Economist & Economic Consultant

Jeff Rowley

PhD graduate specialising in causal inference and working in antitrust

About

I hold a PhD in Economics from University College London (UCL), specialising in microeconometric application and theory. My academic research centred on causal inference and the application of statistical decision theory: identifying the effect of different policy interventions on groups of individuals under weak assumptions, and learning which mix of interventions achieves the best outcome using sequential machine learning methods like contextual bandits algorithms.

I currently work as an independent contractor, engaged by the Dante Quaglione Practice at Berkeley Research Group (BRG) in ongoing litigation in the UK and abroad. During quieter periods, I collaborate with Dante on research in competition and innovation.

I enjoy the challenge that frontier methods present—the challenge of understanding advanced theory and translating it to useable estimators and code that can be applied to measure real-world effects. My past experience of teaching Economics—including an advanced course in mechanism design at PhD level—and academic speciality demonstrate my adaptability to work in and pursuit of deep knowledge of different areas of Economics and beyond, and of my commitment to rigour.

Research

Published

Bandit Algorithms for Policy Learning: Methods, Implementation, and Welfare-Performance

📖︎ Japanese Economic Review

✍︎ with Toru Kitagawa

Studies how a sequential bandit algorithm (EXP4.P) can be used to assign individualised treatments in real time, rather than learning a policy after a trial has ended. Establishes welfare-regret guarantees and tests the proposed approach on data from the US Job Training Partnership Act Study sample.

Stochastic Treatment Choice with Empirical Welfare Updating

✍︎ with Toru Kitagawa & H. Lopez

Proposes a way to choose assignment rules (who should be treated, and in which way) by assuming a prior over assignment rules and updating this prior using an empirical welfare criterion rather than a likelihood. Derives an optimal updating rule and a tractable variational approximation—with convergence guarantees—applied to data from the US Job Training Partnership Act Study sample.

Von Mises–Fisher Distributions and their Statistical Divergence

✍︎ with Toru Kitagawa

Derives analytical measures of statistical divergence between von Mises–Fisher distributions—a family of distributions with support on the hypersphere that has broad potential in microeconometric problems involving treatment assignment and discrete choice.

Also in progress:

Consulting & Teaching

CV & contact

✉︎ Email — jeffreyrowley@duck.com