Francisco Rosales, Ph.D.

Research Projects

• Conversational Hybrid Inoculation Mechanism for Epistemic Resilience Architecture (CHIMERA). CHIMERA is an LLM-based, conversational simulation platform designed to study how individuals respond to persuasive misinformation in interactive settings. The system operationalizes misinformation exposure as a game between a human participant and an AI persuasion agent, with outcomes governed by transparent, incentive-compatible scoring rules implemented via tokenized rewards. By embedding well-documented misinformation strategies—such as emotional manipulation, narrative coherence, and authority appeals—within a controlled adversarial interaction, CHIMERA enables the systematic evaluation of epistemic resilience, belief updating, and susceptibility to persuasion. The platform is intended as a research testbed for computational social science, AI alignment, and misinformation studies, allowing reproducible experimentation on how conversational agents influence belief formation under real incentives.
• Interpretable Deep Reinforcement Learning for Dynamic Truck Dispatching in Open-Pit Mining. Dynamic resource allocation under uncertainty remains a central challenge in operations research, particularly in capital intensive industries such as open pit mining, where inefficient dispatching decisions can result in major productivity losses and higher operational costs. This paper presents a Deep Reinforcement Learning framework for the dynamic truck assignment problem in open pit mining, with a dual emphasis on performance optimization and policy interpretability. The problem is modeled as a Markov Decision Process and implemented within a discrete event simulation environment that captures the stochastic behavior of truck and shovel interactions. A Deep Q Learning approach with neural network function approximation is used to learn adaptive dispatching policies directly from simulated experience. To promote transparency, Explainable Artificial Intelligence techniques based on Shapley values are applied to interpret and validate the learned decision strategies. Experimental results demonstrate that the proposed framework substantially outperforms heuristic and random dispatching methods in terms of resource utilization and operational efficiency.
• Data-Driven Price Elasticity Estimation for Decision Support in the Global Tin Market. In this document we estimate supply and demand elasticities in the global tin market with the objective of supporting production planning, investment evaluation, and strategic decision-making. Using a unified dataset combining financial indicators, geopolitical time series, industrial activity indices, and proprietary production data, we develop a causal inference pipeline based on instrumental variables and two-stage least squares (2SLS). Our approach integrates standard econometric identification techniques with data science practices such as feature engineering, model validation, and structural robustness evaluation. We estimate monthly and annual elasticities for major producing and consuming regions between 2000--2024. Demand is systematically inelastic across all regions, while supply responses show strong heterogeneity: Peru exhibits the highest responsiveness, Indonesia displays elastic short-run adjustments driven by artisanal mining dynamics, and China shows near-zero responsiveness. We interpret these results within a decision-science framework, highlighting implications for production optimization, risk management, and price forecasting. The study illustrates how combining econometric causal identification with data-driven analytics can enhance decision processes in extractive industries and critical minerals markets.

Selected Publications

Journal Articles
• Rosales, F., G. Campos. Assessing out-of-sample performance of orthogonal portfolio rules in emerging markets. Finance Research Letters, 2025.
• Perez, I., F. Rosales, and L. Duffaut. Feedback dynamic control for exiting a debt-induced spiral in a deterministic Keen model. PLOS ONE, 2024.
• Dietrich, S., A. Meysonnat, F. Rosales, V. Cebotari, and F. Grassmann. Economic development, weather shocks and child marriage in South Asia: A machine learning approach. PLOS ONE, 2022.
• Serra, P., T. Krivobokova, F. Rosales, and K. Klockmann. Joint non-parametric estimation of mean and auto-covariances for Gaussian processes. Computational Statistics and Data Analysis, 2022.
• Chávez-Bedoya, L., and F. Rosales. Orthogonal portfolios to assess estimation risk. International Review of Economics and Finance, 2022.
• Chávez-Bedoya, L., and F. Rosales. Reduction of estimation risk in optimal portfolio choice using redundant constraints. International Review of Financial Analysis, 2022.
• Martini, J., F. Rosales, N. Ha, T. Kneib, J. Heisse, and V. Wimmer. Lost in translation: On the impact of data coding on penalized regression with interactions. G3: Genes, Genomes, Genetics, 2019.
• Duffaut, L., F. Rosales, and A. Posadas. Embedding spatial variability in rainfall reconstruction. International Journal of Remote Sensing, 2018.
• Fuenzalida, D., S. Mongrut, M. Nash, and F. Rosales. Anticipating turbulent periods in Latin American emerging markets. Revista de Contaduría y Administración, 2009.
• Rosales, F., A. Posadas, and R. Quiroz. Multifractal characterization of spatial income curdling: Theory and applications. Advances in Complex Systems, 2008.
• Chávez-Bedoya, L., and F. Rosales. A non-Gaussian approach in value-at-risk computation in non-liquid markets. Revista de la Superintendencia de Banca y Seguros del Perú, 2007.
• Books & Book Chapters
• Cortez, R., and F. Rosales. 340 Exercises in Microeconomics (in Spanish). Research Center of Universidad del Pacífico, 2005.
• Tecuapetla-Gomez, I., F. Galicia, and F. Rosales. Characterization of NDVI time series in Mexican forests via functional data analysis applied to satellite images (in Spanish). In: Geomatics Applications in the Study of Sustainability, Vol. 2. Universidad de Nuevo León, 2023.
• Conference Talks & posters
• F. Rosales and A. Diaz. Interpretable Deep Reinforcement Learning for Dynamic Truck Dispatching in Open-Pit Mining. In: 12th International Conference on Management and Industrial Engineering., Bucharest, Romania, 2025.
• Martinez, D., C. Garcia, S. Delgado, M. Flores, R. Rodriguez and F. Rosales. Algorithmic Trading for Portfolio Optimization in Financial Markets. In: 4th Congress on Smart Computing Technologies, Sikkim, India, 2025.
• Rosales, F. Revisiting “Embedding Spatial Variability in Rain-field Reconstruction”. In: Joint Statistical Meetings of the American Statistical Association, Atlanta, USA, 2020.
• Rosales, F. Detrending Dependent Data with Bayesian Adaptive Splines. In: 32nd Annual Conference of the Mexican Statistical Society, Mexico City, Mexico, 2017.
• Rosales, F. Wavelet-based Techniques for Reconstructing Signals with Missing Data. In: International Financial Conference, Santiago, Chile, 2017.
• Rosales, F. Empirical Bayes Smoothing Splines with Correlated Noise to Study Mortality Rates. In: Mathematical and Statistical Methods for Actuarial Sciences and Finance, Paris, France, 2016.
• Rosales, F. Analysis of Price Transmission using a Nonparametric Error Correction Model with Time-Varying Cointegration. In: International Conference of Agricultural Economists, Milan, Italy, 2015.
• Rosales, F. Instant Trend-Seasonal Decomposition with Splines. In: German Probability and Statistics Days, Ulm, Germany, 2014.
• Rosales, F. Time Series Decomposition in Illiquid Markets. In: Bachelier Finance Society Conference, Sydney, Australia, 2012.
• Rosales, F. Multifractals in Spatial Income Distribution. In: Congreso Latinoamericano de Probabilidad y Estadística Matemática, Lima, Peru, 2007.
• Rosales, F. Multifractal Characterization in Spatial Income Curdling. In: European Conference on Complex Systems, Dresden, Germany, 2007.
• Rosales, F. Notes on the Adequacy of the Brownian Motion Hypothesis in Non-Liquid Markets: Are Jump Processes or Discrete-Time Ones a Better Alternative? In: Conferencia Internacional de Finanzas, Santiago de Chile, Chile, 2006.
• Rosales, F. Anticipating Turbulence in Latin-American Emerging Markets: The Mexican Crisis of 1994. In: Conferencia Internacional de Finanzas, Santiago de Chile, Chile, 2006.

Consulting

Francisco Rosales works with organizations on artificial intelligence strategy, analytics transformation, and executive-level capability building. His consulting practice focuses on helping senior leadership teams translate data and AI investments into measurable business value, aligning advanced analytics initiatives with strategic priorities, governance frameworks, and operational decision-making. He advises organizations on how to scale data science responsibly, integrate AI into core processes, and build internal analytical capabilities that support sustainable growth, risk management, and competitive advantage.

Contact

Email: francisco.rosales-marticorena@protonmail.com
Linkedin: https://linkedin.com/in/fraroma
GitHub: https://github.com/lfrm