Thread by @DinaPomeranz

Next up at NBER DEV: Natalia Rigol on "Targeting High Ability Entrepreneurs Using Community Information: Mechanism Design in the Field"

With Reshmaan Hussam and Ben Roth:
papers.nber.org/conf_papers/f1…

This paper is a nice application of some of the concepts described by Esther Duflo in her Master Lecture yesterday, combining field work, causal inference and machine learning.

Research context: could information provided by community members help target grants to entrepreneurs who may have the highest economic returns from receiving them?

Grants were randomly assigned in a public lottery. Researchers then tested whether their peers had correctly predicted who had the highest economic return to the grant.

Indeed, those entrepreneurs whom peers had ranked positively not only had higher profits in general, but also benefitted more from the grant in terms of increasing their profits even further.

Those ranked on the top responded differently to the grant. More additional work put in, etc.

Average return to the grant was 8%. Return for those in the top group was 23%.

Next step: testing whether they could predict the returns as effectively using observable characteristics when using the observations by peers.

For this, they use machine learning.

Since the sample was small, they did not have power to throw out part of the sample as the training sample for the prediction. Instead they trained it on a sample from a similar intervention in another country.

They can also detect that sometimes entrepreneurs are strategically lying.

They create an incentive mechanism to encourage more truthful reporting, which doesn't rely on ex-post data collection and allows them to pay people right away.

Main takeaways:

Discussion by Chris Woodruff.

Main comment: Wow!

The authors do many different things very effectively in this project: