From Payday Loans to Pawnshops Fringe Banking, the Unbanked, and Health


Jerzy comes to us from Tufts University and i’m really excited about this work First I’ll just give an outline of the
talk so I’ll start out with just a brief history of the fringe banking industry
talk about who uses fringe services and why then I’ll go through methods results
and discussion including a discussion of a lot of the sensitivity analyses we did
to deal with some of the limitations of our data and I’ll talk about some policy
implications of our results for financial regulations and social welfare
programs and then I’ll take questions and if you have any questions during the
talk to shout out my name and I’ll try to answer so first a history of the
fringe banking industry so fringe banking the fringe banking industry
includes payday lenders who give customers short-term loans pending their
next paychecks pawn brokers who buy customers property allow them to
repurchase it later at a higher cost car title lenders who hold customers as car
titles as collateral for short-term loans I’m in check cashers who cash
checks for a perchick fee I’m so fringe lenders charged annual percentage
interest rates of 400 to 600 percent I’m in credit checks where the services are
not generally required until they’re marketed as one-time emergency loans
borrowers frequently borrow repeatedly and they often can’t our discharge the
debts quickly so the average payday loan borrower is indebted for five months and
pays five hundred and twenty dollars for interest of loans of just three hundred
and seventy-five dollars one in five car title borrowers has their vehicle seized
after default so deregulation really believed the groundwork for the fringe
banking industry in the 1970s federal monetary policy doing control inflation
increased long-term commercial interest rates and that made the the high cost of
funds made operating within existing state interest rate caps difficult for
banks and other lenders in in response to political pressure from banks many
states altered their caps or granted exemptions for certain lenders in 1978
at a Supreme Court decision allowed federally chartered banks to charge
their home state interest rates to customers in other states subsequently
state chartered banks lobbied for the same export rights in states weakened
rate caps to attract business or kind of led to this race to the bottom between
states trying to attract business other regulatory changes around this time a
lot of banks to diversify their investment activities so you may have
heard of the repealed the glass-steagall Act and they also were able to more
easily expand across state lines this contributed to consolidation in the
financial sector so throughout much of the 20th century
the banking and credit needs of the working poor had been met by local
commonly owned institutions like credit unions and savings and loans
associations but as local institutions merged with larger national banks that
reduced less profitable services like small dollar loans that may be more
catered to the needs of local communities and banks hide a need for
revenue due to competition between banks increase the fees on deposit accounts
that rendered them prohibitive for many low income people so the fringe banking
industry has grown precipitously over the last thirty to forty years
so the payday lending industry which we can in the early 1990s extended ten
billion dollars in credit in 2001 and forty eight billion dollars in credit in
2011 the check cashing industry was nearly non-existent before the mid 1970s
that almost 60 billion dollars in transactions in 2010 similar growth has
occurred in the pawnbroker and car title ending industries as well as other forms
of consumer lending and also like more mainstream forms of lending like credit
cards student loans and mortgages I’m so due in part to the regulatory changes in
the 70s as well as cheap credit made available to primarily white men through
programs like the GI Bill throughout the latter half of the 20th century
household debt increased substantially so this graph shows
of household debt to income and household debt to assets ratios from
1950 to 2010 an alongside deregulation were cuts and social services rising
costs of necessities in certain cases stagnant wages and condiment declines in
personal savings rates so this graph shows an approximately three fold
decrease in the average personal savings rate among Americans from 1980 to 2017
and with a spike after the recession I’m so inequities in fringe borrowing so
I’ll just go over some quick descriptive statistics about who uses fringe loans
so about 8% of the overall population has used fringe loans in the past year
but as you might expect fringe borrowing is more common among those with low or
volatile incomes I’m in mirroring patterns and wealth and equity past your
fringe borrowing is two to three times more common among minorities and racial
and ethnic groups and among whites it’s more common among those with less
education the disabled female-headed households in the unbanked which are
people who lack access to a household bank account so as I mentioned and
probably unsurprisingly inequities in fringe borrowing by race ethnicity track
and equities and wealth by race ethnicity so this graph shows absolute
how absolute and relative gaps in wealth by race ethnicity have increased
substantially over the last six decades despite the graph starting right after
the end of formal Jim Crow segregation in other that causes of the wealth
inequities are due to an array of factors beyond the scope of this talk
inequities can be a part attributed to the inequity well access to credit
that’s been available to many many middle class whites throughout the
United States is history so two discriminatory practices and credit
markets have contributed to inequities and wealth and in turn fringe borrowing
so some well-known discriminatory practices include redlining which was a
practice initiated by the Federal Housing Administration 1934 that
involved marking maps with red lines to delineate neighborhoods were federally
insured and regulated mortgages would be denied to marginalize and racialized
groups and the picture on the right shows a map of redlining at Seattle so
it’ll be washed out but you can see the Central District was in historically
black neighborhood it was one of the primary red lined areas and then
mortgage also often require divorce singled and
widowed women to secure mortgages with a man’s signature and though reverse
redlining was made illegal in the 1960s today under reverse redlining accessible
loans for marginalised groups are often high-cost and risky example African
Americans were more likely to be steered into subprime mortgages and whites with
similar even much lower credit scores or incomes and people of color particularly
women were disproportionately dispossessed of wealth during the
subprime mortgage crisis I’m so regarding fringe banks they frequently
locate in poorer non-white communities where people lack access to more
mainstream forms of credit and in our data we can see that conditional on a
variety of like demographic and socio-economic variables minoritized
groups are more likely than whites to use french services and so they lack
access to say four forms of credit so people primarily use fringe loans to pay
for basic living expenses like rent or to make up for lost income from missed
work and they’re also sometimes used to cover unexpected expenses like medical
bills and sometimes I’m also stuff like house repairs car repairs etc so the
unbanked as I mentioned before are those who lack access to a household checking
account or a savings account I’m at about 7% of US households are unbanked
and they’re about twice as likely to use fringe loans five times as likely to use
transactional friends services like check cashing and bill paying services
and those who are banked them so most households go on banks because they lack
sufficient money for an account they want this they want privacy and distrust
banks or they cannot afford the fees associated with bank accounts so
overdraft fees were pretty rare before deregulation in the 1970s but they
generated thirty three billion dollars for banks in 2015 which often sequence
withdrawals from largest to smallest to maximize overdraft fees so that means
basically if you have $100 in your account and three withdrawal is one
that’s $200 one that’s ten dollars and another that’s ten dollars the bank
could withdraw the two hundred dollars for us to trigger an overdraft then
withdraw the two smaller withdrawals after that to trigger two more
overdrafts or as if they had withdrawn the first smaller one first it would
have only triggered one overdraft fee with that large one at the end I’m so
construed the construed as light loans to the account holder or the typical
overdraft fee would be a the equivalent of 17,000 percent annual
percentage interest rate which is much larger than a fringe loan the material
costs of being unbanked are also high however so the average household of four
earning $25,000 per year spends roughly two and a half thousand dollars on check
cashing bill paying and money orders so now I’ll just connect fringe banking and
unbanked to health so the motivation for our analysis so clearly fringe low news
can exacerbate the well-documented effects a financial hardship on health
I’m using fringe loans for recurring expenses can lead to spiraling debt in
bankruptcy and there’s some evidence that the true costs of fringe borrowing
are hidden by lenders underestimated by borrowers or that lenders target
consumers in a more predatory fashion like has been documented in the subprime
mortgage industry no no I support often lack other options and fringe loans can
be cheaper than overdraft fees or delaying payments for necessities so the
material consequences of Fringe borrowing have to be balanced against
the consequences of alternatives like foregoing necessities or defaulting on
other high-cost loans and as I’m sure as many of you probably know the stress of
debt can harm mental and physical health so indebtedness is often shamed fringe
debt may be especially stigmatized so so social isolation looming default
grassman from debt collectors can contribute to anxiety depression and
suicide this has been written about a lot in regards to the recent subprime
mortgage crisis meanwhile being unbanked and a largely non-cash economy generates
its own sources of stress so bills must be paid in person at certain locations
within certain hours irrespective of transportation costs wait times or
conflicting obligations so chronic stress can put individuals at risk for
metabolic and cardiovascular diseases by just regulating systems that respond to
stress like the hypothalamic-pituitary-adrenal axis and
the immune and inflammatory systems I mean inequities and the distribution
of stress and invulnerability to stress across demographic groups may exacerbate
health inequities I mean fringe borrowers frequently face
other chronic stressors like discrimination that might amplify the
health effects of financial strain once again however the net stress
from fringe dead or being unbanked needs to be balanced against the stress of
alternatives so our basic research question was what is the association
between fringe borrowing on bank status and self-rated health which is the
outcome we had access to and note that we’re not investigating and don’t really
have information about how fringe borrowing affects other aspects of
household well-being so it could be that while the material consequences of or
stress effects of fringe borrowing or harmful for the borrower they’re good
for other aspects of household well-being like the family member that
gets medical care because the borrower took out a fringe loan to pay for it so
we’re just focused on the borrower themselves so why is this work important
so many studies of assess the relationship between wealth and health
rarely do studies distinguish between certain types of debt and so certain
types of debt can be used to build wealth like regulated mortgages while
other sources of debt can drain wealth like payday loans for example and though
clearly thinking about structural inequities is essential it’s also
important to think about how these structural inequities are reproduced
through lower-level systems like the credit market and embodied in social
outcomes like poor health and then this builds on recent work in other fields
like sociology the documents that pecuniary penalties faced by those
living in poor neighborhoods called poverty taxes these include monetary
sanctions for those living in heavily policed communities as well as fines and
tickets etc that can reproduce poverty across generations and then finally
there are policy implications of this work both in the short term since Reagan
can be addressed through smaller welfare programs and changes to financial
institutions as well as through large or structural changes to programs and the
economy so now I’ll go over methods for our primary analysis so data came from
the Census Bureau’s current population survey which is Census Bureau survey
conducted to go out to workforce data on employment for example so we created a
person level ID demerged the June FDIC supplements in 2011 2013 and 2015 that
focus on finances with the following marches 2012 2014 and 2016 a six up
laments that focus on health so nine months separate our exposure and our
outcome and as you can see our exposure an
outcome in our primary analyses we’re only measured once which is clearly
problematic and we try to address that in sensitivity analyses later so we use
a data set of non proxy respondents who are household financial decision-makers
we did this both to avoid Mis classification of the exposure and the
outcome by proxy response and because we thought the material and stress effects
of the exposure might be most pronounced among those who bore household financial
decision-making responsibilities I’m so we had about 15,000 people in each of
our data sets the sample sizes differ a little bit just because the questions at
different universes for unbanked versus Fringe borrowing so this picture shows
the various sampling rotations in the current population survey no it’s not
very intelligible but basically households in the CPS are sampled eight
times total four to four months periods separated by an eight-month break I mean
our analysis focused on the rotation rotation group see that’s highlighted in
red and their sample that they’re surveyed in both the FDIC supplement in
June and the asic supplement at nine months later in March so they’re the
only rotation group that’s sampled in both of those surveys I mean not many
people in the health field know that you can follow people longitudinally in the
current population survey or at least not many people do it and that might be
because it’s kind of a nightmare to merge across supplements but it’s
frequently done in the in the economics literature because you stuff like income
is asked in every survey so theoretically you could get information
about someone income for eight waves of the survey
I’m sorry exposures were past your household pawnshop payday or car title
loan use an unbanked was defined as living in a household without a bank
account I’m so clearly the fringe low in exposure is pretty crude we don’t know
how often people are taking out loans for how much etc our outcome was
dichotomize self-rated health as using the standard question which is shown to
be independently predictive of morbidity and mortality unconditional and other
health conditions and SES factors however it’s still a pretty crude
measure of health and has less predictive ability among certain
demographic groups like Spanish speakers no Takata maizing it can improve
reliability which is why we did that so here are the confounders we use and I
will go through all of them but one important thing we were missing is
information about household wealth or other sources of household debt so we
don’t know people are carrying others like credit card debt for example
however we’re hopeful that the information on use of other fringe
services that we were able to adjust for so I also check cashing use rent to own
use refund anticipation loans may services may serve as proxies for the
unmeasured confounders for unbanked readjusted for the same variables except
for French service use which we hypothesize were mediators of the
unbanked health relationship rather than confounders so something to think about
during the analysis ya know so we didn’t have any information about that we try
to address that in a sensitivity analysis later I’ll explain a little bit
more that’s definitely a good point so somebody to think about and I just
added this slide so I don’t have like a great answer for it yet but what’s the
correct counterfactual contrasts in the fringe borrowing analysis in particular
for example is it people who took out a loan compared with people who didn’t
because they lived in a state where fringe lending was illegal they had
access to low-cost credit and banking services they got help from family and
friends they weren’t duped into taking out a
loan by a predatory lender they didn’t take out a loan because they didn’t need
to I mean would we expect the measure of Association to be the same if we use
each of these groups as our comparison group so we use a propensity score
matching approach for analyses and our primary rationale for using repent’ see
score matching approach so we’re concerned about structural differences
between the exposed and the unexposed since fringe borrowers than the unbanked
tend to be really low SES may just be totally different demographically from
people who don’t use fringe loans I’m so no overly dramatic example of this is
shown in the figure when there’s no region of common support between
exposure groups there’s no possibility of causal inference there’s no exchange
ability I mean Cove area it’s across exposure groups so in traditional
methods this problem may be hidden propensity score estimation can make the
problem clear the possibility of off support and off support inference I’m a
propensity score matching with a tight caliper makes off support in inference
almost impossible and I’ll explain more about what
propensity score estimation is what propensity score matching is later
in addition identifying the health effects of Fringe borrowing or being
unbanked on Fringe borrowers or the unbanked which is the average treatment
effect on the treated was more relevant than identifying the health effects of
Fringe borrowing or being unbanked on all participants so the average
treatment effect some of whom are really high SES and had a low probability of
ever being exposed so propensity score matching estimates the average treatment
effect on the treated which I’ll explain more later
so here’s a brief overview of propensity scores and apologies for those of you
for whom this is a review so propensity the propensity scores the estimated
probability of exposure predicted from observed confounders typically the
propensity score is estimated from a logistic model but it could be estimated
from a probit model or other sorts of models I mean it’s basically just an
alternative approach to traditional methods that combines multiple
covariates into one variable the propensity score to calculate the
exposure outcome association one can match respondents on the propensity
score which is what we did adjust for the propensity score weight observations
by the inverse of the propensity score etc so we use nearest neighbor
propensity score matching in our analysis
so each exposed participant was matched and I’m sorry a nearest neighbor
matching each exposed participant is matched and unexposed participants who
have the nearest or the most similar propensity score on the one-to-one
matching procedures shown in the figure below propensity score matching attempts
to make the exposure independent of measured confounders and frequently you
specify a maximum distance within which the exposed can be matched to the
unexposed so that’s called the calliper and due to the matching procedure so I
mentioned this before which matches exposed unexposed respondents to expose
respondents on the propensity score and disc discards people who are unmatched
propensity score matching Elsie’s estimate the average treatment effect on
the treated rather than the average treatment effect like in a normal
regression model assuming no on measured confounding so our principal matching
approach was based on recommendations and a sort of how-to paper by Elizabeth
Stewart that came out a couple years ago so first each participants propensity
score was estimated using logistic models that included that Afra mention
confounders squared age and income terms then we
perform nearest-neighbor matching without replacement to match the maximum
of two unexposed participants to one exposed participant on the propensity
score within 0.05 standard deviations of the propensity scores so the caliper
then we estimated prevalence ratios for the exposure outcome Association using
par saw models and I call these prevalence ratios because again our
outcome was only measured one so this might have been prevalent poor for
health rather than incident poor fare health so we adjusted four variables in
the propensity score matching dress residual Kovarian imbalance that we
thought might remain after matching and because we thought certain variables are
strong confounders we want to make sure those were adjusted for in our analysis
and we didn’t adjust for all the confounders due to problems with model
convergence originally we had used a sample that had clustering by household
and our GE models particularly weren’t converging and then we’re also concerned
about positivity and then finally we bootstrapped the matching in regression
a thousand times to correctly estimate the variance and i’ll explain what i
mean by that um so we bootstrapped for two primary reasons first of our pencil
score is estimated from a model rather than being a measured variable this
point seems a little bit controversial and maybe less important since it seems
like some evidence suggests that not taking into account the fact that the
propensity score is estimated results in an over estimate of the standard error
so it’s an T conservative I’m in second and more importantly the order that
exposed participants are chosen and matched to unexposed participants can
impact which participants are chosen for the analysis and this is particularly
problematic when doing K to one matching without replacement which is what we did
so our bootstrap procedure proceeded in several steps first we drew a random
sample with replacement from the data that was the size of our original data
set we perform the propensity score matching procedure estimated the post on
models I described on the previous page repeated steps one two three a thousand
times and then calculate confidence intervals based on the sampling
distribution the propensity of the prevalence ratio I mean this procedure
took a really long time it took like eight hours for each of our data sets
which is kind of frustrating and ended up just making the whole analysis take
way longer than it would have otherwise so it’s a downside to doing
so this just shows how what I meant by saying that the order of matching
matters so for example if we’re doing two to one matching without replacement
with a caliper of zero within a caliper of 0.2 propensity score units and when
we start with we start matching with the exposed participant that’s first in the
top row they’d match to two unexposed participants then we move to this person
that’s second in the top row they matched to unexposed participants
however if we start with the second exposed participant as shown in the
second scenario they match the two unexposed participants one of whom is
different from whom they match to before and then if we match that first exposed
participant they’d only match to one unexposed participant I mean if the
outcomes differ for the chosen matched pairs and the prevalence ratio estimate
will differ so not taking to it listen to account would underestimate the
variance so something else to think about I also just added this slide so in
addition to the historical and social context we provide four analyses how
could we have more explicitly analyzed whether the health consequences of these
exposures disproportionately burdened certain demographic groups I’m gonna
have something in mind something that we talked about in epi five twelve and I’m
wondering if others have thoughts I mean part of the reason why we didn’t do this
is because we already have so much going on the paper we didn’t really have space
for more analyses but I think would have been important to incorporate analyses
of inequity empirically into our study rather than just providing context for
it so we did a variety of sensitivity analyses first we were concerned about
unmeasured confounding so we’re concerned by own measure about on
measure confounding by factors like wealth other sources of debt baseline
health status so we ran propensity score matching alice ease with check cashing
use and refund anticipation loans control exposures so these are used by
similar populations to those who use fringe loans but are non debt creating
they’re just transactional so we have pathi size they’d be less harmful for
health if unmeasured confounding we’re minimal we expected these exposures to
have smaller or even no health effects relative to fringe borrowing we’re also
concerned about reverse causation because again our exposure and outcome
are only measured once in the analyses this is a concern because fringe loans
are sometimes you to pay for medical debt or cover fallout
from fallout from illness like missed work and similarly people may become
unbanked through the financial consequences of illness so if you recall
from earlier on in the presentation this is the sampling structure of the current
population survey and rotation group C shown in red which the rotation group we
focused on in this analysis actually was also sampled in the asic three months
prior to baseline and asked about their health status um so we didn’t know this
when we originally started the analyses and on Jim did a presentation at CS de
back in June and someone brought this up that this it would be possible to merge
in that prior asic supplement into our main data set so we merge the asic
supplement three months prior to baseline with our main sample and
excluded those reporting poor fare health income from disability benefits
are being uninsured on the march asic three months prior to baseline and then
we estimated prevalence ratios we apply saw models on non propensity score match
their bowls since the exclusions and then conditioning on another wave of the
survey chopped our sample size by about half I mean if reverse causation were
minimal we expected these explosions to have small health effects on the prep or
small effects on the prevalence ratio estimates note that the temporal
relationship between fringe borrowing and the exclusions is still ambiguous
since our fringe borrowing exposure was past year fringe borrowing yet we’re
excluding people who report poor for health and complem disability benefits
are being uninsured three months prior to baseline so there’s still a little
bit of ambiguity there what’s our final sensitivity analysis were insure was an
instrumental variable analysis i’m conducted on the fringe Barot for the
fringe borrowing analysis so some econometric terminology and motivation
for the instrumental variable analysis so if the exposure outcome relationship
is confounded or the outcome causes the exposure which are problems I described
earlier the exposure is endogenous and ordinary methods will be biased so in
given and dodging a’ti we can estimate unbiased effects with an instrumental
variable which is a variable that predicts exposure but it’s no direct
effect on the outcome and no prior cause is in common with the outcome so an
example of this would be treatment assignment and a randomized control
trial since clearly that’s a cause of the exposure and if randomization
done correctly it shouldn’t have any prior causes in common with the outcome
and no direct effect on the outcome aside from through its effect on the
exposure so the top picture is a dag of a valid instrument since the instrument
Z affects the exposure X but has no direct effect on the outcome Y and there
were no unmeasured confounders of the instrument outcome relationship on the
bottom picture meanwhile shows an invalid instruments there’s an
unmeasured confounder of the instrument outcome relationship so we conducted two
stage least squares intrumental variable analyses which are one of the more
common types of instrumental variable analysis so the 2’s OS analysis is two
primary steps of first stage model in a second stage model and the first stage
fringe borrowing was regressed on the instrument and the aforementioned
confounders in an OLS model and then the second stage self-rated health her
breasts on predicted values of Fringe borrowing generated from the first stage
model along with covariance to calculate prevalence differences rather than
prevalence ratios so – SOS provides an estimate of the local average treatment
effect which in this case is the health effective or the effective fringe
borrowing on self-rated health for those who’s borrowing behavior was affected by
the by the instrument in this case regulations which I’ll explain so this
is the COS also called the group of compliers this is a small portion of the
population we can’t identify exactly who these people are which is a drawback of
instrumental variable analysis so our instruments were state-level regulations
of paid a pawn and check cashing outlets the prior research shows that these
regulations are associated with reductions in fringe borrowing since
fringe lending becomes improv unprofitable after the regulations are
put in place so states were considered considered to have payday loan
regulations if they kept the annual percentage interest rates at less than
or equal to 36% and they’re considered to have pawn regulations moderate pawn
regulations that they cap the interest rate below 120 percent in strong
regulations if they tap the interest rate and less than 36 percent on then
states are considered to have check cashing regulations if they had any cap
on fees at check cashing outlets and these instruments are being valid if
fringe banking regulations are associated with unmeasured factors that
influence self-rated health for example progressive state
level policies if we didn’t adjust for those in the analysis and they’re
associated with fringe lending regulations and clearly that’s an
unmeasured confounder of the instrument outcome relationship so since having a
single strong instrument is preferable preferable to having multiple weak
instruments both in terms of bias and precision at least from what I can tell
we combined the three instruments into a single weighted instrumental variable
normalized to the range zero one and it had 12 possible values I’m so to do this
we use a method by some folks at Stanford Sanjay Basu just came out with
a paper where he did this so first we regress fringe borrowing on each
regulation and logistic models so we had fringe borrowing as the outcome and the
regulation as the exposure we record the deviance for each instrument from the
logistic models which is just a proxy for the discriminatory power of the
instrument so how strongly associated the regulation was with fringe borrowing
then we created a single weighted instrument which was an average of the
instrument individual instruments weighted by their partial deviances I
mean exactly how we did this isn’t so important but basically instrument range
from zero which meant that the state had no regulations all the way to one which
meant to say that the strictest regulations and you can see the deviance
numbers aren’t direct aren’t really interpretable either aside from in
relation to each other so you can see the pawn regulations were most strongly
associated with reductions in fringe borrowing payday regulations were
associated with moderate reductions and fringe borrowing and check cashing
either at least the least strong effect on fringe borrowing this is a final
slide about instrumental variable analysis so we implemented the two stage
least squares model and propensity score matching bowls to increase instrument
strength since we hypothesized that fringe borrowing regulations would be
unlikely to effect the borrowing behavior of high SES participants who
would be unlikely to take out a fringe loan regardless of the regulatory
environment I mean for the same reason that I described before we bootstrapped
estimates of our coefficients and standard errors so there are several
assumptions of instrumental variable analysis that need to be verified I mean
if any are violated bias tends to be in the red deck in the direction of the
ordinary least squares estimate which is obviously problematic so the first
assumption the instrument is associated with the
exposure we calculate which we tested by calculating robust F statistics for the
significance of the weighted instrument in that first stage regression that I
described before and we did that for each of the thousand bootstrap
repetitions the second assumption which is in as easily isn’t empirically
verifiable like the first assumption is that the instrument is independent of
the outcome conditional on exposure and covariance so first we tested this by
calculating descriptives for the sample stratified by quartiles of the
instrument so basically hypothesize that if measured confounders were unevenly
distributed across levels of the instrument might be true that unmeasured
confounders were also unli unevenly distributed across levels of the
instrument which would render the instrument invalid since there be
unmeasured confounders of the instrument outcome relationship we also conducted a
variety of sensitivity analyses the instrumental variable analysis to test
the robustness of our results to alternative modelling specifications and
I’ll describe them the results in a few slides
the final assumption of IV analyses is not testable so it’s this final
assumptions monotonicity this would be violated if unlike most respondents some
respondents took out loans and states with regulations and refused to take out
loans and states without regulations we believe this is unlikely since the
instance the regulations that we studied are essentially bans on Fringe loans so
there wouldn’t be any fringe loans available even if someone wanted to take
out a loan once the regulation was put in place all right so our results so
these are just a script like kind of crude descriptives of the prevalence of
poor for health in our study I’m sorry about 13% of our total sample reported
poor for health 23% of fringe borrowers reported poor
fare health and 31% of those who are on banks reported poor for health and those
are shown in purple so here’s the district density plots of propensity
scores on the log scale across the exposed and the unexposed I mean you can
see there’s actually pretty good overlap in the propensity scores across the
exposed and the unexposed so our concerns about structural differences
and covariance between exposure group wasn’t really supported by the data so
here are the median standardized mean differences and confounder means and
proportions between exposure groups its exposure groups across the bootstrap
samples before propensity score matching and gold and then after propensity score
matching in purple I’m so standardized mean differences are just a way to
compare the means and proportions of variables between exposed and unexposed
groups and they’re frequently used for propensity score matching standardized
mean differences of less than 0.1 are generally considered to indicate a
negligible and balanced and covariance across exposure groups so that’s just
kind of a general guide and we can see that after matching all standardized
mean differences aside from renta own use in the fringe borrowing analysis at
standardized mean differences of less than 0.1 I’m indicating that the
propensity score matching procedure succeeded in making exposing unexposed
groups exchangeable unobserved confounders so here are the results from
the main analysis and some sensitivity analyses focus on fringe loan use so
fringe borrowing was associated with a 38% higher prevalence of poor fare
health with a 95% confidence interval of about one point one four two one point
six eight excluding those with prior poor fare health disability income or
being uninsured on the march asic three months prior to baseline didn’t really
change the prevalence ratio estimates nonetheless the precision was
considerably worse for the analysis that excluded those with poor fare health
this is because at least I think there are a few respondents who switch from
good health three months prior to baseline to bad health or to poor fair
health nine months after baseline so only so those those responders would
have to switch from poor for health to good I’m sorry good health no getting
confused yeah respondents would have to switch from good health to poor health
over the course of just one year so there are a few respondents with the
outcome of interest in that model and then you can see the results for the
control exposures at the bottom and both can both check-cashing use and refund
anticipation loans had small no health effects unless we think that both on
measure confounding and reverse causation didn’t have necessarily undue
influence on our results at least based on these since
Timothy analyses here results from the unbanked analysis so when the main
unbanked analysis on bank status was associated with the 17% higher
prevalence of poor health the 95% confidence interval of 0.99 to 1.3 9
it’s clearly a much smaller health effect than we observed for fringe
borrowing I mean as with fringe borrowing the exclusions had little
effect on the estimates though excluding those reporting poor for a health prior
to baseline increase the prevalence ratio estimate to 1.4 but clearly you
can see the precision was really poor with the confidence interval overlapping
on the confidence interval for a main analysis basically totally and now we’ll
go over the instrumental variable results so this is a map of the
distribution of values of the weighted instrument by state with darker colors
indicating stronger fringe Landing regulations I mean you can see there’s a
potentially problematic clustering of stronger regulations in the Northeast
this is potentially problematic since people living in the Northeast and
report better health and people living in other states and they’re also
progressive state level policies in the Northeast that don’t necessarily exist
in states like the south so that instrument may be correlated with
unmeasured factors that influence health so their first Diagnostics for the
strength of the IV on the estimated coefficient across the bootstrap samples
from the first stage regression was negative 0.16 which corresponds to a 40%
reduction the probability of fringe borrowing and states with the strictest
regulations relative to states with the weakest regulations in the median F
statistic for the significance of the weighted instrument in the first stage
regression across the bootstrap samples is 20.6 I’m an F statistics greater than
10 are generally considered to indicate sufficient instrument strength though
again this isn’t a hard rule I think based on this information the instrument
was sufficiently correlated with the exposure to continue with instrumental
variable analysis here the standardized mean difference is for the confounders
across quartiles of the instrument and you can see that most of the
standardized mean difference is in the confounders across levels of the
instrument or less than 0.1 but there are larger differences in metro area
race ethnicity and us-born indicating that the instrument might be
somewhat correlated with measure and potentially unmeasured confounders this
means instrumental variable analysis should be interpreted cautiously since
the instrument outcome relationship may be confounded still even the largest
standardized mean differences are pretty small so the largest ones are about
point two so it’s a little bit yeah there isn’t clear evidence either way I
would say so here are the prevalence differences estimated from the two SLS
analysis or remember these are prevalence differences it’s not
prevalence ratios I’m the main to SLS analysis the absolute difference in
prevalence support for health for Fringe loan users versus non fringe loan users
was 33% with fringe loan users having the higher prevalence support for health
you can see across these analyses precision was really poor which is often
a problem in to SLS analyses since the exposure is estimated from a model
rather than measured it’s always going to be worse than what you have in an OLS
model or an ordinary regression so our first our first sensitivity analysis was
run to test if correlation between unmeasured factors in any one instrument
was biasing the results so dropping payday pawn or check cashing regulations
from our weighted instrument didn’t substantially change the prevalence
difference estimates though clearly the precision was worse particularly when we
dropped pawn regulations since that was our strongest instrument and the
dropping those individual instruments made our weighted instrument weaker than
it was in our primary analysis next we tested if correlation between unmeasured
factors within the instrument with any one region was biasing our results so we
reran the analyses dropping one census region at a time from our analyses I
mean as with the prior sensitivity analyses this didn’t change the
prevalence difference estimates though it really blew up the precision
particularly dropping the Northeast from the analysis where the prevalence
difference estimate still may be 0.3 but the confidence interval is like negative
0.72 over 1 which doesn’t make any sense because prevalence differences are
bounded between negative 1 and 1 I mean I think this happened partially because
it just decreased our sample size a ton and then also most of the states with
the strictest fringe fringe lending regulations were
clustered in the Northeast there was no one left with an instrument value of one
or very few states left with an instrument value of one after we did
that and then finally we adjusted for year specific state level policies that
may be associated with the instrument and with health indicators of state
level Medicaid expansion the existence of a state level Earned Income Tax
Credit program I’m in the proportion of eligible state residents participating
in the supplemental nutritional assistance program once again this
didn’t change the prevalence just sand difference estimates though precision
worsened particularly when we adjusted for the state level you ITC programs and
just one note these are pretty huge prevalence difference estimates so
converting the prevalence ratio estimates from a primary propensity
score matching ala C’s to prevalence differences using adjusted adjusted
Average adjusted predictions that the means demonstrates that the to us
estimates or five or six times larger than the propensity score match it’s
nonetheless – SOS calculates local average treatment effects rather than
average treatment effects among the treated so the results aren’t directly
comparable necessarily so we believe that these results coupled with our main
analyses support our primary findings but the
actual effect sizes here are in a primary interest given that they’re only
relevant for a small portion of the population those whose fringe borrowing
behavior was affected by the regulations and then once again they should be
interpreted cautiously because of the potential correlation between odd
measured factors in our instrument so now we’ll go through the discussion so
in summary we found that fringe borrowing and unbanked status were
associated with worse self-rated health and sensitivity analyses generally
supported these findings despite the major limitations with our data the
fringe borrowing health effect was significantly larger than the unbanked
health effect so strengths of the analysis include the fact that this is
to our knowledge the first empirical analysis of the exposures and the health
literature few public health studies have leveraged the CPS is panel
structure and followed people over time and then the findings were supported by
our sensitivity analyses which I already mentioned
however there a number of limitations to our analyses first our outcome was
self-rated health which is although predictive of morbidity and mortality
and certain groups is less so in certain low SES groups which are the primary
people exposed in our analysis so that’s obviously problematic and then I’ve
mentioned throughout the presentation just the limitations of our data the
limited exposure outcome in covariate data we had one limitation I haven’t
talked as much about is that we only had information about Pat any past year
fringe Barling not information about fringe borrowing frequency or amount so
we couldn’t really analyze dose effects which I think would have been
interesting I’m an important I’m so now we’ll go over some policy implications
and there are a number of ways in which the social economic and health effects
of these exposures might be addressed so one option is financial regulations
which we think are unlikely to sufficiently address the both economic
and health consequences of the exposures so many states have said Apr and feed
caps on fringe loans typically a cap of 36 percent which I discussed before this
is the amount recommended by consumer groups in some states have reported
positive regulatory effects effects after implementing such regulations
nonetheless lenders can often skirt regulations by disguising their services
and moving online I mean as discussed previously the needs of low-income folks
may be unmad or satisfied at greater cost if they just suddenly lose access
to fringe loans some nos meanwhile have argued for regulations then enable
mainstream banks to re-enter the small dollar loan market since recent
legislation has effectively legislated away the ability of mainstream banks to
offer small dollar loans I’m however lending to the poor and
banking the poor isn’t particularly profitable low-income folks often hold
small deposits they frequently default on loans given that banks really only
care about their bottom line it’s unclear that they’d want to reenter the
small dollar loan market in sufficient numbers to reduce the need for fringe
borrowing moreover recent scandals regarding discriminatory lending
fraudulent accounts like it Wells Fargo an overdraft fee’s mandate caution about
the role of mainstream banks and low-income lending so it’s really hard
to see mainstream banks as like the solution or the necessarily the good
guys in this verse Asian so another option is
building alternative financial institutions and to date most government
government initiatives to bank the poorer like the CD ba and the CRA have
tried to enlist private profit motivated banks to bank and lend to the poor and
have thus been largely ineffective since it’s not really a profitable endeavor
i’mso loans made through the CRA are often subprime since the only way to
make one of the only ways to make money off of low-income folks is by charging
them really high interest rates and banks created through the CD ba have
struggled to survive due to limited government support so successful
initiatives should be built outside the mainstream banking sector like credit
unions and savings and loans associations were in the past options
include the creation of a postal banking system which actually existed in the
u.s. from 1910 to 1963 as analogues in other countries so postal banking would
address the geographic barriers to banks given the ubiquity of post offices and
low-income communities and the cost barriers to banks given and honored the
cost barriers of low-income lend and given a nonprofit mission municipal
banks community-led lending circles could serve similar functions however
with half of Americans reporting they’d be unable to produce $400 cash for an
emergency and the common use of Fringe loans to pay for necessities clearly the
core of the frenching fringe banking problem stems from financial instability
scarce resources and the high cost of necessities so robust public provision
of necessities and labor protections would actually address the root causes
of fringe service use there’s pretty good data to support this so a recent
study that was published by the Federal Reserve found that each $1 increase in
the state level minimum wage was associated with a 40% reduction and the
probability of fringe borrowing I’m another study that was just published in
Health Affairs found that early medically Medicaid expansion in
California was associated with 11 percent decrease in fringe loan use and
clearly these these programs would have other salutary effects on other social
determinants of health and health equity I’m and then addressing broader
structural factors that deep in poverty for marginalized groups like segregation
or mass incarceration are also clearly important and that’s

Leave a Reply

Your email address will not be published. Required fields are marked *