Please use this identifier to cite or link to this item: https://hdl.handle.net/10419/241940 
Year of Publication: 
2021
Series/Report no.: 
cemmap working paper No. CWP04/21
Publisher: 
Centre for Microdata Methods and Practice (cemmap), London
Abstract: 
We develop empirical models of hedonic prices and derive hedonic indices for measuring changes in customer welfare based upon deep learning. We first generate abstract product attributes, or "features," from text descriptions and images using deep neural networks, and then use these attributes to estimate the hedonic price function. Specifically, we convert textual information about the product to numeric product features using the ELMO or BERT language models, trained or fine-tuned using Amazon's product descriptions. We convert the product image to numerical product features by a pre-trained ResNet50 image model. To produce the estimated hedonic price function, we use a multi-task neural network again, trained to predict the price of a product simultaneously in all time periods. We apply the models to Amazon's data for first-party apparel sales to estimate hedonic prices. The resulting models have high predictive accuracy, with R2 ranging from 80% to 90%. We also construct hedonic price indices: over the period 2013-2017 the hedonic Fisher price index decreased, providing improvement in customer welfare.
Persistent Identifier of the first edition: 
Document Type: 
Working Paper

Files in This Item:
File
Size





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.