19
Endnotes
1 “Tax filers” are not the entire US population. In 2011, for
example, around 10 percent of the population may not have
been represented in any tax filing document, because their
households did not file a return with the IRS. Using other
administrative information, Cilke (2014) finds evidence that
most of the income of these non-filers comes from government
transfers. Nonetheless, for the small minority of the US
population who are non-filers, we do not know if they would
have received refund payments if they had filed, nor how
any such payments would have aected their out-of-pocket
healthcare spending behavior.
2 The JPMCI HOSP data asset was constructed using a sample
of de-identified core Chase customers for whom we observe
financial attributes, including out-of-pocket healthcare spending
between 2013 and 2016. For the purposes of our research, the
unit of analysis was the primary account holder. We focused
on accounts held by adults aged 18 to 64, as adults 65 and
older were more likely to make payments using paper checks,
which we could not categorize. To provide better visibility into
income and spending, we selected accounts which had at least
five checking account outflows each month, at least $5,000 in
take-home income each year, and used paper checks, cash, and
non-Chase credit cards for less than 50 percent of their total
spending. The JPMCI HOSP data asset includes customers who
resided within the 23 states in which JPMorgan Chase has a
retail branch presence. We re-weighted our population to reflect
the joint age and income distribution among the 18-64 year old
population within each state. See Farrell and Greig (2017a) for a
full description of the JPMCI HOSP data asset.
3 Other eorts to estimate the impact of tax refunds on consumer
spending have done so on a monthly basis and documented
higher total spending and specifically durable spending in
February among families eligible for the Earned Income Tax
Credit (Barrow and McGranahan, 2000). A number of studies
have measured the impacts of changes in tax rebates on
household spending with higher frequency. See Parker (2017)
and Broda and Parker (2014) for recent summaries of this
literature and evidence using Nielson Consumer Panel that
weekly household spending increased by 9-10 percent after
receiving the 2008 Economic Stimulus Payment. Notably, Baugh
et al. (2014), based on daily transaction data, provide evidence
that in the week following receipt of their tax refund, households
increase their restaurant spending by 8 percent, retail spending
by 12 percent and, ATM withdrawals by 16 percent.
4 The seasonality of influenza, which is a significant driver of
healthcare costs (Molinari, Ortega-Sanchez, et al., 2007), is
closely tracked by the US Centers for Disease Control and
Prevention. In the 2016/2017 season, influenza activity peaked
in late February, and prevalence fell sharply starting in the third
week of March (Blanton, Alabi, et al., 2017). Nationally, inpatient
discharges are consistently highest in March (NCHS, 2010) and
the daily rate of outpatient visits to hospitals in the state of New
York peak in March (NYSDH, 2016).
5 The average value of all tax refunds received in a year in the
JPMCI sample was $3,100; this includes directly deposited
federal and state tax refunds. This is roughly comparable
to national estimates. The average federal tax refund (i.e.,
not including state refunds) was $2,860 ($2,995 for directly
deposited tax refunds) in 2016, $2,797 ($2,957 for directly
deposited tax refunds) in 2015, and $2,792 ($2,918 for directly
deposited tax refunds) (IRS, 2017a; IRS, 2017b).
6 The cyclicality in the unadjusted (green) series in Figure 6 is
driven by the fact that healthcare spending on weekdays is
naturally elevated relative to weekends, and the fact that the IRS
does not distribute tax refund payments on weekends. As a
result, day 0 is a weekday for all 1.2 million accounts in our
sample, which therefore means that days 0+/- 7, 0+/- 14, and so
on also fall on weekdays for 100 percent of the sample. By
contrast, days 4+/-7, 4+/-14, and so on fall on weekends for 48
percent of the sample. Therefore, we compute the weekday-
adjusted (blue) series in Figure 6 as follows, for each day t:
Where x
t
is average out-of-pocket healthcare expenditure per
account on day t, x
p
is “typical” average daily expenditure per
account (where “typical” is identified by the 100 days prior to
the tax refund payment), and r
t
is the ratio of the fraction of
accounts for whom day t is a weekday to the “typical” fraction of
account-days that fall on weekdays (i.e., the 100 days prior to
the tax refund payment). Therefore, if day t is more likely to fall
on a weekday than is typical, then r
t
>1, so we adjust the average
for that day downward by a proportion of typical expenditure.
Conversely, if it is more likely to fall on a weekend than typical,
then r
t
<1, so we adjust the average for that day upward. Based
on this, we compute “tax refund-triggered additional spending”
over any period between day s and day t by:
Or, equivalently:
This reflects the fact that any dierence in average expenditure
in the period from day s to day t compared with an equivalent
number of days during the pre-refund period might be an
artifact of dierences in weekday versus weekend composition.
We use the r
τ
x
p
term in the summation above to sweep out that
artifactual component. Finally, we note that the adjusted (green)
series in Figure 6 is countercyclical with the unadjusted (blue)
series during the period prior to the refund payment. This
indicates that this approach somewhat over-corrects, in that it