Informal risk sharing arrangements and the crowding out of formal insurance markets: A cross-sectional analysis from Vietnam by Fiona Wainwright and Carol Newman* Abstract Using cross-sectional household survey data from Vietnam, this paper examines the extent to which the use of informal insurance mechanisms “crowds out” the take-up of formal insurance products. We analyse formal insurance demand within rural Vietnam in order to determine the likely barriers to household participation and include measures of reliance on informal information networks and social organisations to determine the extent to which the formal insurance market is crowded out by the informal market. A significant negative impact of informal insurance information sources on formal insurance participation and demand decisions provides evidence of a “crowding out” effect. In contrast, seeking information and assistance from formal social organisations has a positive association with insurance demand suggesting that these channels may have an important role to play in matching households to formal insurance markets. Overall, results indicate that while there is significant underexploited growth potential in the formal Vietnamese insurance market, the presence of the informal market should also be taken into consideration when formulating an appropriate policy response. Key Words: formal insurance demand, crowding out, informal risk sharing, Vietnam JEL Codes: O12, D53, D85, * Carol Newman (cnewman@tcd.ie) and Fiona Wainwright (wainwrif@tcd.ie) are at the Department of Economics, Trinity College Dublin. Our special thanks go to Prof. Finn Tarp at the Department of Economics and the Development Economics Research Group, University of Copenhagen for permitting the use of the data. 1 1. Introduction Recent literature suggests that the lack of insurance, both in terms of market availability and actual take-up, is one of the key drivers of persistent levels of poverty in developing countries (Morduch, 2002). Across the developing world the majority of low income households are engaged in farming-type activities, and are therefore potentially subject to material losses as a consequence of fluctuations in agriculturally derived income from external shocks such as drought, floods, pest infestation and livestock disease.1 The purchase of a comprehensive insurance package, however, is a complex process (Showers and Shotick, 1994). Households evaluate their financial needs and determine various insurance schemes to insure against adverse income shocks. According to modern financial theory, a household should hold a diversified market portfolio that minimizes non-systematic risk. In practice, however, (and especially in developing economies) many idiosyncratic and aggregate level risks are not insured and households often remain exposed to movements in local weather, regional house prices, prices of commodities like rice, gasoline and heating oil and national income fluctuations. In the absence of well developed financial markets in developing countries, households tend to engage in costly informal risk mitigating strategies to reduce income fluctuations.2 These income-smoothing activities can succeed in reducing the volatility of income, but at a very high cost. As such, in many developing countries community based informal risk-sharing and risk-coping arrangements are common whereby fluctuations in household income or idiosyncratic shocks do not have any effect on the consumption of individual households (Ray, 1998). The role of informal risk-sharing arrangements and their interaction with formal insurance markets is an important area within the development economics literature. For example, Townsend (1994) using data from three high-risk villages in India, finds that household consumption co-moves with village average consumption and that once village level risk is controlled for, household consumption is not influenced by contemporaneous own income, sickness, unemployment, or other idiosyncratic shocks. Overall, the evidence suggests that informal risk-sharing arrangements are important. However there is also much evidence to suggest that informal risk-sharing is typically partial and inefficient and as a result expected utility is likely to increase under formal insurance arrangements (Cox and Jimenez, 1992; Grimard, 1997; Ligon 2001). Furthermore, shocks such as weather events tend to affect all households in a local geographic area making community-based risk-sharing mechanisms, such as interhousehold transfers and local credit and asset markets, less effective at easing the impact of an income shock. In addition, Morduch (2001) highlights the fact that informal insurance is typically limited by the inability to write binding, enforceable long term contracts (Morduch, 2001). Instead, the arrangements stay together only for as long as the expected value of staying true to the arrangement exceeds the value of defecting and facing the risk alone (i.e. self-insuring). The arrangements weaken as the self-insurance option improves. Townsend and Lim (1998) support this finding 1 For example Morduch and Sicular (2001) found that while average incomes in villages in the Shandong Province in northern China were growing at 8 per cent per year, one quarter of the population in any given year was suffering losses of about 20 per cent. 2 For example, Morduch (1995) finds that Indian farmers near subsistence level spatially diversify their plots, and devote a larger share of land to low-yield traditional varieties of rice and castor. 2 that the most effective approach to risk-coping at the household level is not via informally based risk-sharing but by self-insurance through in-kind saving (for example, building up grain reserves and drawing them down as required). Rosenzweig (1988) provides evidence that formal risk-sharing opportunities are crowded out by informal networks. Using data from Karnataka State in India, the study highlights the importance of private transfers within networks of families and friends, and shows that such transfers frequently crowd out formal loans. On the other hand, research also suggests that the crowding out of formal networks by the informal networks might not be the case. Indeed, evidence from Taiwan demonstrates that informal and formal networks co-exist (Levenson and Beasley, 1996). These authors highlight the persistence of informal financial mechanisms despite the development of financial markets, possibly for the reason that they may offer more cost-effective finance. The ongoing debate regarding the persistence of informal risk-coping within rural communities in developing economies presents an opportunity to examine how the informal sector impacts upon the formal insurance market within the Vietnamese context. This paper contributes to the literature by focussing specifically on the role of formal insurance products in insuring against risk in rural Vietnam and identifying causal relationships between informal risk sharing arrangements and the demand for formal insurance. We use cross-sectional data from the Vietnamese Access to Resources Household Survey 2006. We find that informal risk-sharing negatively impacts on formal insurance demand while reliance on formal social organisations has a positive effect.3 This paper is organised as follows. Section 2 details the methodological framework used for the analysis while Section 3 describes the data. Section 4 discusses the empirical findings and Section 5 concludes. 2. Methodological Framework Ehrlich and Becker (1972) demonstrate that an individual’s need for insurance is no different to their need for any other good or service. Our insurance demand model is developed within a simple neo-classical framework under symmetric information. It is assumed that the household chooses whether to purchase insurance (and the amount to purchase) based upon expected utility values. Any factor that increases the expected utility of purchasing formal insurance is therefore expected to increase the probability of purchase. An individual purchases an insurance contract so as to alter his/her pattern of income across states of nature (Rothschild, Stiglitz, 1976). We begin by considering a riskaverse household with the following standard constant absolute risk aversion (CARA) utility function: U i (C ) = − 3 1 α exp{− αC} (1) This latter result supports the findings of Newman et al. (2009) who find evidence of a role for community governance, in the form of formal socio-political organisations, in correcting for both infrastructural and information failures in financial markets at the commune level. 3 Where U (.) is increasing and strictly concave ( U ' (.) → 0 as C → 0 ), and where α ≥ 0 measures risk aversion and C denotes household consumption. We assume increasing relative risk aversion (IRRA) in that the coefficient of relative risk aversion, R A = αC is increasing in consumption goods as measured by C : dR A =α dC (2) Households can be exposed to many different states, s = 1,2,..., S , and the household assigns a probability, π s , to each different state. Ys denotes the household level of income associated with each state. I denotes the quantity of insurance demanded per household while p (I ) denotes the price per unit coverage. p (I ) * I thus represents the total insurance premium paid. L denotes the amount to be insured or the potential loss incurred. Abstracting the analysis to two states of nature ( s = 1 representing a state where a loss occurs and s = 2 representing a state where a loss does not occur) so π 1 + π 2 = 1 , in any given time period the household is faced with the following budget constraint if s = 1 : Y1 = Y + I − L − p (I ) * I (3) and if s = 2 : Y2 = Y − p (I ) * I (4) To keep the model simple we assume that there is no savings or borrowing and so C1 = Y1 and C 2 = Y2 . Placing the budget constraints directly into the generalized form of the utility function, the household seeks to maximise the following von-Neumann Morgenstern Expected Utility function: EU i (I ) = π 1U i (C1 ) + π 2U i (C 2 ) = π 1U i (Y + I − L − p(I ) * I ) + π 2U i (Y − p(I ) * I ) (5) The insurance demand function is therefore: I i * (π , p, L, Y , α ) = Arg max EU i ( I ) (6) In a developing country context, the observed level of insurance purchased, I 0 , may deviate from this optimum. This may be due to, for example, information failures or the existence of informal risk-sharing arrangements that crowd out the demand for formal insurance products. As such, the amount of insurance that each household demands will not only depend on the outcome of this optimization problem, it may also be influenced by any informal risk-sharing arrangements (measured by ε ) that the household is a part of within the local commune. I 0 ( π , p , L, Y , ε , α ) = I * ( π , p , L, Y , α ) − ε 4 (7) Equations (6) and (7) form the basis for our empirical model. In the absence of price information, the only observable factors in the insurance demand model are income and the extent of informal risk sharing. We include, however, a number of variables to control for the level of risk aversion (which will determine L and α ). The empirical model for household insurance demand is: I i = β 0 + β1Yi + β 2 N i + β 4 Zi + vi (8) where I i is measured as the insurance premium paid, Yi is income, N i is a measure of informal risk-sharing arrangements that may crowd out the demand for insurance and Z i is a vector of control variables including socio-economic characteristics of households and regional indicators to control for supply-side variations. An important empirical consideration is that many households report a zero willingness to buy formal insurance. The reasons could be that households do not positively value this type of insurance or because they lack the financial resources to pay any positive amount for it. Furthermore, it may be the case in rural areas that households are not even aware that formal insurance products exist. Failure to take this into account will result in biased and inconsistent estimates. We attempt to control for this potential bias by modeling the demand for insurance using a censored tobit model that conditions on the fact that households are aware of the existence of formal insurance products. In addition, as a robustness check, we separately model the insurance participation decision within a standard probit framework. Denoting the decision to purchase insurance with a zero-one dummy variable d i and the vector of explanatory variables determining this decision as Wi , the participation decision can be modeled in a standard probit framework where d i * is an unobserved underlying utility associated with the purchase of insurance: d i * = δWi + u i d i = 1 if d i * > 0 d i = 0 otherwise (9) Denoting the observed level of insurance purchased as insi and the vector of explanatory variables determining this decision as X i , the expenditure decision can be modeled in a standard tobit framework where insi * is an unobserved underlying utility associated with the level of insurance purchased: insi * = λX i + ei insi = insi * if ins i * > 0 ins i = 0 otherwise 5 (10) Assuming that the error terms are distributed as u i ~ N (0,1) for the probit and ( ) ei ~ N 0, σ 2 for the tobit model, each can be estimated using Maximum Likelihood Estimation. The consistency of Maximum Likelihood Estimator for the censored tobit model depends critically on the assumption of homoscedastic errors. To allow for the possibility of heteroscedasticity we consider a multiplicative heteroscedasticity structure where the error variance is assumed to be related to a sub-set of exogenous variables, gathered in a vector H i (not including a constant) (Verbeek, 2004).4 V {ei | X i } = σ 2 exp{ωH i } (11) The tobit model is also heavily parametric in character, with error terms assumed normally distributed. Such strong assumptions may be costly when the data do not fit this distribution, resulting in inconsistent maximum likelihood estimators. A conditional moments test against the null hypothesis of normal errors is conducted on the demand equation. This test was derived by Pagan and Vella (1989). In the event of non-normality the dependent variable is transformed by an Inverse Hyperbolic Sine (IHS) transformation algorithm. This transformation accommodates zero, negative, and positive values for the random variable and is known to better handle extreme values than other transformations such as Box-Cox. The IHS transformation on the insurance demand variable insi is: T (θinsi ) = sinh −1 (insi ) θ [ = log θinsi + (θ 2 insi2 + 1) 12 ]θ (12) The final tobit demand log likelihood equation that corrects for heteroscedasticity and adjusts for non-normality with (IHS) transformation is given by: ⎡ ⎛ βX log L = ∑0 ⎢1 − Φ⎜⎜ i ⎝ σi ⎣ ⎞⎤ ⎟⎟⎥ ⎠⎦ ⎡ −1 2 ⎛ T (θins i ) − βX i + ∑1 ⎢ θ 2 ins i2 + 1 φ ⎜⎜ σi ⎝ ⎣ ( ) ⎞⎤ ⎟⎟⎥ ⎠⎦ (13) where Φ (.) and φ (.) refer to the standard normal probability and density functions, respectively. A limitation of the probit participation and tobit demand models within the context of this study is that they do not explicitly control for whether or not the household is aware of the existence of insurance products. To accommodate this, all estimates are conditional on awareness. 4 Likelihood ratio tests are performed to determine the subset of continuous explanatory variables that are causing heteroscedasticity and appropriate corrections are made to the Log Likelihood function. 6 3. Data The data are taken from the Vietnam Access to Resources Household Survey (VARHS). This survey was carried out in rural areas of 12 provinces of Vietnam between August and September 2006 thereby producing cross-sectional data on more than 2,300 households. The survey was developed by the Development Economics Research Group, Department of Economics, University of Copenhagen and the Institute of Labour Studies and Social Affairs, Hanoi Vietnam. The VARHS explores issues surrounding Vietnamese rural households’ access to resources and the constraints that these households face in managing their livelihoods. The households are spread over 161 districts and 466 communes. Along with detailed demographic information on household members, the survey includes sections on household savings, credit (both formal and informal), formal insurance, risk response, informal safety nets and the structure of social capital. In this paper we are interested in using the empirical model given in equation (8) to test, in particular, the extent to which informal risk sharing arrangements crowd out the demand of formal insurance products. The data indicate that 65 per cent of households surveyed hold formal insurance. Among the 11 types of formal insurance listed in the survey, health insurance schemes (including health insurance for children) and vehicle insurance schemes have the highest participation rates. In contrast, there are no households in our sample that have agricultural insurance. 5 Table 1 disaggregates the total insurance purchased by households into its constituent parts. The survey also reveals the reasons why households do not demand each type of formal insurance. Seven per cent of households are not aware of the existence of formal insurance products, 58 per cent state that they have no need for insurance while 18 per cent of households have no information regarding insurance products. The data also reveal that of the 39 per cent of households that experienced an income shock between 2002 and 2005 only 2.9 per cent of these losses were covered by formal insurance. Eighteen per cent of households claim that they rely on informal mechanisms such as borrowing from a friend or relative. While these observations from the raw data do not provide evidence of crowding out they do suggest that despite the fact that 65 per cent of households hold formal insurance, informal risk coping measures remain important. INSERT TABLE 1 HERE To measure the extent of informal risk coping among households in our sample we follow a similar approach to Jowett (2004). We consider two proxy measures of social capital: one that captures informal arrangements and the other formal networks. The informal proxy measures the importance to households of informal networks (Friends/Neighbours/Family) as sources of information regarding credit and insurance. The measure is binary, recording 1 if informal sources of information are important and 0 if informal sources of information are not important. The formal proxy measure of social capital is a binary measure reflecting whether the household expects to receive future help from formal groups/organisations, recording 1 if the household expects to receive future help and 0 if not. These groups exist at a more 5 This supports the informal statistic that only 1 per cent of farming areas in Vietnam hold insurance (see Vietnamnet (www.vnn.vn), “Agricultural Insurance: Where is the State?”, 16/07/2004). 7 structured/organised level and consist of the Communist Party, Women’s Union, Farmers Union and Veteran’s Union. We explicitly exclude more informal groups (Neighbourhood Committee, Sports Club) from this classification. The inclusion of these measures of informal risk sharing into the model could potentially introduce endogeneity yielding inconsistent estimates. It could be argued, for example, that the extent to which households rely on informal risk sharing networks will depend upon unobserved household characteristics such as risk aversion which will be included in the error term of the formal insurance demand model ( ei in equation (10)). As the theoretical model (equation (7)) predicts, formal insurance demand is increasing in risk aversion. We propose by extension that reliance on informal risk sharing networks is also increasing in risk aversion. If this is the case then the crowding out effect of informal risk sharing may be understated since the effect of this variable will be dampened by the fact that it also captures the positive effect of unobservable risk aversion on the demand for formal insurance. In contrast, if complementarities exist between informal risk sharing and formal insurance demand, the correlation with unobserved risk aversion will lead to this effect being overstated. We address the endogeneity issue by using an instrumental variables approach. We consider two instruments for the informal proxy. First, we use a measure of the distance from the commune to the nearest state bank on the basis that the further a commune is located from a formal financial institution the more likely households are to rely on informal networks for information and risk sharing. Second, we construct an ordinal measure of informal risk coping intensity using data on sources of credit accessed between 2002 and 2005. Households borrow from a range of different sources including formal banks (such as, the Social Policy bank, or the Bank for Agriculture and Rural Development) and informal sources (including relative and friends, informal credit schemes and rotating savings and credit schemes). Households borrowing from relatives and friends are given a high score while those borrowing from formal sources score lowly. This measure thus reflects the strength of informal financial networks at the household level. We argue that this instrument is not contemporaneously correlated with unobserved risk aversion on the basis that it measures only from which source, not how much a household borrows. This reflects an income smoothing decision that a rational household will make (regardless of their risk aversion) without incorporating the magnitude of any borrowing undertaken (which could be correlated with risk aversion). We also consider two instruments for the formal measure of social capital. First, we use the distance from the household to the local people’s committee office on the basis that the shorter the distance the more active formal groups are likely to be in the commune and the more likely households are to expect help from these organisations in the future. Second, we use the number of formal organisations that the household is an active member. The more organisations that they are ‘active’ members of the more likely they are to expect to receive help. We argue that membership of such formal social organisations represents a measure of how the state still maintains an important function in the mobilization of resources with respect to rural organisational life within Vietnam and that this is not connected to a household’s level of risk aversion. 8 Aside from the crowding out hypothesis our model also allows us to test a number of other interesting predictions. The IRRA utility function, which underpins our model, predicts that the amount of insurance demanded should be increasing in the level of wealth. In the model we allow wealth to equate to consumption, which in the absence of savings and borrowing could be proxied by income. However, since we also treat insurance like any other good, standard income effects might also be present. If insurance is a normal good, then we would expect a positive relationship between income and the amount of insurance demanded (i.e. β1 > 0 ). We proxy income in the model with total expenditure incurred by the household less the amount spent on insurance. Having controlled for income, our model predicts that household wealth should also have a positive effect on insurance demand. We measure wealth as a wealth quintile indicator using per capita household consumption. Much of the empirical literature investigating insurance demand finds that household characteristics should be included as important controls of heterogeneity in the degree of risk aversion across households (Magrabi et al., 1991). Showers and Shotick (1994) hypothesise that the specific characteristics related to household demand for insurance include family size, age, number employed and income. They also consider how the number of household earners interacts with income to influence insurance demand. They propose that multi-earner households perceive less risk of loss of income than single-income households, even at the same income level. In contrast, families with many members may in fact be more risk averse as they are more exposed to shocks. Lazear and Micheal (1988) suggest that there may also be economies of scale in insurance purchase and so as a family size increases we would expect demand for insurance to also increase, but at a decreasing rate. We use the age of the head of household as a proxy for the overall lifecycle stage of a family unit. As families age, the level of income and number of dependents tend to rise. This should result in an increase in risk aversion and so an increase in household demand for insurance. However, in the declining years of a family, we would expect the demand for insurance to decrease as the family reaches the end of their lifecycle with less insurable risk. This relationship may be curvilinear (Duker, 1960), a consideration also made in our model. The full list of variables considered for this analysis is given in Table 2. INSERT TABLE 2 HERE 4. Empirical Results A censored tobit demand model for formal insurance, conditional on awareness, with adjustments for heterosedasticity and non-normality (IHS transformation) is estimated using STATA.6 All continuous variables enter in log form to smooth extreme value volatilities. The baseline model includes income, wealth plus other interaction and control variables to control for the unobserved measure of risk aversion and initially exclude the measures of informal risk sharing. We estimate a baseline tobit insurance demand model and a standard probit decision model. The latter captures the extent to 6 The Stata code to perform this estimation is non-standard and was written using maximum likelihood estimation procedures. Tests for heteroscedasticity and non-normality are performed with homoscedastic and normal errors both rejected at the 1 per cent significance level. 9 which causal relationships are effective at the margin rather than in the level of demand thus serving as a robustness check on the model. The results are presented in Table 3. INSERT TABLE 3 HERE The theoretical model predicts that the extent of household risk aversion is increasing in wealth. When explicitly controlling for risk aversion, household wealth should have a positive effect on insurance demand assuming insurance is a normal good. This is reflected in the positive coefficients on the upper wealth quintiles of the tobit demand estimates. The associated probit model also returns a significant and positive relationship between two upper wealth quintiles and the probability that a household participates in the formal insurance market. We also hypothesise that a household is more risk averse in savings therefore financial savings should also positively impact on insurance demand. We find that the level of household financial savings does have a positive and significant effect on the demand for formal insurance and on the probability of purchase.7 We also find that, as hypothesised, the greater the size of the family, the greater the demand for formal insurance as there tend to be more insurable risks in larger families. This result is also reflected in the participation equation The level of education of the head of household has a positive and significant effect on the demand for formal insurance. This could be due to the fact that educated households are more aware of risks or else simply that they have more information about formal insurance products. Farming households have a negative and significant impact on the amount of insurance demanded. This is surprising given that farmers are more exposed to weather related risks. Finally, a prior adverse income shock has an insignificant effect on both the insurance demand and the probability of purchase decisions. We note that while gender has a negative and significant impact on insurance demand, it is not significant in the probit participation decision. This inconsistency may reflect the fact that certain factors while important to a household’s insurance market participation decision, can become less significant when deciding on the amount of insurance to purchase. Focussing specifically on the insurance demand equation we note that while wealth records positive and significant effects on demand, income records an insignificant effect contrary to the theoretical model predictions. An insignificant relationship between the number of earners and insurance demanded is also found, thereby rejecting Showers and Shotick (1994) hypothesis. Family size has a positive and significant impact on insurance demand as proposed by Lazear and Micheal (1988) and economies of scale in insurance purchase are also evident from the negative coefficient against the square term. The age and age squared coefficients are both insignificant thereby rejecting the Duker (1969) proposition that age can be used as a proxy for the lifecycle stage of a family unit. These inconsistencies might be partially explained by the social and economic differences between a developed economy such as the US (the focus of Duker’s research) and a developing economy such as Vietnam where insurance markets are less well developed and families tend to be more cohesive and supportive of older members. 7 It could also be argued that savings are a form of self-insurance and so could crowd out the demand for formal insurance. This does not seem to be the case. 10 Next, we extend our analysis to consider the extent to which informal risk sharing arrangements crowd out the demand for formal insurance. We present the results for the un-instrumented and the instrumented cases in Table 4. Only the informal risk sharing measures are presented for ease of illustration. The results for the other variables remain robust to their inclusion. INSERT TABLE 4 HERE The two measures of informal risk sharing included are: an informal proxy measuring the importance to households of informal networks as sources of information regarding credit and insurance; and a formal proxy measure of social capital measuring whether the household expects to receive future help from formal groups/organisations. In the un-instrumented case we find that informal information sources negatively and significantly impact on the demand for insurance (and on the participation decision). In contrast, the formal measure has a positive and significant impact on both decisions. To correct for possible bias as a result of the endogeneity of these variables we instrument using four different instruments as discussed above.8 We run first stage regressions on each of the potentially endogenous regressors, the results of which are presented in the appendix, and find that all are significant at least at the 5 per cent level. We then estimate the probit and tobit models using the fitted values from this first stage regression in the second stage model. 9 Results are presented in Table 5. INSERT TABLE 5 HERE Results confirm that the instrument for social capital measuring whether the household expects to receive future assistance from formal groups/organisations has a positive effect on both the probability of purchase and on the insurance amount demanded. This result suggests that social groups may fill an important information gap in the insurance market by directing households toward formal insurance products in times of need. In contrast, we confirm that the reliance on informal information networks has a significant negative effect on both the participation in formal insurance markets and on the amount of formal insurance demanded. This suggests that informal information sources ‘crowd out’ formal sources. Overall, these results suggest two conflicting effects of informal networks and social capital on insurance demand in rural Vietnam. On the one hand, as found in much of the other literature on developing countries, there is some evidence to suggest that informal risk-sharing arrangements crowd out the demand for formal insurance. On the other hand, membership of formal social groups linked to the state may play an important role in linking households to formal insurance markets. 8 We test for endogeneity using the ivreg GMM approach. Both informal risk sharing regressors were found to be endogenous with GMM C statistics of 6.68843 and 9.71733 respectively. Applying our instruments results in a test of overidentifying restrictions that fails to reject the null of exogenous instruments with a Hansen’s J statistic of 0.277245. All other coefficients are robust to these tests. 9 We also implement as an additional robustness check the Stata module TOBITIV by Jonah B. Gelbach that implements the method of Whitney Newey, 'Efficient Estimation of Limited Dependent Variable Models with Endogenous Explanatory Variables', Journal of Econometrics (1987). The standard errors require adjustment but the point estimates will be consistent. All coefficients remain robust to this estimation. 11 While we have sought to eliminate the endogeneity present in the informal risk measures through the use of suitable instrumental variables, we will continue to seek additional instruments to verify the robustness of our existing estimates. In addition, we propose to extend the analysis to include marginal effects in order to evaluate the magnitude of the effects of our measures of informal risk-sharing on formal insurance demand - both when instrumented and when not. 6. Conclusion Formal insurance markets are underdeveloped in developing countries. In Vietnam, there is reason to believe that the rural economy may benefit greatly from sustainable insurance products. Targeting specific market segments with information about the advantages of purchasing formal insurance and the specific types of insurance products available could potentially stimulate formal insurance demand and thus provide rural communities with a guaranteed safety net in the face of adverse income shocks. As the composition of rural Vietnam evolves, changes in household circumstances and characteristics will influence the demand for formal insurance. In this study, the determinants of household formal insurance demand are explored together with the extent to which informal information channels and formal social organisations crowd out the formal insurance market. Using data from the Vietnamese Access to resources household survey 2006, a probit purchase decision and a tobit model of formal insurance demand is estimated conditional upon awareness. We find that wealthier households with greater financial savings, higher net incomes and better education, demand more formal insurance. Farming households are less likely to purchase. We find conflicting evidence on the role of social networks in insurance demand decisions. A significant negative impact of informal insurance information sources on the decision to purchase and the amount of insurance demanded provides evidence that informal arrangements ‘crowd out’ formal insurance demand. In contrast, associations with formal social organisations have a positive effect on insurance demand suggesting that they may have an important role to play in matching households to formal insurance markets. A major consideration of this study is the endogeneity of the measures of informal risk sharing used in this model. We attempt to control for risk aversion of the household to as great an extent as possible in the model and select suitable instruments to correct for this endogeneity bias. Future work will attempt to find additional instruments in order to enrich our existing measures of social capital. In extracting policy implications from this research, a deeper understanding of why households elect to purchase different insurance types can provide useful insights concerning future needs for both the private and public sector alike. Overall, our results suggest that targeting rural farming households through local formal groups could be an effective way to increase demand for formal insurance products. In particular, social organisations may have the potential to act as a positive conduit for disseminating product information. It should also be noted, however, that if encouraging insurance purchase is desirable from a policy perspective, information quality is important in seeking the desired market response. 12 References Vince E. Showers and Joyce A. Shotick (1994) “The Effects of Household Characteristics on Demand for Insurance: A Tobit Analysis”, The Journal of Risk and Insurance, Vol 61, No.3, 492-502. Matthew Jowett (2003) “Do informal risk sharing networks crowd out public voluntary health insurance? Evidence from Vietnam”; Applied Economics, 35, 11531161. Carol Newman, Finn Tarp and Katleen Van den Broeck (2008) “Network effects and household savings: Evidence from Vietnam”. Carol Newman, Maeve Henchion and Alan Matthews (2003) “A Double-Hurdle Model of Irish Household Expenditure on Prepared Meals”. Peter G. Moffatt (2003) “Hurdle Models of Loan Default”; School of Economic and Social Studies; University of East Anglia, Norwich, UK. Andrew M Jones (1989) “A Double-Hurdle Model of Cigarette Consumption”, Journal of Applied Econometrics, Vol. 4 No.1, pp 23-39. Michelle A. Danis and Anthony Pennington-Cross (2005) “A Dynamic Look at Subprime Loan Performance”, Federal Reserve Bank of St. Louis, Research Division, 411 Locust St, St Louis MO, 63102. Jonathan Morduch (2002) “Micro-insurance: the next revolution”; Forthcoming in ‘What we have learned about poverty? Edited by Abhigit Banerjee, Roland Benabou and Dilip Mookherjee; Oxford University Press. Steven T. Yen and Andrew M. Jones (1997) “Household Consumption of Cheese: An Inverse Hyperbolic Sine Double-Hurdle Model with Dependent Errors”, American Journal of Agricultural Economics, Vol 79, No1, pp 246-251. Pengfei Li (2005) “Box-Cox Transformations: An Overview”, Department of Statistics, University of Connecticut. Roberto Martinez-Espineira (2004) “A Box Cox double hurdle model of wildlife valuation: the citizens perspective”, Department of Economics, St Francis Xavier University. Debraj Ray “Development Economics”, Princeton University Press John F. McDonald and Robert A. Moffitt (1980) “The Uses of Tobit Analysis”, The Review of Economics and Statistics, Vol 62, No 2, pp 318-321. Takeshi Amemiya (1973) “Regression Analysis when the Dependent Variable is Truncated Normal”,Econometrica, Vol 41, No 6, pp 997-1016. 13 John B. Burbidge, Lonnie Magee, A. Leslie Robb (1988) “Alternative Transformations to Handle Extreme Values of the Dependent Variable”, Journal of the American Statistical Association, Vol 83, No. 401, pp 123-127. David Mayers and Clifford W. Smith Jr, (1983) “The Interdependence of Individual Portfolio Decisions and the Demand for Insurance”, The Journal of Political Economy, Vol 91, No2, pp 304-311. Robert M. Townsend (1994) “Risk and Insurance in Village India”; Econometrica, Vol 62, No 3, pp 539-591. Ali Asgary, Ken Willis, Ali Akbar Taghvaei, Mojtaba Rafeian (2004) “Estimating Rural Households Willingness to pay for Health Insurance”; The European Journal of Health Economics, Vol 5, No 3, pp 209-215. Christopher Udry (1994) “Risk and Insurance in a Rural Credit Market: An empirical investigation in Northern Nigeria”; The Review of Economic Studies, Vol 61, No 3 pp 495-526. David F. Babbel (1985) “The Price Elasticity of Demand for Whole Life Insurance”, The Journal of Finance, Vol 40 No1, pp 225-239. Eric P. Briys and Henri Louberge (1985) “On the Theory of Rational Insurance Purchasing: A Note”; The Journal of Finance, Vol 40 No. 2, pp 577-581. Jan Mossin (1968) “Aspects of Rational Insurance Purchasing”, The Journal of Political Economy, Vol. 76, No 4, Part 1, pp. 553-568. Neil A. Doherty (1984) “Portfolio Efficient Insurance Buying Strategies”, Journal of Risk and Insurance, 51: 205-224. Jacob M. Duker (1969) “Expenditures for Life Insurance Among Working-Wife Families”, Journal of Risk and Insurance, 36: 525-533. I Ehrlich and Gary S. Becker (1972) “Market Insurance, Self-Insurance and SelfProtection”,1972, Journal of Political Economy, 80: 623-648. Milton Friedman (1957) “The Theory of the Consumption Function”, Princeton University Press, Princeton, N.J. J.D Hammond, D.B Houston and E. R. Melander (1967) “Determinants of Household Life Insurance Premium Expenditures: An Empirical Investigation”, Journal of Risk and Insurance, 34: 397-408. Lazear, E. P. And R.T. Micheal (1988) “Allocation of Income within the Household,” Chicago University Press. Margabi, F.M, Y.S. Chung, S.S. Cha and S. Yang (1991) “The Economics of Household Consumption”, New York: Praeger. 14 Marno Verbeek (2004) “Modern Econometrics”, 2nd Edition (2004). 15 Tables Table 1: Explanatory Variables and descriptions Explanatory Variable Description Formal Social Organisations Help Measure Informal Information Sharing Measure Gender Debt Net Income Wealth Financial Savings Age Age Squared Family Size Family Size Squared Number of Earners Number of Children Children Supporting (1 Yes, 0 No) Married (1 Yes, 0 No) Education Farmer (1 Yes, 0 No) Prior Income Shock (from 2002 to 2005) Net Income * Number of Earners Provincial Dummies Whether household expects to receive future help from formal social organisations (Communist Party, Farmers Union, Veterans Union). The importance to households of informal sources of information on insurance and credit (1 important, 0 not important). Gender of head of household (1 Male, 0 Female). Household financial debt amount. Household net income (less insurance premiums paid). Wealth Quintile Measure based on per capita food expenditure. Formal financial savings of household. Age of household head. Age squared measure for hypothesis testing. Size of family. Size of family squared for hypothesis testing. Number of family members earning an income. Number of children in the household. Whether household receives financial support from children. Whether household head is married. A Measure from 1 to 5 where 1 = Cannot read and write; 2 = Can read and write but did not finish primary school; 3= Finished Primary School; 4 = Finished Lower Secondary School; 5 = Finished Upper Secondary School; 6 = Third Level. Whether household head of household is a farmer. Has the household suffered from prior income shock between the years 2002 to 2005 (1 Yes, 0 No). Interaction term for hypothesis testing. Dummy variable representing each of the 12 provinces surveyed. Used for control purposes. Table 2: Types of Insurance against percentage of participating households Types of Insurance Fire Life Agriculture Social Farmers Social Health Free Social Free Health for Children Vehicle Education Other 16 Percentage of households 0.1 7.0 0.0 15.4 1.0 53.3 3.0 36.2 29.9 7.2 3.9 Table 3: Baseline Tobit Demand Equation and Probit Participation equation Tobit Demand Equation Wealth Measure of Household – Quintile 2 Wealth Measure of Household – Quintile 3 Wealth Measure of Household – Quintile 4 Wealth Measure of Household – Quintile 5 Financial Savings of Household Gender of Head of Household (1 Male, 0 Female) Debt Net Income Net Income * Number of Earners Age Age Squared Family Size Family Size Squared Number of Earners Number of Children Children Support (1 yes, 0 no) Married (1 Yes, 0 No) Education Measure Farmer (1 Yes, 0 No) Prior Income Shock (1 yes, 0 no) Provincial Dummies Coefficient 0.4070 0.8982*** 1.8078*** 2.6899*** 0.0672*** -0.5872* -0.0141 0.2844 1.1992 0.0873 -0.0008 1.2324*** -0.0678*** -0.3713 -0.2695* -0.4070 0.2719 0.7738*** -0.5460*** 0.0354 Yes Standard Error 0.4568 0.4581 0.4541 0.4868 0.0319 0.4014 0.0480 1.3857 1.3651 0.0728 0.0007 0.3750 0.0316 0.4197 0.1879 0.3203 0.4608 0.1237 0.2621 0.2741 Heteroscedasticity equation Coefficient Standard Error Financial Savings -0.0249*** 0.0071 Net Income -0.1112*** 0.0256 _Cons 2.5630 30.6278 *** denotes significance at 1% level; ** denotes significance at 5% level; *denotes significance at 10 level% Log Likelihood = -3029.4335 Number of Observations = 1751 IHS estimate = 1.21409 Probit Participation Equation Wealth Measure of Household – Quintile 2 Wealth Measure of Household – Quintile 3 Wealth Measure of Household – Quintile 4 Wealth Measure of Household – Quintile 5 Financial Savings of Household Gender of Head of Household (1 Male, 0 Female) Debt Net Income Net Income * Number of Earners Age Age Squared Family Size Family Size Squared Number of Earners Number of Children Children Support (1 yes, 0 no) Married (1 Yes, 0 No) Education Measure Farmer (1 Yes, 0 No) Prior Income Shock (1 yes, 0 no) Provincial Dummies _cons 17 Coefficient 0.0259 0.1314 0.3074*** 0.6105*** 0.0155** -0.0937 0.0027 0.1238 0.1730 0.0231 -0.0002 0.2867*** -0.0167*** -0.0658 -0.0677* -0.0654 0.0787 0.1660*** -0.0857* 0.0016 YES -5.064*** Standard Error 0.1051 0.1082 0.1099 0.1224 0.0099 0.1099 0.0130 0.0130 0.3526 0.3470 0.0187 0.0002 0.0949 0.0080 0.1074 0.0486 0.0827 0.1220 0.0328 0.0690 0.6691 Table 4: Informal Risk Sharing Effects Tobit Demand Equation Informal Information Sharing Measure Formal Social Organisations Measure Coefficient -0.6689** 1.5154*** Standard Error 0.2576 0.3038 Probit Participation Equation Informal Information Sharing Measure Formal Social Organisations Measure Coefficient -0.2055*** 0.3730*** Standard Error 0.0678 0.0789 Table 5: Instrumented Informal Risk Sharing Effects Tobit Demand Equation Informal Information Sharing Instrument Formal Social Organisations Instrument Coefficient -10.5860 *** 3.8406*** Standard Error 3.0140 0.6234 Probit Participation Equation Informal Information Sharing Measure Formal Social Organisations Measure Coefficient -2.6688*** 1.0612*** Standard Error 0.7800 0.1778 18 Appendix Stage 1 Reduced form Estimates: Coefficient Standard Error Informal risk sharing networks Wealth Measure of Household – Quintile 2 0.0233 0.0375 Wealth Measure of Household – Quintile 3 0.0482 0.0390 Wealth Measure of Household – Quintile 4 0.0759** 0.0399 Wealth Measure of Household – Quintile 5 0.0218 0.0438 Financial Savings 0.0029 0.0035 Gender 0.0006 0.0392 Debt 0.0125*** 0.0046 Net Income -0.1720 0.1201 Net Income * Number of Earners 0.1377 0.1162 Age -0.0069 0.0067 Age Squared 0.0000 0.0000 Family Size 0.0552 0.0336 Family Size Squared -0.0053* 0.0028 Number of Earners -0.0190 0.0361 Number of Children 0.0227 0.0174 Children Supporting (1 yes, 0 no) 0.0361 0.0302 Married (1 yes, 0 no) -0.0448 0.0439 Education Measure 0.0012 0.0117 Farmer (1 yes, 0 no) 0.0084 0.0248 Prior Income Shock (1 yes, 0 no) 0.0450** 0.0257 Instruments for informal risk-sharing networks Distance from commune to local state bank 0.0782** 0.0480 Index of informal risk-coping intensity 0.0093*** 0.0046 Instruments for formal risk-sharing networks Distance of household to local peoples committee 0.0068*** 0.0029 office Formal group membership 0.0041 0.0109 _cons 0.5255*** 0.2408 *** denotes significance at 1% level; ** denotes significance at 5% level; *denotes significance at 10 level% 19 Appendix (continued) Stage 1 Reduced form Estimates: Coefficient Standard Error Formal risk sharing networks Wealth Measure of Household – Quintile 2 0.0177 0.0289 Wealth Measure of Household – Quintile 3 -0.0001 0.0301 Wealth Measure of Household – Quintile 4 0.0188 0.0308 Wealth Measure of Household – Quintile 5 0.0218 0.0337 Financial Savings 0.0033 0.0027 Gender 0.0309 0.0302 Debt 0.0048 0.0036 Net Income -0.2488*** 0.0926 Net Income*Number of Earners 0.2230*** 0.0896 Age -0.0039 0.0052 Age Squared 0.0000 0.0000 Family Size 0.0603*** 0.0258 Family Size Squared -0.0032 0.0022 Number of Earners -0.0547** 0.0279 Number of Children -0.0228** 0.0135 Children Supporting (1 yes, 0 no) 0.0228 0.0233 Married (1 yes, 0 no) -0.0122 0.0339 Education Measure -0.0052 0.0090 Farmer (1 yes, 0 no) -0.0038 0.0191 Prior Income Shock (1 yes, 0 no) 0.0395** 0.0198 Instruments for informal risk-sharing networks Distance from commune to local state bank 0.0711*** 0.0370 Index of informal risk-coping intensity -0.0076*** 0.0035 Instruments for formal risk-sharing networks Distance of household to local peoples committee 0.0025* 0.0022 office Formal group membership 0.1779*** 0.0084 _cons 0.6126*** 0.1856 *** denotes significance at 1% level; ** denotes significance at 5% level; *denotes significance at 10% level 20