Alexander Michalka

As an undergraduate, I studied applied math at Berkeley. I had an interest in optimization, and began taking OR courses around my third year. Between college and grad school I worked for three years as a quantitative researcher at a startup called WeatherBill in San Francisco, building statistical weather models for pricing parametric weather insurance contracts.
In 2009 I started my PhD at Columbia, where my work with Dan Bienstock focused on cutting plane methods for optimization problems with convex objective functions but non-convex domains. Problems of this sort arise in signal processing, semiconductor manufacturing, and portfolio optimization, often with an objective that is a convex quadratic, representing a measure of noise, power, or variance.
We were able to derive polynomial-time separation procedures in some special cases where the objective is quadratic and the feasible region is defined by the complement of a polyhedron, the complement of an ellipsoid, or a union of polyhedra. We also wrote a paper on a solution method for a generalization of the Trust-Region Subproblem from nonlinear programming that includes convex polyhedral constraints.

After finishing my PhD I took a position within the Global Stock Selection research team at AQR Capital Management, a systematic asset manager based in Greenwich CT. At AQR I work mostly on research projects involving portfolio construction and optimization.
Optimization is a key part of our process at AQR - it is how we translate our "model", which is constructed in a world with no costs or constraints, into actual traded portfolios. I've worked on a variety of OR-related projects, specifically: methods of controlling certain types of risk exposures in our portfolios, allocating leverage to strategies within a multi-strategy fund, and optimal weighting of signals within our model.