Determining consumer preferences and utility is a foundational challenge in economics. They are central in determining consumer behaviour through the utility-maximising consumer decision-making process. However, preferences and utilities are not observable and may not even be known to the individual making the choice; only the outcome is observed, often in the form of demand. Uncovering the shape of utility functions is important, as its curvature governs a consumer's willingness to substitute between goods, providing the key causal mechanism for predicting their real-world responses to price changes, but is largely left unexplored.
In this talk, Marta Grześkiewicz will present an algorithm for uncovering a utility function based on observational consumption data. The algorithm, Preference Extraction and Reward Learning (PEARL) is able to uncover a representation of the utility function that best rationalises observed consumer choice data given any specified functional form. Towards this, she introduces a flexible utility function, the Input-Concave Neural Network which is a neural network with concave activation functions that is able to capture complex relationships across goods, including cross-price elasticities. The method is shown to obtain near-zero errors on counterfactual predictions on simulated noise-free and noisy data.
About the speaker
Marta Grześkiewicz is a College Assistant Professor (Early Career Fellowship), Director of Studies, and Fellow in Economics at St John’s College, University of Cambridge. She holds a BA in Economics from the University of Cambridge, and an MSc in Data Science and PhD in Economics and Machine Learning from UCL. Her research interests lie at the intersection of economics and machine learning. Her current projects involve developing algorithms to model choice and decision-making by economic agents, with applications in consumer theory and behavioural finance, agent-based modelling with machine learning in banking and finance, and the integration of economic theory into machine learning models for economic forecasting.
Registration
Please register using the form below and contact complexity@inet.ox.ac.uk if you have any queries. Our seminars are run in person and online but are not recorded.