1

DRAFT PAPER

Chronic Poverty Research Centre 2010 Conference - Ten Years of ‘War against Poverty’:

What have we learned since 2000 and what should we do 2010-2020?, University of

Manchester, 8 – 10 September 2010

Uptake of Micro-Life Insurance in Rural Ghana

Lena Giesbert (GIGA Hamburg and Humboldt University of Berlin)

Abstract:

This paper investigates the conditions under which households decide to participate in the emerging microinsurance market. It estimates the cross-sectional determinants of households’ decision to take up a micro life insurance by using data from a survey conducted among 1030 insured and non-insured households in rural Ghana in 2009. Insurance participation is examined against a simple neoclassical benchmark model, which argues that life insurance take-up increases with risk aversion, the prospect variability of risk, initial wealth, and with the ‘intensity for bequests’. In line with earlier studies on micro health insurance and agricultural insurance, it is found that participation in micro life insurance is explained by a number of factors which are at odds with the predictions of the conventional model. The evidence confirms the outstanding role of trust and social networks for the probability of purchasing micro life insurance. This is underlined by a strongly negative association of the idiosyncratic risk assessment within the household with the uptake of micro life insurance, suggesting that households view the microinsurance policy itself as a risky option. Keywords: Vulnerability, Household Behaviour, Life Insurance, Ghana JEL classification: O16, G21, D12

Author contact:

Lena Giesbert, GIGA Hamburg, Neuer Jungfernstieg 21 20354 Hamburg, Germany email: giesbert@giga-hamburg.de phone: +49-40-42825-566, fax: +49-40-42825-511

1

1. Introduction

During the last decade, microinsurance markets have been growing rapidly in the developing world. This accommodates increasing recognition of the fact that the poor desire and operate with a whole range of financial services to accumulate capital and manage with risk (Collins et al. 2009). Extending the concept of microfinance into the realm of insurance, microinsurance is seen as a formal tool that enables the poor to better operate with the consequences of shocks such as death, illness, droughts or floods, which often cause severe drawbacks in their attempt to escape their vulnerable livelihoods (Siegel et al. 2001; Churchill 2002; Cohen et al. 2005; Dercon 2005; Dercon et al. 2008b).

Quite a substantive body of literature has shown that in the light of incomplete conventional financial and insurance markets poor households engage in income smoothing activities that reduce temporary income fluctuations, but often come at the cost of lower total returns to wealth (Rosenzweig and Binswanger 1993; Murdoch 1995; Platteau 1997). A range of informal risk sharing mechanisms has been identified, that balance consumption variability to some extent, but usually not entirely. In addition, the degree of consumption smoothing via informal risk management strategies seems to be higher for wealthier households, as compared to poorer ones (Townsend 1994; Murdoch 1995; Townsend 1995; Dercon 2002). At the same time, public social security systems and safety nets are typically weak and often cover less than 10 percent of the population in developing countries, the majority of whom are employees in the formal sector (ILO 2001). In this context, microinsurance not only offers a direct welfare benefit through a payout in the case that an insurable loss occurs, it is also seen as an effective tool to prevent households from engaging in insufficient and costly alternative ways of coping with shocks.

As demonstrated by a number of studies, the most frequent and stressful risks in developing countries are those of illness, death of an income earner in the household, and property loss as a result of theft or fire as well as risks in agriculture, such as droughts and floods (Dercon 2002; Cohen and Sebstad 2003; Tesliuc and Lindert 2004; Cohen et al. 2005; Dercon et al. 2008a). Indeed, these risks correspond highly to the types of microinsurance products offered today. In order of frequency, they include life insurance, accident and disability cover, property and index insurances, and health insurance.1 Notwithstanding the potential of microinsurance offering secure protection at affordable prices for poor households to hedge against some of the biggest risks they are exposed to, take-up rates have so far remained low. Although microinsurance products have been identified in 77 out of the 100 poorest countries of the world, in most of the countries they cover less than 5 percent of the total population (Roth et al. 2007).2

What prevents low-income households to participate in formal insurance markets, once they are available? Or, in other words, what are the major determining factors of microinsurance uptake and, hence, which are the households that are currently reached by this new tool to manage risk? This paper addresses these questions by studying the patterns of household

1 Although live insurance clearly outnumbers all other types of microinsurance, it is important to note that there exist quite numerous compound products, for instance covering life, hospitalization and disability at the same time. In addition, more than 60 percent of the total number of life insurance products is, in fact, tied to a loan. These are often seen as something that benefits the lender rather than the policyholder in satisfaction surveys (Roth et al. 2007). 2 Of the 78 million insured people, the majority of 67.2 million has been located in India and China, nonetheless amounting to only 2.7 % of the total low-income population in these countries. The coverage of the poor in Africa and Latin America is at about 0.3 percent and 7.8 percent, respectively (Roth et al. 2007).

2

participation in the most frequent type of microinsurance, i.e. a micro life insurance in rural areas in Southern Ghana.

The success of life insurance in the microinsurance business is due to a range of factors. First, it is one of the most demanded forms of coverage, as indicated by the range of highly severe risks mentioned above. Second, it is easier to provide than other types of insurance, such as property-, health-, or index-based weather insurance. Due to the clear-cut nature of the loss event, for instance, it is relatively uncomplicated to price, mostly resistant to fraud and moral hazard and not dependent on the existence and efficient functioning of complex other infrastructure, such as hospitals, or rain gauge systems and the like. Third, it is easy to link to other microfinance savings and loan products and to distribute via delivery channels of microfinancial institutions that have already built up good client relations to the target group (Roth et al. 2007).3

In the Ghanaian context, the relevance of life insurance is threefold. Not only may there be a long-term permanent loss in total household income if a working household member dies. There is also an immediate need for funds to cover funeral costs, which are often substantial. Contrary to older traditions, where people were buried soon after their death, in many Ghanaian communities there has evolved a custom of stocking corpses of deceased relatives for long periods, sometimes months, before they are buried. This is due to economic interests of the involved facilities, but also for the reason that funerals are seen as an opportunity to show and enhance social status and prestige, the mortuary system being part of the ritual norms to be followed. If access to sources of cash - such as loans and donations from social networks or remittances from migrants - is limited, these events may result in ruinous consequences for the remaining household (Arhin 1994; Geest 2006; Muzzucato et al. 2006). Eventually, endowment type micro life insurance may play an important role in filling gaps in existing public old age security systems. Participation in Ghana’s national pension scheme (SSNIT), for instance, is almost exclusively confined to formal sector employees.4

While there is now increasing interest in the study of microinsurance, empirical contributions from academia are still limited and have mainly concentrated on the analysis of health insurance (Preker et al. 2002; Dror et al. 2006; Wagstaff et al. 2007; Wang and Rosenman 2007; Chankova et al. 2008; Ito and Kono 2010) and, to a lesser extent, agricultural insurance (Sakurai and Reardon 1997; Giné et al. 2008; Cai et al. 2009; Cole et al. 2009). From the existent evidence, however, it can be expected that participation patterns in (any) microinsurance markets are not necessarily consistent with predictions of a simple conventional benchmark model. Giné et al. (2008), followed by Cole et al. (2009), show that participation in a rainfall insurance for farmers in India indeed matches with some of the predictions of such a benchmark model augmented with borrowing constraints. That is, insurance take-up is decreasing in basis risk between insurance payouts, expected income fluctuations and credit constraints faced by a household, and increasing in household wealth. Contrary to neoclassical insurance theory, however, is their finding that risk averse households are significantly less likely to take-up insurance. Altogether, their results suggest that households which are unfamiliar with the insurance product and the distributing institution, or its staff, view purchasing insurance as a risky endeavour, rather than a decision for safety. Hence, trust between the insurance provider and a potential client seems to be a powerful explanatory factor in microinsurance take-up behaviour. This has also been confirmed by Cai et al. (2009) in the context of a government-subsidized product, where

3 It is worth noting that this applies not only to the microinsurance market, but as well to the emerging conventional insurance markets in developing countries (Oetzel and Banerjee 2003) 4 The overall coverage of the scheme remains at about 10 percent of the total workforce. While informal sector workers are not per se excluded from the scheme, their share is negligibly low at around 1% of the total active members (Boon 2007).

3

farmers are found to draw away from the insurance as their level of trust in the local government in China is low due to frequent experience of policy delivery failures.

Although most widespread in practice, empirical studies on the purchase of micro life insurance products are so far not available and analyses of conventional (life) insurance markets in developing countries are confined to a number of (cross-country) studies based on macroeconomic data (Browne and Kim 1993; Beck and Webb 2002; Nakata and Sawada 2007). According to the conventional theoretical framework, participation in life insurance markets increases with risk aversion, the life time expectancy (or probability of risk), initial wealth as well as the ‘intensity for bequests’ (measured by the demographic composition of the household), and decreases with the fraction of total premium exceeding the true expected loss.

Against this benchmark, the analysis evaluates the determining factors of micro life insurance uptake based on household data from a survey among 1030 insured and non-insured households in three areas of Southern Ghana. By taking into consideration alternative theoretical frameworks such as adverse selection models and prospect theory, however, it places specific emphasize on the role of the idiosyncratic risk assessment within the household versus its objective probability of risk and the past experience of shocks. Additionally, it tests for the impact of trust and social networks as well as information and the existence of other risk management strategies, for the probability of purchasing micro life insurance.

Section 2 gives an overview on the major theoretical determinants of (life) insurance participation. Section 3 gives an overview on the features and distribution of the life insurance product followed by a description of the data in section 4. Section 5 describes major summary statistics and the construction of variables. Section 6 presents the empirical results and section 6 concludes.

2. Theoretical Predictions of the Determinants of Life Insurance

Participation

Tracing back to the seminal work of von Neumann and Morgenstern (1944), expected utility theory and the concept of risk aversion have become the basic framework in the analysis of risk and insurance. According to this theoretical framework, it is assumed that people are risk-averse and, as they exhibit diminishing marginal utility with respect to wealth, they purchase insurance because they prefer the certainty of paying small premiums to secure future income streams to the risk of suffering a large financial loss when a shock occurs (Mossin 1968). In a slightly different perspective, it has also been postulated that instead of paying for future security, by purchasing insurance individuals transfer resources from low marginal utility of income states to future income states where marginal utility is high (Debreu 1959; Arrow 1971; Nyman 2001; Karni 2006). Central to this theoretical work further is the assumption that with decreasing absolute risk aversion and rising wealth, the willingness-to-pay for insurance, i.e. the additional amount paid to exchange the prospect of risk against a certain level of wealth (risk premium), decreases, hence, insurance may be an inferior good (Pratt 1964; Mossin 1968; Arrow 1971).5 However, as the effect of wealth on the decision to

5 Note that most theoretical work presented in this section refers to the demand of insurance, i.e. the amount of insurance purchased in markets where the amount of coverage is theoretically not constrained. Other than that, this study is concerned with the purchasing decision of consumers (households, in this case), or, in other words, the mere choice of whether or not to participate in the insurance market. However, as this can be viewed as a precondition for the question of demand, most determinants specified in the respective frameworks should hold in the case of the participation question as well.

4

purchase life insurance depends on the consumers risk aversion function it can be ambiguous and is to be established empirically (Outreville 1996).

In a full information setting and in line with the above outlined expected utility framework, standard theoretical models of life insurance assume that consumers – and their households - maximize utility by reducing uncertainty in their income streams due to the possibility of a premature death of a primary income earner. Guided by the pioneering work of Yaari (1965) and Hakansson (1969), most of the early theoretical studies focus on the decision to purchase life insurance of the primary income earner, who maximizes the weighted sum of her own lifetime utility from consumption and that of bequests. In Yaari’s life-cycle model, a consumer with uncertain lifetime purchases life insurance to increase her expected utility in the following manner:

∫ +=T

TSTdttCgtTUE0

)]([)()([)()]([ ϕβα (1)

where T is a random variable with a known probability reflecting the consumer’s lifetime, u[c(t)] is the immediate utility c at any moment t, and φ[S(T)] is the immediate utility of bequests. Further, there are discount factors α(.) and β(.), which reflect the factor by which future consumption is discounted. Therefore, β can be seen as a subjective weighting function for bequests, which is expected to increase noticeably as consumers marry or have offspring and to take on a hump-shape curve, as the importance of bequests is greatest when the consumer dies at prime age, i.e. the middle years. Within this framework, the demand of insurance is a function of wealth and the expected income over the lifetime, the size of premiums (including the loading factor), the subjective discount rate for current over future consumption, and the consumers’ lifetime probability.

In an extension of Yaari’s model, Lewis (1989) partly endogenizes shifts in the demand for insurance by explicitly incorporating the preferences of the dependents and beneficiaries into the model. Thus, offspring purchase life insurance as they face an uncertain income stream resulting from a parents’ uncertain lifetime. By specifying the utility maximization problem for the offspring and spouse separately, Lewis shows that total life insurance ownership can be written as:

−

−

−=− 0,

)1(

1max)1(

1

WTCpl

lpFlp

δ

(2)

where l is the policy loading factor (the ratio of the additional cost of an insurance policy to its actuarial value), p is the probability that the primary wage earner dies, and F is the face value of all insurance written on that persons’ life. The parameter δ is some measure of the beneficiaries’ relative risk aversion, TC represents the consumption of the spouse from the current period to the predicted end of his or her lifetime and that of the offspring from the current period until he or she leaves the household, and W denotes the households’ wealth net of the spouses own bequest at a certain age.

It is important to note that the theoretical framework presented so far is directly applicable only to the mortality risk component of life insurance, referred to as term life insurance. However, the determinants of the savings and annuity components of life insurance policies may be somewhat different. As pointed out by Pissaridis (1980) in a life-cycle model with uncertain age of death, additional to the motive for a bequest one should in general expect a motive of saving for retirement.6 In the context of the industrial countries usually considered in studies on conventional insurance demand, life insurance often combines features that

6 In the context of developing countries, this would rather refer to the period after regular productive work of primary income earners, which is normally not officially set at a certain age, but depends on the individual’s context conditions, such as his or her health condition and physical ability to work.

5

serve both these motives simultaneously. As individuals normally survive until retirement age, life insurance is considered primarily a “pension”, while the bequest motive would be satisfied by the fact that, in exchange for a reduction of the pension, the insured amount is made available to the consumer’s dependents by the insurance company should the policy holder die before retirement. Therefore, Pissaridis points out that preferences on the utility from bequests and that of consumption may differ over the lifetime, as the former may be discounted more heavily than the latter, if the bequest motive of insurance purchases diminishes with age – and, consequently, the retirement motive becomes more important.7 Empirical contributions have underlined that term and whole life forms of insurance are not necessarily substitute goods (Babbel and Ohtsuka 1989; Outreville 1996). As the microinsurance under study includes features of term and whole life insurance (due to a voluntary savings component added to the mortality cover), the potential heterogeneity in life insurance contracts with regard to bequest and savings motives will be taken into account in the subsequent analysis.

In summary, the standard elementary benchmark model predicts the decision to purchase life insurance in a simple setting without information asymmetries to be a function of the household’s risk aversion, the accumulated wealth of the household and the expected wealth over the consumer’s lifetime, the expected lifetime (in other words, the probability of the risk), the costs (often proxied by the loading factor of a policy, i.e. the fraction of the total premium that is in excess of the true expected loss), and the household’s intensity for bequests.

However, several authors have pointed to remaining insurance puzzles, as many households remain uninsured against substantive income risks, or “over-insure” given what the probability of the risk would justify. Particularly in the context of emerging (micro) insurance markets alternative theoretical approaches may be of great importance, as full information may not necessarily be given.

As one explanation for observed deviations from the standard model, a number of studies have analyzed the role of adverse selection and moral hazard for insurance purchasing behaviour (Rothschild and Stiglitz 1976; Chiappori 2000; Dionne et al. 2000; Winter 2000; Abbring et al. 2003). Central to asymmetric information models is the assumption that contracts with a more comprehensive coverage are chosen by agents with higher expected costs from the realization of an insurable risk. In their seminal work, Rothschild and Stiglitz (1976) postulate that there is a separating equilibrium with two classes of individuals based on their high and low risk probabilities, respectively. Conditional on individual preferences and the risk probability, the competitive equilibrium in the insurance market lies at different points. The concept of moral hazard is of limited applicability in the context of life insurance, as it is unlikely that the party with more information has the tendency or incentive to provoke the insured event (death) to happen.8

Adverse selection, however, can indeed be expected to take place in the context of the microinsurance contract under study. While patterns of life expectancy and health outcomes in communities are publicly observable to some extent, it is still likely that consumers have an informational advantage vis-à-vis the insurer on their individual riskiness. This is particularly

7 Note that a specific demand function of insurance is not specified in this framework, as Pissaridis rather considers the life-cycle allocation problem with uncertain life expectancy and resulting wealth-age profiles of individuals. Therefore, insurance only enters the function as a net return in different specifications of the allocation maximization process in worlds with perfect or imperfect insurance (Pissaridis 1980). 8 It may only present a problem to the extent that such behaviour takes place with regard to the additional components of hospitalization and accident benefits included in the policy under study. Evidence does not suggest that this is a problem in practice, however, as there is very limited understanding of these additional components in the policy, as will be outlined in greater detail later.

6

the case if detailed health checks are not necessary or technically not possible before buying the insurance. Hence, if households that are more prone to the risk of pre-retirement death and bad health are more likely to purchase insurance on could thus infer problems of adverse selection in the market.

Different from the preference-for-certainty hypothesis of the standard insurance theories, Kahnemann and Tversky have developed prospect theory as an alternative framework postulating the asymmetric evaluation of gains and losses by consumers (1979; 1981). These are assumed to react in a risk-loving way when confronted with the probability of a loss and to act risk-averse only towards gains from a reference point of wealth. An important insight of this theory also is that individuals tend to overvalue high-probability events, whereas they undervalue medium-probability and low-probability events. Thus insurance is purchased only in the case that the subjective risk perception (i.e. in the case of an overestimation of a high-probability event) compensates for the under-valuation of a loss relative to the reference point. In addition, the heterogeneity with regard to the subjective expectation of a risk realization (translating into a potential insurance payout) may be shaped to a great extend by the household’s real experience with corresponding shocks, leading to a higher wariness towards them.9

Existing empirical studies on the participation in microinsurance markets regarding health insurance and crop insurance products have further pointed to the important role of trust in explaining deviating insurance purchasing behaviour from the standard model (Giné et al. 2008; Cai et al. 2009; Cole et al. 2009). While trust has not been explicitly considered in models of insurance consumption, it has received some attention in theoretical work on financial market participation, more specifically, stock market participation. Including a measure of (mis)trust in their theoretical model on stock market participation, Guiso, Sapienza and Zingales (2007) show that the perception of risk is not only a function of the objective characteristics of the stock, but also the consumer’s subjective probability to be cheated. Less trusting individuals are thus less likely to participate in the stock market. Related to the issue of trust, Hong et al. (2004) propose that stock market participation is influenced by social interaction, in that “social” consumers find it more attractive to invest in stocks when more of their peers participate.

Micro life insurance is still a very young financial product in the empirical setting of the study, which has only been offered for five to six years. Given expectedly low levels of financial literacy and the limited exposure with formal insurance in the survey region it can be assumed that many households are indeed uncertain about the terms and conditions of the policy and/or do not really trust the insurer and the distribution channels. Enhanced access to information on, experience with other financial services offered by the same agencies and/or the appreciation and guidance by trusted third parties, e.g. social peers, may thus be of central importance for the uptake of the insurance product. However, household’s engagement in social networks may also be ambiguous as this could enhance their access to informal support networks in the case of shocks. Generally, households which have access to (effective) alternative strategies to cope with death or illness, such as remittances or transfers from family members or friends, may have less incentives to buy a formal insurance policy.

It is important to note that life insurance purchasing behaviour is not entirely driven by factors on the side of the consumers. There are important supply-side factors, which affect the availability and the price of life insurance, for instance. Among these are the market conditions that insurance companies or intermediates face and information which guides the actuarial specifications of the product. The data in this analysis does not allow identifying such supply-side factors in a strict sense, due to the facts that a) the loading expenses of the

9 Note that the empirical data does not allow controlling for different risk preferences towards gains or losses, but only the subjective evaluation of the risk exposure towards different types of risk.

7

different insurers in the market are not known, b) there is limited variation in prices of life insurances in the survey areas since the insurance market in rural areas in Ghana has only started to develop and competition has remained low, and c) exact information on the access conditions of certain insurance products other than the microinsurance product specifically under study is not available. The empirical analysis will thus have to rely on a reduced-form approach. Potential problems of omitted variable bias will be taken into consideration by giving special emphasize on the description of the conditions under which the microinsurance product under study is sold, i.e. features of the product, marketing strategies and provision channels. Further, it will take into account potentially different market conditions across the survey areas by controlling for local fixed effects.

3. Distribution and Marketing of the Anidaso Policy10

The insurance market in the survey areas includes the microinsurance under study, called Anidaso policy ( “anidaso” meaning “hope” in Twi), and few other insurances. These entail private health, life, and property insurances offered by commercial insurance companies that are not specifically targeted to low-income households. Otherwise, national health insurance is provided by the National Health Insurance Scheme (NHIS), which was launched in 2004 and replaced the former cash-and-carry health care system. It provides medical care at public hospitals, recognized private hospitals, and health centers for contributors and their dependents. Premiums are graded by income, and particular groups, such as the elderly, indigent people and pregnant women are covered free of charge. The NHIS is well received, especially in rural areas, where a majority of people had hitherto gone without health services as a result of lacking resources and insurance alternatives.

The Anidaso policy is provided by the commercial Gemini Life Insurance Company (GLICO). The product was developed with initial support of CARE International between 2001 and 2004. After the product development and trial phase, GLICO took full responsibility for the product and does not receive subsidies of any kind for this insurance today.

The policy offers term life insurance up to age 60, accident benefits, and hospitalization benefits (calculated per day spent in hospital) for the policyholder, the spouse, and up to four children. Contributions towards a so-called investment plan, which serves as a savings scheme and pays the accumulated amount at the maturity of the term, can be added on a voluntary basis. During the research it became obvious that most policyholders are actually unaware of the accident and hospitalization benefits and consider Anidaso to be a pure life insurance or, to a lesser extent, a savings device.11 The latter option may either be motivated by the intention to receive a pension, or to have the opportunity of a partial withdrawal, which is possible after several years of payment. The policy is specifically targeted at low-income people in both urban and rural areas.

For the sale and distribution of the policy, GLICO started to cooperate in early 2004 with six rural and community banks (RCBs).12 It currently collaborates with 20 RCBs, five microfinance institutions (MFIs) and one savings and loan company in six regions of southern Ghana. The number of Anidaso policyholders per financial institution ranges from around 200

10 This section borrows heavily from a respective section in another paper based on a pilot study of the same insurance product written together with Susan Steiner and Mirko Bendig (Giesbert et al. 2010). 11 The fact that GLICO has hitherto only received claims upon death of policyholders but no claims in relation with any of the additional policy components underlines our impression that policyholders consider the Anidaso policy to be a pure life insurance. 12 In general, RCBs are unit banks owned by members of the community. While they do not exclusively target low-income people, their business is by and large microfinance orientated because the majority of the population in their service areas can be classified as low-income (Basu et al. 2004; Steel and Andah 2008).

8

to over 1,000, and the total number of policyholders had reached 15,000 by December 2008. In each of the partnering financial institutions, GLICO assigns one Personal Insurance Advisor (PIA), who is in charge of marketing the Anidaso policy and mediating all running operations between the bank and the insurance company. In addition, there is usually a team of a minimum of two sales agents that joins forces with the PIA in the marketing process. The PIA and the sales agents are typically recruited locally, but they are trained at GLICO’s headquarters.

GLICO’s marketing strategy includes approaching group and opinion leaders in the communities, who are then mobilized to spread the word about the product and to help organize marketing meetings. Furthermore, PIAs and sales agents attend group meetings of microfinance groups of the rural banks or other (financial) self-help groups, accompany rural banks’ mobile bankers13, make individual door-to-door marketing rounds and approach visitors at the bank. Less frequently, GLICO holds large and widely announced product launches at community centers and bank offices. Interested individuals can usually apply on the spot.

There are no clearly defined eligibility criteria for policyholders except that they have to be adults below the age of 55 and that they have, or are willing to open up, an account with the local financial institution. This latter condition is necessary because the insurance premiums are directly deducted from policyholders’ accounts; or from group accounts (if policyholders are organized in groups).14 No detailed health check or information on the health condition of applicants or other household members is required. In fact, this feature of the policy is used for promotion purposes in the Anidaso policy information flyer. The monthly premiums start at 2 Ghana Cedi and may be as high as 10–15 Ghana Cedi if the savings component is chosen.15

4. Source of Data

The analysis is based on a household survey of 1030 households conducted by the author in cooperation with the Institute of Statistical, Social and Economic Research (ISSER) of the University Legon in southern Ghana from January till March 2009. The survey was undertaken in the context of a research project on household risk management and the participation in microinsurance markets of low-income households in sub-Saharan Africa. In an ex ante selection process, the Anidaso policy was chosen as GLICO had been identified as the only known insurance provider in sub-Saharan Africa offering voluntary life insurance to low-income households.16

13 These operate in the same (but formalized) way as so-called susu collectors in the informal financial sector, to which we refer below. 14 Financial groups are very common in Ghana. In the formal financial market, they usually have a joint savings account and accumulate savings from their members in order to qualify for a loan. In the case a loan was granted, the group handles the collection of repayments, acts as a mediator between the loan officers and the individual group members, and bears responsibility for recovery (Steel and Andah 2008). 15 In our sample of 321 Anidaso policyholders, the mean monthly premium is 5.3 Ghana Cedi and the median is 3.1 Ghana Cedi. The exchange rate at the time of our survey (February 2009) was 1.00 Ghana Cedi = 1.25 US Dollar. 16 This selection was done in the year 2007. At that time, all other providers of which information was available had an insufficient number of clients, offered only compulsory (mostly credit life) insurance, or provided health or heavily subsidized agricultural insurance. However, since information on microinsurance providers and products is fragmentary, it may well be that voluntary microinsurance products besides GLICO’s Anidaso policy existed that we were not aware of. Due to the dynamic nature of the market, it can be assumed that there are many more voluntary life insurance products today.

9

In a next step, we selected the survey areas by choosing communities in three services areas of the 26 financial institutions that distribute the policy. In doing so, we ensured regional variation by selecting the three service areas from three different regions. We only considered RCB’s which served semi-urban or rural areas17 because we intended to make sure that we would find a high share of low-income people in the overall population, assuming that people in these areas are on average poorer than people in highly urban areas. We also paid attention to a sufficient density of bank clients holding an Anidaso insurance contract. Out of eight possible survey sites according to the above criteria, we randomly chose three RCB’s and their service areas situated in a) the Agona West Municipal District in the Central Region, b) the Akuapim North District in the Eastern Region and c) the South Tongu District in the Volta Region. Within these service areas, the communities were deliberately chosen to include an equal share of communities with insured clients and communities comparable in size, infrastructure and access to the rural bank’s services without any insured clients.

Of the total sample, a third of all households were randomly drawn from Anidaso client-lists in the localities with policyholders.18 In the same localities, another third of households which were not insured was randomly selected according to a counting procedure in each of the localities, where the counting interval was set according to the official total number of households obtained from the National Census in 1998/1999. We only included localities with a sufficient number of policyholders to obtain a meaningful amount of insured and non-insured interviewees in each locality. Finally, another third of households was randomly selected in the comparable communities without Anidaso policyholders using the same counting procedure as described above. Thereby we included a total number of 17 communities in the sample across the three regions. Table A1 in the Appendix shows the number and share of Anidaso policy holder and non-policy holder survey households across these survey sites.

While external validity in a strict sense may not be fully given, it is assumed that scope for generalization goes beyond the local areas of the survey itself. First, the results should be at least representative for semi-urban locations of the eight service areas of RCB’s in the South of Ghana where microinsurance is available. Second, there is little reason to assume non-random program placement based on particular characteristics of cooperating institutions. In principle, GLICO would distribute the Anidaso policy through any formal financial institution that has the capacity to do so in terms of the deduction of premiums from policyholders’ accounts and is willing to cooperate. Such financial institutions, mainly in the form of RCBs or MFIs, are represented in every district capital and many other towns in the Southern Regions of Ghana. However, it is important to note that GLICO has started to distribute the policy only with some of the existent financial institutions and there remains some lack of clarity as to why these showed interest and were selected and others not. One reason may simply be to start at some point and expand from there.

5. Summary Statistics and Construction of Variables

Summary statistics are presented in Table 2. Wealth and demographic characteristics of the sample underline that the survey areas cover semi-urban locations in the southern regions of Ghana. It includes low and middle-income households with an average total income of 78.687 Ghana Cedi per month per adult equivalent (around $ 62.95), which is about twice the total

17 Out of the 26 financial institutions, 12 were located in an urban setting and 4 had only inactive clients or a limited number of clients. The Ashanti Region was not included in the selection as it is a relatively wealthier region due to widespread large scale cash crop production, for instance, and the respective RCB’s serve either cities, such as Kumasi, or very large towns (the remaining 3 cases). 18 In the two towns in the Volta Region with a meaningful number of insured clients, their overall number was still relatively limited so that we had to use a full sample of the policyholders here.

10

national consumption poverty line. Similar to the national average, around 30 percent of the households fall below the poverty line.19 Even though all survey locations are locally considered to be towns, around 50 percent of the households are engaged in farming with mean landholdings of 5.119 acres. However, non-farm activities are more widespread with 85 percent of the households being engaged in at least one such activity. On average, household heads have 8 years of formal education. Nevertheless, 26 percent of the heads have not completed primary education and 33 percent of the heads report to be illiterate or unable to read and write properly.

Microinsured households show significant differences in certain characteristics as compared to non-microinsured households. Note that all averages are weighted by population size and survey stratification so that averages of non-microinsured households are close to full-sample averages due to the low take-up rates of microinsurance. On average microinsured households feel less exposed towards risk, they have higher mean asset levels and they are much more engaged in non-farm activities (98 percent to 85 percent) than in farming (33 percent to 50 percent). More of their heads are married; they have a 23 percent higher share of own children and a three times lower share of old dependents in the household. They live in communities with a higher ratio of rural bank clients before the Anidaso policy was introduced and a lower ratio of susu clients. Although all survey locations are within the service areas of rural banks, this gives some indication of an access barrier in terms of formal financial services in certain locations, which will be elaborated in more detail in the later estimations. Microinsured households have used services of a rural bank for more than three times as long as the population as whole (in years), 10 percent and 15 percent more of them read the newspaper and listens to the news in the radio often, respectively, and their head is on average member in 0.22 more groups. All of these differences in means are significant at the one percent level.

The main estimations below will be on the determinants of the decision to purchase a micro life insurance, i.e. the Anidaso policy. However, one could argue that the respective categorization of households implicitly classifies households that did not buy the Anidaso policy as non-insured, or at least assumes these to behave differently to the predictions set out in the theoretical framework of (life) insurance consumption above. Therefore, additional robustness checks include the NHIS insurance and the few other (private) insurance policies available in the area. Table 1 reports the distribution of the different insurance categories considered. The Anidaso policy exhibits the lowest take-up rates with 2.09 percent of the households in the survey areas, while it is underlined that the NHIS is quite well received with half of the households being covered.

Table 1: Types of insurance used by households

Type of insurance Number of

households in the

sample

(total = 1031 )

Estimated number of

households in the

survey area

(total = 24310.5)

Estimated

proportion in the

survey area (%)

Anidaso policy 321 507.37 2.09 National health insurance (NHIS) 562 12602.00 51.84

Any insurance 738 14536.80 59.80

Private insurance 409 3349.98 13.78

Source: Author’s calculation based on survey data

19 The total national food and non-food consumption poverty line based on the most recent Ghana Living Standard Survey (GLSS) 2005/2006 is set at 370.80 Ghana Cedi per adult equivalent per year (Ghana Statistical Service 2007). Note that this is not an income-based poverty line so that figures for the sample are here likely to be underestimated, as they are based on total income per adult equivalent and not expenditures.

11

Variables are generally measured on the level of the household, with the exception of some, which refer to the household head as the major decision-maker in the household.

In terms of risk aversion, researchers have had some success in measuring risk aversion on the individual level, often by use of gambling experiments. It has been cautioned that this is not without difficulties, as gambling experiments may actually only predict future gambling behaviour, rather than revealing real-world risk-taking behaviour involving other risky economic decisions (Greene 1963; 1964; Rabin and Thaler 2001). However, there is some general agreement on the fact that individuals possess a basic set of attitudes towards risk, which are apparent in decisions towards different types of risky economic alternatives (e.g. Cox and Harrison 2008; Harrison and Rutström 2008). While the survey did not include a pure lottery type experiment to derive a measure of risk aversion, it generated data from a decision experiment that involves the chances of an additional payment (analogue to an insurance payout) in a hypothetical future scenario depending on the possibility to become ill or to remain healthy. Via a quadratic scoring rule, utility and personal beliefs (on the probability to become ill) may be jointly estimated. Under the assumptions of expected utility theory on risk preferences, it is then possible to determine degrees of relative risk aversion of individuals.

Especially in the context of a developing country, the success of experimental games depends largely on the level of understanding of the participants. Given that education levels are often much lower than in industrial countries, where such experiments have mostly been conducted, this may be one reason why Ito and Kono (2010) in their study on sow insurance, for instance, found shifts in the decisions of a typical lottery that were consistently at odds with their expectations leading to difficulties in the specification of actual parameters of risk aversion. Either due to similar such reasons or due to the fact that actual risk preferences do not strictly follow the assumptions of expected utility theory, the values of the constructed risk aversion measure in this study deviate highly from the expected patterns. Thus, at this stage of the analysis the risk aversion measure cannot confidently be included in the below estimations.20 Subsequent results thus have to be treated with caution due to potential omitted variable bias.

As a measure of the probability of the risk, or the size of the insured risk, the analysis includes a variable on the household’s health status, measured as the share of (severely) ill household

members in the last twelve months. Even though the insurance of interest is not a health but a life insurance, the current health status is assumed to be a proxy also of the probability of death.

Although the data at hand does not allow controlling explicitly for a discount rate, a larger number of young dependents in a household and a head in his or her prime age should result in a larger weight attached to bequests leading to an increased willingness to buy mortality coverage type of insurance. Similarly, marital status should point into the same direction, as this indicates responsibility for a remaining spouse. Thus, the analysis includes the household head’s age and age squared to identify respective life cycle effects. Additionally, a variable indicating the share of own kids in the household and a dummy variable indicating whether or not the head is married are considered as a proxy for the need to bequeath. On the opposite, the average age of all household members and its square term as well as the share of old

dependents in the household are included to indicate the relevance of an additional motive to save for retirement.

20 Most hopefully, this will be possible in future versions of this paper after a revision of the calculation procedure of the risk aversion measure. This will take into account the actual objective probability of risk and allow preferences to deviate from expected utility assumptions.

12

Table 2: Summary statistics

Variable

Microinsured households

mean

Non-microinsured households

mean

Level of significance of difference. if any

(percent) Full sample

mean Std. Err. Min. Max. Number of

observations Probability of risk occurence

Share of severely ill HH members (last 12 mths) 0.208 0.208 0.208 0.013 0 1 1031 Share of ill HH members (last 12 mths) 0.665 0.692 0.691 0.015 0 1 1031 Past experience of shocks No. of health shocks (last 5 years) 0.589 0.482 0.485 0.040 0 16 1031 No. of deaths (last 5 years) 0.221 0.284 10 0.282 0.025 0 6 1031 No. of economic shocks (last 5 years) 0.555 0.369 5 0.373 0.047 0 20 1031 Subjective probability of risk Risk assessment index -0.195 0.056 1 0.051 0.047 -1.752 3.652 1031 Wealth and income activities Lagged asset index 0.395 -0.089 1 -0.079 0.042 -1.657 3.451 1031 Landsize in acres 4.441 5.133 1 5.119 0.534 0 158.080 1031 HH engaged in farming 0.332 0.507 1 0.503 0.021 0 1 1031 HH engaged in non-farm activities 0.986 0.847 1 0.850 0.016 0 1 1031 Head is a farmera 0.118 0.285 1 0.281 0.019 0 1 1031 Head engaged in non-farm activitiesa 0.884 0.685 1 0.689 0.020 0 1 1031 Remittances per month (Ghana Cedi) 8.898 10.701 10.663 1.527 0 800 1028 Transfers per month (Ghana Cedi) 6.610 7.778 7.754 0.956 0 200 1031 Total income per adult equivalent per month (Ghana Cedi) 91.244 78.419 10 78.687 3.486 1.808 821.918 1031 Bequest motive Head is marrieda 0.643 0.532 1 0.535 0.021 0 1 1031 Share of own kids in HH 0.427 0.347 1 0.349 0.012 0.000 0.857 1031 Share of old dependents in HH 0.019 0.060 1 0.059 0.008 0.000 1.000 1031 Age of head 44.630 49.146 1 49.052 0.657 19 95 1031 Average age of all HH members 26.451 31.304 1 31.202 0.706 8.6 94 1031 Trust Ratio of RCB clients in community before Anidaso was introduced 0.544 0.425 1 0.427 0.007 0.024 0.867 1031 Ratio of susu clients in community 0.163 0.189 1 0.189 0.004 0.012 0.474 1031 Relationship to RCB before Anidaso was introduced in years 3.216 1.001 1 1.047 0.146 0 33 1031 Networks and information No. of groups head is member of 1.166 0.942 1 0.947 0.039 0 10 1011 Head reads newspapera 0.522 0.412 1 0.414 0.021 0 1 1011 Head listens to news in radio oftena 0.701 0.566 1 0.569 0.021 0 1 1011 Other variables Female headed HHa 0.359 0.455 1 0.453 0.021 0 1 1031 Years of schooling of head 10.090 8.126 1 8.167 0.223 0 26 1031 HH experienced that loan was denieda 0.203 0.094 1 0.097 0.012 0 1 1031

Unweighted number of observations 321 710 1031 Weighted number of observations 507.37 23803.2 24310.5

a Dummy variable where 1 = yes

Source: Author’ calculation based on survey data

13

Several variables are intended to reflect a household’s level of wealth. Two dummy variables indicate whether or not the household head is engaged in non-farm activities or a farmer, respectively. The analysis further takes into account the income received from remittances and from transfers per month as potential alternative risk coping strategies. All of these variables are further considered to reflect the variation of income sources as well as the reliability and steadiness of income streams. In order to control for potential endogeneity the lagged version of an asset index created by principal component analysis is used, which captures asset ownership five years ago.

As discussed above, other variables go beyond testing the predictions of the benchmark model. The past experience of shocks is reflected by three variables indicating the number of

health shocks, the number of deaths and the number of economic shocks a household experienced in the last five years. Instead of an explicit measure of risk aversion, for instance, the analysis considers the subjective assessment of risk by a risk perception index created through principal component analysis using polychoric correlations, which are able to adequately address the ordinal structure of the underlying variables (Kolenikov and Angeles 2008). These include information on the subjective exposure to illness, accidents and economic shocks, relative to other households in their community rated by the household on a scale from one (much less exposed) to five (much more exposed).21

The variables ratio of RCB clients in community before Anidaso was introduced and ratio of

susu clients in the community are proxies for the relative level of familiarity and popularity of formal financial services offered by RCBs and the prevalence of informal financial services, respectively. Both of these community-level variables are expected to relate to the level of trust into the providing institution of the Anidaso policy and to reflect underlying structures of access to formal financial services for communities, which are within the service areas of a rural bank, but lack any microinsurance clients. As will be explained in greater detail below, these variables are only included in specifications based on the full sample, while other specifications refer to a sub-sample of policyholder locations only. As a measure of trust on the household level, the variable relationship to RCB before Anidaso was introduced in years refers to the number of years a household has used services of an RCB. The role of social networks is examined by introducing the number of groups the head is a member of, including for instance social community groups, occupational groups, or self-help groups.

Aside from that, the analysis controls for gender, education and the degree of information of the household head, the experience of a loan denial, as well as potential local or regional effects (by community or region dummies). The former variables follow particularly from the review of the empirical microfinance and microinsurance literature. In terms of the role of gender in microfinancial markets it has been typically found that they have less access to (formal) credit (Khandker 1998; Kimuyu 1999; Diagne et al. 2000; Jabbar et al. 2002) and are thus, for instance, more likely to participate in savings schemes as an alternative to accumulate resources for investment or in order to cope with shocks (Kimuyu 1999). In addition, there is a common focus of microfinance programmes on women, which is based on the assumption that women are more likely to spend resources in a more responsible way, for instance on the health condition and educational attainment of their children (Duflo 2005). This would give reason to assume that females are likely to be more interested in purchasing insurance for their family to cope with shocks than men.

Interestingly, in terms of education there are several studies that do not find a significant effect on microinsurance uptake, which might be due to the fact that it is rather the level of specific knowledge on insurance or financial literacy in general that determines the decision to buy insurance (Wang and Rosenman 2007; Giné et al. 2008). As a very broad measure of the level of information, including information on financial matters, the analysis considers 21 The index is calculated using the polychoricpca command from STATA written by Stas Kolenikov.

14

two variables indicating whether or not the household head reads the newspaper or listens often to the news in the radio. While the analysis cannot control for credit constraints as in Giné et al. (2008) in a strict sense, the estimation specification includes a measure reflecting the experience that a loan from a financial institution was denied. Other variables, such as employment status, assets, and remittances or transfers, which are different measures of the wealth of a household, may serve as an indication for liquidity constraints as well.

5. Estimation Results

In order to investigate the correlates of households’ uptake of microinsurance a reduced-form probit model is estimated. Accordingly, the results can only be interpreted as conditional on the prevailing supply-side conditions described in detail above. The dependent variable takes on the value of 1 if the household purchased the Anidaso microinsurance and 0 otherwise.

Table 3 presents the estimation results. The first column shows results of a full sample estimation, which includes policyholder locations as well as non-policyholder locations of the RCB service areas. All further specifications shown in the subsequent columns report results from a reduced sample containing only the policyholder locations. This is due to the fact that there seems to be a clear selection into the microinsurance scheme based on access conditions regarding the services of the RCBs in general, as reflected by the highly significant effects of the ratio of RCB clients in the community before Anidaso was introduced and the ratio of susu clients in the community. Both variables have the strongest effect in the overall model, showing that for a household at the mean population take-up probability of 0.0209, with one more RCB-client per non-client in the community the probability of purchasing insurance increases by a factor of 4 (4.0 percentage points) and with one more susu client per non-client in the community the probability decreases by a factor of seven (-7.4 percentage points).22 This raises concerns of a systematic neglect of certain communities by the agents of rural banks, or the presence of other constraints, such as high transaction costs associated with geographical distance to the rural banks. Inhabitants of these communities may hence be prevented from consuming formal financial services at the same rate as the inhabitants of the policyholder communities in the sample and rather resort to informal ones, as shown by the susu variable. In order to ensure comparability of households in the sample, the non-policyholder locations are thus dropped in all further estimations.

The results seem to confirm the predictions of the benchmark model to a large extent. As set out in the first hypothesis it is expected that the participation in micro life insurance increases with the objective size of the risk. Indeed, the results show a significant positive association between the share of severely ill household members and the uptake of microinsurance. An increase of the share of severely ill household members by half a standard deviation (which translates into an approximate increase of severely ill members by 15 percent), i.e. the centred standard deviation change effect, raises the probability to purchase micro life insurance by about 1 percent (the marginal effect is 0.032). However, given that the health status of the household is not public information and that customers are not obliged to report this in the Anidaso policy admission application, this could be an indication for asymmetric information in the market at the same time, i.e. adverse selection.23

22 Scaling the marginal effects by the mean population insurance participation rate indicates the percentage change in the take-up probability for a one unit change of a respective explanatory variable for a household whose initial probability to purchase the Anidaso policy is at the population average. The respective values are not reported in Table 3. 23 Note that all references made to coefficients, marginal effects and significance levels refer to the estimation results of the second specification reported in the second column of Table 3 if not declared otherwise. Effects of a discrete change in specific values, such as a standard deviation change, are calculated additionally and are not reported in the table.

15

Table 3: Probit Estimation Results of Anidaso Microinsurance Participation

Microinsurance (1) Microinsurance (2) Microinsurance (3) Independent variables

Coeff. t-stat. Marg. effect Coeff. t-stat. Marg. effect Coeff. t-stat. Marg. effect

Probability of risk occurence

Share of severely ill HH members (last 12 mths) 0.387*** 2.650 0.012 0.483*** 3.024 0.032 0.381** 2.390 0.025 Past experience of shocks No. of health shocks (last 5 years) 0.044 0.964 0.001 0.031 0.620 0.002 0.047 0.962 0.003 No. of deaths (last 5 years) -0.053 -0.651 -0.002 -0.038 -0.407 -0.003 -0.012 -0.141 -0.001 No. of economic shocks (last 5 years) 0.085*** 2.698 0.003 0.109** 2.555 0.007 0.115*** 2.628 0.008 Subjective probability of risk Subjective risk perception index -0.105*** -2.836 -0.003 -0.106*** -2.625 -0.007 -0.084** -2.169 -0.006 Wealth and income activities Lagged asset index 0.096** 2.050 0.003 0.025 0.439 0.002 0.023 0.442 0.002 Landsize in acres (log-scale) -0.034 -1.387 -0.001 -0.020 -0.766 -0.001 -0.028 -1.096 -0.002 Head engaged in non-farm activitiesa 0.433*** 3.858 0.014 0.474*** 3.694 0.031 0.535*** 3.941 0.036 Remittances per month (Ghana Cedi) 0.001 0.688 0.000 0.002 1.632 0.000 0.001 0.724 0.000 Transfers per month (Ghana Cedi) 0.000 0.258 0.000 0.001 0.671 0.000 0.000 -0.002 0.000 Bequest motive Head is marrieda -0.030 -0.282 -0.001 -0.048 -0.391 -0.003 0.071 0.646 0.005 Share of own kids in household 0.507*** 2.892 0.016 0.514*** 2.636 0.034 Age of head 0.060*** 2.714 0.002 0.086*** 3.356 0.006 Age of head squared -0.001*** -2.944 0.000 -0.001*** -3.730 0.000 Trust Ratio of RCB clients in community before Anidaso introduced 1.318*** 6.978 0.042 Ratio of susu clients in community -2.799*** -4.072 -0.088 Years HH used RCB services before Anidaso introduced 0.064*** 4.781 0.004 0.055*** 4.485 0.004 Networks and information No. of groups head is member of 0.078* 1.837 0.002 0.103** 2.223 0.007 0.121*** 2.599 0.008 Head reads newspapera -0.221** -2.016 -0.007 -0.265** -2.199 -0.017 -0.290** -2.447 -0.019 Head listens to news in radio oftena 0.092 0.903 0.003 0.042 0.375 0.003 0.058 0.529 0.004 Other variables Female headed HHa -0.194* -1.714 -0.006 -0.311** -2.365 -0.021 -0.228* -1.854 -0.015 Years of schooling of head 0.009 0.816 0.000 -0.001 -0.084 0.000 0.000 -0.042 0.000 HH experienced that loan was denieda 0.339*** 2.795 0.011 0.315** 2.106 0.021 0.302** 2.161 0.020 Region 1a 0.185* 1.770 0.006 Region 2a -0.092 -0.757 -0.003 Share of old dependents in HH -0.167 -0.395 -0.011 Average age of HH members 0.026 1.606 0.002 Average age of HH members squared -0.001** -2.354 0.000 Community controls Yes Yes

Constant -4.189*** -7.639 -4.271*** -6.800 -2.719*** -7.305

Observations 1028 686 672

Notes: a Dummy variable where 1 = yes. Households in the sample are weighted according to their sampling probabilities. The asterisks indicate level of significance: *** significant at 1 percent, ** significant at 5 percent, * significant at 10 percent. Source: Author’s calculation based on survey data.

16

While the theoretical prediction on the relationship between a household’s level of wealth and the uptake of insurance is somewhat ambiguous, the empirical results point into a positive direction. However, the asset index is only significant in the full-sample estimation and looses significance when controlling for the relationship with the rural bank and community effects, which, again, underlines differences in terms of welfare between policyholder locations and non-policyholder locations in the sample. Nonetheless, whether or not the head is engaged in non-farm activities significantly increases the probability of purchasing micro life insurance by 3.2 percentage points. For a household with an initial take-up probability at the population mean, having a head who is engaged in non-farm activities thus triples the probability to take-up microinsurance. This result is underlined when exchanging the non-farm activity variable by a dummy variable indicating whether or not the head is a farmer, which shows a negative significant coefficient (result not shown). Together with the negative, but insignificant effect of the size of land used by the household, this suggests that activities in the non-farm sector facilitate greater ability to afford the regular monthly premium payments for the Anidaso policy through more steady and reliable incomes. Pointing into the same direction, the main self-declared reason for households not to purchase any insurance is that it is considered too expensive (see Table A2, Appendix). Surprisingly, as an alternative coping strategy and source of cash, remittances or transfers by family members or friends neither seem to increase the willingness to purchase micro life insurance, nor do they seem to function as substitutes for it.

As for the bequest motive hypothesized above, estimates show that the marital status of the head does not seem to influence the decision to buy micro life insurance. All other variables intended to measure the motive to bequest show significant coefficients of the expected sign. Increasing the share of own kids by half a standard deviation (translating to an approximately 14 percent higher share of own kids), results in a 1 percent higher probability to purchase micro life insurance (the marginal effect is 0.034). Apparently, there is a life cycle effect in terms of the age of the household head. The coefficient of the head’s age shows a positive significant sign. More precisely, a one-year increase of the head’s age results in a 30 percent higher likelihood to buy the Anidaso policy (0.6 percentage points). However, this effect only holds up to a turning point, which is at 44.2 years of age. The relatively low level of the turning point suggests that the bequest motive outweighs any potential motive to save for retirement. The results here match well with the self-reported reasons of households to buy the Anidaso policy shown in Table A2 in the Appendix. While the majority of households (57.58 percent) reports the very broad reason “to secure against future shocks”, the second most common reason is “to protect family in case of illness/death” (23.80 percent). In addition, this finding might also indicate that with rising age there is a rising experience with financial matters and an increase in family responsibility, yet, only up to a certain point. At the same time, after this point, i.e. for the older generation, the cost of evaluating and accepting new products and technical procedures may also become higher.

The predominant motivation to bequest is underlined by the results of the third model specification shown in the third column of Table 3. This specification includes the average age of all household members and its squared term as well as the share of old dependents in the household. Both variables are expected to elicit an additional saving-for-retirement motive in the decision to buy micro life insurance, as with rising age there is higher need for retirement income and the bequest motive is assumed to diminish. The share of old dependents shows a negative, but insignificant coefficient. The age variables, on the other hand, indicate the same relationship as in the specification with the household head’s age, with a significant negative coefficient of the squared term. With 25.5 years of age, the turning point of the effect is even lower than that of the head’s age, thereby strongly highlighting the bequest motive in favour of the saving-for-retirement motive. However, it is hardly possible

17

to exclude the option that the insurance might still be seen as a savings instrument, though not as a pension, as some households have evidently chosen the additional savings component of the policy. Accordingly, 8.79 percent of the households report that they have bought the Anidaso policy for investment reasons (Table A2, Appendix). Instead of saving for retirement, this might rather be motivated by the option of a partial withdrawal of the insured amount mentioned earlier.

Beyond the predictions of the benchmark, it is hypothesized that the past experience of shocks and the subjective risk perception by the households are important factors in the micro life insurance take-up decision. Contrary to the expectation, the number of health or death shocks a household has experienced during the past five years does not seem relevant for the choice to buy a policy. Interestingly, however, there is a positive effect of the number of economic shocks (i.e. dramatic increase/decrease of input/output prices, inability to sell products, or loss of job), a household experienced in last 5 years. Other than health and death shocks, economic shocks are not directly linked to the examined type of insurance coverage. While there seems to be no straightforward explanation for this finding, a tentative interpretation could be that economic shocks are more frequently experienced and thus create higher wariness to build a buffer against times of hardship.

In contrast to a priori expectations, households which consider themselves to be more exposed to risk than others in their neighbourhood are less likely to purchase the Anidaso policy. As found in the earlier studies on micro rainfall insurance (Giné et al. 2008; Cole et al. 2009) households which are risk averse (although this is not directly measured by the subjective exposure towards risk) might also be averse to uncertainty associated with the policy. In fact, due to a limited understanding of the insurance, the policy itself might be considered as a risk and thus not be perceived as helpful in dealing with the consequences of death or illness.24

Also the estimates on the role of trust and networks strongly confirm the results of the studies on micro insurance geared towards crop failure, which found that participation in insurance increases strongly with household’s familiarity with the vendor and membership in village networks, or, with the endorsement from a trusted third party about the insurance policy (Giné et al. 2008; Cai et al. 2009; Cole et al. 2009). The relationship to the rural bank exhibits a highly significant positive coefficient. Specifically, with one more year a household has used services from the rural bank before the Anidaso policy was introduced, the probability to participate in the insurance increases by about 20 percent in view of the initial population take-up probability. The variable measuring the number of groups the household head is a member of is positive and statistically significant at the 5 percent level. One more group the head is a member of raises the probability to take up the Anidaso policy by 33 percent. It has been cautioned by Giné et al. (2008), however, that the strong explanatory power of social networks might in part reflect an omitted variable bias in terms of the intensity of marketing, which in the case of their study is strongly linked to the advertisement of the product via reaching out to community opinion leaders and existing customers. As outlined above, in the context of the Anidaso policy GLICO takes on a very similar marketing approach, including the specific targeting of groups. This potential bias has to be born in mind in the above interpretation. Either way, the evidence presented here is at odds with the full-information benchmark model, where neither the intensity of marketing, nor the intensity of networks should have an effect on households’ decision to purchase insurance.

24 Although it is not possible to identify the direct causation due to the cross-sectional nature of the data, at least a reverse causation seems unlikely here. In principle, households which do not have access to insurance might feel more exposed to risk. Yet, the provision of micro life insurance is such a young phenomenon that it is unlikely that households feel the absence of this option as such a strong threat.

18

As to the expectations, a higher level of education should enhance the ability to understand a new financial product and to evaluate the terms and conditions associated with it. Yet, results resemble those of the earlier literature, which finds no significant relationship between education and the uptake of microinsurance (Wang and Rosenman 2007; Giné et al. 2008). The evidence deviates, however, from that of others in terms of the degree of financial knowledge. Whether or not the head reads the newspaper shows a negative coefficient, which is significant at the 1% level. Switching on this variable reduces the probability to purchase micro life insurance by 86 percent (1.8 percentage points). On the one hand, this could be caused by measurement error, as it seems likely that this variable is much too broad to measure the degree of financial information – let alone financial literacy. On the other hand, the variable might also reflect that reading the newspapers does not only create a higher level of information and awareness on financial matters (among other things). It could also (negatively) bias the personal impression of financial institutions, including insurance companies, depending on the focus of the respective reports on respective issues.

Opposite to the a priori expectation on the role of gender, female headed households are significantly less likely to take up insurance. A change from the minimum to the maximum (i.e. 0 to 1) in this variable reduces the probability of insurance participation by a factor of two (-2.2 percentage points). All other things being equal, this gives some indication that females are more excluded than men from participation in the formal financial sector in general. One reason for this might be that operating with formal financial services is viewed rather as a men’s domain, while women resort to other (informal) strategies to deal with risk (Bortei-Doku and Aryeetey 1995; Ankomah 1996).

Whether or not the household has ever experienced that a formal loan was denied is positively related to the uptake of micro life insurance. As indicated before, this variable cannot reflect credit constraints as such. However, it shows that those households which have tried accessing a loan were denied and may thus try to resort to other options to prepare against risk. In addition, from discussions with GLICO’s sales agents, the rural bank’s staff, and focus group discussions conducted by the author in the Central Region there is some indication that clients view the Anidaso policy as one option to gain reputation at the bank as an alternative for collateral, which will later allow them to access a loan. Furthermore, some sales agents have obviously used this as an argument to convince people to buy a policy. Correspondingly, at least 5.13 percent of the households self-reportedly bought the policy “to obtain collateral for a loan” (Table A2, Appendix).

When comparing the results of the microinsurance probit model with those of the purchase of other insurance categories, including a) the NHIS, b) any insurance and c) private insurance there are some outstanding differences, but also similarities in the estimates (Table A3, Appendix).25 Not surprisingly, the number of health shocks experienced in the last five years is positive significant in the NHIS equation. At the same time, the number of highly severe household members becomes insignificant, as these two variables are highly collinear. However, when ignoring the severity of illness, a variable reflecting the mere number of ill household members in the last twelve months seems more robust and shows the expected sign and significance level (see the third column in Table A3). This effect is present throughout all insurance categories and can be a sign of adverse selection, at least in the case of the NHIS, which is heavily subsidized and faced by strong financial challenges. Interestingly, for all of the other insurance categories the subjective risk perception index shows the expected

25 Note that for these other insurance categories estimations are based on the full sample, excluding few observations due to missing values in some of the variables. An exception is the private insurance probit, where observations from one of the communities, Mankrom Nkwanta in the Central Region, are dropped as membership in this community predicts failure completely.

19

positive significant coefficient, as opposed to the microinsurance equation. This gives reason to assume, that the NHIS as well as other types of private insurance are better understood and perceived as adequate mechanisms to deal with the associated risk as compared to the Anidaso policy.

Another deviating finding is that the level of wealth contributes positively to membership in the NHIS as well as the purchase of private insurance in general to quite an extend. Half a standard deviation change in the asset index increases the likelihood to have national health insurance by 12.0 percentage points. Pointing into the same direction, the amount of transfers received per month is positively and significantly associated with membership in the NHIS. In contrast to the Anidaso policy, the relationship between remittances and the purchase of any private insurance is positive, indicating that remittances could indeed act as substitutes for insurance. Surprisingly, whether or not the head has a spouse reveals a positive and significant coefficient for the NHIS, while the coefficient of the share of own kids is negative and significant. While the first effect could be explained by the fact that it is generally possible to insure family members, i.e. the spouse and children, in the NHIS under the same policy, the second effect does not give much support to this explanation. Again different to the estimation of the microinsurance purchase, the age variables show the opposite sign. Households with a younger head are more likely to participate in the NHIS, but apparently older heads as well, shown by a significant positive coefficient of the square term (the turning point is at age 48). Not surprisingly, the relationship to a formal bank26 is not significantly related to the uptake of national health insurance. Though, similarly to the Anidaso policy, there is a positive relationship of this variable with the uptake of any private insurance on a low significance level. While the variables measuring the integration into networks and information are altogether insignificant for the other insurance categories, whether or not the household has ever experienced that a loan was denied is significant and negatively associated to membership in the NHIS. One may thus have to consider that this variable in fact captures to a large extend the access to formal loans in the first place. In other words, the category of households which have experienced a denial of a formal loan have at least tried to access one, while in the opposite category those which have never tried outweigh the group of households which received a loan, the effect thereby showing a more general exclusion from access to formal financial services and also formal social security options.

7. Conclusion

Following the enormous success of microcredit throughout the last decades, microinsurance has expanded rapidly throughout the developing world. Despite high expectations on the role of microinsurance for the improvement of household risk management, however, take-up rates have remained low and there is still limited empirical evidence on the impediments to trade of different types of microinsurance. Different from the products examined in previous studies, this paper investigates the correlates of households’ decision to take up a micro life insurance. The analysis is based on household survey data collected in semi-urban locations within three service areas of rural banks in Southern Ghana.

The evidence confirms several predictions of the benchmark model of life insurance consumption. Specifically, there is a strong indication of a motivation to bequest, which outweighs the potential saving-for-retirement motive. This underlines the recognition of the long-term and short-term consequences of the death of a major breadwinner with regard to 26 The relationship to the rural bank is here exchanged by a variable indicating the relationship to any bank, as all of the respective insurances are not specifically linked to rural banks. The argumentation to include this variable here rather is that it reflects a relationship to the formal financial sector in general, hence, a higher exposure to formal financial services of any type.

20

permanent losses in total household income and the immediate need for funds to cover funeral costs. It also shows that the Anidaso policy indeed fills a gap in the formal risk management options available to the low income population, which has hitherto been dominated by credit and savings instruments, or, the NHIS. Yet, as a number of households have apparently chosen the optional investment component of the insurance package, it is still likely that some households intend to save via the Anidaso policy, for instance by making use of the option of a partial withdrawal. Another reason for households to buy a micro life insurance is denoted by a positive relationship between the experience of a loan denial and the decision to purchase a policy. As indicated by the self-reported reasons to buy the Anidaso policy and as suggested by the results of focus group discussions and information from sales agents, a number of clients might view the Anidaso policy as one option to gain reputation at the bank as an alternative for collateral. However, the results are here to be treated with caution due to the suboptimal variable included to measure a loan denial, which may in fact rather capture the access to formal loans in the first place.

The issue of limited outreach of financial services, including micro life insurance, is also underlined by the fact that there seems to be a systematic neglect of locations in the service area of rural banks associated with geographical distance and respective transaction costs. Apparently, although mobile banking via the susu system is very widespread in Ghana, this has not been exploited to a great extend by the sales agents of the insurance.

Furthermore, up to now there seems to be limited outreach of the insurance to the poorest and most vulnerable households which are assumed to have the least access to other options to manage with risk. For instance, households engaged in the non-farm sector associated with more steady and reliable incomes are more likely to participate in micro life insurance. However, the positive relationship of the level of wealth with the uptake of insurance is even more pronounced in the case of the NHIS or other (non-micro) insurances. Moreover, different to the a priori expectation that females would be more likely to purchase micro life insurance due to a lower access to formal loans and a higher sense of responsibility towards their family, female headship significantly reduces the probability to purchase a policy. This signifies that in the Ghanaian context operating with formal financial services is typically viewed as a men’s domain, while women resort to other (informal) strategies to deal with risk. To further investigate the role of gender in the access to and use of (micro) insurance seems to be an interesting topic for future research.

Going beyond the predictions of the neoclassical benchmark model, the analysis highlights some other important explanatory factors in the decision to participate in micro life insurance. There is indicative evidence of adverse selection, as households with a higher share of (severely) ill household members in the last twelve months are more likely to take up micro life insurance. In line with earlier studies on households’ microinsurance participation behaviour the results also underline the important role of trust into and familiarity with the providing institution and its financial services as well as the integration into networks. Take-up rates are higher for households which have used financial services offered by a rural bank for a longer period of time and for those which are integrated into a higher number of groups. Interestingly, the level of information measured by the consumption of news via the radio or the newspapers is negatively associated with the uptake of micro life insurance. This might capture the fact that the image spread via the media of financial institutions in general, and insurance companies in particular, may be negatively biased.

While the data do not allow controlling explicitly for the role of adverse selection, a subjective risk perception index included in the analysis shows that households which feel more exposed to risk compared to other households in their community are surprisingly less likely to participate. As similarly found by Giné et al. (2008) in the case of micro rainfall insurance, households seem to consider the Anidaso policy to be a risky choice itself, because

21

they do not fully understand the insurance and all its terms and conditions. This is underlined by the finding that the subjective risk perception index has the expected positive effect on the participation in other types of insurance, such as the NHIS and other private insurances. Taking these factors together, once again, this suggests that more proactive and intensive education on insurance and information on the specific product are crucial for accessing the market in its full potential.

Interestingly, while remittances do not seem to compete with the choice of micro life insurance, there is a negative effect on the use of private insurance in general. Transfers from family members or friends, on the other hand, only exhibit a positive relationship with the membership in the NHIS. The role of remittances and transfers as alternative risk management options to insurance thus remains ambiguous and leaves scope for further investigations.

22

Bibliography

Abbring, Jaap H., Pierre-André Chiappori, James J. Heckman and Jean Pinquet (2003). "Adverse Selection and Moral Hazard in Insurance: Can Dynamic Data Help to Distinguish?" Journal of the European Economic Association, Papers and Proceedings of

the Seventeenth Annual Congress of the European Economic Association (Apr. - May), 1(2/3): 512-521

Ankomah, Augustine (1996). "Premarital relationships and livelihoods in Ghana." Gender &

Development 4(3): 39 - 47.

Arhin, Kwame (1994). "The Economic Implications of Transformations in Akan Funeral Rites." Africa: Journal of the International African Institue, 64(3): 307-321.

Arrow, Kenneth J. (1971). Insurance, Risk and Resource Allocation. Essays in the Theory of

Risk Bearing. Chicago, Markham Publishing Company. Chapter 3: 90-133.

Babbel, David F. and Eisaku Ohtsuka (1989). "Aspects of Optimal Multiperiod Life Insurance." The Journal of Risk and Insurance, 56(3): 460-481.

Basu, Anupam, Rodolphe Blavy and Murat Yulek (2004). "Microfinance in Africa: Experience and lessons from selected African countries." IMF Working Paper, 04/174. International Monetary Fund, Washington D.C.

Beck, Thorsten and Ian Webb (2002). "Economic, Demographic, and Institutional Determinants of Life Insurance Consumption across Countries." World Bank Economic

Review, 17(1): 51-88.

Boon, Emmanuel K. (2007). "Knowledge Systems and Social Security in Africa: A Case Study on Ghana." Studies of Tribes and Tribals, Special Volume No. 1, Indigenous Knowledge Systems and Sustainable Development: Relevance for Africa: 63-76

Bortei-Doku, Ellen and Ernest Aryeetey (1995). Mobilizing Cash for Business: Women in Rotating Susu Clubs in Ghana. Money-Go-Rounds. The Importance of Rotating Savings

and Credit Associations for Women. Shirley Ardener and Sandra Burman. Oxford, Herndon, Berg Publishers: 77-94.

Browne, Mark J. and Kihong Kim (1993). "An International Analysis of Life Insurance Demand." Journal of Risk and Insurance, 60(4): 616-634.

Cai, Hongbin, Yuyu Chen, Hanming Fang and Li-An Zhou (2009). "Microinsurance, Trust and Economic Development: Evidence from a Randomized Natural Field Experiment." PIER Working Paper. Penn Institute for Economic Research, Philadelphia.

Chankova, S., S. Sulzbach and F. Diop (2008). "Impact of mutual health organizations: evidence from West Africa." Health policy and Planning, 23(4): 264-276.

Chiappori, Pierre-André (2000). Econometric Models of Insurance Under Asymmetric Information. Handbook of Insurance. Georges Dionne. Boston, Dordrecht, London, Kluwer Academic Publishers.

Churchill, Craig (2002). "Trying to Understand the Demand for Microinsurance." Journal of

International Development, 14: 381-387.

Cohen, Monique, Michael J. McChord and Jennefer Sebstad (2005). "Reducing Vulnerability: Demand for and Supply of Microinsurance in East Africa." Journal of International

Development, 17(3): 319-325.

Cohen, Monique and Jennefer Sebstad (2003). "Reducing Vulnerability: The Demand For Microinsurance." MicroSave Africa, Nairobi.

Cole, Shawn, Xavier Giné, Jeremy Tobacman, Petia Topalova, Robert M. Townsend and James Vickery (2009). "Barriers to Household Risk Management: Evidence from India."

23

Federal Research Bank of New York Staff Reports, 373. Federal Research Bank of New York, New York.

Collins, Daryl, Jonathan Murdoch, Stuart Rutherford and Orlanda Ruthven (2009). Portfolio's

of the poor: how the world's poor live on two dollars a day. New Jersey, Woodstock, Oxfordshire, Princeton University Press.

Cox, James C. and Glenn W. Harrison (2008). Risk Aversion in Experiments: An Introduction

West Yorkshire [u.a.], Emerald Group Publishing Limited

Debreu, Gerard (1959). Theory of Value. New York, John Wiley & Sons, Inc.

Dercon, Stefan (2002). "Income Risk, Coping Strategies and Safety Nets." The World Bank

Research Observer, 17(2): 141-166.

---, Ed. (2005). Insurance Against Poverty. A study prepared by the World Institute for Development Economics Research of the United Nations University (UNU-WIDER). Oxford, New York, Oxford University Press.

Dercon, Stefan, Tessa Bold and Cesar Calvo (2008a). Insurance for the Poor? Social

Protection for the Poor and Poorest. Risk, Needs and Rights. Armando Barrientos and David Hulme. Houndmills, Basingstoke, Hampshire, Palgrave Macmillan.

Dercon, Stefan, Martina Kirchberger, Jan Willem Gunning and Jean-Phillippe Platteau (2008b). "Literature Review on Microinsurance." Microinsurance Paper 1. International Labour Organization, Geneva.

Diagne, Aliou, Manfred Zeller and Manohar Sharma (2000). "Empirical Measurement of Households' Access to Credit and Credit Constraints in Developing Countries: Methodological Issues and Evidence." FCND Discussion Paper 90. International Food Policy Research Institute, Washington, D.C.

Dionne, Georges, Neil Doherty and Nathalie Fombaron (2000). Adverse Selection in Insurance Markets. Handbook of Insurance. Georges Dionne. Boston, Dordrecht, London, Kluwer Academic Publishers.

Dror, David Mark, Ruth Koren and David Mark Steinberg (2006). "The Impact of Filipino Micro Health Insurance Units on Income-Related Equality of Access to Healthcare." Health Policy, 77(3): 304-317.

Duflo, Esther (2005). "Gender Equality in Development." BREAD Policy Paper, 11. Bureau for Research in Economic Analysis of Development.

Geest, Sjaak van der (2006). "Between Death and Funeral: Mortuaries and the Exploitation of Liminality in Kwahu, Ghana." Africa: Journal of the International African Institute 76(4): 485-501.

Ghana Statistical Service (2007). "Patterns and Trends of Poverty in Ghana 1991 - 2006." Ghana Statistical Service, Accra.

Giesbert, Lena, Susan Steiner and Mirko Bendig (2010). Participation in Micro Life Insurance and the Use of Other Financial Services in Ghana, Mimeo.

Giné, Xavier, Robert M. Townsend and James Vickery (2008). "Patterns of Rainfall Insurance Participation in Rural India." The World Bank Economic Review, 22(3): 539-566.

Greene, Mark R. (1963). "Attitudes toward Risk and a Theory of Insurance Consumption." The Journal of Insurance, 30(2): 165-182.

--- (1964). ""Insurance Mindedness". Implications for Insurance Theory." The Journal of Risk

and Insurance, 31(1): 27-38.

Guiso, Luigi, Paola Sapienza and Luigi Zingales (2007). "Trusting the Stock Market." Journal of Finance, forthcoming.

Hakansson, N. H. (1969). "Optimal Investment and Consumption Strategies Under Risk, An Uncertain Lifetime, and Insurance." International Economic Review, 10: 443-466.

24

Harrison, Glenn W. and E. Elisabet Rutström (2008). Risk Aversion in the Laboratory. West Yorkshire [u.a.], Emerald Group Publishing Limited.

Hong, Harrison, Stanley Kubik and Jeremy Stein (2004). "Social Interaction and Stock Market Participation." Journal of Finance, 59(1): 137-163.

ILO (2001). "Social security: A new consensus." International Labour Organisation, Genf.

Ito, Seiro and Hisaki Kono (2010). "Why is the take-up of microinsurance so low? Evidence from a health insurance scheme in India " The Developing Economies, 48(1): 74-101.

Jabbar, M.A., S.K. Ehui and R. von Kaufmann (2002). "Supply and Demand for Livestock Credit in Sub-Saharan Africa: Lessons for Designing New Credit Schemes." World

Development, 30(6): 1029-1042.

Kahnemann, Daniel and Amos Tversky (1981). "The Framing of Decisions and the Psychology of Choice." Science, 211: 453-458.

Kahnemann, Daniel and Amos Tversky (1979). "Prospect Theory: An Analysis of Decision under Risk." Econometrica, 47(263-291.).

Karni, Edi (2006). "Subjective expected utility theory without states of the world." Journal of

Mathematical Economics, 42(3): 325-342.

Khandker, Shahidur R. (1998). Fighting Poverty with Microcredit: Experience in Bangladesh. Washington, D.C., World Bank.

Kimuyu, Peter Kiko (1999). "Rotating Saving and Credit Associations in Rural East Africa." World Development 27(7): 1299-1308.

Kolenikov, S. and G. Angeles. (2008). "Socioeconomic status measurement with discrete proxy variables: Is principal component analysis a reliable answer?" Retrieved 05. August, 2010, from http://web.missouri.edu/~kolenikovs/papers/polychoric-technical-4.pdf.

Lewis, Frank D. (1989). "Dependents and the Demand for Life Insurance." The American

Economic Review, 79(3): 452-467.

Mossin, J. (1968). "Aspects of Rational Insurance Purchasing." Journal of Political Economy, 79: 553-568.

Murdoch, Jonathan (1995). "Income Smoothing and Consumption Smoothing." Journal of

Economic Perspectives, 9(3): 103-114.

Muzzucato, Valentina , Mirjam Kabki and Lothar Smith (2006). "Transnational Migration and Economy of Funerals: Changing Practices in Ghana." Development and Change, 37(5): 1047-1072.

Nakata, Hiroyuki and Yasuyuki Sawada (2007). "Demand for non-life insurance: a cross-country analysis." CIRJE-Discussion Papers, F-461,1-5. Centre for International Research on the Japanese Economy, University of Tokyo, Tokyo.

Nyman, John A. (2001). "The demand for insurance: expected utility theory from gain perspective." University of Minnesota.

Oetzel, Jennifer M. and Sudeshna Ghosh Banerjee (2003). Liberalization of Insurance Services in Emerging Markets: Do Open Markets Improve Local Insurance Service? Strategic Management Society’s (SMS) Annual Conference. Baltimore.

Outreville, J. Francois (1996). "Life Insurance Markets in Developing Countries." Journal of

Risk and Insurance, 63(2): 263-278.

Pissaridis, C.A. (1980). "The Wealth - Age Relation with Life Insurance." Economica, 47: 451-457.

Platteau, Jean-Phillippe (1997). "Mutual Insurance as an Elusive Concept in Traditional Rural Communities." Journal of Development Studies, 33(6): 764-796.

25

Pratt, John W. (1964). "Risk Aversion in the Small and in the Large." Econometrica, 32(1/2): 122-136.

Preker, A.S., G. Carrin, D.M. Dror, M. KJakab, W. Hsiao and D. Arhin (2002). "Effectiveness of Community Health Financing in Meeting the Cost of Illness." Bulletin of the World

Health Organization, 80(2): 143-150.

Rabin, Matthew and Richard H. Thaler (2001). "Anomalies: Risk Aversion." The Journal of

Economic Perspectives, 15(1): 219-232.

Rosenzweig, Mark R. and Hans P. Binswanger (1993). "Wealth, Weather Risk and the Composition and Profitability of Agricultural Investments." The Economic Journal, 103(416): 56-78.

Roth, Jim, Michael J. McChord and Dominic Liber (2007). "The Landscape of Microinsurance in the World's 100 Poorest Countries." The Microinsurance Centre, LLC, Appleton, WI.

Rothschild, Michael and Joseph E Stiglitz (1976). "Equilibrium in Competitive Insurance Markets: An Essay on the Economics of Imperfect Information." The Quarterly Journal of

Economics, 90(4): 630-49.

Sakurai, Takeshi and Thomas Reardon (1997). "Potential Demand for Drought Insurance in Burkina Faso and Its Determinants." American Journal of Agricultural Economics, 79(4): 1193-1207.

Siegel, Paul B., Jeffrey Alwang and Sudharshan Canagarajah (2001). "Viewing Microinsurance as a Social Risk Management Instrument." Social Protection Discussion

Paper, 0116. Worldbank, Washington.

Steel, William and David Andah (2008). Rural & Microfinance Regulation in Ghana: Implications for Development of the Industry. The Economy of Ghana. Ernest Aryeetey and Ravi Kanbur. Accra [u.a.], Woeli Publishing Services: 195-221.

Tesliuc, Emil D. and Kathy Lindert (2004). "Risk and Vulnerability in Guatemala: A Quantitative and Qualitative Assessment." Social Protection Discussion Paper Series, 0404. The World Bank Social Protection Unit Washington D.C.

Townsend, Robert M. (1994). "Risk and Insurance in Village India." Econometrica, 62(3): 539-591.

--- (1995). "Consumption insurence: An evaluation of risk-bearing systems in low- income economies." Journal of Economic Perspectives 9(3): 83-102.

von Neumann, J. and O. Morgenstern (1944). Theory of Games and Economic Behaviour. Princeton, NJ, Princeton University Press.

Wagstaff, A., M. Lindelow, G. Jun, X. Ling and O. Juncheng (2007). "Extending Health Insurance to the Rural Population. An Impact Evaluation of China's New Cooperative Medical Scheme." World Bank Policy Research Working Paper, 4150. The World Bank, Washington D.C.

Wang, Holly H. and Robert Rosenman (2007). "Preceived Need and Actual Demand for Health Insurance Among Rural Chinese Residents." China Economic Review, 18(4): 373-388.

Winter, Ralph (2000). Optimal Insurance Under Moral Hazard. Handbook of Insurance. Georges Dionne. Boston, Dordrecht, London, Kluwer Academic Publishers.

Yaari, Menahem E. (1965). "Uncertain Lifetime, Life Insurance, and the Theory of the Consumer." Review of Economic Studies, 32(2): 137-150.

26

Appendix

Table A1: Insured and Non-Insured Survey Households across Survey Sites

Number of non-insured

households in sample

Number of insured

households in sample

Estimated share of

insured households in

population (weighted, in

percent)

Central Region Policyholder locations Nsaba 41 24 0.17 Duakwa 32 25 0.18 Kwanyako 44 62 0.35 Non-policyholder locations Mensakrom 27 0 0 Mankrom Nkwanta 27 0 0 Asafo 61 0 0

Eastern Region

Policyholder locations Mamfe 28 28 0.25 Mampong 43 44 0.36 Larteh 44 36 0.29 Non-policyholder locations

Tingkong/Nyame Bekyere 44 0 0

New Mangoase 43 0 0 Asenema 28 0 0

Volta Region

Policyholder locations Sogakope 99 72 0.34 Dabala 34 30 0.15 Non-policyholder locations Adutor 59 0 0 Hikpo 32 0 0 Kpotame 23 0 0

Total 709 321 2.09

Source: Authors’ illustration based on household survey data.

Note: Due to oversampling of insured households, the share of insured households in the respective total community population is much smaller than the relative number of these households in the sample suggests.

27

Table A2: Self-Declared Reason to Buy or Not to Buy Insurance

Reason Number of

households

Estimated households in

population (weighted. in

percent)

Reason to buy the Anidaso policy

To secure against future shocks 180 57.58

To protect family in case of illness/death 77 23.80

For investment reasons 28 8.76

To obtain collateral for loans 17 5.13

Old age security 5 1.47

To finance medical care 4 1.24

Other 3 1.05

To finance funeral costs 2 0.58

Education 1 0.39

Total 317 100.00

Reason not to buy any insurance

Too expensive 145 49.62

Not important to me 40 13.29

No information on insurance facilities 29 12.95

Don' t trust insurer 26 6.64

No knowledge on insurance 16 6.05

Not enough time/can't be bothered 12 4.63

Did not think about it 10 2.80

Insurance provider too far away 8 2.31

Other 2 0.75

Procedures too difficult 2 0.41

Not eligible 1 0.29

Insurance not considered effective 2 0.26

Total 293 100.00

Source: Authors’ illustration based on household survey data.

28

Table A3: Probit Estimation Results of Participation in Different Types of Insurance

Microinsurance NHIS (1) NHIS (2) All insurance Private insurance

Independent variables Coeff.

t-

stat.

Marg.

effect Coeff.

t-

stat.

Marg.

effect Coeff.

t-

stat.

Marg.

effect Coeff.

t-

stat.

Marg.

effect Coeff.

t-

stat.

Marg.

effect

Probability of risk occurence Share of severely ill HH members (last 12 mths) 0.483*** 3.024 0.032 0.176 0.842 0.069 Share of ill HH members (last 12 mths) 0.505*** 2.997 0.198 0.530*** 3.068 0.187 -0.024 -0.105 -0.004 Past experience of shocks No. of health shocks (last 5 years) 0.031 0.620 0.002 0.105* 1.681 0.041 0.106* 1.691 0.041 0.109* 1.761 0.038 -0.011 -0.156 -0.002 No. of deaths (last 5 years) -0.038 -0.407 -0.003 0.121 1.162 0.048 0.112 1.088 0.044 0.109 1.016 0.038 -0.020 -0.166 -0.003 No. of economic shocks (last 5 years) 0.109** 2.555 0.007 0.042 0.700 0.016 0.022 0.368 0.009 -0.006 -0.099 -0.002 0.024 0.379 0.004 Subjective probability of risk Subjective risk perception index -0.106*** -2.625 -0.007 0.141*** 2.619 0.056 0.121** 2.221 0.048 0.124** 2.113 0.044 0.158** 2.252 0.025 Wealth and income activities Lagged asset index 0.025 0.439 0.002 0.302*** 3.625 0.118 0.305*** 3.658 0.120 0.370*** 4.251 0.130 0.329*** 3.392 0.053 Landsize in acres (log-scale) -0.020 -0.766 -0.001 -0.025 -0.698 -0.010 -0.017 -0.479 -0.007 -0.045 -1.204 -0.016 -0.038 -0.844 -0.006 Head engaged in non-farm activitiesa 0.474*** 3.694 0.031 0.042 0.279 0.016 0.045 0.305 0.018 0.171 1.145 0.060 0.660*** 2.953 0.106 Remittances per month (Ghana Cedi) 0.002 1.632 0.000 0.001 0.760 0.000 0.001 0.708 0.000 0.000 -0.087 0.000 -0.007*** -2.980 -0.001 Transfers per month (Ghana Cedi) 0.001 0.671 0.000 0.007*** 2.688 0.003 0.008*** 2.655 0.003 0.006* 1.938 0.002 0.002 0.587 0.000 Bequest motive Head is marrieda -0.048 -0.391 -0.003 0.410*** 2.612 0.161 0.428*** 2.715 0.168 0.302* 1.890 0.106 0.625*** 3.316 0.100 Share of own kids in household 0.514*** 2.636 0.034 -0.565** -2.326 -0.222 -0.526** -2.189 -0.206 -0.364 -1.451 -0.128 -0.737** -2.575 -0.118 Age of head 0.086*** 3.356 0.006 -0.049** -2.056 -0.019 -0.049** -2.066 -0.019 -0.064*** -2.607 -0.023 -0.026 -0.831 -0.004 Age of head squared -0.001*** -3.730 0.000 0.001** 2.145 0.000 0.001** 2.142 0.000 0.001*** 2.903 0.000 0.000 0.619 0.000 Trust Years HH used services of RCB before Anidaso was introduced 0.064*** 4.781 0.004 Years HH used services of any bank before Anidaso was introduced 0.019 1.479 0.007 0.021 1.626 0.008 0.036* 1.887 0.013 0.020 1.466 0.003 Networks and information No. of groups head is member of 0.103** 2.223 0.007 0.013 0.179 0.005 -0.010 -0.145 -0.004 -0.119 -1.618 -0.042 0.197*** 2.686 0.032 Head reads newspapera -0.265** -2.199 -0.017 -0.042 -0.263 -0.016 -0.091 -0.563 -0.036 0.131 0.776 0.046 0.252 1.423 0.040 Head listens to news in radio oftena 0.042 0.375 0.003 -0.041 -0.282 -0.016 -0.065 -0.446 -0.025 0.022 0.154 0.008 0.171 1.029 0.027 Other variables Female headed HHa -0.311** -2.365 -0.021 -0.056 -0.330 -0.022 -0.049 -0.290 -0.019 -0.145 -0.830 -0.051 -0.046 -0.241 -0.007 Years of schooling of head -0.001 -0.084 0.000 0.013 0.868 0.005 0.015 1.015 0.006 0.025 1.571 0.009 0.040** 2.142 0.006 HH experienced that loan was denieda 0.315** 2.106 0.021 -0.525*** -2.620 -0.206 -0.556*** -2.704 -0.218 -0.299 -1.464 -0.105 0.204 1.015 0.033 Community controls Yes Yes Yes Yes Yes

Constant -4.271*** -6.800 0.809 1.252 0.548 0.845 0.613 0.926 0.613 0.926

Observations 672 1008 1008 1008 983

Notes: a Dummy variable where 1 = yes. Households in the sample are weighted according to their sampling probabilities. The asterisks indicate level of significance: *** significant at 1 percent, ** significant at 5 percent, * significant at 10 percent.

Source: Author’s calculation based on survey data