Working Paper: NBER ID: w31620
Authors: Mia Ellis; Cynthia Kinnan; Margaret S. McMillan; Sarah Shaukat
Abstract: Not all firms have equal capacity to absorb productive credit. Identifying those with higher potential may have large consequences for productivity. We collect detailed survey data on small- and medium-sized Tanzanian firms who borrow from a large commercial bank, which in turn raises funds via international capital markets. Using machine learning methods to identify predictors of loan growth, we document, first, that we achieve high rates of predictive power. Second, “soft” information (entrepreneurs’ motivations for entrepreneurship and constraints faced) has predictive power over and above administrative data (sector, age, etc.). Third, there is a different and larger set of predictors for women than men, consistent with greater barriers to efficient capital allocation among female entrepreneurs.
Keywords: small firms; loan growth; Tanzania; machine learning; capital allocation
JEL Codes: G14; J16; O16
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| soft information about entrepreneurs' motivations and constraints (L26) | loan growth (G21) |
| record-keeping and financing business growth with retained earnings (D25) | loan growth (G21) |
| loan documentation issues and need for local government signatures (H74) | loan growth (G21) |
| predictors of loan growth differ by gender (J16) | barriers to efficient capital allocation (D61) |