Poverty and Migration in the Digital Age: Experimental Evidence on Mobile Banking in Bangladesh Jean Lee, Jonathan Morduch, Saravana Ravindran, Abu Shonchoy, Hassan Zaman April 26, 2017 1
Context Migration Urbanization Mobile Money Poverty and risk 2
Two sites Dhaka: Capital city, home to garment factories Gaibandha Dhaka Gaibandha district, Rangpur One of poorest regions of Bangladesh, with exposure to monga (seasonal famine, September through November). Rangpur has significantly lower rates of food consumption per capita than other regions. 3
Focus: impact of remittances via mobile money Factory workers Remittances 4
5
6
Urbanization From rural to Dhaka (lifetime net migration rate 2001-11) Absolute rural decline Working age, 20-44 7
8
9
Mobile Money in Bangladesh [E]xperts at Bangladesh Bank, the country s central bank, describe mobile money as a key strategy to expand financial access in this nation of 160 million people, where fewer than 30% have a bank account. - Wall Street Journal, 2015 Wide range of bank-based, interoperable mobile money providers: Dutch Bangla Bank, bkash, etc. Potential to mitigate economic shocks (Jack and Suri, AER 2014; Batista and Vicente 2016) 10
bkash Leading mobile money service provided by BRAC Bank Mobile wallet and person-toperson transfers Individuals deposit and withdraw money through agent network Launched in 2011. Handles about 70 million transactions per day (Wall Street Journal, 2015) 11
Rapid Expansion Microfinance: After 4 decades, 21 million users. 90% women. Mobile money: In 5 years, 21 million accounts. 18% women. Leesa Shrader, CGAP 2015 12
13
bkash 14
Partner: Gana Unnayan Kendra (GUK), a local NGO in Gaibandha. Works to train garment workers and place them in jobs in Dhaka. 15
Research Questions Context: Seasonal variability in incomes, as well as the relatively higher incomes in factory jobs in Dhaka, motivates migration. Project: Introduce mobile banking accounts to a sample of migrants and rural families. Questions: Does the technology improve wellbeing of rural households (transfer recipients), particularly through the annual pre-harvest (monga season) famine. Reduce food consumption variability? Improve education? Improve financial conditions? Improve recipient household members health through the difficult famine season? 16
Experimental Design 17
Linked samples Rural families: Rural families of migrants Urban migrants: Migrants to Dhaka from these same rural households (70% male, 30% female). Rural households trained through GUK Targeted for this intervention after identified as ultrapoor 99% have mobile phones 11 % have bank accounts Avg land: about 0.1 acre Many have incomes < $1 per day per person Encouragement design: Half of the sample is experimentally introduced to the technology 18
Overview of Design Rural household-urban migrant pairs are the unit of randomization 817 household-migrant pairs in final sample Recruited through use of a garments training program (SHIREE) roster and snowball sampling Three cross-randomized experiments: 1. bkash training vs. no bkash training 2. Messaging about individual benefits vs. messaging about family/social benefits 3. In a willingness-to-pay survey, priming to think about bkash or cash We focus here on Experiment 1 19
Training Intervention 30- to 45-minute intervention. How to sign up and use bkash. Information about bkash mobile money (hard copy) Technical assistance with enrollment: locating necessary identification, completing application locating vendor who could accept application 200 Taka (<3 USD) compensation for participation in the training 20
Timeline Study recruitment (Sept 2014 to Feb 2015) Baseline survey (Dec 2014 to March 2015) Introduce bkash (April 2015 to May 2015) Treatment: 415 households (bkash training and incentive) Control: 400 households Marketing: Within treatment arm, cross-randomized order in which households and migrants were approached whether or not migrant is first mover and pro-social marketing strategy Midline survey (August 2015 to September 2015) Endline survey (January 2016 to March 2016) 21
Administrative Data In addition to survey data, we collected administrative data from bkash on accounts held by households and migrants in our sample Data range from shortly following the end of the treatment phase to one year later (July 2015 to July 2016) 22
Estimation 23
First Stage Compare active accounts by treatment status, where active is defined as having at least one transaction in the period July 2015 to July 2016 Transactions include transfers (sent or received), withdrawals, deposits and airtime top-ups Data are from bkash admin files 24
Frequency of Transactions 73%% of bkash account holders make more than one transaction in their account per month 25
First Stage: Rural Controls: gender, age, primary school completion of head of the household, and household size 26
First Stage: Urban Controls: gender, age, primary school completion of head of the household, and household size 27
Average Month-End Balances Low Balances Average month-end balances are low (< $3) $1 = 78 taka PPP$1 = 31 taka Rising over time 28
Number of Remittances Treatment households send about double the number of remittances over the period as control households 29
Value of Mobile Remittances Aus Local boro HYV boro Similar pattern for value of remittances sent Peaks and valleys somewhat correspond to festivals and harvest seasons 30
Migrant Income and Remittances Average monthly income: 7830 Taka per migrant. $100, PPP$300, (PPP$10/day) Remittances in past 7 months: 17,279 Taka $222, PPP$557 2468 Taka per month ($32, PPP$80) Large fraction of monthly income (2468/7830 = 31.5%). 31
Key findings: impact of treatment Rural households: Reduced borrowing and increased savings Improved education, health and agricultural outcomes Improved resilience to negative economic shocks Urban migrants: Increased savings and reduced poverty Increased employment in formal sector Worsened self-reported health indicators 32
Results: Rural Households 33
Estimating Impacts We then estimate Intent-to-Treat and Instrumental Variables estimates of the treatment effect using the following specification: Our outcomes include individual outcomes and indices of outcomes within an outcome locus, such as education or health, as constructed following Kling, Liebman and Katz (2007) 34
Borrowing At baseline, the total size of loans taken by rural households over the last 12 months was 6798 Taka. Monthly remittances are large in comparison to the size of total loans (2486/6798 = 36.6%) 35
Borrowing Significant reduction in reported need to borrow over past 1 year Reduction in total value of loans (pvalue = 0.107) 36
Borrowing Index 37
Savings 44 percentage points more likely to save on a control mean of 43% 38
Education Significant increases in daily hours spent studying and aspirations for children Insignificant increases in rates of passing last exam, enrollment, and attendance No effect on education expenditures 39
Education Index 40
Health Significant decrease in number of sick household members Insignificant decreases in weeks ill over past year and average medical expenses 41
Health Index 42
Agriculture Fewer negative agricultural productivity shocks Insignificant positive increase in agricultural productivity 43
Agriculture Index 44
Consumption No significant impacts on consumption levels 45
Consumption Index 46
Resilience to Shocks To compare the outcomes of rural households in the treatment group hit by shocks with outcomes of rural households in the control group hit by shocks, we estimate: is the coefficient of interest (on treatment group * shock) 47
Shocks Remittance Flows Large and significant increases in remittances sent by migrants via bkash Pattern very similar to estimates for consumption Illustrates mechanism for consumption smoothing when hit by shocks 48
Shocks - Consumption Significant increase in consumption when hit by agricultural shocks (32%) Impact larger than that estimated by Jack and Suri, 2014 (7%) Insignificant increase in consumption when hit by health shocks 49
Shocks Food Consumption Similar pattern when we look at food consumption Magnitudes are slightly larger, since food consumption is likely hardest hit when households are faced with shocks 50
Positive Shocks 51
Are Migrants Key for Resilience to Shocks? To compare outcomes of rural households in the treatment group hit by shocks whose paired migrants are also hit with shocks, with rural households in the control group hit by shocks whose paired migrants are hit with shocks: is the coefficient of interest (rural and urban pairs experience shocks at same time) 52
Household Shock + Migrant Shock No impact on consumption for rural households hit by shocks, when migrant is hit by a health shock Migrants are key for provision of insurance in bad times 53
Household Shock + Migrant Shock Migrants unable to send more remittances when they are hit by health shocks 54
Results: Urban Migrants 55
Remittances 56
Remittances 57
Savings 58
Formal Employment 59
Work in Garments Industry 60
Poverty 61
Migrant Health Decreases in selfreported health indicators in terms of physical health problems, ease of daily work, bodily pain, social activities, emotional problems and severe emotional problems 62
Health Index 63
Higher Income at Expense of Health 64
Results: Robustness Checks 65
Spillover Analysis - Rural Impact on control group of adoption based on having more villagers in the treatment group 66
Spillover Analysis - Urban Impact on control group of adoption based on having more people locally in the treatment group 67
Network Spillovers - Urban Discussed bkash with social network, but that did not affect adoption. 68
Conclusion 2 big trends gaining speed: Movement of people Movement of money Extreme poverty will become more difficult to reduce Migration with remittances as a poverty reduction strategy 69
Strategy 1: Local development Labor (Capital, Human capital, X) Microfinance Graduation/ultrapoor programs Training SME
Strategy 2: Migrate and connect Labor (Capital, Human capital, X)
72
Summary Statistics 73