The Impact of Inventory Advances (Zero-Interest Loans) on Women’s Economic Empowerment

Our research addresses a critical gap in understanding how the design of credit and marketing support can better serve women entrepreneurs in emerging markets. While extensive research exists on microfinance, most studies have focused narrowly on traditional cash loans, whether individual or group-based. Far less is known about in-kind credit and the features of an inventory advance that drive sales.

The four experimental conditions are:
1. Control: Brochure only – no demonstration and no instant delivery.
2. Treatment 1: unsealed units at home — can demonstrate and instant delivery available.
3. Treatment 2: sealed units at home — cannot demonstrate, but instant delivery available.
4. Treatment 3: demo unit only — can demonstrate, but no instant delivery.

This design isolates the mechanisms underlying inventory advances. The control group (CG) establishes the baseline under standard sales practices. Comparing T1 versus CG captures the overall effect of providing inventory on credit. Comparing T1 versus T2 isolates the effect of demonstration capability: both arms hold equivalent stock and offer instant delivery, but only T1 can demonstrate the product to customers. Comparing T1 versus T3 isolates the effect of instant delivery: both arms can demonstrate, but only T1 has stock available for immediate handover. Together, these contrasts allow us to separate the value of demonstration from the value of stock availability — two mechanisms that are typically bundled together in inventory-on-credit interventions.

2026-07-12

2027-07-12

o Sales
o Profits
o Consumer demonstrations
o Loan repayment

o Sales will be captured by two variables: a binary (extensive-margin) indicator of whether the entrepreneur completed any sale, and a continuous (intensive-margin) count of the number of sales. Both will be measured in total and disaggregated by product type and product category.
o Profit will be captured by continuous variable by product type, product category and in total - to indicate the profit made by the entrepreneur.
o Consumer demonstrations captures whether, the frequency, and the type of product demonstrations that the entrepreneur engages in during the sales process. Under this outcome, we will construct (i) binary & continuous variables by type of demonstration event to capture whether and the frequency at which the entrepreneur conducts this event, (ii) the number of (potential) consumers attending the event, and (iii) sales completed at the event.
o Loan repayment will be captured by both a binary and continuous variable to indicate whether and the amount the entrepreneur has repaid.

These measures will be captured through the sales registration application provided to the entrepreneurs and will be cross-verified through structured phone-based surveys and random spot checks.

o Financial health
o Gender empowerment

o Financial health outcomes include measures related to access to financial resources. Access to credit will be captured through questions on formal and informal borrowing, loan terms, repayment behavior, and perceived access to financial services. Additionally, income will be measured by asking respondents to list all sources of household and personal income over a defined reference period. Savings will be defined as cash or assets set aside for future use, reported separately for personal and business purposes.
o Gender empowerment outcomes include confidence, self-efficacy, autonomy and agency, gender norms and gender-based restrictions, financial decision making, participation in public life, and social standing within community. These constructs will be captured through a mix of standardized survey items and behavioral measures.

To capture the above mentioned constructs, we will include sets of standardized survey items during the endline survey and through standardized video interviews.

This study employs a randomized controlled trial (RCT), to explore how zero-interest loans, in the form of inventory, can empower rural women entrepreneurs, selling socially beneficial products, economically and socially.

The study targets a sample of 280 female entrepreneurs, drawn from 280 distinct villages in Bihar, India. Our sample consists of only one entrepreneur per village to eliminate intra-village spillover effects. The sample size has been determined through power calculations based on prior studies, providing sufficient statistical power to detect meaningful differences across four experimental arms.

We will conduct individual-level randomization, stratified by district and current employment status (excluding farming, dairy, and daily wage work). The four arms form a 2 × 2 factorial design, crossing demonstration capability (yes/no) with instant delivery (yes/no).

We will provide model-free evidence and results from the intent-to-treat (ITT) regressions. We will also consider the effect of the treatment on the treated (TOT) specifications.

The project builds on our previous research (pilots), conducted in Rajastan.

Not available

Randomization done in office by a computer (STATA)

Individual randomization

No

N/A

280 individuals (women)

CG: 70 participants
T1: 70 participants
T2: 70 participants
T3: 70 participants

We use outcome data on sales from two past RCTs to estimate sample size requirements. For our key outcome, we focus on sales outcomes at the intensive and extensive margin. Our first pilot RCT showed that 14% in the control versus 97% in the treatment arm completed at least one sale over the pilot period (a cumulative measure, in contrast to the monthly rate reported for the second pilot below). Effect sizes on sales outcomes are substantial: Mean sales are 0.46 (SD = 1.35) in the control group and 5.54 (SD = 2.34) in the treatment group. Pooling across arms, the overall standard deviation of sales in this pilot was 3.1. Since this pilot in 2018, we have improved data quality through better IT systems and reduced measurement error; we therefore conservatively assume this overall SD has dropped slightly, to 2.7. We complement this and use outcome data on sales at the month level from a second pilot RCT (N = 96 respondents, four arms, conducted 2019) to estimate sample size requirements. For our key outcome, we focus on sales outcomes at the intensive and extensive margin. The pilot showed that 13.1% of respondents in the control arm completed at least one sale in a given month, compared with 41.7–49.3% across the three treatment arms. Effect sizes on sales outcomes were substantial: mean monthly sales were 0.19 (SD = 0.55) in the control group and ranged from 2.07 to 3.28 (SD = 3.72–5.80) across the treatment arms. We will implement an individual-level randomization and stratify randomization by district and current employment status (excluding farming, dairy or daily wage work). We plan to use a regression specification that includes district-fixed effects and individual-level exogenous covariates (e.g., household size, age) to increase power. Importantly, the new RCT will collect respondent-month panel data over 3–5 months, which substantially raises effective sample size; we adopt a conservative intra-respondent correlation of ρ = 0.05. All calculations follow J-PAL/World Bank standards with α = 0.05 and power = 0.80. The study includes 280 respondents: a control group and three treatment arms of 70 participants each. Assuming T = 4 monthly observations per respondent (ρ = 0.05). Below, we present the resulting minimum detectable effects (MDEs): 5. % Completing a Sale: a. Across our two pilots, with a control completion rate of 13 to 14%, we are able to detect an MDE of approximately 9.7 to 10.0 percentage points (~0.29 SD of the control rate) in the share of respondents completing at least one sale, with 80% power. b. 6. Number of Sales: a. Across our two pilots, the observation-level standard deviation of monthly sales ranges from about 2.7 (Pilot 1) to 4.1 (Pilot 2). Given this range, we are able to detect an MDE of approximately 0.7 to 1.0 additional monthly sales (~0.25 SD) with 80% power. For our key outcome, we focus on sales outcomes at the intensive and extensive margin. Our pilot RCT showed that 3% in the control versus 97% in the treatment arm completed at least one sale. We conservatively assume that the SD has slightly dropped slightly from SD = 3.1 to SD = 2.7. Additionally, we are in a position to implement an individual-level randomization and thus do not need to implement cluster randomisation, which increases our power. Finally, we plan to use a regression specification that includes district-fixed-effects and individual-level exogenous covariates (e.g., household size, age) to increase power. Key outcomes: - % Completing a Sale: Based on pilot data, we conservatively estimate that 5% of the control group would make a sale. Based on this assumption, with a sample of 60 Dharmalife women entrepreneurs per arm, we can detect an MDE of 0.65 SD (19.36% increase in the number of individuals completing at least one sale) with 80% power. - Number of Sales: Based on a sample of 60 Dharmalife women entrepreneurs per arm, we are able to detect an MDE of 0.50 SD increase (33.69% increase in the number of products sold) with 80% power.