James A. Brofos quantitative strategies researcher

I am a quantitative strategies researcher at SESCO Enterprises, which engages in the proprietary trading of electricity products in the United States.

Before working in quantitative strategies, I was a Ph.D. student in the Department of Statistics and Data Science at Yale University supported by a NSF Graduate Research Fellowship. My research interests revolved around designing intelligent systems that leverage new computational methods in statistical inference, and in developing algorithms that would generalize, comprehend uncertainty, and scale to problems of practical significance. I graduated with a Ph.D. in Statistics and Data Science in 2022.

Before coming to Yale, I was a data scientist at The MITRE Corporation, which operates federally-funded research and development centers. At MITRE, I assisted the United States government with scientific research, analysis, development, acquisition, and systems engineering.

I did my undergraduate education at Dartmouth College, where I studied mathematics and computer science. I graduated with a B.A. in 2015.


I have historically been interested in statistics and machine learning with a Bayesian flavor. My research encompassed statistical research in engineering disciplines, cybersecurity, and classical antiquity. I have worked in scalable estimation for graphical models, variational inference, robustness to errors-in-variables, Bayesian machine learning, and methods of Markov chain Monte Carlo.


Journal Articles

Conference Articles

Workshop Articles


I have been involved in software development in graduate school and in my experiences in government contracting. Below are some of my favorite software projects which span applications including algorithmic finance, Bayesian methods statical modeling, and procedures for black-box function optimization.

Iliad and Odyssey: Geometric Hamiltonian Monte Carlo

Iliad implements geometric methods of Hamiltonian Monte Carlo that incorporate second-order geometric information about the target probability distribution into the Markov chain. The software is pure NumPy and SciPy and gives implementations of Hamiltonian Monte Carlo, Riemannian Manifold Hamiltonian Monte Carlo, and Lagrangian Monte Carlo. The Odyssey software gives implementations of several baseline Bayesian inference tasks including Neal's funnel distribution, a log-Gaussian Cox-Poisson model, a stochastic volatility model, and inference in a differential equation model.

Thor: Bayesian Optimization

Implements an API and user interface that facilitates Bayesian optimization of machine learning systems. Includes state-of-the-art advances in hyperparameter search such as portfolios of acquisition functions, distributed optimization, and low-discrepancy pseudo-random restarts for kernel parameter tuning. Thor possesses client-side interfaces for Python, R, and MATLAB.

Sif: Gaussian Processes

Implements Bayesian non-parametric inference via Gaussian processes. Sif includes acquisition functions for Bayesian optimization, kernels and approximate kernel inference, and support for fully Bayesian inference with elliptical slice sampling. Sif's modules give low-level access to Bayesian linear and logistic regression models and develops this machinery into Gaussian regression and classification processes.

Odin: Algorithmic Trading Infrastructure

Implements an event-driven live trading (via Interactive Brokers) and backtesting infrastructure in Python. Supports a modular design that allows retail traders to leverage low-level control over signal generation, portfolio management, equity rebalancing, and data streams. Odin integrates closely with a dedicated Postgres equities database and incorporates modern portfolio performance metrics.