June 2026 – The influential Banerjee and Newman (1993) model of poverty traps has long shaped development research and policy. Its elegant framework suggests that with borrowing constraints, households face two possible paths: subsistence activities requiring no capital, or more profitable occupations requiring lumpy investments. This dynamic produces the classic S-shaped curve, where starting assets determine whether households converge to poverty or prosperity.
A landmark experiment by Balboni et al. (2022) in rural Bangladesh reignited interest in this theory. By giving extremely poor women a cow valued at around $490, along with training and support, researchers observed asset dynamics that appeared to confirm the model. They estimated a threshold at $504: households above it saw assets grow, while those below saw them shrink. Their conclusion was bold — large asset transfers could be an effective way to break poverty traps.
Yet new evidence complicates this picture. A recent working paper by Dean Karlan, Amol Singh Raswan, and Christopher Udry, provocatively titled “The Sisyphean pursuit of evidence for poverty traps”, challenges the robustness of these findings. Analyzing seven other graduation experiments across countries including Ethiopia, Ghana, Honduras, India, Pakistan, and Peru, they found no evidence of the S-shaped dynamics.
Two key issues stand out. First, the S-shape in Balboni et al. appears highly sensitive to functional form choices, particularly the use of log transforms to handle zero assets. Alternative specifications show consistent asset growth across households, undermining the threshold effect. Second, cow values varied by village, meaning households in wealthier areas with higher cow prices also had better growth prospects. What looked like wealth-based differences may instead reflect geographic heterogeneity — areas with stronger infrastructure, NGO presence, and education levels.
This debate raises broader questions. If poverty traps exist, they may be more about geography and migration than livestock. As Banerjee et al. (2021) found in India, long-term impacts of ultra-poor programs often come from households diversifying into wage employment and migration opportunities.
The search for poverty traps may not be futile, but it is complex. Thresholds may differ across households, contexts, and sectors. Rather than focusing narrowly on cows or chickens, researchers suggest looking at structural constraints such as education, location, and access to broader labor markets.
The renewed debate underscores both the appeal and the difficulty of proving poverty traps empirically. While the S-shape remains a powerful metaphor, the evidence suggests poverty is shaped by a wider set of constraints — and breaking free may require more than asset transfers. It may demand mobility, skills, and systemic change.
