EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
1237. Giving Deep Attention to Consumer Preferences with Large Language Models
Contributed abstract in session MA-59: Pricing and applications 3, stream Pricing and Revenue Management.
Monday, 8:30-10:00Room: S08 (building: 101)
Authors (first author is the speaker)
| 1. | Joshua Foster
|
| Ivey Business School, Western University | |
| 2. | Fredrik Odegaard
|
| Ivey Business School, Western University |
Abstract
We use the natural language descriptions of exotic and collectible vehicles from online auction markets to semi-nonparametrically estimate the primitives of demand (private valuations, number of potential bidders) for each individual vehicle in our dataset. The first stage of our estimation method fine-tunes a large language model (RoBERTa, Liu et al., 2019) to predict the inputs of our demand estimators using the descriptions provided for each vehicle. We then append a multi-layer perceptron to the tuned language model that projects into the parameter space of these estimators so that a second stage training can recover the primitives of interest. Our identification method relies on the structural features of specific bids submitted in these markets, which can proxy for valuations under our data generation assumptions. Finally, we demonstrate how this model can generate counterfactual analyses using new natural language descriptions within the defined product space.
Keywords
- Auctions / Competitive Bidding
- Machine Learning
- Economic Modeling
Status: accepted
Back to the list of papers