Monetary theory and electronic money: reflections on the Kenyan experience.
Jack, William ; Suri, Tavneet ; Townsend, Robert 等
In 2007, the leading cell phone company in Kenya, Safaricom Ltd.,
launched M-PESA, a short message service (SMS)-based money transfer
system that allows individuals to deposit, send, and withdraw funds from
a virtual account on their cell phones and that is separate from the
banking system. M-PESA has grown rapidly, currently reaching more than
seven million users, approximately 38 percent of Kenya's adult
population, and it is widely viewed as a success story to be emulated
across the developing world. Indeed, similar products have recently been
launched in a growing number of countries across Africa, Asia, and Latin
America, with the intent of expanding financial services to previously
unreached populations. (1)
M-PESA is used not only for remittance purposes, but also to save,
to purchase pre-paid phone credit and other goods and services, to pay
bills, and to execute bank account transactions. However, consumers do
not need bank accounts in order to use M-PESA, and Jack and Suri (2009)
found it was used by more than half of the unbanked in their sample. It
is used by a broad cross section of Kenyan society, but has increasingly
been adopted by those at the lower end of the income distribution, as is
evidenced by the steady reduction in the average transaction size since
its inception. A large part of M-PESA's success is attributed to
the broad and dense network of over 16,000 agents across Kenya, which
provides the retail interface with consumers.
In this article, we examine the role of monetary theory in
understanding this new generation of mobile banking products, especially
those that, like M-PESA, do not simply provide electronic access to
existing bank accounts. Deposits of money in a mobile phone-based
account reflect holdings by the account owner of a commodity we refer to
as e-money. Because e-money can be easily transferred from one
individual to another, as long as it is expected to retain its value, it
can be used in equilibrium as a means of exchange, as well as to
transfer purchasing power between individuals.
Indeed, at the time of the launch of M-PESA, the service was seen
as a means of overcoming the high transactions costs associated with
sending cash remittances that faced the 80 percent of individuals in the
economy without bank accounts. (2) But since then, e-money has been
increasingly used both as a store of value and as a means of exchange,
with users able to pay utility bills, make loan repayments, and even pay
for taxi rides with it. The co-existence of essentially two forms of
cash, even if closely related and linked, raises certain theoretical
modeling issues in itself. But when one form of cash is issued by a
profit-maximizing entity and the other by the central bank, further
issues of competition, regulation, and coordination naturally emerge.
Although mobile banking is in its infancy in the United States,
payroll cards have provided a similar payment function, albeit without
the geographic reach of mobile phone communication. Funds are typically
deposited by an employer into the account of an employee, who can either
withdraw cash at an automated teller machine (ATM) or use the card to
make purchases at stores possessing debit card machines. As in the case
of mobile banking, payroll card users do not need a bank account, and
transactions are executed using an existing communication network. In
addition, payroll cards in the United States are generally cheaper than
check-cashing services and money orders, just as M-PESA in Kenya is
cheaper than most alternatives. Foster et al. (2010) describe the use of
various payments systems in the United States--they find that 93.4
percent of consumers in the United States have adopted a payment card,
but only 17.2 percent have a prepaid card.
This article presents a first look at how existing models of
monetary theory can be used to think about the impact of mobile banking
on the operations of the financial system and the implications for
monetary and regulatory policy decisions that face the central bank. We
are not yet in a position to develop a fully articulated model of mobile
banking, but we hope this discussion will be a first step in this
process. In addition, this article is not an exhaustive discussion of
all models of money, but more of a focus on a subset of models that have
different implications for the role of e-money in an economy.
Most theoretical models of money and credit include both a temporal
dimension and some kind of generalized locational heterogeneity.
Sequential trades over time require promises to be made (and kept) and
records to be maintained. On the other hand, spatial separation can mean
that it is not always possible for two parties to a trade to meet each
other at the right time, so more complicated multilateral chains of
individuals are required to effect the desired net trades.
In these environments, financial instruments such as fiat money and
private debt can sometimes improve the efficiency of resource
allocations by facilitating intertemporal and interspatial trades.
However, equilibrium allocations may continue to be inefficient without
the intervention of either a public institution (such as a central bank)
or a well-regulated private agent (such as a clearinghouse).
Mobile banking has the potential to effectively reduce the
distances that separate individuals, both literally and figuratively,
thereby lessening the frictions that characterize models of incomplete
intermediation, relaxing liquidity constraints, and reducing the need
for monetary interventions. On the other hand, new liquidity constraints
could arise that are binding for individuals who trade with the new
financial instrument, e-money.
The article proceeds as follows: Section 1, which draws heavily on
Jack and Suri (2009), provides background on the recent evolution of
mobile technology and mobile banking in Kenya and on the practical
operational features of M-PESA. Section 2 reviews a number of strands of
the literature and discusses the specific lessons that we might learn
regarding both the equilibrium impact of mobile banking and its
implications for policy. Section 3 presents some empirical facts from a
survey on M-PESA customers and agents that provide some insights into
the implications from the models and lessons in Section 2. Section 4
concludes.
1. BACKGROUND ON M-PESA
Mobile Money in Kenya: An Introduction
Mobile phone technology has reduced communication costs in many
parts of the developing world from prohibitive levels to amounts that
are, in comparison, virtually trivial. Nowhere has this transformation
been as acute as in sub-Saharan Africa, where networks of both fixed
line communication and physical transportation infrastructure are often
inadequate, unreliable, and dilapidated. While mobile phone calling
rates remain high by world standards, the technology has allowed
millions of Africans to leap-frog the landline en route to 21st century
connectivity. As the number of landlines in Kenya fell from about
300,000 in 1999 to around 250,000 by 2008, mobile phone subscriptions
increased from virtually zero to nearly 17 million over the same time
period (Figure 1). Assuming an individual has at most one cell phone,
(3) 47 percent of the population, or fully 83 percent of the population
15 years and older, have access to mobile phone technology. In March
2007, following a donor-funded pilot project, Safaricom launched a new
mobile phone-based payment and money transfer service, known as M-PESA.
(4) The service allows users to deposit money into accounts linked to
their cell phones, to send balances using SMS technology to other users
(including sellers of goods and services), and to redeem deposits for
regular money. Charges, deducted from users' accounts, are levied
when e-money is sent and when cash is withdrawn. (5)
[FIGURE 1 OMITTED]
In particular, Safaricom accepts deposits of cash from customers
with a Safaricom cell phone SIM (subscriber identity module) card and
who have registered with Safaricom as M-PESA users. Registration is
simple, requiring an official form of identification (typically the
national ID card held by all Kenyans, or a passport) but none of the
other validation documents that are typically necessary when a bank
account is opened. Formally, in exchange for cash deposits, Safaricom
issues a commodity known as "e-money," measured in the same
units as money (denominated in shillings), which is held in an account
under the user's name. This account is operated and managed by
M-PESA and records the quantity of e-money owned by a customer at a
given time. There is no charge to a customer for depositing funds into
his account, but a sliding tariff is levied on withdrawals from M-PESA
accounts (for example, the cost of withdrawing $100 is about $1). (6) An
M-PESA user who sends e-money is charged a flat fee of about 40 U.S.
cents if sending to another registered user, and a sliding fee if
sending to a phone number that is not registered for M-PESA. (7) Figure
2 illustrates the schedules of total net tariffs for sending money by
M-PESA, including the cost of withdrawing the funds incurred by the
recipient, and compares these with two other money transfer
services--Western Union and Postapay (operated by the Post Office). The
M-PESA tariffs shown include both the sending and withdrawal fees and
are differentiated according to receipt by registered and nonregistered
user. Fees are charged to the user's account, from which e-money is
deducted. Additional cash fees are officially not permitted, but there
is evidence that they are sometimes charged on an informal basis by
agents. We return to this issue below.
[FIGURE 2 OMITTED]
E-money can be transferred from one customer's M-PESA account
to another's using SMS technology or sold back to Safaricom in
exchange for money. Originally, transfers of e-money sent from one user
to another were expected primarily to reflect unrequited, internal,
within-country remittances, but nowadays, while remittances are still an
important use of M-PESA, e-money transfers are often used to pay
directly for goods and services, from school fees to the wages of
domestic staff. (8)
The Growth of Mobile Money
M-PESA has spread quickly and has become one of the most successful
mobile phone-based financial services in the world (9) The average
number of people opening up an M-PESA account (i.e., new registrations)
per day exceeded 5,000 in August 2007 and reached nearly 10,000 in
December that year (see Figure 3). By August 2009, 7.7 million M-PESA
accounts had been registered. Ignoring multiple accounts and those held
by foreigners, this means 38 percent of the adult population of Kenya
had gained access to M-PESA in just over two years.
[FIGURE 3 OMITTED]
Since the launch of M-PESA, wary of regulation by the Central Bank
of Kenya, Safaricom has been at pains to stress that M-PESA is not a
bank. However, the ubiquity of the cell phone across both urban and
rural parts of the country, and the lack of penetration of regular
banking services, (10) led to hopes that M-PESA accounts could
substitute for bank accounts and reach the unbanked population. Data
reported in Jack and Suri (2009) suggest this is partially true,
although M-PESA has been adopted by both the banked and unbanked in
roughly equal proportions (11) In addition, more recently, M-PESA users
have been able to withdraw funds from their M-PESA accounts at ATMs
operated by one of the commercial banks (Equity Bank) and some banks
have begun to use M-PESA as their mobile banking platform. However,
deposits cannot (yet) be made at ATMs, and the network of ATMs and bank
branches, while growing, remain limited: In the long run they could
replace agents, but both capital costs and the costs of security,
operation, and maintenance suggest agents will continue to play an
important role for some time. (12)
The average size of M-PESA transactions has fallen over time as it
has reached more of the population and has been used more extensively,
as shown in Figure 4. In the two years following its introduction, the
average transaction size fell about 30 percent, having started at KShs
3,300 (about $50). Most of this decline has probably been because of the
expansion of take-up among the poorer individuals and households.
[FIGURE 4 OMITTED]
Table 1 shows the various types of transactions for which M-PESA is
used, which include not just sending and receiving money, but also
storing or saving money, purchasing airtime (the prepaid credit used for
voice and text communications), and paying bills.
Table 1 What Do Individuals Use M-PESA For?
Fraction of Sample
(Based on Multiple Responses)
Receive Money 28.40%
Send Money 25.08%
Store/Save Money for 14.39%
Everyday Use
Buy Airtime for Myself 13.58%
Buy Airtime for Someone Else 8.30%
Store/Save Money for Emergencies 6.69%
Store/Save Money for 0.27%
Unusually Large Purchases
Pay Bills 1.35%
Receive Money for a 0.77%
Bill/Else Pay Bills
Notes: Each entry is the share of registered M-PESA users in our
sample who reported the corresponding function to be the most
commonly used. The bill payment service had only just started at
the time of the survey and has since become rather popular.
While the sustained growth in M-PESA registrations is notable, the
volume of financial transactions mediated through M-PESA should not be
exaggerated. Table 2 reports that the volume of transactions effected
between banks under the RTGS (Real Time Gross Settlement) method is
nearly 700 times the daily value transacted through M-PESA; and, maybe
more relevant, the daily value transacted through the check system
(automated clearinghouse, or ACH) is about 85 times the daily value
transacted through M-PESA. Related, the average mobile transaction is
about 100 times smaller than the average check transaction (ACH) and
just half the size of the average ATM transaction. (13) M-PESA is not
designed to replace all payment mechanisms, but has effectively filled a
niche in the market.
Table 2 Daily Financial Transactions, Oct. 2007-Sept. 2008
RTGS ACH ATM Mobile
Value Per Day (billion KShs) 66.3 8.5 1.0 0.1
Transactions Per Day (thousands) 1.0 39.2 180.2 107.2
Value Per Transaction (million KShs) 64.67 0.216 0.006 0.003
Notes: KShs = Kenyan Shillings.
Source: Central Bank of Kenya (2009).
The Agent Network
To facilitate purchases and sales of e-money, and in light of low
rates of bank account coverage among a widely dispersed population,
M-PESA maintains and operates an extensive network of more than 16,000
agents across Kenya. These agents are like small bank branches, often
manned by a single person. As can be seen in Figure 5, the growth of
this network lagged behind that of the customer base for the first year
of M-PESA's operation, during which time the number of users per
agent increased five-fold from a low of 200 to a high of 1,000. But
since mid-2008, agent growth has accelerated and the number of users per
agent has fallen back to about 600.
[FIGURE 5 OMITTED]
Registered M-PESA users can make deposits and withdrawals of cash
(i.e., make purchases and sales of e-money) with the agents, who receive
a commission on a sliding scale for both deposits and withdrawals. (14)
Clearly, withdrawals of cash can only be effected if the agent has
sufficient funds. But symmetrically, cash deposits can only be made if
the agent has sufficient e-money balances on his/her phone. Agents face
a nontrivial inventory management problem, having to predict the time
profile of net e-money needs. Figure 6 shows a representation of the
flow of money and e-money among individuals, Safaricom, commercial
banks, and the central bank, and illustrates the core workings of
M-PESA. The role of what we call the "coordinator," which in
practice is a head office, an aggregator, or a super agent, is described
in more detail below.
[FIGURE 6 OMITTED]
The network of commercial bank branches across Kenya, while
growing, remains much smaller. As of November 2009, the Central Bank of
Kenya (15) reports that 44 commercial banks had 849 branches across
Kenya (about one branch for every 40,000 Kenyans), with 50 percent of
branches concentrated in the largest (by size of branch network) four
banks. As of 2008 in the United States, there were 7,086 institutions
with 82,547 branches that came under Federal Deposit Insurance
Corporation protection, yielding a density of bank branches about 10
times that in Kenya (whose population is about 10 percent of the United
States).
In practice, M-PESA agents are organized into groups. Originally,
M-PESA required that agent groups operate in at least three different
locations, so that the probability of cash or e-money shortfalls could
be minimized. This diversification within the group would only be
effective, of course, if the inventories of money and e-money were
efficiently re-allocated across agents in the group accordingly. There
are now three agent models in operation, in which there is a central
body that manages and coordinates the operations of a group of
subsidiary agents These models are differentiated with regard to the
formal status of the coordinating body and the ownership structure of
the group, and whether the central body conducts direct transactions
with individual users, as shown in Figure 7.
[FIGURE 7 OMITTED]
In the first model, one member of the agent group is designated as
the "head office," which deals directly with Safaricom, while
subsidiary agents that are owned by the head office manage cash and
e-money balances through transactions with the head office. (16) Both
the head office and the agents can transact directly with M-PESA users.
The second model is the aggregator model, with the aggregator acting as
a head office, dealing directly with Safaricom, and managing the cash
and e-money balances of agents. However, the agents can be independently
owned entities with which the aggregator has a contractual relationship.
A final and much more recent model (17) allows a bank branch, referred
to as a "super agent," to make cash and e-money transactions
with agents on an ad hoc basis. However, the bank does not trade e-money
with M-PESA customers. The super agent model is one example of the
integration of M-PESA services into the banking system. Other
developments in this vein include the ability to transfer funds, often
via ATMs, between a user's M-PESA account and accounts at certain
commercial banks with which M-PESA has forged partnerships. But even as
M-PESA has facilitated transactions for the approximately 72 percent of
user households in Jack and Suri's sample with bank accounts, it
remains popular with the un -banked, of whom more than half (54 percent)
used M-PESA. (18)
The cash collected by M-PESA agents in exchange for sales of
e-money is either kept on the premises or deposited in the agent's
(or head office's) bank account. When they wish to replenish their
e-money balances, agents transfer money via the banking system to one of
two bank accounts held by Safaricom. Safaricom is required to limit the
quantity of e-money it issues to the amount of money it receives from
agents--that is, e-money is 100 percent backed by deposits in commercial
banks. However, these deposits are subject only to the regular 6 percent
Kenyan Central Bank reserve requirement.
2. MODELS OF MONEY AND MEANS OF PAYMENT WITH SPATIAL SEPARATION
M-PESA's rapid expansion means that a large share of the
Kenyan population now conducts at least some of their financial
transactions by phone. In this section we discuss the implications of
this new kind of payment system for the management of the financial
system as a whole and of central bank regulatory and monetary policies
in particular. To address these questions, we describe in some detail a
number of models of money, the payment system, and clearing and
settlement. The purpose is to focus on the features of the models that
can provide insights into the operational design of mobile banking and
inform policy choices facing regulators and monetary authorities.
Therefore, we follow the summary of each model with a discussion of its
implications for mobile banking. We proceed incrementally, beginning
with simple but surprisingly rich models of money, then progressively
review more complex models that we believe reflect particular features
of the Kenyan financial environment.
Townsend Model of Financial Deepening and Growth
This model focuses directly on improvements in the technology of
communication and links the degree of financial interconnectedness of
agents with the level of economic development in a cross section and
also over time. The idea is that as connectedness increases, with
electronic payments connecting otherwise spatially separated agents,
there is an increase in the specialization of labor, an increase in the
consumption of market-produced goods, and a shift toward e-money
relative to fiat money. This is the story of how financial deepening and
growth are intertwined and how M-PESA could help Kenya increase gross
domestic product over time at the same time as it increases monetized
exchange.
Each household of type i can produce (by supplying labor) only good
i, and each has a utility function over its own consumption of good i
and a good it cannot produce, i + 1, as well as leisure. When households
are in autarky, without physical or electronic contact, no trade is
possible, so each household consumes all its production of good i only.
In this situation, there is no need for a means of payment. In contrast,
with some travel, as in the Cass and Yaari (1966) or Lucas (1980)
versions of the Wicksell (1935) triangle applied many times, household i
can only trade either with household i + 1 (whose good i values) or with
household i - 1 (who values good i). But because of the structure of
preferences (e.g., because household i + 1 does not value good i and
instead wants goods i + 1 and i + 2), narrow bilateral exchange between
i and i + 1 is in no one's self interest. This is the key lack of
double coincidence of wants. Decentralized trade would give rise to
autarky if it were not for valued fiat money. (19)
The timing-location is shown in Figure 8 where, in any given
period, household i has two members, a shopper and a seller, who can
only move horizontally to trade with households i + 1 and i - 1,
respectively. Between time periods, members of household i, i even,
shift down one line, and households i, i odd, stay put. Thus, debt
issued by a household of type i, i even, can only be passed along to a
household vertically above the issuer and so has no value. Only fiat
money is used and it can have value. Specifically, one member of each
household i travels to the market with i + 1 and purchases some of good
i + 1 at price pi + 1 with fiat money acquired previously; a second
member travels to the market with i - 1 and sells some good i for money
at price [p.sub.i] * Note that it takes one period for goods produced
and sold to come back via money holding in the interim as goods
purchased. With constant prices across time and space and with a
positive intertemporal discount rate, this makes it less beneficial to
supply labor. This is a crucial aspect of this and other related models
below.
[FIGURE 8 OMITTED]
In a Walrasian, centralized exchange regime with electronic debits
and credits, households can now hold intraperiod debt for within-period
purchases and, at the same time, send and receive electronic credits. At
the end of the period, accounts are cleared. Intuitively, when one
member of household i travels to market (i, i + 1) to buy good i + 1
from household i + 1, it is as if that member were using a credit card
(or phone) linked electronically to a central account, which will not be
paid until the end of the period. The second member of household i who
travels to market (i, i - 1) and sells good i is paid with a credit card
from household i - 1. At the end of the period, these electronic debits
and credits are cleared and accounts must balance (we return to
interperiod debt in the Lacker model below). Note that goods produced
and sold can be transformed in this way to goods purchased within the
same period, so there is no inefficiency associated with holding idle
money balances. In fact, in the equilibrium of this electronic
accounting system, fiat money plays no role and its price is zero. The
prices of goods themselves are in some (arbitrary) unit of account.
Related, though households remain separated in space, it is as if they
are transacting with one another in a centralized market that ignores
the spatial segmentation as far as prices and values are concerned.
However, this system works only if households are allowed to overdraft
their electronic accounts and there is enough commitment or punishment
to make sure they honor debts accrued within the period.
In summary, if we then assume that substitution effects dominate
income effects and focus on prices, the cost of consumption of the
nonproduced good in terms of labor is infinite in autarky and high in
the fiat money regime relative to the centralized Walrasian electronic
clearing e-regime. Moving from autarky to the decentralized money regime
and then to the centralized Walrasian regime, the model predicts that
labor supply increases, output of the produced commodity rises,
consumption of the nonproduced good rises, consumption of the produced
good drops, trade volume increases, and welfare increases. If an economy
has a mix of decentralized and centralized regimes, as with some
fraction of "lines" (see Figure 8) using fiat money and other
"lines" using Walrasian credit, and these fractions vary
across countries, then per capita national income rises as financial
interconnectedness increases, fiat money decreases, and per capita
private debt increases, but the ratio of fiat money to income decreases
and the ratio of credit to income increases. This pattern tends to be
what we see in cross-sectional data. Similar comparisons are valid for
an economy that is becoming more financially integrated over time, like
Kenya, where forward-looking households in the fiat currency part of the
economy treat financial integration into the Walrasian e-system as an
exogenous random event that happens with positive probability
(essentially changing the discount rate). Note, however, that thus far,
in this particular model, no household needs to use multiple means of
payment.
Implications for Mobile Banking
What are the implications of this kind of model of financial
deepening for a system like M-PESA? It is clear that M-PESA will change
the financial connectedness of the individuals in the economy, which in
the model above will cause higher economic development. Therefore, the
main takeaway from this model is that M-PESA can be viewed as a
technological innovation that lowers trading costs or, better put,
allows financial transfers (credits and debits) across agents who are
still separated in space. This improves welfare, at least in the model
economy without government and no vested interests in the current
intermediation system (and without other heterogeneity). Fiat money and
electronic payments can co-exist if some households have access to
M-PESA and some do not. However, in the model, but perhaps not in the
M-PESA system, the household buying goods in effect creates a net
increase in e-money within the period. If e-money were essentially only
a debit card, then an initial deposit of currency would have to underlie
the debit transaction, undercutting this key advantage. In other words,
the theory argues that we might see features of net credit creation in
the functioning of the actual M-PESA system, though perhaps at an
aggregated or agent level and not necessarily at the level of individual
households. However, for this feature to exist there must be a (harsh)
means of preventing reneging or default so that accounts actually clear
at the end. Even that requires foresight of the overall equilibrium,
e.g., here the shopper knows the prices at which the seller is receiving
credits. Again, we come back to this mismatch and interperiod carryovers
in the other models below.
Manuelli and Sargent Turnpike Model with Currency and Debt
A closely related model of Manuelli and Sargent (2009) rationalizes
the coexistence of fiat money and private credit. As in Townsend's
turnpike models, agents meet in pairs and, while they have long enough
relationships to undertake some efficiency-enhancing intertemporal
trades via the extension of private credit, they do not stay together
long enough to effect fully Pareto-efficient allocations. More
specifically, time is divided into periods (think of these as
"years"), each composed of four subintervals (e.g.,
"seasons"). Individuals meet for just half a year only, i.e.,
two consecutive subintervals, and then move on--some to the east, some
to the west (see Figure 9). In the first subinterval of a half-year, one
person in a given pair has a positive endowment of the single perishable
consumption good and the other has none, and in the second subinterval
these roles are reversed, giving rise to short-term (two-subinterval)
private credit arrangements. However, the positive endowments in each
subinterval can be either high or low (for example, a > 0, b > 0,
and a/b > 1), while aggregate output in each half-year (a + b) is
constant, and each individual's annual aggregate endowment is
constant, also equal to (a + b). Because agents remain together for only
two subintervals (one half-year), they cannot implement trades across
half-years--that is, they cannot issue long-term debt. Fiat money plays
a role in facilitating the trades that such debt would effect. Manuelli
and Sargent generalize this to include labor supply, so that output is
endogenous.
[FIGURE 9 OMITTED]
One interpretation of Manuelli and Sargent's model is as a
generalization of Townsend's original turnpike in which endowments
fluctuated with a periodicity of two and meetings lasted only one
period. Instead of meeting for two periods, we can interpret Manuelli
and Sargent as a model where households continue their travels after one
period but remain linked electronically for two periods (though we
ignore the requisite costly shipping of goods in the second period--some
of the models below are more complicated so as to eliminate this flaw in
our attempted interpretation). As the time and spatial limitations of
communication fall (e.g., with the expansion of the network of M-PESA
agents, accounts, and the use of cell phones), debts of increasingly
long maturity can, in principle, be issued and repaid.
Implications for Mobile Banking and Monetary Policy
To the extent that mobile banking facilitates the operation of the
private (often informal) credit market, a model that accommodates such
products with nontrivial implications for policy can be informative. To
start, as in the Townsend models, the laissez faire, non-interventionist
monetary equilibrium (without debt) is not Pareto optimal. Essentially,
the wedge that we discussed in the earlier Cass-Yaari model, where money
is earned through production and held without interest for one period,
can be eliminated with intervention by paying interest on cash balances.
This equates intertemporal substitution in consumption to the natural
rate of time discount and ensures that no household hits a binding
corner, running out of cash.
But the impact of monetary policy interventions in the form of
changes to base money depends on whether private credit is allowed or,
under the interpretation here, whether e-money that allows borrowing and
lending is in the system. An increase in the growth rate of the money
supply has ambiguous effects on the average level of output but
increases the volatility of output when there are no restrictions on
private borrowing and lending. However, in economies where individuals
do not have access to private loan markets, say because they move on
without cell phones, the results are quite different: An increase in the
rate of money growth decreases mean output and has no effect on
volatility of output (which remains zero). Likewise, if the economy is
liberalized, or otherwise experiences a surprise innovation that allows
private borrowing and lending, then prices increase and output becomes
more volatile. Financial innovation is welfare-improving but intimately
connected with the impact variables that central banks typically monitor
or attempt to control.
As Manuelli and Sargent (2009) emphasize, the potential
destabilizing effects of actual financial liberalizations are
highlighted in both the academic and policy literatures. More generally,
the effects of monetary policy depend on the way private credit markets
are operating, even if in the process of borrowing and lending there is
no net creation of e-money. Thus, when formulating monetary policy, the
central bank will need to take into account the effective change in
financial regimes that M-PESA has brought with it. Indeed, in the above
class of models, optimal monetary policy in terms of control over fiat
base money is still relatively straightforward but not without interest.
Specifically, the allocation achieved under optimal policy differs from
the one associated with the corresponding economy with no locational
restrictions and centralized trades permitted at time zero. While both
allocations are Pareto optimal, they are not the same, implying that
efficient monetary policy has redistributive consequences. Further,
optimal government-issued currency continues to play an essential role
even when interest is optimally paid on holdings of such currency. And,
the interest-on-currency policy does not work in a way that can be
replicated by free banking in a Walrasian world. Related, moving from a
suboptimal policy to one with interest on currency may redistribute
income and not be Pareto improving. In this model, unlike the first,
e-money does not drive out fiat money nor the need for an optimal
monetary policy. This is reminiscent of a class of related models of
monetary management in which implementation of policy depends on the
ability of agents to trade in asset markets. (20) Financial market
segmentation relies on costs that may be arguably decreasing.
Townsend's Models of Activist Monetary Policy and Money as a
Communication Device
A generalization of Manuelli and Sargent would allow credit
arrangements to be used to implement trade among individuals who remain
in their home location and deal with each other repeatedly over time.
This is also similar to the Walrasian accounting system of the first
model above, except that here again the trade is intertemporal, with
borrowing and lending over time, so that any individual's balance
does not have to net to zero at the end of each period. In the next set
of models, credit is identifiable as direct communication and promises.
Fiat money then co-exists with credit and serves as a communication
device for dealing with strangers across locations. (21) But the models
here feature Diamond and Dybvig (1983)-style preference shocks with
patient and urgent households generating the desired intertemporal
trade. Moreover, the models here deliver welfare gains from an activist
monetary authority responding to shocks and managing liquidity needs.
More generally, the quantity and kinds of money in the system are
determined optimally in an effort to compensate for missing credit and
insurance markets. In this way, one can build on the platform of
e-transfers to create a highly effective recordkeeping system in which
electronic accounts allow for a rich variety of financial instruments.
Townsend's (1989) model envisions a scenario where there are N
islands each with N inhabitants (the case of N = 3 is shown in Figure 10
but, more generally, N is a large number) (22) Preference shocks that
are correlated among a segment of each island's residents occur in
the first period. That is, some fraction of the residents are patient,
in principle willing to lend, and the residual fraction are urgent,
wanting to borrow. However, a share (1 - [lambda]) of the population of
each island moves, spreading out across all the other islands in such a
way that no mover encounters anyone from his home island at his new
destination. This creates a problem if recordkeeping is limited to
locations, that is, if there is no cross-island communication or
accounting system so that only nonmovers can borrow and lend: Promises
involving movers (either among themselves but going to different
locations or between them and nonmovers), on the other hand, are not
credible as they cannot be consummated at a later date.
[FIGURE 10 OMITTED]
As movers are effectively excluded from the credit market, a social
planner could attempt to implement efficient intertemporal consumption
profiles by asking movers at each date to report their preferences,
allocating consumption accordingly. But if the information reported
cannot be credibly transmitted to other islands without a recordkeeping
device, then the only incentive-compatible mechanism is one that gives
all movers the same level of consumption in both periods, independent of
their preferences. Portable fiat money allocated to movers, and
monotonically related to their first period announcements, can
facilitate the transmission of information across time and space to the
strangers they meet at their destinations. In this interpretation, fiat
money is a portable token. By allowing side trades between individuals,
monotonicity can be strengthened to linearity, delivering a price of
fiat money or tokens for goods. Of course, the initial nominal price
level remains arbitrary, as that is simply a matter of the denomination
of the unit of account.
However, if additional periods are added to the model (e.g.,
another round of movers), future movers must also be allocated fiat
money in order to engage in intertemporal trade. The purchasing power of
each unit of money allocated to second-round movers must, for efficiency
reasons, be the same as that offered to first-round movers, but the
quantity will be increasing in the number of movers and the proportion
who are patient. (23) As the preferences of new generations of consumers
are revealed, planned consumption levels supported by allocations of
fiat money in early periods may be revised. Since the purchasing power
as previously explained is constant across early movers in a given
period, the associated adjustment to consumption levels is effected
through changes in the price level. That is, inflation eats up the
purchasing power of first-round movers if it is judged that they should
be getting less given what new information tells the monetary authority
about the way they and second-round movers should be treated. Note that
this activist policy is quite different from, say, a Friedman rule, as
described in the earlier class of models above in which a constant rate
of deflation can remove the distorting wedge. This is not enough here.
Optimal policy is state contingent (Manuelli and Sargent [2009]
anticipate such results in the concluding section of their article).
Note, however, that fiat money as tokens conveys only the
information that a household has been patient in the past, not that the
household has been a first- or second-round mover. If even more
information, such as the dates and the nature of past transactions, was
encoded in the system, then the distinction between local and
inter-island accounts could disappear. That is, one can imagine one kind
of fiat money--e.g., red tokens for first-round movers, green tokens for
second-round movers, and electronic accounts for those who stay home.
Indeed, accounts that distinguish all these space/time transactions
could be accomplished with the electronic recordkeeping that mobile
technologies and markets allow, at least in principle. Indeed, with all
of that, we could in theory go further and here again completely mimic
the outcome of a perfect Walrasian accounting system in which changing
locations per se has no consequences. The fraction of agents leaving an
island would be exactly the same as the fraction arriving and,
financially speaking, there would be no strangers.
Townsend (1987) generalizes this idea of multiple monies (or
differentiated e-accounts) in a similar framework with four agents,
spatial separation, and private information on preference shocks. In
particular, suppose there are two islands with two individuals each, as
illustrated in Figure 11. In period 1, agents a and b live on the left
island and agents a' and V live on the right island. In period 2, b
and b' switch places, while a and a' (who are subject to
shocks) remain on their home islands. Agents b and V are risk neutral
and in principle can provide insurance to agents a and a', who are
risk averse. With one good, preference shocks determine not only the
degree of risk aversion but also relative patience. With two goods,
there can be preference shocks for each good over time (e.g., patient
for good one and urgent for good two) and an overall intertemporal shock
determining utility in period 1 versus period 2.
[FIGURE 11 OMITTED]
Townsend then examines the properties of trade facilitated by
alternative communication devices in this environment, both for the
cases of a single good as well as for multiple goods. First, oral
communication can take place only between agents in the same location
and so cannot be used to convey credible information across time to
strangers (if agents cannot carry tokens, commodities, or messages). The
equilibrium is thus Pareto inefficient. On the other hand, tokens
(money) that are appropriately distributed in period 1 can be used to
verify information in period 2, helping with incentives to reveal
information correctly and acting again as a technology for storing that
information. The previous model provides intuition for the case of one
good. However, with two goods, one type of token may not be enough.
Intuitively, one wants to convey the full history of shocks for each
good in the first period, yet ensure incentive compatibility in the
second when agents can turn out to be very desirous of consumption
overall. For example, one type of token, say green, is handed out in
period 1 given a certain realization of preference shocks, while the red
token is handed out given another realization of these shocks again in
period 1 (alternatively, these are different "credits" in
different cell accounts). Then in the second period, the agent is
required to show not just the correct number of tokens, but also the
correct colored token (or have the requisite balances in a specific cell
account). Indeed, much can be done even with n-commodities and
m-combinations of shocks using combinations of red and green tokens (two
types of e-money) as an encryption system. The point more generally is
that multiple monies are used to convey the history of trade, borrowing
and lending, and insurance, not simply a means of payment or transfer
system.
Implications for Mobile Banking
The bottom line of these models of money as a communication device
is that the better the communication of past shocks or transactions, the
more efficient can be the allocation of consumption; however (with
initial heterogeneity), this may be wealth redistributing. The model
features tokens or fiat money but, again, portable cell devices linked
to some of the account history of earlier transactions would provide
similar features. To achieve an efficient allocation there can arise, as
in these models, the need for active liquidity management. We can see
that in a scenario where M-PESA emerges as the entity behind a large
fraction of transactions, e-money could substitute for fiat money or
tokens. This would not necessarily replace the need for an activist
monetary policy, but it would alter that policy so that the level of
tokens created on net by the financial system ideally responds to
mobility and the state of demand, as would electronic credits if allowed
optimally to function that way. Here a distinction between private
credit and public money becomes blurred as we consider questions about
optimal market design. The social good is served by having mutually
agreed upon and collectively enforced rules.
Another lesson from these models is that electronic records of past
transactions allow new financial instruments, in this case better
borrowing/lending and insurance over space and time. Tying fiat money to
e-money and thinking of both as solely facilitating payments may lead
one to miss otherwise beneficial arrangements that have to do with
insurance against spatial and intertemporal idiosyncratic and aggregate
shocks. Indeed, under the current M-PESA system, the prices at which
money trades for e-money are supposed to be fixed over time and across
space; e-money and cash trade for each other one for one (as described
above, however, there are nonlinearities in the transactions costs by
amount traded)--yet these fees can be seen as allowing in principle a
trading price between cash and e-money that is different from one.
Whether or not one wants to allow money prices and the rate of exchange
of money for e-money to move with the state of local demand and
inventory of the actors again begs the question of what e-money is
supposed to be: a means of payment only, if it facilitates an expansion
of the monetary base, or a partial substitute for missing, more
centralized economy-wide insurance and credit markets
Townsend and Wallace--Circulating Private Debt and a Coordination
Problem
There is yet another way to think of money, namely as an object
that, even if privately issued, appears frequently in exchange, i.e.,
with a high velocity. We can understand this by simply extending the
model environment in the previous section to four periods with
households b and b' continuing to switch locations from one period
to another, back and forth, and with households a and a' remaining
in a single location. Townsend and Wallace (1982) replace preference
shocks with time-varying endowments of a single good, but with different
profiles for the different agents, to induce the desire for
intertemporal trade. They also assume there are many agents of each type
in any given location to justify price-taking behavior. In one of the
equilibria, the first period household b makes a deposit of goods (but
could be money) to (that is, lends to) agent a, as if agent a were a
bank issuing long-term debt (or at least debt payable on demand).
However, household b does not hold this debt but rather moves in the
second period to a different location inhabited by agent a'. At
this new location, neither party is physically connected to bank a.
Subsequently, in the third period, agent a' will pass the debt to
b', who in turn redeems it in the last period since b' meets
up with the original issuer, agent a. Note that long-term debt is also
the debt that circulates, that has a high velocity. In that sense
circulating debt has something to do with maturity transformation.
Short-term debt (e.g., two-period debt) between agents a and b (or
a' and b') not only extinguishes sooner but it also does not
circulate.
With private debt transferable electronically, one has the same
equilibria but with the more realistic interpretation of agent a as an
M-PESA agent who issues debt (in this case in exchange for goods, not
fiat money, but see below). That is, household b uses the e-money
account to trade with subsequent households and, again, the e-money is
netted out back to zero when a third party comes to agent a to redeem
it. In this model, agent a' can also play this role as banker, or
M-PESA agent, instead of a. However, without coordination, another
problem emerges. The amount of e-money issued by agents a and a'
has to be coordinated so as to be consistent with the overall
equilibrium. If M-PESA agents a and a' are not communicating across
space, then it is hard to imagine how this would happen. Townsend and
Wallace refer to various historical episodes such as the crash of
markets in bills of exchange as evidence that the model with
coordination problems is picking up problems that may occur in practice.
Implications for Mobile Banking
Electronic debits can be transferred across agents in spatially
separated locations and have a high velocity. This seems to capture a
big part of the Kenyan M-PESA reality. This comes, however, with
potential coordination issues that need to be thought through. In the
current model with two locations, four agent types, and four periods,
one achieves the first-best with the right combination of circulating
private debt and other short-term noncirculating debt. In that
equilibrium, prices/interest rates are moving around over time and space
and all markets in goods and financial instruments are clear. Again,
fixing the price at which one object trades against another would seem
to create additional problems. But even if prices were flexible, it
appears that agents need to coordinate on the overall credit issue. Lack
of initial coordination could show up as an over-issue or under-issue of
the correct financial instrument, or of the combination of instruments
that is supposed to give the correct maturity structure, showing up in
turn later on as sharp movements in prices. This could even lead to
doubts about the commitment or ability of agents to achieve the
requisite transfers of purchasing power necessary for liquidity in
intermediate periods or to ensure redemption of debt at maturity.
Manuelli and Sargent ponder whether fiat money can help solve this type
of coordination problem.
Lacker's Model of Clearing and Settlement and Inter-Agent
Markets
Lacker (1997) focuses on clearing and settlement via a central bank
and the impact of certain central bank policies such as reserve
requirements and interest paid on reserves. Building on the earlier
models of Townsend (1983, 1989), Lacker develops a model in which there
is a large number, N, of islands, on each of which live N individuals.
Each island produces a single perishable good that must be consumed on
the island. This geography is illustrated for the case N = 3 in Panel A
of Figure 12, in which the islands are labeled A, B, and C, and the
individuals are 1, 2, and 3. In each period, all but one of the
individuals who live on a given island travel to all the other islands
at random, one to each, with one staying behind. In Panel B of Figure
12, individual 3 remains home. All individuals consume the good that is
produced on the island they visit (so the one who remains consumes the
good produced on his island). As in Townsend (1989), before they leave
"home," travelers entrust their endowment of goods to the
individual who stays behind (called the merchant banker) and is
responsible for handing it out to arrivals from other islands. The
record of the amount entrusted to the stay-at-home agent is an
individual's "deposit" and the merchant banker is thought
of as operating a bank that holds his island's deposits.
[FIGURE 12 OMITTED]
As illustrated in Figure 12, each island receives a fully
diversified group of visitors each period, one from each other island.
If preferences and endowments were suitably fixed (e.g., if each
individual consumed 1/N units of the good of the island s/he visited),
consumption and income would balance on a person-by-person basis.
However, Lacker assumes, like some of the models above, that each period
the islands are hit by Diamond-Dybvig idiosyncratic independent
identically distributed preference shocks that affect the urgency of
consumption. All individuals from a given island get the same shock.
Since each island is visited by an individual from every other island
and since shocks are independent across islands, there is no aggregate
uncertainty about the demand for each island's good (as N goes to
[infinity]). However, an island that suffers a run of large urgent
shocks over time consumes more over time than an island that suffers a
run of small shocks.
Because goods do not move between islands, there is no possibility
to directly exchange one for another. Instead, an individual purchases
consumption from the merchant on the island he visits by providing a
bill or check drawn on his deposit held by his own merchant who stayed
at home. (This could be an electronic charge to the e-account but,
again, not one paid instantaneously.) In turn, each merchant collects
bills or e-credits from all other islands, one for each visitor. In any
given period, some islands will consume more than they produce (i.e.,
issue more bills than they collect or be left with negative e-balances),
while the opposite will be true for others. Intertemporal trade between
islands across periods, i.e., interbank borrowing and lending of
e-balances, is thus efficient.
In the final stage of the period, all the merchant bankers travel
with their bills to a central location and submit them (to each other)
for payment (Panel C of Figure 12). With cell technologies, physical
meetings would not be necessary. Payment is effected through an
accounting mechanism, with each island's account being credited and
debited according to the bills or e-money presented to and by it. The
residual that does not clear is carried over, in surplus or deficit.
Implications for Mobile Banking
Lacker refers to the central institution that keeps the accounts of
each island as the central bank and these accounts are thought of as
reserve accounts. However, this could equally be a private clearinghouse
run by Safaricom or some other independent entity, as the model focuses
on the account-keeping and clearing functions of the institution, not
the issuing of money per se. Positive account balances with the
institution are the liabilities of that institution, while negative
balances represent overdrafts. In the model, bills are cleared (i.e.,
accepted by the clearinghouse/central bank) at the end of the period and
settled (i.e., deposits transferred by the institution from one
island's reserve account to another's) at the end of the
period for across-period borrowing and lending.
Beyond Lacker's model, limits on within-period bank overdrafts
with the clearinghouse/central bank can induce some banks (ones with a
positive preference shock) to borrow from others. Likewise, limits on
overnight overdrafts can induce residents of islands that have had a
string of positive, urgent shocks to constrain their consumption below
the efficient level, as they are unable to borrow enough. Lacker's
model is a useful motivation for thinking about another aspect of the
M-PESA system, in particular overall clearing and the related inventory
management problem faced by agents. The kind of contractual conditions
Safaricom might want to specify would be crucial given the reality of
the actual economy in which the distinction between within-period and
across-period clearing and borrowing/lending is hard to maintain.
We identify each M-PESA agent with a merchant banker in
Lacker's model, although individuals are not bound to agents like
residents are to islands. An M-PESA agent's trading account at
Safaricom corresponds directly to the reserve account held by each bank
with the central bank. To parallel the model, individuals deposit their
endowments (of cash) with an agent each period, which requires that the
agent hold sufficient e-money. An agent would take out an overdraft loan
from Safaricom if he were required to issue e-money to a customer before
having presented the equivalent amount of cash to Safaricom. Because
transferring a bank note or cash is slower than transferring e-money, it
seems likely that there could be demand for such overdrafts.
E-money is sent between individuals (i.e., checks are exchanged)
and recipients present their e-money (i.e., checks) to agents. This
happens at the end of the period in Lacker's model. If agents have
enough cash to purchase the e-money from customers, their trading
accounts are credited with the relevant amounts. In reality, as
individuals visit the agent over the course of the day, his net demand
for e-money will fluctuate and he might require short-term overdrafts
from Safaricom or need to acquire cash in some other way. In the absence
of such a facility, he will need to trade off the costs of holding
"zero-interest reserves" (e-money balances on his trading
account) against the costs of reduced trade (and commissions).
Alternatively, agents could lend e-money to one another, creating the
equivalent of an interbank market as envisioned in Lacker's model.
Again, this might be organized by another institution (like a
clearinghouse) that itself purchased e-money from Safaricom and lent it
out to agents at some interest rate. Likewise, the head offices or super
agents could perform this role, though neither appears to charge
interest. Each head office or super agent would face a similar inventory
management problem of course, having to hold enough e-money and/or cash
to lend out during the day/period.
Freeman and Green's Models of Liquidity--Optimal Base-Money
Management
Freeman's (1996) model and Green's (1999) reformulation,
related to Townsend (1989) as exposited above, focus on getting money
and circulating debt in the same setting simultaneously because of
imperfect meetings between creditors and debtors. This, again, has
implications for liquidity and monetary policy.
In Green's overlapping generations model, there are two types
of individuals (creditors and debtors) who live for two periods each. A
creditor is someone who in equilibrium will be willing to defer
consumption, while a debtor will wish to borrow. We follow tradition and
refer to young and old agents, simply to imply the first and second
periods of the two-period, dynamic transactions profiles of the
households. When young, creditors and debtors are endowed with
perishable goods x and y, respectively. In the first period, old
creditors are endowed with fiat money and old debtors have nothing.
Creditors and debtors also differ in their preferences: Creditors wish
to consume when they are young and old, while debtors wish to consume
only while young. Both types prefer to consume a mix of goods x and y
instead of just their own.
Suppose young debtors meet young creditors first and only then go
on to meet old creditors. Young debtors purchase x in return for debt d
that they issue to young creditors, as illustrated in the first panel of
Figure 13. Subsequently, young debtors sell their own good y to old
creditors in exchange for money. At the beginning of the next period
(the second panel of Figure 13), now-old (previously young) debtors
settle their debts using money with now-old (previously young)
creditors. Once the debt is settled the process repeats, with the
now-old creditors holding money and the new young cohorts endowed with
goods.
[FIGURE 13 OMITTED]
Nontrivial monetary dynamics can arise when creditors and debtors
do not necessarily meet at the "right" time. With various
waves of movers, old agents either arrive late or leave early: In
particular, a fraction (1 - [delta]) of debtors arrive late and a
fraction [gamma] of creditors leave early. This naturally complicates
the debt settlement process and can lead to inefficiencies. In
particular, while efficiency requires that all creditors consume the
same quantity of goods when old (purchased from young debtors), those
who leave early may in equilibrium consume less than this amount while
those who leave later consume more. Thus, early-leaving creditors can
end up facing liquidity shortages that constrain trade.
This situation is illustrated in Figure 14, in which period t + 1
is divided into two subperiods (t + 1 in the second panel down, and t
+ 1' in the third panel). Agents not in the market at the
relevant moment in time are shown as boxes with broken lines. It is
assumed that the creditors are fully diversified at t + 1, holding debt
issued by each and every old debtor. At the beginning of t + 1, all the
old creditors are present, as are a fraction, [delta] of old debtors.
The debtors settle their debts and each creditor receives a share,
[delta] < 1, of the amount owed to him. At date t + 1', the
late-arriving old debtors are able to settle their debts in full with
old creditors who remain. Early-leaving old creditors will consume less
than their efficient level of consumption and late-leaving ones will
consume more.
[FIGURE 14 OMITTED]
An alternative scenario illustrated in Figure 15 allows old
creditors to exchange debt. Period r + 1 is now divided into 1 + 1, t +
1', and t + 1". At the beginning of t + 1 (the second panel),
early-arriving debtors repay their debts to all creditors. As before,
all creditors continue to hold outstanding debt issued by debtors who
have not yet arrived. Next, at time t + 1' (the third panel),
early-leaving creditors sell their remaining holdings of debt to
long-lived creditors in exchange for money. If [delta] < [gamma],
with a relative scarcity of debtors and an abundance of early-leaving
creditors, the creditors spend all of their money on debt, which has a
price less than one. Early-leaving creditors then purchase goods from
young debtors and quit the market. In the final panel, at time t +
1", late-arriving debtors settle their debt with those creditors
who remain, including the debt those creditors bought from early-leaving
creditors. They then purchase goods from young debtors. The amount of
consumption they enjoy is (1 - [delta])/(1 - [gamma]) times the
efficient level. Thus, if not enough debtors show up in time, even if
creditors trade their debt, the allocation is inefficient.
[FIGURE 15 OMITTED]
As Freeman observed, a central bank can remedy the inefficiency by
issuing money to some or all creditors and then withdrawing it from
circulation later, say by taxing young creditors as they enter the
second period. The important issue is that new money is issued to
creditors and is issued before the early-leaving creditors depart.
Implications for Mobile Banking
In Green's version of the model, debtors consume nothing in
the second period of their lives, but creditors do. The only reason for
old debtors to come to the market is to pay off their debts. So if
e-money allows them, or some of them, to do this without coming to the
market, then the share that are "late" is smaller. In the
extreme case, there would be no late-arriving debtors and no liquidity
problems for creditors. But if some old debtors still didn't pay
off their debts in time (maybe because they couldn't find an agent
with e-money), then it would be possible that early-leaving creditors
wouldn't have enough money to finance the efficient level of
purchases from young debtors.
These models can inform our thinking about mobile banking in a
couple of ways. First, focusing on the reduction in transactions costs
associated with transferring e-money, mobile banking might reduce the
proportion (1 - [delta]) of debtors who "arrive late" and the
proportion of creditors, [gamma], who "leave early." Debtors
who previously had to physically meet their creditors in order to settle
their debts can now settle them with e-money and no longer need to be
present. On the other hand, even if not all debts are repaid at the
beginning of the period (i.e., if there remain some late-arriving
debtors), the existence of e-money could relax the liquidity constraint
faced by early-leaving creditors and make central bank intervention less
necessary.
However, it is overly simplistic to assume that mobile banking
allows individuals to send money costlessly: it allows them to send
e-money costlessly (or at least at low cost) but they must acquire it
first. A more complete model would thus include individuals holding
optimal mixes of money and e-money and would describe the production
process whereby each is converted into the other. In practice, this
conversion is effected by M-PESA agents who simply perform the role of a
technological black box--a black box that is sometimes out of service.
Although this feature is not part of the Freeman and Green set-up,
if money and e-money are both used in equilibrium, then a
"late-arriving debtor" might correspond to an individual who
is otherwise "on time" and has sufficient financial resources
(money and/or e-money) to repay his debts, but who is frustrated in not
being able to find an M-PESA agent with sufficient e-money or money.
(24) Similarly, an early-leaving creditor in this environment could be
one who has in fact been repaid, say in e-money, but who must use money
to purchase the consumption good. (25) If he cannot find an M-PESA agent
with sufficient cash, then he could be liquidity constrained as above.
First, if he is stuck with e-money but must trade with money, he will
suffer a loss equal to his excess e-money holdings. On the other hand,
even if he cannot find an M-PESA agent to trade with, he might trade his
e-money with another creditor for cash, just as early-leaving creditors
sell their debt to late-leaving creditors for cash in the third panel of
Figure 15. But such trades must take place between locationally
proximate agents, and if there is an excess supply of e-money locally,
the allocation of consumption might remain inefficient.
3. LINKING THEORY WITH DATA: RESULTS FROM HOUSEHOLD AND AGENT
SURVEYS
In this section we present data that speaks to some of the issues
raised by the models of money summarized above, especially as regards
shortages of e-money and cash and whether there are indeed credits in
the system because of the operational logistics of agents as described
in Section 1. These data, some of which are reported in Jack and Suri
(2009), derive from a survey of 3,000 households and 250 M-PESA agents
in Kenya in late 2008. (26) We focus on issues related to M-PESA agents
as reported by consumers and the agents themselves, as motivated by the
models.
First, 10 percent of all consumers reported facing at least one
problem with the agents they had visited. Of those who reported
problems, Table 3 shows the breakdown of the problems they had. By far,
the most common problems are agents' lack of cash and e-money. The
first four rows in the table, in fact, suggest that in some cases agents
have been able to increase the price of e-money by varying the fees they
charge consumers. This is an important implication of the models
discussed above--fixing prices for cash and e-money will require an
accompanying policy stance. However, the penultimate two rows in the
table confirm that this strategy is employed nowhere near enough to
clear the market.
Table 3 The Problems Consumers Have Had with Agents
Most Used Agent Closest Agent
Agent Gave Less Money/
E-Money Than I Was Owed 2.63% 2.80%
Agent Charged Me to Deposit 1.11% 1.68%
Agent Overcharged Me 1.17% 1.84%
Agent Undercharged Me 0.52% 0.78%
Agent Was Absent 0.74% 0.91%
Agent Refused to 0.80% 0.61%
Perform the Transaction
Agent Was Unknowledgeable 1.04% 1.57%
Agent Was Rude 3.66% 6.02%
Agent Had No E-Money/ 43.60% 22.81%
Not Enough E-Money
Agent Had No Cash/ 34.78% 51.33%
Not Enough Cash
Other 9.95% 9.65%
In addition, in the survey, consumers were specifically asked if
they were either unable to deposit money or unable to withdraw money
from the M-PESA agent closest to them or from the agent they used the
most. Table 4 (27) shows that approximately 6 percent of M-PESA users
are unable to deposit money with an agent, i.e., the agent does not have
any e-money to give the consumer in return. Also, as many as 15 percent
of consumers were unable to withdraw money from the closest agent, i.e.,
the agent had no cash to give the consumer in exchange for e-money.
Table 4 Unable to Deposit Cash (No E-Money) or Unable to Withdraw
Cash (No Cash)
Most Used Agent Closest Agent
Have You Ever Been Unable to Deposit
Money at this Agent?
Yes 6.63% 6.22%
No 93.37% 93.78%
Total 100% 100%
Have You Ever Been Unable to Withdraw
Money from this Agent?
Yes 6.63% 15.33%
No 93.37% 84.67%
Total 100% 100%
In the survey of agents themselves, respondents were asked how
often they run out of e-money and how often they run out of cash--these
results are reported in Tables 5 and 6. On average, about 29 percent of
agents run out of e-money once a month or more frequently and indeed a
nontrivial fraction (14 percent) run out about once a week. Similarly,
about 26 percent of agents run out of cash once a month or more
frequently than that and about 10 percent run out once a week (and, in
fact, about 8 percent run out on a daily basis). Clearly there are
liquidity issues, both in terms of cash as well as in terms of e-money.
This is anticipated from the discussion of Lacker (1997), Freeman
(1996), and Green (1996)--models in which such liquidity constraints are
evident.
Table 5 How Often Do Agents Run Out of E-Money?
Fraction
More Than Once a Day 3.2%
Once a Day 6.4%
Once a Week 14.0%
Once a Month 5.6%
Once Every Three Months 1.2%
Once Every Six Months 0.4%
Less Often Than That 12.0%
Never 57.2%
Table 6 How Often Do Agents Run Out of Cash?
Fraction
More Than Once a Day 3.2%
Once a Day 8.4%
Once a Week 10.0%
Once a Month 4.8%
Once Every Three Months 1.2%
Once Every Six Months 0.4%
Less Often Than That 22.4%
Never 49.6%
Safaricom initially required all M-PESA agents to pre-purchase
e-money before they could trade it for money to the public. If an agent
runs out of e-money, he is required to purchase more, either from
Safaricom or from the public when they redeem cash, before being able to
take cash deposits from the public. This suggests that there are no
credits or debits involved between agents and Safaricom and therefore no
role for a formal settlement system. Indeed, even as the agent model has
evolved, this feature has been maintained, at least with respect to the
relationship between the "coordinating bodies" of Figure 7 and
Safaricom. On the other hand, cash and e-money transactions between
agents and their head offices or aggregators need not remain in
continuous balance, and the parties can operate in a net credit or debit
position vis-a-vis each other. These imbalances are of little concern
for the head office model (Panel A of Figure 7), to the extent that the
agents are owned and controlled closely enough that internal financial
arrangement of the group does not affect its viability. However, the
more arm's length relationship between agents and an aggregator
(Panel B) suggests that chronic imbalances with such a group could prove
problematic.
We note that while the potential exists for persistent financial
imbalances within a group under the aggregator model, in principle
M-PESA users on the one hand and Safaricom on the other do not face any
risk associated with the bankruptcy of any particular agent or agent
group, as deposits of cash are matched at the level of the coordinating
body by transfers of e-money and vice versa. If an individual user finds
that all agents within a reasonable distance go out of business, he will
likely face a liquidity constraint unless he is able to use his e-money
to directly purchase goods and services.
The survey asked agents how they pay for e-money when they request
it from their head office. In well over half the cases, agents receive
transfers from their head offices without any immediate corresponding
payments (see Table 7).
Table 7 How is E-Money Paid For When an Agent Request It?
Fraction
It is a Direct Purchase of E-Money From the 36.2%
Head Office (Involves a Cash Transfer)
Receive a Direct Transfer From the 31.2%
Head Office with No Concurrent Payment
Receive a Direct Transfer From the 18.1%
Head Office with No Payment Required
Other 12.6%
Refused to Answer 2.0%
Total 100%
Similarly, agents were asked how they get cash for M-PESA
transactions when they run out. As reported in Table 8, in more than
half the cases agents do not immediately exchange e-money for cash
received.
Table 8 How Do Agents Get Cash When They Run Out of It?
Fraction
Redeem From Head Office in Exchange for E-Money 17.6%
Redeem Direct Transfer of Cash From 20.4%
Head Office with No E-Money Exchange
Use Own Savings 42.8%
Other * 18.0%
Don't Know 1.2%
Total 100%
Notes: * Of which 27 percent is from "sale of credit card,"
27 percent is "wait for deposits," 21 percent is "borrow from
management," and 11 percent is "from the other business."
The statistics in Tables 7 and 8 suggest that credit arrangements,
explicit or otherwise, between agents and their head offices or
aggregators appear to be widespread. We do not know the maturity of
these credits but, given the large number of agents reporting them, it
is possible that at least some of them are longer than simple overnight
positions. Indeed it seems inevitable that there may be a nontrivial
amount of credit in these transactions as the "supply chain"
of e-money involves exchanges in spatially and temporally separated
markets, leading naturally to one-way transfers of either cash or
e-money. Given the mechanics of M-PESA, these credits cannot be issued
between agents and individual users or between the coordinating body and
Safaricom. But, the evidence indicates that such net credits/debits do
exist between agents and their coordinating bodies. Again, all the
models above allow for credit and debits, which are often welfare
improving. In addition, some of the models above illustrate the
welfare-improving nature of broader financial integration, which such
credits and debits would encourage.
4. CONCLUSION AND FURTHER MODELING
The most successful version of mobile banking in Kenya (and perhaps
the world), M-PESA, is--quite literally--everywhere. In many cases, the
scenarios envisioned in existing monetary theory models appear to match
the reality of M-PESA and, as such, these models promise to inform
decisions taken by both Safaricom in managing M-PESA and the central
bank in managing the Kenyan economy. The empirical evidence presented,
from surveys of both M-PESA users and agents, further serves to
illustrate the importance of these lessons.
For example, just as the central bank may intervene to relax
liquidity constraints, it is arguable that Safaricom should actively
manage "e-liquidity" by issuing e-money that is at times, in
some locations, unbacked by money deposits, assuming that such activism
would be costless and allowed by the central bank. In fact, the data
suggest that some M-PESA agents are engaging in such e-liquidity
management already (for example, when they receive e-money transfers
from their head offices without a corresponding transfer back of cash).
This has implications, of course, for the measurement and meaning of
monetary and debt aggregates. Improved systems, however, require that
the company have better information on net demands for e-money across
agents than the agents themselves have, or at least be better able to
act on this information without the space/time coordination problems
that the models suggest. Not surprisingly, Safaricom has been changing
their agent model over time to better deal with cash and e-money
liquidity issues. Time series and geographically disaggregated data on
fluctuations in demand would be useful for further evaluating these
issues and making improvements to their system.
On the modeling side, understanding the operations of the M-PESA
agent network seems key to the development of an overall comprehensive
model of e-money. For example, modeling the decisions and constraints of
agents would potentially allow us to endogenize the timing patterns
assumed in some of the existing models. Frictions that impede efficient
and immediate reallocations of money and e-money balances across agents
would thereby replace these timing assumptions as the fundamental source
of liquidity constraints. Similarly, realistic heterogeneity across
consumers--for example, in terms of phone ownership, access to mobile
coverage, safety of the local environment, frequency of market access,
access to M-PESA agents--could be modeled more explicitly.
REFERENCES
Cavalcanti, Ricardo, and Neil Wallace. 1999. "A Model of
Private Bank-Note Issue." Review of Economic Dynamics 2 (January):
104-36.
Cass, David, and Menahem E. Yaari. 1966. "A Re-Examination of
the Pure Consumption Loans Model." Journal of Political Economy 74
(August): 353-67.
Central Bank of Kenya. 2009. Presentation at Conference on Banking
and Payment Technologies, East Africa, Nairobi (February).
Diamond, Douglas, and Philip Dybvig. 1983. "Bank Runs, Deposit
Insurance, and Liquidity." Journal of Political Economy 91 (June):
401-19.
Financial Sector Deepening. FinAccess Reports, various. Kenya: FSD.
Foster, Kevin, Erik Meijer, Scott Schuh, and Michael A. Zabek.
2010. "The 2008 Survey of Consumer Payment Choice." Federal
Reserve Bank of Boston Public Policy Discussion Paper 09-10 (April).
Freeman, Scott. 1996. "The Payments System, Liquidity, and
Rediscounting." American Economic Review 86 (December): 1,126-38.
Fuerst, Timothy S. 1992. "Liquidity, Loanable Funds, and Real
Activity." Journal of Monetary Economics 29 (February): 3-24.
Fujiki, Hiroshi, and Migiwa Tanaka. 2009. "Currency Demand,
New Technology and the Adoption of Electronic Money: Evidence Using
Individual Household Data." IMES Discussion Paper Series
(December).
Green, Edward J. 1999. "Money and Debt in the Structure of
Payments." Federal Reserve Bank of Minneapolis Quarterly Review 23
(Spring): 13-29.
Grossman, Sanford J., and Laurence Weiss. 1983. "A
Transactions Based Model of the Monetary Transmission Mechanism."
American Economic Review 73 (December): 871-80.
Ireland, Peter N. 1994. "Money and the Gain From Enduring
Relationships in the Turnpike Model." Federal Reserve Bank of
Richmond Working Paper 94-07.
Jack, William, and Tavneet Suri. 2009. "Mobile Money: The
Economics of M-PESA." Working Paper.
Jack, William, and Tavneet Suri. 2010. "The Risk Sharing
Benefits of Mobile Money." Working Paper.
Kocherlakota, Narayana. 2005. "Discussion of 'From
Private Banking to Central Banking: Ingredients of a Welfare
Analysis.'" International Economic Review 46 (May): 633-6.
Kocherlakota, Narayana, and Neil Wallace. 1998. "Incomplete
Record-Keeping and Optimal Payment Arrangements." Journal of
Economic Theory 81 (August): 272-89.
Lacker, Jeffrey M. 1997. "Clearing, Settlement and Monetary
Policy." Journal of Monetary Economics 40 (October): 347-81.
Lucas, Robert E., Jr. 1980. "Equilibrium in a Pure Currency
Economy." Economic Inquiry 18 (April): 203-20.
Lucas, Robert E., Jr. 1990. "Liquidity and Interest
Rates." Journal of Economic Theory 50 (April): 237-64.
Manuelli, Rodolfo, and Thomas J. Sargent. 2009. "Alternative
Monetary Policies in a Turnpike Economy: Vintage Article." New York
University Working Paper (June).
Mas, Ignacio, and Olga Morawczynski. 2009. "Designing Mobile
Money Services: Lessons from M-PESA." Innovations 4 (April): 77-91.
Perez-Verdia, Carlos. 2000. "Transaction Costs and
Liquidity." University of Chicago, PhD dissertation.
Romer, David. 1987. "The Monetary Transmission Mechanism in a
General Equilibrium Version of the Baumol-Tobin Model." Journal of
Monetary Economics 20 (July): 105-22.
Rotemberg, Julio J. 1984. "A Monetary Equilibrium Model with
Transactions Costs." Journal of Political Economy 92 (February):
40-58.
Townsend, Robert. 1983. "Financial Structure and Economic
Activity." American Economic Review 73 (December): 895-911.
Townsend, Robert. 1987. "Economic Organization with Limited
Communication." American Economic Review 77 (December): 954-71.
Townsend, Robert. 1989. "Currency and Credit in a Private
Information Economy." Journal of Political Economy 97 (December):
1,323-44.
Townsend, Robert, and Neil Wallace. 1982. "A Model of
Circulating Private Debt." Federal Reserve Bank of Minneapolis
Staff Report 83.
Wallace, Neil. 2000. "Knowledge of Individual Histories and
Optimal Payment Arrangements." Federal Reserve Bank of Minnesota
Quarterly Review 24 (Summer): 11-21.
Wallace, Neil. 2005. "From Private Banking to Central Banking:
Ingredients of a Welfare Analysis." International Economic Review
46 (May): 619-31.
Wicksell, Knut. 1935. lectures on Political Economy. London:
MacMillan.
William Jack is at the Department of Economics, Georgetown
University; Tavneet Suri is at MIT Sloan School of Management; and
Robert Townsend is at the Department of Economics, MIT. The authors
would like to thank the editor, Ned Prescott; the referees, Huberto M.
Ennis and Kartik Athreya; and Ignacio Mas, Carlos Perez-Verdia, and Tom
Sargent for detailed comments and feedback. This research is supported
by the Gates Foundation through the University of Chicago Consortium on
Financial Systems and Poverty. The views expressed in this article are
those of the authors and do not necessarily reflect those of the Federal
Reserve Bank of Richmond or the Federal Reserve System. E-mails:
[email protected];
[email protected];
[email protected].
(1) For example, M-PESA is similar to the service called G-CASH in
the Philippines.
(2) The original pilot program, supported by the U.K. government
and the mobile phone provider Vodafone, was aimed at increasing the
efficiency of microfinance products by allowing borrowers to make
repayments more easily. However, by the time of the full launch, the
focus had shifted to facilitating the sending of remittances more
generally.
(3) This is not quite true, as some individuals own two (or more)
phones so as to take advantage of the different tariff policies of
competing providers.
(4) Pesa is Kiswahili for "money"--hence M[obile]-Money.
A second mobile banking service called ZAP has since been launched,
operated by Zain, the second largest mobile phone operator in Kenya.
ZAP's market share remains very small at this point in time.
(5) The marginal cost of depositing and sending money is very low.
These fees cover the costs of maintaining and expanding the agent
network and physical infrastructure, marketing, and profits.
(6) The complete tariff schedule is available at
http://www.safaricom.co.ke/fileadmin/template/main/downloads/Mpesa_forms/14th%20Tariff%20Poster%20new.pdf
(7) Nonregistered individuals can receive money sent by a
registered user as long as they have a cell phone. The recipient
receives a text message with a code that can be taken to an M-PESA agent
who provides the cash less any fees. The fee schedule is designed so as
to encourage recipients to register. Note that a nonregistered user
cannot send e-money to a third individual from his phone.
(8) Transactions are not always limited to innocent trades. For
example, there are reports of people using M-PESA to pay bribes to
traffic police. Even worse, rumors have circulated in Nairobi that
kidnappers are requesting ransom to be paid by M-PESA. although these
rumors have not been confirmed.
(9) Similar services in Tanzania and South Africa, for example,
have penetrated the market much less. See Mas and Morawczynski (2009).
(10) In 2006 it was estimated that 18.9 percent of Kenyan adults
used a bank account or insurance product, and by 2009 this had increased
to 22.6 percent (see Financial Sector Deepening, Finaccess I).
(11) These data are from a survey fielded in late 2008. Since then,
there has been some growth in the number of individuals and households
with a bank account because of the expansion of such institutions as
Equity and Family Bank.
(12) In 2003 there were 230 ATMs in Kenya (see Central Bank of
Kenya [2003] at http://www.centralbank.go.ke/downloads/nps/nps%20old/psk.pdf). Recent data suggest there are around 1,200.
(13) These data refer to a period before M-PESA could be used at
ATMs.
(14) The commission amounts are nonlinear (and concave) to (he size
of the transaction. Some reports suggest that in response to this,
agents may encourage customers to split their transactions into multiple
pieces, thereby increasing the overall commission.
(15) http://www.centralbank.go.ke/financialsystem/banks/Register.aspx
(16) M-PESA requires that each coordinating body has a bank account
so that funds can be transferred easily between them. In order to open
an M-PESA business, the coordinating body must have a minimum balance in
a bank account, which is used to purchase initial holdings of e-money.
(17) This model started after the first round of the Jack and Suri
(2009) survey.
(18) About 50 percent of households had at least one member with a
bank account. Of banked households in the survey, about 60 percent used
M-PESA, compared with the 54 percent of un-banked households reported
above.
(19) These models rule out private debt and future contracts in
fiat money by assuming there are no pairings such that debt can be
redeemed by the issuer. See below.
(20) For example, see Grossman and Weiss (1983). Rotemberg (1984).
Romer (1987), Lucas (1990). Fuerst (1992), and Perez-Verdia (2000).
(21) See also Ireland (1994), Kocherlakola and Wallace (1998),
Cavalcanti and Wallace (1999), Kocherlakota (2005), and Wallace (2005),
and the review in Wallace (2000) in which outside money and inside money
issued by banks with known trading histories co-exist.
(22) In this model, the agents all pre-commit to arbitrary tax and
transfer schemes over time and to all institutions and resource
allocation rules. In the language of the models, they commit to an
economy-wide credit arrangement that specifies consumption and transfers
to agents conditional on aggregate states and on individual specific
location shifters (that are public) and individual announcements of
preference shocks (private). Apart from these plans, there is no
government and no distinction between private and public.
(23) Individuals who are patient consume more later and therefore
need more units of money to confirm this to future strangers.
(24) Whether the debtor needs to find an M-PESA agent with money or
e-money will depend on what form of financial wealth the debtor has on
hand and how the creditor wishes to be paid. This in turn will depend on
the specific features of the two monies.
(25) Again, whether the creditor needs money or e-money depends on
how the seller wants to be paid.
(26) Part of these data form the basis of a confidential report
issued by Financial Sector Deepening to the Central Bank of Kenya. In
addition. Jack and Suri (2010) look at some of the microeconomic
risk-sharing impacts of M-PESA. Other papers have also looked at the
more microlevel impacts of e-money on currency demand (for example, see
Fujiki and Tanaka [2009]).
(27) Note that the questions asked for Tables 3 and 4 are quite
different. Table 3 asks about the main consumer-reported problems with
agents while Table 4 asks about the incidence of two specific problems.
