LLM Agents Predicting the Valuation of Goods

How to cite

Kazinnik, Sophia; Nguyen, David; Enriquez, Jose Ramon. 2024. "LLM Agents Predicting the Valuation of Goods." AEA RCT Registry. AEARCTR-0014817. https://doi.org/10.1257/rct.14817-1.2

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Sophia Kazinnik* , David Nguyen , Jose Ramon Enriquez

* Primary investigator

Registry ID
AEARCTR-0014817
DOI
https://doi.org/10.1257/rct.14817-1.2
First registered
Last updated
Status
On going
Trial dates
2024-11-01 – 2025-05-31
Countries
United States of America ()
Registry record
http://www.socialscienceregistry.org/trials/14817

Abstract

This study evaluates the capability of Large Language Models (LLMs) to predict individuals' valuation of digital and analog goods, specifically their willingness to accept (WTA) compensation for foregoing services like Facebook. Using survey data from Coyle and Nguyen (2023), which includes demographic information and WTA valuations for various goods, we investigate whether LLMs can simulate individual responses when provided with personal characteristics and prior valuations.

To further assess the applicability of LLMs in predicting valuations, we conducted an extension by surveying participants on Prolific using a new subset of digital and analog goods. We then replicate this survey with LLMs employing the same methodology. The comparison between human and LLM responses provides additional insights into the strengths and limitations of LLMs in capturing individual valuations.

Experimental design

We replicate the survey structure using a number of large language models, and compare the accuracy of the synthetic outcomes with the ones produces in the actual survey(s).

Design details

Randomization method
Randomization is applied within the survey structure and is replicated with our synthetic set-up.
Randomization unit
Individual; this is a survey + synthetic replication.
Clusters
This is a survey at an individual level, there are around 10,000 observation in total (not all will be included/used in the main analysis)
Observations
This is a survey at an individual level, there are around 10,000 observation in total (not all will be included/used in the main analysis)
Arms
This is a survey at an individual level, there are around 10,000 observation in total (not all will be included/used in the main analysis)
Primary outcomes
The key outcome is the accuracy with which our set of LLMs replicate the survey outcomes.

Cited by

No citation data available for this registration yet.

Keywords