Yizhou Jin

A U.S. auto insurer offered drivers a discount for sharing their driving data: safer drivers selected in, and drivers became 30% safer while monitored. Our model shows that mandating data portability, or restricting how the monitoring data are priced, would reduce monitoring and hurt consumers.

Italian auto insurers differ in how precisely they rate risk. Insurers that rate risk more precisely, including some smaller and more specialized firms, win safer customers despite higher costs, and a shared risk bureau would erode their edge yet raise consumer surplus by 15.7%.

Using second-by-second smartphone records from rideshare drivers, I find that handheld phone use more than doubles the odds of a collision, yet drivers barely cut it when rain or a higher deductible raises their risk or cost; a one-time warning reduces it by about 19%, with the largest proportional responses among the heaviest users.

On a platform where more than 20,000 sellers register each day, fewer than 1% become viable businesses. Entrants shown beside established sellers in the same search win buyers’ purchases, yet reach far fewer buyers’ consideration sets, pointing to discovery rather than product quality as the barrier to entry.

In a randomized experiment with 712,118 new sellers, free, algorithmically customized training raised monthly revenue by 2.2%, much of it through advertising while sellers had few ratings.

Using behavioral driving data, we estimate a model of risky action, attention and insurance that separates responses to financial incentives from the effect of making risk salient, and use it to design contracts that price a directly measured behavior while keeping insurance against accidents.

A large e-commerce platform sells customer and market analytics to its sellers. Small sellers’ demand for analytics is especially price-sensitive, sellers induced to buy them achieve higher conversion and more buyers, and in our model cheaper access raises sales and lowers concentration within the estimation sample.