The Value of Fintech for Retail Consumers 


Sumit Agarwal 


Sangheum Cho 


Hyun-Soo Choi 


Leora F. Klapper* 


 


PRELIMINARY DRAFT. PLEASE DO NOT CITE!!! 


 


Abstract: 


We measure the value for retail consumers of using a Fintech platform to send international 
remittance payments. The value of Fintech mostly comes through overcoming various 
frictions such as the high cost of sending payments or the time-constraints of transactions. 
Using detailed transaction-level international remittance data from a leading Fintech firm 
in Korea, we find that the Fintech significantly lowers the cost for low-income workers to 
send money home. On average, the Fintech platform lowers remittance cost by 10.6 percent, 
as compared to traditional commercial banks. Yet, while the Fintech enhances consumer 
welfare by increasing the flexibility of transactions, the flexibility offered by a cancellation 
feature of the Fintech product might not always lead to the optimal exchange rate timing of 
remittance transactions and can also harm consumers by amplifying their behavioral bias. 
We also construct social networks among workers in our sample using the Fintech and use 
this data as an identification strategy to show learning effects among these workers. 


 


 


 


 


 


 


 


*Aggarwal: National University of Singapore, ushakri@yahoo.com; Cho: Korea Advanced Institute 
of Science and Technology, sh.cho@kaist.ac.kr; Choi: Korea Advanced Institute of Science and 
Technology, hschoi19@kaist.ac.kr; Klapper: World Bank, lklapper@worldbank.org. We are 
grateful to Sentbe for providing the data used in this paper. This paper’s findings interpretations, 
and conclusions are entirely those of the authors and do not necessarily represent the views or 
policies of the World Bank or International Monetary Fund, their Executive Directors, or the 
countries they represent. 



1. Introduction 


Globalization has increased labour migration and the number of international migrants reached 
nearly 272 million in 2019, up from 153 million in 1990 (United Nations, 2019). Many migrants 
are motivated by higher wages and better opportunities in foreign countries and the growth in 
migration has been accompanied by an increase in the value of international remittance sent 
back to migrants’ home countries (Lucas and Stark, 1985; Funkhouser, 1995; and Clemens and 
Tiongson, 2017). According to World Bank estimates, 
remittances 
will total $573 billion dollars 
in 
2019, of which US$422 billion went to developing countries that involved 250 million migrant 
workers. For some individual recipient countries, 
remittances 
can be as high as a third of their 
GDP (World Bank, 2019). 


Sending international remittance can be expensive, unsafe, and difficult for workers. The global 
average cost of sending $200 internationally has only fallen from 9% in 2011 to 7% in 2019, 
which is still a financial burden for migrant workers (World Bank, 2019). Most money is sent 
through cash couriers, informal methods of transferring money (like Hawala), money wiring 
services (like Western Union), or banks. For example, according to a 2019 survey of migrant 
workers living in Korea, 81% of Nepali respondents informally sent money home mainly 
through individual brokers, known as Hundi, despite the high risk (IOM, 2019). However, the 
use of Financial technology (Fintech) to make international remittances has grown over the past 
few years with the use of mobile apps and mobile money services such as TransferWise and 
PayPal Zoom. 


In this paper, we estimate the value of Fintech for retail consumers using transaction-level data 
of international remittances sent using a Fintech platform. The value of Fintech is mostly 
derived by relaxing various frictions such as the high cost or time (spatial) constraints of making 
a transaction. Our dataset allows us to study: First, the value of Fintech to reduce the costs of 
international remittances; second, the consequence of greater flexibility in the timing of 
transactions; and third, the influence of social networks on timing decisions. 


Our paper contributes to the literature on the value of Fintech. Chen et al. (2019) studies the 
value of Fintech in the view of innovators and financial institutions. While they find substantial 
value creation to innovators, they find mixed results for other financial industry participants by 
their willingness to adopt new technology. Berg et al. (2020) and Agarwal et al. (2020) discover 


the benefit of Fintech from constructing credit score out of digital footprints. Jagtiani and 
Lemieux (2017) finds the benefit of Fintech from the information efficiency achieved by online 
lending platform. Our paper complements this literature by focusing on additional benefits of 
Fintech: the reduction in cost and the flexibility in the timing of payment transactions. 


We use detailed transaction-level international remittance data from ‘Sentbe’, a Korean digital 
money transfer operator, which sent more than USD$880 million in accumulated remittances 
from Korea over the 2019 calendar year.1 Until recently, the remittance market in Korea was 
controlled entirely by large financial institutions—leading to high foreign transfer fees and 
costly delays. However, regulatory changes introduced in 2017 allowed non-bank Fintechs to 
offer consumer remittance services. Sentbe has drastically reduced the cost of remittance to 1.2% 
of the total remitted amount, while the total remittance costs of sending money through a bank 
remains at about 8%. Typically, money sent via Sentbe is available for family members to collect 
in 1-hour, as compared to 2-3 days for conventional bank services. Recipients can choose their 
payout option: a home delivery or cash pick-up point. 


1 For more information see: https://www.sentbe.com/en/. 


Fintech platforms provide both a cheaper solution for international remittances and additional 
flexibility in the timing of remittance transactions. Compared to traditional banks, Fintech 
platforms enable senders to overcome time and physical constraints. For example, typically, 
workers would need to visit a bank branch during the bank’s working hours. This constraint is 
especially binding for those who are tied to their jobs during the bank’s working hours, e.g. low 
skilled workers and workers who need to travel significant distance to visit a bank branch, such 
as workers living in sparsely populated areas. Fintech allows users to execute remittance 
transactions regardless of time and physical location. 


There are several advantages of using this data to study our research questions. First, detailed 
information on the timing and amounts of international remittances, combined with rich 
demographic information about the consumers, allows us to examine individual remittance 
decisions. Second, we can exploit two novel features of the platform: A “cancellation” feature 
that allows users to cancel their order within 24 hours, for no fee. In addition, we can exploit a 
cash bonus for referring the product to friends to construct “social networks” of users. 



We use this data, plus daily spot exchange rate for a window around the executed payment, to 
construct a measure of how well users time the exchange rate market, which we call our 
“optimality” score. We find that users are more likely to optimize the timing of their payment 
when: (i) transaction amounts are larger; (ii) the exchange rate appreciated on the previous day; 
and (ii) and the cancelation feature was used during the optimization period. We also show that 
this information is shared among social networks. 


Most of our sample are foreign workers from Northeast and Southeast Asia working in Korea, 
who are typically low-skilled workers and need to transfer a significant part of their earnings to 
support their families back home.1 Although these workers having little formal education or 
experience using formal financial services, international remittance are a large portion of their 
income and they would exert their best efforts in making remittance decisions. 


1 Sentbe focuses on sending money to the Southeast and Northeast Asia corridors and foreign workers in our sample 
are from Bangladesh, China, Indonesia, India, Cambodia, Malaysia, Philippine, Pakistan, Thailand, and Vietnam. 


Our data also shows that workers are more likely to send an international remittance payment 
the day after the spot exchange rate of their home country appreciates (relative to the KRW). In 
economic terms, for every one-standard deviation increase in spot exchange rate on the 
previous day, the likelihood of remittance transaction increases 4.2%. This result is robust to 
using the appreciation (percentage change) of spot exchange rate on the previous day. 
Importantly, this shows that workers respond to changes in exchange rates when deciding when 
to send international remittances. 


We find two other features of the Fintech platform that improve the welfare of workers. First is 
the Cancellation feature that allows users to hold multiple remittance orders up to 24 hours and 
users can decide at any time which orders to execute before they expire. We find that workers 
from countries with higher levels of financial development are more likely to use this feature. 
Use of the Cancellation feature is also associated with a higher optimality score: the average spot 
exchange rate applied for completed remittance transactions is 1.84% higher than the average 
spot exchange rates of those cancelled orders associated with the completed order. 


Finally, we exploit a unique feature that is associated with Fintech platforms to test whether 
workers learn through their social networks the optimal timing for their international 


remittance payments. Workers that use Sentbe can recommend the platform to their friends in 
exchange for bonus credits that can be used for future remittance transactions by the referrers. 
We use this referral data to construct social networks among the workers in our sample and use 
this data as an identification strategy to show learning effects among these workers. 


2. The use of fintech for overseas remittance payments 


Over the last decade, international remittance sent to low- and middle-income countries has 
grown fast (Figure 1). The total amounts of international remittances sent to low- and middle-
income countries increased from about 150 billion dollars in 2004, to 500 billion dollars in 2019 
(Panel A). Furthermore, excluding china, international remittances sent to low- and middle-
income countries exceeds the total amounts of foreign direct investment to these countries 
(Panel B). International remittances also make large contributions to the local GDP of many 
countries, i.e. international remittances sent to Philippines contributed 10.2% of total GDP in 
2018 and contributed more than 5% of GDP in Pakistan, Vietnam, Cambodia, and Bangladesh 
(Figure 3). 


Despite fast growth, sending international remittance often remains difficult and expensive 
(Figure 2). For example, according to the World Bank, the global average cost of sending $200 
is 7% of the transaction amount, a decline of only 2 percentage points from 2011 (Panel A). The 
cost of remittance is also sensitive to the competitiveness of the industry, as shown by the 
average cost, by type of service provider, from 2011 to 2020 (Panel B). Banks on average charge 
the highest rate (11.55%), while public post offices charge the lowest (5.25%). Money transfer 
operators charge, on average, 6.4%--but this charge increases to 8.66% when the money 
transfer operator has an exclusive partnership arrangement with the local post office. 


To address the challenges to workers sending money home, there has been a global surge in the 
creation and growth of Fintech firms servicing international remittance payments. These 
Fintech solutions offer cheaper and easier solutions, e.g. TransferWise or PayPal. In Korea, a 
long history of strict regulations on foreign exchange transactions (the Foreign Exchange 
Transactions Act), limited formal international remittance payments to regulated banking 
institutions, who charged high prices with limited competition. In June, 2017, the Ministry of 
Economy and Finance of Korea relaxed the regulation and allowed non-bank financial firms to 


compete in the international remittance market. Our data is from one of the Fintech firms who 
entered the market. 


3. Data and summary statistics 


We use detailed transaction-level international remittance data and individual-level 
demographic data from one of the leading Fintech firms in Korea, the Sentbe, for our analysis. 
The firm’s target customers are mainly foreign workers in Korea and includes individuals mostly 
from nine Asian countries: Bangladesh, Indonesia, India, Cambodia, Malaysia, Philippine, 
Pakistan, Thailand, and Vietnam. A key feature of our data is that it includes low-skilled workers 
who send a significant portion of their earnings back home to support their families. Although 
these workers are financially unsophisticated, they exert their best efforts in making remittance 
decisions, which have a large impact on their utility and wealth. 


The average age of our sample individuals is about 31 but it varies from 27 to 33 across 
countries, etc. Table 1 reports summary statistics of our daily individual-level data, by 
destination country. Our sample has 25,994 individuals who make 476,659 payment 
transactions from February 2016 to March 2020.1 The largest percent of workers in our sample 
are from the Philippines with 8,231 users (32%) who account for 244,297 transactions (52%) 
with the average transaction amounts of 497,465 KRW, about USD$415. The next largest 
country represented in our sample is Vietnam, with 7,743 users (30%), followed by workers 
from Indonesia with 4,994 users (19%). 


1 We limit our sample to individuals whose nationality is matched to the remitted currency. For example, we do not 
include Koreans sending money to the countries. The only exception is for Cambodians, since the Cambodian Riel 
is pegged to the US dollar and Cambodians send US dollars home to Cambodia. There exist some users with multiple 
transactions within a day that we collapse to construct daily transactions. 


Workers in our sample send international remittance payments, on average, about 2.1 times a 
month with 721,912 KRW, about USD$600, per remittance. There are notable patterns between 
sending amounts and remittance frequency by country: Workers from the Philippines, 
Indonesia, Malaysia tend to remit more frequent, smaller denomination payments, while 
workers from India tend to remit less frequent, larger denomination payments. 



Finally, for the foreign spot exchange rates (foreign currency units per KRW), we use the spot 
exchange rate that Sentbe quoted within their app to their consumers.1 Figure 5 plots the spot 
exchange rates for our sample countries during our sample period.2 


1 Sentbe shows users high frequency foreign exchange spot rates, updated every 10-minutes. This information is 
provided primarily from KEB-Hana Bank, who specializes in the foreign exchange market, and also from global 
benchmarks such as Xe.com and investing.com. 


2 Sentbe introduced international remittance services to most countries in 2018, except for Indonesia, Philippines, 
and Vietnam which were introduced to the platform in 2016. 


3 We use the cost for workers to send international remittances at a bank branch. Some banks now offer online 
apps with cheaper solutions, but these were not available during our early sample period and are still more 
expensive than the cost of using a Fintech solution. 


4. Measuring the benefits of fintech 


In this section we estimate the cost savings of using a Fintech, relative to traditional banking 
products. Banks charge five different layers of fees for international remittances: The first is a 
fixed fee per remittance (‘Telegraphic Charge’), the second is the variable fee as a percentage of 
the remittance amount, and the third is the margin on the foreign spot exchange rate. Additional 
fees are charged by the intermediary bank for using SWIFT (Brokerage Fee), and the foreign 
receiving bank. 


To compare fees, we use the list fees for two major banks in Korea, IBK bank and Woori Bank, 
which are commonly used by workers to send international remittances. Table 2 reports the 
cost structure.3 Panel A compares the fee structures of IBK bank and Woori Bank to the fee 
charged by the Fintech firm. All three institutions charge 5,000 KRW (about USD$4) as 
Telegraphic Charges. Only the traditional banks charge a Brokerage fee of about 10,000 KRW 
(USD$8). Exchange rate margins are about 1% of the amount sent, but varies by bank and 
destination country (as shown in Panel B.) The Fintech firm charges a fixed rate 1% margin for 
all countries. 


Based on the fee structures in Table 2, Figure 6, Panel A compares the remittance cost of the 
Fintech firm to the remittance cost of traditional banks for a sample of sending amounts. The 
solid line reports the cost of remittance of the Fintech platform, by sending amounts, and the 
dotted line reports the cost of remittance of the traditional banks, by sending amounts. The 
green bar reports the distribution of sending amounts of remittance transactions, which we use 


as a weight to compute the average cost reduction for Fintech users across the distribution. The 
difference between two lines shows the relative cost reduction for each sending amounts. For 
example, the fees for sending the median international remittance payment amount in our 
sample, 374,690 KRW (USD $314), the Fintech platform charges 5,000 KRW (USD $4) for 
Telegraphic Charge and 3747 KRW (USD $3) for the exchange rate margin. This adds up to 8747 
KRW (USD $7), which is approximately 2% of the sending amount. However, traditional banks 
charge 6.3% for the same amount of remittance. 


We find that the Fintech solution charges are on average 10.6 percent lower than traditional 
banks. This sizable effect is largely because Fintech’s have a relative advantage in sending 
smaller denomination payments, typical in our sample: Workers send on average about 
USD$600 (721,906 Korean Won (KRW)) per month. Note that the difference is largest in 
transactions with smaller amounts. A large number of workers in our sample earn a low-income 
and send low-denomination payments—that incur the highest fees (as a percent of amount 
sent). 


Panel B reports the similar results by the country of destination.1 The reduction is largest among 
workers from Indonesia (13.1 percentage points) and smallest among workers from India (4.1 
percentage points). 


1 We do this analysis for seven countries, excluding Bangladesh and Pakistan since neither bank offers remittance 
service to these two countries. 


2 Weekend service is available at banks for a significant additional fee. 


However, using a Fintech may not always be the preferable means to send international 
remittances: When the remitted amount is large, workers may prefer to send their payment 
through a traditional bank, since the cost is similar, and the reputation of safety and reliability 
is higher. In general, Fintech platforms appear to have a comparative advantage for low-
denomination, high-frequency international remittance payments. 


We also find that a large number of payments are made on the weekend and outside of typical 
bank operating hours. Since workers using traditional bank services can only transact during 
lunch breaks during the week or on weekends2, this finding suggests a severe constraint that 
may be imposed on individual workers if a Fintech option did not exist. 



5. Timing international remittance payments 


Fintech platforms not only provide a cheaper solution for international remittance payments, 
but also provide additional flexibility in the timing of remittance transactions. Compared to 
traditional banks, Fintech platform enable workers to overcome time and physical constraints. 
To send an international remittance payment using a bank, workers are required to travel, in-
person, to a bank branch during its opening hours. This constraint is particularly challenging for 
workers who are tied to their jobs during the bank’s working hours, e.g. low skilled workers, or 
those who need to travel significant distance to visit a bank branch, e.g. workers in sparsely 
populated areas. The only alternative to banking during working hours was to pay even higher 
fees at special bank branches exploiting this friction.1 In comparison, workers can send 
international remittances using a Fintech app on their phone any time of day or night, weekday 
or weekend, from the comfort of their home. 


1 To service Southeast Asian workers, special bank branches open on the weekend to send international remittances 
at a higher fee. 


2 Service was suspended from July 17th-December 7th 2017, while Sendbe acquired a license for overseas 
remittance, required by the government’s deregulation in July 2017.There was another short stoppage of the 
service from February 15th-February 18th 2018 due to the Lunar New Year Holidays in Korea. We use the complete 
sample period from 2016-2020 for our main analysis, but including only the post-2018 period does not change our 
main results. 


Figure 4 shows the distribution of our sample of transactions over time. Panel A reports the 
number of remittance transactions by calendar day and shows that the number of transactions 
gradually increases as the Fintech platform matures.2 We find that the calendar days with the 
highest number of transactions are between the 10th-14th of the month, which are the typical 
salary days in Korea. Panel B reports the number of remittance transactions by the day of a 
month. As seen in the monthly peaks in Panel A, remittance transactions are more likely to occur 
between the 10th-14th of the month. Panel C reports the number of remittance transactions by 
the day of a week. We find that the remittance transactions are lower during the weekends and 
highest on Monday. 


Panel D reports the number of Sebtbe remittance transactions, averaged over time by the time-
of-day (in 10-minute intervals). Although the greatest number of transactions are during 


lunchtime (12:30), we find a high number of transactions done in the evening, after traditional 
bank working hours. 


5.1 Day and time of day 


Our analysis uses daily individual-level transactions from February 2016 to March 2020. We 
construct a balanced sample by replacing missing observations for 1 month before the day of 
first remittance to 1 month after the day of the last remittance payment with zero. The final 
dataset has 10,623,364 observations. 


Table 3 reports the results from a linear probability model estimating daily remittance 
transaction patterns. The dependent variable is a dummy variable D{Remittance}i,t that equals 
one if an individual i transacts on day t and 0 otherwise. D{Remittance}i,t has a mean of 0.04 
indicating that users remit funds using the Fintech platform on about 4% of sample days. 
Column (1) shows result using dummy variables for the days of a week, i.e. Monday, Tuesday, 
Wednesday, Thursday, Friday, and Saturday are included as independent variables. We include 
individual fixed effects and country-year-month fixed effects to control for potential differences 
in individual characteristics and fluctuations in the foreign exchange market. All standard errors 
are clustered by individual and country-year-month level. 


We find that remittance transactions are more likely to happen on Monday and the likelihood of 
remittances gradually decreases daily through the week. The likelihood of remittance is 
significantly lower on Saturday and Sunday. This is contrary to our expectations that digital 
remittance transactions would be more convenient—and more commonly used—during non-
working days/hours. 


Column (3) adds a dummy variable Salary Dayst that is equal to 1 for the 10th-14th days of the 
month, and 0 otherwise. We find that remittance transactions are 40% more likely to occur 
during salary days, as compared to the unconditional probability of remittance. Many workers 
are supporting families back home and are likely to send money home soon after their salary is 
paid. 



5.3 Cancellation feature 


5.2 Exchange rate appreciation 


Next, we examine the relationship between the daily remittance decision and the lagged 
appreciation (percentage change) in spot exchange rate of currency c: [ΔSPOTc,t−1 = 
log(SPOTc,t−1/SPOTc,t−2)]. In Column (2), we find that the likelihood of making a remittance 
transaction increases with the appreciation in spot exchange rate on the previous day. ΔSPOTc,t−1 
has a mean of -0.01 with standard deviation of 0.42. In economic terms, for every 1 standard 
deviation increase in spot exchange rate on the previous day, the likelihood of remittance 
transaction increases by 4.2% ((0.004*0.42)/0.04). 


Similarly, we find that workers are more likely to transact after the appreciation of spot 
exchange rate on the previous day, regardless of the magnitude of the change. Column (3) uses 
a dummy variable of D {ΔSPOTc,t−1 > 0} that equals to 1 if ΔSPOTc,t−1 is positive and 0 otherwise. 


What makes individuals trade more after the appreciation of spot exchange rate? One potential 
story is due to the behavioral bias of our sample individuals. Individuals are known to suffer the 
“law of small numbers” (Tversky and Kahneman (1971)) and likely to expect reversion to the 
mean, i.e. that an appreciation of the spot rate on the previous day will be followed by a 
depreciation of the spot rate on the following day. In this case, we would expect more remittance 
transactions following the appreciations of spot exchange rate. 


Next we examine a unique features of the Fintech platform, which are not available to workers 
making bank payments: the Cancellation feature, which allows users to hold multiple remittance 
orders up to 24 hours and users can decide which orders to execute before they expire. Since 
spot exchange rate quotes do not change over the weekend, the importance of the Cancellation 
feature on remittance decision is limited during weekends. We focus on the sample observations 
during weekdays so that our sample size reduces from 10,623,364 to 7,591,671 with 368,189 
number of actual remittances.1 


1 For robustness we include weekends observations and find no qualitative differences. 


Table 4 reports the descriptive statistics on the characteristics of 368,189 transactions by the 
use of Cancellation. The Cancellation feature is more typically used by women, young workers, 


and for larger remittance amounts. Usage varies by the nationality of individuals: about 10% of 
remittances by Thai workers use the cancellation feature, as compared to about 5% of 
remittances by workers from Cambodia. To help explain these trends, Figure 7 plots the 
Financial Development Index of a country in 2018 from IMF and the average usage of 
Cancellation by the individuals from the country. We find that the use of the Cancellation feature 
is positively correlated with the Financial Development Index of the home/receiving country. 
The slope of regressing Cancellation on Financial Development Index is 6% and statistically 
significant at 5% level. 


Table 5, Column (1) reports our results using a linear probability model of Cancellation on the 
likelihood of a remittance payment. The dependent variable is a dummy variable 
D{Remittance}i,t that is equals to 1 if an individual i transacts on day t and 0 otherwise. We 
include Salary Dayst as a control variable with individual fixed effects and country-year-month 
fixed effects. Our main independent variable of interest is a dummy variable Cancellationi,t that 
equals to 1 if an individual i uses Cancellation for any remittance transaction on day t. We find 
that the likelihood of remittance transaction increases by 35% when Cancellation is used. 


In Column (4), we put both D{ΔSPOTc,t−1 > 0} and Cancellationi,t into the regression to find that 
two effects are not inter-dependent in a sense that individual effects remain same. 


In Columns (5)-(8), we report similar results using the amounts of remittances. Since we 
observe remittance amounts only if the remittance occurs, our sample size reduces to 368,189. 
We find that greater payment amounts are associated with larger payment amounts, following 
spot exchange rate appreciations, and with the use of the cancellation feature. 


6 Optimizing remittance transactions 


Next, we analyze whether workers in our study optimize the timing of their remittance 
payments by using daily exchange rates to compute a set of counterfactual amounts, assuming 
the transaction occurs on another day within the window of [-5, +5] around the actual 
transaction. We normalize the amount of the original remittance, in receiving currency, to that 
at t = 0, the amount is equal to 1. 



The measure indicates whether a remitted payment receives the optimal exchange rate within 
a 2 week horizon, or not. If the highest value occurs at day 0, this indicates that the worker was 
able to pick the best day to transact within a 2-week window. If the peak occurs prior to day 0, 
this implies that the worker should have sent the transaction earlier and a peak following day 0 
suggests that the worker should have made the transaction few days later in order to optimize 
the exchange rate. 


Figure 8, Panel A plots average returns using all remittance transactions in the window of [-
5,+5]. We find that the peak generally occurs at day 0, although the magnitude is small. On 
average, our sample individuals outperform around 0.04% compared to the exchange rates in 
previous 5 days of the remittances and outperform around 0.02% compared to the exchange 
rates after 5 days of the remittances. 


6.1 Definition of Optimality Score 


To understand the link between the determinants and the optimality in overseas remittance 
timing, we define an optimality measure for each remittance transaction by computing a set of 
counterfactual amounts in receiving currency if the transaction occurs in the other days in the 
window of [-5, +5] around the actual transaction. Normalizing the amounts by the original 
amounts of remittance in receiving currency at t = 0, the measure equals to 1 at t = 0. Having the 
peak at t = 0 means that the individual was able to get the best exchange rate in 2-weeks window. 
Having the peak before (after) t = 0 means that the individual could have done better if he/she 
transacted few days earlier (later). 


We define the average difference in receiving amounts between t = 0 and the other days in the 
window of [-5,+5] as the Optimality Score [−5,+5]. We find high optimality score among younger 
users with significant variation by their nationalities. For the individuals with the optimality 
score in the top 1/3 among our sample individuals, the receiving amounts at t = 0 are on average 
0.54% higher to the amounts in the window of [-5,-1] and 0.49% higher to the amounts in the 
window of [1,5]. 


We find that the sending amounts are positively associated with the optimality score but only 
during the pre-transaction period of [-5,-1]. This indicates that the individuals put extra effort 


in picking remittance timing for the larger remittances since the effort only can affect on the 
optimality in pre-transaction period. 


Why do workers make more transactions after the appreciation of spot exchange rate? One 
possible reason is behavioral biases: For example, individuals are susceptible to the “law of 
small numbers” (Tversky and Kahneman, 1971), that is, they tend to generalize from small 
amounts of data. In this case, the appreciation of spot exchange rate on the previous day would 
cause workers to expect the spot rate to continue to appreciate (and their wages to depreciate). 
However, if the home-country’s exchange rate recovers (depreciates), workers could lose money. 


When a new technology arrives and relaxes pre-existing constraints on consumers, is it always 
beneficial to consumers? In theory, rational agents should be better off with additional flexibility 
in their choice set. However, Barber and Odean (2000) find that individual investors who hold 
common stocks directly pay a tremendous performance penalty for active trading because of 
overconfidence in their timing of the market. Similarly, workers with additional flexibility on the 
timing of international remittance payments—and greater exposure to exchange rate 
fluctuations— may be worse off, particularly when the workers have low financial capability. 


The small magnitude in the outperformance is partly due to the difference in individuals’ 
characteristics. We construct a measure of optimality to formally investigate the optimal 
behavior of individuals’ remittance decisions. Panel A of Figure 9 shows an example of a 
remittance transaction by a Vietnamese user on Aug 28, 2018. The black solid line represents 
the spot exchange rates from -5 to +5 days of the remittance transaction relative to the spot 
exchange rate of the actual remittance. The relative spot exchange rate on day 0 is 1 since it 
serves as a reference point. We compute the average difference between the reference point and 
the relative spot exchange rates to construct the Optimality Scorei,t [−5,+5]. In this example, the 
Optimality Scorei,t [−5,+5] is 0.0083 indicating that the spot exchange rate applied to the 
remittance was higher than the average spot exchange rates from -5 to +5 days of the remittance 
by 0.8%. 


But the optimality score can vary widely across remittance transactions. We provide other 
examples in Panel B of Figure 9. The top-right figure reports a remittance transaction by an 
Indian user on Nov 16, 2018 with the Optimality Scorei,t [−5,+5] of 0.0028. This means that the 


spot exchange rate applied to the remittance was higher than the average spot exchange rates 
from -5 to +5 days of the remittance by 0.3%. We see that the user would get better rate if he/she 
transacted few days earlier. 


The optimality score can be negative if the remittance transaction was not optimally done. 
Bottom-left figure reports a remittance transaction by an Indonesian user on Jun 26, 2018 with 
the Optimality Scorei,t [−5,+5] of -0.0018 and this indicates that the spot exchange rate for the 
remittance was lower than the average spot exchange rates around the remittance by 0.2%. 
Bottom-right figure reports a remittance transaction by a Vietnamese user on Jan 22, 2019 with 
the Optimality Scorei,t [−5,+5] of -0.0068 indicating that the spot exchange rate for the 
remittance was lower than the average spot exchange rates from -5 to +5 days of the remittance 
by 0.7%. Indeed, the remittance transaction occurs at the worst timing in 2-weeks window. 


To check the robustness of the Optimality Scorei,t [−5,+5] measuring optimality in remittance 
timing, Table 6 reports the results comparing the Optimality Scorei,t [−5,+5] and the relative 
ranking of spot exchange rates within 11-days window. For each currency, we sort 11 spot 
exchange rates from -5 to 5 days of the remittance transaction to assign a ranking for the spot 
exchange rate of day 0, i.e. the ranking equals to 11 if the spot exchange rate of day 0 is the 
highest within the window period and the ranking equals to 1 if the spot exchange rate of day 0 
is the lowest. If correctly measured, Optimality Scorei,t [−5,+5] should be positively correlated to 
the probability of having higher ranking. 


Panel A reports the logistic regression results using the dummy variables of the relative ranking 
as dependent variables. The dependent variable in Column (1) is D {Rank = 11} (Highest) which 
equals to 1 if the ranking of the spot exchange rate on day 0 is the highest and 0 otherwise. We 
include country-year-month fixed effects. We find that Optimality Scorei,t [−5,+5] is positively 
associated with the probability of the spot exchange rate of day 0 being the best rate in the 
window period. Column (2) uses D {Rank ≥ 9}, which equals to 1 if the ranking of the spot 
exchange rate on day 0 is within top 3 and 0 otherwise, to find similar results. 


Alternatively, Column (3) uses D {Rank ≤ 3}, which equals 1 if the ranking of the spot exchange 
rate on day 0 is within bottom 3 and 0 otherwise, to find that the Optimality Scorei,t [−5,+5] is 
negatively correlated with the probability of the spot exchange rate of day 0 being the best rate 


in the window period. Column (4) uses D {Rank = 1} (Lowest), which equals to 1 if the ranking 
of the spot exchange rate on day 0 is the lowest and 0 otherwise, to find that the probability of 
the spot exchange rate of day 0 being the lowest is lower with lower Optimality Scorei,t [−5,+5]. 
In Panel B, we find similar results using the Pearson correlation between Optimality Scorei,t 
[−5,+5] and the dependent variables in Panel A. 


When we sort our sample individuals by the average Optimality Scorei,t [−5,+5], we find clearer 
view on the small magnitude occurring in Panel A of Figure 8. Panel B of Figure 8 plots the 
average returns using all remittances by the individuals with the optimality score in top 1/3. We 
again find that the peak occurs just at day 0 but with much larger effect. On average, our sample 
individuals outperform around 0.54% compared to the spot exchange rates in previous 5 days 
of the remittances and outperform around 0.49% compared to the spot exchange rates after 5 
days of the remittances. 


6.2 Determinants of Optimal Remittance Transactions 


What explains the optimality in the timing of remittance transactions? How does the optimality 
in the timing of remittance transactions relate to the determinants of remittance decision? 


In Table 7, we first report the average Optimality Scorei,t [−5,+5] by various individual 
characteristics. We find that the individuals in their 30s show higher optimality score than other 
age groups. We do not find much difference between female and male users. By country, 
individuals from Pakistan, Bangladesh, and India have higher average optimality score while 
individuals from Philippines, Cambodia, and Indonesia have lower average optimality score. 


 Interestingly, we find that the optimality score increases from 0.02 to 0.08 when the users use 
Cancellation for their remittance transactions and it increases from -0.08 to 0.15 if the users 
make a remittance payment following the appreciation of spot exchange rate on previous day. 
In other words, a higher optimality score is correlated with workers that used the cancelation 
option in prior transactions, suggesting that canceling orders helps workers better time the 
exchange rate market. 


Table 8 reports the summary statistics of variables that we use for our analysis of the optimality 
score. Our variable log(SendAmounti,t) has a mean of 12.83, which is about 373,249 KRW (USD 


$313), with a standard deviation of 1.27. About 7% of our sample remittance transactions 
included a Cancellation option, 46% are associated with the appreciation of spot exchange rate 
on the previous day, and about 22% are transacted on Salary Dayst. Optimality Scorei,t [−5,+5] 
has mean of 0.03 with standard deviation of 0.41. 


Table 9 reports the panel regression results of the determinants of remittance decision on the 
optimality score. In Panel A, we use Optimality Scorei,t [−5,+5] as main dependent variable. 
Column (1) reports the effect of sending amounts on the optimality score. After controlling the 
individual fixed effect and country-year-month fixed effect, we find that larger sending amounts 
are associated with higher optimality score. A one standard deviation increase 
inlog(SentAmounti,t) increases 3% of a 1 standard deviation of the optimality score 
(0.011*1.27/0.41=0.03). This may indicate that individuals put extra effort to time larger 
payments. 


We decompose Optimality Scorei,t [−5,+5] into the optimality score in the pre-transaction period, 
Optimality Scorei,t [−5,−1] in Panel B, and the optimality score in post-transaction period, 
Optimality Scorei,t [+1,+5] in Panel C. We find that the amount of the payment is significantly 
related to a higher optimality score only in the pre-transaction period. This again supports the 
explanation that individuals put extra efforts for the larger remittances since the effort only can 
affect on the optimality in pre-transaction period. 


Column (3) reports the effect of Cancellation on the optimality score. By using the option to hold 
multiple remittance orders up to 24 hours, individuals can enhance their optimal timing in 
remittance transactions. With the Cancellation option, we find that individuals can increase 7% 
of a 1 standard deviation of the optimality score (0.029/0.41=0.07). Better performance 
associated with Cancellation is not surprising since individuals can hold multiple remittance 
orders and only need to pick the remittance with best exchange rate. Based on our calculation, 
the spot exchange rate applied for remittance transactions is higher than the spot exchange rate 
applied for the cancelled orders associated with the remittances by 1.84% on average. 


Consistent with our expectations, we find that improvements in the timing of payments occurs 
only in the pre-transaction period but not in the post-transaction period. In Panel B and C, we 
find that Cancellation improves the optimality score only in the pre-transaction period. We also 


find similar results in Panel C of Figure 8. The optimality of remittance timing is significantly 
better with Cancellation in the pre-transaction period but not in the post transaction period. 


Column (4) reports the effect of lagged percentage change in spot exchange rate on the 
optimality score. ΔSPOTc,t−1 is the log return of spot exchange rate of currency c from t−2 to t − 
1 and D {ΔSPOTc,t−1 > 0} is a dummy variable that equals to 1 if ΔSPOTc,t−1 is positive and 0 
otherwise. In Panel A, we find that the optimality score increases when individuals transact 
following the appreciation of the spot exchange rate on the previous day. The appreciation of 
spot exchange rate increases 56% of a 1 standard deviation of the optimality score 
(0.228/0.41=0.56). 


After controlling for the usage of Cancellation, we still find that D{ΔSPOTc,t−1 > 0} significantly 
increases the optimality of remittance transactions (Column (5)). When we include all the 
independent variables in Column (6), we confirm that all the results remain same indicating that 
the effects are not mutually dependent. When the adjusted R2 in Column (6) is 0.136, the 
adjusted R2 only with D {ΔSPOTc,t−1 > 0} is high as 0.131 and this indicates the dominant role of 
D {ΔSPOTc,t−1 > 0} in explaining the optimal remittance timing. 


Earlier in Table 5, we found that workers in our sample are more likely to make remittance 
transactions after the appreciation of spot exchange rate. If the behavior is purely due to the 
behavioral bias, we would expect to find negative performance when individuals follow the 
behavioral bias. However, we find significant improvement in optimal remittance timing when 
individuals follow the appreciation of spot exchange rate. How should we understand the results? 


We examine the short-term behavior of foreign spot exchange rates. Since workers in our 
sample are not professional traders in foreign exchange markets nor individual traders who aim 
to profit from the trading foreign exchange contracts, they are likely to have a narrow time 
window for remittance due to the nature of their remittance needs, i.e. monthly remittance for 
family’s living, and the behavior of short-term changes in spot exchange rates are particularly 
important for the sample individuals. 


Main dependent variables in Table 10 are the cumulative changes in spot exchange rates from 
t−1 to t+s for s = 0,1,2,3 and 4. In Panel A, we use ΔSPOTc,t−1 as main independent variable and 
we control for country-year-month fixed effects. When the spot exchange rate appreciates from 


t−2 to t−1, we find that the cumulative return from the spot exchange rate in following days are 
negative. In economic terms, for every 1 standard deviation increase in spot exchange rate from 
t − 2 to t − 1, the cumulative return of the spot exchange rate from t−1 to t+s decreases about 
0.09 to 0.22 standard deviations. We use D {ΔSPOTc,t−1 > 0} as main independent variable in 
Panel B and find that the appreciation of spot exchange rate from t − 2 to t − 1 significantly 
reduces the cumulative return of spot exchange rate in the following days. 


If the improvement in remittance timing of individuals comes through the mean-reverting 
tendency of spot exchange rates in short-term period, we should find a strong effect of D 
{ΔSPOTc,t−1 > 0} on the post-transaction optimality score. Panel C of Table 9 reports the results 
using Optimality Scorei,t [+1,+5] to find that the appreciation of spot exchange rate indeed 
explains the post-transaction optimality. 


The results related to D {ΔSPOTc,t−1 > 0} indicates that the appreciation in spot exchange on the 
previous day leads to a reversal in spot exchange rate in a near future. As a result, the remittance 
decisions by the sample individuals following D {ΔSPOTc,t−1 > 0} (as in Table 5) could be resulted 
in higher optimality in remittance timing (as in Table 9). We then arrive at two-alternative 
explanations for the results. First is the case when our sample individuals understand the short-
term mean-reverting behavior of spot exchange rate and follow the signal for their remittance 
transactions. Second is the case when our sample individuals suffer the behavioral bias 
believing the mean reversion in spot exchange rates but the spot exchange rates turned out to 
be favorable to them. To distinguish the two alternative hypothesis, we use the social networks 
among sample individuals in the Fintech platform to examine the learning effects of these 
determinants in the social networks. 


7 Learning effect through social network 


Social networks have been a key factor to explain various phenomena, including the effect of 
learning through network behavior (Hirshleifer, 2020; Munshi, 2003). One of the key features 
of Fintech platforms is the social network that are associated with growth of the services. Like 
many Fintechs, Sentbe offers cash incentives (1,500 Won, or about $4.20) for users to 
‘recommend’ the service to their friends in exchange for bonus credits that can be used to pay 


for future remittance payments. We use this referral data to construct social networks among 
our sample of workers. 


First, we report an example of a social network among Indonesian users in our sample. Table 
11, Panel A shows 11 Indonesian users with registration date, area of residence, and type of 
occupation. We label users in alphabetical order by the time of registration in the system. We 
find that this network of workers live in a similar residential area and report a similar 
occupation. Figure 10 reports the referral relationship between 11 users. User A recommends 
the service to other users (B, C, D, and H); next, user C recommends the service to additional 
users (K, J, and E); followed by user E recommending it to other users (F and I). Lastly, user F 
recommends it to user G. We define this group of 11 individuals as a social network. 


We report the summary statistics of the social networks in our sample in Table 12. Panel A 
reports the size distribution of the social networks. We define those networks with more than 4 
members as social networks1 and identify 361 social networks in our sample with an average 
number of seven users in the social network. Panel B reports summary statistics of variables 
related to social networks. Among the 24,687 workers in our sample, about 11% are associated 
with some social network. 


1 Our results on social networks are robust with any cutoff for the network size from 2 to 4 but the number of social 
network decreases as we increase the cutoff. 


As discussed earlier, on average, workers use the Cancellation feature at 65% point in the length 
of their usage periods. For example, if an individual use Cancellation in his 10th remittance 
transaction and the total number of remittance transactions of him is 20, then Time to 1st 
Cancellation is 0.5. Credit Balance of individuals has mean of KRW 2,924 (USD $2.5) with a 
standard deviation of KRW 4,773 (USD $4.0). 


In Panel C, we find that individuals in social networks make a larger number of transactions than 
workers not in social networks. Workers in social networks are also more likely to use the 
Cancellation feature significantly earlier. The fraction of members of social network being in a 
same area is higher than unconditional fraction of individuals with same nationality in the area. 
Credit Balance also seems to be higher among the individuals in social networks. Panel D shows 
that among workers in social networks, workers with larger Credit Balances are located at the 


central nodes in the networks, using measures of Centralityd and Centralitye from Banerjee et al. 
(2013). 


7.2 Clustering of remittance transactions within social network 


We first test whether remittance transactions within a social network are clustered. Table 11, 
Panel B shows an example of the time stamps of remittance transactions that occurred in the 
social network shown in Panel A on Feb 25th, 2020. Following the remittance order issued by 
user B at 16:16, different users sequentially submitted remittance orders on the same day. 


To further examine the clustered remittance transactions within a social network, we construct 
for each social network a hypothetical matching group with the same number of individuals who 
are individually matched to individuals in the social network in terms of nationality, total 
number of remittance, and the average amounts of remittances in that month. Figure 11 
compares the daily number of remittances in February 2020 between the social network in 
Figure 10 and its matching group of individuals. The blue-dashed bar shows the daily number 
of remittances in the social network and the orange-solid bar shows the daily number of 
remittances in the matching group. We find that the remittance transactions seem to be more 
clustered in the social network. 


We compute the Herfindahl-Hirschman index (HHI) of the remittance transactions in social 
network and its matching group for each trading months. For example, the HHI of the remittance 
transactions of the month in Figure 11 is 0.0089 for the social network and 0.0035 for the 
matching group. When we calculate the average HHI for all social networks in all trading months, 
the average HHI of social networks is 0.0123 while the average HHI of the matching groups is 
0.0082 so that the average HHI of social networks is 50% higher than the average HHI of the 
matching groups with t-statistics of 3.65. 


7.3 Learning through social network 


We use the learning effect through the social network to distinguish hypothesis discussed earlier. 
We find that two determinants, the usage of Cancellation and the appreciation of spot exchange 
rate on the previous day, increase the likelihood of remittance transaction and the optimality of 
the remittance timing. 



While Cancellation is unarguably a feature that helps users to improve their remittance timing, 
the role of the appreciation of spot exchange rate is less clear since the result can be driven by 
the realization of favorable behaviors in spot exchange rates to individuals’ behavioral bias. If 
individuals recognize the determinants as valuable signals for remittance transactions, we 
would expect a learning occurring in the social networks. By testing whether the determinants 
are learnt through the social network or not, we can interpret the meaning of the determinants. 


Table 13 reports the learning of Cancellation by members of a social network. The dependent 
variable is a dummy variable, Cancellationi,t , which equals 1 if an individual i uses Cancellation 
on day t and 0 otherwise. In Column (1), the main independent variable is a dummy variable of 
I_SocialNetworki,t which equals to 1 if an individual i is affiliated to any social network at day t 
and 0 otherwise. We include a dummy variable to control for the appreciation of lagged spot 
exchange rate (D {ΔSPOTc,t−1 > 0}), log(SendAmounti,t), a dummy variable for Salary Dayst , and 
country-year-month fixed effects. We find that being in a social network does not change the 
likelihood of using Cancellation. We continue to also find that the use of Cancellation is 
associated with younger users, women, and large transaction amounts. 


In Column (2), we use CancellationHistory INi,t, the ratio of the cumulative number of 
cancellations to the cumulative number of remittance transactions up to day t by individual i, as 
the independent variable. We find that individuals use Cancellation more when they have more 
experience with it. For minimizing look-ahead bias due to the individual fixed effects, we do not 
include individual fixed effects but include individual characteristics as control variables. 


In Column (3), we use CancellationHistory SNi,t, the ratio of the cumulative number of 
cancellations to the cumulative number of remittance transactions up to day t by all individuals 
in the social network s that individual i belongs to excluding the individual i’s own cancellations 
and remittances, as independent variable. We find that being in a social network actually lowers 
the likelihood of using Cancellation—but when the social network has cumulative experience 
among users of using the Cancellation feature, social network members use Cancellation more. 
This result suggests that individuals are learning about Cancellation through their social 
networks. Column (4) include both CancellationHistory INi,t and CancellationHistory SNi,t to find 
that both effects are still significant. We find that the effect from the social network experience 
is similar to the effect from individual experience. 



In Table 14, we investigate the learning through social network on the appreciation of spot 
exchange rate on the previous day. The dependent variable is a dummy variable of D 
{Remittance}i,t which equals 1 if an individual i transacts on day t and 0 otherwise. The result in 
Column (1) uses the regression specification that is similar to Table 5 except we exclude 
individual fixed effects but include individual control variables and an additional variable to 
proxy for learning. The main independent variable is SpotHistory INi,t, the ratio of the cumulative 
number of remittances on the following day with the appreciation in spot rate change to the 
total number of remittance transactions by day t. 


As in Table 5, the likelihood of remittance transaction increases with D {ΔSPOTc,t−1 > 0}, 
Cancellationi,t, and Salary Days. We also find that the likelihood of remittance transactions 
increases with SpotHistory INi,t. That is, an individual is more likely to do remittance transactions 
as he has more experience of remittance transactions associated with the appreciation of spot 
exchange rate on the previous day. This may be due to individuals’ overconfidence through their 
superior past performances. The individuals with more experience of remittance transactions 
when the spot exchange rate appreciates have higher optimality score due to the mean-reverting 
behavior of spot exchange rate in our sample. 


Column (2) includes the interaction term between SpotHistory INi,t and D {ΔSPOTc,t−1 > 0}. We find 
that the interaction is not significant. That is, individuals who make remittance transactions when 
the spot exchange rate appreciates on the previous day do not show a higher probability of 
making a remittance transaction when the appreciation actually occurs. This indicates that 
individuals are not taking the appreciation of spot exchange rate as a signal for remittance 
decisions. It seems to be a behavioral response to the appreciation of spot exchange rate given a 
positive coefficient on D {ΔSPOTc,t−1 > 0}. 


Column (3) uses as the independent variable SpotHistory SNi,t: the ratio of the cumulative 
number of remittance transactions on the following day with the appreciation in spot rate 
change to the total number of remittances before day t by all individuals in the social network s 
excluding the individual i’s own remittances. While SpotHistory SNi,t increases the likelihood of 
making a remittance payment, the interaction term between SpotHistory SNi,t and D {ΔSPOTc,t−1 
> 0} does not show a significant effect in Column (4). This indicates that individuals in social 


networks do not learn about the usage of D {ΔSPOTc,t−1 > 0} as a signal for remittance timing 
through the network. 


What does this imply to the value of Fintech to retail consumers? Additional flexibility in Fintech 
platform can allow users to improve their financial decisions though constraint-free 
transactions with some unique features in the technology such as Cancellation in our case. And 
the information can be spread out through the social networks to improve the decisions of other 
consumers in the networks. However, the additional flexibility can exacerbate the effect of the 
behavioral bias of retail consumers on their optimal decisions. Considering the general 
consensus of unpredictable short-term foreign exchange rates shown in a large body of 
empirical literature (i.e. Meese and Rogoff, 1983), the flexibility of remittance payments may 
harm consumers’ welfare when the short-term behavior of spot exchange rate changes, but 
workers remain overconfident on their ability of optimal decisions. 


8 Conclusion 


We study the value of Fintechs for retail consumers using transaction-level data of low-income 
workers in Korea sending international remittances through a Fintech platform. We find that 
the value of Fintech is mostly derived from overcoming various frictions, such as high cost or 
time/spatial constraints of bank/brick-and-mortar transactions. We find that the Fintech 
platform lowers remittance cost by 10.6%, on average, as compared to traditional commercial 
banks. However, we find mixed results regarding the time/spatial flexibility provided by the 
Fintech platform that it may not always lead to the optimal timing of remittance transactions. 
While the Fintech platform can enhance consumer welfare by allowing constraint-free 
transactions with some advanced features in the platform, the flexibilities can also harm 
consumers by amplifying their behavioral bias. 


 


 


 



References 


Agarwal, S., S. Alok, P. Ghosh, and S. Gupta. 2020. Financial inclusion and alternate credit scoring 
for the millennials: Role of big data and machine learning in Fintech. Working Paper . 


Banerjee, A., A. G. Chandrasekhar, E. Duflo, and M. O. Jackson. 2013. The diffusion of microfiance. 
Science 341:363–370. 


Barber, B. M., and T. Odean. 2000. Trading is hazardous to your wealth: The common stock 
investment performance of individual investors. The Journal of Finance 55:773–806. 


Berg, T., V. Burg, A. Gombovi´c, and M. Puri. 2020. On the rise of Fintechs: Credit scoring using 
digital footprints. The Review of Financial Studies 33:2845–2897. 


Chen, M. A., Q. Wu, and B. Yang. 2019. How valuable is Fintech innovation? The Review of 
Financial Studies 32:2062–2106. 


Choi, H.-S., and R. Loh. 2019. Physical frictions and digital banking adoption. Working Paper. 


 


Clemens, M. A., and E. R. Tiongson. 2017. Split decisions: Household finance when a policy 
discontinuity allocates overseas work. Review of Economics and Statistics 99:531 – 543. 


Funkhouser, E. 1995. Remittances from international migration: A comparison of El Salvador 
and Nicaragua. The Review of Economics and Statistics 77:137–146. 


Hirshleifer, D. 2020. Presidential address: Social transmission bias in economics and finance. 
The Journal of Finance . 


International Organization for Migration. 2019. Migration in Nepal: A Country Profile 2019. IOM: 
Switzerland. 


Jagtiani, J., and C. Lemieux. 2017. Fintech lending: Financial inclusion, risk pricing, and 
alternative information. Working Paper . 


Lucas, R. E. B., and O. Stark. 1985. Motivations to remit: Evidence from Botswana. Journal of 
Political Economy 93:901–918. 


Meese, R. A., and K. Rogoff. 1983. Empirical exchange rate models of the seventies: Do they fit 
out of sample? Journal of International Economics 14:3–24. 


Munshi, K. 2003. Networks in the modern economy: Mexican migrants in the US labor market. 
The Quarterly Journal of Economics 118:549–599. 


Tversky, A., and D. Kahneman. 1971. Belief in the law of small numbers. Psychological Bulletin 
76:105–110. 


United Nations. 2019. International migrant stock 2019. United Nations Department of Economic 
and Social Affairs, Population Division New York . 



World Bank. 2019. Migration and remittances: Recent developments and outlook. Migration and 
Development Brief 31 . 





Table 1: Individual Remittance Transactions by Nationality 


We report the summary statistics of individual remittance transactions by customers’ nationality. Our sample includes individuals-transactions from 
February 2016 to March 2020 for the remittances to 9 Southeast and Northeast Asian Countries including Bangladesh, Cambodia, India, Indonesia, 
Malaysia, Pakistan, Philippine, Thailand, and Vietnam. We report the number of sample customers, their number of remittance transactions, average 
sending amounts per transaction in USD, and average age by customers’ nationality. We also report average number of remittance transactions per month. 


Country 


Date of 1st 
transaction 


Number of 
Customers 


Number of 
Remittance 
payments 


Transaction Amount 
(USD): 


Mean Median 


Average Number 
of Monthly 
Remittances 


Average 
age of 
Sender 


Bangladesh 


Jun-18 


530 


3,583 


552 


383 


1.57 


31 


Cambodia 


Jul-18 


500 


5,041 


1,226 


905 


1.99 


28 


India 


Jul-18 


1,765 


17,498 


607 


286 


1.61 


33 


Indonesia 


Sep-16 


4,994 


91,149 


686 


391 


2.13 


30 


Malaysia 


Jul-18 


167 


2,233 


817 


527 


2.55 


27 


Pakistan 


Jul-18 


1,290 


12,346 


418 


224 


1.70 


32 


Philippines 


Feb-16 


8,231 


244,297 


1,056 


840 


2.31 


32 


Thailand 


Apr-18 


1,074 


15,182 


903 


588 


2.03 


32 


Vietnam 


Jun-16 


7,443 


85,330 


552 


383 


1.72 


28 


Total 


Feb-16 


25,994 


476,659 


1,226 


905 


2.08 


31 





Table 2: Cost Structure of Commercial Banks and Fintech 


We report the cost structures of oversea remittance of two major commercial banks in Korea and the Fintech. 
Panel A reports fees by type. Telegraphic Charges is the fixed fee charged per remittance request. Sending Amount 
Fees is the variable fee charged by remittance amounts. The commercial banks charge the fee by three different 
groups of sending amounts: up to $500, from $500 to $2000, and from $2000. Margin (%) is the average margin 
charged on the spot exchange rate of each currency. Brokerage Fees is the fixed fee by intermediary bank for using 
SWIFT. Panel B reports the margin on spot exchange rate (%) by country of destination. IBK Bank serves 
remittances to Indonesia, Malaysia, Thailand, and Cambodia. Woori Bank serves remittances to Indonesia, India, 
Philippine, Thailand, Cambodia and Vietnam. Sentbe serves the remittance transactions to Bangladesh, Indonesia, 
India, Malaysia, Philippine, Pakistan, Thailand, Cambodia, and Vietnam. The data is from banks’ websites and the 
fees are reported in KRW (KRW) except the margin. 


 



Table 3: Daily Remittance Decisions 


We report the panel regression results of individual daily remittance decisions. Starting from our data on daily 
user-level transactions from February 2016 to March 2020, we fill zeros for all users without any remittance 
transaction from 1 month before the day of first remittance to 1 month after the last day of remittance. As a result, 
the dataset has 10,623,364 observations. The dependent variable is a dummy variable D {Remittancei,t} which 
equals to 1 if an individual i transacts on day t and 0 otherwise. Column (1) reports the result using dummy 
variables of Monday, Tuesday, Wednesday, Thursday, Friday, and Saturday as independent variables. We include 
individual fixed effects and country-year-month fixed effects. Column (2) reports the result using a dummy 
variable of Weekendst as an independent variable. Column (3) reports the result using a dummy variable of Salary 
Dayst as an independent variable. Column (4) reports the result using Weekendst, Salary Dayst, and the interaction 
of Weekendst and Salary Dayst as independent variables. All the standard errors are clustered at the individual 
and country-year-month level. ***, **, * denotes 1%, 5%, and 10% statistical significance. 


 


 



Table 4: Descriptive Statistics of Cancellation Usage 


We report the descriptive statistics of the usage of Cancellation by different characteristics of individual users. We 
report the average usage of Cancellation by age, sending amounts, gender, and nationality. We divide the sample 
by age into three groups of Below 30s, 30s, and Above 30s. We divide the sample by sending amounts into three 
groups of below 30th quantile (Below Q30), between 30th to 70th quantile (Q30-Q70), and above 70th quantile 
(Above Q70). 


Group 


Average Usage 


 


Mean 
Std. Dev. 


 


Number of 


Users 


Cancellations 
Remittances 




Bangladesh 


0.077 


0.267 


442 


206 


2,659 


Cambodia 


0.050 


0.219 


453 


181 


3,597 


India 


0.088 


0.283 


1,708 


1,370 


15,587 


Indonesia 


0.057 


0.232 


4,686 


3,832 


66,906 


Malaysia 


0.095 


0.293 


155 


167 


1,764 


Pakistan 


0.078 


0.269 


1,218 


805 


10,294 


Philippines 


0.059 


0.235 


7,914 


10,732 


182,447 


Thailand 


0.104 


0.305 


1,035 


1,345 


12,995 


Vietnam 


0.077 


0.267 


7,076 


5,538 


71,940 


Total 


0.066 


0.248 


24,687 


24,176 


368,189 




Age 


 


Below 30s 
0.067 


0.250 


12,542 


10,776 


160,496 


30s 
0.065 


0.247 


10,450 


11,388 


173,917 


Above 30s 
0.060 


Sending Amounts 


 


0.237 


1,695 


2,012 


33,776 


Below Q30 
0.049 


0.216 


6,987 


5,414 


110,456 


Above Q70 
0.085 


0.278 


8,919 


9,347 


110,455 


Q30-Q70 
0.064 


Gender 


0.245 


8,781 


9,415 


147,278 


Female 
0.077 


0.267 


6,357 


8,457 


109,133 


Male 
0.061 


0.239 


18,267 


15,683 


258,748 




Nationality 



Table 5: Determinants of Individual Remittance Decisions 


 


 





 


Table 6: Validity of Optimality Score 


We report the results of logistic regression and Pearson correlation between the optimality score and the rank of spot exchange rate in 2-weeks window. 
Panel A reports the logistic regression results using the dummy variables of the relative ranking as dependent variables. The dependent variable in Column 
(1) is D {Rankc,t = 11} (Highest) that equals to 1 if the spot exchange rate of day 0 is the highest in the window and 0 otherwise. Main independent variable 
is Optimality Scorec,t [−5,+5] which is the optimality score of the day t of currency c measured in the window of [−5,+5]. We include country-year-month 
fixed effects. Column (2) uses D {Rankc,t ≥ 9} that equals to 1 if the ranking of the spot exchange rate in day 0 is within top 3 in the window and 0 otherwise. 
Column (3) usesD {Rankc,t ≤ 3} that equals to 1 if the ranking of the spot exchange rate in day 0 is within bottom 3 in the window and 0 otherwise. Column 
(4) uses D {Rankc,t = 1} (Lowest) that equals to 1 if the ranking of the spot exchange rate in day 0 is the lowest in the window and 0 otherwise. Panel B 
reports the Pearson correlation of the dummies of ranking (D {Rankc,t = k}) and Optimality Scorei,t [−5,+5]. We report the p-value of the Pearson correlation 
in angular bracket. All standard errors in logistic regressions are clustered at country-year-month level. ***, **, * denotes 1%, 5%, and 10% statistical 
significance. 



Table 7: Descriptive Statistics of Optimality Scores 


We report the descriptive statistics of the optimality score, Optimality Scorei,t [−5,+5], by different characteristics 
of individual users. We first report the average optimality score by age, gender, nationality, and sending amounts. 
We divide the sample by age into three groups of Below 30s, 30s, and Above 30s. We divide the sample by sending 
amounts into three groups of below 30th quantile (Below Q30), between 30th to 70th quantile (Q30-Q70), and 
above 70th quantile (Above Q70). We also report the average optimality score by the determinants of remittance 
decisions such as SalaryDays, the usage of Cancellation, and the sign of ΔSPOTc,t−1. 


 


 


 



Table 8: Summary Statistics of Individual Daily Remittances 


We report the summary statistics of individual daily remittances. log(SendAmounti,t) is the log of sending amounts 
(KRW) of an individual i at day t. Cancellationi,t is a dummy variable that equals to 1 if an individual i use the 
Cancellation for the remittance on day t and 0 otherwise. ΔSPOTc,t−1 is lagged daily change of spot rate of currency 
c from t − 2 to t − 1. D {ΔSPOTc,t−1 > 0} is a dummy variable that equals to 1 if ΔSPOTc,t−1 > 0 and 0 otherwise. Salary 
Dayst is a dummy variable that equals to 1 for days between 10th and 14th of each month and 0 otherwise. 
Optimality Scorei,t [−5,+5] is the average percentage difference between actual spot rate charged on a remittance 
at day t and the spot rates in the 2-weeks window of [-5, +5]. Similarly, Optimality Scorei,t [−5,−1] is the average 
percentage difference between actual spot rate charged on a remittance at day t and the spot rates in the window 
of [-5, -1] and Optimality Scorei,t [+1,+5] is the average percentage difference between actual spot rate charged on 
a remittance at day t and the spot rates in the window of [+1, +5]. Financial Development Index is the IMF financial 
development index by country and we match it to individuals by their nationality. We winsorize 
log(SendAmounti,t), ΔSPOTc,t−1, Optimality Scorei,t [−5,+5], Optimality Scorei,t [−5,−1], Optimality Scorei,t [+1,+5], and 
Age at 1% and 99% level. 


 


 



Table 9: Transaction, Product & Market Features Associated with Optimality of 
Remittance Timing 


We report the panel regression results of the determinants of optimal remittance timing. Panel A use Optimality 
Scorei,t [−5,+5] as main dependent variable. Optimality Scorei,t [−5,+5] is the average percentage difference 
between actual spot rate charged on a remittance at day t and the spot rates in the 2-weeks window of [-5, +5]. 
Column (1) reports the result using log(SendAmounti,t) as independent variable where log(SendAmounti,t) is the 
log of sending amounts (KRW) of an individual i at day t. We include individual fixed effects and country-year-
month fixed effects. Column (2) reports the result using SalaryDayst, a dummy variable that equals to 1 for days 
between 10th and 14th of each month and 0 otherwise, as independent variable. Column (3) reports the result 
using Cancellationi,t, a dummy variable that equals to 1 if an individual i use the Cancellation for the remittance 
on day t and 0 otherwise, as independent variable. Column (4) report the result using D {ΔSPOTc,t−1 > 0}, a dummy 
variable that equals to 1 if ΔSPOTc,t−1 > 0 and 0 otherwise, as independent variable. Column (5) include both 
Cancellationi,t and D {ΔSPOTc,t−1 > 0} as independent variables. Column (6) includes all the covariates. Panel B use 
Optimality Scorei,t [−5,−1] as main dependent variable. Optimality Scorei,t [−5,−1] is the average percentage 
difference between actual spot rate charged on a remittance at day t and the spot rates in the window of [-5, -1]. 
Panel C use Optimality Scorei,t [+1,+5] as main dependent variable. Optimality Scorei,t [+1,+5] is the average 
percentage difference between actual spot rate charged on a remittance at day t and the spot rates in the window 
of [+1, +5]. All standard errors in panel regressions are clustered at individual and country-year-month level. ***, 
**, * denotes 1%, 5%, and 10% statistical significance. 


 


 


 



Table 9 Continues 


 


 


 



Table 10: Short-term Behavior of Foreign Exchange Rate 


We report the panel regression results of short-term behavior of foreign exchange rates. We use the exchange 
rates of 9 countries that Sentbe is servicing. The foreign exchange rate is expressed as foreign currency unit per 
KRW. Depende nt variables are ΔSPOTc,t−1→t+s for s = 0,1,2,3,4, which are the log ratio of spot exchange rates 
between day t−1 and t+s (log(SPOTc,t+s/SPOTc,t−1)). Column (1) uses ΔSPOTc,t−1→t, Column (2) uses ΔSPOTc,t−1→t+1, 
Column (3) uses ΔSPOTc,t−1→t+2, Column (4) uses ΔSPOTc,t−1→t+3, and Column (5) uses ΔSPOTc,t−1→t+4. In Panel A, 
main independent variable is ΔSPOTc,t−1, which is the log ratio of spot exchange rate between day t − 2 and t − 1. 
We also include country-year-month fixed effects. In Panel B, main independent variable is D {ΔSPOTc,t−1 > 0}, 
which is a dummy variable that equals to 1 if ΔSPOTc,t−1 > 0 and 0 otherwise. All the standard errors are clustered 
at the country-year-month level. ***, **, * denotes 1%, 5%, and 10% statistical significance. 


Panel A: 


Short-term Behavior of Foreign Exchange Rate 


(1) 


 
(2) 
(3) 
(4) 


ΔSPOTc,t−1→t+s 


(5) 


Variable 


s = 0 


1 


2 


3 


4 


ΔSPOTc,t−1 


-0.117*** 


-0.086*** 


-0.140*** 


-0.171*** 


-0.215*** 


 


(-5.82) 


(-4.16) 


(-5.84) 


(-6.93) 


(-8.93) 


Observations 


6,657 


6,657 


6,657 


6,657 


6,657 


Adjusted R2 


0.017 


0.059 


0.100 


0.138 


0.173 


Country-Year-Month FE 


Yes 


Yes 


Yes 


Yes 


Yes 




 


Panel B: Short-term Behavior of Foreign Exchange Rate When It Appreciates 


 
(1) 
(2) 
(3) 
(4) 
(5) 


ΔSPOTc,t−1→t+s 


Variable 


s = 0 


1 


2 


3 


4 


D {ΔSPOTc,t−1 > 0} 


-0.070*** 


-0.086*** 


-0.119*** 


-0.152*** 


-0.168*** 


 


(-6.09) 


(-5.65) 


(-5.97) 


(-6.42) 


(-6.39) 


Observations 


6,657 


6,657 


6,657 


6,657 


6,657 


Adjusted R2 


0.008 


0.059 


0.098 


0.136 


0.169 


Country-Year-Month FE 


Yes 


Yes 


Yes 


Yes 


Yes 




 


 



Table 11: An Example of Social Network 


We report a real example of social network among Indonesian users with the list of users and the clustering 
behavior in users’ remittance within the network. In Panel A, we report the list of users in the network. We label 
the users by the order of users’ registration date. We report the area of residence and occupation type for the 
users. Panel B reports an example of the remittance transactions by the users in the network on February 25th, 
2020. 


 


 


 



Table 12: Descriptive Statistics of Social Networks 


We report the descriptive statistics of social networks and the characteristics of social networks. Panel A reports 
the summary statistics of the number of users in the social networks. We limit our definition of social network 
for those networks with more than 4 users in it. Panel B reports the summary statistics of individual-level 
variables that are related to the social networks. I SocialNetwork is a dummy variable that equals to 1 if the 
individual is in any of the social networks and 0 otherwise. Number of Remittances is the total number of 
remittances by individuals in our sample. Time to 1st Cancellation is the relative time of the first usage of 
Cancellation to the total number of remittances by individuals. % of Users in Same Area is the fraction of users in 
the social network who has same area of residence. For those without social network, we use the fraction of the 
users in the same area of residence with same nationality. Credit Balance is the average credit amounts of users. 
Centralityd and Centralitye are the measures of the centrality of an individual in the network from Banerjee et al. 
(2013). These measures are only defined for users in any social network. Centralityd measures the degree 
centrality of individuals in social networks and Centralitye measures the eigenvector centrality of individuals in 
social networks. Panel C reports cross-sectional regression results of individual characteristics with social 
networks. The dependent variables are Number of Remittances in Column (1), Time to 1st Cancellation in Column 
(2), % of Users in Same Area in Column (3), and Credit Balance in Column (4). Panel D reports the Pearson 
correlation between Credit Balance, Centralityd and Centralitye. ***, **, * denotes 1%, 5%, and 10% statistical 
significance. 


 



Table 13: Learning About the Usage of Cancellation 


We report the panel regression results of the usage of Cancellation in a social network on the individual usage of 
the Cancellation. The dependent variable is Cancellationi,t, which is a dummy variable that equals to 1 if an 
individual i use the Cancellation for the remittance on day t and 0 otherwise. Column (1) reports the result using 
I SocialNetworki,t as main independent variable. We control D {ΔSPOTc,t−1 > 0}, log(SendAmounti,t), Salary Dayst, 
Age, and Male with country-year-month fixed effects. Column (2) reports the result using CancellationHistory INi,t 
as main independent variable. CancellationHistory INi,t is the ratio of the cumulative number of remittances 
involving cancellations before day t to the total number of remittances before day. We also control I SocialNetworki,t. Column (3) re- 


ports the result using CancellationHistory SNi,t as main independent variable. CancellationHistory SNi,t is the ratio 
of the cumulative number of remittances involving cancellations before day t in a social network where the 
individual i is in to the total number of remittances before day t in the social network excluding the individual i’s 
own cancellations and remittances, . Column (4) 
reports the result using both CancellationHistory INi,t and CancellationHistory SNi,t. All standard errors in panel 
regressions are clustered at individual and country-year-month level. ***, **, * denotes 1%, 5%, and 10% 
statistical significance. 


 


 


 


 



Table 14: Learning About the Short-Term Behavior of Spot Exchange Rate 


We report the panel regression results of the history of transactions with the appreciation in the spot exchange 
rate on individuals’ remittance decision with the appreciation in the spot exchange rate. The dependent variable 
is D {Remittancei,t}, which is a dummy variable that equals to 1 if an individual i transacts on day t and 0 otherwise. 
Column (1) reports the result using SpotHistory INi,t as the main independent variable. SpotHistory INi,t is the ratio 
of the cumulative number of remittances with the appreciation of spot exchange rate on the previous day to the 
total number of remittances before day . We 
control D {ΔSPOTc,t−1 > 0}, Cancellationi,t, Salary Dayst, Age, and Male with country-year-month fixed effects. 
Column (2) reports the result using the interaction term of D {ΔSPOTc,t−1 > 0} and SpotHistory INi,t. Column (3) 
reports the result using SpotHistory SNi,t as the main independent variable. SpotHistory SNi,t is the ratio of the 
cumulative number of remittance with the appreciation of spot exchange rate on the previous days to the total 
number of remittances before day t in a social network s where the individual i is in excluding the individual i’s 
own remittances, . Column (4) reports 
the result using the interaction term of D {ΔSPOTc,t−1 > 0} and SpotHistory SNi,t. All standard errors in panel 
regressions are clustered at individual and country-year-month level. ***, **, * denotes 1%, 5%, and 10% 
statistical significance. 


 


 



Figure 1: Overseas Remittance toward Low- and Middle-Income Countries 


We plot the overseas remittance toward low- and middle-income countries. The sample period is from 1990 to 
2018. Panel A plots the aggregated amounts of remittance flows to the full sample of low- and middle-income 
countries and Panel B plots the aggregated amounts excluding China. We also plot the amounts of Foreign Direct 
Investment (FDI) for comparison. The data is from World Bank (2019). 


 


 



Figure 2: Cost of Overseas Remittance 


We plot the cost of overseas remittance using the quarterly data from 2011 Q1 to 2020 Q1. Panel A shows the 
time-series plot of the global average cost of sending $200. Panel B plots the average costs of remitting $200 by 
type of provider. The left Panel reports the average cost of worldwide full sample including 48 countries and the 
right panel reports the average cost of 9 countries with the remittance service provided by our Fintech platform, 
Sentbe, such as Bangladesh, Cambodia, India, Indonesia, Malaysia, Pakistan, Philippine, Thailand, and Vietnam. 
The data is from World Bank (2019). 


 


 


 


 



Figure 3: Importance of the Overseas Remittance in east Asian Region 


We plot the total remittance amounts in 2018 and the percentage of remittance amounts in the countries’ GDP in 
2018 for the countries with the Fintech remittance service such as Bangladesh, Cambodia, India, Indonesia, 
Malaysia, Pakistan, Philippine, Thailand, and Vietnam. The left panel plots the total amounts of remittance in 
2018, and the right panel plots the percentage of remittance amounts in the countries’ GDP. 


 


 



Figure 4: Time-Series of Individual Remittance Transactions 


We plot the time-series of the number of individual remittances. Panel A plots the daily number of remittances in 
our sample period. The circle markers indicate the days with monthly peak of remittance transactions. The 
Fintech service was not available 2 times in our sample period. The first is from July 17th to December 7th of 
2017, and the second is from February 15th to 18th of 2018. Panel B plots the number of remittances in each day 
of a month. Panel C plots the number of remittances in each day of a week. Panel D plots the number of 
remittances in each 10 minutes in a day. The marker indicates 12:30PM when the maximum number of 
remittances in a day occurs. 


 


 


 



Figure 4 Continues 


 





Figure 5: Spot Exchange Rates 


We plot the spot exchange rates of the currencies for 9 countries in our sample. The sample period is from February 2016 to March 2020. Countries may 
have different starting dates due to the different starts of the Fintech service for those countries. 


 


 



Figure 6: Comparison of Remittance Cost between Fintech and Commercial Banks 


We compare the remittance cost between Fintech and commercial banks. Panel A uses all the remittance 
transactions for 9 countries in our sample. We report the sending amounts in x-axis and the cost of remittance 
for the amounts in y-axis. Solid line reports the remittance cost associated with the sending amounts using 
Fintech Platform and dotted line reports the remittance cost associated with the sending amounts using 
commercial banks. The difference between two lines is the difference in remittance cost for the amounts. We also 
report the remittance transactions in our data by the bins of 10,000 KRW. We compute and report the 
transaction-weighted difference between two lines for the average benefit from cost reduction using Fintech. 
Panel B plots the similar results by country. 


Panel A: Difference in Remittance Costs between Fintech and Commercial Banks 


 


 


 


 



Figure 6 Continues 


Panel B: Difference in Remittance Costs between Fintech and Commercial Banks by Country 


 


 



Figure 7: Financial Development Index and the usage of Cancellation 


We plot the financial development index and the probability of using the Cancellation feature in the Fintech 
platform. The x-axis is the Financial Development Index in 2018 from IMF. The y-axis is the average probability 
of using Cancellation by the individuals from the country. We include the linear-fitted line. 


 


 



Figure 8: Optimality in Remittance Timing 


We plot the average optimality of remittance transactions. For each remittance transaction, we compute 
hypothetical remittance amounts in receiving currency associated with different exchange rates before or after 
the actual remittance. We normalize these amounts from -5 to +5 days with the original amounts of remittance 
in receiving currency on day 0. Panel A reports the average relative spot exchange rate using full sample. We 
report the 95% confidence interval around the line. Instead of using full sample, Panel B only use the sample 
individuals in top 1/3 among all sample individuals in terms of the optimality. The box plots report 10th, 25th, 
50th, 75th, and 90th of the distribution. Panel C plots the average optimality of remittance transactions with the 
usage of Cancellation. The solid line reports the relative spot exchange rate of users with Cancellation and the 
dashed line reports the relative spot exchange rate of users without Cancellation. We report 95% confidence 
intervals associated with the two lines. All the standard errors are clustered at the individual, and currency-year-
month level. 


Panel A: Average Optimality Score in Full Sample 


 


 



Figure 8 Continues 


Panel B: Average Optimality Score of Top 1/3 


 


Panel C: Average Optimality Score with the Usage of Cancellation. 


 



Figure 9: Measure of Optimal Remittance Timing 


We report some examples of calculating optimality score using real examples in the data. In Panel A, we plot the 
spot exchange rate in the window of [-5, +5] relative to the spot exchange rate of day 0 using an example of a 
Vietnamese user on August 28th, 2018. The difference of average relative rate from 1, which is 0.0083 in this case, 
is defined as the optimality measure. Panel B reports other examples of optimality score. While top-left figure 
reports the same example as in Panel A, we report other examples with various optimalities in remittance timings. 


Panel A: Example of Optimal Remittance Timing 


 


 


 





Figure 9 Continues 


Panel B: Other Examples 


 


 


 


 



Figure 10: An Example of Social Network 


We report an example of a social network among Indonesian users. We label the users by the order of their 
registration dates to the Fintech platform. In this example, user A recommends the platform to B, C, D, and H. The 
user C recommends it to E, J, and K and the user E recommends it to F and I. And the user F recommend it to G. 


 



Figure 11: Clustered Remittance Transactions within a Social Network 


We plot an example of clustered remittance transactions within the social network in Figure 10 in February 2020. 
We plot the daily aggregated number of remittance transaction by the users in the social network (blue dash bar) 
and their matching sample without social network (orange solid bar). We also plot the spot exchange rate of 
Indonesian Rupiah in the unit of 1 KRW. 


 


60