Graduation and Climate Resilience: Evidence from Bangladesh

Arm 1: Climate Graduation – This arm has two components: First, a loan from BRAC which will cover the cost of either fattening a cow owned by the beneficiary or for the purchase of a new cow. The loan will also cover the cost of an insurance policy that covers the beneficiary's cow (loan repayments will include insurance premiums that cover the replacement of the cow in event of loss). Second, a community-adaptation component that will organize monthly mobilization meetings for village households to facilitate coordination and cooperation on the community response to climate shocks (i.e., flood protections, new well construction, but also achieving sufficient demand for the loan/insurance). This will be delivered by community meeting facilitators who have undergone facilitator training with NGO Nijera Kori and have adapted the curriculum from those meetings to the setting of this project.
Arm 2: Private Adaptation Only – This arm includes only the livestock loan and insurance bundle described in the ‘Climate Graduation’ arm.
Arm 3: Community Adaptation Only – This includes only the community-adaptation intervention described in the ‘Climate Graduation’ Arm
Arm 4: Control
Note: In addition to the four-arm structure we will cross randomize the structure of the community coordination intervention. Half of villages will hold community meeting which only include the ultra-poor. The other half of villages will hold community meetings which include both the ultra-poor as well as village members which form a representative sample of the village population.

2025-10-15

2026-10-13

- Extensive margin take-up of loan
- Extensive margin take-up of community meetings (i.e. attendance)
- Intensive margin take-up of loan
- Intensive margin take-up of community meetings (i.e. meeting length, meeting participation, and index of adaptation investments agreed in the meetings)
- Index of Productive assets
- Share of assets covered by formal insurance
- Diversification of earnings (household member migration)
- Adaptation investment and behavior
- Consumption
- Household durables
- Savings
- Surrogate welfare index

Surrogate index for long-run outcomes constructed from recent evaluations of ultra poor graduation programs in Bangladesh.

We will explore impacts on our primary outcomes using (i) the full experimental sample and (ii) subgroups defined on the basis of:

- Ultra Poor Graduation status
- Climate shocks (pre- and post-treatment)
- Gender

The primary outcomes span a theory of change that has three parts.

Part I is the take-up of the two interventions:
- Extensive margin take-up of loans
- Extensive margin take-up of community meetings (i.e. attendance)
- Intensive margin take-up of loan product
- Intensive margin take-up of community meetings (i.e. meeting length, meeting participation, an index of adaptation investments agreed in the community meetings)

Part II captures the mechanisms activated by the interventions: higher investment in productive assets and insurance protection of those assets for the Private Adaptation intervention, and more diffuse and robust community adaptation measures in response to the Community Adaptation intervention. The specific outcomes included are:
- Adaptation investment and behavior (i.e. monetary or time investments in projects generated by the community meetings)
- Index of productive assets
- Share of assets covered by formal insurance
- Diversification of earnings (e.g. household member migration)

Part III is welfare measures and resilience proxies, which allow us to understand the extent to which the interventions allow beneficiaries to withstand climate shocks.
- Consumption
- Household durables
- Savings
- Surrogate welfare index

10500 households are divided into 430 clusters based on geographic proximity. Each cluster is made up of approximately 24 respondents. We use stratified randomisation to allocate clusters to four groups: (1) climate graduation, (2) private adaptation only, (3) community adaptation only, (4) control. The makeup of community meetings is cross-randomized. In half of clusters the meetings will be made up of only UPG beneficiaries; in the other half of cluster, meetings will include a full sample of village members

Not available

Stratified randomization on a computer. Strata are BRAC branch and cow ownership.

Geographical cluster level (Approximately one village per cluster)

Yes

430

10500 households

Control: 70 clusters
Insurance Only: 144 clusters
Insurance + Full Community Meetings: 72 clusters
Insurance + Ultra Poor Only Community Meetings: 72 clusters
Full Community Meetings: 36 clusters
Ultra Poor Only Community Meetings: 36 clusters

Treatment effects on assets. We have an MDE of 13.8% (0.14 standard deviations, or $69 USD) for control mean assets of $500 USD, for an omnibus comparison of all individuals assigned to any insurance treatment arm versus control.