Access to Finance

Forum
Reports by CGAP and Its Partners
No. 7, June 2013

Microcredit Interest Rates and Their Determinants

2004–2011

Richard Rosenberg, Scott Gaul, William Ford, and Olga Tomilova

a
 Acknowledgments
This paper and the research behind it have been jointly produced by the
Microfinance Information Exchange (MIX), KfW, and CGAP.
The authors are grateful to KfW for financial support, to MIX for data and
data processing, and to CGAP for analytical models and publication services.
We thank Matthias Adler, Gregory Chen, Alexia Latortue, and Kate McKee for
insightful comments. Of course, it is the authors, not the commentators or
sponsoring agencies, who are responsible for the conclusions and views
expressed here.

© CGAP, MIX, and KfW, 2013

CGAP
1818 H Street, N.W.
Washington, DC 20433 USA

Internet: www.cgap.org
Email: [email protected]
Telephone: +1 202 473 9594

All rights reserved. Requests for permission to reproduce portions of it should be sent to CGAP at the address in the
copyright notice above. CGAP, MIX, and KfW encourage dissemination of their work and will normally give permission
promptly and, when reproduction is for noncommercial purposes, without asking a fee. Permission to photocopy
portions for classroom use is granted through the Copyright Center, Inc., Suite 910, 222 Rosewood Drive, Danvers,
MA 01923 USA.
 Contents

Introduction 1
Section 1. Level and Trend of Interest Rates 4
Section 2. Cost of Funds 9
Section 3. Loan Loss Expense 11
Section 4. Operating Expenses (and Loan Size) 13
Section 5. Profits 18
Section 6. Overview and Summary 21
References
23
annex Data and Methodology 24

i
Introduction

F
rom the beginning of modern microcredit,1 its defer most discussion of methodology until the An-
most controversial dimension has been the in- nex, one point is worth making here at the begin-
terest rates charged by microlenders—often re- ning. The earlier CGAP paper used data from a con-
ferred to as microfinance institutions (MFIs).2 These sistent panel: that is, trend analysis was based on 175
rates are higher, often much higher, than normal profitable microlenders that had reported their data
bank rates, mainly because it inevitably costs more to each year from 2003 through 2006. This approach
lend and collect a given amount through thousands gave a picture of what happened to a typical set of
of tiny loans than to lend and collect the same amount microlenders over time.
in a few large loans. Higher administrative costs have This paper, by contrast, mainly uses data from
to be covered by higher interest rates. But how much MFIs that reported at any time from 2004 through
higher? Many people worry that poor borrowers are 2011.3 Thus, for example, a microlender that entered
being exploited by excessive interest rates, given that the market in 2005, or one that closed down in 2009,
those borrowers have little bargaining power, and would be included in the data for the years when
that an ever-larger proportion of microcredit is mov- they provided reports. We feel this approach gives a
ing into for-profit organizations where higher inter- better picture of the evolution of the whole market,
est rates could, as the story goes, mean higher returns and thereby better approximates the situation of a
for the shareholders. typical set of clients over time. The drawback is that
Several years ago CGAP reviewed 2003–2006 fi- trend lines in this paper cannot be mapped against
nancial data from hundreds of MFIs collected by the trend lines in the previous paper, because the sam-
Microfinance Information Exchange (MIX), look- ple of MFIs was selected on a different basis. (We
ing at interest rates and the costs and profits that did calculate panel data for a consistent set of 456
drive those interest rates. The main purpose of that MFIs that reported from 2007 through 2011; we
paper (Rosenberg, Gonzalez, and Narain 2009) was used this data mainly to check trends that we report
to assemble empirical data that would help frame from the full 2004–2011 data set.)
the question of the reasonableness of microcredit
interest rates, allowing a discussion based more on 3. F
 or readers interested in the composition of this group, we can
summarize the distribution of the more than 6000 annual ob-
facts and less on ideology.
servations from 2004 through 2011. Note that this is the distri-
In this paper, we review a better and fuller set of bution of MFIs, not of customers served. Category definitions
MIX data that runs from 2004 to 2011. Though we can be found in the Annex:
Region: SSA 14%, EAP 13%, ECA 18%, LAC 34%, MENA 5%, S.

1. I n this paper, “microcredit” refers to very small, shorter-term, Asia 16% (for abbreviations see Figure 1).
usually uncollateralized loans made to low-income microen-  rofit status: for-profit 39%, nonprofit 59%, n/a 2%. (Note
P
trepreneurs and their households, using unconventional tech- that for-profit MFIs serve the majority of borrowers, because
niques such as group liability, frequent repayment periods, they tend to be larger than nonprofit MFIs.)
escalating loan sizes, forced savings schemes, etc.
2. MFIs are financial providers that focus, sometimes exclusive- Prudentially regulated by financial authorities? yes 57%, no

ly, on delivery of financial services targeted at low-income 41%, n/a 2%
clients whose income sources are typically informal, rather  egal form: bank 9%, regulated nonbank financial institution
L
than wages from registered employers. Among these financial 32%, credit union/co-op 13%, NGO 38%, rural bank 6%, other
services, microcredit predominates in most MFIs today, but or n/a 2%
savings, insurance, payments, and other money transfers are  arget market: low micro 42%, broad micro 49%, high micro
T
being added to the mix, as well as more varied and flexible 5%, small business 4%
forms of credit. MFIs take many forms—for instance, informal
Financial intermediation (voluntary savings): >1/5 of assets

village banks, not-for-profit lending agencies, savings and
39%, up to 1/5 of assets 17%, none 44%
loan cooperatives, for-profit finance companies, licensed spe-
cialized banks, specialized departments in universal commer- Age: 1–4 years 10%, 5–8 years 19%, >8 years 69%, n/a 2%
cial banks, and government programs and institutions. Borrowers: <10k 48%, 10k–30k 23%, >30k 29%

1
 The data set and the methodology used to gener- is profit (or loss). A simplified version of the relevant
ate our results are discussed further in this paper’s formula is
Annex. Our main purpose here is to survey market Income from loans = Cost of funds + Loan loss
developments over the period; there will not be expense + Operating Expense + Profit4,5
much discussion of the “appropriateness” of interest
In other words, interest income—the amount of
rates, costs, or profits. A major new feature of this
loan charges that microlenders collect from their
paper is that it is complemented by an online data-
customers—moves up or down only if one or more of
base, described later in the paper, that readers can
the components on the right side of the equation
use to dig more deeply into the underlying MIX
moves up or down.
data—and in particular, to look at the dynamics of
That formula provides the structure of this paper:
individual country markets.
Not surprisingly, five more years of data re- • Section 1 looks at the level and trend of micro-
veal some important changes in the industry. For lenders’ interest rates worldwide, and breaks
instance, them out among different types of institutions
(peer groups).
• Globally, interest rates declined substantially
through 2007, but then leveled off. This is partly • Section 2 examines the cost of funds that micro-
due to the behavior of operating (i.e., staff and lenders borrow to fund their loan portfolio.
administrative) costs, whose long-term decline • Section 3 reports on loan losses, including worri-
was interrupted in 2008 and 2011. Another fac- some recent developments in two large markets.
tor has been a rise in microlenders’ cost of
funds, as they expanded beyond subsidized re- • Section 4 presents trends in operating expens-
sources and drew increasingly on commercial es, and touches on the closely related issue of
borrowings. loan size.

• Average returns on equity have been falling, and • Section 5 looks at microlenders’ profits, the most
the percentage of borrowers’ loan payments that controversial component of microcredit interest
go to profits has dropped dramatically. This is rates.
good news for those who are worried about ex- • A reader without time to read the whole paper
ploitation of poor borrowers, but may be more may wish to skip to Section 6, which provides a
ambiguous for those concerned about the finan- graphic overview of the movement of interest
cial performance of the industry. rates and their components over the period and
• For the subset of lenders who focus on a low- a summary of the main findings.
end (i.e., poorer) clientele, interest rates have • The Annex describes our database and method-
risen, along with operating expenses and cost ology, including the reasons for dropping four
of funds. On the other hand, low-end lenders large microlenders6 from the analysis.
are considerably more profitable on average
than other lenders (except in 2011, when the A dense forest of data lies behind this paper. To
profitability of the group was depressed by a avoid unreasonable demands on the reader’s pa-
repayment crisis in the Indian state of Andhra tience, we have limited ourselves to the tops of
Pradesh). some of the more important trees. But MIX has
posted our data files on its website, including Excel
The percentage of borrowers’ interest payments
that went to MFI profits dropped from about one-
4. “ Operating expense” is the term MIX uses to describe person-
fifth in 2004 to less than one-tenth in 2011. nel and administrative costs, such as salaries, depreciation,
As in the 2009 paper, we will look not just at in- maintenance, etc.
terest rates but also at their components—that is, the 5. A fuller formula is
Income from loans + Other income = Cost of funds + Loan loss
main factors that determine how high interest rates
expense + Operating expense + Tax + Profit
will be. Lenders use their interest income to cover 6. BRI (Indonesia), Harbin Bank (China), Postal Savings Bank
costs, and the difference between income and costs of China, and Vietnam Bank for Social Policy.

2
pivot tables where readers can slice the data any way more than 800 different data cuts that were avail-
they like (http://microfinance-business-solution. able. Most of the information presented here is in
mixmarket.org/rs/microfinance/images/Interest the form of global cuts, often broken out by peer
Rate Paper Supporting Data.zip). The pivot tables groups, such as region, for-profit status, loan meth-
allow a user to select among 14 financial indicators odology, etc. But for many readers, the most rele-
and display 2004–2001 adjusted or unadjusted re- vant peer grouping will consist of the micro-
sults (weighted averages and quartiles) broken out lenders operating in a particular country. We
in any of nine different peer groupings, including strongly encourage these readers to use the online
individual countries. pivot tables to customize an analysis of what has
In choosing which groupings of these data to in- been happening in any specific country.
clude in the paper, we have had to select among

3
 1
Section

Level and Trend of Interest Rates

How to measure microcredit microborrowers are really paying, interest yield is

inferior to APR in important ways. For instance,
interest
• In 2011, about a third of microborrowers were
Before presenting data and findings, we need to served by lenders that use compulsory savings—
discuss two different ways to measure interest that is, they require borrowers to maintain a per-
rates on microloans: interest yield and annual per- centage of their loan on deposit with the lender.
centage rate (APR). Understanding the distinction This practice raises the effective interest rate,
between these two is crucial for a proper interpre- because the deposit requirement reduces the net
tation of the interest rate data we present in this loan disbursement that the borrower can actu-
section. ally use, while the borrower pays interest on the
From a client standpoint, a typical way to state full loan amount. APR incorporates this effect,
interest rates is to calculate an APR on the client’s while interest yield does not.
particular loan product. APR takes into account the
amount and timing of all the cash flows associated • MIX’s calculation of interest yield lumps the
with the loan, including not only things that are ex- lender’s entire portfolio together, even though
plicitly designated as “interest” and “principal,” but that portfolio may contain loan products with
also any other expected fees or charges, as well as quite different terms, and may even include
compulsory deposits that are a condition of the products that are better characterized as small
loan. This APR indicator is a good representation of business loans rather than microloans.
the effective cost of a loan for borrowers who pay as • The denominator of the MIX interest yield ratio
agreed. APR can be substantially different from is GLP—the total amount of all outstanding loans
(usually higher than) the stated interest rate in the that has neither been repaid nor written off. But
loan contract. some of those loans are delinquent—the borrow-
MicroFinance Transparency (MF Transparency) ers are behind on payments. The effect of this
is building a database with APR information on difference can be illustrated simply. Suppose
some or all of the significant microlenders in a that total interest income is 200, and GLP is
growing range of countries. Collection of these data 1000, producing an interest yield of 20 percent
is labor-intensive and depends on the willing coop- that the “average” borrower is paying. But if the
eration of microlenders who might occasionally portion of the loans that is actually performing is
find the publication of these pricing specifics em- only 800, then the average borrowers are really
barrassing. As of this writing, the MF Transparency paying 25 percent.8
website displays data from 17 countries.7
In contrast, the MIX database we draw from in An internal MIX analysis in 2011, based on seven
this paper cannot generate APRs. What MIX pro- countries for which MF Transparency also had
vides is “interest yield,” which expresses the total of data, found that the MIX interest yield understat-
all income from loans (interest, fees, other loan
8. M
 IX is building better information about compulsory depos-
charges) as a percentage of the lender’s average an-
its, and makes adjustments that attempt to represent net port-
nual gross loan portfolio (GLP). From the vantage folio more accurately, but we found that these MIX data were
point of the lender, interest yield is clearly mean- not yet consistent enough to produce reliable results at pres-
ingful. But as an indication of what individual ent. A very rough analysis of these data suggests that compul-
sory deposits in some MFIs might add something like 3 per-
7. http://data.mftransparency.org/data/countries/ cent to the worldwide average APR. The average impact of
adjusting for nonperforming loans is harder to decipher.

4
ed the MF Transparency APR by an average of Finally, we emphasize that the issue discussed above
about 6 percentage points. However, the sample applies only to data about interest rates. It poses no
was too small to allow for much generalization of problem for the majority of our analysis, which deals
this result. with the determinants of interest rates, namely cost
Given the limitations of the MIX interest yield of funds, loan losses, operating expenses, and profit.
measure, why are we using it in this paper? One
reason is that the MIX’s much broader coverage
provides a better sample of the worldwide micro- Level of Interest Yields in 2011
credit market: more than 105 countries for 2011,
compared to MF Transparency’s 17. An even more Figure 1 shows a global median interest yield of
important reason is that MIX, having started col- about 27 percent. Distribution graphs like this one
lecting data long before MF Transparency, has remind us that there is wide variation in microcred-
many more years of data, allowing trend analysis it rates, so any statement about a median (or aver-
that is not yet possible for the latter. We think it age) rate is a composite summary that veils a great
highly likely that interest yield trends and APR deal of underlying diversity. The regional distribu-
trends would move approximately in parallel tion indicates that rates vary more widely in Africa
over a span of years. A detailed discussion of this and Latin America than in other regions. Also, we
point will be posted along with our under- notice that rates are substantially lower in South
lying data (http://microfinance-business-solution. Asia than elsewhere: the relative cost of hiring staff
mixmarket.org/rs/microfinance/images/Interest tends to be lower there, and—at least in Bangla-
Rate Paper Supporting Data.zip). desh—the political climate and the strong social ori-
How, then, should the reader regard the mean-
ingfulness of interest yield data? Here is our view:
1. Actual effective rates paid for specific loan prod- figure 1
ucts at a point in time. Interest yield probably un-
MFI Interest Yield Distribution, 2011
derstates these by varying and often substantial
100
amounts.
90
2. Peer group differences (e.g., how do rates at for-
profit and nonprofit microlenders compare on 80

average?). We think that substantial differences 70

in interest yield among peer groups are very 60
likely a meaningful indication of a difference
Percent

50
among the groups in what their average bor-
rowers pay. However, some caution is appropri- 40
ate here, because the gap between interest yield 30
and true APR can vary from one peer group to
20
another.9
10
3. Time-series trends. Trends in interest yields (the
0
main focus of this section) are probably quite a WORLD Africa EAP ECA LAC MENA S. Asia
good indicator of trends in what typical borrow-
Note: Interest and fee income from loan portfolio as % of average GLP,
ers are actually paying, on the plausible assump- 866 MFIs reporting to MIX. The thick horizontal bars represent medi-
tion that the gap between interest yield and APR ans; the top and bottom of the solid boxes represent the 75th and 25th
percentiles, respectively; and the high and low short bars represent the
stays relatively stable on average from one year 95th and 5th percentiles, respectively. So, for example, 95 percent of
to the next. the MFIs in the sample are collecting an interest yield below about 70
percent. Data here are unweighted: each MFI counts the same regard-
less of size. EAP = East Asia and Pacific, ECA = Europe and Central
9. T
 his is particularly true when comparing MFIs that focus on
Asia, LAC = Latin America and the Caribbean, MENA = Middle East
smaller loans to poorer clients, as against MFIs with a broad
and North Africa.
suite of loan products, some of which serve clients that might
not fit one’s particular definition of “micro.”

5
 entation of the industry have probably led manag- On the assumption that the microcredit market
ers to focus more on keeping rates low.10 is getting more saturated and competitive in quite a
few countries, we might have expected a different
result. Analysis of individual countries where the
Global average interest rates have market is thought to be more competitive shows
stopped declining in recent years continued interest rate decline post-2006 in some
(e.g., Bolivia, Nicaragua, Cambodia) but not in oth-
Figure 2 shows a drop in average global microcredit ers (e.g., Mexico, Bosnia/Herzegovina, Indonesia).
rates through 2007, but not thereafter. (Inflation- Sorting out the evidence on the effects of competi-
adjusted rates fell in 2008 because few micro- tion would require more detailed country analysis
lenders raised their rates enough to compensate for than we were able to do for this paper.
the spike in worldwide inflation that year.)11 The
analysis of interest rate determinants later in the
paper suggests that the main reason world average Peer group patterns
rates didn’t drop after 2007 is that operating (i.e.,
staff and administrative) costs stayed level.12 The regional breakout in Figure 3 shows that over
the full 2004–2011 period, Latin America is the only
region with no significant decline in average inter-
figure 2 est yield. However, there is important regional vari-
ation since 2006: Africa and East Asia/Pacific show
Global Interest Yield Trends, 2004–2011 substantial continued declines—perhaps because
35 they were the least developed markets in 2006. At
Nominal any rate, these two regions are the ones that sub-
30
stantially improved their operating expenses since
25 2006 (see Figure 12). But reported average rates ac-
tually went back up in Latin America, the most
20 Real commercialized of the regions.
Percent

15 Figure 4 illustrates the unremarkable finding

that for-profit microlenders collect higher average
10
interest yields than nonprofit microlenders. How-
5 ever, for-profit interest rates have dropped more
than nonprofit interest rates: the average difference
0
2004 2005 2006 2007 2008 2009 2010 2011 between the two peer groups dropped from 5 per-
centage points in 2004 down to 1.7 percentage
Note: Global interest and fee income from loans/average total GLP,
weighted by GLP, both nominal and net of inflation. points by 2011. By way of illustration, on a $1000
loan in 2011, the annual difference between the for-
profit and nonprofit interest charges would amount
10. F igure 1 and subsequent figures showing percentile distribu- on average to $17, or less than $1.50 per month.
tions are unweighted; in other words, each MFI counts the When we separate microlenders by the target
same regardless of its size. Not surprisingly, the median in market they serve (Figure 5), we find that in institu-
such a distribution may be different from the weighted aver-
age (e.g., Figure 3) where large MFIs count for proportion- tions focused on the low-end market (smaller aver-
ally more than small MFIs. However, in the particular case of age loan sizes, and thus presumably poorer borrow-
the 2011 global interest yield, the weighted average (see Fig- ers) interest rates are actually higher in 2011 than
ure 2) and the median are very close, about 27 percent.
11. The same effects show up in panel analysis where we tracked
they were in 2004.13
the 456 MFIs that reported consistently to MIX every year
from 2007 to 2011. 13. L
 oan sizes here are measured as a percentage of countries’
12. As we will see later (compare Figures 3 and 12), the correla- per capita national income. People with wide on-the-ground
tion between interest yield and operating cost shows up at experience of many MFIs agree that their average loan sizes
the regional level: Africa and EAP, the two regions with inter- bear some rough relation to client poverty—poorer clients
est rate declines since 2006, also had lower operating costs. tending to take smaller loans—but the relationship is very far
from perfect. See, for instance, Schreiner, Matul, Pawlak, and
Kline (2006) and Hoepner, Liu, and Wilson (2011).

6
 Figure 6, comparing regulated and nonregulated
microlenders,14 seems to point in the same direc- figure 3
tion. Regulation refers here to licensing and/or
Interest Yield Changes 2004–2011
prudential supervision by the country’s banking
45
authorities. Most of the regulated microcredit port- 2004
40 39% 2006
folio is in banks, and most of these are for-profit. 37%
35% 34% 2011
35
The regulated lenders tend to have lower rates: they 30% 30% 30% 30% 30%
30 28%
tend to offer larger loans, while the nonregulated 26% 27%

Percent
26% 27% 25%26%
25%
25 23% 23%
MFIs tend to make smaller loans that require high- 22%
21%
20
er operating costs per dollar lent. Rates among non-
15
regulated microlenders have been rising substan-
10
tially since 2006.
5
0
14. “ Regulated” refers to banks and other finance companies that WORLD Africa EAP ECA LAC MENA S. Asia
are subject to prudential regulation and supervision by the –0.4% –2.4% –1.3% –1.0% –0.0% –0.6% –1.1%
county’s banking and financial authorities. The rest of the +0.1% –2.5% –1.5% 0% +0.7% +0.2% –0.4%
MFIs are categorized as “nonregulated”: like any other busi- Avg. change per year, 2004–2011 Avg. change per year, 2006–2011
ness, they are subject to some regulation (e.g., consumer pro-
tection) but not to prudential regulation whose objective is to Note: Interest and fee income from loans as percentage of average GLP for
guard the financial health of an institution taking deposits the period, weighted by GLP. The Africa series begins in 2005 rather than
from the public. MFIs are categorized based on their status 2004.
in 2011.

figure 4 figure 5

For-Profit vs. Nonprofit Interest Yields, Interest Yields by Target Market, 2004–2011
2004–2011 40
Low End
35 35
For-profit MFIs
30 30
Broad
25 25
Percent

Nonprofit MFIs 20
20 High End
Percent

15
15
10
10
5
5 0
2004 2005 2006 2007 2008 2009 2010 2011
0
2004 2005 2006 2007 2008 2009 2010 2011
Note: Total interest and fee income / average total GLP, weighted by
GLP, nominal. MFIs are grouped by “depth”—average loan balance
Note: Total interest and fee income/average total GLP, weighted by per borrower as % of per capita gross national income. For the “low
GLP. MFIs are assigned to “for-profit” or “nonprofit” depending on end” market, depth is <20% or average loan balance < US$150. For
their legal status in 2011. “broad,” depth is between 20% and 149%. For “high end,” depth is
between 150% and 250%. For the “small business” market, which is
not included in this graph, depth is over 250%.

7
 The two preceding figures show higher rates for
figure 6 lenders that tend to focus on smaller borrowers. At
first blush, this looks like bad news for low-end cli-
Regulated vs. Nonregulated Interest Yields,
ents. However, the trend probably reflects some
2004–2011
shifting of low-end clientele: if banks and broad-
40 market microlenders have been capturing more of
Nonregulated
35
the easier-to-serve portion of poor borrowers, then
the unregulated and low-end microlenders would be
30
left with a somewhat tougher segment of clients, and
25 their rising interest rates might simply reflect the
Regulated
higher expenses of serving this segment.15 Another
Percent

20
factor is that funding costs for low-end lenders have
15
been rising, as we will see in Figure 8.
10 The fact that costs and thus interest rates are ris-
5 ing for microlenders who focus on poorer clients
has a bearing on the perennial argument over
0
2004 2005 2006 2007 2008 2009 2010 2011 whether to protect the poor by imposing interest
Note: Total interest and fee income/average total GLP, weighted by rate caps. As costs rise for low-end microlenders, a
GLP. given fixed-interest rate cap would put (or keep)
more and more of them out of business as the years
go by.
Having sketched a few important patterns and
trends in interest rates, we now turn to the princi-
pal elements that determine (or “drive”) those rates.
To repeat, the simplified description of this rela-
tionship is
Income from loans = Cost of funds + Loan loss
expense + Operating Expense + Profit
After looking at these determinants individually,
we will put them back together again in Section 6
to show how the trends in these elements combine
to produce the trends in interest yields.

15. I f this conjecture is true, we might expect to see average loan

sizes decreasing in both broad-market and low-end MFIs, as
well as in both regulated and nonregulated MFIs. This is in-
deed what has happened—average loan sizes have declined
by roughly five percentage points among all these groups
since 2006. And operating expense ratios have been rising for
MFIs aimed at the low-end clientele.

8
 2
section

Cost of Funds

M
icrolenders fund their loans with some Peer group analysis
combination of equity (their own money)
and debt (money borrowed from deposi- Figure 8 shows another piece of bad news for mi-
tors or outside lenders). In a sense, the equity is free, crolenders focused on low-end borrowers: the aver-
at least for a not-for-profit lender that has no share- age cost of funds is growing faster for this peer
holder owners who collect dividends. But borrowed group than for others. Funding costs for micro-
funds entail a cost in the form of interest expense. lenders that focus on high-end borrowers have
stayed fairly level, while funding costs have climbed
Funding costs have been rising. substantially for broad-market microlenders and
Figure 7 shows a slow, steady climb in the nominal especially for low-end microlenders.18 This rise in
costs at which microlenders can borrow money to funding costs is part of the reason that average
fund their loan portfolios. This climb is both less worldwide interest yields paid by microborrowers
pronounced but more jumpy when we look at the have not been declining in the past few years, and
real (i.e., net of inflation) cost of funds.16 The most interest yields paid by customers of low-end lend-
probable explanation of the rise in borrowing costs ers have actually grown, as we saw in Section 1.
is that as microlenders expand, they can fund less of
18. F
 or definitions of the three target market designations, see
their portfolio from the limited amounts of heavily the note below Figure 5.
subsidized liabilities from development agencies,
and they have to turn increasingly toward more ex-
pensive commercial and quasi-commercial debt
figure 7
from local and international markets.
Some people hope that funding costs will decline Cost of Funds, Nominal and Real, 2004–2011
substantially as more and more microlenders mobi- 8
lize voluntary deposits, but such a result is far from Nominal
guaranteed. Over the time span of our study, aver- 6
age funding costs actually look slightly higher for
lenders that rely heavily on voluntary savings than 4
for lenders that take no such savings.17 Also note
Percent

2
that any decrease in funding cost produced by sav- Real
ings mobilization can be offset by increases in oper- 0
2004 2005 2006 2007 2008 2009 2010 2011
ating costs to administer the savings function, espe-
cially for small-sized liquid deposits that are aimed –2
at the microclientele.
–4
16. T he sharp changes in real rates in 2008 and 2009 probably
reflect the time it took for interest contracts to reprice fol- Note: Financial expense as % of liabilities, weighted by liabilities, both
lowing the world inflation spike in 2008. nominal and adjusted for each country’s inflation.
17. The difference, about 0.1 percentage points, is probably not
statistically significant.

9
 Not surprisingly, regulated institutions like
figure 8 banks and licensed finance companies have been
able to borrow money an average of 1.5 percentage
Cost of Funds (Nominal) by Target Market
points cheaper than nonregulated lenders. Most of
2004–2011
the regulated microlenders can take savings, and in-
12 terest cost for their savings is lower than for large
commercial borrowings.19 Regulated institutions
10 Low End have some cost advantage even on large commer-
cial loans: lenders see them as safer because they
8
Broad
are licensed and supervised by the banking authori-
ties. Also, regulated microlenders on average can
Percent

6
absorb larger borrowings, which can reduce their
4 High End interest and transaction costs.

2
19. A
 t first blush, this may seem inconsistent with the preceding
finding that MFIs who take voluntary deposits have higher
0 funding costs that those who do not. The explanation is that
2004 2005 2006 2007 2008 2009 2010 2011
funding costs have been particularly high for unregulated
Note: Financial expense as % of liabilities, weighted by liabilities. deposit-takers.

10
 3
section

Loan Loss Expense

M
ost microloans are backed by no collateral, The loan levels in Figure 9 are calculated from
or by collateral that is unlikely to cover a microlenders’ reports to MIX, usually but not al-
defaulted loan amount once collection ex- ways based on externally audited financial state-
penses are taken into account. As a result, outbreaks ments. However, microlenders, especially the un-
of late payment or default are especially dangerous regulated ones, use many different accounting
for a microlender, because they can spin out of con- policies for recognizing and reporting problem
trol quickly. loans. Microlenders (like other lenders!) often err in
When a borrower falls several payments behind estimating their credit risk. Their errors are seldom
on a loan, or something else happens that puts even- on the high side, and many external auditors are re-
tual collection of the loan in doubt, the sound ac- markably generous when it comes to allowing opti-
counting practice is to book a “loan loss provision mistic approaches to loan loss accounting. MIX
expense” that reflects the loan’s loss in value—i.e., the makes an analytical adjustment to reported loan
lowered likelihood it will be collected in full. This losses, in effect applying a uniform accounting poli-
practice recognizes probable loan losses promptly cy to recognition of those losses.21 The point of this
rather than waiting for the full term of the loan to adjustment is uniformity, not fine-tuning to the par-
expire and collection efforts to fail before booking ticular circumstances of a given lender; thus the
the loss. If the lender books a provision expense for a MIX loan loss adjustment might not accurately re-
loan, but the loan is later recovered in full, then the flect the risk of each institution’s portfolio. However,
provision expense is simply reversed at that point. In we have no doubt that when looking at broad groups
this section, we look at the quality (i.e., collectability) of microlenders, the MIX adjustments generate a
of microloan portfolios through the lens of net loan picture that is closer to reality than the financial
loss provision expense. We stress that this indicator statement figures submitted by the institutions.
approximates actual loan losses over the years, not As shown in Table 1, MIX’s adjustment has only
just levels of delinquency (late payment). a small effect on Mexican loan loss rates, suggest-
ing that the Mexican loan loss accounting may be
Loan losses have recently been climbing fast in fairly close to realistic. However, the adjustment
India and Mexico, but the average for the rest almost triples India’s average 2011 loan loss from a
of the world has been fairly stable. self-reported 9.7 percent to an adjusted figure of
The spike in India is due mainly to the recent col- almost 29 percent. The authors have not gone back
lapse of microcredit repayment in Andhra Pradesh.20
The apparently serious problem in Mexico has been Table 1 Effect of MIX Adjustments on 2011
longer in the making. But in the rest of the world, Loan Loss Expense
average loan loss has declined from a worrisome
Unadjusted Adjusted
level of almost 4 percent in 2009 back toward a safer
level a bit above 2 percent in 2011. MEXICO 11.9% 12.1%

INDIA 9.7% 28.9%

20. See, for example CGAP (2010) on Andhra Pradesh. 21. MIX’s loan loss adjustment protocol is described in the Annex.

11
 figure 9 figure 10

Loan Loss Provision 2004–2011 Loan Loss Expense by Profit/Nonprofit

14 Status, 2004–2011

12 5.0
4.5
10 Mexico
4.0
8 3.5
Percent

3.0 Profit

Percent
6
2.5
All Other Countries
4 2.0
1.5 Nonprofit
2
India 1.0
0 0.5
2004 2005 2006 2007 2008 2009 2010 2011
0
2004 2005 2006 2007 2008 2009 2010 2011
Note: Net annual provision expenses (unadjusted) for loan impairment
as % of average GLP, weighted by GLP.
Note: Net Loan loss expense (unadjusted) as % of GLP, weighted
by GLP.

to review the individual financial statements of Peer group analysis

the Indian microlenders in MIX, but the prima
facie hypothesis would be that there might be a The only clear pattern we’ve noticed in the peer
massive overhang of under-reported loan losses group breakouts for this indicator is that on aver-
that will continue to depress overall Indian profit- age for-profit microlenders have had higher loan
ability in subsequent years.22 losses than nonprofits do (Figure 10), this would
seem to be a prima facie indication of a tendency
toward riskier lending and collection practice
22. W
 e understand that India’s central bank has relaxed some
among for-profit MFIs on average. However, the
loan-loss accounting rules for MFIs in 2011. The probable
motive is to let Indian commercial banks reduce the losses gap seems to be narrowing, except for the for-prof-
they have to recognize on loans they have made to the MFIs. it spike in 2011, which is almost entirely due to
loan losses of Indian for-profits.

12
 4
section

Operating Expenses (and Loan Size)

O
perating expenses include the costs of im-
plementing the loan activities—personnel figure 11
compensation, supplies, travel, deprecia-
Operating Expense Ratio, 2004–2011
tion of fixed assets, etc. Operating expenses con-
sume the majority of the income of most micro- 18
lenders’ loan portfolios, so this component is the 16
World
largest determinant of the rate the borrowers end 14
up paying. 12
Percent 10
Declines in operating expenses (i.e., improve-
ments in efficiency) have slowed recently. 8

Much of the hope for lower interest rates is based 6

on an expectation that as microlenders acquire 4
more experience they learn to lend more efficiently. 2
Standard economic theory tells us that, in young in- 0
dustries, one normally expects to see cost improve- 2004 2005 2006 2007 2008 2009 2010 2011
ments as firms (or the whole industry in a given Note: Operating (i.e., staff and administrative) expense as % of aver-
market) acquire more experience. Eventually, age GLP, weighted by GLP.
though, the most powerful efficiency lessons have
been learned, and the learning curve flattens out: at
this point efficiency improves slowly if at all in the
figure 12
absence of technological breakthroughs.23 In addi-
tion to the learning curve, there is hope that the Operating Expense Ratio Changes, 2004–2011
pressure of competition will force lenders to find
30
more efficient delivery systems. 28% 28% 2004
2006
Figure 11 shows that global average operating 25 24%
23% 2011
costs for MIX microlenders fell substantially 20%
20 19%
through 2007, but the downward trend was inter- 18%
Percent

17% 17%
16% 16% 16% 16%
15%
rupted in 2008 and again in 2011. Are microcredit 15 15%
13%13%
14%
12%
operating costs getting toward the bottom of their 11% 11%
10
learning curve? Or are we seeing temporary bumps
with further improvement in efficiency yet to come? 5
No conclusion can be drawn at this point—certainly 0
not on the basis of worldwide average behavior. Ef- WORLD Africa EAP ECA LAC MENA S. Asia
–0.3% –1.5% –1.3% –0.7% –0.1% –0.6% –0.2%
ficiency trends differ a lot from one region to an- –0.2% –1.8% –1.5% 0% 0% +0.4% 0%
other (Figure 12). Since 2006, operating efficiency
Avg. change per year, 2004–2011 Avg. change per year, 2006–2011
has improved substantially in relatively immature
Note: Total operating expense/average GLP, weighted by GLP, nominal.
The Africa series begins with 2005 rather than 2004.
23. This is especially the case with microfinance, where there
are relatively few economies of scale after MFIs grow past
5,000 or 10,000 clients (Rosenberg, Gonzalez, and Narain
2009).

13
 markets like Africa and EAP, but has been flat or It is common to equate this kind of “efficiency”
even increased in the other regions. A further com- with the quality of management. But this can be se-
plication, the impact of loan sizes, is discussed later riously misleading, especially in comparing differ-
in this section. ent kinds of microlenders. Managers at the low-end
microlenders and the unregulated microlenders
lend and collect much smaller loans,24 which tend
Peer group analysis of operating to cost more to administer than large loans do,
costs, including the impact of when measured per dollar lent, even with the best
possible management.
loan sizes
Figure 15 uses Philippine data to illustrate two
Thus far, the measure of administrative efficiency points. The main point is that operating cost per
that we have used is operating expense as a percent- dollar lent (the lower plotted curve) does in fact
age of average outstanding GLP. This ratio can be tend to be higher for tiny loans. The secondary
thought of as the operating cost per dollar outstand- point is that interest yield (the upper plotted curve)
ing. It is meaningful for many purposes, but using it parallels the operating cost curve: as we said, oper-
to compare the “efficiency” of different micro- ating cost is typically the most important determi-
lenders can be problematic. We will illustrate this nant of the interest that borrowers pay.25
important and widely overlooked point at some The cost per dollar lent, which we have used so
length, using as examples a comparison among far an as efficiency indicator, penalizes lenders
lenders serving different target markets, and a com- making smaller loans, because their operating costs
parison between regulated and unregulated lenders.
24. S ee Figure 18.
Figures 13 and 14 seem to show not only that 25. The Philippines plot was selected because it was a particularly
both low-end lenders and unregulated lenders are clean and striking illustration of the points being made here.
less efficient than others (i.e., have higher average The relationships are quite a bit looser in most countries, and
occasionally even run in the other direction. Nevertheless
operating costs per dollar of portfolio lent), but also these points are true as statements of general tendency, and
that they are losing efficiency over time. the correlations are substantial.

figure 13 figure 14

Operating Expense Ratio 2004–2011, Operating Expense Ratio by Regulatory

by Target Market Status, 2004–2011
25 25

Low End
20 20
Nonregulated

Broad
15 15
Percent

Percent

Regulated
High End
10 10

5 5

0 0
2004 2005 2006 2007 2008 2009 2010 2011 2004 2005 2006 2007 2008 2009 2010 2011
Note: Operating (staff and administrative) expenses/average GLP. (For Note: Operating (staff and administrative) expenses/average GLP
definitions of the three target market designations, see the note below
Figure 5.)

14
will always tend to be higher as a percentage of
each dollar outstanding. However, we can compen- figure 15
sate (to some extent) for the effect of loan size by
Pricing and Cost Curves for the Philippines
changing our indicator from cost per dollar lent to
cost per loan outstanding—in other words, we di-
125
vide operating costs not by the amount of the aver-
age outstanding loan portfolio, but rather by the
average number of active loans outstanding over
Nominal yield
the year, regardless of how large those loans are. 100 Operating expense
Table 2 illustrates the difference in these indica-
tors with two hypothetical lenders that have the
same size loan portfolio but very different adminis-
75
trative costs. We posit that both institutions are

Percent
managed with the lowest possible operating cost
given their loan sizes and other circumstances.
Using the standard efficiency measure, cost per 50
dollar outstanding (5), the low-end lender looks
bad by comparison, but this is a meaningless result
given the difference in loan sizes. The low-end
25
lender’s efficiency looks better when presented as
(6) cost per loan outstanding.26
But using this latter measure makes the high-
end lender look worse. Are its managers really less
0 25 50 75 100
efficient? No: making a single large loan does tend to Percent
cost more than making a single small loan—for in- Note: Operating (staff and administrative) expenses/average GLP. (For
stance, the larger loan may require additional anal- definitions of the three target market designations, see the note below
Figure 5.)
ysis or a more skilled loan officer. The point is that
as loan size increases, operating cost per loan also
increases but at a less than proportional rate. This
leaves us with the same statement that we made at Table 2 Two Measures of Efficiency
the beginning of the paper: it usually costs more to
lend and collect a given amount of money in many Low-End MFI High-End MFI
small loans than in fewer big loans. 1. Avg number of active loans 100,000 10,000
Now let us return to our efficiency comparison 2. Avg outstanding loan size $200 $2,000
between regulated and unregulated microlenders.
3. Avg loan portfolio [ (1) x (2) ] $20 million $20 million
The cost-per-dollar measure we used in Table 2
made it look as if the unregulated lenders were less 4. Operating expense $4 million $2 million

efficient, and that their efficiency was actually get- 5. Cost per dollar o/s [ (4) ÷ (3) ] 20% 10%
ting worse. But if efficiency is taken as a measure of 6. Cost per loan o/s [ (4) ÷ (1) ] $40 $100
management quality, the comparison is unfair, be-
cause unregulated loan sizes average roughly half of
regulated loan sizes, and are getting smaller over

26. T
 he dynamic would be the same if cost per borrower were
used instead of cost per loan.

15
 time.27 Figure 16 uses cost per loan, which can be a Mission drift; savings mobilization
more useful measure of the evolution of efficiency
over time. This presentation suggests a probability As more and more of the microcredit portfolio moves
that cost management in the unregulated micro- into regulated banks and other for-profit institutions,
lenders is actually improving.28 a common concern is that these commercialized mi-
Turning back to target market peer groups (Fig- crolenders will lose their focus on poor customers
ure 17), we see that by a cost per loan metric, low- and gradually shift to larger (and supposedly more
end lenders no longer look relatively inefficient, profitable) loans. However, it is hard to find support
and their average cost levels have been quite stable for this concern in the MIX data. To begin with, the
in relation to per capita income. At the other end of assumption that larger loans will tend to be more
the spectrum, high-end lenders show improved ef- profitable doesn’t appear to be true, as we will see in
ficiency since 2005 (though some of this is probably the following section when we discuss lenders’ prof-
a result of their declining average loan sizes). its. In fact, the average loan size in for-profit and
Some readers may have found this discussion of regulated MFIs has been dropping steadily since
efficiency measures annoyingly convoluted. By way 2004 (Figure 18).29,30 This doesn’t necessarily mean
of apology, we offer instead a simple take-home that concerns about mission drift are unfounded. But
message: be very cautious when using either effi- if commercialization is producing mission drift, that
ciency measure—cost per dollar or cost per loan—to mission drift does not seem to be playing itself out in
compare the cost-control skills of managers of dif- any widespread shift to larger loans.
ferent institutions.

27. S
 ee Figure 18. 29. T
 he same pattern shows up in data using a consistent panel
28. How can unregulated MFIs’ operating cost be improving in of MFIs, so this result is not driven by entry of new MFIs into
relation to the number of loans, while at the same time it is the for-profit or regulated peer groups.
getting worse in relation to the amount of the loan portfolio? 30. We repeat here our earlier warning that the correlation be-
Both of these can happen because loan sizes in the unregulat- tween loan size and client poverty is very rough, especially
ed MFIs have been dropping. when applied to changes over time in an MFI.

figure 16 figure 17

Cost per Loan by Regulatory Status, Cost per Loan 2004–2011 by Target Market
2004–2011 30
20
High End
18 25
Regulated
16
20
14
Percent

12 15
Percent

10
Broad
8 10
6
Unregulated 5 Low End
4
2 0
2004 2005 2006 2007 2008 2009 2010 2011
0
2004 2005 2006 2007 2008 2009 2010 2011
Note: Operating costs/number of active loans averaged over the year
Note: Operating costs/number of active loans averaged over the year and expressed as % of per capita gross national income.
and expressed as % of per capita gross national income.

16
 Not surprisingly, smaller (and presumably poor-
er) borrowers tend to have less access to deposit figure 19
services from their microlenders. Figure 19 shows
Average Loan Size by Degree of Voluntary
that loan sizes are much higher in institutions that
Savings Mobilization, 2004–2011
offer significant voluntary savings services than in
institutions that offer little or no voluntary savings. 50
What is more, loan size is climbing in the former 45 High Savings Mobilization

but shrinking in the latter.31 40

35
31. A
 lert readers may note that the two findings in this subsec-
30 Low Savings Mobilization
tion (mission drift; savings mobilization) don’t have much to

Percent
do with operating costs, or indeed with any aspect of interest 25
rates. But we thought they were interesting anyway. 20 None

15
10
figure 18 5

Average Loan Size 2004–2011 by Regulated 0

2004 2005 2006 2007 2008 2009 2010 2011
and For-Profit Status
Note: Annual average of loan portfolio divided by annual average of
numbers of active loans, expressed as % of per capita gross national
2004 income, weighted by loan portfolio. “High” means voluntary savings
60 2006
55% >20% of total assets, “low” means <20%, “none” means 0%.
2011
50
45%
40% 41%
40
Percent

34%
30%
30
25%
23% 23%
20 18% 19%
16%

10

0
Nonprofit Profit Nonregulated Regulated
Note: Annual average of loan portfolio divided by annual average
of numbers of active loans, expressed as % of per capita gross
national income, weighted by loan portfolio.

17
 5
Section

Profits

P
rofit is a residual: the difference between in- Notably, the impact of profit on interest rates is
come and expense. In financial institutions, falling. Profit as a percentage of interest income de-
net profit is often measured as a percentage clined fairly steadily from about 20 percent in 2004
of assets employed or as a percentage of the share- to about 10 percent in 2011.
holder’s equity investment.

Level and trend of microlender

Profits in perspective profits
Before looking at level and trend of MFI profits, Profit levels in the industry vary widely (Figure 21).
we first clarify profit’s impact on the borrower. In 2011, about a quarter of microlenders earned an-
Microcredit profits are so controversial that it can nual returns greater than 20 percent on sharehold-
be easy to overestimate how much they affect the ers’ investment. About 5 percent produced profits
interest rates that borrowers pay. Figure 20 shows higher than 40 percent. In 2011, out of a total sam-
how much microcredit interest rates would drop if ple of 879 MFIs, 44 had returns on equity higher
all lenders chose to forgo any return on their own- than 40 percent, and only seven of those were sig-
er’s investment—an extreme supposition indeed. nificant lenders with over 100,000 clients.
The impact of profits is not insignificant, but rates At the other end of the spectrum, plenty of mi-
would still be very high even without them. Of crolenders lost money, especially in Africa and in
course, this figure presents average results: there South Asia (where some lenders working in Andhra
are many microlenders whose profits constitute a Pradesh had a very bad year).
larger percentage of the interest that they charge. Of the various components of interest rates,
profits are the most controversial. Some think that a
microlender has no right to claim it is pursuing a
figure 20
“social” mission if it is extracting profit, or anything
Impact of Profit on Global Interest Rates, 2004–2011 beyond a very modest profit, from its services to
poor clients. Others argue that high profits will en-
35
29.6% courage innovation and faster expansion of servic-
30 28.4%
26.4% 25.8%
27.2%
26.1% 25.5%
26.9% es, and that competition will eventually squeeze out
5.8% 4.9%
25 4.4% 4.0%
3.4% 2.7% 3.2%
2.6% excesses. It is very hard to parlay empirical data
into a quantification of a “reasonable” profit level
Percent

20
19.6%
17.3%
for microcredit, and we will not attempt to do so
15 16.6% 12.4% 12.4%
15.5% 10.3% 9.7% here.32 We limit ourselves to comparing the average
10

5
23.8% 23.5% 22.0% 21.8% 23.8% 23.5% 22.3% 24.3%
0 32. T
 he Social Performance Task Force has tried to address stan-
2004 2005 2006 2007 2008 2009 2010 2011
dards of reasonableness for microfinance profits, but does
Actual Interest Yield minus Net Profit = Breakeven Interest Yield
not seem close to being able to define any quantitative
Profit as Percentage of Interest
benchmarks for evaluating appropriate returns, even for or-
Note: Profit (net income – taxes) is calculated as a % of GLP; all results ganizations that profess to have a “double bottom line.” See,
weighted by GLP. e.g., http://sptf.info/sp-task-force/annual-meetings

18
 figure 21 figure 22

Return on Average Equity 2011, 2011 Profits—MFI vs. Commercial

World and Regions Bank Returns on Average Assets
80
and Equity

60 20
17.8%
40 18
16
20
14
0

Percent
12
Percent

10.2%
–20
10
–40 8
–60 6

–80 4
2.0% 1.69%
2
–100
0
–120 ROAA ROAE
WORLD Africa EAP ECA LAC MENA S. Asia
MFIs Banks
Note: After-tax net profit as % of average shareholders’ equity or non-
profit net worth, unweighted. The thick horizontal bars represent me-
Note: MFI data from MIX. Bank data from BankScope,
dians; the top and bottom of the solid boxes represent the 75th and
including only those countries where MIX MFIs are
25th percentiles, respectively; and the high and low short bars repre-
present. Country-by-country results weighted by MFI
sent the 95th and 5th percentiles, respectively.
GLP.

profitability of microlenders with that of commer-

cial banks (Figure 22). figure 23
When measured against assets, profit is slightly
Global Return on Average Equity, with and
higher on average for microlenders than for banks
without India, 2004–2011
in the same countries. But compared with micro-
lenders, commercial banks have more scope to le- 50
verage their capital structure: that is, they fund 40
India
more of their assets with other people’s money—de- 30
posits and borrowings—rather than with their own 20
equity. As a result, microlenders, despite their high- 10 Global, with India
Percent

er returns on assets, tend to do markedly less well 0

than banks in producing returns on their owners’ –10 Global, without India

equity investments. –20

When we look at overall trends in MFI profit- –30
ability, it is useful to disaggregate India (Figure –40
23), a huge market where some institutions had –50
2004 2005 2006 2007 2008 2009 2010 2011
disastrous years in 2010 and especially 2011, due
to the crisis in Andhra Pradesh. If India is includ- Note: After-tax net profit as % of average shareholder’s equity, weight-
ed by equity.
ed, average profits show a pronounced decline
from 2004 to 2011. If India is excluded, the aver-
age level of profits is much lower, but the rate of
decline is less.

19
 figure 24 figure 25

Return on Equity by For-Profit Status, Profitability of Assets by Market Segments,

2004–2011 2004–2011
25 7

Profit 6
20
5
Low End
15
4
Percent

Percent
Nonprofit
3 Broad
10
2
High End
5
1

0 0
2004 2005 2006 2007 2008 2009 2010 2011 2004 2005 2006 2007 2008 2009 2010 2011

Note: Return on average shareholders’ equity, weighted by equity. Note: Return on average assets, weighted by assets.

International investment funds that funnel Peer group analysis

commercial and quasi-commercial money to mi-
crolenders have not generated impressive re-
Unremarkably, for-profit microlenders produce
sults: annual returns peaked at about 6 percent
higher returns on equity than nonprofit MFIs, ex-
in 2008 but have languished between 2 percent
cept for 2010–2011, when the performance of Indi-
and 3 percent in 2009–2011 (Lützenkirchen
an for-profits dragged the group down (Figure 24).
2012). Returns have been well below what the
More surprisingly (to some, at least), low-end
funds could have earned by investing, for in-
lenders on average have been distinctly more prof-
stance, in commercial banks.
itable than broad-market or high-end lenders, ex-
cept for 2011, when most of the Indian institutions
that took a beating were ones that served low-end
markets (Figure 25).

20
 6
section

Overview and Summary

H
aving broken interest yield into its main
components, we now reassemble them in figure 26
Figure 26, which presents their evolution
Drivers of Interest Yields, as % of Yield, 2004–2011
from 2004 to 2011.33 What happened over the peri-
od, on average, is that 35 2004 2014
Interest Interest
• Operating expenses declined as microlenders 30
Yield Yield

became more efficient, 29.6% 5.8% Profit 2.6% 26.9%

25 3.6%
2.4% Loan Losses
• Financial expenses grew significantly as micro-
Percent
20 5.2% Financial Expense 7.8%
lenders took on more commercial funding,
15 16.8%
14.0%
• Loan losses increased (probably by more than 10 Operating Expense
the unadjusted amount shown here), and
5
• Profits dropped, with the result that 0
2004 2011
• Interest yield dropped by 2.7 percentage points
Note: All data as percentage of average GLP, weighted by GLP.
over the period.
We saw earlier (Figures 3 and 12) that most of the
decline in operating costs and interest yields oc-
curred early in the period. Cost of Funds
Here by way of review are some of the other • Funding costs have climbed substantially as mi-
main conclusions of this paper: crolenders fund more of their portfolio from
commercial borrowing.
Interest Rates
• Funding costs have risen most for microlenders
• MFIs’ nominal interest yield averaged about 27 serving the low-end clientele.
percent in 2011, having declined in 2004–2007,
• So far at least, voluntary savings mobilization
but not in 2007–2011.
has not necessarily lowered funding costs.
• Rates have been rising for microlenders focused
on low-end borrowers. Loan Losses
• Rates have dropped for banks and other regulat- • Two large markets, India and Mexico, have seen
ed microlenders, but risen for NGOs and other sharp rises in bad loans in recent years; but aver-
unregulated microlenders. age loan losses for the rest of the world have
been fairly steady.
33. In both years, the components add up to slightly more than
the interest income from the loan portfolio. The discrepancy • Analytical loan loss adjustments by MIX suggest
is the result of taxes as well as other income not from the loan that the 2011 financial statements of some Indian
portfolio, neither of which are represented among the com-
microlenders may have substantially underesti-
ponents. The discrepancy is bigger in 2011 mainly because
MFIs were earning more nonportfolio income then, from mated their probable loan losses, creating an
investments and from other financial services. overhang that may continue to depress their
profitability in subsequent years.

21
 Operating Expenses • Not surprisingly, low-end microborrowers have
considerably less access to savings services than
• Operating cost is the largest determinant of in-
high-end microborrowers.
terest rate levels.
• The decline of average operating expense (i.e., Profits
improvement in efficiency) has slowed recently,
• The percentage of borrowers’ interest payments
though trends differ by region. Since 2006, cost
that went to microlender profits dropped from
per dollar outstanding has dropped rapidly in
about one-fifth in 2004 to less than one-tenth in
Africa and EAP, but stagnated or risen in the oth-
2011.
er regions.
• Microlenders’ returns on assets average slightly
• It remains to be seen whether the plateau in op-
higher than commercial bank returns, but mi-
erating costs over the past few years will be fol-
crolenders average much lower than commer-
lowed by further declines, or whether this pla-
cial banks in producing returns on shareholders’
teau represents the bottoming out of the learning
investment.
curve effect.
• Microlender returns to shareholders’ equity
• Cost per dollar outstanding is the prevalent mea-
dropped substantially over the period; much but
sure of operating efficiency, but it can be very
not all of this drop is due to severe recent prob-
misleading if used to compare different micro-
lems in the Indian state of Andhra Pradesh.
lenders in terms of management’s effectiveness
at controlling costs. • Low-end markets were substantially more prof-
itable than others during the period, except for
• Average loan size trends do not support a hy-
2011 where low-end microlender profits were
pothesis of mission drift in commercialized mi-
depressed by the Andhra Pradesh crisis.
crolenders: over the period, average loan sizes
dropped much more among for-profit micro-
lenders and regulated microlenders than among
nonprofit and unregulated microlenders.

22
References

CGAP. 2010. “Andhra Pradesh 2010: Global Implications of the Crisis in Indian Microfinance.”
Focus Note 67. Washington, D.C.: CGAP, November. http://www.cgap.org/sites/default/files/
CGAP-Focus-Note-Andhra-Pradesh-2010-Global-Implications-of-the-Crisis-in-Indian-
Microfinance-Nov-2010.pdf
Hoepner, Andreas G. F., Hong Liu, and John O. S. Wilson. 2011. “The Outreach Measurement
Debate in Microfinance: Does Average Loan Size Relate to Client Poverty?” http://papers.
ssrn.com/sol3/papers.cfm?abstract_id=1956569
Lützenkirchen, Cédric. 2012. “Microfinance in Evolution: An Industry between Crisis and
Advancement.” Deutsehe Bank Research, 13 September.
Rosenberg, Richard, Adrian Gonzalez, and Sushma Narain. 2009. “The New Moneylenders:
Are the Poor Being Exploited by High Microcredit Interest Rates?” Occasional Paper 15.
Washington, D.C.: CGAP, February. http://www.cgap.org/sites/default/files/CGAP-
Occasional-Paper-The-New-Moneylenders-Are-the-Poor-Being-Exploited-by-High-
Microcredit-Interest-Rates-Feb-2009.pdf
Schreiner, Mark, Michal Matul, Ewa Pawlak, and Sean Kline. 2006. “Poverty Scorecards: Lessons
from a Microlender in Bosnia-Herzegovina.” http://www.microfinance.com/English/Papers/
Scoring_Poverty_in_BiH_Short.pdf

23
 A NNE X

Data and Methodology

By Scott Gaul

What data did we use? largely determine those charges. Those links are
weakened in lenders that have access to large con-
Data for this analysis were drawn from the MIX Mar- tinuing subsidies.34 This focus, along with data
ket database for the years 2004–2011. Yield data are availability issues, led us to exclude a few large
not widely available before 2004 in the database. In- lenders from the dataset.
stitutions were dropped from the analysis if data were
• BRI. We left Bank Rakyat Indonesia (BRI) out of
not available for all of the indicators used in the analy-
the analysis because it blends microcredit with a
sis, to ensure that differences in indicators are not due
significant portfolio of commercial lending ac-
to differences in the samples for those indicators.
tivity, but does not provide the disaggregated
In total, the dataset consists of 6,043 observa-
revenue and expense data that would be neces-
tions for 2004–2011, each covering 48 variables
sary for the analysis in this paper.
(including descriptive information about the insti-
tution—name, country, legal status). The full data • Harbin Bank. Harbin is a large Chinese bank
set includes any institution that provided data in a with a massive microcredit portfolio (in 2011
given year, subject to some exclusions described Harbin alone had 19 percent of global portfolio
below. Consequently, this dataset reflects both in MIX’s dataset). MIX Market has only two
changes in the market—from the entry and exit of years of data for Harbin Bank. Given the poten-
participants—as well as changes in the voluntary tial distortion of trend data, as well as uncertain-
reporting of data to MIX Market. For summary ty about its activities and mission, we did not in-
statistics, we feel that this dataset still provides an clude Harbin in the final dataset.
accurate read on the relative levels of interest rates
• PSBC. Postal Savings Bank of China (PSBC) is a
in a given market at a given point in time, as well as
large microlender in China. As with Harbin
the changes over time.
Bank, the scale of its activities (GLP of US$14 bil-
In addition, a balanced panel data set is also used
lion in 2011) has a significant influence on global
for some analysis. In the balanced panel, only insti-
figures and any peer groups in which it is includ-
tutions that provide data for all years of the period
ed, but MIX has no data on PSBC before 2010,
are included. Thus, changes in indicators for the
and the data have only a one-star quality ranking.
panel data are due to changes at those institutions,
In addition, the government linkage increases
not changes in the composition of a peer group or
the likelihood of subsidized pricing.
market. The longer the period used for the panel
dataset, the fewer institutions make the cut. We • VBSP. Vietnam Bank for Social Policy (VBSP) is a
chose a five-year panel, covering 2007–2011, which large state bank that receives substantial govern-
let us use 456 institutions. We used the panel data ment subsidies. Interest rates at VBSP are well
mainly as a cross-check against results from the full below what would be needed to cover costs, so
data set.
34. O
 ne problem with large subsidies is that they can substan-
We tried to focus as much as possible on micro- tially distort the operational picture presented by a lender’s
lenders whose mission included financial sustain- financial statements if—as is common—the subsidies are not
ability, because we are exploring links between correctly segregated as nonoperating income. More general-
ly, we wanted this paper to focus mainly on the vast majority
interest charges and the cost components that of MFIs that have to respond to market conditions and costs.

24
 we also dropped it given its influence on global when aggregated. Medians and weighted averages
and regional results.35 are the most frequently used metrics in the paper.
Informally, medians describe the “typical MFI”
We also excluded a few other institutions whose in-
since they report data on the MFI at the 50th per-
terest income, as well as substantial continuing loss-
centile of the distribution. Weighted averages de-
es, strongly suggested a policy of subsidized pricing
scribe something closer to what is “typical” for cli-
and absence of an intent to reach financial sustain-
ents since larger institutions serve more clients and
ability. These institutions are so small that their
also receive more weight in the results. Calculations
treatment does not materially affect our results.
for both match the methods used on MIX Market.
MIX applies a set of standard adjustments to
The data files on which the paper is based can
MFI data.35 By default, data used in the paper are
be found at http://microfinance-business-solution.
unadjusted. Since the adjustments require several
mixmarket.org/rs/microfinance/images/Interest
data points as inputs, the sample for unadjusted
Rate Paper Supporting Data.zip. Most of the data are
data is larger than for adjusted data (the latter cov-
displayed in Excel pivot tables, which make it easy
ering 4,389 observations). In addition, adjusted
to conduct detailed analysis of individual country
data are not disclosed for individual MFIs on the
markets as well as any other peer group of interest.
MIX Market site, while unadjusted data are. Thus,
the analysis from this paper can be largely replicat-
ed by users of the MIX Market site for unadjusted
data. When adjusted data are used in the paper,
Loan Loss Adjustments
they are explicitly referenced as such. MIX’s policy on analytical adjustment of loan loss
Peer groups were calculated from MIX Market provisioning is found at http://www.themix.org/
data based on the definitions below. For each peer sites/default/files/Methodology%20for%20Bench-
group, the count (number of observations), median, marks%20and%20Trendlines.pdf:
minimum, maximum, simple average, and weight- “Finally, we apply standardized policies for loan
ed average are reported. Weighted averages are loss provisioning and write-offs. MFIs vary tremen-
computed using the denominator of the ratio, un- dously in accounting for loan delinquency. Some
less indicated otherwise. For instance, return on count the entire loan balance as overdue the day a
(average) equity is weighted by the average equity payment is missed. Others do not consider a loan
delinquent until its full term has expired. Some
35. F
 or description of MIX’s adjustments, see http://www.
themix.org/sites/default/files/Methodology%20for%20 MFIs write off bad debt within one year of the ini-
Benchmarks%20and%20Trendlines.pdf tial delinquency, while others never write off bad

Definitions of Indicators, Peer Groups, and Loan Loss Provision Adjustments

Indicator Derivation

Average loan size Average gross loan portfolio / average number of active loans

Cost of funds Financial expense / liabilities

Cost per loan Operating cost / average number of active loans

Gross loan portfolio Total outstanding balance on all active loans

Interest yield (nominal) All interest and fee revenue from loans / average gross loan portfolio

Interest yield (real) Nominal interest yield adjusted for inflation

Loan loss expense Net annual provision expense for loan impairment / average gross loan portfolio

Operating expense ratio Total operating (i.e., personnel and administrative) expense / average gross loan portfolio

Return on average assets (Net operating income - taxes) / average assets

Return on average equity (Net operating income - taxes) / average equity

25
 loans, thus carrying forward a defaulted loan that ing purposes only. We do not recommend that all
they have little chance of ever recovering. MFIs use exactly the same policies.) In most cases,
“We classify as ‘at risk’ any loan with a payment these adjustments are a rough approximation of
over 90 days late. We provision 50 percent of the risk. They are intended only to create an even play-
outstanding balance for loans between 90 and 180 ing field, at the most minimal of levels, for cross in-
days late, and 100 percent for loans over 180 days stitutional comparison and benchmarking. Never-
late. Some institutions also renegotiate (refinance theless, most participating MFIs have high-quality
or reschedule) delinquent loans. As these loans loan portfolios, so loan loss provision expense is not
present a higher probability of default, we provision an important contributor to their overall cost struc-
all renegotiated balances at 50 percent. Whereever ture. If we felt that a program did not fairly repre-
we have adequate information, we adjust to assure sent its general level of delinquency, and we were
that all loans are fully written off within one year of unable to adjust it accordingly, we would simply
their becoming delinquent. (Note: We apply these exclude it from the peer group.”
provisioning and write-off policies for benchmark-

MIX Peer Groups

Group Categories Criteria

Age New 1 to 4 years

Young 5 to 8 years
Mature More than 8 years

Charter Type Bank

Credit Union
NBFI
NGO
Rural Bank

Financial Intermediation (FI) Non FI No voluntary savings

Low FI Voluntary savings <20% of total assets
High FI Voluntary savings >20% of total assets

Lending Methodology Individual

Solidarity Group
Individual/Solidarity
Village Banking

Outreach Large Number of borrowers > 30,000

Medium Number of borrowers 10,000 to 30,000
Small Number of borrowers < 10,000

Profit Status Profit Registered as a for-profit institution

Not for Profit Registered in a nonprofit status

Region Africa Sub-Saharan Africa

Asia South Asia, East Asia and the Pacific
ECA Eastern Europe and Central Asia
LAC Latin America and Caribbean
MENA Middle East and North Africa

Scale (Gross Loan Portfolio Large Africa, Asia, ECA, MENA: >8 million; LAC: >15 million
in USD) Medium Africa, Asia, ECA, MENA: 2 million–8 million; LAC: 4 million–15 million
Small Africa, Asia, ECA, MENA: <2 million; LAC: <4 million

Sustainability Non-FSS Financial Self-sufficiency <100%

FSS Financial Self-sufficiency =100%

Target Market (Depth = Avg. Low end Depth <20% OR average loan size (<USD 150)
Loan Balance per Borrower/ Broad Depth between 20% and 149%
GNI per Capita) High end Depth between 150% and 250%
Small business Depth over 250%

26