Polina Borisova

Working Papers

"TikToks vs. Movies: How Content Length Shapes Engagement" (Job Market Paper) (current draft)

When should content be short or long? I answer this question with a model in which a creator develops a sequence of videos that users consume one at a time. The creator’s quality determines how likely each video is to be good, so watching one video provides information about both the video itself and the creator. Short content lets users watch more videos from the same creator and learn about the creator’s quality faster. Long content keeps them with each one longer, allowing more learning at the video level and greater value extraction when a video proves good. In light of this trade-off, I show that the optimal content length depends on the cost of initiating consumption and on whether the user is still searching for a high-quality creator or has already identified one. This mechanism links platform monetization to content length and rationalizes the use of short content as an entry point into longer-form consumption. 

"Self-Prospecting: Optimal Experimentation Under Present-Bias" (with Nikhil Vellodi), SSRN.

We study dynamic information provision to a present-biased decision maker (DM) who faces an experimentation problem. In our model, rewards arrive independently of the hidden state, so that under full information, beliefs remain constant prior to conclusive news arrival. Our main finding is that coarsening the DM’s information to induce a continuously increasing path of beliefs during active experimentation is optimal. Doing so links current and future experi- mentation choices in a complementary fashion, boosting incentives and ameliorating the DM’s lack of self-control. We apply our findings to various settings, including motivational design for talent discovery, short-termism in organizations and policy-making, and optimal self-delusion. 

Work in Progress

"Targets That Motivate, Targets That Inform" 

I study a dynamic principal-agent model in which targets simultaneously function as motivational reference points and as Bayesian screening devices. The principal trades off incentive provision against learning about the agent's type. This creates a novel dynamic pattern: after an unmet target, it can be optimal to raise subsequent targets rather than relax them. 

Other Publications(Mathematics)

"Separation of Variables for Type D_n Hitchin Systems on Hyperelliptic Curves"

Russian Mathematical Surveys 76(2), 181–182 (2021). https://doi.org/10.4213/rm9935, ArXiv.

"Hitchin Systems on Hyperelliptic Curves" (with Oleg Sheinman)

Proc. Steklov Inst. Math. 311, 22–35 (2020). ArXiv.