THE IMPACT OF STOCK MARKET FLUCTUATIONS ON THE MENTAL AND PHYSICAL WELL-BEING OF CHILDREN.
Cotti, Chad ; Simon, David
THE IMPACT OF STOCK MARKET FLUCTUATIONS ON THE MENTAL AND PHYSICAL WELL-BEING OF CHILDREN.
I. INTRODUCTION
Severe fluctuations in the stock market could potentially have
widespread implications for the well-being of American families and
children. In particular, the U.S. stock market crash of 2008 was
extremely severe, as the Dow Jones Industrial Average (DJIA) fell over
50% from late 2007 to early 2009. The direct impacts of this collapse
were widespread, as over 60% of the U.S. population owned investment
stock at the beginning of the 2008 recession (Gallup Inc. 2011). Not
surprisingly, the substantial variation in wealth and financial
stability created by this market decline has promoted the study of many
interesting behavioral and psychological questions. This paper attempts
to quantify the impact of the stock market crash on one important
measure of family well-being: the health of children. One reason for
focusing on child health is that children are a particularly vulnerable
and relatively understudied demographic group. Studies on child health
are additionally of interest because child health is a strong predictor
for future labor market productivity (Almond and Currie 2011).
Earlier work reveals the depth and breadth of the impact of
financial market fluctuations on health outcomes and behavior, but these
papers have only focused on adult populations (e.g., Cotti, Dunn, and
Tefft 2015; Deaton 2012; Engelberg and Parsons 2016; McInerney, Mellor,
and Nicholas 2013; Schwandt 2014). In particular, researchers have found
that the market crash led to large declines in life evaluation (Deaton
2012), meaningful increases in psychological stress (McInerney, Mellor,
and Nicholas 2013), and increases in negative health/behavioral effects,
such as the health and survival of the elderly, increased adult
drinking, smoking, accidents, and sickness (Cotti, Dunn, and Tefft 2015;
Engelberg and Parsons 2016; Schwandt 2014). If there are notable impacts
on adult health and well-being, the impact of fluctuations in financial
markets may also impact the health of children. Parental stress from a
financial crisis may spill over to affect the mental and physical
well-being of children. If a stock market crash acts as a signal for
decreased expected future income then the resulting household income
effect could reduce health investments in children. Furthermore,
increases in negative parental health behaviors, such as drinking, could
weaken child health through decreased oversight and resulting injuries.
For example, early life exposure to cigarette smoke has been shown to
increase sick days from school and medical care utilization (Simon
2015). However, it remains unclear if the negative impacts identified on
adult health and behavior would carry over to children's health and
well-being. Lost future wealth (or the perception of wealth) may
conversely lead to a reduction in children being exposed to risky
activities, greater parental attention, and/or parental time
investments, all of which may improve the immediate health outcomes of
children.
In this paper, we attempt to directly address these issues by
investigating the impact of stock market fluctuations on the
contemporaneous health of children through capitalizing on the large
swings in value characterized by stock prices from 2004 to 2012. Our
primary findings suggest that large declines in the stock market harm
child health, as seen in increases in hospitalizations, decreases in
self-reported health, and increases in sick days from school. We find no
statistically significant effect of fluctuations on emotional
difficulties, though the sign on this variable is consistent with the
crash harming mental health. As we will demonstrate, this pattern of
results is robust to a myriad of robustness and falsification tests.
Our empirical specification compares fluctuations in the stock
market (measured with the DJIA index) with contemporaneous changes in
child health (measured as part of the National Health Interview Survey
[NHIS]), while independently controlling for the unemployment rate,
child demographics, and parental demographics. The coefficient on the
log Dow Jones is identified off of the sharp swings in stock value over
this time: the increase in stock prices leading up to the 2008 financial
crisis, followed by a steep decline during the financial crisis and a
subsequent rebound beginning in late 2009. A careful graphic analysis
shows that a number of different child health outcomes follow a similar
pattern as the stock market. There is no evidence that these results are
driven by preexisting trends in health before the 2008 stock market
crash. Using the Survey of Consumer Finance (SCF) we predict likely
stock holders and nonstock holders allowing us to test whether the
channel is primarily through a wealth effect of holding stock. A careful
set of falsification exercises (such as estimating the effect of leads
in the market) and robustness checks (such as controlling for aggregate
economic conditions in different ways) helps alleviate further empirical
concerns. Section II discusses background and motivation. Section III
summarizes the data. Section IV discusses the empirical methodology,
model specification, and overall identification strategy. Section V
presents the results and discusses the findings. Section VI concludes.
II. BACKGROUND AND MOTIVATION
Research on the impact of financial markets on population health is
related to a larger literature that investigates the impact of aggregate
economic conditions on health outcomes and behaviors. A study of the
impact of financial markets must consider the potential for a concurrent
influence of aggregate economic conditions, since the two series are
correlated with each other. Much of the research on aggregate economic
conditions and health focuses on changes in unemployment rates and has
found that adult health outcomes (e.g., all-cause mortality) and health
behaviors (e.g., drinking, smoking, etc.) are positively impacted by
recessionary events. (1) That being said, studies in this literature
that include variation from more recent recessions find a weaker
relationship between business cycles and mortality (Ruhm 2015; Stevens
et al. 2015). There is also heterogeneity in effects based on the cause
of death (2) and the level of geographic aggregation considered: for
example, Lindo (2015) finds that, when including recent recessions and
aggregating at the county level, the beneficial health effects of
recessions are smaller than previously found but very precisely
estimated. (3) Related research on children's health has found that
recessions lead to improvements in early life health (Dehejia and
Lleras-Muney 2004), and that the estimated coefficient of unemployment
on decreased mortality is largest for very young children relative to
adults (Stevens et al. 2015). Furthermore, the health of children
declines with a decline in male predicted employment growth, but
improves with a decline in female predicted employment growth (Page,
Schaller, and Simon 2017). In total, the work in this area has resulted
in a large but perplexing literature, with the majority of studies
suggesting that a decline in aggregate economic conditions improves
health but with the exact mechanisms for these effects remaining unclear
and with there being significant heterogeneity in the magnitude of the
impacts by time period, cause of death, age of death, and level of
geographic aggregation.
On the surface, improved health during economic downturns would
analogously suggest that health may also improve in response to a
decline in stock prices. Yet, recent research on this specific issue on
adult populations has shown the opposite effect. Specifically, during
the 2008 stock market crash, Americans reported large declines in life
evaluation (Deaton 2012); exhibited increased symptoms of depression and
poor mental health (McInerney, Mellor, and Nicholas 2013); experienced a
spike in hospitalizations for psychological disorders (Engelberg and
Parsons 2016); and engaged in more cigarette smoking, binge drinking,
and fatal car accidents involving alcohol (Cotti, Dunn, and Tefft 2015).
Recent research on stock market fluctuations has also found that
declines in the stock market causes physical and mental health to
deteriorate among the elderly (Schwandt 2014).
What could explain differences between effects of the stock market
on health and the impact of aggregate business cycle fluctuations?
Becker (2007) argues that exogenous events that impact individual
attitudes about the future will impact behavioral choices. To the extent
that fluctuations in stock indices influence expectations of future
economic conditions, then the identification of a negative relationship
between rates of depression, overall life valuation, and important
health-related behaviors with the stock market is sensible. So, a
possible explanation for the difference between the recessionary
(unemployment rate) effects and stock market effects is that, while
fluctuations in the unemployment rate may more generally capture
contemporaneous economic constraints faced by households, fluctuations
in the stock market may more predominantly convey information about the
future economic environment (Cotti, Dunn, and Tefft 2015). However, we
will consider other possibilities as well: such as the potential that
the unemployment rate proxies for changes in economic activity that
spill over to improve child health (such as through decreased pollution)
that are not present with the stock market. Ultimately, the exact
mechanism for explaining the differences between the two series remains
unclear, though we shed some light on the reason for these differences
after presenting our main results (see Section V.E).
One might hypothesize that the stock market induced changes in
psychological stress and depression among adults (Deaton 2012;
McInerney, Mellor, and Nicholas 2013) or sharp increases in negative
health outcome and poor heath behaviors among adults (Cotti, Dunn, and
Tefft 2015; Engelberg and Parsons 2016) will lead to reductions in
children's health outcomes. However, the impact on children could
also be very different from that of adults. Furthermore, the
subpopulations of children that are impacted may differ based on the
mechanisms by which market fluctuations impact behavior. For example, if
the stock market influences child outcomes due to a decrease in family
assets, then the effects are likely to be concentrated on stock holders:
particularly children from higher socioeconomic status (SES) households.
On the other hand, if a crash in the stock market impacts children
through generating stress caused by new information about health of the
economy, then this is likely to more strongly impact those children who
are most vulnerable: children from low SES backgrounds. Naturally, these
mechanisms are not mutually exclusive and both stock holders and
nonstock holders could reasonably be impacted by fluctuations in the Dow
Jones, and for different reasons. Given the emerging research on stock
market indices and negative health outcomes/behaviors, investigating how
market fluctuations impact child health outcomes, utilization, and
behavior (after accounting for contemporaneous economic conditions) is a
natural area of further study.
III. DATA
In this paper, we use repeated cross sections from 2004 to 2012 of
the NHIS to investigate the impact of stock market fluctuations on child
health. The NHIS is a large, nationally representative, monthly survey
used to track the health trends of the U.S. population. The NHIS
provides data on 34,000-40,000 families in a year. A significant
advantage of the survey is that it occurs at the household level, so we
are able to link children with the characteristics of their mothers and
fathers.
Household characteristics that we utilize include child's age,
mother's age, mother's education, race, and marital status.
Controlling for these characteristics allows us to absorb observable
differences between families that might be correlated with health. When
the child is too young or otherwise unable to participate in the survey,
answers are provided by a knowledgeable adult, which is the child's
parent over 90% of the time. We include all children between the ages of
0-17 in the sample depending on the outcome. (4)
When looking at indicators of child health, we attempt to limit
ourselves to measurable outcomes that are likely easy to measure and/or
recall, as well as address different aspects of health (e.g., physical
health, mental health, overall health, etc.). Depending on the health
measure under investigation, we make use of health outcomes available in
two of the surveys embedded in the NHIS. First, we use outcomes from the
Person-Core questionnaire which includes demographic and health data on
every member (and child) in the household. Specifically, in the
person-core questionnaire we measure an indicator of a child being
hospitalized in the past 12 months. Hospitalizations have the advantage
of being relatively objective and based on events that are unlikely to
be forgotten. (5) In addition, the person-core file has information on
parent-reported health of their children (on a scale of 1-5, with 1
being excellent and 5 being poor). Of note, roughly 50% of the sample
reports that their children are in excellent health, making
"excellent health" a natural way to bifurcate the data. The
results follow a similar pattern when we use the 1-5 point scale of
reported health as the dependent variable.
Additional information on child health outcomes is gathered in the
sample child questionnaire, where one child per household is randomly
selected and asked more detailed questions about their health and
well-being. In this survey, there is a count of the number of sick days
from school in the last 12 months. We additionally look at an outcome
that captures whether the child has experienced definite emotional
difficulties in the last 6 months.
Lastly, given that we are testing the effects of the stock market
on several outcomes, we also found it helpful to pool our outcomes
together into a single health index (similar to indexes used by Kling,
Liebman, and Katz 2007 and Anderson 2008). Such an index provides a
convenient summary of the overall pattern in child health across our
different models; additionally, the index boosts statistical power
through using information from many outcomes in a single test (Kling,
Liebman, and Katz 2007). We constructed this index by normalizing our
outcomes such that a higher value represents a negative outcome for
child health (a value of 1 in the case of our bivariate outcomes), and
such that each outcome has a mean of 0 and a standard deviation of 1. We
then took the average of each of our four outcomes to create the index.
Table 1 shows means and standard deviations for the variables used in
our paper.
For our principal measure of fluctuations in the U.S. stock market,
we selected the DJIA index, a market index constructed from the stock
prices of 30 manufacturers of industrial and consumer goods, to
summarize the market. The DJIA is highly correlated with other broad
stock market indices, for example, the NASDAQ and S&P500, and it is
the most widely cited market index in newspapers, television, and the
internet. (6) To create our main independent variable, we use the
natural log of the DJIA monthly mean daily market closing index,
aggregated by month. (7) The DJIA data series was downloaded from the
St. Louis Fed's FRED Economic Data website. (8)
In relating the DJIA index to the NHIS health outcome data, we must
recognize that the NHIS questions are generally retrospective. For
example, "hospitalizations" tracks if the child visited the
emergency room in the past 12 months. As a result, we utilize the
mid-point value of the DJIA index time series in the corresponding
12-month period (i.e., the 6-month period of the DJIA index) for our
baseline analysis. Results prove largely robust to variations on this
scheme. (9)
Lastly, we also merge in data on national unemployment rates,
extracted from the Bureau of Labor Statistics, U.S. Department of Labor,
to account for impacts of fluctuations in the unemployment rate on
health outcomes (also lagged 6 months in the same fashion as the DJIA).
(10) Figure 1 graphs the quarterly average of the log Dow Jones closing
value concurrently with the quarterly average of the monthly national
unemployment rates. It is important for our study that we separately
identify the effect of the stock market from changes in the unemployment
rate. Looking at Figure 1, while the two series are correlated, there is
also independent variation between them. Notably, the stock market
recovered much more quickly after the financial collapse, while the
unemployment rate remains high through the early 2000s. (11) Also, while
the unemployment rate was trending downwards before the financial
collapse, it did not do so at near as fast of a rate as the upward trend
in the value of the Dow Jones. Ultimately, our regression models are
testing for a similar sudden decrease followed by a rebound in child
health outcomes that occurs concurrently with changes in the stock
market.
IV. IDENTIFICATION AND EMPIRICAL STRATEGY
The large subset of this literature that considers the impacts of
recessions/expansions on health, inspired by Ruhm (2000), leverages
regional (typically state-level) variation in the intensity of
unemployment rates for identification. A similar approach is not an
option for estimating the effects of stock market fluctuations on health
outcomes, because of the purely time-series nature of variation in the
stock market index. In other words, a focus on national stock indexes
lacks a geographically assigned control group (e.g., a
"second-difference") and, as such, requires more careful
consideration of how to handle concerns about confounding factors. We
recognize that this makes a study of the health effects of the stock
market more challenging; however, these methodological issues lead us to
a broader approach to alleviating these concerns via employing a range
of careful robustness specifications and tests to improve confidence in
the baseline empirical identification.
Similar to related papers (e.g., Cotti, Dunn, and Tefft 2015;
Engelberg and Parsons 2016; McInerney, Mellor, and Nicholas 2013) we
leverage the rise and fall of the stock market as our principle
treatment variable. Specifically, we undertake an empirical approach
that is similar to the model used by Cotti, Dunn, and Tefft (2015), but
demonstrate that we can overcome concerns that the relationship is
spuriously identified by using a comprehensive series of falsification
and sensitivity tests. The baseline strategy utilized in estimating how
changes in the DJIA impact a range of child outcomes takes on versions
of the following empirical model:
(1)
[Y.sub.imy] = [[beta].sub.1] [S.sub.my] + [[beta].sub.2] [U.sub.my]
+ [[beta].sub.imy][X.sub.imy] + [[gamma].sub.m] + [[epsilon].sub.imy]
where [Y.sub.imy] is a measure of health of child i which is
reported at each year-month (my). The primary variables of interest are
represented by [S.sub.my] which summarizes the U.S. stock market.
Specifically, we define [S.sub.my] as the natural log of the monthly
average daily close of the DJIA index lagged 6-months at year-month, and
[U.sub.my] is the corresponding national unemployment rate lagged
6-months. [[beta].sub.1], is the coefficient of interest, identified off
of fluctuations in the stock market, [gamma] is a vector of month-fixed
effects. Including month of year-fixed effects helps insure that the
impacts of the stock market on child health accounts for seasonal
changes in child health. (12)
[X.sub.it] includes demographic and socioeconomic controls. Our
principle specification includes controls for child age (in categories:
0-2, 5-7, 8-11, 12-14, and 15-17), mother's age (in categories:
18-25, 26-35, 36-45, 46, and older), mother's education (high
school dropout, high school or some college, and college education or
more), race (black, white, Hispanic, and other), and an indicator for
the child's mother being married. (13) NHIS survey weights are used
in all estimates, and standard errors are clustered at the year-month
level to account for within time correlations in child health. (14)
Throughout our empirical analysis, the dependent variables are often
dichotomous indicators, such as whether the child is in excellent child
health, but can also be a level measure, such as the number of sick days
from school. For clarity, we will estimate linear models for all of our
outcomes. However, we also demonstrate that the results for our
dichotomous outcomes are robust to using probit models. Another concern
with this strategy is that if there is a downward trend in child health
prior to the stock market crash, this could cause a spurious correlation
between health and stock market fluctuations. We address these both with
a graphic analysis of the trends in health relative to the stock market
crash and by specifying controls for time in different ways as a
robustness check.
Overall, the key identifying assumptions of this empirical approach
is that our model: (a) appropriately accounts for the impact of
aggregate economic activity on health, and (b) that there are no other
trends or covariates that correlate with the stock market and health in
a way that confounds our estimates. To test the first assumption, we use
different measures of the unemployment rate such as regional
unemployment rates and gender-specific rates (as well as not including a
measure for the unemployment rate, in case unemployment itself is an
endogenous control). We also look at alternative business cycle
measures, which include measures of housing prices and the
employment-to-population ratio. In regards to assumption (b), unlike
other papers in the literature we cannot use differences in regional
variation in stock market. Instead we carefully ensure the timing,
magnitude, and robustness of the estimated treatment effects on child
health are directly related to the stock market crash of 2008.
Ideally, by controlling for [U.sub.my], we can get an estimate of
[[beta].sub.1], that captures the impact of stock market fluctuations
that is independent of effects of the unemployment rate. In practice,
these two series are likely endogenously codetermined to some degree: a
crash in the stock market could lead to declines in investment activity
that also causes the unemployment rate to increase, which in turn leads
to declines in the stock market (and so on). If both series have
inter-related effects on child health, then controlling for [U.sub.my]
could introduce bias into our estimates of [[beta].sub.1] because it
would be an "endogenous" control. (15) To address this
concern, we first note in Figure 1 that it does not notably appear that
the stock market crash is correlated with an immediate impact on
unemployment: the increase in the unemployment rate starts before the
market crash, evolves smoothly over this time, and does not appear to
deviate suddenly from its trend directly after the crash. Given these
dynamics, we consider it to be at least plausible that the impact of the
stock crash and recovery on child health is largely independent of the
more slowly and smoothly evolving unemployment rate. We formally test
for bias from including unemployment as an endogenous control by
estimating a version of Equation (1) where we drop the unemployment
rate, next we re-estimate this model but drop the stock market measure
and only include the national unemployment rate, and we then compare
these two results to our baseline. We get similar effects of the
unemployment rate and stock market regardless of whether we include both
in the same model (Table 2), estimate effects of the DJIA excluding the
unemployment rate (panel E of Table 5), or estimate the effects of the
unemployment rate excluding the DJIA (panel F of Table 5). It seems that
even if [U.sub.my] is endogenous in Equation (1), the associated bias on
child health is small and does not alter the overall conclusions of our
paper.
When considering the overall validity of our empirical approach, it
helps to consider how the literature has estimated the impact of
business cycles on health at different time periods and levels of
aggregation. Lindo (2015) uses more-disaggregated measures of economic
activity (such as at the county level) to estimate effects on health
that are smaller in magnitude, but more precisely measured. They argue
that this is because important spillover effects for neighboring
geographic areas are missed in the disaggregated model, hence reducing
the magnitude of the estimates. These findings imply that estimates that
capitalize on a national-level measure of the U.S. stock market would
provide a broad national "average" effect of the market across
individuals in our sample, but the lack of geographic variation will
reduce power and increase estimated standard errors. However, as
discussed in Section II, the literature has also estimated different
effects of business cycles over time (e.g., Lindo 2015; Ruhm 2015).
Nevertheless, an advantage of using a national time series for
identification is that inter-state migration is not a concern for
biasing our estimates; and important covariates such as the unemployment
rate are measured much more accurately than at more disaggregated levels
(such as the county), which could introduce measurement error and biases
of its own. Given that there are reasons to expect differences in
business cycle effects by time-period and aggregation level, we also
separately estimate effects of the unemployment rate on child health by
dropping the stock index (Smy) from Equation (1): allowing us to
directly document the relationship between the nationally aggregated
unemployment rates and child health outcomes.
V. RESULTS
A. Baseline Estimates
Before discussing our regression results, we first present a
graphic analysis of how our child health measures fluctuate with the log
of the Dow Jones. As discussed above, in this analysis we compare the
6-month lagged Dow Jones with retrospective reports of health, allowing
us to roughly match the changes in the two variables. Figure 2 presents
this in year-quarter time for the "excellent health" outcome.
Figure 2 shows that reported child health and the Dow Jones series
follow fairly similar patterns. From 2004 through the first quarter of
2008, reported excellent health is increasing with the log of the Dow
Jones. After the stock market crash, reported excellent health continues
to increase for several months before declining with the Dow Jones,
reaching its minimum point several months after the trough of the crash.
This is consistent with a somewhat lagged correlation between the crash
and excellent health (and is also consistent with the retrospective
nature of many of the NHIS questions). Finally, both health and the
stock market improve together in the final part of the sample. It is
particularly important to note that we do not see a downward trend in
health in the quarters leading up to the stock market crash. Such a
confounding preexisting trend would be one sign that our empirical
estimates were driven by something other than the stock market. Although
Figure 2 shows no evidence of a confounding preexisting trend, we will,
nevertheless, verify this in a formal sensitivity analysis.
Figure 3 performs the same analysis as Figure 2, but across all our
child health outcomes of interest. Many of the outcomes are more extreme
and have smaller means than excellent health status, making the graphs
noisier. We dealt with this by aggregating the health outcomes into
yearly rather than year-quarter bins, though regardless of the
aggregation method the graphs follow the same overall pattern.
Generally, across these outcomes the pattern is one such that we see
health declining around the time of the market crash and then improving
with the recovery. In the case of reported excellent health this is the
more aggregated version of Figure 2. For sick days and emotional
difficulties we also see these indicators of poor health increasing with
the market crash. It is worth noting that, relative to our other
outcomes, hospitalizations seem to be correlated less with the stock
market in this graphical analysis.
In order to more formally investigate the relationship between
market fluctuations on child health, we now switch to our regression
analysis. Table 2 reports results for a set of NHIS outcomes and their
association with stock market fluctuations. Overall, the results suggest
that declines in the stock market negatively impact child health.
Specifically, estimates suggest that during poor market performance,
overall "excellent" health status meaningfully declines.
Similarly, measures of poor health: hospitalizations, emotional
difficulties, sick days from school, and a standardized aggregate poor
health index all increase during periods of market decline (although the
estimated effect on emotional difficulties is not statistically
significant). Interestingly, these outcomes are notably consistent with
recent work studying the health effects of market declines on adults
(Engelberg and Parsons 2016; McInerney, Mellor, and Nicholas 2013). Our
estimates suggest that, during a month in which the DJIA is 10% lower,
the likelihood of hospitalization increases by 0.09 percentage points,
the likelihood of a parent reporting their child as having excellent
health status decreases by 0.46 percentage points, and a child misses an
average of 0.114 more days of school. These are meaningful effects when
considered against the mean of each value and in the context of a very
large stock market crash. For example, these estimates suggest that a
65% decline in the DJIA (as was observed between 2008 and 2009), would
result in (relative to the mean) a 10% increase in the hospitalization
rate, a 22% increase in sick days from school, and a 5% decrease in
having "excellent" overall health.
As our aim is to isolate stock market effects independent of
business cycle factors, in all baseline models we control for the
unemployment rate. (16) Many of the unemployment rate coefficients are
not precisely estimated, but when statistically significant they are
consistent with previous work reporting that individuals generally
participate in healthier behaviors as economic conditions, proxied by
unemployment, tend to worsen (e.g., Ruhm 2000).
B. Robustness
We explore the robustness of the above results to reasonable
changes in our empirical specification and estimation approach. First,
we demonstrate in Tables 3 and Al that the results are very similar when
using alternate specifications. Specifically, we recognize that in
typical quasi-experimental models it is of key importance to capture
changes over time in the outcomes that might be spuriously related to
the stock market. Therefore, in Table 3 we demonstrate the robustness of
our results to controlling for time in different ways. The first column
of Table 3 begins with our baseline specification. The second column
replaces the baseline month-fixed effects with quarter of the year-fixed
effects. The third column adds to the baseline specification a linear
time trend, and the fourth column replaces this linear time trend with
calendar year-fixed effects. While the estimates are highly robust to
the inclusion of quarter of the year and calendar year-fixed effects,
the estimated effects are weaker when we control for linear time trends.
Lastly, in the final column we show what happens when we replace
"calendar year"-fixed effects with "regime"-fixed
effects. We define each so called "regime" by dividing the
DJIA and the unemployment rate time series based on the different
relative relationships that exist between these two variables. For
example, in the first part of our sample stock prices are rising while
unemployment rates are falling, after these stock prices continue to
rise while unemployment rates now rise as well: these two periods of
relative trends represent the first two regimes we define. (17) The
results are generally quite robust to including these
"regime"-fixed effects. Overall, our estimates remain largely
unchanged and statistically significant across these different
alternative specifications. (18)
In Table A1, we also investigate the sensitivity of our results to
the inclusion of three potential endogenous covariates: mother's
insurance status, employment status, and family income. As we
systematically introduce these covariates there are only very small
changes in the coefficients on the Dow Jones, demonstrating that there
seems to be little evidence of omitted variable bias affecting our
estimates. (19)
Next, in order to help further address potential concerns with the
identification strategy outlined in Section IV, we engage in a set of
falsification tests where we lead the DJIA treatment variable in
different ways. By putting in a lead of the stock market measure, we are
essentially specifying that all fluctuations happened sooner than they
did in reality. Hence, finding similar estimates to our main results
presented in Table 2 would lend doubt to the validity of our current
findings; as it would suggest that unknowable fluctuations in the DJIA
that occur in the future had impacts on child health in the past. As can
be seen in Table 4, estimates from this falsification analysis
consistently show no relationship. These results help alleviate concerns
that our findings are spurious, and are particularly important given the
nature of the identification strategy utilized.
For our empirical strategy to be convincing, it is imperative to
identify the impact of stock market fluctuations separately from
business cycles. To the degree that the unemployment rate is a noisy or
poor measure of economic conditions, then even when controlling for the
unemployment rate the coefficient on the stock market could suffer from
omitted variables bias. To this end, in Table 5 we test the sensitivity
of our results to controlling for a range of other measures of aggregate
economic activity. If there is little change in the coefficient after
controlling for these different measures then that suggests that our
models are not biased by some unobserved factor related to business
cycle fluctuations. First, in panel A we replace the national
unemployment rate with measures of regional unemployment rates
(aggregated from state measures), (20) next in panel B we use
gender-specific national average unemployment rates, and then in panels
C and D we utilize two alternative measures of national business cycles:
the employment-to-population ratio and a housing price index. (21) In
the penultimate set of results in panel E, we completely exclude
business cycle controls. With the exception of hospitalizations, the
estimates and statistical inference presented in panels A-E of Table 5
are very similar across all specifications to those presented in Table
2. In regards to hospitalizations, we will show in the subgroup analysis
below, that the results presented in Table 2 are driven by the very
young children. Given that only a relatively small part of the sample
responds to hospitalizations this may explain why this outcome is
sensitive to some of our robustness tests. (22) In panel F of Table 5,
we remove the DJIA index and run a model using the national unemployment
rate as the only macro-measure. We talk about the relevance of this test
in Section V.E.
Importantly, all of these business cycle measures are known from
the economics literature to be strongly correlated with aggregate
economic activity. Since we see no substantial change in results from
moving from a model with no controls for economic activity across these
different measures, it is difficult to imagine that there is some
excluded covariate related to economic activity that would have a
stronger impact on our results. Still, we believe it is important to be
clear that concern about an omitted variable biasing our results cannot
be completely alleviated. An alternate approach is to attempt to sign
the bias that would result from failing to control fully for aggregate
economic activity. The literature on business cycles and health
discussed above typically finds that health is counter cyclical. To the
degree that the stock market is positively correlated with economic
activity, this suggests that failing to control fully for business
cycles would bias our results towards zero.
Lastly, we investigate the sensitivity of our results to some of
our other empirical choices. We have been using a linear specification
for our baseline estimates, so we re-estimate our primary results using
a probit (when appropriate). The results are robust to the specification
selected (see Table A2). Also, given that stock market fluctuations
occur at the national level, it may not be appropriate to assume that
errors are independent across year and month. There is no meaningful
change in statistical inference when standard errors are clustered at
the year level, although the standard errors become a bit larger (see
Table A3 for details).
C. Subgroup Analysis
In this section, we investigate whether there is heterogeneity in
our results across different subgroups. Understanding which specific
groups are responding can provide insight on the underlying mechanism
impacting behavior. Table 6 reports sample subgroup results that vary by
child's age, sex, and mother's education.
We first investigate heterogeneity depending on child age, as shown
in the first panel of Table 6. Results indicate that estimates are
relatively consistent across age groups, with the exception that
hospitalization effects seem to be isolated among very young children.
This is potentially because hospitalizations are a relatively extreme
outcome and earlier work suggests that early life is a time when
children are particularly susceptible to shocks to parents' SES
(Currie 2009).
Next, in the middle panel of Table 6, we estimate separate models
for male and female children. Results indicate similar estimated impacts
of market fluctuations across almost all outcomes. However, estimates
are statistically significant for all measures for females, and this is
only the case for sick days and health status for males. Moreover, the
female subsample is the only group to provide a statistically
significant estimate on the emotional difficulties measure, suggesting
that as the stock market declines there are increases in emotional
difficulties observed among female children. Overall, the general
estimated effect of market fluctuation on child health is the same for
both genders, but the impacts of market declines is more precisely
estimated for girls.
Lastly, we stratify our results by mother's educational
attainment. Specifically, we split the sample by children whose mother
dropped out of high school (very low human capital) versus a college
degree (high human capital). (23) In part we consider mother's
education in this context to be a proxy for a child's SES. Results
presented in the last panel of Table 6 show that, with the exception of
hospitalizations, children whose mother was a high school dropout are
impacted more during a market decline. However, an important caveat to
this discussion is that most of these estimated coefficients are not
statistically different from each other. Estimates on high SES children
do indicate that they see increases in hospitalization and sick days,
but the latter is much smaller in magnitude relative to low SES
children. Children from low SES backgrounds may be more vulnerable to
the general stress and uncertainty induced by a market crash even though
those families are less likely to hold stock. On the other hand, high
SES families are likely to have greater resources to respond to a health
shock generated by a stock market crash and this may dampen the negative
effects on child health. We investigate this more in Section V.D, where
we use data from the SCF to predict the likelihood of a family holding
stock.
In general, across these subgroups we see the strongest impact
among lower SES, female children. But, effects are observed in most
subgroups to some extent. We find it interesting that effects are
generally larger and more broadly observed among children of
low-education mothers, which is a group that one may not expect to be
explicitly impacted by stock market fluctuations as they are less likely
to own stock. While we take this issue up in more detail in the next
section, it does speak to the hypothesis that stock market fluctuations
may impact overall population health and behavior by increasing stress
and providing information about future outcomes, which could impact
personal discount rates (Becker 2007).
D. Extension: Analysis of Stock Holders versus Nonstock Holders
If stock market fluctuations are impacting child health, an open
question is what are the mechanisms? In particular, is the mechanism an
income effect from lost stock value, which would be isolated to the
segment of the population who owns stock? While a wealth effect is
intuitive, it is not necessarily what is driving our findings,
particularly if the impacts of stock market fluctuations on health are
the result of a shift in behavior caused by a change in assessment of
national wealth and stability by the general population.
In order to attempt to provide some insight into these questions,
we calculate predicted probabilities of holding stock for NHIS
respondents using the 2007 cross-section SCFs (Board of Governors of the
Federal Reserve System 2013). We regressed a dichotomous measure of any
stock holdings (24) (probit model) on SCF variables that could be mapped
to corresponding NHIS control variables. These include measures of sex,
age, education, race/ethnicity, marital status, employment status, and
income. For each NHIS child between 2004 and 2012 we predicted whether
their parents owned stock. The 2007 SCF model was assumed to be
representative of stock holdings among the population prior to the
2008-2009 stock market crash.
The top two panels of Table 7 stratify the sample by predicted
stock holders and nonstock holders. (25) Generally speaking, the
estimated relationship between the DJIA and health are similar across
both groups, although the estimates are much more precisely measured for
the nonownership group. In the third panel of Table 7, we estimate an
interaction between being a predicted stock owner and the log DJIA. This
shows explicitly that, with one exception, the estimated effects are not
statistically different across groups, and in all cases tell a similar
qualitative story. These results suggest that the mechanism by which the
market crash influenced health is one that operates broadly through the
economy, not just through a wealth effect on stock holders. At the same
time, this raises a puzzle: if the market crash produces an economy wide
treatment, why are its effects so different from our estimated impact of
business cycle fluctuations? In the next section, we consider this
question along with a closer investigation of the potential mechanisms
explaining our results.
E. Distinguishing between Aggregate Business Cycle Fluctuations and
Stock Market Shocks
Why does an increase in the national unemployment rate improve
child health while the stock market crash is harmful? (26) Following a
model of time use, we could assume unemployment rates reflect the
household's availability of employment and the prevailing wage
rate, therefore capturing changes in both income and the opportunity
cost of time. If the effect of a decreased opportunity cost of time
dominates, then parental time investments in child health could increase
with an economic downturn. (27) For the stock market, we could assume an
extreme case where the market acts purely as a signal of future economic
activity. Here, a decline in stock value signals a decrease in expected
future financial stability, leading to decreased investments in child
health. The decline in expected income could also act as a stress shock,
which may directly harm health.
Overall, this explanation for the relative difference between the
two series fits with some of our evidence. If the stock market harms
health due to it being a signal of expected future stability then it
makes sense that we see strong effects on nonstock holders as these
families are likely more vulnerable. This also matches a larger
literature which documented that the 2008 crash negatively affects
well-being potentially through stress-related channels (Deaton 2012;
Engelberg and Parsons 2016; McInerney, Mellor, and Nicholas 2013).
Likewise, earlier work has shown that adult health investments decline
with the market crash (again affecting nonstock holders potentially due
to increased stress) which could directly harm child health through
smoke exposure from greater adult smoking, adult drinking, and other
poor health habits (Cotti, Dunn, and Tefft 2015).
In our data, we can directly test whether a decline in the stock
market is associated with decreased market investments in child health.
Specifically, we put an indicator for a child having had medical care
delayed in the past 12 months due to costs on the left-hand side of
Equation (1). Our results show that a decline in the stock market is
associated with an increase in the incidence of a child's medical
care being delayed (see panel A of Table 8). (28) Importantly, this
effect of the stock market on care is consistent regardless of the
inclusion of the unemployment rate. In results not shown, we stratify by
stock holders and nonstock holders and found that the impact on delayed
care is strong for the children of nonstock holders: supporting the
hypothesis that low income families respond to increases in economic
uncertainty by making fewer health investments in children. (29)
While the above explanation is appealing, it is overly restrictive
to assume that the unemployment rate only influences health through a
time-substitution effect. Research on state level unemployment rates and
health suggest that procyclical mortality could be at least partially
due to business cycles changing economic activity in ways that are
external to a household (in the sense they do not affect labor inputs)
but that spill over to influence health. For example, the literature has
documented that increases in the unemployment rate decreases pollution
and respiratory health, decreases traffic fatalities, and improves
nursing home inputs (30) (Cotti and Tefft 2011; Heutel and Ruhm 2016;
Stevens et al. 2015). If these spillovers cause improvements in health,
and a decline in the stock market is not associated with such
spillovers, then this could explain why the two series have opposite
effects on child health. To some degree this seems intuitive: it is
difficult to imagine that the stock market is as strongly associated
with changes in pollution/traffic accidents as the national unemployment
rate. However, given the data limitations of the NHIS it is difficult to
systematically test for the role of different potential spillovers.
It also seems overly restrictive to assume that the stock market
does not have any influence on "own-household" economic
conditions. Just like child health, demand for labor inputs is likely to
decline with a decrease in the expected future strength of the economy;
and this could result in changes in wages and employment that impart
time-substitution effects of their own. We test this by estimating
another variation on Equation (1) with a child's father's
employment and hours worked last week (conditional on being employed) as
outcome variables. Similar to panel A of Table 8, we first include both
the unemployment rate and DJIA index on the right hand side and then
only include each individually. We calculate the magnitude of the stock
market crash relative to the recession by multiplying the coefficient of
each series by the peak to trough change in the Dow Jones and
unemployment rate (respectively). (31) The results (presented in panels
B and C of Table 8) show that the increase in the national unemployment
rate during the recession resulted in sizable decreases in employment
status and conditional hours worked. However, the effect of the stock
market crash on these measures was much smaller and not as robust. Thus,
the unemployment rate is a substantially stronger proxy for effects on
current household labor market inputs than the stock market. (32)
It is also possible that both the unemployment rate (in addition to
the stock market) conveys signals about the future of the economy:
making it unclear why the stock market crash would have a different
effect on health. One potential explanation is that the stock market
conveys qualitatively different (or better) information about future
economic conditions than the unemployment rates: such as which
industries are most likely to be affected. For example, the 2008 stock
market crash was related to a decline in mortgage backed securities and
the housing market: and housing is a source of financial security for
many families (even those who do not hold stock). A final potential
explanation is that stock market prices and the unemployment rate are
endogenously and jointly codetermined by each other in their
relationship to child health. If this is the case, we bias our
coefficients by controlling for both of these in our baseline model, and
the opposite effects of the two series could be driven by this bias.
However, as we show in panels E and F of Table 5, this does not seem to
be the case: we get similar effects on both series regardless of whether
we estimate the unemployment rates and stock market individually or
jointly. (33)
Ultimately, we feel that the best evidence that fits our results
suggests that the stock market crash harmed child well-being through
either a channel of stress or through discouraging investments in child
health. However, due to the complexity of how economic conditions
influence health, the mechanisms at work are not completely clear. We
believe that fully understanding these mechanisms could be a fertile
avenue for future work and one that would lead to a better understanding
of both how the stock market influences the behavior of nonstock holders
and an understanding of how and why recessions affect health.
VI. CONCLUSION
The stock market crash of 2008 caused a severe impact to households
across the United States. The impacts of a stock market crash on family
welfare and behavior has been identified for life well-being,
psychological stress, and adult health behaviors. We have attempted to
add to this literature by documenting impacts of stock market
fluctuations on a range of child outcomes, including effects on both
mental and physical health. Specifically, we show the negative effect of
a stock market crash on hospitalizations, child-reported excellent
health status, sick days from school, and an overall poor health index.
A graphic analysis suggests that our results are not driven by a
spurious preexisting trend of declining child health before the market
crash. Similarly, our regression results are robust to a range of
specifications, robustness tests, and falsification exercises.
We have additionally explored a number of mechanisms and
relationships by which the market crash could have influenced outcomes.
Widespread effects across subgroups suggests that more than just stock
holders were impacted. This lends credence to the interpretation of the
stock market crash as affecting widespread behavior, and potentially
emotional stress, through changes in information about the future health
of the economy.
APPENDIX A
TABLE A1
Impact of the 2008-2009 Stock Market Fluctuations in the DJIA
on Children's Health Outcomes
(A) Hospitalization
Log Dow Jones -0.095 * -0.010 ** -0.010 ** -0.012 **
(0.005) (0.005) (0.005) (0.005)
Unemployment -0.001 *** -0.001 *** -0.001 *** -0.002 ***
rate (0.000) (0.000) (0.000) (0.000)
N 202,445 202,445 202,445 202,445
Insurance status No Yes Yes Yes
Employment No No Yes Yes
status
Income measures No No No Yes
(B) Health Status Excellent
Log Dow Jones 0.046 *** 0.045 *** 0.045 *** 0.037 ***
(0.015) (0.015) (0.015) (0.014)
Unemployment 0.003 *** 0.003 *** 0.003 *** 0.003 ***
rate (0.001) (0.001) (0.001) (0.001)
N 202,718 202,718 202,718 202,718
Insurance status No Yes Yes Yes
Employment No No Yes Yes
status
Income measures No No No Yes
(C) School Sick Days
Log Dow Jones -1.144 *** -1.153 *** -1.161 *** -1.265 ***
(0.255) (0.255) (0.252) (0.255)
Unemployment 0.030 * 0.029 0.025 0.011
rate (0.018) (0.018) (0.018) (0.018)
N 64,710 64,710 64.710 64,710
Insurance status No Yes Yes Yes
Employment No No Yes Yes
status
Income measures No No No Yes
(D) Emotional Difficulties
Log Dow Jones -0.002 -0.002 -0.002 -0.006
(0.008) (0.008) (0.008) (0.008)
Unemployment 0.001 ** 0.001 ** 0.001 ** 0.001
rate (0.001) (0.001) (0.001) (0.001)
N 70,299 70,299 70,299 70,299
Insurance status No Yes Yes Yes
Employment No No Yes Yes
status
Income measures No No No Yes
(E) Poor Health Index
Log Dow Jones -0.085 *** -0.086 *** -0.086 *** -0.093 ***
(0.021) (0.021) (0.021) (0.021)
Unemployment 0.001 0.001 0.001 -0.001
rate (0.002) (0.002) (0.002) (0.002)
N 64.050 64,050 64,050 64,050
Insurance status No Yes Yes Yes
Employment No No Yes Yes
status
Income measures No No No Yes
Notes: All models include month-fixed effects and controls for
race, gender, child's age, mother's age, mother's education,
and mother's marital status. Robust standard errors clustered
by year-month are in parentheses.
*** p <.01; ** p <.05; * p <.1.
TABLE A2
Probit Models
Hospitalization School Health Status:
Sick Days Excellent
Ln average daily -0.092 * 0.120 **
close, DJIA (0.053) NA (0.039)
N 202,445 202,718
Emotional Health Index
Difficulties (z-Score)
Ln average daily -0.018
close, DJIA (0.075) NA
N 70,295
Notes: All models include month-fixed effects and controls for the
national unemployment rate, race, gender, child's age, mother's age,
mother's education, and mother's marital status. Robust standard
errors clustered by year-month are in parentheses.
*** p <.01; ** p < .05; * p <.l.
TABLE A3
Standard Errors Clustered at the Year Level
Hospitalization School Health Status:
Sick Days Excellent
Ln average daily -0.009 * -1.144 ** 0.046 *
close, DJIA (0.005) (0.399) (0.025)
N 202,445 64,710 202,718
Emotional Health
Difficulties Index (z-Score)
Ln average daily -0.002 -0.087 **
close, DJIA (0.005) (0.035)
N 70,299 64,050
Notes: All models include month-fixed effects and controls for the
national unemployment rate, race, gender, child's age, mother's age,
mother's education, and mother's marital status. Robust standard
errors clustered by year-month are in parentheses.
*** p <.01; ** p<.05; * p <.1.
APPENDIX B: TIMING OF THE STOCK MARKET AND CHILD HEALTH IN OUR
MODEL
In assigning the timing of the Dow Jones, notably we do not average
over the 12-month reference period of our outcome and instead (for our
baseline specification) use a 6-month lag. Because we rely on a time
series, and the stock market crash was a fairly sudden and extreme
event, averaging over 12 months washes some of the variation from the
trough of the market crash. At the same time, using the 6-month lag
captures the same median timing as a 12-month average resulting in a
time series with a similar pattern. We demonstrate this graphically
below, where we show a graph of the 6-month-lagged Dow Jones overlaid
with a graph of the 12-month-averaged Dow Jones focusing on the years
around the market crash.
As can be seen, the patterns of the two time series are very
similar to each other. The main difference between the two is that the
12-month average Dow Jones is somewhat smoother but also the apex of the
financial crisis is less distinct. The variation in the Dow Jones from
the depths of the "crash" itself is an important part of the
story, and that even though outcomes questions are asked over a 12-month
reference period, we mechanistically fail to measure the most extreme
part of the crash when averaging over this period.
That being said, Table A4 presents the robustness of the main
results to alternative ways assigning timing for the lagged Dow Jones.
The first three rows show the results for different ways of lagging the
DJIA: the 3-, 6-, and 9-month lags, respectively. While the last four
rows present the average of the Dow Jones for different windows: the
average of months 1-6, 1-12, 5-8, and 7-12. Jointly these specifications
provide different ways of capturing how the retrospective nature of the
health questions asked in the paper interact with the timing. Generally
there is not a great deal of difference in estimates across different
timing assumptions. Statistical significance remains similar as the
baseline estimates for all measures except hospitalization.
TABLE B1
Impact of the Lagged DJIA on Measures of Children's Health Outcomes
Hospitalization Health Status School Sick
Excellent Days
Log Dow Jones, -0.007 0.020 -0.996 ***
3-month lag (0.005) (0.018) (0.270)
Log Dow Jones, -0.009 * 0.046 ** -1.144 ***
6-month lag (0.005) (0.015) (0.255)
Log Dow Jones, -0.005 0.052 *** -1.153 ***
9-month lag (0.004) (0.014) (0.278)
Log Dow Jones, -0.007 0.020 -0996 **
average of past (0.006) (0.018) (0.309)
6 months
Log Dow Jones, -0.007 0.043 ** -1.398 ***
average of past (0.005) (0.016) (0.301)
12 months
Log Dow Jones, -0.008 * 0.040 ** -1.135 ***
average of past (0.005) (0.016) (0.251)
5-8 months
Log Down Jones, -0.006 0.056 *** -1.279 ***
average of past (0.005) (0.014) (0.293)
7-12 months
Mean of outcome 0.058 0.555 3.378
N 202,445 202,718 64,710
Emotional Poor
Difficulties Health Index
Log Dow Jones, 0.002 -0.066 **
3-month lag (0.008) (0.026)
Log Dow Jones, -0.002 -0.085 ***
6-month lag (0.008) (0.021)
Log Dow Jones, 0.002 -0.081 ***
9-month lag (0.007) (0.021)
Log Dow Jones, 0.002 -0.065 **
average of past (0.008) (0.028)
6 months
Log Dow Jones, 0.001 -0.094 ***
average of past (0.009) (0.024)
12 months
Log Dow Jones, -0.000 -0.079 ***
average of past (0.008) (0.021)
5-8 months
Log Down Jones, 0.002 -0.088 ***
average of past (0.008) (0.022)
7-12 months
Mean of outcome 0.051 -0.004
N 70,299 64,050
Notes: All models include month-fixed effects and controls for the
(corresponding lagged) national unemployment rate, race, gender,
child's age, mother's age, mother's education, and mother's marital
status. Robust standard errors clustered by year-month
are in parentheses.
*** p <.01; ** p < .05; * p <.1.
ABBREVIATIONS
DJIA: Dow Jones Industrial Average
NHIS: National Health Interview Survey
SCF: Survey of Consumer Finance
SES: Socioeconomic Status
REFERENCES
Almond, D., and J. Currie. "Human Capital Development before
Age 5," in Handbook of Labor Economics, Vol. 4B, edited by D. Card
and O. Ashenfelter. Amsterdam, The Netherlands: Elsevier, 2011,
1315-486.
Anderson, M. L. "Multiple Inference and Gender Differences in
the Effects of Early Intervention: A Reevaluation of the Abecedarian.
Perry Preschool, and Early Training Projects." Journal of the
American Statistical Association, 103(484), 2008, 1481-95.
Angrist, J., and J.-S. Pichke. Mostly Harmless Econometrics: An
Empiricist's Companion. Princeton, NJ: Princeton University Press,
2009.
Becker, G. S. "Health as Human Capital: Synthesis and
Extensions." Oxford Economic Papers, 59, 2007, 379-410.
Board of Governors of the Federal Reserve System. "Survey of
Consumer Finances." 2013. Accessed 2013. http://
www.federalreserve.gov/econresdata/scf/scf_2007.htm.
Charles, K. K, and P. DeCicca. "Local Labor Market
Fluctuations and Health: Is There a Connection and for Whom?"
Journal of Health Economics, 27, 2008, 1532-50.
Cotti, C, and N. Tefft. "Decomposing the Relationship between
Macroeconomic Conditions and Fatal Car Crashes during the Great
Recession: Alcohol- and Non-Alcohol-Related Accidents." The B.E.
Journal of Economic Analysis & Policy, 11(1), 2011, Art. 48.
Cotti, C., R. A. Dunn, and N. Tefft. "The Dow Is Killing Me:
Risky Health Behaviors and the Stock Market." Health Economics,
24(7), 2015, 803-21.
Currie, J. "Healthy, Wealthy, and Wise: Socioeconomic Status,
Poor Health in Childhood, and Human Capital Development." Journal
of Economic Literature, 47(1), 2009, 87-122.
Deaton, A. S. "The Financial Crisis and the Well-Being of
Americans: 2011 OEP Hicks Lecture." Oxford Economic Papers, 64,
2012, 1-26.
Dehejia, R., and A. Lleras-Muney. "Booms, Busts, and
Babies' Health." Quarterly Journal of Economics, 119(3), 2004,
1091-130.
Engelberg, J., and C. Parsons. "Worrying about the Stock
Market: Evidence from Hospital Admissions." Journal of Finance, 71,
2016, 1227-50.
Ettner, S. L. "Measuring the Human Cost of a Weak Economy:
Does Unemployment Lead to Alcohol Abuse?" Social Science &
Medicine, 44, 1997, 251 -60.
Evans, W. N., and T. J. Moore. "Liquidity, Economic Activity,
and Mortality." Review of Economics and Statistics, 94,
2012,400-18.
Freeman, D. G. "A Note on 'Economic Conditions and
Alcohol Problems'." Journal of Health Economics, 18, 1999,
661-70.
Gallup Inc. "In U.S., 54% Have Stock Market Investments,
Lowest Since 1999." 2011. Accessed 2013. http://www
.gallup.com/poll/147206/stock- market- investments-lowest-1999.aspx.
Heutel, G., and C. Ruhm. "Air Pollution and Procyclical
Mortality." Journal of the Association of Environmental and
Resource Economists, 3(3), 2016, 667-706.
Kling, J. R., J. B. Liebman, and L. F. Katz. "Experimental
Analysis of Neighborhood Effects." Econometrica, 75(1), 2007,
83-119.
Kruger, N. A., and M. Svensson. "Good Times Are Drinking
Times: Empirical Evidence on Business Cycles and Alcohol Sales in Sweden
1861-2000." Applied Economics Letters, 17, 2008, 543-46.
Lindo, J. M. "Aggregation and the Estimated Effects of
Economic Conditions on Health." Journal of Health Economics, AO,
2015, 83-96.
McInerney. M., J. Mellor, and L. H. Nicholas. "Recession
Depression: Mental Health Effects of the 2008 Stock Market Crash."
Journal of Health Economics, 32(6), 2013, 1090-104.
Neumayer, E. "Recessions Lower (Some) Mortality Rates:
Evidence from Germany." Social Science & Medicine, 58, 2004,
1037-47.
Page, M., J. Schaller, and D. Simon. "The Effects of Aggregate
and Gender-Specific Labor Demand Shocks on Children's Health."
Journal of Human Resources, 2017. [Epub ahead of print],
https://doi.Org/10.3368/jhr.54.l.0716.8045R
Ruhm, C. J. "Economic Conditions and Alcohol Problems."
Journal of Health Economics, 14, 1995, 583-603.
--. "Are Recessions Good for Your Health?" Quarterly
Journal of Economics, 115, 2000, 617-50.
--. "Healthy Living in Hard Times." Journal of Health
Economics, 24, 2005, 341-63.
--. "Recessions, Healthy No More." Journal of Health
Economics, 42, 2015, 17-28.
Ruhm, C. J., and W. E. Black. "Does Drinking Really Decrease
in Bad Times?" Journal of Health Economics, 21,2002, 659-78.
Schaller, J., and M. Zerpa. "Short-Run Effects of Parental Job
Loss on Child Health." NBER Working Paper No. 21745,2015.
Schwandt, H. "Wealth Shocks and Health Outcomes: Evidence from
Stock Market Fluctuations." SSRN Scholarly Paper No. ID 2465333,
Social Science Research Network, Rochester, NY, 2014. Accessed January
25, 2017. http://papers.ssrn.com/abstracts2465333.
Simon, D. E. "Does Early Life Exposure to Cigarette Smoke
Permanently Harm Childhood Welfare? Evidence from Cigarette Tax
Hikes." American Economic Journal: Applied Economics, 8(4), 2015,
128-59.
Stevens, A. H., D. L. Miller, M. E. Page, and M. Filipski.
"The Best of Times, the Worst of Times: Understanding Pro-Cyclical
Mortality." American Economic Journal: Economic Policy, 4, 2015,
279-311.
Sullivan, D., and T. Von Wachter. "Job Displacement and
Mortality: An Analysis Using Administrative Data." Quarterly
Journal of Economics, 124(3), 2009, 1265-306.
Tapia Granados, J. "Recessions and Mortality in Spain,
1980-1997." European Journal of Population/Revue Europeenne de
Demographie, 21, 2005, 393-422.
CHAD COTTI and DAVID SIMON *
* We would like to thank the participants at the University of
Connecticut Health Economics seminar series for their insightful
comments on this paper. We would like to thank Mark McInerney and Mike
DiNardi for their excellent research assistance. All errors are our own.
Cotti: Professor, Department of Economics, University of Wisconsin
Oshkosh, Oshkosh, WI 54901. Phone 920-203-4660/4661, Fax 920-424-1329,
E-mail cottic@uwosh.edu
Simon: Assistant Professor, Department of Economics, University of
Connecticut, Storrs, CT 06269. Phone 336-482-6009, Fax 860-486-4463,
E-mail david.simon@uconn.edu
doi: 10.1111/ecin.12528
(1.) This pattern has been documented in the United States (Ruhm
2000), the European Union (Kruger and Svensson 2008; Neumayer 2004;
Tapia Granados 2005), and Japan (Tapia Granados 2005). It is
hypothesized that behaviors associated with increased mortality, such as
consumption of alcohol and cigarettes, are sufficiently normal so that
health actually improves when the economy worsens. This explanation is
consistent with some empirical studies showing that the decline in
mortality during times of higher unemployment is concentrated in acute
causes, for example, motor vehicle accidents and injuries, rather than
slowly developing causes, such as cancer or kidney disease (Evans and
Moore 2012). Empirical investigation of such behaviors confirms that
changes in these "pathway" causes of acute mortality do indeed
fall. Specifically, studies have shown that risky behaviors such as
alcohol consumption (Cotti, Dunn, and Tefft 2015; Ettner 1997; Freeman
1999; Ruhm 1995; Ruhm and Black 2002), cigarette consumption (Charles
and DeCicca 2008; Ruhm 2000, 2005), and drunk driving (Cotti and Tefft
2011), are negatively related to the unemployment rate in the United
States.
(2.) For example while Ruhm (2015) finds that with recent data
there is a weak overall impact of recessions on mortality, the
relationship between business cycles on traffic fatalities remains
pro-cyclical while the impact on cancer and accidental poisonings has
become highly counter cyclical.
(3.) Related to this literature, studies that look at the impact of
job loss on health find evidence that health deteriorates after job loss
(Sullivan and Von Wachter 2009), which suggests that the health effects
of business cycles are not fully operating through the channel of own
employment.
(4.) We are limited to looking at ages 4-17 for emotional
difficulties because this question is not asked of younger children.
Similarly, we look at ages 5-17 for school-related outcomes such as sick
days from school.
(5.) A disadvantage of hospitalizations is that an increase in
hospitalizations could reflect an increase in access rather than a
decrease in health. Theoretically, we believe the stock market crash is
most likely affecting child health through either: an income shock,
increased stress, or a change in discount rates through additional
information on future events. We do not believe these should result in
an increase in access and because of this we consider an increase in
hospitalizations in this context to be from declines in health.
(6.) For a more complete summary of the DJIA, see http://
www.djaverages.com/index.cfm?go=industrial-overview.
(7.) The natural log is used instead of the level for ease of
interpretation. Deflation of the market index is not necessary because
when logged the inflators are transformed to annual constant shifts in
the log index, which are then absorbed by the year indicator variables
included in each regression model.
(8.) http://research.stlouisfed.org/fred2/series/DJIA/.
(9.) We discuss in detail the decision to model the Dow Jones as a
6-month lag (rather than some average) in Appendix B. Estimates are
robust to several different lagged time frames used for the DJIA (e.g.,
3- and 9-month, as well as several different averages across these time
frames). The one exception to this is the hospitalization outcome, which
loses statistical significance in some cases, although the estimates
remain qualitatively the same. These results are also shown in Appendix
B.
(10.) We also demonstrate the robustness of the results to a broad
set of alternative business cycle measures as well.
(11.) Because of this independent variation, the correlation
coefficient between the log monthly closing Dow Jones and the monthly
unemployment rate in our sample is only -0.21.
(12.) Below we demonstrate the robustness of outcomes to accounting
for time effects in variety of alternative ways.
(13.) Mother's insurance status, maternal employment, and
family income are not included in the main models of the paper because
they are potentially endogenous variables. While these covariates will
likely impact our Y variables, it is doubtful that they are correlated
with the value of the national stock market, so there is not a strong
concern that their omission will confound our findings. Nevertheless, we
will demonstrate that the results are robust to their inclusion in an
appendix table.
(14.) The correct level of clustering is not obvious in this
context. The variation from the stock market occurs at the time-series
level. However, because there are only 9 years of data, clustering on
year leads to a small clusters problem. Clustering on household is an
option; however, some of our outcomes come from the sample child file
which only samples one child per household. Ultimately, clustering at
the year-month level captures the time-series dimension of the stock
market and results in enough clusters. We have also separately run our
models clustering at the year and household level as well and our
results are robust to these specifications.
(15.) For a detailed discussion of endogenous controls see Angrist
and Pischke (2009, 64-68).
(16.) We explore the robustness of this aspect of our base line
specification in detail in Table 5 below.
(17.) In the data, we identify the respective peak (or trough) of
the log Dow Jones and unemployment rate, and then code a fixed effect
for each of the different periods of increase and decrease relative to
each other. When doing this exercise, the data ends up breaking down
into five of these different periods. When including regime fixed
effects identification is isolated to comparing within-regime
correlation between the Dow Jones and child health. For example, if
children tend to be on average healthy in the period when the
unemployment rate is declining and stock is improving (regime 1) the
fixed effects will correct for this: since here we are only comparing
within regime changes.
(18.) We recognize that in showing results with calendar year fixed
effects in Table 3 that the definition of a year in this context is
somewhat arbitrary. To test this, we ran models where we redefined a
"year" as beginning in one of the 12 different months. That
is, in addition to looking at the typical calendar year
(January-December), we define a year as February-January,
March-February, and so on. Because we are estimating a large number of
models (12 for each outcome) we do not include these results in the
paper, but they are available upon request. Regardless of exactly how
"a year" is defined, coefficients on hospitalizations,
excellent health, and definite difficulties are generally similar to
those in the baseline specification and our overall conclusions remain
the same.
(19.) We experimented with different combinations of introducing
these covariates, but the coefficient estimates remain markedly similar.
(20.) In using regional unemployment rates, we now also add
regional fixed effects to these models.
(21.) All business cycle measures were collected from the Bureau of
Labor Statistics, except for data on housing prices which were collected
from the federal housing agency state level series which was then
averaged to get a national measure.
(22.) We also re-estimated our baseline model on the levels of the
Dow Jones (rather than log), to make sure that how we were specifying
functional form was not driving the robustness of our results to
controlling for business cycles. Results are consistent.
(23.) The number of observations in this analysis is notably
smaller because high school graduates and those with "some"
college were excluded.
(24.) Reported as stocks or stock mutual funds.
(25.) Children were assigned as part of a stockholder's
household if the household's predicted ownership value exceeded
0.50.
(26.) Overall, our findings add another dimension of complexity to
the literature on the relationship between economic conditions and
health. As we discuss in Section II, the literature on the impact of
state unemployment rates on mortality already finds large differences in
effects when looking by time period, aggregation, age at which mortality
is measured, and cause of death.
(27.) Page, Schaller, and Simon (2017) find evidence that declines
in female specific labor market demand leads mothers to both spend more
time in the upkeep of the home and improves child health outcomes. This
is consistent with a story of a dominant time-substitution effect
improving child outcomes. However, when comparing our national
unemployment rate results to the state-level literature, it is important
to recognize that, as explored in Lindo (2015), there could be different
impacts based on the level of aggregation of the economic indicator.
(28.) This table also shows that there is a reduction in delayed
care due to an increase in the unemployment rate. While understanding
this phenomenon fully would require additional work, it is consistent
with an improvement in child health decreasing the need for medical care
for the marginal case.
(29.) Moreover, the stronger estimates for non-stock holders
further suggests that the behavior of wealthier families are not driving
these results, even though these households are more explicitly impacted
by stock market fluctuations, which is consistent with the idea that
these families' wealth (or possibly correlated educational
attainment) insulates their children from harm to some extent. Due to
the large number of results already shown, results on mechanism outcomes
divided by stock holders and non-stock holders are available on request.
(30.) Stevens et al. (2015) show that at the state level the
largest number of deaths occur among the elderly during an economic
expansion. While the number of deaths among the elderly is driving
pro-cyclical mortality, Stevens et al. (2015) document large coefficient
estimates of effects on children. This is consistent with an increase in
the national unemployment rate being good for children's health.
(31.) Over the time of our sample, the log Dow Jones reached a max
of 9.539 and a min of 8.887 for a change of: -0.65. The unemployment
rate reached a minimum of 4.4 and a maximum of 10 for a change of 5.6.
We multiplied these changes by the relevant coefficients to get the
"Stock Market Crash Effect" and "Recession Effect"
for each of the outcomes, shown in the bottom panel of Table 8.
(32.) A related question is whether or not changes in household
conditions are themselves a viable mechanism for improving health. While
the job loss literature suggests that there is a negative effect on
health for job displacement (Sullivan and Von Wachter 2009), we show
there is an intensive margin change in hours worked in addition to a
change in employment. Further, it is not as clear that the negative
health effects of job displacement hold for children: Schaller and Zerpa
(2015) find that while paternal job loss is bad for child health,
maternal job loss actually decreases incidences of infectious disease
among children.
(33.) This pattern is consistent with the results presented in
panels E and F of Table 5 when compared to the main results in Table 2.
Caption: FIGURE 1
Fluctuations in the Dow Jones and the Unemployment Rate by Quarter
2004-2012
Caption: FIGURE 2
Fluctuations in the 6-Month-Lagged Dow Jones and Reported Health of
Child
Caption: FIGURE 3
Fluctuations in the 6-Month-Lagged Dow Jones and Child Health
Outcomes: (A) Excellent Health, (B) Sick From School, (C)
Hospitalizations, and (D) Reported Emotional Difficulties
TABLE 1
National Health Interview Survey 2004-2012 Summary Statistics
Demographic Covariates
Mean/SD
White 0.71
(0.46)
Black 0.15
(0.36)
Other race 0.15
(0.35)
Mother dropout 0.16
(0.36)
Mother high school 0.24
(0.42)
Mother some college 0.31
(0.46)
College 0.27
(0.45)
Black 0.15
(0.36)
White 0.71
(0.46)
Other race 0.15
(0.35)
Child's age 8.41
(5.20)
Mother's age 36.45
(7.94)
Outcome Variables
Mean/SD N
Hospitalization in past 12 months 0.06 202,445
(0.23)
Excellent health 0.55 202,718
(0.50)
Sick days from school in past 12 months 3.49 64,710
(6.46)
Definite emotional difficulties in 6 months 0.05 70,299
(0.22)
Poor Health Index -0.004 64,050
(0.57)
Notes: For each outcome, we include all available observations on
children ages
0-17. Depending on the outcome variable looked at,
sample sizes change due to missing observations. Information on
hospitalizations and excellent health are contained in the NHIS
person core questionnaire which asks questions of all children in
the household. Sick days from school and emotional difficulties are
asked about in the NHIS sample child questionnaire which only
interviews one child per household. SD, standard deviation.
TABLE 2
Impact of the 2008-2009 Stock Market Fluctuations in the DJIA
on Children's Health Outcomes
Hospitalization School Health
Sick Days Status:
Excellent
Ln average daily -0.009 * -1.144 *** 0.046 ***
close, DJIA (0.005) (0.255) (0.015)
Unemployment -0.001 *** 0.030 * 0.003 ***
rate (0.000) (0.018) (0.001)
N 202,445 64,710 202,718
Mean of 0.057 3.38 0.555
dependent
variable
Emotional Poor Health
Difficulties Index (z-Score)
Ln average daily -0.002 -0.085 ***
close, DJIA (0.008) (0.021)
Unemployment 0.001 ** 0.001
rate (0.001) (0.002)
N 70,299 64,050
Mean of 0.051 -0.004
dependent
variable
Notes: All models include month-fixed effects and controls for race,
gender, child's age, mother's age, mother's education, and mother's
marital status. Robust standard errors clustered by year-month are
in parentheses.
*** p <.01; ** p <.05, * p <.1.
TABLE 3
Impact of the 2008-2009 Stock Market Fluctuations in the
DJIA on Measures of Children's Health Outcomes, Robustness
Across Specifications
(A) Hospitalization
Log Dow Jones -0.009 * -0.010 ** -0.014 *
(0.005) (0.005) (0.007)
N 202,445 202,445 202,445
Month-fixed effects Yes No Yes
Quarter of the year-fixed effects No Yes No
Linear time trend No No Yes
Calendar year-fixed effects No No No
Regime-fixed effects No No No
Log Dow Jones -0.037 *** -0.011
(0.013) (0.008)
N 202,445 202,445
Month-fixed effects Yes Yes
Quarter of the year-fixed effects No No
Linear time trend No No
Calendar year-fixed effects Yes No
Regime-fixed effects No Yes
(B) Health Status Excellent
Log Dow Jones 0.046 *** 0.048 *** -0.009
(0.015) (0.017) (0.020)
N 202,718 202,718 202,718
Month-fixed effects Yes No Yes
Quarter of the year-fixed effects No Yes No
Linear time trend No No Yes
Calendar year-fixed effects No No No
Regime-fixed effects No No No
Log Dow Jones 0.061 * 0.060 **
(0.033) (0.030)
N 202,718 202,718
Month-fixed effects Yes Yes
Quarter of the year-fixed effects No No
Linear time trend No No
Calendar year-fixed effects Yes No
Regime-fixed effects No Yes
(C) School Sick Days
Log Dow Jones 1.144 *** -1.147 *** -0.658 **
(0.255) (0.295) (0.302)
N 64,710 64,710 64,710
Month-fixed effects Yes No Yes
Quarter of the year-fixed effects No Yes No
Linear time trend No No Yes
Calendar year-fixed effects No No No
Regime-fixed effects No No No
Log Dow Jones -0.470 -1.376 *
(0.707) (0.625)
N 64,710 64,710
Month-fixed effects Yes Yes
Quarter of the year-fixed effects No No
Linear time trend No No
Calendar year-fixed effects Yes No
Regime-fixed effects No Yes
(D) Emotional Difficulties
* Log Dow Jones -0.002 -0.002 -0.004
(0.008) (0.008) (0.012)
N 70,299 70,299 70,299
Month-fixed effects Yes No Yes
Quarter of the year-fixed effects No Yes No
Linear time trend No No Yes
Calendar year-fixed effects No No No
Regime-fixed effects No No No
* Log Dow Jones -0.038 * -0.034 *
(0.021) (0.017)
N 70,299 70,299
Month-fixed effects Yes Yes
Quarter of the year-fixed effects No No
Linear time trend No No
Calendar year-fixed effects Yes No
Regime-fixed effects No Yes
(E) Poor Health Index
Log Dow Jones -0.084 *** -0.087 *** -0.031
(0.021) (0.022) (0.030)
N 64,050 64,050 64,050
Month-fixed effects Yes No Yes
Quarter of the year-fixed effects No Yes No
Linear time trend No No Yes
Calendar year-fixed effects No No No
Regime-fixed effects No No No
Log Dow Jones -0.116 * -0.114 **
(0.064) (0.051)
N 64,050 64.050
Month-fixed effects Yes Yes
Quarter of the year-fixed effects No No
Linear time trend No No
Calendar year-fixed effects Yes No
Regime-fixed effects No Yes
Notes: Besides as specified in each panel/column, all models include
controls for national unemployment rate, race, gender, child's age,
mother's age, mother's education, and mother's marital status.
Robust standard errors clustered by year-month are in parentheses,
*** p <.01; ** p <.05; * p<.1.
TABLE 4
Impact of Leading DJIA on Measures of Children's Health Outcomes
Hospitalization Health School
Status Excellent Sick Days
Log Dow Jones, -0.007 0.016 -0.546 *
6-month lead (0.004) (0.014) (0.305)
Log Dow Jones, -0.007 0.007 -0.234
9-month lead (0.005) (0.015) (0.272)
Log Dow Jones, -0.005 0.002 0.221
12-month lead (0.005) (0.017) (0.243)
Log Dow Jones, -0.006 0.006 -0.296
average of 7-12-month (0.005) (0.015) (0.284)
leads
Emotional Poor
Difficulties Health Index
Log Dow Jones, 0.001 -0.012
6-month lead (0.008) (0.021)
Log Dow Jones, 0.012 0.009
9-month lead (0.008) (0.020)
Log Dow Jones, -0.014 * 0.034
12-month lead (0.008) (0.020)
Log Dow Jones, 0.012 0.007
average of 7-12-month (0.008) (0.023)
leads
Notes: All models include month-fixed effects and controls for the
(corresponding lagged) national unemployment rate, race,
gender, child's age, mother's age, mother's education, and mother's
marital status. Robust standard errors clustered by year-month
are in parentheses.
*** p <.01; ** p <.05; * p <.1.
TABLE 5
Alternative Business Cycle Specifications
Hospitalization School
Sick Days
(A)
Ln average daily close, DJIA -0.0130 ** -0.941 ***
(0.006) (0.286)
Regional unemp. rate -0.001 *** 0.044
(0.000) (0.019)
N 178,149 56,874
(B)
Ln average daily close, DJIA -0.006 -0.948 ***
(0.006) (0.293)
Female unemployment rate -0.004 ** -0.102
(0.002) (0.098)
Male unemployment rate 0.0018 0.105
(0.0013) (0.073)
N 202,445 64,710
(C)
Ln average daily close, DJIA -0.006 -1.238 ***
(0.005) (0.246)
Employment to pop. ratio 0.001 *** -0.022
(0.000) (0.018)
N 202,445 64,710
(D)
Ln average daily close, DJIA -0.009 * -1.189 ***
(0.005) (0.252)
Home Price Index 0.0001 *** -0.002
(0.0000) (0.001)
N 202,445 64,710
(E)
Ln average daily close, DJIA -0.004 -1.274 ***
(0.005) (0.242)
N 202,445 64,710
(F)
National unemp. rate -0.001 *** 0.051 **
(0.000) (0.019)
N 202,445 64,710
Health Emotional Health
Status: Difficulties Index
Excellent (z-score)
(A)
Ln average daily close, DJIA 0.053 *** 0.005 -0.074 ***
(0.015) (0.009) (0.025)
Regional unemp. rate 0.005 *** 0.002 *** 0.002
(0.001) (0.001) (0.002)
N 178,387 61,705 56,294
(B)
Ln average daily close, DJIA 0.079 *** 0.003 -0.076 ***
(0.014) (0.009) (0.025)
Female unemployment rate -0.007 -0.002 -0.007
(0.004) (0.003) (0.008)
Male unemployment rate 0.009 *** 0.002 0.006
(0.003) (0.002) (0.006)
N 202,718 70,299 64,050
(C)
Ln average daily close, DJIA 0.037 ** -0.005 -0.089 ***
(0.014) (0.008) (0.020)
Employment to pop. ratio -0.003 *** -0.001 ** -0.000
(0.001) (0.001) (0.002)
N 202,718 70,299 64,050
(D)
Ln average daily close, DJIA 0.037 ** -0.005 -0.087 ***
(0.016) (0.008) (0.021)
Home Price Index -0.0001 -0.000 -0.000
(0.0001) (0.000) (0.000)
N 202,718 70,299 64,050
(E)
Ln average daily close, DJIA 0.031 ** -0.007 -0.089 ***
(0.017) (0.008) (0.020)
N 202,718 70,299 64,050
(F)
National unemp. rate 0.003 ** 0.001 ** 0.003
(0.001) (0.001) (0.002)
N 202,718 70,299 64,050
Notes: All models include month-fixed effects and controls for race,
gender, child's age, mother's age, mother's education, and mother's
marital status. Robust standard errors clustered by year-month are
in parentheses. Models presented in Panel A also
include regional-fixed effects.
*** p <.01; ** p <.05; * p <.1.
TABLE 6
Impact of Stock Market Fluctuations on Child Health by Subgroup
Independent Variable of Interest: Ln Average Daily Close, DJIA
Age Groups
Dependent Less than 5 Age 5-12 Age 13-17
Variable
Hospitalization -0.031 ** -0.001 -0.002
(0.012) (0.004) (0.005)
N 56,486 79,433 66,526
School sick days -1.136 *** -1.154 ***
(0.293) (0.383)
N 32,730 31,980
Health status: 0.041 * 0.054 *** 0.039 *
excellent (0.021) (0.020) (0.020)
N 56,549 79,546 66,623
Emotional -0.001 -0.003
difficulties (0.010) (0.013)
N 33,110 32,128
Poor Health -0.087 *** -0.082 **
Index (0.025) (0.033)
N 32,427 31,623
Independent Variable of Interest: Ln Average Daily Close, DJIA
Child Gender Mother's Education
Dependent Male Female High School College
Variable Dropout
Hospitalization -0.008 -0.011 * -0.008 -0.018 *
(0.006) (0.006) (0.008) (0.010)
N 103,284 84,169 41,904 47,908
School sick days -1.285 *** -0.996 *** -1.459 ** -0.888 ***
(0.370) (0.331) (0.591) (0.278)
N 33,095 31,615 11,552 16,258
Health status: 0.036 ** 0.056 *** 0.113 *** -0.004
excellent (0.016) (0.019) (0.004) (0.023)
N 103,434 99,284 41,902 47,906
Emotional 0.013 -0.017 ** -0.029 0.003
difficulties (0.013) (0.008) (0.021) (0.011)
N 35,952 34,347 12,649 17,626
Poor Health -0.080 ** -0.090 *** -0.160 *** -0.008
Index (0.033) (0.028) (0.050) (0.035)
N 32,767 31,283 11,395 16,108
Notes: All models include month-fixed effects and controls for the
national unemployment rate, race, gender, child's age, mother's age,
mother's education, and mother's marital status. Robust standard
errors clustered by year-month are in parentheses.
*** p <.01; ** p <.05; * p <.1.
TABLE 7
Impact of Stock Market Fluctuations on Child Health Analysis Using
Predicted Probability of Stock Ownership
Hospitalization School Health
Sick Days Status:
Excellent
(A) Predicted stock
ownership--yes
Ln average daily -0.014 -0.681 0.066 *
close, DJIA (0.012) (0.436) (0.034)
N 23,657 7,840 23,649
(B) Predicted stock
ownership--no
Ln average daily -0.009 * -1.215 *** 0.063 ***
close, DJIA (0.005) (0.305) (0.017)
N 178,788 56,870 179,069
(C) Predicted stock
ownership--interaction
Ln average daily -0.009 * -1.178 *** 0.059 ***
close, DJIA (0.005) (0.311) (0.017)
Predicted stock 0.058 -2.128 0.915 **
ownership--yes indicator (0.123) (5.552) (0.363)
DJIA stock owner -0.006 0.225 -0.099 **
interaction (0.012) (0.596) (0.039)
N 202,445 64,710 202,718
Emotional Poor Health
Difficulties Index (z-Score)
(A) Predicted stock
ownership--yes
Ln average daily 0.012 -0.025
close, DJIA (0.021) (0.048)
N 8,464 7,751
(B) Predicted stock
ownership--no
Ln average daily -0.004 -0.093 ***
close, DJIA (0.009) (0.026)
N 61,835 56,299
(C) Predicted stock
ownership--interaction
Ln average daily -0.002 -0.089 ***
close, DJIA (0.009) (0.025)
Predicted stock -0.047 -0.312
ownership--yes indicator (0.224) (0.586)
DJIA stock owner 0.006 0.034
interaction (0.024) (0.063)
N 70,299 64,050
Notes: All models include month-fixed effects and controls for the
national unemployment rate, race, gender, child's age, mother's
age, mother's education, and mother's marital status. Robust
standard errors clustered by year-month are in parentheses.
*** p <.01; ** p <.05; * p <.1.
TABLE 8
Impact of Stock Market Fluctuations on Delayed Care
and Own Household Economic Constraints
(A) Delayed Care
Ln average -0.018 ** -0.013 **
daily close, (0.006) (0.006)
DJIA
Unemployment -0.0012 ** -0.0008 **
rate (0.0004) (0.0004)
N 202,634
Mean of 0.038
dependent
variable:
Standardized
effects (a)
Stock market 0.012 ** 0.009 **
crash effect
Recession -0.007 ** -0.004 **
effect
(B) Employment Status
Ln average 0.006 0.051 ***
daily close, (0.011) (0.015)
DJIA
Unemployment -0.011 *** -0.011 ***
rate (0.001) (0.001)
N 146,496
Mean of 0.872
dependent
variable:
Standardized
effects (a)
Stock market -0.004 -0.033 ***
crash effect
Recession -0.062 *** -0.063 ***
effect
(C) Hours Worked
Ln average -0.253 0.655
daily close, (0.532) (0.587)
DJIA
Unemployment -0.223 *** -0.218 ***
rate (0.030) (0.030)
N 124,032
Mean of 45.0
dependent
variable:
Standardized
effects (a)
Stock market 0.164 -0.426
crash effect
Recession -1.249 *** 1.22 ***
effect
Notes: All models include month-fixed effects and controls for
race, gender, child's age, mother's age, mother's education, and
mother's marital status. Robust standard errors clustered by
year-month are in parentheses.
(a) To standardize effects across the stock market crash and the
great recession, we first calculated the peak to trough change in
the log Dow Jones and unemployment rate respectively. Over the time
of our sample, the log Dow Jones reached a max of 9.539 and a min
of 8.887 for a change of: -0.65. The unemployment rate reached a
minimum of 4.4 and a maximum of 10 for a change of 5.6. We
multiplied these changes by the relevant coefficients to get the
"Stock Market Crash Effect" and "Recession Effect" for each of the
outcomes in the above table.
*** p <.01; ** p <.05; * p<.1.
COPYRIGHT 2018 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2018 Gale, Cengage Learning. All rights reserved.
