Impact Evaluation of Agricultural Microcredit Product in Myanmar

The intervention is the provision of a seasonal agricultural microcredit loan.

2019-12-13

2020-06-15

Our theory of change is that seasonal agricultural microcredit loans should primarily be invested in agricultural activity. Since the loans are provided very close to planting season, they should lead to a genuine increase in access to finance for the winter agricultural season, rather than providing a slightly lower-cost source of financing. Hence loan recipients should have greater capital to invest in agriculture, rather than substituting for other loans.

Given their timing and purpose, we expect loans to be used on variable inputs such as seeds, fertilizer, and labor inputs. Hence we should see impacts on these agricultural investments, both in the monetary value invested in inputs, and the likelihood of accessing key types of inputs. It is also possible that access to loans would lead to a change in cropping patterns, e.g., devoting more land or investment to a more input intensive crop. So we will study the crop mix, but we don't have strong priors here.

In principle, increases in agricultural investment should lead to increases in agricultural productivity and outcomes, in particular, total yield, which can be measured in quantity terms (e.g., kilograms or volume-based measures of yield). Increases in yields should lead to increases in agricultural revenue. While loan recipients might also increase expenditure through use of loan proceeds, on net they should have higher profits than non-recipients.

Total agricultural revenue will be measured by adding up the revenue from individual crops. We will look also look at the revenue from the "main" (largest revenue) crop.

Total agricultural expenditure will be measured by adding up the expenditure from individual crops. We will look also look at the expenditure on the "main" (largest revenue) crop.

To calculate agricultural profit, we will subtract agricultural expenditure from agricultural revenue.

In some cases, the endline survey will be conducted before all farmers have harvested their winter season crop. In that case we will calculate their yield as the sum of what they have harvested so far, with what they still expect to harvest. Similarly, we will calculate their revenue as the sum of what they have sold so far, and what they still plan to sell.

Secondary outcomes concern off-farm outcomes: non-farm business and earnings activity, household labour market outcomes, and broader outcomes for the household.

With respect to other economic activity, we will focus on investments and returns from non-farm businesses, and labour earnings.

With respect to other household outcomes, we will construct the Probability of Poverty Index (based on the Myanmar PPI), and calculate other monetary measures of household outcomes including expenditure on health and education, transfer receipts, and food security. In particular, we will look at a measure of 4-weeks' consumption, and likelihood of households needing to ration meals.

Access to agricultural microcredit is provided to a randomly-selected treatment group and not to a control group, within a set of farmers who showed interest in the agricultural microcredit product.

The study was originally designed to study the impacts of a digital agricultural microcredit product, where farmers would be selected to receive loans based on a credit score calculated using call data records (CDR), aka, mobile phone meta data. This is the recruitment strategy that was originally socialized with farmers: they were asked for their interest in releasing their mobile phone meta data for credit scoring, with the prospect of receiving a seasonal agricultural microcredit loan through a digital financial transfer. The sample for the study is drawn from the set of 453 farmers who signed up to potentially receive a loan through this method.

However, the credit scores were delayed, and hence loans were provided through an alternative method. First, farmers were randomized into treatment-control based on the last digit of their national ID card number being even or odd. 188 farmers were automatically selected for treatment through this method. Then loan officers of the microfinance partner visited the remaining households to do conventional credit scoring. Out of this an additional 168 farmers were selected to receive the loans. So, in total, 356 farmers received loans, 188 through randomization, 168 through selection by credit officers, and (453 - 356) = 97 farmers did not.

The randomization component is done based on the last digit of farmers' National Registration Card (NRC) number, which is effectively random. Treatment-control was divided based on even-odd last NRC digits.

Individual borrower (representing an agricultural household).

No

21 villages.

453 potential borrowers (leaders of agricultural households).

There are 356 treatment and 97 control. The distribution of treatment-control varies by cluster (village) as randomization was done at individual rather than cluster level.

At the time of the baseline survey, 342/453 households were successfully surveyed, a 75% response rate. Due to Covid-19, the endline survey was implemented as a phone survey, as in-person surveying was not legal in Myanmar at the time. At the time of the endline survey, we resurveyed baseline respondents, and also attempted to survey as many respondents who were missed at baseline (from the 25%), successfully reaching 94, so another 20.7% of the original sample of 453. We implemented a modified survey instrument at endline with this group, which also asked about a subset of variables that would have been asked at baseline. This allowed us to partially reconstruct a baseline-endline panel.

Of the 342 baseline respondents, only 1 attritted in the endline phone survey, so we resurveyed 341 baseline respondents at endline. Added to the 94 who were missed in the original baseline, that gives us 435 total respondents (a 96% overall response rate), 341 who have full baseline-endline panel data, and 94 who have partial baseline-endline panel data.

We can also break down these figures by sample group, which we describe as group 1 (188 randomized treatment), group 2 (168 credit officer selected treatment), and group 3 (97 control).
-Group 1. 186 completed, 2 not completed = 188. Of the 186, 152 responded at baseline and endline, and 34 only at endline.
-Group 2. 162 completed, 6 not completed = 168. Of the 162, 124 responded at baseline and endline, and 38 only at endline.
-Group 3. 87 completed, 10 not completed = 97. Of the 87, 65 responded at baseline and endline, and 22 only at endline.