FIN 450F/550F Final Exam: Banking the Unbanked and Learning a New Technology
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
FIN 450F/550F
Final Exam: Banking the Unbanked and Learning a New Technology
Due: May 5, 2025 (midnight CST)
• Please submit a write-up of results, along with your R code (or a R Markdown file). You can use a diferent statistical programming software, e.g., Stata or Python, if you like.
• Answer the questions in the order given here.
• This final exam has to be done individually. No help can be solicited or provided.
• 150 points (10 bonus points).
Background: Many financial economics as well as market researchers ask how customers adopt new financial technologies when they had no previous experience with such technologies. Breza, Kanz, and Klapper (2024) conduct a field experiment with a group of unbanked factory workers in Bangladesh, who previously received all their wages in cash. The workers were randomized, and 56% of workers were introduced to digital payroll accounts (either bank or mobile – also randomized) and started receiving electronic wage payments. Some workers who were assigned to the cash wages group were introduced to either mobile money or bank accounts (equivalent to checking accounts). The authors found that exposure to the payroll accounts leads to “increased account use, accelerated learning, and avoidance of common consumer protection risks.” The measures of interest were outside transactions (transactions outside of the workplace) and direct transactions (transactions without an intermediary), from which one can estimate account use and learning outcomes from the use of the new technology as well as financial well-being (measured by spending and savings).
Task at large: Using the survey and administrative data collected by Breza, Kanz, and Klapper (2024), estimate the efects of the bank and mobile account introduction on transactions outside of the workplace, transactions without an intermediary, and the ability to accumulate savings, consumption, and remittances.
The data is available in 3 CSV files that you need to download:
1. summary__statistics + balance.csv - data used for the sample summary statistics;
2. survey__transaction__data.csv - data from the survey of the factory workers;
3. admin__transaction__data.csv - administrative data containing information about account use by experiment participants.
Q0 (10 pts): Please describe why we are using an experiment to assess whether or not digital payroll accounts are good or bad for people. Could we simply survey factory workers that do or do not have bank accounts and regress outcomes of financial well-being (e.g., savings and consumption) on an indicator for whether or not an individual has a bank account? Would we be able to learn something from that regression?
Q1 (20 pts): Generate summary statistics for the survey data in summary__statistics + balance.csv. We are interested in the marital status, years worked at the factory, family com- position (having kids), and variables for digital account usage.
Q2a (25 pts): Use the following OLS regression specification (which utilizes administrative data in admin__transaction__data.csv):
Yi,T,t = αt + αT + βpayrollTi,payroll + Xi(0)√ + εi,T,t ,
where Yi,T,t is an outcome of interest, i indexes individuals, T indexes the number of months since treatment, and t indexes calendar month-by-year. Ti,payroll is a treatment indicator equal to one if individual i was assigned to the bank or mobile payroll account treatment condition. Because we only have data from the partnering financial service providers for the treatment arms that received payroll accounts, members of the checking account only treatments (who continued to receive their wages in cash) comprise the omitted category. We include calendar month-by-year and months- since-treatment fixed efects to absorb all confounding time trends in usage. Xi is a vector of randomization strata fixed efects and εi,T,t is a stochastic error term. Standard errors are clustered at the worker level. Because workers only appear in the administrative data once they receive their account, all observations capture post-treatment information.
Estimate treatment efects on the account use variables: account withdrawals, transactions exclud- ing withdrawals, sending money, purchases, deposits, and account balances.
Q2b (25 pts): Estimate treatment efects on transactions at the workplace and outside of the workplace (withdrawals, transactions excluding withdrawals, deposits, and all transactions).
Q3a (25 pts): When we use the survey data (from survey__transaction__data.csv), one can additionally compare the payroll and account only groups to workers in the pure control condition and incorporate both pre- and post-treatment observations. Use the following panel regression specification (which utilizes survey data and administrative data):
Yi,r = ηr + ηi + θacctT
r(o),a(st)cct + θpayrollT
r(o),p(s)tayroll + µi,r ,
where Yi is an outcome of interest measured in the endline, midline follow-up, and baseline surveys.
The variable T
r(o),a(st)cct is a binary indicator for whether individual i has been assigned to the account
only treatment and survey round r occurs after the treatment has been administered to i. Similarly,
T
r(o),p(s)tayroll is a binary indicator for whether individual i has been assigned to the payroll account
group and survey round r occurs after the treatment has been administered to i. The variables ηi and ηr denote worker and survey round fixed efects, respectively. µi,r is a stochastic error term. When outcomes are only measured in the endline surveys, we include randomization strata fixed efects in the place of the worker and survey round fixed efects.
Estimate treatment efects on direct (no intermediary) transaction variables from the administrative data using survey data variables in the above specification: agent-to-agent transactions, agent- to-person transactions, person-to-agent transactions, person-to-person transactions, indirect, and direct transactions.
Q3b (25 pts): Estimate treatment efects on savings and consumption variables from the survey data: indicator for having savings, formal savings, informal savings, total amount of savings, total remittances, total consumption, food and non-food consumption, and buying items over TK 1,000.
Q4 (20 pts): What conclusions can you make about the efects of digital payroll and payment accounts on account usage and savings and consumption variables? Please discuss the OLS and panel regression results in detail.
Bonus question (10 pts): Something that Breza, Kanz, and Klapper wanted to track from the study is whether participants have trust in the new technology (namely, mobile and bank accounts with digital payrolls). Create the following graphs from the endline survey data:
1. A bar graph representing the share of participants who feel “comfortable” or “very comfortable” leaving (1) BDT 1,000 and (2) BDT 5,000 if they are in the mobile payroll group. Also create a bar for the control group.
2. A bar graph representing the share of participants who feel “comfortable” or “very comfortable” leaving (1) BDT 1,000 and (2) BDT 5,000 if they are in the mobile account group. Also create a bar for the control group.
2025-05-16