
• integral_domain_matrix_ntru — matrix NTRU over an integral domain (Wijayanti et al., 2023).

Full list on Google Scholar.

• Davis, Sousa, Klüppelberg. Indirect inference for time series using the empirical characteristic function and control variates. J. Time Series Analysis 42 (2021).
• Sousa, Stelzer. Moment-based estimation for the multivariate COGARCH(1,1) process. Scand. J. Statistics 49 (2022).
• Sousa, Haug, Klüppelberg. Indirect inference for Lévy-driven continuous-time GARCH models. Scand. J. Statistics 46 (2019).
• Otiniano, Oliveira, Matsushita, Sousa. Modeling multimodal bivariate extreme values: theory, estimation, and applications. Commun. Nonlinear Sci. Numer. Simul. (2026).

• Assis, Sousa. Matrix polynomial attack on the Megrelishvili key exchange protocol. IACR Communications in Cryptology 2(4) (2026).
• Lagrota, Neto, Sousa, Maia. Optimizing ML-KEM with OpenMP: a multilevel parallelization study. SBSeg (2026).
• Sousa, Neto. Lattice basis reduction attack on matrix NTRU. SBSeg (2024). 🏆 Best full paper in Cryptography.

• Best full paper in Cryptography, SBSeg 2024
• 2nd place, Best Master's Dissertation in Statistics 2013–2014, Brazilian Statistical Association (ABE)

• Referee: Journal of Internet Services and Applications; Journal of Econometrics - 2018
• Referee: SBSEG 2024 - Current

Portuguese, English, German

Anonymization, post-quantum cryptography, random number generator testing, stable distributions, COGARCH, indirect inference, lattice attacks

R, Python, SageMath, C/OpenMP