The Economics and Psychology of Poverty Dean Karlan

1. May 18, 2012
TYING
ODYSSEUS
TO
THE
MAST:
EVIDENCE
FROMA
COMMITMENT
SAVINGS
PRODUCT
IN
THE
PHILIPPINES
AND
NEW
YORK
CITY

2. “…but
you
must
bind
me
hard
and
fast,
so
that
I
cannot
s6r
from
the
spot
where
you
will
stand
me…
and
if
I
beg
you
to
release
me,
you
must
6ghten
and
add
to
my
bonds.”
-­‐-­‐-­‐
The
Odyssey

3. Outline
1. Mo;va;ons
2. The
SEED
Savings
Product
3. The
Super
Saver
CD
Product
4. “Net
PorJolio”
Accounts

4. Mo6va6ons
o
People
tend
to
make
bad
decisions
(mistakes)
in
par6cular,
predictable
direc6ons
o
A
“Nudge”
helps
guide
people
to
decisions
they
themselves
would
say
they
want
Common
Cogni6ve
Biases
Limited
ARen6on
Planning
Fallacy
Present
Bias
(Self
Control)
Exponen6al
Growth
Bias

5. Why
does
it
maRer?
•
SeXng
goals,
mee6ng
goals
and
s6cking
to
plans
is
hard,
psychologically
and
cogni6vely
o
Staying
out
of
debt
o
DraYing
and
s6cking
to
a
budget
o
Get
started
and
con6nue
saving
•
Leads
to…
o
…
low
household
savings
rates,
low
net
worth,
low
financial
resiliency
o
…
high
debt
loads
and
debt
reliance
in
emergencies
•
Biases
are
not
limited
to
low
and
middle
income
consumers
o
But
LMI
consumers
have
less
room
for
mistakes

6. Commitment
Contracts
• Make
your
vices
more
expensive
• Make
your
virtues
cheaper
• The
obvious:
o Savings
o Weight
loss
o Smoking
o Exercise

7. s6ckK.com
The
less
than
obvious
(actual
contracts!)
Smoke
less
weed.
No
emailing
or
talking
to
him
un6l
at
least
the
morning
of
Jan
7
Speak
more
slowly
to
foreigners
in
NYC
No
More
Da6ng
Losers…
and
Susie
is
my
Referee
Monday
through
Friday.

8. Outline
1. Mo6va6ons
2. The
SEED
Savings
Product
3. The
Super
Saver
CD
Product
4. “Net
PorJolio”
Accounts

9. SEED:
A
Commitment
Savings
Product
in
the
Philippines
• “SEED”
for
the
Green
Bank
of
Caraga
in
Mindanao
• Simplest
“commitment:”
no
withdrawals
un6l
goal
reached
• 28%
opened
an
account
• 300%
increase
in
savings
for
those
who
opened
an
account
• Similar
to
“Christmas
Clubs”
in
the
USA,
popular
in
the
past

10. SEED:
A
Commitment
Savings
Product
in
the
Philippines
Freq
Percent of SEED Clients
Offered
SEED
Account
Opened
SEED
Account
710
202
AMOUNT
GOAL
62
31%
$10
-­‐
$200
39
19%
$200
-­‐
$400
15
7%
$400
-­‐
$600
5
2%
>$600
3
1%
DATE
GOAL
140
2
-­‐
6
months
51
7
-­‐
12
months
62
13
-­‐
24
months
16
11
69%
25%
31%
8%
5%
>
24
months
BOX
167
83%
WITHOUT
BOX
35
17%
SEED
Take-­‐up
Op6ons

12. The
Commitment
Contract

13. Impact
• All
results
rela6ve
to
a
control
group.
– Why?
– Defini;on
of
Impact
• 80%
increase
in
savings
aYer
one
year
for
all
offered
• ~300%
increase
for
those
who
take-­‐up
• Improve
power
in
household
– Compare
to
credit

14. Magnitude
in
Real
Dollars
• Doctor’s
visit:
150
pesos
• Public
school
fees
are
150
pesos/
year,
plus
~200
pesos/month
for
special
projects
• 1
month
supply
of
rice
for
a
family
of
5:
1000
pesos

15. Outline
1. Mo6va6ons
2. The
SEED
Savings
Product
3. The
Super
Saver
CD
Product
4. “Net
PorJolio”
Accounts
5. Next
Steps

16. Super
Saver
CD
Commitment:
Savings
Goal
o Iden6fy
savings
goal
o Select
maturity
date
(by
when
do
you
want
to
save
this
amount)
o Frequency
of
deposits
(weekly,
monthly)
Balance-­‐building
o Minimum
ini6al
deposit
o Members
then
make
contribu6ons
in
order
to
build
savings
balance
and
reach
savings
goal
to
which
s/he
commiRed
o Does
not
require
full
lump
sum
in
order
to
open
the
CD
Compe;;ve
dividend
rate
o Varied
during
course
of
study
Minimum
ini;al
deposit
o $15
Maximum
account
balance
per
member
o $10,000
Fixed
maturity
term
o 3
months,
6
months,
12
months,
18
months
o Maturity
term
is
chosen
by
the
member
o CD
closes
on
maturity
date
(or
next
business
day)
and
all
funds
are
transferred
to
the
member’s
main
account
Contribu;ons
o Automa6c
or
manual
contribu6ons
o Automa6c
transfers
from
another
NTFCU
account
or
external
account
(switches
to
manual
if
specified
account
has
insufficient
funds)
o Manual
if
member
makes
them
in
person
or
by
mail
o NTFCU
recommends
contribu6on
amount
so
member
can
reach
goal
Withdrawals
and
Penal;es
o Penalty
for
withdrawal
before
reaching
commiRed
goal
is
forfeited
dividends
and
forfeited
$15
of
ini6al
deposit
o If
member
reaches
maturity
date
but
does
not
meet
savings
goal,
the
member
only
forfeits
dividends
upon
withdrawal
o If
savings
goal
has
been
met,
there
is
no
penalty
upon
withdrawal

17. Sample
crea6on
o
All
par6cipants
are
members
of
Neighborhood
Trust
Federal
Credit
Union
o
Par6cipants
were
enrolled
in
the
study
via
a
request
to
complete
a
baseline
survey
in
the
credit
union
o
Upon
survey
comple6on,
the
electronic
survey
tool
created
a
random
treatment
assignment:
•
1/3
-­‐
Super
Saver
CD
•
1/3
-­‐
Financial
Counseling
•
1/3
-­‐
Control
Group
(survey
only)
o
Enrolled
1167
credit
union
members
between
March
2010
and
May
2011
(staff
turnover
resulted
in
12
months
of
enrollment
during
this
period)
Full
Sample
Control
Super
Saver
CD
Offer
Counseling
Offer
1167
389
381
397
Treatment
Assignments

18. Sample
characteris6cs
Demographics
o Annual
Income:
55%
less
than
$20,000
,
86%
less
than
$40,000
o Educa6on:
51%
high
school
or
less,
43%
some
college,
7%
bachelors
degree
o Gender:
67%
female
o Mean
Age:
49
o Spanish
Language
Surveys:
74%

19. Sample
characteris6cs,
cont’d
Financial
Situa;on
o Household
Financial
Situa6on:
24%
“good
or
beRer”,
40%
“ok”,
36%
“not
very
good"
or
“bad“
o Financial
Distress:
19%
unable
to
make
rent,
u6li6es,
or
mortgage;
14%
skipped
meals;
24%
turned
down
for
credit;
40%
discouraged
from
applying
o Sketchy
credit:
18%
have
refund
an6cipa6on
loan,
payday
loan,
auto
6tle
loan,
pawn
loan,
loan-­‐shark
loan,
or
rent-­‐to-­‐own
arrangement

20. Sample
characteris6cs,
cont’d
Behavioral
Biases
o Self-­‐control:
regrets
spending
lacks
financial
discipline
48%
“strongly
agree”;
80%
agree
at
least
somewhat
o Exponen6al
Growth
Bias
(savings
compounding):
48%
underes;mate,
27%
linear
approxima6on,
7%
missing/don’t
know,
13%
overes6mate,
5%
correct
o Pa6ence
over
different
6me
horizons:
9%
“standard”
reversal
(impa6ent
now,
more-­‐pa6ent
in
future),
3%
nonstandard
reversal,
15%
always
impa6ent,
69%
always
pa6ent

21. Take-­‐up
Stats
o Takeup:
21.3%
of
individuals
take-­‐up
Super
Saver
CD
product
(N=381)
o Maturity
Term:
mean
=
13.6
months,
median
=
18
months,
N=
81
o Goal:
mean
=
$1701
(sd
=
2046),
median
=
$
1000,
N
=
81
o Counseling
takeup:
15.9%
showed
up
to
appointment,
while
48.9%
signed
up
(room
to
improve
on
this
gap)

22. CD
Take-­‐up
Analysis:
Demand
Model
o Women
were
11%
more
likely
to
take
up
than
men
o Lower-­‐income
respondents
were
WAY
more
likely
to
take
up.
o Basically,
take-­‐up
very
low
amongst
>$60k,
and
much
higher
among
everyone
else
• So
(certain)
demographics
may
be
useful
for
direct
marke6ng/messaging
o Demand
for
the
CD
is
correlated
with
individual-­‐level
proxies
for
being
“behavioral”.
o Surprisingly,
strongest
results
here
is
that
guys
who
understand
compounding
are
much
less
likely
to
take-­‐up.
Are
these
guys…
o Yield-­‐chasers?
o Rela6vely
sophis6cated,
or
ra6onal?

23. CD
usage
stats
o Percent
of
goal
reached
at
closing:
mean
=
70.3%,
median
=
57%
N=70
o Balances
at
closing:
mean
=
$910,
median
=
$417
o About
30%
of
members
reached
goal…
o Middle
of
pack
compared
to
other
IPA
studies
on
commitment
savings

24. Descrip6ve
Results
for
CD
(not
using
randomiza6on)

25. Measuring
impacts
of
CD:
On
what?
How?
On
what?
1.
Credit
union
balances
o Using
various
defini6ons
of
balances
o What’s
counted
(checking,
net
assets,
etc.)
o How
(maximum
or
average;
dealing
with
outliers,
etc.)
o Measured
at
quarter-­‐end
in
each
of
four
quarters
aYer
someone
enters
sample
o So
snapshots
only:
no
flows
2.
1-­‐year-­‐aer
credit
report
outcomes:
score,
card
balances,
u6liza6on,
delinquencies
How?
o Use
treatment
offer
(inten6on-­‐to-­‐treat)
to
iden6fy
causal
effect
of
CD
or
counseling
treatment.
o I.e.,
compare
outcomes
across
the
3
different
arms
(control,
CD
offer,
counseling
offer),
where
a
member’s
arm
is
defined
based
on
the
ini6al
assignment,
not
based
on
what
she
actually
got

26. Results:
Treatment
Effects
of
Super
Saver
CD
• On
balances,
measured
various
ways:
Imprecisely
es6mated
zeros
• On
credit
report
variables:
Imprecisely
es6mated
zeros,
except:
– Can
rule
out
large
effects
on
credit
scores
in
either
direc6on:
not
much
ac6on
here

27. Treatment
effects
on
savings
balances
Table
3:
Treatment
Effects
on
Savings
Balances
(Super
Saver
CD
Account
Balances
+
Savings
Account
Balances)
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent
Variable:
Maximum
End
of
Quarter
Balance
Maximum
End
of
Quarter
Balance
Log
(Max
End
of
Quarter
Balance
+
1)
Mean
End
of
Quarter
Balance
Mean
End
of
Quarter
Balance
Log
(Mean
End
of
Quarter
Balance
+
1)
Level
Balance
at
End
of
Quarter
Log
(Level
Balance
at
End
of
Quarter
+
1)
Estimator: OLS Median OLS OLS Median OLS OLS OLS
Savings
Treatment -­‐146.4 48.71 0.0821 -­‐131.7 29.73* 0.0790 -­‐21.73 0.197
(242.0) (34.03) (0.165) (161.5) (16.67) (0.157) (190.4) (0.222)
Counseling
Treatment 339.4 38.35 -­‐0.134 159.8 16.68 -­‐0.166 247.2 0.0515
(408.9) (34.32) (0.176) (283.6) (16.34) (0.169) (343.4) (0.220)
Number
of
Observations 1167 1167 1165 1167 1167 1161 5767 5747
R-­‐Squared 0.178 0.132 0.255 0.141 0.003 0.069
Dependent
Variable
Baseline
Mean 819.02 819.02 819.02 819.02 819.02 819.02 819.02 819.02
[Median]
[46.34]
[46.34]
[46.34]
[46.34]
[46.34]
[46.34]
[46.34]
[46.34]
Control
variables
included:
Timing
of
Entry
into
Study Yes Yes Yes Yes Yes Yes No No
Timing
of
Observation No No No No No No Yes Yes
Baseline
Dependent
Variable Yes Yes Yes Yes Yes Yes No No
Individual
FEs No No No No No No Yes Yes
Notes
Standard
errors
in
parentheses,
Huber-­‐White
for
intent-­‐to-­‐treat
specifications,
clustered
on
individual
for
fixed
effect
specifications.
Sample
sizes
differ
for
regressions
with
logged
dependent
variables
because
of
some
negative
values
for
the
level
dependent
variables.
Maximum
or
mean
savings
balance
is
the
dependent
variable
in
cross-­‐sectional
regressions,
and
is
defined
over
4
observations
taken
at
treatment-­‐quarter
end
dates;
level
balance
is
the
dependent
variable
in
panel
regressions,
and
is
simply
the
observed
balance
levels
at
treatment-­‐quarter
end
dates.
All
regressions
also
control
for
baseline
balances,
and
for
month-­‐year
of
entry
(and
hence
treatment
assignment)
into
our
sample,
except
in
the
individual
fixed
effects
specifications,
where
instead
we
control
for
the
month-­‐year
of
each
observation.
*p<0.10
**
p<0.05
***
p<0.01.

28. Treatment
effects
on
credit
scores
Table
4:
Treatment
Effects
on
Credit
Scores
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Dependent
Variable:
Has
Report
and
Credit
Score
(dummy)
Credit
Score
(level)
Active
Trade
Lines
(count)
Credit
Card
Balance
(level)
Credit
Card
Balance
(level)
Credit
Card
Balance
(log
level)
Credit
Utilization
(level)
Negative
Trades
(count)
Has
Delinquent
or
90
Days
Past
Due
Account
(dummy)
Delinquent
or
90
Days
Past
Due
Account
(count)
Estimator: OLS OLS OLS OLS Median OLS OLS OLS OLS OLS
Savings
Treatment 0.054 -­‐3.183 -­‐0.100 -­‐656.047* -­‐94.000 -­‐0.315 0.015 0.055 0.004 0.657
(0.034) (7.675) (0.430) (368.444) (133.029) (0.309) (0.034) (0.212) (0.040) (0.586)
Counseling
Treatment 0.026 -­‐7.996 0.670 92.626 -­‐94.000 -­‐0.289 -­‐0.014 0.117 0.040 1.044*
(0.034) (7.513) (0.460) (474.450) (139.580) (0.313) (0.033) (0.216) (0.040) (0.596)
Number
of
Observations 1,167 780 921 921 921 921 669 921 921 921
R-­‐Squared 0.015 0.009 0.018 0.029 0.009 0.018 0.004 0.014 0.023 0.018
Dependent
Variable
Baseline
Mean 0.64 652.5 4.62 2696.2 2696.2 4.68 0.44 1.45 0.48 3.55
[Median] [1.00] [656.50] [3.00] [523.00] [523.00] [6.26] [0.41] [0.00] [0.00] [0.00]
Control
variables
included:
Timing Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Huber-­‐White
standard
errors
in
parentheses.
Median
regression
reports
bootstrapped
standard
errors
with
1000
repititions.
All
regressions
control
for
month
of
entry
into
sample.
Sample
size
differs
because
credit
reports,
credit
scores,
and
utilization
rates
are
unavailable
for
some
individuals.
*p<0.10
**
p<0.05
***
p<0.01.
Notes:

29. Next
steps:
Future
Studies?
Improving
commitment
features?
• Offer
op6on
of
stronger
commitment
• Decision
aids:
(are
people
failing
because
they
overshoot?)
• How
important
is
goal-­‐seXng
per
se?
(test
a
pure
goal-­‐seXng
treatment,
no
commitment)
• Some
exit
interviews
with
members
who
held
to
maturity
but
didn’t
reach
goal
would
help
generate
ideas
Improving
counseling
sign-­‐up
-­‐>
take-­‐up?
o Offer
commitment?
(e.g.,
performance
bond
for
showing
up)
o Change
throughput
process?
(E.g.,
way
to
offer
intake
session
on-­‐demand?)
o Follow-­‐up
messaging?
o Why
does
month-­‐of-­‐entry
maRer
so
much?
Counseling
content
(and
reinforcing
it)?
• Should
credit
scores
be
moving?
• Failure
to
complete
follow-­‐up
tasks?
o Treat
these?
o Delegate
to
counselor?

30. Outline
1. Mo6va6ons
2. The
SEED
Savings
Product
3. The
Super
Saver
CD
Product
4. “Net
Por]olio”
Accounts

31. The
Money
Pump
• Common
prac6ce:
– Someone
in
debt.
Teach
them
to
save.
Huge
long
term
benefits.
Is
this
right?
• What
is
the
“right”
financial
answer?
– Borrow
high
and
save
low?
No.
– So
why
hold
both?
• Liquidity
• Limited
liability
(op6on
for
default/bankruptcy)
• Transac6on
costs
• Behavioral
– Mental
accoun6ng
– Lack
of
knowledge/awareness,
ie,
simply
put
“mistakes”
– Social
norms
– What
else?

32. The
PorJolio
Account
• Banks
serve
two
roles:
– Intermedia6on
of
capital
• Intermedia6on
of
capital
within
a
person
at
one
point
in
6me
is
not
wriRen
down
as
a
model
of
why
banks
exist
in
ANY
textbook.
– Facilitate
payments
• Net
porJolio
account:
– Home
equity
example,
one
account
– Add
on
mental
accoun6ng
features
necessary
(?)

34. Escalator:
Funtheory.com

35. Thank
you!
Dean
Karlan
dean.karlan@yale.edu
hRp://www.poverty-­‐ac6on.org
hRp://karlan.yale.edu