Remittances, transaction costs, and informality

8/11/2019 Remittances, transaction costs, and informality

http://slidepdf.com/reader/full/remittances-transaction-costs-and-informality 1/11

Remittances, transaction costs, and informality☆

Caroline Freund a ,⁎, Nikola Spatafora b

a World Bank, United States b IMF, United States

Received 14 October 2005; received in revised form 11 September 2007; accepted 12 September 2007

Abstract

Recorded workers' remittances to developing countries reached $167 billion in 2005, bringing increasing attention to these flows

as a potential tool for development. In this paper, we explore the determinants of remittances and their associated transaction costs. We

find that recorded remittances depend positively on the stock of migrants and negatively on transfer costs and exchange rate

restrictions. In turn, transfer costs are lower when financial systems are more developed and exchange rates less volatile. The negative

impact of transactions costs on remittances suggests that migrants either refrain from sending money home or else remit through

informal channels when costs are high. We provide evidence from household surveys supportive of a sizeable informal sector.

© 2007 Published by Elsevier B.V.

JEL classification: F21; F22; F30

Keywords: Remittances; Migration; Money transfer fees

1. Introduction

Recorded flows of workers' remittances to develop-

ing countries have grown from about $70 billion in 2000

to more than $150 billion in 2005 (World Bank, 2006).

Such high levels and rapid growth rates have brought

increasing attention to remittances as a potential tool for

development. Understanding the determinants of remit-

tances is important because these flows reduce poverty,

allow recipients to smooth consumption, and are also

used for investment purposes (World Bank, 2006,

Chapter 4). In times of severe economic distress, remit-

tances may be the primary source of income for many

families, as in Zimbabwe today (Wines, 2007).

The main motivation of this study is to explore the

determinants of remittances. We find that the number of

migrants is the primary determinant of official remit-

tances: an increase in the stock of migrants residingin OECD countries leads to a roughly proportionate

increase in recorded remittances. We also find that high

transaction costs significantly reduce recorded remit-

tances: a one percentage point reduction in transaction

costs raises recorded remittances by 14–23%.

Our analysis offers three contributions. First, in ex-

ploring the determinants of remittances, we improve on

the previous literature by including migrant stocks and

transaction costs in the estimation. Previous work by

Aggarwal and Spatafora (2005) and Sayan (2006)

Journal of Development Economics 86 (2008) 356 – 366

www.elsevier.com/locate/econbase

☆ We are grateful to Irena Omelaniuk for providing us with IOM

studies of remittances in several countries and for helpful discussions,

to Dean Yang for generously providing data and summary statistics on

the mode of remittance transfer in the Philippines, and to Angela

Espiritu for outstanding research assistance. We are also grateful to

Gordon Hanson, Dilip Ratha, Dean Yang, participants at a World Bank

seminar, and two anonymous referees for comments.⁎ Corresponding author.

E-mail addresses: [email protected] (C. Freund),

[email protected] (N. Spatafora).

0304-3878/$ - see front matter © 2007 Published by Elsevier B.V.doi:10.1016/j.jdeveco.2007.09.002

mailto:[email protected]


http://dx.doi.org/10.1016/j.jdeveco.2007.09.002

http://dx.doi.org/10.1016/j.jdeveco.2007.09.002





examined the effects of income on remittances, but their

work focused on the cyclicality of remittances and did not

include migration nor transactions costs, two important

determinants of remittances. In contrast, Faini (2006)

focuses on differences in remitting patterns between

educated and uneducated migrants, and includes migrant stock but neither transactions costs nor other variables that

we find to be important. IMF (2005) and World Bank

(2006) offer a survey of the literature.

Second, we collect new data on the cost of sending

money home and explore how transaction costs vary

with source and home country features. We find that

transaction costs are systematically related to lack of

financial depth and exchange rate volatility. Policies that

encourage financial development in the migrant's home

country and that reduce exchange rate volatility will

help lower official transactions costs.Third, we provide new evidence about informal re-

mittance flows. The analysis indicates that transactions

costs have a large and statistically significant effect

on recorded remittance receipts. This is consistent with

migrants either refraining from remitting money, or else

remitting large amounts through lower cost informal

channels when official transaction costs are high. Evi-

dence from household surveys and from changes in

remittances over time supports the informal-channel

mechanism. Household surveys find that transaction

costs largely affect the way that remittances are sent and

not the amount that is sent. These surveys also provideevidence of large informal remittance flows in many

countries. In addition, there exists a negative correlation

between annual changes in recorded remittances and

changes in Net Errors and Omissions (NEO) from the

balance of payments. This suggests that increases in

recorded remittances are partially offset by declines in

informal remittances. Finally, market observers suspect

that, globally, informal flows range from 50% to 250%

of recorded flows.1

The paper proceeds as follows. The next section

discusses our data, including in particular a new dataset on the transaction costs associated with sending remit-

tances; we also focus on the problems associated with

measuring remittance flows. Section 3 presents results

on the determinants of both remittances and the asso-

ciated transaction costs, using both cross-sectional and

panel techniques. Section 4 develops some insights into

informal remittance flows, both by examining survey

data on remittances and how they are transmitted, and

by analyzing the relationship between NEO and re-

corded remittances. Section 5 concludes.

2. The data

2.1. Remittance inflows

We collected a panel of aggregate data on remittances

from the IMF's Balance of Payments statistics. The

dataset covers up to 104 countries for which workers'

remittances are reported during 1995–2003. On aver-

age, we have five observations per country. As is stan-

dard practice, we define remittances as the sum of three

items in the Balance of Payments statistics: “Compen-

sation of Employees,” “Workers' Remittances,” and

“

Migrants' Transfers.”

Adjustments are however madefor a number of countries, following the advice of IMF

country desks and national authorities.2

Remittances can in general be sent through either

formal or informal channels. Throughout the paper, we

define informal remittances as money transfer services

that do not involve formal contracts and hence are

unlikely to be recorded in national accounts. Formal

channels include money transfer services offered by

banks, post office banks, non-bank financial institutions,

foreign exchange bureaus, and money transfer oper-

ators (MTOs) like Western Union and MoneyGram.

Informal channels include cash transfers based on per-sonal relationships through business people, or carried

out by unofficial courier companies, friends, relatives or

oneself.3

In addition to not capturing the informal sector,

remittance data in the balance of payments do not

always include transactions through MTOs. One study

of 40 central banks in developing countries around the

world indicates that about 60% do not record data from

small MTOs that do not settle through banks (De Luna

Martinez, 2005). Because of this under-recording, a

portion of formal remittance transactions will end up inthe statistical discrepancy in the balance of payments.

1 Celent (2002) estimated that informal remittances would account

for 35% of total remittances (or 54% of recorded remittances) in 2006.

AITE (2005) reports that total remittances currently exceed IMFestimates by 250%.

2 See the Appendix for a fuller discussion of the definition of

remittances and of the various adjustments made to the data. The IMF

Balance of Payments Manual (IMF, 1993) sets out formal definitions

and classifications, while the IMF Balance of Payments Statistical

Yearbook , Part 3, illustrates the diversity of approaches, sources, and

methods used by different countries. In general, data weaknesses are

due to difficulty in obtaining the precise data. See IMF (2005), Box

2.4, for a detailed discussion of the measurement practices of national

statistical agencies and potential improvements.3

Pieke et. al.(2005) provide a survey of informal market definitions,costs, and systems.

357C. Freund, N. Spatafora / Journal of Development Economics 86 (2008) 356 – 366



Transactions by MTOs that are settled through banks

and are properly identified as remittances will still be

recorded.

In Fig. 1, we present broad trends in remittance

receipts for several different regions. One notable

feature is the rapid increase over the last decade,

particularly for Latin America. This increase likely

reflects rising numbers of migrant workers around the

world. In addition, it may reflect technological devel-

opments and increased competition in the financial-

services industry reducing the cost of sending remit-tances, especially through the formal financial sector.

This could have both encouraged remittances in general,

and led to a shift in transactions from the informal into

the formal sector. Indeed, data from household surveys

(see Section 4) imply that the informal sector is now

quite small and has been declining in several Latin

American countries.4

2.2. Remittance transaction costs

We collected data from Western Union U.S., WesternUnion U.K., and MoneyGram U.S. on the service fees

associated with sending remittances to specific countries

in 2005–06. These transaction costs were computed

assuming a remittance size of $200, and are quoted as a

percentage of this amount.5 Regarding the Western

Union data, we use U.S.-based costs for countries with

large stocks of migrant workers in the United States,

U.K.-based costs for countries with large migrant stocks

in the United Kingdom, and an average of the two costs

in all other cases.6

We then average the Western Unionfees and the Money Gram fees. Results are broadly

similar if we use Western Union data alone.

The cost of remitting small amounts through formal

channels can at times be prohibitively high, owing to

the minimum fee charged by most service providers.

Indeed, in our data (Table 1), the average cost of sending

money through an MTO is around 11%, and the cost of

sending money to Africa is around 13%. Minimum fees

at banks range from $5 to $50 depending on the sending

and receiving countries as well as the product (Sander

and Maimbo, 2003). Fees are usually structured by brackets of transfer values. The average cost declines

sharply as the amount remitted increases, reflecting the

minimum fees.

These data are based on the cost of sending remit-

tances through formal channels. Globally, studies

indicate that informal remittance channels are cheaper

than formal channels, especially banks and MTOs like

Western Union and MoneyGram. For informal remit-

tance channels as a whole, Sander (2003) reports the

average cost of remitting at 3–5% globally, although it

can be higher in specific cases. Likewise, Swanson and

Kubas (2005) report costs of 1–5%. Orozco (2003)estimates the cost of a Hawala/Hundi transaction to be

less than 2% of the value of the principal. Similarly,

remittances through friends, taxi drivers, etc., are also

low-cost compared to formal channels.7

Fig. 1. Remittances received by region1 .

5 We also tried computing the implicit exchange rate spread

(defined as the difference between the exchange rate offered by the

remittance service provider, and the central exchange rate as quoted

by Bloomberg). However, this variable never proved significant,

likely because exchange rates and hence the associated spreads may

fluctuate by large amounts. As a result, values at the end of the period

(when we collected the data) may be a poor guide to period-average

values. Orozco (2003) finds that fixed fees make up the bulk (over

80%) of transactions costs on average.6 The results are robust to using only cost data from the United

States. There is no full sample for the United Kingdom data, as costs

from the U.K. to some Latin American countries were missing.7 For example, in a survey conducted in South Africa, remittances

up to R250 to neighboring countries cost R25 and R50 through,

respectively, friends and taxi drivers, as compared with over R100

through registered banks and over R80 through money transfer agents

like MoneyGram and Western Union (Genesis, 2003). Similarly,

Siddiqui and Abrar (2003) find that costs of informal channels inBangladesh are about 45% of formal costs.

4 In a related vein, the increase in official remittances may also

reflect improved recording of formal flows through MTOs. For

instance, Mexican remittances increased by 95% during 2000–2004.

According to the central bank, a key factor underlying this was that

only after 2002 did Mexico begin recording transactions from MTOsthat do not settle through banks (De Luna Martinez, 2005).

358 C. Freund, N. Spatafora / Journal of Development Economics 86 (2008) 356 – 366



2.3. Other variables

We also gathered data on the stock of migrant workers

around the world. Where such data are available, they

generally refer to the total number of migrant workersresident in a specific country. Given our interest in

remittance receipts, however, we need data on the total

number of migrant workers originating from a specific

country, regardless of which country they migrate to.

Two such datasets are available. The first is the OECD's

Database on Immigration and Expatriates, which

covers the year 2000. The second dataset, compiled by

Docquier and Marfouk (2005), covers both 1990 and

2000, and also has information on the educational attain-

ment of migrants. Both datasets only cover migrants

residing in the OECD.8 This restriction is unlikely to bias

any estimated coefficients below, provided that, for anygiven country, the stock of migrants residing in the

OECD is closely correlated with (and can therefore be

used to proxy for) the total stock of migrants. We return

to this issue in the estimation.

Other variables used in the estimation include: per

capita domestic output; per capita output in the country

where most of a country's migrant workers are located

(“main host output ”); an indicator denoting the presence

of a dual exchange rate system (“dual exchange rate

dummy”); financial development; the three-firm concen-

tration ratio for the banking sector (“ bank concentration”);and a dollarization dummy. The Appendix provides

further details on precise definitions and sources.

Table 1 reports summary statistics for the variables in

the averaged data. Remittances and the stock of mi-

grants are largest in Latin America and Asia. Countries

in Asia enjoy the lowest transaction costs. In contrast,

Sub-Saharan Africa records the highest transaction

costs.

3. Estimation

This section investigates the determinants of remit-

tance inflows in middle-and low-income countries, as

the determinants of remittances to industrial countriesare likely to be very different. The basic methodology

involves cross-country regressions of remittances on

potential explanatory variables, including remittance

transaction costs. In addition, we analyze the factors

driving these transaction costs through a separate set of

cross-country regressions.

Data on transaction costs are only available for a cross-

section. Hence, the above regressions, which directly

involve such costs, are also purely cross-sectional. That

said, the underlying determinants of transaction costs are

available for a panel starting in 1995. Hence, we check

the robustness of our results through panel regressions of remittances on various explanatory variables, including

the underlying determinants of transaction costs (rather

than transaction costs themselves).

3.1. An analysis of money transfer costs

The data on the transaction costs associated with

receiving remittances are novel. In addition, under-

standing the factors driving these transaction costs is

important to interpreting our results. We therefore start

by investigating the determinants of remittance transac-tion costs. We expect that overall financial-sector devel-

opment might lead to greater availability and lower costs

for remittance services. Lack of exchange rate risk, as

would be the case for dollar remittances being sent to

a dollarized economy, should likewise reduce the cost

of providing remittance services. Again, greater com-

petition in the financial-services industry might have

a powerful negative impact on such costs. Conversely,

defective institutions and greater business risk, as

proxied by our measure of financial risk, would instead

be expected to reduce the willingness of agents to pro-

vide remittance services.

Table 1

Summary statistics of averaged data, 1995–2003

Region Total remittances

(million $U.S.)

Total migrant stock

in 2000 (d000s)

Average fee

in 2005/6

Average dual exchange

rate system

NOB

Sub-Saharan Africa 1948 852 13.02 0.21 24

Eastern Europe and Central Asia 8437 4207 11.66 0.15 20East Asia and the Pacific 11,071 2776 8.40 0.20 10

Middle East and North Africa 8467 1209 11.67 0.25 4

Latin America and the Caribbean 17,064 11,800 10.07 0.25 28

South Asia 13,593 2380 7.58 0.17 6

Total 60,581 23,200 10.91 0.21 92

8

According to World Bank (2006), Chapter IV, migrants residing inthe OECD account for about half of the world migration stock.




We therefore run the cross-sectional regression

Cost i ¼ a þ b1FinDevi þ b2Dollar i þ b3Risk iþ b4Conci þ ei; ð1Þ

where Cost denotes the (percentage) transaction costsassociated with receiving remittances; FinDev denotes

financial development, as measured by the ratio of do-

mestic deposits to GDP; Dollar denotes a dollarization

dummy; Risk denotes financial risk; and Conc denotes

bank concentration. All variables are constructed as an

average over 1995–2003, except for transaction costs

where we use values for 2005 (the only available year).

The results are presented in Table 2, column 1. The

coefficients on financial development, dollarization, fi-

nancial risk, and bank concentration all have the ex-

pected sign; the first two are statistically significant at

the 5% or lower level. Overall, the results suggest that a

wide range of policies (including measures to increase

competition among remittance-service providers, to in-

crease financial development, and to reduce exchange

rate volatility) would be expected to reduce the trans-

action costs associated with remittances, and hence to

increase recorded remittances.

Domestic wages (as proxied by domestic income per

capita), if not matched by equivalent service-sector pro-

ductivity, might be associated with greater cost of remit-

tance services. However, any such effects are statistically

insignificant (Table 2, column 2).Market-size effects might also be important. In partic-

ular, greater remittances might reduce fees, because of

more competition in large markets or returns to scale. To

control for such effects, we include in the regression

equation total remittances (columns 3 and 4) as well as the

migrant stock (columns 5 and 6). While both have the

expected negative sign, they are only significant in one

regression out of four, while financial development and

dollarization remain significant throughout. In addition,

when remittances or migrant stock are included, the

coefficient on bank concentration drops in magnitude.

This is consistent with the view that the impact of large

markets arises largely through enhanced competition;

given measurement error, it would then be hard to

estimate separately the effect of market size and of competition. Indeed, when bank concentration is not

included in the regression, the coefficient on the migrant

stock becomes statistically significant (column 6). These

results highlight the importance of measures to improve

competition, especially in small markets, where market

size alone will not boost competition.

3.2. Determinants of remittances: Cross-sectional

estimates

Having developed some understanding of remittancetransaction costs, we now analyze a broader issue: the

determinants of remittance receipts. Aggregate remit-

tances are likely to depend on the total number of

migrants, wages in the host economy, and income in the

source economy. Transaction costs and a dual exchange

rate will also lower remittances if they lead to less money

being remitted or to greater use of informal (and un-

recorded) channels. As discussed, transaction costs are

only available for a cross-section. Our basic estimating

equation is therefore the cross-sectional regression

Ri ¼ a þ b1 yi þ b2 M i þ b3 yi⁎ þ b4DualER iþ b5Feei þ ei; ð2Þ

where R denotes the log of workers' remittances reported

in the BOP statistics; y denotes the log of domestic

income per capita; M denotes the log of the stock of

migrant workers; y⁎ denotes the log of main host income

per capita; DualER is an indicator of the presence of a dual

exchange rate; Fee denotes the service fee associated with

receiving remittances; and i indexes the relevant country.

Table 2Cross-sectional regression results: determinants of transaction costs

Explanatory variables (1) (2) (3) (4) (5) (6)

Financial development −0.023⁎⁎ (2.20) −0.023⁎⁎ (2.23) −0.029⁎⁎ (2.64) −0.031⁎⁎⁎ (3.09) −0.024⁎⁎ (2.26) −0.026⁎⁎⁎ (2.83)

Dollarization −2.329⁎⁎⁎ (5.01) −2.349⁎⁎⁎ (5.05) −2.345⁎⁎⁎ (4.82) −2.391⁎⁎⁎ (5.22) −2.182⁎⁎⁎ (4.63) −2.278⁎⁎⁎ (5.17)

Bank concentration 0.016 (1.45) 0.019 (1.63) 0.003 (0.22) 0.007 (0.55)

Financial risk 0.015 (0.32) 0.045 (0.79) 0.01 (0.19) 0.019 (0.40)

Domestic income per capita 0.231 (0.99)

Remittances −0.161 (1.42) −0.171 (1.63)

Stock of migrants −0.206 (1.48) −0.244⁎⁎ (2.03)

NOB 66 66 60 60 66 66

R-squared 0.55 0.56 0.59 0.59 0.57 0.56

Domestic income per capita, remittances, and the stock of migrants are in logs. ⁎⁎

and ⁎⁎⁎

denote significance at, respectively, the 5% and 1% level.Absolute values of robust t-statistics are in parentheses.




All variables are constructed as an average over 1995–

2003, except for the service fee where we again use values

for 2005.

The results are presented in Table 3, column 1. The

stock of migrant workers residing in the OECD coun-

tries is the main driver of remittances, with an elasticity

close to unity. This suggests that either migration to

the OECD is for most countries a good proxy for over-all migration, or else recorded remittances stem almost

entirely from the OECD.

The service fee has a negative and significant impact

on recorded remittances, consistent with the notion that

higher fees discourage remitters or push them into the

informal sector. The coefficient is economically signif-

icant: a one percentage point increase in fees reduces

remittances by 16%. Put differently, assume that the

average fee to transfer money to developing countries

were reduced from 11% of the transaction amount, the

value observed in our dataset, to 5%, which is close tothe upper estimates of informal costs in Orozco (2003)

and Sander (2003) and to the lowest country-specific

estimates of formal costs in our dataset. Then recorded

remittances would on average almost double.9 At a

regional level, the impact of such a reduction in fees

to 5% would prove especially large for Sub-Saharan

Africa, where costs are currently highest. The next

section discusses in detail how to interpret these results.

The above analysis may be subject to simultaneity

bias if higher remittances lead to lower fees, through

either greater competition or increasing returns to scale

in the provision of money transfer services. To minimize

such concerns, we also use remittances per migrant

and remittances per capita as the dependent variables

(columns 2 and 3), since these can be high even when

overall remittances and migration are low. The servicefee remains significant and the magnitude of the coef-

ficient is roughly unchanged, suggesting that the impact

does not reflect large remittance flows driving down

fees. Finally, since using remittances per migrant and per

capita may not completely eliminate the simultaneity

problem, we also instrument for transactions costs using

beginning-of-period financial development and a dol-

larization dummy, two key determinants of transactions

costs from Table 2. The results are reported in column 4.

The coefficient on the service fee remains negative and

significant.Our measure of economic restrictions, the dual ex-

change rate indicator, also always has a negative impact

on remittances. The effect is not statistically significant,

but it is in the panel regressions discussed later. Overall,

the results suggest that policies aimed at reducing costs

in the remittance marketplace would be associated with

increases in recorded remittances.

Surprisingly, main host-country income per capita has

an insignificant effect on remittances. This likely reflects

the fact that the variable is a very imprecise proxy for

average wages in the full set of countries in which

migrants are actually employed. Domestic income per

Table 3

Cross-sectional regression results: determinants of remittances

Explanatory variables Dependent variable Dependent variable Dependent variable Dependent variable Dependent variable

Remittances Remittances per migrant Remittances per capita Remittances, IVa Remittances, IV b

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

Dual exchange rate −

0.255 (0.65) −

0.292 (0.75) −

0.376 (1.03) −

0.169 (0.39) −

0.122 (0.27)Service fee −0.159⁎⁎ (2.60) −0.143⁎⁎ (2.37) −0.137⁎⁎ (2.30) −0.231⁎ (1.83) −0.228⁎ (1.88)

Stock of migrant workers 0.950⁎⁎⁎ (10.72) 0.881⁎⁎⁎ (8.89) 0.814⁎⁎⁎ (4.12)

Migrant share of source

population

0.701⁎⁎⁎ (8.34)

Main host income per capita −0.192 (1.23) −0.214 (1.50) −0.109 (0.79) −0.153 (0.82) −0.143 (0.75)

Domestic income per capita −1.101 (0.58) −0.104 (0.60) 0.168 (0.91) −0.151 (0.78) −0.127 (0.17)

F-stat for fee 18.2 [0.00] 10.7 [0.00]

F-stat for migrant 5.67 [0.00]

Overidentification test 2.27 [0.32]

NOB 92 92 92 77 77

R-squared 0.65 0.08 0.56 0.65 0.65

All dependent variables, and all independent variables except for dual exchange rate and the service fee, are in logs. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance

at, respectively, the 10%, 5%, and 1% level. Absolute values of robust t-statistics are in parentheses, P -values are in brackets.a Instrument for the service fee using dollar dummy and lagged financial depth. b Instrument for log (stock of migrant workers) using the index of geographical remoteness, and regional dummies. Instrument for the service fee

using dollar dummy and lagged financial depth.

9 Note that since the fee is lower, more money would be received

for any given amount sent. However, this effect can at most explain a7% increase.




capita also has an insignificant effect on remittances. Both

are however significant in the panel estimates.

A final potential source of simultaneity bias is that

remittances may affect migration through family re-

unification and the alleviation of liquidity constraints.

To deal with this potential problem, we instrument for

migration using domestic population, an index of geo-

graphical remoteness (from Djankov et al., 2005), and

regional dummies. The IV results, reported in Table 3,

column 5, are broadly similar to the OLS results.10

3.3. Determinants of remittances: Panel estimates

In order to interpret our findings, including those on

the impact of transaction costs, it is desirable to know

whether the results hold only across countries, or also

over time within a given country. Data on remittances are

available on a panel basis, but data on transaction costs

are only available for a cross-section; as a result, it is not

possible to run directly a panel regression of remittances

on transaction costs. Instead, we now exploit the resultsin Table 2, and allow all variables used there to explain

transaction costs to enter as regressors for remittances in

a panel context. Specifically, we run the panel regression

Ri;t ¼ ai þ b1DualER i;t þ b2FinDevi;t

þ b3Dollar i;t þ b4Risk i;t þ b5Conci;t

þ b6 M i þ b7 yi;t ⁎ þ b8 yi;t þ gt þ ei;t ; ð3Þ

where αi denotes the country-specific fixed effects, γt

the year-specific fixed effects, and all other variables are

as defined previously. The equation is estimated over

1995–2002.

The results are presented in Table 4. The dual ex-

change rate again has the expected sign, and is now

statistically significant at the 10% level. Financial

development (the main determinant of transaction

costs) is significant at the 5% level. As expected, the

stock of migrants has a highly significant effect onremittances, and the coefficient is similar in magnitude to

the one found in the cross-sectional regressions. In

addition, within this sample, remittances are significant-

ly pro-cyclical, as shown by the positive coefficient on

domestic income per capita.11 The elasticity of remit-

tances with respect to main host income always equals or

exceeds unity (and is significant in the regressions

without instruments), that is, increases in host-country

income and hence migrant wages are passed on at least

proportionately to families abroad. All of these results

also hold using remittances per capita as the dependent variable.

One issue is that to the extent that remittances are

channeled through banks, or make their way into bank

accounts, they will directly affect financial develop-

ment. To control for this potential endogeneity, we

instrument for financial development using its twice-

10 We also try using the share of migrants with at least secondary

(or tertiary) education as a determinant of remittances. The estimated

coefficients change sign depending on the specification and are never

significant. In contrast, Faini (2006) finds some evidence that skilledmigrants may remit relatively less.

Table 4

Panel regression results: determinants of remittances

Remittances Remittances per capita Remittances, IVa Remittances per capita, IVa

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

Dual exchange rates −0.179⁎ (1.83) −0.188⁎ (1.92) −0.166⁎ (1.74) −0.175⁎ (1.83)

Financial development 0.012⁎⁎

(2.40) 0.012⁎⁎

(2.38) 0.013⁎⁎

(2.09) 0.013⁎⁎

(2.12)Dollarization 0.178 (0.74) 0.16 (0.67) 0.163 (0.69) 0.147 (0.63)

Bank concentration −0.002 (0.80) −0.002 (−0.75) −0.002 (0.82) −0.002 (0.74)

Financial risk 0.008 (1.11) 0.008 (1.18) 0.007 (0.97) 0.006 (0.92)

Stock of migrants 0.930⁎⁎⁎ (2.92) 1.290⁎⁎⁎ (3.73)

Stock of migrants, as share of population 1.139⁎⁎⁎ (4.06) 1.491⁎⁎⁎

Main host income per capita 1.497⁎⁎ (2.09) 1.424⁎⁎ (2.00) 1.15 (1.60) 1.085 (1.53)

Domestic income per capita 0.796⁎⁎⁎ (5.31) 0.790⁎⁎⁎ (5.25) 0.748⁎⁎⁎ (5.07) 0.722⁎⁎⁎ (4.89)

NOB 523 523 508 508

R-squared 0.28 0.25 0.27 0.25

All dependent variables, and all independent variables except for the dual exchange rate and service fee, are in logs. All regressions include country

fixed effects and a time trend. ⁎, ⁎⁎, and ⁎⁎⁎ denote significance at, respectively, the 10%, 5% and 1% level. Absolute values of robust t-statistics are

in parentheses.a Instrument for financialdevelopmentusing itssecond lag. The F -statistic of the instrument fromthe first stage is just over 3.00 for both regressions.

11 This is consistent with Sayan (2006) who finds that remittances

are generally procyclical in developing countries. In contrast, Yang

(2006) finds that remittances increase following natural disasters,suggesting remittances can help offset bad times.




lagged value; the results continue to hold (columns 3

and 4).12

4. Why are transaction costs so important?

As discussed, empirically observed variations in

service fees have a large impact on recorded remittances.

This result is consistent with either of two hypotheses.

First, migrants refrain from remitting money when trans-

action costs are high. Second, large official transaction

costs encourage migrants to send remittances through

informal channels where transaction costs are lower.

Existing empirical evidence from a number of country

studies suggests that the cost of sending remittances has

little effect on the total amount remitted (formal and

informal), implying that it is the channel of transmission

that is affected. For instance, Gibson et al. (2005) report

that 70% of Tongan migrants would not change theamount sent home if the transfer fee declined.13

Additional evidence that informal channels can be

important comes from household surveys that ask res-

pondents both how much they have received from friends/

relatives abroad, and by what means, including informal

channels. Table 5 reports estimates of the informal sector

as a percentage of total remittances sent, based on surveys

from Armenia, Bangladesh, Dominican Republic, El

Salvador, Guatemala, Mali, Moldova, Philippines, Sene-

gal, and Uganda (see Freund and Spatafora, 2005, for

detailed information about the surveys). There is signif-icant variation in the channel of transmission. In Mali,

Senegal, and Uganda, countries facing extremely high

transaction costs, the bulk of remittances enter informally.

In addition, large informal flows occur in Armenia and

Moldova, where transaction costs are also relatively high.

In contrast, Latin America has a relatively small informal

sector. One explanation is that MTO transmission costs

to Latin America have come down sharply since 1995

(Orozco, 2003). As a result, there may have been a

significant shift from the informal to the formal sector.

Such a shift may account for much of the dramatic increase

since 1995 in recorded remittances to Latin America (seeFig. 1, as well as the discussion in footnote 5).

An alternative way of gauging to what extent the

increase in recorded remittances reflects a move out

of the informal sector is to examine the relationship

between Net Errors and Omissions (NEO), from the

balance of payments, and remittances or migration. If

increases in recorded remittances stem from a move

out of informal channels then NEO should decline as

recorded remittances rise. If instead total remittances

fluctuate for other reasons, such as increased migration,

then recorded and informal remittances are likely tomove in the same direction, resulting in a positive cor-

relation between NEO and recorded remittances. Over-

all, as long as increases in recorded remittances are

Table 5

Informal remittance inflows, selected countries (% of total inflows)

Country Survey data

Mali 70

Senegal 70

Uganda 80Bangladesh 54

Philippines 41

Dominican Republic 15

El Salvador 20

Guatemala 5

Armenia 38

Moldova 47

Sources: Armenia: Roberts, B., and K. Banaian (2004), “Remittances

in Armenia: Size, Impacts, and Measures to enhance their Contribution

to Development,” Special Study for USAID/Armenia, Bearing Point,

October. Based on number of transactions.

Bangladesh: Tasneem Siddiqui and C.R. Abrar (2003), “Migrant Worker

Remittances and Micro-Finance in Bangladesh”.

Dominican Republic: Calculated from “Encuesta de Fuerza de Trabajo,”

2000–2002. Informal sector includes personal and mail.

El Salvador: Calculated from “Encuesta de Hogares de Propositos

Multiples,” 1997. Informal includes family and friends, particular

individuals, and mail.

Guatemala: Survey on the Impact of Family Remittances on Guatemalan

Homes, conducted in 2004 by the Vice-Presidency of Guatemala,

The Central American Bank of Integration, The Bank of Guatemala

and IOM. 2,921 homes were surveyed across 22 “departments” and 170

municipalities.

Mali: ECFIN 2004. Includes only remittances from France.

Moldova: “Moldova Remittance Study,” by CBS AxA, a TNS CSOP

Branch in Moldova, involving inter alia a survey of 3,714 households in

October – November 2004. 1,299 households had at least one member earning a living abroad in 2003 and 2004. Only 65% of remittance

receivers answered question.

Philippines: Survey of Overseas Filipinos, 2000, National Statistics

Office of the Philippine Government. Informal includes others and

brought home by migrant.

Senegal: ECFIN 2004. Includes only remittances from France.

Uganda: Data from IOM, Kampala.

13 Similarly, Yang (2004) shows that, when the Philippine peso

depreciated during the Asian financial crisis, Philippine migrants sent

less money in foreign currencyand roughly the same in pesos,suggesting

that migrants have a target amount they want their family to receive. One

explanation is that many remitters have little option but to send money,

given the severe economichardships faced by their families in the source

country. More generally, existing studies typically find remittances to be

driven by the need to support migrant workers’ families, rather than by

investment considerations alone (e.g., Aggarwal and Spatafora, 2005).

12 Note that measures of financial development may proxy for not only

remittance transaction costs, but also the broader investment climate.

Hence, to the extent that remittances are used for investment purposes,these estimates may overstate the effect of lowering costs on remittances.




primarily associated with shifts out of informal flows,

NEO will be negatively correlated with remittances. In

addition, assuming that migration increases the remit-

tances that enter informally (as well as formally), NEO

will be positively correlated with migration.

Table 6 reports the results of regressing NEO on

remittances and on migration, using panel data from

1981–2003, and controlling for country and year fixed

effects.14 An increase in remittances is associated with a

decline in NEO (column 1), and an increase in the migrant

share of the population leads to an increase in NEO(column 2). Including both remittances and migration in

the regressions leaves the results roughly unchanged

(column 3). Columns 4 and 5 report the results using the

same sample as column 3, but including remittances and

migration separately; again, the results are similar.

Overall, these results support the notion that year-to-

year fluctuations in recorded remittances are associated

with a shift out of the informal or unrecorded sector to the

formal sector, while the increasing trend in migration

is associated with an increase in both recorded and

unrecorded flows.

5. Conclusions

This paper has explored the determinants of remit-

tances and their associated transaction costs. Not sur-

prisingly, recorded remittance inflows depend positively

on the country's stock of migrants residing abroad. In

the sample, remittances are also pro-cyclical, possibly

reflecting their use to finance investments, and not just

to smooth consumption. In addition, remittances depend

negatively on transfer costs and exchange rate restrictions.

In turn, transfer costs are lower when financial systems

are more developed, and exchange rates less volatile.

The statistically significant and economically mean-ingful negative impact of transactions costs on remittances

suggests that, when costs are high, migrants either refrain

from sending money home or else remit through informal

channels. Given the large informal flows reported in

household surveys and the negative correlation between

remittances and net errors and omissions, we interpret the

results as reflecting a large informal sector. That said,

the informal sector appears to have shrunk over time,

especially in Latin America and Asia. To some extent, this

development has generated misleading impressions about

the true speed at which remittances are growing.Reductions in transaction costs would encourage an

increase in remittance flows, and/or a further shift of

remittance flows towards the formal sector. Such a shift

might yield significant benefits to policy makers and

development workers. First, if policy is being designed

to encourage remittances or stimulate investment, it

is important to know the true size of the flows. Inac-

curate information may lead to inappropriate initiatives.

Second, from an efficiency standpoint, a large share of

informal remittances in an economy suggests that rents to

banks and money transfer providers in the official market

are very large, and there may be simple ways to improvecompetition and increase the remittances received. Third,

there may be positive externalities from using formal

channels (and especially financial institutions such as

banks) to transfer money, including increased access to

credit and the use of financial institutions for savings.

Appendix A. Data sources

This appendix provides further details on the data

used in the essay, and in particular discusses the time-

series employed to construct a measure of remittances.Throughout the essay, regional classifications follow

the current IMF groupings.

The most reliable information available on remit-

tance flows is published by the IMF in its annual

Balance of Payments Statistics. Unless otherwise indi-

cated, remittances were defined, as recommended by

Ratha (2003), as the sum of three items in the Balance of

Payments Statistics: “Compensation of Employees”

(part of the income component of the current account),

“Workers' Remittances” (part of current transfers in the

current account), and “Migrants' Transfers” (part of the

capital account). Specifically, the IMF's Balance of

Table 6

Remittances as a determinant of net errors and omissions

Dependent variable: net errors and omissions/GDP

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

Remittances/

GDP

−0.070

(1.48)

−0.080⁎

(1.68)

−0.082⁎

(1.69)Stock of migrant

workers, as

share of source

population

0.083⁎⁎

(2.18)

0.067

(1.50)

0.071

(1.62)

F-test of joint

significance

2.96

(0.05)

Observations 2074 2475 1984 1984 1984

Number of

countries

125 137 123 123 123

R-squared 0.01 0.01 0.02 0.02 0.01

Country and year fixed effects are included in all regressions. ⁎, ⁎⁎,

and ⁎⁎⁎ denote significance at, respectively, the 10%, 5% and 1%

level. Absolute values of robust t -statistics are in parentheses.

14 As noted,we do not have annual data on migration. Rather,migration

is estimated from two data points, and is therefore entered as a country-specific trend.




Payments Manual, Fifth Edition, defines “Workers'

Remittances” as current transfers made by migrants who

are employed and resident in another economy. This

typically includes those workers who move to an

economy and stay, or are expected to stay, a year or

longer. “Compensation of Employees” instead com- prises wages, salaries, and other benefits (cash or in-

kind) earned by nonresident workers for work per-

formed for residents of other countries. Such workers

typically include border and seasonal workers, together

with some other categories, e.g., local embassy staff.

Finally, “Migrants' Transfers” include financial items

that arise from the migration (change of residence) of

individuals from one economy to another.

Following discussions with the IMF Statistical

Department, the IMF country desks, and national

authorities on the precise construction of the abovemeasures in each specific country, “Compensation of

Employees” was excluded from remittances for several

countries.15 In general, the “Other Current Transfers”

item was not included in the definition of total

remittances. However, the Balance of Payments Statis-

tics Yearbook specifies explicitly that migrants' remit-

tances are in fact recorded under “Other Current

Transfers” for Kenya, Malaysia, and the Syrian Arab

Republic. Additional adjustments or additions to the

series for remittances were made on the basis of

information received from IMF country desks and

national authorities.16,17

Details of some other key variables are as follows.

• Domestic output . This is measured as real GDP per

capita. It comes from the Penn WorldTable Version 6.1.

• Main host output . For each country i, this is measured

as real GDP per capita in the country j which con-

tains the largest share of country i's migrant workers.

It is from the Penn World Table Version 6.1 and the

above-mentioned migration data from the OECD.

• Dual exchange rate dummy. This binary indicator

specifies if a country has more than one exchange

rate that may be used simultaneously for different

purposes and/or by different entities. It comes from

the IMF's Annual Report on Exchange Arrange-

ments and Exchange Restrictions, 2003 (ARREAR).

• Financial risk . This indicator assesses a country's

ability to pay its way by financing its official, com-mercial and trade debt obligations. It is based on the

following components: foreign debt as a percentage

of GDP, foreign debt service as a percentage of

exports of goods and services, current account as a

percentage of exports of goods and services, net

liquidity as months of import cover, and exchange

rate stability. It comes from the International Country

Risk Guide.

• Bank concentration. This measure is calculated by

taking the assets of the three largest banks in a

country as a share of the assets of all commercial banks. It is drawn from the World Bank's Financial

Structure Database.

• Dollarization dummy. This binary indicator equals

unity for any year when a country is officially

dollarized.

• Restrictions on foreign-currency deposits held

abroad . This indicator specifies whether resident

accounts that are maintained in foreign currency and

held abroad are allowed. It is drawn from ARREAR.

References

Aggarwal, R., Spatafora, N., 2005. Workers' remittances and economic

development. IMF (2005). IMF, Washington, DC. Chapter II.

AITE, 2005. Consumer Money Transfers.

Celent, 2002. Global Money Transfers.

De Luna Martinez, J., 2005. “Workers remittances to developing

countries: a survey with central banks on selected public policy

issues,” World Bank mimeo.

Djankov, Freund, Pham-si, 2005. “Trading on time,” World Bank

Working Paper No. 3909.

Docquier, F., Marfouk, A., 2005. International migration by education

attainment, 1990–2000. In: Ozden, C., Schiff, M. (Eds.), Interna-

tional Migration, Remittances and the Brain Drain. World Bank,

pp. 151–200.

ECFIN, 2004. EU Survey on Workers' Remittances from the EU to

Third Countries. European Commission.

Faini, R., 2006. “Remittances and the Brain Drain,” CEPR Discussion

Paper No. 5720.

Freund, C., Spatafora, N., 2005. Remittances: transaction costs, deter-

minants, and informal flows. World Bank Policy Research

Working Paper, vol. 3704. World Bank, Washington, DC.

Genesis Analytics, 2003. African Families, African Money Bridging

the Money Transfer Divide. April, http://www.finmarktrust.org.za.

Gibson, McKenzie, and Rohorua, 2005. “How cost-elastic are

remittances? Estimates from Tongan migrants in New Zealand”

Mimeo, World Bank.

IMF, 1993. Balance of Payments Manual, fifth Ed. IMF, Washington,DC.

15 Specifically, Argentina, Australia, Azerbaijan, Barbados, Belize,

Benin, Brazil, Cambodia, Cape Verde, China, Côte d’Ivoire, Dominican

Republic, Ecuador, El Salvador, Guyana, Italy, Panama, Rwanda,

Senegal, Seychelles, Turkey, and Venezuela. For most of these countries,

the Balance of Payments Statistics Yearbook states explicitly that no

information on border and seasonal workers is included in this category.16 See Freund and Spatafora (2005) for details.17 No data on remittances were available for the following countries,

and they were therefore excluded from the analysis: Afghanistan, I.S.

of; Angola; the Bahamas; Bahrain; Bhutan; Brunei Darussalam;

Burundi; Canada; Congo, Dem. Rep. of; Iraq; Kuwait; Liberia; Libya;

Qatar; Singapore; Taiwan Province of China; United Arab Emirates;Uzbekistan; Vietnam; and Zambia.


http://www.finmarktrust.org.za/

http://www.finmarktrust.org.za/



IMF, 2005. World Economic Outlook: Globalization and External

Imbalances, April 2005. IMF, Washington, DC.

Orozco, M., 2003. “Worker remittances: an international comparison,”

Working Paper commissioned by the Multilateral Investment Fund

of the Inter-American Development Bank.

Pieke, F., Van Hear, N., Lindley, A., 2005. Synthesis Study: A Part of

the Report on Informal Remittance Systems in Africa, Caribbeanand Pacific Countries. ESRC Centre on Migration, Policy and

Society.

Ratha, D., 2003. Workers' remittances: an important and stable source

of external development finance. Global Development Finance

2003. World Bank, Washington, DC, pp. 157–175. Chapter 7.

Roberts, B., Banaian, K., 2004. Remittances in Armenia: Size,

Impacts, and Measures to enhance their Contribution to Develop-

ment. Special Study for USAID/Armenia, Bearing Point. October.

Sander, C., 2003. Capturing a market share? Migrant remittance

transfers & commercialization of microfinance in Africa. Paper

prepared for the Conference on Current Issues in Microfinance,

Bannock Consulting.

Sander, C., Maimbo, Munzele, S., 2003. Migrant labor remittances in

Africa: reducing obstacles to developmental contributions. AfricaRegion Working Paper Series, vol. 64. World Bank.

Sayan, S., 2006. “Business cycles and workers' remittances: how do

migrant workers respond to cyclical movements of GDP at home?”

mimeo (Washington, DC: IMF).

Siddiqui, T., Abrar, C.R., 2003. Migrant Worker “Remittances and

Micro-Finance in Bangladesh" Working Paper No. 38, Social

Finance Program, International Labour Office.

Swanson, P., Kubas, M., 2005. Global Money Transfer Report. Piper Jaffray.

Wines, M., 2007. Caps on prices only deepen Zimbabweans' misery.

The New York Times. 2 August 2007.

World Bank, 2006. Global Economic Prospects 2006: Economic

Implications of Remittances and Migration.WorldBank, Washington,

DC.

Yang, D., 2004. “International migration, remittances, and household

investment: evidence from Philippine migrants' exchange rate

shocks,” mimeo (University of Michigan).

Yang, D., 2006. “Coping with disaster: the impact of hurricanes on

international financial flows, 1970–2002,” mimeo (University of

Michigan).


Documents

Remittances, transaction costs, and informality