My substantive research interests are Marketing Analytics, e.g. Retail Analytics and e-Commerce, Pricing and Promotions, Choice Modeling.

My methodological interests are Bayesian Econometrics, Probabilistic Machine Learning and Bayesian Nonparametrics.

Work In Progress


Relaxing Functional Form in Choice Models Through Gaussian Processes


Will They Come Again, and Why? A Bayesian Purchase Forecasting Model

Will They Buy Again? and Why: A Bayesian Purchase Forecasting Model

Managers need to allocate marketing efforts in a timely and effective manner under
scarce data constraints when dealing with retention of recently acquired customers. In our
first essay, we build an interpretable, continuous-time, Bayesian purchase forecasting
model for customer retention using transaction and clickstream data. By fusing together
different types of structured and unstructured data, we are able to cluster customers into
transaction and clickstream segments, which are used to predict customer activity. Our
model allows us to automatically discover probabilistic relationships between the timing
of transactions and clickstream activity, and predict distributions of time-to-event at the
individual level.

Working draft available upon request.

Relaxing Functional Form in Choice Models Through Gaussian Processes

we show how expansion in a product category moderates variety
seeking. Consumers rapidly reach satiation for low quality products and switch to higher
quality products. The imposed functional form of utility in choice models strongly
influences the range of estimable substitution patterns across goods. Using Gaussian
process priors on direct utility functions, we relax the functional form of utility and estimate a
nonhomothetic choice model with simultaneous purchases under constrained utility
maximization. Our model captures nonlinear rates of satiation across low and high
quality products, that traditional models fail to capture. We provide more precise and
robust statistical inference.

Working draft available upon request.