STA 290 Seminar: Jay Wang (Hewlett Packard)

Statistics Seminar: STA 290

Thursday, May 23rd, 2013 at 4:10pm, MSB 1147 (Colloquium Room)
Refreshments at 3:30pm in MSB 4110 (Statistics Lounge)

Speaker: Jay Wang (Hewlett Packard Labs, Palo Alto)

Title: "Boosted Semiparametric Multinomial Logit Model for Leveraging Aggregated Sales Data"

Abstract: Building a powerful and reliable demand model is critical to strategic and tactical pricing and portfolio management. Practical questions like {how customers perceive the difference between 8GB and 4GB RAM} or {what happens if the price of a product category decreases by 50 dollars} often arise when optimizing price or product mix. Motivated by the need to flexibly estimate demand, we propose two semiparametric choice models, namely the varying-coefficient multinomial logit (MNL) model and partially linear MNL model. Both models are estimated via boosted trees which use a sequence of trees to approximate the nonparametric component of the model. The performance of the proposed methods are compared with linear and quadratic MNL models on a real-world aggregated sales data. Industrial applications in pricing and portfolio management as well as some practical concerns are commented on.