In the hungry season, Zambian farmers sell their labour to eat. One trial showed what happens when they do not have to.

Every year, in the weeks before harvest, a smallholder household in Zambia's maize belt runs short of two things at once: cash and grain. Last season's stores are empty. The new crop is in the field but not ready. Those same weeks are when the field most needs work, because weeding and care in the rains help decide how much the crop will yield.

Households settle the clash the only way open to them. In a large Zambian study, a majority of households said that to cover short-term needs they would sell their own labour in the local market, casual work known as ganyu (Fink, Jack and Masiye, 2020). The hours that could have gone into the household's own field go into someone else's, because food is needed today.

That is a financial problem that looks like an agricultural one. The question worth asking is what the shortage of cash costs the farm, and what happens to a whole village when the shortage is lifted.

What the trial did

Günther Fink, B. Kelsey Jack and Felix Masiye, an economist at the University of Zambia, ran a randomised trial with 3,139 farming households in 175 villages in Chipata District, Eastern Province, over two seasons. Seasonal loans were offered in January 2014 and January 2015, in the middle of the hungry season, to randomly chosen villages. Other villages were not offered one. The results were published in the American Economic Review in November 2020.

Take-up was close to universal: 98 percent of eligible households took the loan. Repayment was about 95 percent in the first year.

The more interesting results are about labour. In the first year, the chance that a household sold any ganyu labour in the hungry season fell by 4.8 percentage points, a 14 percent drop, and hours sold fell by about a quarter. Treated households put 4.9 more hours of family labour a week into their own farms, an increase of 10.6 percent. About 65 percent of the loan's value went on additional labour inputs on the farm. Agricultural output rose by 9 percent. The effects were larger for poorer households.

Why this is a village story, not a household story

Labour sold by one household is labour hired by another. In the trial, households in treated villages became 43 percent more likely to hire ganyu labour, and as many households stopped selling their own hours, wages in the local labour market rose. The authors report this directly: higher output on farms, and higher wages in the village.

So the loan did not only work through the person who borrowed. Part of its effect travelled through the village labour market to people who never took it. Our reading, and it is an interpretation rather than a finding of the paper, is that this is one reason seasonal cash is hard to supply on commercial terms. A lender can bill only the borrower. It sees a share of the benefit it creates.

The same shape elsewhere

The hungry season binds household decisions in other regions too. In rural Bangladesh, Gharad Bryan, Shyamal Chowdhury and Ahmed Mushfiq Mobarak gave randomly chosen households a small incentive to send a member to work elsewhere during the lean season. 22 percent of households sent a seasonal migrant, consumption at home rose, and treated households were 8 to 10 percentage points more likely to send a migrant again one and three years later (Econometrica, 2014). Different country, different instrument, same underlying fact: before harvest, a household's use of its own labour is shaped by how much cash it has.

What the trial did not settle

The first open question is who carries the weather. Repayment, about 95 percent in the first year, fell to about 80 percent in the second, which the researchers speculate reflects volatile rainfall and lower output in 2015 (Innovations for Poverty Action policy brief and ATAI project page). A bad season hits every borrower in a district at once. That is the risk a lender cannot spread across borrowers, and the trial was not designed to answer it.

The second is who delivers. The loans were administered by Innovations for Poverty Action under the Chipata Loan Project, and the paper states that repayment risk was borne by the project. This was a research programme, not a lender, and the trial measured output and labour, not a case for a lender. A recent test in Kenya, run by the Digital Credit Observatory with Acre Africa and Pezesha, shows how far demand in a trial can sit from a delivered loan. After SMS outreach failed, phone calls reached 1,131 farmers and 11 percent expressed interest. Only 61 farmers ended up borrowing, and about half of those paid late. The researchers concluded that smallholders were “either not credit constrained, or likely wary of digital credit even when loans are insured.” The product, country and delivery route differ from Zambia's, so it does not contradict the Zambian result. It does show that a loan farmers want in a trial does not become a loan farmers take up by itself.

Who should care

Lenders and development finance teams designing seasonal products should treat timing as part of the product: cash that arrives in the hungry season does a different job from cash that arrives before it. Programme designers should count the labour market effect when they weigh what a seasonal loan is worth. Researchers have a clear gap, because no trial yet varies the delivery route, the lender or the weather exposure.

What we see from the field

DNC lends cash to farmers during the season in villages around Petauke, in the same province as the trial, repaid in maize at harvest, and records each loan and repayment against the individual farmer on Panjiia. What the trial points us towards is a narrow thing. A repayment record that spans more than one weather year carries information a single good season cannot. A farmer who repaid in the dry year has shown something a farmer who repaid in the good one has not. That is the kind of evidence a lender would need to separate the weather from the borrower.

The open question is who should hold the dry-year risk, and what record would let someone carry it knowingly.