“If we knew what it was we were doing, it would not be called research, would it?” - Albert Einstein
Current Research Projects
Non-Experimental Estimates of Individual Treatment Effects via Machine Learning
I am currently working on combining techniques from Machine Learning and Econometrics to develop methods for providing non-experimental estimates of (causal) Individual Treatment Effects. This is useful in settings in which there is a lot of data on a single individual available, for example Smart Meter consumption data in the context of Energy Efficiency and Demand Response.
Modeling of Consumer Behavior using Smart Meter Data
This project is based on a data set of consumers participating in residential demand response. Users registered in the program receive incentives (monetary and non-monetary) to reduce their energy consumption at certain times of the day (e.g. when Locational Marginal Prices are unusually high). The resulting flexibility in demand can then be bundled and sold in the wholesale electricity market. The overarching goal of this project is to develop statistical models for customer behavior based on large amounts of Smart Meter data and other data sources (e.g. weather and location data) in order to increase the predictability of user behavior.
Regret Minimization on Large Spaces and Learning Equilibria in Continuous Games
In this project we explore the problem of decision-making under uncertainty in large spaces. Many online learning algorithms (such as Exp3, Exp4, UCB, etc.) are designed for finite action spaces. However, in many cases, the decision space is continuous, for example in pricing problems. Starting from a general analysis of Dual Averaging on Reflexive Banach spaces, we have developed an online learning algorithm, which, in an adversarial setting with complete observation, achieves sub-linear regret under rather weak assumptions (Lipschitz continuity and local fatness of the action set). The results can be used to learn Nash Equilibria via repeated play of Continuous Games.
Incentive Design for Aggregating Demand-Side Resources in Electricity Markets
This project we investigate the incentive design problem that an energy aggregator who participates in the wholesale electricity faces. Specifically, buildings agree to adjust their power consumption (e.g. HVAC) according to a signal sent by the aggregator. In return, the aggregator offers monetary rewards to incentivize the individual buildings. This allows the aggregator to bundle the flexibility of different buildings to offer it in form of a product that is compatible with the wholesale market. We frame the problem of determining the optimal incentives as an optimal contract design problem.
Past Research Projects