Consumption and Debt Response to Unanticipated Income Shocks

Discussion of
Consumption and Debt Responses to
Unanticipated Income Shocks: Evidence
from a Natural Experiment in Singapore
By Sumit Agarwal and Wenlan Qian
AEA Annual Meetings
Philadelphia PA, January 2013
Jonathan A. Parker
MIT Sloan
A. Summary
B. Compare to Broda Parker (2012)
C. Comments/concerns
A. Summary
o Growth dividend program in 2010
Payment to Singapore citizens not foreigners (40% of pop)
Amount based on 2010 income and Dec. 2010 home value
Payments avg. 12% of avg. monthly income
Other programs also; more to low income & age>35, medical
account grants to older than 45, income tax rebate
o Bank data: sample of 100,000+ individuals
• (Unnamed:) bank has 80% of population as customers
• Monthly spending credit and debit cards: 30% of PCE
• Income, nationality, ethnicity, occupation, zip
o Infer amount of payment from income
o Using propensity score matching, compare spending of
recipients to non-recipients by month . . .
Average log spending
(matched sample)
Main findings
Cumulative change from announcement, disbursement in month 2
Total cumulative response: 90% of payment
On announcement spending more on credit cards
On disbursement and shortly thereafter more on debit cards
No pre-treatment effects
• Individuals with low account balances spent
only after disbursement and mostly by debit
• Individuals with high account balances spent
only on credit (stat. insig.)
• Low income spend more than high income
(stat. insig.)
• Low credit card limit spend more
• Men spend 100%, women 62% (stat. insig.)
Compare to Broda Parker (2012)
Key difference: BP use random timing not
common date of announcement/disbursement
1. Stimulus payment arrival causes jump up in
weekly spending, then smaller but
persistently higher level
2. No detectable spending when people learn
about stimulus payment
3. Concentrated among those with two months
income “available in cash, bank accounts, or
easily accessible funds”
Weekly spending
1.4% of ESP spent
on UPC household
No pre-treatment
Rapid dropoff in
spending rate, but
spending over 7
Consistent with
lower Parker,
Souleles, Johnson,
Spending upon learning about the ESP
Weekly spending
Treatment effect by enough liquid funds
B. Comments/concerns
These comments are concerns that are
unlikely to change the qualitative results but
might change the quantitative findings
1. Identification from nationality and
propensity score matching
2. Identification from payment amount
3. Quantitative implications
4. Individuals vs. households
5. Substantive implications
1. Identification from nationality and
propensity score matching (PSM)
Key identifying assumption: spending by foreigners is like
spending by citizens except due to policy
• PSM: “conditional on predictors of treatment status”
• Can perfectly predict treatment vs. control status
– No foreigners in treatment, no citizens in control
– Perfect prediction would violate common support assumption
Concern: different religion/culture and seasonal expenditures
– Chinese new year is February 3 2011 (year of Rabbit)
– Singaporeans have British culture -- drive of the correct side of
the street
Ethnicity control goes a long way towards dealing with these
– Plot propensity score densities of treated and untreated
– If differences between the maxima and minima of the density
distributions, remove cases that lie outside the support of the
other distribution. Estimates LATE only within overlap sample.
2. Identification from payment
Key identifying assumption: spending by individuals with
different incomes and home values is alike except for policy
• Propensity score matching in the paper does not address
this at all
Concern: high income households have wealth/income
increases relative to low income households
• Particularly acute concern for the inferences involving
heterogeneity in treatment response
Paper has some robustness tests that address some of these
concerns one by one
Possible identification strategy: compare households with
similar incomes and house values but on either side of the
policy discontinuity
3. Quantitative implications
• Payouts 12% of average (18% of median)
monthly income, 1% (1.5%) of annual
• Data capture 30% of consumption
• Ratio of sample average consumption to
average income 0.15
Yet average spending response is 90% of
4. Individuals vs. households
Payment depends on income and home value
• Income: total household or individual?
• Housing total value?
Spending on cards presumably for household
Concern: spending is joint, payment is individual,
• Spouse gets funds from other government stimulus
programs that affect only citizens – response biased
• Some cards spend spouses payment: “control group”
spending rises; some cards do not spend own payment:
“treatment group spending lowered” – response biased
Estimates similar for sample of single households
5. Substantive implications
Policy: stimulus transfer leads to spending within 9
months of 90 cents on the dollar
• The opposite from Ricardian equivalence
• Given we get a lot of spending either way, should
households get to choose how to spend stimulus
instead of government?
Theory: reject LCPIH
• But some forward-looking behavior among people
that spend 90% of a transitory income
• Can the paper give us more on anticipatory
spending? What characteristics/situation predict
anticipatory spending?

similar documents