Jesse Silbert

Hi! I am a sixth year Ph.D. student at the Princeton University Department of Economics interested in industrial organization, market design, labor markets, online platforms, and matching theory.

In addition to economics, I am passionate about rock climbing, weird music, and good food.

Email: jesseas@princeton.edu

Research:

Large Language Models and Signaling in Online Labor Markets (with Anaïs Galdin). [Work in progress]

Abstract:

Large language models have the potential to transform the landscape of online labor marketplaces, not just through the channels of supply and demand but also the matching technology itself. In this paper, we use data from Freelancer.com, a major online labor platform, to investigate exactly how the advent of large language models (LLMs) affects the matching process in online labor markets. Our first major contribution is to develop a new measure of “content fit” that uses a LLM to quantify how well a cover letter responds to a given job description. We show that this measure is significantly predictive of labor demand, and that job posters have a high willingness to pay for increases in the “content fit” of cover letters. Motivated by this finding, we develop a Spence (1973) style model of signaling, in which Freelancers spend time (costly effort) to increase their cover letters’ levels of “content fit”, and the freelancers with a greater match quality with a given job posting are more easily (less marginally costly) able to expend that effort. This model along with our descriptive findings lend evidence to the theory that job posters are using “content fit” as a noisy but predictive signal of freelancer match quality. With this theory in hand, we then ask what happens to the efficiency of this signaling equilibrium when LLMs make it cheaper to increase “content fit”, thus garbling the signal. To do so, we estimate how the relationship between time spent writing and “content fit” changes after the introduction of LLMs onto the platform, and then simulate a signaling equilibrium, in which supply and demand are held fixed as they are in the pre-LLM time periods, but the signaling technology of the post-LLM time periods is used instead. One of our key contributions is to develop estimation and identification strategies of scoring auctions where one of the dimensions of the score is a signal for quality whose weighting in the score is determined by the equilibrium itself. These strategies allow us to recover both supply and demand, as well as the relevant parameters of the signaling equilibrium.

Job Matching without Price Discrimination (with Wilbur Townsend). [Previously circulated as “Stable Matching in Monopsonistic Labor Markets”]

Abstract:

In many labor markets, firms do not price discriminate among their workers. In this paper, we study how a labor market with uniform salaries matches workers to jobs. To do so, we construct a job matching model in which each firm views workers as interchangeable and must pay all its workers the same salary. While an efficient stable outcome always exists, inefficient outcomes can be stable as well. Workers’ preferred stable outcome is efficient. In contrast, firms prefer inefficient stable outcomes in which they pay lower salaries. Though a strategyproof mechanism that implements an efficient stable outcome can elicit how workers value employment, it cannot elicit firms’ production technologies.