Precautionary Saving and Health Insurance: A Portfolio Choice Perspective

Transcription

1 Front. Econ. China 2016, 11(2): DOI /s RESEARCH ARTICLE Jiaping Qiu Precautionary Saving and Health Insurance: A Portfolio Choice Perspective Abstract This paper analyzes the impact of health insurance on household portfolio choice. Using the U.S. Survey of Consumer Finance and Health Retirement Survey databases, it finds that insured households are more likely to own stocks and invest a larger proportion of financial assets in stocks than uninsured households do. The results remain strong even after controlling for household characteristics and reverse causality. Further, the results are robust across different survey years and data sources. It suggests that a precautionary motive is strong in household portfolio choice decisions. Keywords precautionary motive, health insurance, portfolio choice JEL Classification G11, I10, I18 1 Introduction Health expenditure risks could be quite serious among certain groups of people. For example, Feenberg and Skinner (1994) documented that health expenditures accounted for more than 20% of the household income of 4.3% of non-elderly households. Health expenditure risk could be even more serious for uninsured households. According to the Employee Benefit Research Institute, as of 1996, 17.4% of the US population had no health insurance, a number that had grown by 15% between 1988 and The percentage of uninsured Americans remains high. In 2007, the Census Bureau reported that about 15% of the total population did not have any form of health insurance. Health expenditure risk, therefore, is a potentially important background risk and might have a significant impact on a household s consumption, savings and asset allocation. This paper investigates systematically, the relationship between household Received October 02, 2014 Jiaping Qiu ( ) DeGroote School of Business, McMaster University, Hamilton, Ontario L8S 4M4, Canada qiu@mcmaster.ca

2 Precautionary Saving and Health Insurance 233 health insurance status and portfolio choice decisions. This issue is of great importance for at least two reasons. First, the empirical results help us understand the link between social security systems and financial markets. If household portfolio choice is associated with health insurance status, the direct policy implication would be that the growth of the equity market is related to the evolution of social security systems. Second, it offers new insights into the effect of the precautionary motive on household saving behavior. Although moral hazard and adverse selection problems have prevented households from fully insuring against health expenditure uncertainty, health insurance certainly reduces the risk of out-of-pocket health expenditure. If the precautionary motive is strong, it should be expected that uninsured households would tend to hold more risk-free and liquid assets than insured households. Therefore, the empirical results regarding the relationship between health insurance and portfolio choice could provide further evidence on the impact of the precautionary motive on household savings decisions. This paper addresses two closely related issues that have been studied extensively but separately in the literature: the impact of health insurance on precautionary savings; and the effect of the precautionary motive on household portfolio choices. The impact of health insurance on household precautionary savings has been the focus of great attention. Palumbo (1999) found that uncertainty about out-of-pocket medical expenses plays an important role in the motive for precautionary savings among the elderly. Gruber and Yelowitz (1999) investigated the effect of Medicaid on household consumption expenditure and found a strong positive relationship between health insurance and consumption. Chou et al. (2002) also documented a negative relationship between health insurance and household savings in Chinese Taiwan. Using the Survey of Consumer Finance 1989 data, McCluer (1996) examined the interaction between health insurance and the saving behavior of non-retired people. She found that households with health insurance have more wealth than those without insurance; this finding is consistent with the notion that the precautionary motive is weak in uninsured households savings decisions and the endogeneity between health insurance and wealth as well. The theory regarding the precautionary motive suggests that future expenditure risk could affect household saving behavior not only through the choice between consumption and saving, but also through the allocation of

3 234 Jiaping Qiu savings, that is, the choice among assets with different degrees of risk. Kimball (1993) showed that, under standard risk aversion, the zero-mean background risk will induce a household to allocate a larger share of wealth to risk-free assets. Koo (1991) and Bertaut and Haliassos (1992) also demonstrated that the demand for risky assets will decrease as the background risk of income increases. Eechoudt (1996) further presented a necessary and sufficient condition of the property of a utility function which guarantees more risk-averse behavior when facing first and second order stochastic dominance changes in risk. Gormley, Liu and Zhou (2010) presents a model to show that insufficient insurance against large negative shocks could create strong precautionary motive and explain the limited stock market participation puzzle (e.g., Campbell, 2006; Huang and Wang, 2007). The empirical studies of the effect of background risk on household portfolio structure have focused on labor income risk, and found that labor income uncertainty can reduce the incentive to hold risky assets (see, among others, Chakraborty and Kazarosian, 1996; Guiso et al., 1996; Hochguerted, 1997; Heaton and Lucas, 1999, 2000). Less attention, however, has been paid to the impact of health expenditure uncertainty on household portfolio choice. On the other hand, the empirical research on the relationship between health expenditure uncertainty and household financial wealth has focused on the relationship between health insurance and savings. Yet, a subtler question remains unanswered: What is the impact of health insurance on household portfolio choice? A closely related stream of study is to investigate the impact of health status on household portfolio choice. Edward (2003) argued that health status could affect household portfolio choice through its effect on risk aversion if health status affects household utility. Rosen and Wu (2003) pointed out that there are several channels through which health status could affect household portfolio choice, such as risk aversion, planning horizon, and life expectancy. The authors tried to identify the channels through which health status can affect household portfolio choice, but found that the health status effect remains significant even after controlling for factors such as risk attitude, planning horizon, life expectancy, and bequest motive. The paper by Berkowitz and Qiu (2003) found that the effect of health status on the riskiness of household portfolio disappears after controlling for household financial assets, labor income, age and education.

4 Precautionary Saving and Health Insurance 235 Health insurance status is closely related to uncertainty about households future health expenditure, everything else (health status, wealth, age, etc.) being the same. Given the US government s increasing involvement in providing and ensuring access to medical care, it might be more important, from a policy point of view, to understand the relationship between health insurance and portfolio choice. This paper uses the U.S. Survey of Consumer Finance (SCF) and the Health and Retirement Survey (HRS) to examine the relationship between health insurance status and household asset allocation. It finds that uninsured households are less likely than insured households to invest in stocks. Moreover, insured households tend to allocate a greater portion of their wealth to risky assets than do uninsured households. These results are strong even after controlling for household observable and unobservable characteristics, reverse causality; and robust cross different SCF survey years and two independent data sources, the SCF and the HRS. The evidence on the relationship between health insurance and portfolio choice indicates that households portfolio choices are influenced by their health insurance status and suggests that precautionary motives are strong in household saving decisions. The results imply a positive link in the development between social security systems and financial markets. The paper proceeds as follows. Section 2 uses a simple conceptual model to illustrate how health insurance can affect portfolio choices through precautionary motives. Section 3 discusses the data sources. Section 4 uses information from the Survey of Consumer Finance to analyze the relationships among health insurance, wealth, and wealth composition. Section 5 presents the regression results regarding the effect of health insurance on the household portfolio structure using information from the Health and Retirement Survey. Section 6 summarizes and concludes the paper. 2 A Conceptual Model To understand how uncertain health expenditure can, through a precautionary motive, influence household consumption, savings and the allocation of savings between risk-less and risky assets, I consider a simple problem of a representative household that is maximizing expected inter-temporal utility faced with choices among consumption, saving, and the allocation of assets between a risky and a risk-less asset. In order to capture the effect of health expenditure uncertainty, the household is assumed to have a certain income, Y, but is

5 236 Jiaping Qiu uncertain about health expenses, M t. The household lives infinitely with a time additive isoelastic utility. At the beginning of period t, after observing income and health expenditures, the household chooses consumption, C t, saving in bond, B t, and saving in stocks, S t, to maximize its lifetime-expected utility. Hence, the household solves the following problem, subject to, { Bt, St} 0 t= 0 1 γ t t C 1 MAX E β ; γ> 0 (1) 1 γ C+B+S W, (2) t t t t s t+ 1 t t+ 1 t f t+ 1 W =S (1 +r ) +B (1 +r ) +Y M, (3) 2 M M ~ N( M,σ ), (4) t C> 0, W> 0, (5) t where E 0 is expectation at time 0. C t is the consumption at time t. M t is the health expenditure at time t and is assumed to be I.I.D. normal distribution 2 s with the mean M and variance σ M. β is the discount rate. r t+ 1 is the return of stocks and rf is the return of risk-less assets. In addition to the flow of wealth constraint, the household is assumed to face a short sale and borrowing constraints in the equity and bond markets respectively, that is, B 0, (6) t t S 0, (7) t The coefficient of relative risk aversion, γ, measures the strength of both risk aversion and prudence. Two Euler equations determine the optimal choices of bonds and stocks. γ γ γ B t : ( Wt St B t) =MAX[( Wt St), βet[ rf ( Wt+ 1 St+ 1 Bt+ 1) ]. (8) γ γ s γ S t : ( Wt St B t ) =MAX[( Wt Bt ), βet[ rt+ 1( Wt+ 1 St+ 1 Bt+ 1) ]. (9) I numerically find the optimal holdings of bonds and stocks. The bond return is assumed to be risk free with an annual return 2%. The stock returns are assumed to take on the values 0.08 and with an equal probability to match the U.S. stock market historical return of the mean 7.75% and the standard deviation of 15.7%. The rate of time preference β is set to 0.9. The coefficient

6 Precautionary Saving and Health Insurance 237 of relative risk aversion is set equal to 8. 1 The income is normalized to be 1. The purpose of this model is to qualitatively explore the impact of expenditure uncertainty on portfolio choice. The quantitative assessment will be left as an empirical matter. I will therefore not attempt to calibrate the household health expenditure process. Instead, I consider several artificial situations with different degrees of health expenditure risks and examine how they can affect household consumption and portfolio choice. Specifically, I consider the household portfolio choice under three scenarios with increasing degrees of health expenditure risks. In situation 1, the household is fully insured and faces no medical expenditure risk. Situation 2 is a first-order stochastic deterioration of the health expenditure risk of situation 1, so that households face a certain level of medical uncertainty. The mean medical uncertainty is assumed to be 0.3 with the standard deviation The process of health expenditure is set to take on the values 0.45 or 0.15 with equal probability. Situation three is a mean preserving shift of the health expenditure uncertainty of situation 2. The mean of medical expenditure is still 0.3 but the standard deviation is doubled to 0.3. In this case, the process of health expenditure is set to take on value 0.60 or 0 with equal probability. The optimal policy functions of consumption and asset allocation with respect to wealth are presented in Figure 1. Figure 1a shows the household policy function of consumption. The policy functions of consumptions show that the household will spend all of its wealth when wealth is below certain levels. For example, in situation 1, when wealth is below 0.9, the household will spend all of its wealth on consumption and will not engage in saving. The intuition behind this behavior is simple: because the household will have greater wealth from income in the future, it would like to borrow and increase consumption today to 1 Since the result might be sensitive to the risk aversion parameters, I also experiment with coefficients that are equal to 2, 4, and 10. All conclusions remain unchanged. 2 Given that income is normalized to 1, 0.3 mean health expenditure implies the average health expenditure accounts for 30% of the income. Since the focus of this model is the comparative static analysis of the impact of different degrees of expenditure risk on the portfolio choice, this number is chosen only to reflect a FOSC change of health expenditure. Changing the mean of health expenditure does not affect the conclusion as long as there is an increase of health expenditure. Feenberg and Skinner (1994) found that health expenditure accounted for more than 20% of the household income of 4.3% of non-elderly households. Since the health shocks could lead to a large decline of household labor income (Wu 2000), 30%%health expenditure/ income ratio could be reasonably viewed as the health expenditure level for the sick and uninsured households.

7 238 Jiaping Qiu equalize the inter-temporal marginal utility of consumption. However, due to a borrowing constraint, it will not be able to do so and can only consume all the wealth it has. Similar behavior could be found in situations 2 and 3, but we can see that the households will engage in saving at a lower level of wealth and that the consumption function is higher when the household faces a lower expenditure uncertainty, indicating that household will save more when facing greater expenditure uncertainty. This result suggests that a precautionary motive Figure 1a The Household Policy Function of Consumption Figure 1b The Policy Function of s to Wealth Ratio Figure 1 The Optimal Policy Functions of Consumption and Asset Allocation

8 Precautionary Saving and Health Insurance 239 will lead the household to save more when facing greater health expenditure uncertainty, which is consistent with previous empirical findings on the impact of health insurance on precautionary saving. To see the impact of expenditure uncertainty on the asset allocation of savings, Figure 1b plots the policy function of stocks to wealth ratio. When households start saving, they will first put all of their savings in stocks due to the return premium of stocks over bonds. 3 However, from Figure 1b, one can see that households will start to invest in bonds at a lower wealth level when facing greater expenditure risk. Further, they will invest a smaller proportion of wealth in stocks when the expenditure risk increases for a given level of wealth. These results show that the precautionary motives will not only lead households to save more but also to allocate fewer assets to risky assets when facing a greater expenditure uncertainty. In this study, health insurance reduces the risk of health expenditure risk, which discourages households precautionary savings and leads them to save more in risk-less assets. If the precautionary motive is strong, other things being equal, the insured household will invests a greater proportion of financial assets into stocks than uninsured households do. 3 Data Although the impact of health insurance through a precautionary motive on household saving has been examined extensively, the issue of how health insurance can affect household saving allocation remains unexploited. One important reason might be the difficulty in finding data sets with detailed information on households portfolio structure as well as on their health insurance status. The U.S. Survey of Consumer Finance (SCF) and Health and Retirement Survey (HRS) are two data sources which include comprehensive information on household wealth and health. Each database, however, has its own strengths and drawbacks. 3 The results are similar to the results from a portfolio choice model with income uncertainty and liquidity constraint, such as in Haliassos and Michaelides (2003) and Heaton and Lucas (1997). The reason is simple: the health expenditure essentially changes the household income process. Hence, situation 2 could be viewed as a first order stochastic dominated shift of the income process of situation 1, while situation three is a second order stochastic dominated shift of the income process in situation 2.

9 240 Jiaping Qiu The SCF has been regarded as a distinguished comprehensive and reliable source of information on household wealth. The SCF provides richer information on household portfolio information than the HRS. For example, it offers a clear classification of mutual funds into equity funds and bond funds, which allows for a more precise measurement of stock invested through equity fund; better information on the household investments through retirement accounts. Although the survey questions in SCF regarding the allocation of assets held in retirement accounts are categorical and have a certain noise in household portfolio share, the HRS does not indicate asset types in the retirement accounts. Ameriks and Zeldes (2001) found that more households invest in stocks through retirement accounts than through direct stock holdings. Therefore, a better measurement of asset allocation through retirement accounts might be important. SCF also includes detailed information on household health status and health insurance coverage The major shortcoming of the SCF is that the surveys do not follow the same set of households over time and have only cross-sectional features. This issue may be serious in regression analysis since it is not able to control for unobservable household characteristics. The HRS has both cross-sectional and time-series information. The HRS consists of a nationally representative sample of people who were born between 1931 and 1941 in the first wave of the survey; interviews have been conducted every two years since The HRS includes comprehensive questions about household wealth, health and health insurance status. As discussed above, the information of financial assets in the HRS is less detailed than in the SCF and should be noisier in the measurement of household portfolio shares. Based on these considerations, I will use both the SCF and the HRS to investigate the relationship between health insurance and portfolio choice and to compare the consistency of the results from two data sources. 4 Evidence from the Survey of Consumer Finance 4.1 Health Insurance, Wealth and Wealth Composition This section describes the basic relationship between household health insurance status, wealth and wealth composition using information from the SCF. The

10 Precautionary Saving and Health Insurance 241 results reported here are from the SCF Table 1 provides summary statistics on household wealth and health insurance coverage. The population weights are applied to make the statistics represent the population. Column 3 reports information regarding health insurance coverage among different groups of households. The first row of column 3 shows that 19.6% of U.S. households are not fully covered by health insurance. 5 An inspection of column 3 reveals that poor or young households are less likely to have health insurance; and that better education level is associated with higher rate of insurance coverage. The pattern indicates that health insurance coverage is associated with household social status. I use three measures of household wealth to discover the relationship between the household wealth level and insurance status: liquid assets, financial assets, and net worth. The liquid assets include checking accounts, money market deposit accounts, savings accounts, call accounts, certificates of deposit, and savings bonds. The financial assets include liquid assets, stocks, bonds, mutual funds, individual retirement accounts, Keogh accounts, defined contribution pension plans, trust funds, and the cash value of life insurance policies. The net worth is the net value of the household s assets minus its total liabilities. Columns 4 through 9 compare the mean/median wealth of insured and uninsured households. For given income, age, and education categories, the mean/median wealth of households with insurance is greater than the wealth of households without insurance. The results clearly indicate a positive relationship between wealth levels and insurance coverage status. Using the SCF 1989 data, McCluer (1996) presented a similar result and argued that this positive relationship is striking. Since the insured households hold a much higher level of wealth than uninsured households, it implies that insured households save more than the uninsured households and suggests that the precautionary motive for household saving is weak among insured households. However, the positive relationship between wealth level and insurance status could simply be due to the endogeneity between wealth and insurance; in other words, wealthier households are more likely to have health insurance. 4 All results reported in this paper are robust using information from other survey year SCFs. The results from other survey years (1989, 1992, 1995) could be obtained from the author upon request. 5 I define the insured household as the household in which all the members are insured by some type of public or private insurance. The same definition was applied in McCluer (1996).

11 Table 1 Wealth of Insured and Uninsured Households Percent of Households (2) Percent Uninsured (3) All Households Uninsured (4) Mean/Median Household Wealth Liquid Assets Financial Assets Net Worth Insured (5) Uninsured (6) Panel A: Household Income < $15K ,075/70 3,744/490 10,705/250 Insured (7) Uninsured (8) Insured (9) 18,531/1,290 30,339/2,700 62,467/13,050 $15 30K ,443/500 10,144/2,000 11,779/1,002 51,653/13,100 42,985/9, ,298/53,800 $30 50K ,172/1,100 10,168/3,000 25,519/5,700 68,457/24,700 72,212/29, ,443/85,500 $50 100K ,858/3,050 17,176/6,560 56,704/17, ,125/66, ,731/86, ,067/163,400 >100K ,963/10,750 86,951/22, ,646/107, ,728/271,000 1,220,666/259,970 1,928,184/620,300 Panel B: Age of Household Head (years) Under ,484/250 6,877/1500 6,808/660 4,3048/6,450 20,222/3,850 87,185/15, ,273/400 11,804/3,000 21,741/1, ,864/28,400 72,239/14, ,508/81, ,183/1,001 21,004/4,900 42,319/4, ,176/49, ,819/39, ,003/122, ,831/1,300 28,527/4,730 96,634/4, ,248/51, ,646/43, ,362/169, above ,055/200 23,436/5,000 82,208/2, ,854/36, ,709/70, ,949/141,000 Penal C: Education of Household Head Less than High ,442/60 6,725/750 9,041/200 30,330/3,620 46,795/5,800 95,863/37,400 School High School ,330/530 11,874/2,700 20,475/1,500 98,153/21,000 72,121/13, ,166/82,450 College Degree ,907/1,500 29,960/6, ,956/6, ,789/65, ,583/32, ,833/162,600 (To be continued)

12 Percent of Households (2) Percent Uninsured (3) Uninsured (4) Mean/Median Household Wealth Liquid Assets Financial Assets Net Worth Insured (5) Uninsured (6) Insured (7) Uninsured (8) (Continued) Insured (9) All Households Panel D: Working Status of Household Head Employed by ,259/530 11,942/3,000 20,962/1, ,171/24,800 46,838/9, ,334/71,100 Others Self Employed ,808/ ,252/7,500 71,199/4, ,284/65, ,308/61,000 1,102,181/317,670 Not Working ,281/100 20,507/3,000 44,795/ ,149/23, ,042/6, ,073/110,500 Note: This table provides the relationship between wealth and health insurance status for different categories of households, using the information from the SCF 1998 database. Households are divided into different categories according to household income (Panel A), age of household head (Panel B), education of household head (Panel C) and working status of household head (Panel D). Column 2 shows the percentage of households in each category. Column 3 shows the percentage of uninsured households in each category. Columns 4 and 5 compare liquid assets between insured and uninsured households. Columns 6 and 7 compare financial assets between insured and uninsured households. Columns 8 and 9 compare net worth of insured and uninsured households.

13 244 Jiaping Qiu The question that is of interest is the following: keeping wealth and other household characteristics constant, does health insurance affect the allocation of the household wealth? To answer this question, I will focus the analysis on the relationship between health insurance and the composition of household financial assets. I examine the proportion of two types of financial assets to total financial assets: stocks and liquid assets. The stocks are used to assess the risk of household financial assets and include direct stockholding, stock mutual funds, annuities and trusts consisting primarily of stocks, investment retirement accounts and defined contribution pension accounts invested in stocks. This definition follows Bertaut and McCluer (2000), Ameriks and Zeldes (2001) and Heaton and Lucas (2000). 6 The second type of financial assets are liquid assets that ensure the liquidity of household financial assets. The definition of liquid assets used here is the same as that of Faig and Shum (2002), and includes checking and savings accounts, call accounts at brokerages, and money market accounts either in deposits or in mutual funds. Table 2 presents the difference between insured and uninsured households in the average of the share of stocks and liquid assets in their financial assets. In Panel A, one can see that households increase their proportion of financial assets invested in stocks and reduce the percentage of liquid assets as income increases; this is true both for insured and uninsured households. Yet, among households in the same income category, except those with incomes greater than $100K, the insured households allocate a greater portion of their financial assets to stocks and less to cash than uninsured households do. The results appear to be reasonable. For high-income households, health expenditures on the average account for a relatively smaller portion of their wealth. The impact of health expenditure uncertainty might be less serious for this category of households. Hence, the differences in portfolio choice between insured and uninsured households are weak among members of the high-income group. 7 6 In the earlier survey in 1992 and 1995, annuities and trusts are not differentiated and are generally called other managed assets. In principle, private business, housing, and other highly non-liquid assets should be treated as risky assets. However, housing and some real estate holdings are not necessary for investment purposes, and may be highly non-liquid and unresponsive to change in background risk. Since the analysis is restricted to the allocation of financial assets, the above assets are not taken into consideration. 7 From Figure 1b, one can see that, when wealth increases, the difference in the share of stocks in financial assets among different degrees of health expenditure risk decreases.

14 Table 2 The Composition of Financial Assets of Insured and Uninsured Households A: Household Income <15K 15K 30K 30K 50K 50K 100K >100K Uninsured Insured Uninsured Insured Uninsured Insured Uninsured Insured Uninsured Insured of of Cash <35 Uninsured Insured Uninsured Insured of B: Age of the Household Head >65 Uninsured Insured Uninsured Insured Uninsured Insured of Cash C: Education of the Household Head Less than High School High School College Degree Uninsured Insured Uninsured Insured Uninsured Insured of of Cash D: Household Head Working Status Self-employed Work for others Not working Uninsured Insured Uninsured Insured Uninsured Insured of of Cash Note: This table provides the basic relationship between household portfolio choice and health insurance status for different categories of households, using the information from the SCF 1998 database. Households are divided into different categories according to household income (Panel A), age of household head (Panel B), education of household head (Panel C) and working status of household head (Panel D).

15 246 Jiaping Qiu Panel B presents the difference in portfolio choice between insured and uninsured households, according to different age categories. An inspection of Table 2 shows that the young and old households hold less stock and more liquid assets, while the middle-aged households hold more stocks and less cash. Several studies have tried to explain this reversed U shape of U.S. household stock holding (Bodie et al., 1992; Constantides et al., 1998; Viceira, 2001). Of interest for this study is the fact that, for all age categories, the insured households hold more stocks and less cash than the uninsured households. Similar patterns can be found in Panel C and Panel D, in which the households are categorized by education level and employment status. The basic summary statistics present a pattern that insurance is associated with riskier assets and less cash holding among income, age, education, and employment status categories. 4.2 Regression Analysis The summary statistics show a strong relationship between household portfolio choice and health insurance status. Of course, other household characteristics might affect the household portfolio choice. In this section, I use information from the SCF to examine the impact of health insurance on household demand for stocks and liquid assets but controlling for the effects of other household characteristics. Two models are used to gauge the impact of health insurance status on household portfolio choice. First, a probit model is used to estimate the effect of health insurance on the ownership of stock holdings. The dependent variable is equal to 1 if the household own stocks, and 0 otherwise. Second, a tobit model is used to analyze the share of stocks and liquid assets in financial assets. Since not all households own stocks and liquid assets and some households invest all of their financial assets in stocks or liquid assets, the tobit model with the lower limit zero and upper limit 1 is appropriate in handling this problem. The dependent variable for the tobit model is the share of stocks or liquid assets of the financial assets. Hence, the sample households are restricted to those that have positive financial assets in the tobit model estimation. 8 The control variables in the basic regression include: (1) a dummy variable indicating the household health insurance status that is equal to 1 if the household is fully insured and 0 otherwise; (2) a dummy variable indicating the household head s health status. In the SCF, households were asked, Would you 8 In a later section using the information from the HRS, I will use a panel methodology to control for unobservable household characteristics.

16 Precautionary Saving and Health Insurance 247 say your health is excellent, good, fair, or poor? The dummy variable is equal to 1, indicating that a household is sick if it reported that its health is fair or poor, and 0 otherwise; 9 (3) since the analysis focuses on financial asset allocation, I use financial assets to control for the wealth effect. A logarithm of household financial assets is adopted to control for the nonlinearity of the wealth effect; 10 (4) a logarithm function of household head labor income. Some authors (for example, Rosen and Wu 2002) use total income to control for the income effect. The reason for using labor income rather than total income to control for the income effect is to avoid endogeneity between the portfolio choice and capital income, which is a part of household total income. A logarithm function is used to model the non-linearity of the income effect; (5) age of household head; (6) a dummy variable for the education level that is equal to 1 if the head has a college degree and 0 otherwise; (7) a dummy variable for the education level that is equal to 1 if the head s education level is less than high school and 0 otherwise; (8) a dummy variable for race which is equal to 1 if the respondent is white and 0 otherwise; (9) a dummy for marriage status which is equal to 1 if the head is married and 0 otherwise; (10) a dummy variable for the gender of the household head, which is equal to 1 if the household head is female and 0 otherwise; (11) a control variable for family size. The SCF dataset consists of five implicates due to the multiple imputation technique used to handle missing data. The analysis in this paper uses the information contained in all five implicates. The coefficients and t-statistics are adjusted for the imputation using the program provided in the SCF code book. 11 Table 3 reports the regression results from the probit and tobit models. The estimated coefficients of household wealth, income, education, and race are all significant with the sign that is consistent with common wisdoms. Wealthier, higher-income, better educated and white households are more likely to own 9 The results will not be affected if sick households are defined as those that report their health as poor. 10 Some may want to use a quadratic function to control for the nonlinearity. A quadratic function, however, has a major drawback since it assumes decreasing absolutely risk averse once wealth is above certain level. The results here are robust to linear, quadratic and cubic functions of wealth effect. 11 See the Survey of Consumer Finance Code Book 1998 for details.

17 248 Jiaping Qiu stocks and invest more in stocks. Gender and the number of household numbers appear to have no significant impact on household portfolio choice. The estimated coefficient of health insurance from stock ownership equation (probit Table 3 Regression Analysis Using the SCF: Basic Models s Liquid Assets Probit (2) Tobit (3) Tobit (4) Household Insured *** (0.076) *** (0.025) *** (0.018) Head Sick (0.069) ** (0.022) (0.016) Log(Financial Assets) *** (0.015) *** *** Log(Income) *** (0.007) *** (0.002) Age *** (0.002) *** *** (0.0001) College Degree * (0.057) *** (0.017) *** (0.013) Less than High School ** (0.090) *** (0.031) (0.020) Married (0.078) (0.023) (0.017) White *** (0.072) *** (0.023) (0.017) Female (0.080) (0.026) *** (0.018) Family Number (0.025) (0.007) (0.006) Constant *** (0.191) *** (0.054) *** (0.036) N 21,515 20,300 20,300 R-squared Note: This table analyzes the effect of health insurance status on the households portfolio choice, using the information from the SCF 1998 database. Column 2 reports the impact of health insurance status on household stock ownership using the probit model. Column 3 reports the impact of health insurance status on the share of stocks of financial assets using the tobit model. Column 4 reports the impact of health insurance on the share of liquid assets in financial assets using the tobit model. Asymptotic standard errors are reported in parentheses. Numbers with * are significant at the 10-percent level. Numbers with ** are significant at the 5-percent level. Numbers with *** are significant at the 1-percent level. N is the number of observations.

18 Precautionary Saving and Health Insurance 249 model) is and is significant at the 1% level, which implies that the probability of holding stock for households with health insurance is about 9.51% higher than the probability of households without health insurance, when other characteristics are at the means of the sample. The coefficient of health insurance from the stock share equation is in the tobit model and is significant at the 1% level. The calculation of the marginal effect indicates that insured households hold a 4.02% greater portion of their financial assets in stocks than do uninsured households. Column 4 reports the impact of health insurance status on the household holding of liquid assets. The coefficient of household health insurance is and is significant at the 1% level. The point estimator indicates that health insurance is associated with a decrease of 4.22% in the proportion of financial assets to be held in liquid assets. While the above results are consistent with the precautionary motives story, health insurance may affect household portfolio choice through other underlying attitudes in the household. For example, households with health insurance may expect a longer life expectancy than uninsured households. Health insurance may allow households to plan their investments for a longer horizon. Moreover, health insurance could even change households underlying attitudes to risk aversion. To explore the possibility that the link between household insurance status and portfolio choices may through household s underlying attitudes, I augment the basic regression in Table 3 with several important household self-assessed attitudes which may be related to household insurance status as well as portfolio choices. These variables include: (1) household head self-reported life expectancy. This variable is based on the answer to the question: About how old do you think you will live to be? ; (2) dummy variables indicate household planning horizons; these variables are constructed based on the answers to the question: In planning your family's saving and spending, which of the time periods listed on this page is most important to you and your (spouse/partner)? The choices to answer this question are the next few months, next year, in the next few years, the next 5 10 years, and longer than 10 years. (3) dummy variables for household head s risk tolerance. In SCF, the respondents were asked: Which of the statements on this page comes closest to the amount of financial risk that you and your (spouse/partner) are willing to take when you save or make investments? The SCF codes the answer to this question on a 1 to 4 scale, with 1 representing the taking of a substantial financial risk; 2 representing the taking of an above average financial risk; 3 representing the taking of an average financial risk and 4 representing an unwillingness to take

19 250 Jiaping Qiu any financial risk; (4) dummy variables for household head s bequest motives. These are based on the question: Do you and your (spouse/partner) expect to leave a sizable estate to others? There are three possible answers: yes, possibly, or no ; (5) dummy variables for household shopping habits. Since acquiring private insurance for the whole family requires (like investing in stocks) a significant search effort, the shopping experience that the household might acquire from the health insurance purchase might make investment in stocks more likely. The SCF asks households how intensively they comparison shop for the best terms when making major saving and investment decisions. Households could choose one of the five scales to describe their shopping intensity. Scale 1 is almost no shopping. Scale 3 is moderate shopping, and scale 5 is a great deal of shopping. Scales 2 and 4 indicate the intensity between 1 and 3, or 3 and 5, respectively. Table 4 presents the results from the augmented models. The household planning horizon, risk attitude, bequest motives and shopping habits appear to have a significant impact on household portfolio choices. The longer the planning horizon, the more risk loving, stronger bequest motives and stronger shopping habits are associated with more stocks investment in financial assets. From the results of the health insurance ownership equation, it is interesting to see that health insurance is not related to life expectancy, planning horizon, risk attitude and bequest motives. At the same time, though, it is highly correlated with household shopping habits, consistent with the notion that acquiring insurance for the whole family requires some search effort. An inspection of the coefficients on health insurance reveals that including self-assessed household characteristics has little explanatory power of the relationship between health insurance and household portfolio choice. The overall coefficients estimated from the augmented model are slightly smaller than those from the basic models but all remained significant at the 1% level. In sum, the results from the SCF indicate that household portfolio choice decisions are associated with health insurance status and the results are robust across different SCF survey years. 12 The relationship is not generated through underlying attitudes of households, such as risk aversion, life expectancy, shopping habits and so on. The results are consistent with previous studies on the impact of health insurance on household saving, which suggests that precautionary motives are strong in household saving behavior. 12 The results are not reported here but are available upon request.

20 Precautionary Saving and Health Insurance 251 Table 4 Regression Analysis Using the SCF: Augmented Models s Liquid Probit Tobit Tobit (2) (3) (4) Household Insured *** (0.077) ** (0.025) ** (0.017) Head Sick (0.072) * (0.022) (0.017) Log(Financial Assets) *** (0.016) *** *** Log(Income) *** (0.007) *** (0.002) Age *** *** *** College Degree (0.059) (0.017) *** (0.013) Less than High School * (0.092) (0.031) (0.020) Married (0.081) (0.023) (0.017) White *** (0.074) ** (0.022) (0.017) Female (0.082) (0.026) ** (0.018) Family Number (0.025) Life Expectancy * (0.002) Planning Horizon Next Year (0.101) Next Few Years (0.082) Next 5 Years * (0.085) More Than 10 Years ** (0.098) Risk Attitude Above Average Risk (0.125) Average Risk (0.119) No Risk *** (0.123) Bequests Motives Yes * (0.064) (0.007) (0.032) (0.025) *** (0.026) ** (0.027) (0.031) *** (0.030) *** (0.034) *** (0.020) (0.005) (0.023) (0.019) (0.019) (0.022) (0.025) (0.023) (0.027) *** (0.014) (To be continued)

21 252 Jiaping Qiu s (Continued) Liquid Probit Tobit Tobit (2) (3) (4) Possibly (0.070) (0.021) (0.015) Shopping Habits Degree (0.105) ** (0.030) ** (0.022) Degree 3: Moderate Shopping ** (0.071) * (0.021) *** (0.015) Degree ** (0.100) ** (0.025) (0.020) Degree 5: A Great Deal of Shopping ** (0.081) (0.023) (0.017) Constant *** (0.300) *** (0.083) *** (0.059) N R-squared Note: This table analyzes the effect of health insurance status on the households portfolio choice with additional control variables, using the information from the SCF 1998 database. The additional control variables include life expectancy, planning horizon, risk attitude, bequest motives and shopping habits. Column 2 reports the impact of health insurance status on household stock ownership using the probit model. Column 3 reports the impact of health insurance status on the share of stocks of financial assets using the tobit model. Column 4 reports the impact of health insurance on the share of liquid assets in financial assets using the tobit model. Asymptotic standard errors are reported in parentheses. Numbers with * are significant at the 10-percent level. Numbers with ** are significant at the 5-percent level. Numbers with *** are significant at the 1-percent level. N is the number of observations. 5 Evidence From Health and Retirement Survey 5.1 Basic Results The results from the SCF indicate a strong relationship between household health insurance status and portfolio choice. But one should be cautious about the interpretation of the results due to the cross-sectional future of the SCF data. Health insurance status might also be a choice variable for households and determined by the process that generates their portfolio choice decisions. Therefore, the health insurance variable may be correlated with the residues in the above regressions and may bias the results. To mitigate the endogenous

22 Precautionary Saving and Health Insurance 253 concern, I employ a panel database, the Health and Retirement Survey, to address and mitigate endogeneity concerns. Although the information on household assets in the HRS is not as detailed as in SCF, it traces the same set of households over time which allows for the employment of panel methodology to control households' unobservable characteristics. The data that I use are the first five waves of HRS which were conducted in years 1992, 1994, 1996, 1998 and Although the data are a representative of 51 to 69 year-old households, it might be preferable to focus on the elderly since health insurance could be more important for this group of people, given they are generally less healthy than the youth. As shown in Wu and Rosen (2), the data include detailed information on a variety of economic and demographic variables. In each wave of the HRS, respondents were asked about their health insurance status. Responds are characterized as insured if they have public insurance (such as Medicare or Medicaid), and employer-sponsored or have individually purchased insurance. Again, I examine the impact of health insurance on household financial assets allocation. s include directly held stocks and mutual funds. 13 Liquid Assets include checking and saving accounts, CDs, money market funds and T-bill and bonds. The definitions of other household characteristics are similar to those in SCF. 14 Columns 2 to 4 in Table 5 show the health insurance effect on portfolio choice using random effect probit and tobit models. The results from the HRS are very similar to those from the SCF in the stock ownership and stock share equations. Wealthier, higher-income, better educated and white households are more likely to own stocks and to invest more financial assets in stocks. The magnitude of the estimated coefficients of health insurance in ownership and stock share equations are close to those estimated from the SCF data ( vs and vs in the probit and tobit models respectively). But, the coefficient on the liquate asset share is much smaller than the one estimated from the SCF ( vs ) and only significant at the 5% level in the HRS data. The marginal effects calculated from the probit coefficients imply that being insured increases 13 HRS does not provide information on stock investment through retirement accounts. One way to get around this problem is to allocate retirement assets into different types of assets by using information for SCF. This approach, however, might add much more noise to the measure of asset shares. 14 Information on family size is not clear in HRS, so it is not included in regression

23 254 Jiaping Qiu the probability of owning stocks by 1.5 percentage points. The marginal effects calculated from the tobit coefficients indicate that insured households are associated with an increase of 1.3 percentage held in stocks. These results show that the impact of health insurance on portfolio choice is statistically and economically significant as well. Further, the relationship between health insurance and household portfolio choice remains strong even after controlling for the constant household unobservable characteristics. Table 5 Impact of Health Insurance on Household Portfolio Choice: Evidence from HRS All Households Ownership (2) (3) Liquid Assets (4) Household Insured *** (0.048) *** (0.012) ** Head Sick (0.047) (0.012) Log(Financial Assets) *** (0.012) *** *** Log(Income) * ** *** (0.0002) Age *** (0.007) *** (0.002) *** (0.0005) College Degree *** (0.056) *** (0.014) *** Less than High School *** (0.057) *** (0.0147) *** Married *** (0.049) ** (0.012) ** White *** (0.0731) *** (0.019) *** (0.005) Female (0.039) (0.009) Control for Year Effect Yes Yes Yes N 26,876 22,874 22,874 Note: This table analyzes the effect of health insurance status on households portfolio choice, using the information from the first four waves of the HRS. Column 2 reports the impact of health insurance status on household stock ownership using the random effect probit model. Column 3 reports the impact of health insurance status on the share of stocks of financial assets, using the random effect tobit model. Column 4 reports the impact of health insurance on the share of liquid assets of financial assets, using the random effect tobit model. Standard errors are in parentheses. Numbers with * are significant at the 10-percent level. Numbers with ** are significant at the 5-percent level. Numbers with *** are significant at the 1-percent level. N is the number of observations.