risk sharing and the economics of m-pesa payment... · risk sharing and the economics of m-pesa...

Risk sharing and the economics of M-PESA

William Jack Georgetown University

Tavneet Suri MIT Sloan

With support from the Consortium on Financial Systems and Poverty

Impact and Policy Conference August 30 – September 1, 2012

Bangkok

The solution:

Jack - M-PESA

The problem:

M-PESA as a risk spreading tool

• Formal insurance is limited

• Informal insurance exists, but is often incomplete…….why?

• Moral hazard: information asymmetries

• Limited commitment: contract enforcement

• Transaction costs

Jack - M-PESA

Summary of findings

• The consumption of households who don’t use M-PESA falls by about 7% - 10% when they suffer negative shocks

• Lower transaction costs allow households who use M-PESA to smooth these risks perfectly

The M-PESA concept

• Remote account storage accessed by simple SMS technology

• Cash-in and cash-out services provided by M-PESA agents

Jack - M-PESA

Customers

Customer and Agent growth

Agents

Cu

sto

mer

s

Age

nts

2007

2008

2009 2010

Jack - M-PESA

?

2011

0

5,000

10,000

15,000

20,000

25,000

30,000

0

2

4

6

8

10

12

14

16

Oct-06 Apr-07 Nov-07 Jun-08 Dec-08 Jul-09 Jan-10 Aug-10 Feb-11 Sep-11

Mill

ion

s

Nairobi

Lake Victoria

Mombasa

June 2007 Note: partial data only

Jack - M-PESA

Dec 2007 Note: partial data only

Jack - M-PESA

Nairobi

Lake Victoria

Mombasa


Jack - M-PESA

Nairobi

Lake Victoria

Mombasa

Our household survey

Tanzania

Indian Ocean

Uganda Somalia

Nairobi

• 3,000 households across most of Kenya

• Four rounds: 2008, 2009, 2010, 2011

Jack - M-PESA

Who is using M-PESA?

0%

25%

50%

75%

100%

2008 2009 2010 2011

>$2/day $1.25-$2/day <$1.25/day

Households outside Nairobi Median consumption ~$2 per day

Jack - M-PESA

Banking for the unbanked?

0%

25%

50%

75%

100%

2008 2009 2010 2011

Unbanked Banked

Households outside Nairobi Median consumption ~$2 per day

Jack - M-PESA

How do people use M-PESA?

0%

20%

40%

60%

80%

100%

2009 data

Share of households

Transactions

Jack - M-PESA

How often do people use M-PESA?

2%

5%

6%

43%

14%

4%

4%

24%

0% 10% 20% 30% 40% 50%

Daily

Weekly

Every 2 weeks

Monthly

Every 3 months

Every 6 months

Once a year

Less often

Transaction Costs

0

200

400

600

800

1,000

1,200

1,400

0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000

Ta

riff

Amount deposited and sent

Postapay M-PESA: Reg to reg Western Union

Empirical strategy

Shock No shock

Consu

mpti

on

User

Non-user (a)

Shock status

Users are richer ()

Shocks hurt ()

Shocks don’t hurt

users so much (b)

c = a + Shock + User + bUser * Shock + controls

Basic Results OLSA PanelB PanelC Without NairobiC

M-PESA User 0.553*** -0.090** -0.016 -0.008

[0.037] [0.036] [0.047] [0.049]

Negative Shock -0.207*** 0.241** 0.232 0.120

[0.038] [0.116] [0.169] [0.141]

User*Negative Shock 0.101** 0.176*** 0.156** 0.150**

[0.050] [0.050] [0.062] [0.065]

Shock, Users -0.105*** 0.052* 0.055 0.050

[0.033] [0.028] [0.035] [0.037]

Shock, Non-Users -0.207*** -0.069** -0.068 -0.056

[0.038] [0.032] [0.043] [0.045]

Jack - M-PESA

A: Full sample with time Fes; B: Full sample with controls + interactions C: Full sample, controls + interactions, time and time x location FEs

0

0.5

1

1.5

2

2.5

3

3.5

4

Mean Distance

(km)

5th Percentile 25th Percentile 50th Percentile 75th Percentile

Round 1

Round 2

Improving Agent Access

22% Change

40% Change

33% Change

28% Change

14% Change

Distance to the closest agent (km)

Jack - M-PESA

Using Agent Roll Out

Agents w/in

1km

Agents

w/in 2km

Agents w/in

5km

Agents

w/in 20km

Distance to

Agent

Negative Shock 0.152 0.122 0.148 -0.176 0.619***

[0.152] [0.153] [0.160] [0.140] [0.203]

Agents -0.022 -0.003 0.018 -0.002 0.051

[0.039] [0.031] [0.024] [0.006] [0.054]

Agents*Shock 0.055*** 0.050*** 0.021** -0.002 -0.058***

[0.019] [0.015] [0.010] [0.005] [0.019]

Jack - M-PESA

Mechanisms

• Consumption smoothing could be effected through

– Remittances

– Savings

– Information/communication

• We find remittances are the dominant factor

– More likely, More often, More

– Larger network Jack - M-PESA

Charity and Reciprocity in Mobile Phone-‐

Based Giving: Evidence from Rwanda

Joshua Blumenstock, University of Washington

joint with Marcel Fafchamps (Oxford) & Nathan Eagle (Santa Fe Institute)

2 of 17

Context

September 2012 Joshua Blumenstock ([email protected])

The “Mobile Phone Revolution” 3.5 billion subscribers in developing countries Mobile Money: $200 million sent per day in Kenya 1.7 billion “unbanked” phone owners

0

10

20

30

40

50

60

70

80

90

100

2000 2002 2004 2006 2008 2010

Phones per 100 people

Rwanda USA

3 of 17

Background


Limited evidence on economic impacts of mobile phones

Published work focused on prices and markets Jensen (2007), Aker (2010), Klonner and Nolen (2008)

Small set of unpublished studies explore other services Risk sharing and remittances (Jack & Suri, 2012) Household decision-‐making (Aker et al, 2012) Communication between counter insurgents and citizens (Shapiro & Weidemann, 2012) Migration (Aker et al, 2012) Handful of others…

Several ongoing RCT-‐based studies Understand determinants of adoption and use Impact of Mobile-‐based products and services

Savings, payments, insurance, m-‐Health, monitoring,

4 of 17

This Talk: Takeaways


Understanding the role and importance of phone-‐based transfers in Rwanda

1. Empirical evidence on Mobile Money precursor Observe entire universe mobile phone activity in Rwanda Vast disparities in use and access to technology

2. Used for intra-‐national remittances and risk sharing Cf. Jack & Suri (2012) Vs, “traditional” methods:

Distance: Udry (1994), Fafchamps & Gubert (2007), Kurosaki & Fafchamps (2002)

Covariate vs. idiosyncratic shocks: Townsend(1995), de Vreyer(2010), Gine & Yang (2009)

3. Provides insight into motives for risk sharing Cf. Leider et al. (2009), Ligon & Schechter (2011), Cabral (2011)

5 of 17

Data: Anonymous Phone Usage


Records from of all phone-‐based activity, 2005-‐2009 10 terabytes of data 1.4 millions individuals, 4 years Every call, SMS, …, and “Mobile Money” transaction

Panel A: Aggregate traffic Number of phone calls ~10 billion Number of unique users ~1 million Number of “Mobile Money” transfers ~10 million Number of “Mobile Money” dyads ~1 million Panel B: Basics of MM use Transactions per subscriber 6.05 Average distance per transaction (km) 13.51 Average transaction value (RWF) 223.58

6 of 17

Data: Demographics


Some info can be inferred Phone surveys to fill in gaps

2,200 phone interviews (Rwanda) ~80 questions, 20-‐30 minutes (Details) Derive “wealth index” for each subscriber

7 of 17

Demographics of phone access & use


8 of 17 September 2012 Joshua Blumenstock ([email protected])

Motivating Observation: Transfers and Disasters

Lake Kivu Earthquake

9 of 17

Identifying affected individuals


Measuring location of individual i on day t Only have intermittent, approximate location

10 of 17

Measuring the earthquake’s impact


Empirical questions 1. How much is sent?

rt = 1 1 Shockrt + t + r + rt

2. Who benefits? irt = 4 + 4 (Ri*Shockirt) + 4 NearEpicenterirt + t+ i + irt

3. Why is it sent? Charity:

Reciprocity:

Details

( ) ( )it i it ijt j jt ijtU uu x x

1single period utility

continuation value of relationship

( ) ( ) [ ( ) ( )]s tit i it ijt j jt ijt i is ijs j js ijs

s tU u x u x E u x u x

11 of 17

(1) (2) (3) (4) District Cell Tower Subscriber Dyad

Earthquake Shock 14169*** 2832*** 9.48*** 11.92*** (1951.30) (177.02) (0.74) (0.59)

Near epicenter 1.256*** 1.073*** (0.187) (0.39)

Day Dummies Yes Yes Yes Yes

Fixed Effects District Tower Subscriber Dyad Unconditional mean 19006.940 2436.192 5.900 3.692 Unconditional mean (earthquake region)

6355.942 1245.27 3.770 3.190

N 1800 16020 6619440 10566000 R2 0.904 0.630 0.052 0.056

Results: How much is sent?


Notes: Outcome is gross airtime received by affected district/tower/subscriber. “Earthquake shock” takes value 1 for people near epicenter of the day of the earthquake. “Near epicenter” is defined as towers 20 miles of the epicenter. Results hold with “near epicenter” re-‐defined anywhere in interval 10– 50 miles.

12 of 17

Results: How much is sent?


Total effect is small: 42,000 RWF = $84 USD (Much larger effect on calls: $2,400 USD)

Consider growth of network 400-‐fold increase in # users since 2/2008 $25,000 -‐ $33,000 projected today $11 million projected in Kenya

What benefit? Avg balance = $0.10 32% of users had < $0.01 on account

13 of 17

Results: Who Benefits?


Heterogeneity The wealthy receive more (but are not more likely to send) As do individuals with more contacts, connections to Kigali Transfers occur between “reciprocal” pairs (details)

Normally: i is less likely to send to j if j sent to i in past After quake: i is more likely to send to j

Partial Interpretation Predicted: Charity

Predicted: Reciprocity Actual ( 4)

ij / xi Wealth of i (recipient) Negative Positive Positive

ij / Tijt Past j to i transfers Positive Negative Negative

ij / Dij Geographic distance -‐-‐ Negative Negative

ij / xj Wealth of j (sender) Positive -‐-‐ -‐-‐

ij / Sij Social proximity of i and j Positive Positive Positive

14 of 17 September 2012 Joshua Blumenstock ([email protected])

Results: Sending money over distance Transfers come from 20km-‐120km away Rwandans have limited alternatives for transfer (details)

15 of 17

Summary


Empirical results Mobile Money sent in response to shocks Benefits are heterogeneous Transfers more consistent with reciprocity (not charity)

Results in context Early evidence of how and why Mobile Money (MM) can be used to for risk sharing But no direct evidence on welfare effects (cf. Jack & Suri)

16 of 17

Policy Implications


1. Immediately after launch – and while still very rudimentary – transfers used for risk sharing Good news: long distances, covariate shocks Bad news: Benefits accrue to the “elite”

2. Understand existing disparities in deciding how to target/subsidize expansion of network

3. Leverage novel forms of data in policy design and evaluation Use phones to identify people victims in need, transmit MM “Digital footprints” to measure poverty, labor mobility, migration, … Other opportunities abound!

The end. Supplemental slides follow

risk sharing and the economics of m-pesa payment... · risk sharing and the economics of m-pesa...

Documents