Shama Research

Don't fund winners.

Two women apply to the same small fund on the same day. Both want ₹40,000 for a tailoring shop. One of them has some savings and a cousin who works at a bank. The other has nowhere else to go.

Most funds pick the first woman. She is the safe choice. She is also the one who would have been fine without the loan.

The applicant most likely to succeed is usually the one who needs you least.

A working paper and an open source engine. Free, non-commercial.

Two women sit side by side across a table from a fund officer, waiting on his decision.

The one it changes

For the woman with nowhere else to go, the loan is the whole difference. Her shop grows or it does not, and that comes down to you. For the safe applicant, the loan mostly speeds things up. She was getting there anyway.

So sort people by that. How much does the money actually change what happens? The paper calls it additionality.

A hand drops one coin into an already-full open palm on the left; the same coin is the only thing in an empty open palm on the right.
The same coin barely registers in a full hand. In an empty one it is the only thing there. That is the hand it should go to.
Two shops, a year after the same ₹40,000 loan. The safe applicant mostly, it was coming anyway No other way in without the loan, nothing
Blue is the part the ₹40,000 built. Take the loan away and it disappears. For the safe applicant that is a sliver. For the other woman it is almost the whole shop. The rule sorts people by the size of that gap.

Nobody can rig it

The rule is written down and made public before applications open. A computer applies it. There is no meeting where someone quietly moves a friend up the list.

Every decision goes on the record afterward, with the reason each person got a no.

Round 8 · published

A‑102funded
A‑118funded
A‑093not funded. Your case was close; a stronger application would clear it next round.
A‑140not funded. Your district was oversubscribed this round. Try again next time.
A‑076not funded. Below the fund's minimum this round. The money waits rather than going to a weaker applicant.

Every other applicant gets the same. Anyone can check the maths.

What the engine puts out after every round. A rejected applicant sees the exact bar they missed and why. A year later, anyone can run the same decision again and check it comes out the same.

What exists

  • A paper, Don't Fund Winners, on SSRN. It also re-analyses a real randomized microcredit trial.
  • An engine, open source and written in Python. It runs a full revolving fund and publishes each round.
  • It is still early. So far everything runs on test data, and it gets real the first time an actual lending circle runs a round through it.

For a fund

You supply the money and the applicants. The engine makes the call and keeps the record. It does not move money or contact anyone; that part stays with you. The loans carry no interest.

Maaz Ahmad, Zachary Adam, and Manuel Gonzalez.