Preliminary Results – Please do not circulate or cite without permission from authors 1 Advantageous or Adverse Selection in Emerging Health Insurance Markets: Evidence from a Micro Health Insurance Program in Pakistan 1 Yi Yao Joan T. Schmit Justin R. Sydnor 1 Yi Yao (corresponding author [email protected]) is a PhD candidate at School of Business, University of Wisconsin at Madison. Joan Schmit is a professor and American Family Insurance Chair in Risk Management and Insurance at University of Wisconsin at Madison. Justin Sydnor is an assistant professor of Risk Management and Insurance at University of Wisconsin at Madison.
21
Embed
Preliminary Results Please do not circulate or cite ...aria.org/meetings/2011 papers/Advantageous or Adverse Selection in...Pauly et.al (2008) used data ... The standard “positive
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Preliminary Results – Please do not circulate or cite without permission from authors
1
Advantageous or Adverse Selection in Emerging Health Insurance Markets:
Evidence from a Micro Health Insurance Program in Pakistan1
Yi Yao
Joan T. Schmit
Justin R. Sydnor
1Yi Yao (corresponding author [email protected]) is a PhD candidate at School of Business, University of Wisconsin at
Madison. Joan Schmit is a professor and American Family Insurance Chair in Risk Management and Insurance at University of Wisconsin at Madison. Justin Sydnor is an assistant professor of Risk Management and Insurance at University of Wisconsin at Madison.
Preliminary Results – Please do not circulate or cite without permission from authors
2
Abstract
Despite widespread interest in microinsurance programs in developing countries, little is known
about the nature of information asymmetries and selection in these emerging markets. We use
data from a micro health insurance program in Pakistan to investigate the degree of adverse
selection in the program. We analyze how claim rates evolve as households renew their policies
and find that households who have larger claims during the policy period are slightly more likely
to renew their policies for the next period. Although that pattern is on the surface consistent with
adverse selection, we instead find that when compared to households joining the insurance in the
same period, renewed households have significantly lower claim frequency and total claim
amounts. Taken together these results suggest that a) households who experience claims
perceive higher value to insurance, b) that households do not seem to be acting (at least on the
margin of renewals) on private information about their risk types and c) that there are forces
affecting insurance demand that lead to advantageous selection in the policyholders that retain
coverage over time.
Preliminary Results – Please do not circulate or cite without permission from authors
3
I: Introduction
In recent years there has been an explosion of interest in the creation of new insurance products in
developing countries. Closely related to the growth of “microfinance,” these new “microinsurance”
programs are often created by non-profit organizations to provide insurance in the developing world. To
achieve their goals, however, these organizations must find a sustainable business model. This paper
focuses on the viability of micro health insurance in particular. Because there are generally data
limitations, high transaction costs (relative to premium and income), and low levels of education in
developing countries, insurance policies of all kinds need to be simple. That need for simplicity in both
underwriting (to limit distribution expenses) and policy language (both to limit administrative expenses as
well as to develop product demand) may be a particular challenge for the development of health insurance
because most private health insurance markets require extensive underwriting and further employ
somewhat complex policy provisions to limit both adverse selection as well as moral hazard.
Although there is a significant body of empirical literature on detecting adverse selection in health
insurance markets in developed countries (see Cohen and Siegelman, 2010 for a review), there is little
empirical evidence on the role adverse selection plays in developing microinsurance markets. The small
existing literature paints somewhat conflicting views of the potential for sustainable micro health
insurance. Pauly et.al (2008) used data from the World Health Survey for 14 developing countries to
compare the risk premium that local people would be willing to pay with likely values for the
administrative expense. Without taking adverse selection and moral hazard into consideration, they
concluded that a voluntary health insurance market might be feasible. In contrast, Biener and Eling (2009)
examined the problems and solutions for microinsurance markets using Berliner’s (1982) set of criteria to
identify significant problems discussed in existing studies on microinsurance. Based on a survey of
studies that looked at hypothetical decisions to join an insurance program, they concluded that problems
of information asymmetry were “epidemic” in micro health insurance, providing evidence ranging from
adverse selection, moral hazard to fraud in the previous literatures.
In this paper, we help address this debate about the viability of micro health insurance by examining
household renewal decisions to understand better the development of a microinsurer’s risk portfolio in its
early years of operations. The results reveal the essence of selection in the given micro health insurance
market.
Resolving this debate is important because of all the microinsurance programs being provided, micro
health insurance programs are among the most needed but least provided. Both acute and chronic
diseases are serious problems for the poor in developing countries, where poor sanitation, low nutrition,
and inadequate preventative care are widespread. As a result, the life expectancy for those living in low
and middle income countries is 20% lower than those living in high income countries. In an extreme
comparison, the life expectancy for people living in Sub-Saharan African is only 46, compared with 75.5
in high income countries2. The consequences of illness for the poor not only include reduced current
standards of living, but may also lead them to reduce investments in their future human capital. With that
backdrop, it is perhaps not surprising that people with low income also have an interest in health
2 Table 2.3 Selected mortality characteristics by sex and World Bank region in 2001, Global burden of disease and
risk factors, Oxford University Press and The World Bank, 2006.
Preliminary Results – Please do not circulate or cite without permission from authors
4
insurance programs. Despite that interest, however, it is estimated that only 20% have access to adequate
health insurance3 (Bockstal, 2008).
For this study we use data from the Aga Khan Agency for Microfinance (AKAM), which has run a simple
micro health insurance program in Pakistan since 2007. Testing for the nature of selection issues in
microinsurance markets presents unique challenges. Most studies of selection issues in insurance have to
overcome the inability to observe outcomes for people who do not purchase any insurance. In developed
markets, researchers typically approach the issues by using tests for adverse selection that analyze
differences in claim experience across insureds who purchase different amounts of insurance (see
Chiapporri and Salanie, 2000). In developing markets, like the AKAM product studied here, customers
are given only one option in coverage in order to hold down administrative costs and keep the products
simple for a population inexperienced in insurance purchasing. The standard “positive correlation” test
between coverage level and risk, therefore, is not feasible.
Other methods to detect adverse selection in the micro health insurance market, however, are available.
In this paper we analyze the nature of selection by measuring the development of claim experience in
renewal policies. We use a two-part model to compare the microinsurer’s new book of business to its
renewed book of business and test whether the risk pool has deteriorated over time, which could
demonstrate the existence of adverse selection. From a theoretical standpoint, the selection and evolution
of renewal policies could result in either adverse or advantageous selection. On the one hand, there are
forces leading to adverse selection in a sense that riskier people might be more likely to stay with the
insurer. One classic version of the adverse selection story here would be if policyholders have private
information about their risk type and suffer unrelated income shocks that affect whether they can afford
insurance. In that case, we would expect that those who know they are especially high risk would be less
likely to cancel insurance due to an income shock, which would result in a deteriorating book of business
when looking at renewed policies. Another possibility is that people might learn about their risk type
over time – for instance learning that they are pregnant. If people act on that new information, we would
expect adverse selection as those who have learned they are higher risk choose to renew while low risks
do not. On the other hand, there are also other forces affecting the decision to be insured that may be
unrelated to risk type and could even result in advantageous selection. In new markets, like that in
Pakistan, people start to learn about the value of insurance and how it protects their financial stability and
flexibility. People who happened to have a claim that was well-handled might be more likely to renew,
not because they are high risk but because they better understand the value of insurance. In addition,
more financially savvy people tend to be healthier ones and might be more likely to stay in the program.
Ultimately, the direction of any selection effects in this type of micro insurance market is an open
empirical question.
We analyze how claim rates evolve as households renew their policies and find that households who have
larger claims during the policy period are slightly more likely to renew their policy for the next period.
Although that pattern is on the surface consistent with adverse selection, we instead find that when
compared to households joining the insurance in the same period, renewed households have significantly
lower claim frequency and total claim amounts. Taken together these results suggest that a) households
who experience claims perceive higher value to insurance, b) that households do not seem to be acting (at
3 Bockstal, Christine. 2008, “HMIS in national social protection strategies: Experiences from francophone African
countries.” Presentation in the 4th International Microinsurance Conference.
Preliminary Results – Please do not circulate or cite without permission from authors
5
least on the margin of renewals) on private information about their risk type and c) that there are forces
affecting insurance demand that lead to advantageous selection in the policyholders that retain coverage
over time.
The remainder of this paper proceeds as follows. Section 2 presents a literature review highlighting work
on adverse selection generally and the small literature on micro insurance. Section 3 introduces the
AKAM micro insurance program and our data, while Section 4 discusses our models and results. We
conclude in Section 5 about our findings.
II: Literature Review
2.1 Introduction to theory of adverse selection
The theory of asymmetric information was first established in 1970s with the seminal work of Akerlof
(1970), Pauly (1974), and Rothchild and Stigliz (1976), and it was further developed by generations of
scholars (see Miyazaki 1977, Wilson 1977, Finkelstein and McGarry 2006). In the classic Rothchild-
Stigliz model, insureds were assumed to be of different risk types, and the resulting market equilibrium
was a separating one where low risk individuals bought partial insurance with reduced welfare, and high
risk individuals bought full insurance. The results of these theories motivated the need of risk
classification as a solution to overcome adverse selection, since if the insurance company could
differentiate the low risk from the high risk, it could offer different contracts to both groups and improve
the welfare of society. In addition, these theories established the rationale of testing for adverse selection
in a given market using the “positive correlation test” between risk type and insurance coverage
purchased. For example, in Puelz and Snow (1994), they used data from auto insurance market of the
United States, and found those with higher accident risk choose lower deductibles (more insurance
coverage), following the “positive correlation test.”
To alleviate the effect of adverse selection, the insurance industry has developed techniques in risk
classification; however, they are still not able to prevent adverse selection completely. Numerous
empirical studies have tested for the existence of adverse selection in different types of insurance markets
around the globe (see Cohen and Siegelman, 2010 for a review), with mixed results. In some cases,
scholars even found empirically it was the low risk individuals who bought more insurance coverage,
which has been referred to as “advantageous selection” (Finkelstein and McGarry, 2006).
2.2 Adverse selection in microinsurance markets
Though the basic predictions from insurance theory should largely function similarly for microinsurance,
the nature of the product imposes a number of challenges in developing the market (Brau et.al, 2009).
Microinsurance as an emerging product for low-income people in developing countries could suffer from
adverse selection in a vital way, in particular because of its limited ability in classifying risks. Due to the
need to reduce administrative costs while keeping the product simple, microinsurance products are often
designed to be universal, without deductible or coinsurance options. There are mainly two reasons for
that. First, it fits the financial literacy of low income individuals. Usually a microinsurance policy is the
first insurance policy that the low income individual purchases, and having deductibles and coinsurance
requires more sophisticated knowledge about insurance. Without proper education and claim handling,
there could be misunderstandings that ruin the reputation of the insurer and insurance overall before the
Preliminary Results – Please do not circulate or cite without permission from authors
6
market can fully develop. Second, insurers are reluctant to implement more sophisticated policy features
because they want to hold down administrative costs. Effectively administering policies with deductibles
and coinsurance requires electronic record keeping and it may be necessary to implement a system for
tracking and verify claims on site at every clinic. Most microinsurers have not yet invested in these
technologies and infrastructure.
2.3 Empirical tests for adverse selection in “traditional” insurance markets
The traditional approach of empirical tests for adverse selection is a “positive correlation test,” i.e., a
positive correlation between risk type and insurance coverage is expected with the existence of adverse
selection.
Scholars tested for adverse selection in various markets using different proxies for risk type and insurance
coverage (see Cohen and Siegelman, 2010 for a review). The most common proxies of insurance
coverage are policies with different levels of deductibles and copayments, the decision to opt in and out of
insurance, and the decision to purchase supplemental insurance. Proxies for risk type ranged from
subjective measurement (self-evaluated health condition) to objective ones (indicators such as age and
medical history) and predicted risk type (see Browne, 1992 & 2006, Gao et.al 2009).
Others (see Fang et.al 2008, Bolhaar et.al 2008, Gao et.al 2009) have found a negative, rather than
positive, relationship between risk type and insurance coverage, which indicates that low risk individuals
purchased more insurance coverage. This negative relationship observed in empirical tests was referred
to as advantageous selection, in contrast to the traditional term of adverse selection. Besides risk
preference, various other sources of advantageous selection have been proposed, including heterogeneity
in income, education, health preferences, financial planning horizons, and cognitive ability.
2.4 Empirical tests for adverse selection in micro health insurance market
Since the development of microinsurance markets remains relatively immature, and data are often
inaccessible and imperfect, the empirical studies on the topic of adverse selection in this market are still
limited. The limited range of studies using various methods, however, all found evidence to support the
existence of adverse selection in micro health insurance markets of different countries in different time
periods.
A series of studies on adverse selection have been conducted using data from a rural mutual healthcare
insurance scheme in China. Wang et al. (2006) followed a voluntary mutual healthcare insurance scheme
in a rural county in China and carried out a panel data analysis over the period of 2002 to 2006. Using
individual level data, they found strong evidence of adverse selection despite a high enrollment rate and
the requirement of having the entire household enroll as a unit. In particular, they found the pre-enrolled
medical expenditure for enrolled individuals were 9.6% higher than average expenditure of all residents.
In addition, the enrolled members of the partially-enrolled family spent 1.7 times more than those non-
enrolled members of the partial enrolled family. Zhang and Wang (2008) also observed that people with
chronic condition history, with fair or poor health were more likely to enroll in the program, showing the
existence of adverse selection. The extent of adverse selection, however, seemed to be stable over the
study period.
Similarly, Ito and Kono (2010) found evidence of adverse selection in a micro health insurance program
in India, based on the result that households with a higher ratio of sick members were more likely to
Preliminary Results – Please do not circulate or cite without permission from authors
7
purchase insurance. Using data from Masisi district in Zaire, Noterman et.al (1995) also found evidence
for adverse selection. They examined the null hypothesis that the increase in hospital utilization by
insureds can be attributable equally to predictable and unpredictable risk conditions and rejected the null
hypothesis.
With a historical survey data of American short-term disability microinsurance back in the early
twentieth-century, Murray (2011) found prima facie evidence of asymmetric information, with evidence
for the presence of adverse selection stronger than that of moral hazard. In addition, it was shown that the
countermeasures taken by the microinsurers, including the enforcement of a trial period and waiting
period, effectively reduced claims.
Besides the effect of adverse selection in explaining the demand for micro health insurance, there are also
other factors affecting the purchase decision which are studied empirically in numerous research studies,
including income, wealth, education, history with insurance, gender, family composition, residence
district, etc. (see Dror et.al, 2006; Ito and Kono, 2010; Bhat and Jain, 2006; Donfouet and Makaudze,
2010, Msuya and Asfaw, 2004; Rao et.al, 2009; Schneider and Diop, 2001).
III: AKAM Micro Health Insurance Program and Data Description
3.1 AKAM program background
AKAM micro health insurance program in Pakistan
The AKAM Microinsurance Initiative commenced in 2006, with support from the Bill and Melinda Gates
Foundation. AKAM is owned by the Aga Khan Development Network (AKDN) and it started its pilot
enrollment period for an annual micro health insurance policy in the Northern Area (NA)4of Pakistan in
Nov 2007. NA is the northernmost political entity located in the mountainous part of Pakistan, with an
estimated population of 1.35 million scattered across six districts. The provision of health insurance is a
milestone in that it is the first effort to make health insurance available to the poor in the area. In the
three years since the program first launched, there were over 100,000 members enrolled, which accounted
for around 7.5% of the local population.
The coverage and premium are the same for every member and the individual annual premium of 400
Pakistan Rupees (PKR) (approximately $5.60) is paid up front5. There is no individual risk classification
in underwriting. In exchange for this premium, the policy provides the following coverage: annual
hospitalization coverage (the core of the product) up to 25,000 PKR (approximately $400)6, life insurance
of 25,000 PKR (approximately $400) on the head of the family and one outpatient voucher valid for a
one-time physician visit. The hospitalization coverage and the life insurance coverage increase to 30,000
PKR for renewed insureds for the same premium, providing motivation for policyholders to renew their
policies. The average annual income per person in NA is around 50,000 PKR in 20097, so the insurance
coverage is equivalent to about half of their annual income.
4 The Northern Areas (NA) is now known as Gilgit-Baltistan(GB).
5 The premium was 350 in Nov 2007, and increased to 400 for Nov 2008, July 2009 and Nov 2009. It increased to
450 PKR for new insureds in ZADO and Danyore in July 2010, but stayed the same for all the other insureds. 6 25000PKR convert to $400 in Nov 2007, $312 in Nov 2008, and $293 in Feb 2011.
7http://www.unicef.org
Preliminary Results – Please do not circulate or cite without permission from authors
8
Distribution of micro insurance and provision of health service
Figure1:Flow chart for AKAM micro health insurance program
AKAM partnered with local supporting organizations (LSOs) to distribute the product to voluntary
groups. There were two purposes for selling group policies. First, the local supporting organizations
have an existing network for disseminating information and hence greatly reduce underwriting and
distribution costs. Second, as the way it works in developed markets, selling group policies helps to
address concerns of adverse selection because these groups are not formed solely for purchasing health
insurance. The major types of group are village organizations (VOs) and women’s organizations (WOs),
and their members could purchase health insurance through the VO/WO if at least 50% of the households
in the organization agreed, in the test survey, to purchase the product if it were available. In addition, the
entire household is required to enroll in the program in order to alleviate adverse selection.
AKAM initiated policies for New Jubilee Life Insurance Company (NJLI), which is a commercial insurer
based in Karachi, Pakistan, owned also by AKDN8. The LSO enters into an agreement with AKAM,
which has been appointed by NJLI to represent it in all matters pertaining to the health microinsurance
program. The LSO contracts with NJLI on behalf of their village member households.
8In regard to the reinsurance arrangement, AKAM arranged a stop loss contract with Swiss Re.
Parent organization
Policyholders
Healthcare providers
Aga Khan Agent of
Microfinance-
Microinsurance
Initiative
(AKAM-MI)
Financial institutions
Reinsurance provider:
Swiss Re
Seller of insurance policy
(insurer): New Jubilee
Life Insurance (NJLI)
Legal representative for
buyers of group contract:
Local supporting organization
(LSO)
Members of LSO: Village
organizations and women’s
organizations (VO/WOs)
Policyholder: Households in
qualified VO/WOs
Other healthcare provider:
Combined military
hospital (CMH) and
government hospital
Main healthcare
provider: Aga Khan
Health Services Pakistan
(AKHSP)
Aga Khan Development
Network (AKDN)
Preliminary Results – Please do not circulate or cite without permission from authors
9
AKAM chose NA to be the first area to provide health insurance because the AKDN intensity is very high.
The main health service provider (Aga Khan Health Services Pakistan, AKHSP), which is a part of
AKDN, operated in NA for over thirty years. It has three hospitals and twenty-five primary care facilities
in NA, and over 90% of the claims from AKAM micro health insurance program are handled within
AKHSP systems. In addition, there are the Combined Military Hospital (CMH) and the government
hospital located in NA, which altogether take less than 10% of the claims, mostly for emergency service.
Background of VO/WO/LSO
AKAM relies on an existing network for village organizations and women’s organizations for product
distribution. By the end of 2005, there were more than 4,000 Village Organizations (VOs) and Women’s
Organizations (WOs) in Northern Areas, covering more than 78 percent of the total households9.
VO/WOs are grass root community organizations that were first built to improve the households’ capacity
for undertaking village development initiative. For example, projects include obtaining funding and
organizing the villagers to work on infrastructure project such as minor irrigation works, flood protection,
erosion control and link roads. In addition, VO/WOs were also involved in organizing informal
community-based micro loans among their members before AKAM started its formal microfinance
service in the area. More recently, those organizations have become key players in providing supportive
networks to enlarge communities’ assets and harness individual skills to generate income in a sustainable
manner.
Local Supporting Organizations (LSOs) are large organizations working for the member organizations in
the area, which may consist of 50 or more villages. They are nonprofit organizations set up in joint effort
with the Aga Khan Rural Support Program (AKRSP) and local population, serving as registered legal
entities under the Pakistan law. LSOs are the umbrellas under whom VOs and WOs could enter into any
legal agreement as a group or sub-group since VOs and WOs are not registered organizations. Gradually,
LSOs have obtained project funding and expanded their service to a wide coverage. Currently, there are
around 40 LSOs in the region.
3.2 Data description
The empirical application utilizes a dataset comprised of household level information, including some
basic demographics on the head of the household’s age and gender, members’ age and gender
decomposition, household size, village organization and local supporting organization to which the
household belongs. It also has some policy level information such as policy limit, renewal status and
enrollment date. Moreover, it contains detailed information on claims made during the policy periods,
including diagnosis, whether it was a pre-existing condition, hospital stayed, date of admission, length of
stay, total bill and bill divided into subcategories of medicine, surgical procedure, surgical supply, bed
charge, inpatient consultation, lab test, and others.
There were in total six enrollment periods since the program first launched in November 2007. Each
policy lasts for one year, and the enrollment window was fixed to November at the beginning, so that it
helped alleviate adverse selection by not allowing people to speculate in purchasing an insurance policy
9The Aga Khan Rural Support programme, an assessment of institutional development of village and women’s
organizations, results of the AKRSP’s Institutional Development Survey 2006.
Preliminary Results – Please do not circulate or cite without permission from authors
10
right before a surgery was needed. In addition, it helped reduce the administrative and distribution costs.
Besides the four November waves of enrollment from 2007 through 2010, a new July enrollment window
was opened in 2009 due to the increasing demand for health insurance as well as the improved
recognition of income cycle for local households.
The first enrollment period starting in Nov 2007 was a pilot period; therefore detailed information was not
collected at the time. As a consequence, we could not include the first period into the data analysis. For
the fifth enrollment period starting from July 2010, claim data is available only through Nov 2010. Since
we only have part-year claim data for that enrollment window, for any analysis based on that period we
scale the available claim data to an annual basis.
To sum up, the data analysis is based on data from four available periods, which are Nov 2008, July 2009,
Nov 2009 and July 2010, with a total member observation of 64,29010
.
Because the two key variables we examine are claim experience (both frequency and severity) and
renewal status, we summarize detailed information over different enrollment periods on those variables in
table 1 and 2 respectively.
Table 1: Summary of household level claim information and loss ratio for all enrollment periods