Data and Code for: Cashing In (and Out): Experimental Evidence on the Effects of Mobile Money in Malawi

Summary:  View help for Summary In this paper, we conducted an RCT with microentrepreneurs in urban Malawi in 2017–2018. Treatment was three-pronged: assistance in opening a mobile money account, training on how to perform basic transactions, and a withdrawal fee waiver. We find that the majority of people opened accounts and used them extensively. We find strong evidence that treated respondents real-located labor from business to agriculture, and we find mixed evidence of an increase in expenditures. In contrast to the existing literature, effects appear to be driven by using the accounts to save rather than to make transfers.

Funding Sources:  View help for Funding Sources IPA Financial Inclusion Program

JEL Classification:  View help for JEL Classification
      E42 Monetary Systems; Standards; Regimes; Government and the Monetary System; Payment Systems
      G21 Banks; Depository Institutions; Micro Finance Institutions; Mortgages
      G51 Household Finance: Household Saving, Borrowing, Debt, and Wealth
      G53 Household Finance: Financial Literacy
      L26 Entrepreneurship
      O16 Economic Development: Financial Markets; Saving and Capital Investment; Corporate Finance and Governance

Geographic Coverage:  View help for Geographic Coverage Blantyre city, Malawi

Time Period(s):  View help for Time Period(s) 3/1/2017 – 7/1/2018

Collection Date(s):  View help for Collection Date(s) 3/1/2017 – 5/1/2017

Universe:  View help for Universe Microentrepreneurs in urban Malawi. We excluded businesses with more than 2 employees (6 percent), businesses in which the owner worked less than 5 days a week (9 percent), and businesses that planned to shut down within 6 months (16 percent). We also excluded any business that was also a mobile money agent (3 per-cent). Finally, we excluded illiterate business owner (20 percent) and owners who could not read written text due to poor eyesight (10 percent)

Data Type(s):  View help for Data Type(s) administrative records data; survey data

Response Rate:  View help for Response Rate 480 microentrepreneurs were enrolled in the study at baseline, 453 were surveyed at least once at follow up. 

Sampling:  View help for Sampling The experiment took place with a representative sample of small entrepreneurs operating in Blantyre, the second largest city in Malawi. While Blantyre is an urban center with a population just over 1 million, the outskirts of the city contain farmland. Blantyre contains 26 wards and 392 enumeration areas (EAs). To construct a sample with coverage across the city, we aimed to randomly select three EAs in each ward, ultimately selecting 77 (one ward did not have 3 EAs).Market structure is heterogeneous across EAs – the number of businesses ranged from 0 to1,649 (mean 104, median 48).10Because of the high number of businesses in some EAs, it was not logistically possible to census every business in those EAs. We therefore decided to divide EAs between those with more than 100 and those with less than 100 businesses. In the smaller EAs, we censused all businesses; in the larger EAs, we counted all businesses but only censused a randomly selected subset of approximately 40% of businesses.11We counted a total of 9,848 businesses and classified 8,078 (82.1%) of these as small businesses.12We attempted to conduct a census survey with 3,857 businesses and completed surveys with 2,842 (74%). Of the 1,012 (26%) businesses that were not censused, 552 (14%) refused to participate (either before or after we were able to explain the study), 346 (9%) were permanently closed, 114 (3%) were not reached (either because the shop was closed after 3 visits or the owner was under 18 years old).

We excluded several classes of businesses in this exercise since they were unlikely to qualify as a small business. This included gas stations, clinics, hospitals, banks, microfinance institutions, manufacturing plants, warehouses, wholesalers and supermarkets. 

After the census, we imposed additional exclusion criteria. First, we excluded any business with more than 2 employees (6% of the census list). Second, we excluded businesses in which the business owner was a mobile money agent (3%) to prevent confounding the mobile money treatment. Third, we excluded businesses in which the owner was not actively involved in running operations (defined as working there at least 5 days per week) since such owners would not be able to reliably answer business-related questions (9%). Fourth, we excluded businesses that were planning to shut down within 6 months (before the project was slated to end – 16%).

Once we had a sample of businesses that met our criteria, we imposed two other exclusion criteria, using data that had been collected either at the census or prior to the baseline survey. First, we removed all polygamous households, which amounted to 5% of the sample. Second, since we initially planned to collect surveys with paper-and-pencil logbooks (we eventually changed to phone surveys), we excluded business owners who were illiterate (about 20% of the sample) and those whose eyesight prevented them from reading a printed page (about 10% of the sample).These exclusion criteria left us with approximately 1,640 eligible businesses from which we drew our final sample, stratified by financial access (defined by having either a mobile money or bank account) and self-reported distance to the nearest mobile money agent (defined as above or below the sample median). In drawing the sample, we chose to oversample businesses connected to the electricity grid: while 26% of eligible businesses were connected to the grid, we sampled 35%. We replaced respondents who could not be found (about 6.5%) or refused to participate (another6.5%) with randomly chosen backups, ultimately yielding a sample of 480  businesses

Data Source:  View help for Data Source Survey data and telecom administrative records on mobile money usage of users in the sample. 

Collection Mode(s):  View help for Collection Mode(s) computer-assisted personal interview (CAPI); face-to-face interview

Scales:  View help for Scales Several Likert-type scales were used. 

Weights:  View help for Weights Sampling weights should be used in all the analyses. 

Unit(s) of Observation:  View help for Unit(s) of Observation business owner

Geographic Unit:  View help for Geographic Unit not applicable