Collateral-Free Loans in Bangladesh Leave Women’s Businesses Hiring Five Times More Workers

In Bangladesh, a woman who owns a small business and wants a bank loan faces a familiar wall: the bank wants collateral, and she almost certainly does not have land or registered assets in her name. Since 2004, Bangladesh Bank, the country’s central bank, has tried to dismantle that wall with a refinancing scheme that lets women who hold at least 51 per cent of an enterprise borrow without putting up any security at all. Participating lenders draw funds from the central bank at 0.5 per cent and on-lend them collateral-free, with the borrower’s interest rate capped by regulation and a 2 per cent incentive paid to any borrower who keeps to her repayment schedule. The fund began at Tk 100 crore and has grown repeatedly, standing at Tk 1,500 crore during the study period and rising to Tk 3,000 crore in 2023, with 46 banks and 27 financial institutions signed on. What has never been established, in two decades of operation, is whether the women who use the scheme actually end up better off than comparable women who do not. A new study now offers the first enterprise-level answer, and the numbers are striking.

The research, published in Discover Global Society, surveyed 193 women entrepreneurs across seven of Bangladesh’s eight administrative divisions: 93 who had borrowed under the scheme, verified against lender disbursement records rather than self-report, and 100 who had never borrowed through any formal channel. Because the women who reach a lender are almost certainly different from those who never do, the researchers could not simply compare the two groups side by side. Instead, they built a statistical comparison. Each woman’s probability of participating was estimated from five characteristics recorded before any loan: her education, her age, whether her business was urban or rural, how long it had been trading, and whether she had received business training. Women were then paired one to one on that estimated probability, producing 61 comparable pairs on which all the outcome comparisons rest.

The headline finding concerns jobs. Inside the matched sample, enterprises run by scheme borrowers employed roughly five times as many workers as the enterprises run by the women they were matched to. Formally, the study reports an incidence rate ratio of 5.108, with a 95 per cent confidence interval running from 3.564 to 7.320, estimated by negative binomial regression, a model chosen because the employment counts were heavily skewed, with a pooled median of four workers but a maximum of 250. A matched non-participant with three employees corresponds to about fifteen inside the scheme. No other predictor in the model, including education, business age, training or investment, came close to significance, which suggests under the study’s assumptions that the difference tracks the loan itself rather than the kind of woman who obtained one.

The authors are careful about how far that number travels. Because the survey was a single round of interviews conducted after the loans were made, there is no pre-loan measure of enterprise size, so the possibility that lenders simply approved women whose businesses were already larger cannot be excluded from the data. Part of the measured difference could reflect approval rather than growth. The employment count also includes the owner and unpaid family workers, so some of the increase may be labour shifting from unrecorded to recorded status as the enterprise formalizes around a credit relationship, rather than genuinely new hiring. The estimate is described throughout as an association consistent with the scheme working, not experimental proof that it does.

What the analysis can do is test how hard the result would be to explain away. Using Rosenbaum bounds, a sensitivity technique that asks how strong an unmeasured difference between the two groups would have to be before the finding lost significance, the employment result holds up to a sensitivity parameter of 3.5 and fails at 4.0. In plain terms, for the fivefold employment gap to be an artefact of hidden bias, there would have to be some unmeasured characteristic, such as determination or appetite for risk, that made a woman nearly four times more likely to join the scheme while being uncorrelated with her education, age, location, business age and training. The strongest measured predictors of participation were location and training, so the hypothetical confounder would need to outweigh both while staying independent of them. That is a demanding standard, and by the conventions of observational research in this field the finding is reasonably durable.

The income result is the puzzle. Household income ran in the opposite direction: scheme borrowers sat in lower income bands than their matched counterparts, with an odds ratio of 0.040. Among the surveyed borrowers, 75.3 per cent fell in the lowest income band, while 97 per cent of the comparison group sat in the second band. The study offers two readings, and its design cannot choose between them. Under the first, the enterprises are on a credit-financed growth path: in the survey year, a much larger wage bill, interest on the loan and money ploughed back into stock and equipment together exceeded the extra revenue the bigger business had so far earned, leaving less for the household. Under the second, the scheme was built to reach women the banks had passed over, and those women came disproportionately from poorer households to begin with, so the coefficient records the target rather than any consequence of the loan. Both readings fit the data, they are not mutually exclusive, and the authors explicitly decline to claim that either is demonstrated. Panel data following the same women over several years would be needed to separate reinvestment from targeting.

One finding needs no statistical caution at all, and it may be the most consequential. Not one of the 100 women in the comparison group had heard of the scheme. Awareness of the circular separated the two groups so completely that it could not even be used in the matching model, because it would have assigned every unaware woman a participation probability of zero. A woman who does not know a facility exists is never assessed by a lender, because she never presents herself to be assessed. Earlier research had identified scheme awareness as a strong independent predictor of formal credit uptake in Bangladesh, compounding the collateral barrier with an information barrier. This study shows that barrier operating at full strength: the loan is already collateral-free, yet the binding constraint on uptake is now simply knowing that it exists.

The pre-matching differences between the groups tell a consistent story about who reaches the scheme. Participants were significantly more urban, more likely to have received business training and older than non-participants, with location carrying a standardized bias of nearly 60 per cent before matching. Refinancing agreements were concluded mainly with institutions in towns and cities, so a rural woman faces a longer journey to apply and her lender faces a longer journey to monitor her. After matching, every covariate fell to an absolute bias of about 10 per cent or below, and no measured characteristic differed significantly between the matched groups, which is the condition the outcome models require. The estimates, however, describe only the region of common support: 32 borrowers for whom no comparable non-participant existed were dropped by the caliper restriction, so the results do not extend to the scheme population as a whole.

For policymakers, the study’s implications are unusually concrete. Outreach at the local level emerges as the cheapest available lever, since the financial design has already removed the collateral requirement that theory identifies as the binding constraint. Expanding participating institution branches and agent banking outlets in rural areas would address the geographic skew in access. And because household income falls in the year of borrowing under either reading of the income result, the pattern argues for support during the first expansion year rather than credit alone. The authors also note that the borrower rate cap has since fallen from 10 to 5 per cent, which would ease the repayment pressure behind the income finding without removing it, and that the fieldwork, conducted in 2018, describes the scheme as it ran before the larger dedicated women’s fund established after 2019.

The contribution is deliberately narrow: one of the first evaluations of this facility to use enterprise-level survey data and a constructed comparison group, reporting a magnitude where the prior literature offered only descriptions. Whether that magnitude is causal depends on an assumption that cannot be verified from a single round of interviews. The obvious next step, the authors write, is to revisit the same enterprises over several years, which would settle both the size of the employment effect and the meaning of the income result. Until then, the study stands as a rare quantified glimpse of what happens when the collateral wall comes down for women entrepreneurs, and a reminder that the most stubborn barrier may not be the bank’s terms at all, but the silence surrounding them.

Subject of Research: Evaluation of collateral-free credit effects on women-led small enterprises in Bangladesh

Article Title: Collateral free credit and women led enterprise outcomes in Bangladesh evidence from a matched evaluation of the Bangladesh Bank refinancing scheme

Article References: Collateral free credit and women led enterprise outcomes in Bangladesh evidence from a matched evaluation of the Bangladesh Bank refinancing scheme. (n.d.). https://doi.org/10.1007/s44282-026-00617-x

Image Credits: AI Generated

DOI: 10.1007/s44282-026-00617-x

Keywords: microfinance, women entrepreneurship, Bangladesh Bank, collateral-free credit, SME finance, propensity score matching, gender financing gap, financial inclusion, program evaluation, employment, household income, credit rationing