WORK INCENTIVES OF MEDICAID BENEFICIARIES AND THE ROLE OF ASSET TESTING

• Author: Ponpoje Porapakkarm,Svetlana Pashchenko
• Date: 01 November 2017
• DOI: http://doi.org/10.1111/iere.12247
• Published date: 01 November 2017

INTERNATIONAL ECONOMIC REVIEW

Vol. 58, No. 4, November 2017

WORK INCENTIVES OF MEDICAID BENEFICIARIES AND THE ROLE OF ASSET

TESTING∗

BYSVETLANA PASHCHENKO AND PONPOJE PORAPAKKARM1

University of Georgia, U.S.A.; National Graduate Institute for Policy Studies (GRIPS), Japan

Should asset testing be used in means-tested programs? Focusing on Medicaid, we show that in the asymmetric

information environment, there is a positive role for asset testing. Our tool is a general equilibrium model with

heterogeneous agents. We find that 23% of Medicaid enrollees do not work in order to be eligible. These

distortions are costly: If Medicaid eligibility could be linked to (unobservable) productivity, this results in

substantial welfare gains. We show that asset testing can achieve a similar outcome when asset limits are allowed

to be different for workers and nonworkers.

1. INTRODUCTION

Should asset testing be used in means-tested programs? The total federal spending on 10

major means-tested programs and tax credits increased more than 10-fold over the last four

decades, reaching $588 billion or 4% of GDP in 2012 (CBO, 2013). Yet, little consensus exists

on certain aspects of these programs’ design, in particular, asset testing. The overall trend over

the last decade was toward abandoning the asset testing policy, but the debate concerning its

use continues. As a recent example, asset testing for the food stamps program was one of the

central issues in the debate over the 2014 Farm Bill.2

In this article, we aim to show that there is a positive role for asset testing in the asymmetric

information environment. Means-tested programs target low-income people by restricting its

enrollees to earn less than a certain limit. This requirement prevents high-income workers from

obtaining public transfers, but it cannot guarantee that nonworkers with potential income above

the income limit do not enroll. Because earning ability is unobservable, once an individual with

high labor income stops working, he is indistinguishable from those whose potential labor

income is low. In this environment, asset testing can be used as an additional tool to improve

the ability of means-tested programs to target the most disadvantaged people.

We focus on Medicaid, which is one of the largest means-tested programs in the United States

and also an important source of health insurance coverage for the nonelderly poor. The fraction

∗Manuscript received May 2015; revised April 2016.

1We thank Orazio Attanasio, Gadi Barlevy, Mariacristina De Nardi, Eric French, Mikhail Golosov, Gary Hansen,

Roozbeh Hosseini, Robert Kaestner, Greg Kaplan, Karen Kopecky, Matthias Kredler, Paul Klein, Vincenzo Quadrini,

Victor Rios-Rull, Yongseok Shin, Kjetil Storesletten, Ija Trapeznikova, Gianluca Violante, Tomoaki Yamada, Pierre

Yared, Eric Young, two anonymous referees, and all seminar participants at the Chinese University of Hong Kong,

the Federal Reserve Bank of Chicago, GRIPS, ETH Risk Center Workshop, IFS, University of Tokyo, EFACR group

in NBER Summer Institute, Mannheim Macro workshop, Midwest Macro meeting in Urbana, NASM in Minneapo-

lis, Nordic Macro Workshop in Sm¨

ogen, SED meeting in Seoul, Pacific Rim Conference in Tokyo, Vienna Macro

Workshop, and Greater Stockholm Macro Group for their comments and suggestions. This work is supported by JSPS

KAKENHI Grant Number 15K03505 and GRIPS’ Research Project Grant. All errors are our own. Please address

correspondence to: Svetlana Pashchenko, Department of Economics, Terry Business School, University of Georgia,

600 South Lumpkin St., Athens, GA 30602, U.S.A. Phone: 706-542-3667. Fax: 706-542-2784. E-mail: svetlana@uga.edu.

2The 2014 Farm Bill reauthorized the Supplemental Nutrition Assistance Program (SNAP), formerly known as the

food stamps program. The House version of the Bill proposed to repeal the broad-based categorical eligibility, which

allows states to bypass asset testing when determining SNAP eligibility. In contrast, the Senate version of the bill made

no changes to the broad-based categorical eligibility.

1117

C

(2017) by the Economics Department of the University of Pennsylvania and the Osaka University Institute of Social

and Economic Research Association

1118 PASHCHENKO AND PORAPAKKARM

of workers among Medicaid enrollees is substantially lower than this fraction among the rest

of the population; on average, nondisabled Medicaid beneficiaries are twice less likely to work

than people with private insurance or the uninsured.3In this article, we ask two questions:

(1) Does Medicaid significantly distort work incentives? (2) Can asset testing improve the

insurance–incentives trade-off of Medicaid without changing the amount of redistribution in

the economy? More specifically, our goal is to quantify the distorting effects of Medicaid on

work incentives, assess its welfare implications, and illustrate how asset testing can mitigate

these distortions. Our important contribution is to show that work-dependent asset testing can

eliminate labor supply distortions without creating significant distortions on savings.

Our approach is a quantitative general equilibrium model with the following key features.

First, we allow for heterogeneity of individuals along the dimensions of health, productivity, and

medical expense shocks. This allows us to capture the insurance role of Medicaid for people

with bad health, large medical shocks, and/or low productivity. Second, we let health affect

productivity and opportunity to access employer-based insurance, which allows us to model

the selection of people with low attachment to the labor force into Medicaid.4Third, people in

our model have several options to insure against medical shocks: self-insurance, public health

insurance, and private health insurance (employer-based and individual). However, private

health insurance is not easily accessible for two reasons. First, employer-based insurance is only

available for the subset of the population working at firms that offer this type of insurance.

Second, the individual market is risk rated, meaning that unhealthy people face high premiums.

People who want to obtain public insurance have to meet an income test and an asset test.

Because labor income is endogenous, Medicaid beneficiaries in our model include those who

have low earning ability and those who have relatively high earning ability but choose not to

work to be eligible. Fourth, we introduce disability shock into the model to be able to separate

disabled and nondisabled individuals in our analysis, i.e., to distinguish between people who

can work (and whose labor supply decisions can be distorted by the Medicaid eligibility rules)

and those who cannot (because they are disabled). Finally, we model other non-Medicaid

government means-tested programs to represent adequately the public safety net existing in

the economy.

We calibrate the model using the MEPS data set. More specifically, we require the model

to reproduce the following key patterns of the data separately for each health group: (i) the

life-cycle profiles of health insurance take-up, (ii) the life-cycle profiles of employment, and (iii)

the average labor income profiles for all workers and for workers without employer-sponsored

health insurance (ESHI). An essential feature of our calibration is that we use our model to

estimate the potential labor income and chances to access ESHI of individuals whom we do not

observe working in the data. This is important for understanding how Medicaid affects labor

supply decisions because a large fraction of Medicaid beneficiaries do not work.

Our findings are as follows. First, around 23% of nondisabled Medicaid enrollees would

choose to work if they were able to keep their access to public insurance. The majority of this

group is unhealthy and has higher medical costs and higher assets than other Medicaid enrollees.

Second, these distortions are important in welfare terms. If we remove the asymmetric

information problem, i.e., link Medicaid eligibility to (unobservable) exogenous productivity as

opposed to (observable) endogenous labor income while keeping the budget of public transfer

programs constant, this will result in ex ante welfare gains equivalent to 1.17% of annual

consumption.

Third, we study how asset testing can be used to reduce the labor supply distortions when

productivity is unobservable. We show that strict asset testing (with the asset limit equal to

$2,000) can almost completely eliminate the moral hazard problem; the percentage of Medicaid

3Own calculations from Medical Expenditure Panel Survey (MEPS) data set; see Section 5 for details.

4In the data, 43.2% of nondisabled Medicaid beneficiaries are unhealthy, whereas the unhealthy among the privately

insured and the uninsured account for only 13% and 24.5%, respectively. In addition, unhealthy people are less likely

to be covered by employer-based health insurance. Only 48% of the unhealthy are covered by employer-based health

insurance compared with 67% among the healthy.

WORK INCENTIVES OF MEDICAID BENEFICIARIES AND THE ROLE OF ASSET TESTING 1119

beneficiaries who stop working to obtain Medicaid decreases from 23% to 1%. However, this

reduction in labor supply distortions comes at the cost of large saving distortions that substan-

tially decrease the welfare gains of this policy. In contrast, if asset limits are allowed to be

different for workers and nonworkers, asset testing can achieve an outcome that is very close

to the “ideal” case of observable productivity. This happens because strict asset testing of non-

workers prevents highly productive individuals from using the following strategy: stop working,

claim Medicaid, and then use their accumulated assets to smooth consumption. In contrast,

loosening asset limits on working beneficiaries relieves saving distortions for individuals who

do not “game” Medicaid rules by lowering their labor supply.5

The results of our policy analysis can reconcile the opposite findings from three recent

empirical studies that examine the effect of public insurance on labor supply using changes in

the Medicaid expansion programs in three states. Garthwaite et al. (2014) and Dague et al. (2013)

find that Medicaid has a large effect on labor supply in Tennessee and Wisconsin, respectively,

whereas Baicker et al. (2014) conclude the opposite for the case of Oregon. Importantly, the

Medicaid expansion programs in Tennessee and Wisconsin had no asset testing, whereas the

program in Oregon imposed a strict asset limit of $2,000. In light of our findings, the different

intensity of the moral hazard problem in these three cases can be attributed to the difference in

the asset testing policies.

The article is organized as follows. Section 2 reviews the related literature. Section 3 introduces

the model. Section 4 explains our calibration. Section 5 compares the performance of the model

with the data. Section 6 presents the results. Section 7 discusses the role of asset testing. Section

8 relates our results to the recent empirical findings. Section 9 concludes.

2. RELATED LITERATURE

Our article is related to several strands of literature. Our positive analysis is motivated by the

literature studying the labor supply effects of public means-tested programs (for an extensive

review, see Moffitt, 2002). A subset of this literature focuses on the Medicaid program. Most of

these studies use data prior to 1996, when adult eligibility for Medicaid was tied to eligibility for

another welfare program, Aid for Families with Dependent Children (AFDC).6,7The close link

between the two programs made it difficult to isolate the effect of Medicaid on labor supply,

and different identification strategies were used. Moffitt and Wolfe (1992) exploit the variation

in the valuation of Medicaid benefits and show that Medicaid has a significant negative effect on

labor force participation. Blank (1989), Winkler (1991), and Montgomery and Navin (2000) use

variations in the generosity of Medicaid by state to evaluate its effect on labor supply. The first

study finds no effect, whereas the last two studies find small effects on labor force participation.

Yelowitz (1995) exploits the delinking of Medicaid from AFDC for children in the late 1980s

and finds that this policy had a positive effect on labor force participation of mothers. Decker

and Selck (2012) and Strumpf (2011) examine the effects of the introduction of the Medicaid

program in the late 1960s and early 1970s on labor force participation; both studies find no

effect. Dave et al. (2013) study the expansion of Medicaid to cover the costs of pregnancy and

childbirth that happened in the late 1980s and find that this policy significantly decreased the

probability that a woman who had recently given birth was employed. Overall, the literature

based on pre-1996 data provides mixed evidence on the effects of Medicaid on labor supply.

However, there is evidence that the decision to participate in welfare programs was noticeably

affected by the availability of health insurance (Ellwood and Adams, 1990; Moffitt and Wolfe,

1992; Decker and Selck, 2012).

5The mechanism behind work-dependent asset testing is analogous to the effect of earnings-dependent wealth

taxation advocated in several studies of optimal taxation; see, for example, Kocherlakota (2005) and Albansei and Sleet

(2006).

6Currently this program is substituted by the Temporary Assistance for Needy Families (TANF).

7At the end of the 1980s, Medicaid was expanded to cover pregnant women regardless of their participation in

welfare.