Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Paper submitted to the Academy of Financial Services conference for presentation at the October, 2007 national conference. Racial/Ethnic Disparities in Risky Asset Ownership: A Decomposition Analysis1 Sherman D. Hanna and Cong Wang, Ohio State University In 2004, only 27% of Hispanic households and 33% of Black households had any risky, high return investments, compared to 64% of White households. This paper examines the disparities of risky asset ownership between Whites and Blacks and between Whites and Hispanics, using the 2004 Survey of Consumer Finances. Logistic models show that differences in risk tolerance levels and household characteristics contribute to the differences in risky asset ownership between households with White respondents and those with Black and Hispanic respondents. This study presents the first application of the Oaxaca decomposition technique to racial/ethnic differences in risky asset ownership, and provides estimates of the relative importance of these factors in accounting for the gaps. Differences in net worth, income, risk tolerance, and education account for a large part of these disparities. Introduction A number of studies have examined households’ disparities in wealth distribution different race/ethnic groups in the U.S., with most studies examining the gaps between Blacks and Whites (Altonji, Ulrich, & Segal, 2000; Barsky, Bound, Kervin, & Lupton, 2002; Blau & Graham, 1990; Smith, 1995; Sharmila, 2002; Wolff, 1998). Investment in risky, high return assets is an important factor in future economic well-being of households, especially in terms of potential retirement adequacy. White households have much higher stock investment ownership rates than minority groups, even after controlling for income and other factors (Zhong & Xiao, 1 This publication was made possible by a generous grant from the NASD Investor Education Foundation. 1 1995; Schooley & Worden, 1996; Wang & Hanna, 2006). However, little is known about to what extent the racial/ethnic wealth gaps are due to differences in income, and household situation, versus differences in risk tolerance levels. In addition to stock investment ownership, we also include investment real estate and business ownership in the category of owning risky, high return assets. Unlike stock investments, it is difficult to estimate the rate of return and standard deviation of investment real estate, though during the period of 1975 to 2005, publicly traded Real Estate Investment trusts are estimated to have had slightly higher rates of return than the S&P 500 stock index (National Association of Real Estate Investment Trusts, 2007). Business ownership might plausibly assumed to have a higher expected return than stocks of publicly traded small companies, with correspondingly higher risk levels (Lai & Hanna, 2004). Bond investments are sometimes assumed to be risky investments, based on having higher mean returns and standard deviations than cash equivalent investments. However, based on the much lower inflation-adjusted mean arithmetic return of long-term corporate bonds, 3.3% versus 9.2% for large company stocks, and much lower standard deviations, 9.8% versus 20.4% for large company stocks (Ibbotson Associates, 2005), it seems reasonable to exclude bond investments from the risky asset category. During the 1926 to 2004 period, bonds were substantially inferior to stock investments for building wealth, as $1 invested as of the beginning of 1926 would by the end of 2004 have turned into an inflation-adjusted value of $239 in a hypothetical large stock fund, $1,222 in a small stock fund, $9 in a corporate bond fund, and $6 in an intermediate government bond fund (Ibbotson Associates, 2005). In this study, we use a technique proposed by Fairlie (2005), which is an extension of Blinder and Oaxaca’s (1973) method, to examine the relative importance of compositional factors in explaining two racial/ethnic differentials in risky asset ownership — between Whites 2 and Blacks and between Whites and Hispanics, and investigate what portion of the observed differences in risky asset can be explained by their household and economic characteristics. Literature Review Overview Households with similar demographic and financial resources might behave quite differently in the allocation of investment portfolios. For financial investments for long-term goals, because of the relationship between risk and return, the portfolio allocation in risky assets will make a substantial difference in the projected accumulation. In the sections below, we first briefly summarize some important literature relevant to the racial differences in terms of stock and other risky asset holdings and wealth accumulation, and then examine what factors affect the choice of holding risky assets. We then explore to what extent those factors contribute the racial disparities between White and Black households and between White and Hispanic households. Investment behavior can be influenced by preferences. Ogden, Ogden and Schau (2004) suggest that subculture, which may be represented by race or ethnicity, might impact preferences. Previous studies have shown that the households who are more willing to take risk are more likely to have stock investments (Wang & Hanna, 2006). Most studies find that Blacks are less willing to take investment risk than Whites (Yao, Gutter, & Hanna, 2005). Based on the SCF investment risk tolerance measure, Sung and Hanna (1996) find no significant differences between Blacks and Whites in willingness to take some investment risk after controlling for other variables, for a sample of employed households in 1989, although they find that Hispanics are significantly less likely than otherwise similar Whites to be willing to take some risks. Yao, Gutter and Hanna (2005) analyze a combination of the 3 1983 to 2001 SCF datasets, and report that Blacks and Hispanics are less willing to take some investment risk than otherwise similar White households, although more willing to take substantial investment risk than Whites. Coleman (2004) analyzes the 1998 SCF models to compare the risk tolerance levels of Whites, Blacks, and Hispanics. In a logistic regression controlling for racial/ethnic group, gender, marital status, education, age, and family size, she finds that Blacks and Hispanics are more likely to be unwilling to take any risk compared to otherwise similar Whites. However, when she also controls for net worth, the predicted net worth difference between Blacks and Whites is not significant. She obtains similar results with a tobit analysis of the risky asset proportion, in that Blacks and Hispanics are have a significantly lower risky asset proportion than otherwise similar Whites when net worth is not controlled, but when net worth controlled, the predicted difference between Blacks and Whites is not significant. A consumer’s access to information and related services in financial markets might also affect its financial behavior. If a consumer can obtain more information and service for financial investment, they probably will be more willing to participate in financial markets. For example, more educated households may be more willing to participate in financial market than less educated people. Haurin and Morrow-Jones (2007) conclude that differences in knowledge of markets might contribute to lower homeownership rates of Black households, so it is plausible that similar factors may contribute to lower risky asset ownership rates. Wealth, Investment and Race The gap between non-Hispanic White households and households with respondents in other racial/ethnic groups narrowed between 1995 and 2001, but then became much wider in 4 2004 (Aizcorbe, Kennickell, & Moore, 2003; Bucks, Kennickell, & Moore, 2006; Kennickell, Starr-McCluer, & Sundén, 1997; Kennickell, Starr-McCluer, & Surette, 2000). Smith (1995) compares the racial and ethnic differences in wealth through 1992 HRS (Health and Retirement Study) and concludes that income is an important reason for racial and ethnic deficits, but that income-conditioned wealth disparities in asset remain large. Differences in risky asset investment ownership may contribute to the wealth differences. Previous researchers all find a racial wealth gap, using a variety of data sets, but have inconsistent results about the influence of various demographic and financial characteristics on the gap. Altonji, et al. (2003) use a decomposition method and find that most or all of the race gap in the wealth level for single men and single women and a substantial portion of the gap for married couples would disappear if Blacks and Whites have the same distribution of income and demographic variables and if the slope coefficients of the White wealth equation hold for Blacks. For instance, they found that single Black men would have 108% of the wealth of single White men if they have the same income and demographics as White men when using the estimated coefficients of the wealth model for Whites. Kaufman (1983) uses a structural decomposition of Black-White earning differentials based on a regression standardization approach and concludes that eliminating all Black-White differences within labor market divisions would still leave a significant earnings gap between Blacks and Whites due to the structure of the labor market. Blau and Graham (1990), using data from the 1976 and 1978 National Longitudinal Surveys of young men and young women (NLSY) to examine the Black-White differences in wealth and asset composition among younger families, find that as much as 75% of racial differences remain unexplained by income and other demographic and locational characteristics. Barsky et al. (2002) use a non-parametric approach to analyze racial differences in earnings and compares the 5 result with the traditional decomposition method. After re-weighting observations to give a sample of Whites a similar earnings distribution as that for Blacks, they found that differences in characteristics explain only 64% of the wealth gap, compared with 97% when the standard (linear) regression decomposition approach is used (with White coefficients). Given the importance of investments in the risky assets in contributing to wealth differences between different racial/ethnic groups, many studies focus on the ownership of risky assets and its relationship with race (Haliassos & Bertaut, 1995; Plath & Stevenson, 2000; Coleman, 2003). For instance, using the 1992 HRS, Choudhury (2002) demonstrates that Whites, Blacks and Hispanics were different in saving behavior, and minority households were notably more disinclined to invest in riskier, higher-yielding financial assets. Gutter, Fox and Montalto (1999) analyze Black/ White racial difference in the likelihood of owing risky assets based on Jackson and Lindley (1989)’s method by using 1995 SCF and conclude that differences in risky asset ownership between Black households and White households are due to racial characteristics in the individual determinants of risky asset ownership, not to race in and of itself. Stevenson and Plath (2006) compare financial service consumption patterns of Hispanics with those of non-Hispanic white, using the 1998 SCF, and find that Hispanics differ markedly from their non-Hispanic White counterparts in terms of financial product preferences and investment asset portfolio composition. Further, Hispanics trail their non-Hispanic White counterparts in terms of breath and depth of financial holdings, particularly in the area of more risky but historically higher return asset categories. An important part of racial/ethnic differences in risky asset holdings is risky asset holdings. While previous researchers' methods vary in analyzing the distribution of wealth, there is agreement that the distribution of wealth in the United States is very unequal and that 6 inequality has worsened in recent decades. Keister and Moller (2000) note that participation in stock and real estate investments markets have very important influences on the unequal distribution of wealth. Wang and Hanna (2006) report that even after controlling for risk tolerance levels and other variables, Blacks and Hispanics are less likely to directly or indirectly hold stock investments than Whites. Hurst, Luoh and Stafford (1998) find that the lower rate of stock ownership among Black families prevent them from benefiting as much as White families from the economic of the 1990s. Therefore, it is important to explore the causes of low risky asset investment ownership by minority households. Gutter and Fontes (2006) found that risky asset ownership was the key to racial differences in portfolio choices, as Black and Hispanic households that owned any risky asset were not significantly different from similar White households in risky asset proportions. Therefore, in this study, we focus on differences in risky asset ownership. Decomposition Method The typical approach to decomposing racial differences in wages, employment, and wealth is to use some extension of what is known as the Blinder-Oaxaca technique, which was first developed by Oaxaca (1973) and Blinder (1973). The Blinder-Oaxaca decomposition technique is widely used to identify and quantify the separate contributions of group differences in measurable characteristics. Fairlie (1999) and Fairlie (2005) note that a linear regression implies that the slope of an racial effect with respect to all other covariates is the same for all races, and only the intercept of the function is shifted up or down. Given this restriction, it is more appropriate to employ non-linear decomposition to give a robust estimate of the contribution of racial effects. Fairlie (1999) and Fairlie (2005) propose an extended non-linear 7 decomposition technique which uses the coefficients directly from a logit or probit model when the outcome is binary. In this study we use a variation of the Blinder-Oaxaca decomposition method as described by Fairlie (2005) to examine how racial/ethnic groups are different in terms of the risky asset ownership and how much of these disparities are due to the observed differences in household characteristics. Oaxaca and Ransom (1994) suggest a way to circumvent the index number problem, whereby the pooled coefficient vector-pooled for the two races being compared-is taken as the non-discriminatory unemployment structure. We illustrate the process of decomposition of the Black-White gap in risky asset ownership as an example of the technique we adopt. In general, the disparity of ownership rate between Blacks and Whites can be expressed as below: ∧ ⎞ ∧ ⎞⎤ ⎡ ∧ ⎞ ∧ ⎞⎤ ⎡ ⎛ ⎛ ⎛ ⎛ B w ⎟⎥ ⎢ B w⎟ B B ⎟⎥ ⎜ ⎜ ⎜ ⎢ F ⎜ X iW β w ⎟ F X β F X β F X β ⎟ NB ⎜ i ⎟⎥ ⎟ NB ⎜ i ⎟ ⎥ ⎢N B ⎜ i ⎢N w ⎜ W B ⎝ ⎠ ⎝ ⎠⎥ ⎝ ⎠ ⎝ ⎠ ⎥+⎢ Y −Y = ⎢ − − B B w B ⎥ ⎢ ⎥ ⎢ i =1 N N N N i =1 i =1 ⎥ ⎥ ⎢ i =1 ⎢ ⎥ ⎥ ⎢ ⎢ ⎦ ⎦ ⎣ ⎣ ∑ Y W ∑ ∑ ∑ B and Y represents the average predicted probability of ownership for White and Black groups respectively. F (.) is the cumulative distribution function from the logistic distribution and N represents the sample size in different groups. X W and X B are row vectors of average value ∧ w ∧ B for the individual characteristics of Whites and Blacks respectively. β and β are the vector of coefficient estimates for Whites and Blacks respectively. For a linear regression, the standard Blinder-Oaxaca decomposition of the White/Black gap in the average of the dependent variable, Y is expressed as: Y W −Y B ∧ ⎤ ⎡ ∧ ⎞⎤ ⎡ W ⎛ ∧ B B = ⎢⎛⎜ X − X ⎞⎟ β w ⎥ + ⎢ X ⎜ β w − β B ⎟⎥ . As Fairlie (2005) illustrates, ⎟⎥ ⎜ ⎠ ⎢⎝ ⎥ ⎢ ⎣ ⎠⎦ ⎦ ⎣ ⎝ this equation is a special case of the equation we employ in our study. 8 The racial gap in this respect can be divided into two parts: one part of explained difference (expressed as the first term in the equation) due to the differences in household characteristics we include in the model, and another part of unexplained difference due to the inability to include unmeasureable variables (expressed as the second term in the equation). The contribution of each variable to the gap is equal to the change in the average predicted probability from replacing the Black distribution with the White distribution of that variable while holding the distributions of the other variable constant. Given that our White sample size is much larger than that of Hispanic and Black samples, we implement the average prediction decomposition method by following the suggestion of Fairlie (2005) to randomly select a White sample to match the Black sample size for the analysis. The Blacks and sampled Whites are then ranked by race and matched based on their predicted stockownership outcomes. This sampling process is repeated 1,000 times, and calculated statistics that are computed with the alternative weights are averaged. In this way, we can largely avoid selection bias from sample differences between different racial groups. In our study, the sample weights are used to estimate the mean outcomes but not the logit regressions. There is no reason for us to prefer the Black or White estimates in this equation, therefore, we follow the common approach of presenting both sets of estimates and also added another set of overall sample estimates as proposed by Oaxaca and Ransom (1994). In this way, we can test the sensitivity and consistency of decomposition specifications from estimates. Therefore, in our study models are estimated separately—overall sample with racial/ethnic controlled, White sample only, Black sample only, and Hispanic sample only, as shown in Table 1. These estimates include the same explanatory variables, with the exception that the race/ethnic coefficients are dropped when they are applied in decomposition specification (1) as shown in 9 Table 4 and Table 5. By incorporating the coefficients from logistic results, decomposition results can show the relative contribution to differences in risky asset ownership from household characteristics such as age, education, and health status. Methods Data and Variables We use the 2004 Survey of Consumer Finances (SCF) to study the White-Black and White-Hispanic gap in risky asset ownership because the SCF is the best source of information on the wealth or financial assets holdings and characteristics of American households (Bucks, et al., 2006). The SCF is a triennial interview survey of U.S families sponsored by the Board of Governors of the Federal Reserve System with the cooperation of the U.S. Department of the Treasury. The race and ethnicity of a household in the SCF are classified according to the respondent to the SCF interview. The surveys before 2004 include one race/ethnic question, with White and Hispanic presented as different categories (Yao, et al., 2005). A few respondents also indicate a second category of racial/ethnic identity, but in the public use datasets before 2004, it is impossible to identify respondents who choose Hispanic as a second category. Starting from the 2004 SCF, there is now a separate question about Hispanic status: whether respondents consider themselves to be Hispanic or Latino in culture or origin. Combining that answer with the primary question gives a higher proportion of Hispanics. However, the results are similar when using the one question racial/ethnic variable, so for consistency with research using previous SCF datasets, we use the one question categorization. The 2004 SCF survey includes 4,519 households, with five implicates for each household (Bucks, et al., 2006). The coding for the race/ethnic status variable (X6809) is different across 10 implicates for 13 households, so those households are excluded from the analyses. Of the remaining 4,506 households, there are 3,511 households with White respondents, 482 with Black respondents, 347 with Hispanic respondents, and 166 with respondents choosing some other racial/ethnic group. (The SCF does not provide detailed breakdowns of this last group in the public dataset.) In analyses of all households, we include the “other” households, but do not present separate analyses. Using the SCF population weight, 73.6% of the households have White respondents, 13.6% have Black respondents, 9.2% have Hispanic respondents, and 3.7% have “other” respondents. We base our multivariate analyses on previous research and on the idea that a household’s decision to hold risky asset investments is related to its risk tolerance, its investment horizon, and its desire to invest for future goals, especially retirement. The appropriateness of risky asset investments for savings goals is strongly related to the investment horizon, given the volatility of risky assets compared to alternate investments such as cash equivalents and shorter term government bonds. Campbell and Viceira (2002) show that the optimal stock proportion of a portfolio should be related to age, for each level of risk aversion. Lower income households may decide to not save much for retirement because of the higher replacement rate for Social Security pensions, so putting funds in any higher return investment should be strongly related to household income. Liquidity concerns may lead some households with low net worth to avoid putting money in high return investments. The focus on much research and normative analysis in finance is on the ratio of risky assets to the total portfolio (e.g., Coleman, 2005). However, most Black and Hispanic households hold no risky assets, so a direct comparison of the risky asset ratio of minority households with White households does not provide much insight, because most of the racial/ethnic differences are due to differences in ownership of risky assets. 11 Our basic models for the multivariate analysis are: For the overall sample: Choice of holding any risky investments = f (race/ethnic group, X) (1) Where X is a vector of household characteristics and risk tolerance. For each of the individual subsamples (White, Black and Hispanic): Choice of holding any risky asset investments = f (X) (2) Where X is a vector of household characteristics and risk tolerance. The dependent variable is dichotomous, and equals 1 if the household holds any risky asset investments, including in mutual funds inside of defined contribution retirement accounts. The explanatory variables in this study are age, age squared, household respondents’ education level, health situation, risk tolerance, household income, net worth, and presence of children age under 19, homeownership, business ownership and gender of respondents (Appendix 1.) Net worth might be endogenous, since over time, households who choose risky assets will tend to accumulate more net worth. Our logistic regression and decomposition analysis results are similar to those presented in this paper when net worth is not included, so if there is an endogeneity bias from using net worth as an independent variable, it does not seem to be large. Analysis Figure 1 shows the aggregate allocation of risky assets. Stock investments, including stocks in mutual funds, account for 38% of all risky assets, business investments account for 37%, and investment real estate assets account for 25%. Figure 2 shows the allocation for 12 households with a total amount of risky assets less than one million dollars. Stock investments account for 52% of all risky assets, business investments account for 19%, and investment real estate assets account for 29%. Households with White respondents are older, more educated, in better health, have higher income, have higher net worth, and are less likely to say they are unwilling to take any risk with investments than are households with Black or Hispanic respondents (Table 1). Only 27% of Hispanic households and 33% of Black households have any type of risky asset investments, compared to 64% of White households (Table 2). The patterns are similar for the new racial/ethnic categories taking into account the separate Hispanic question, so for comparability with previous SCF datasets, we use the older categories. Figure 3 shows the actual pattern of risky asset ownership, as well as the predicted pattern, discussed below. The dependent variable in our model is an indicator of whether or not households hold risky directly and/or indirectly. The explanatory variables included in the study are age, age squared, household heads’ education level, health situation, risk tolerance, household income, net worth, and presence of children aged under 19, homeownership, business ownership and gender of respondents. Decomposition Specifications of Racial Disparities in Risky Asset Ownership The logit empirical analyses (Table 3) show that many household characteristics are related to risky asset investment ownership, and there are some consistent significant factors across three kinds of samples. The logit for the overall sample had a concordance ratio of 92.5%, meaning that the logit correctly classifies over 92% of the households in terms of risky asset investment ownership. Even after controlling for income, net worth, education, risk tolerance, 13 and other variables, Blacks, Hispanics, and those in one of the other categories (largely Asian) have much lower predicted risky asset investment ownership rates than White households. Figure 3 shows the actual and predicted risky asset rates. The gaps between White households and Black, Hispanic households narrow after controlling for the independent variables, though there still substantial differences. In the following section, we examine the fraction of the gaps that can be explained by those characteristics. Black-White Disparity. The difference between White and Black risky asset rates is 31 percentage points. Table 4 shows the decomposition analyses for White-Black differences. Household net worth is the biggest contributor to the racial disparity in risky asset ownership, accounting for 50% of the differences in the full sample, 48% in the Black sample, and 50% in the White sample. Risk tolerance contributes about 19% of the Black-White disparity in estimates based on three samples. Black risky asset investment ownership would be much higher if Black households had the same net worth or the same risk tolerance as White households. The other important contributors are household income and education. Our result indicates that the gap between Black and White risky asset investment ownership is almost completely determined by differences in education, income, net worth, risk tolerance, and other household characteristics. The overall explained difference is 106.5% for the full sample, which implies that if Black households had the same characteristics and risk tolerance as White households, they would have higher risky asset ownership than White households. Figure 4 shows the relative contributions of factors that are related to lower Black risky asset investment ownership, based on the analysis of the overall sample shown in Table 4. 14 Hispanic-White Disparity. The Hispanic-White decomposition shows a pattern similar to the Black-White decomposition. The difference between Hispanic and White households in risky asset investment ownership rates is 37 percentage points. Net worth, risk tolerance, education and income are the most important contributors to the gap. Risk tolerance’s contribution is very close in three specifications and education is the next important contributor, based on the overall sample. Figure 5 shows the relative contributions of factors that are related to lower Hispanic risky asset investment ownership, based on the analysis of the overall sample shown in Table 5. If Hispanic households had the same wealth, risk tolerance, income, education, and other characteristics as White households, their risky asset ownership would be almost as high as White households based on the Hispanic only estimates and close to the White rate based on the full sample and the White-only sample. Conclusions Differences in demographic and economic characteristics explain virtually all of the total differential between the Black and White risky asset investment ownership, and explain more than 80% of the total differential between the Hispanic and White risky asset. Household economic characteristics, net worth and income play the most significant role in causing the racial gap in risky asset investment ownership between both Black-White and Hispanic-White disparity. Obviously, improving those conditions of minority groups may improve their situation in risky asset ownership, but risk tolerance also plays an important role in explaining both sets of racial/ethnic disparities in risky asset ownership. Therefore, education targeted at increasing the risk tolerance of minorities should increase their likelihood of owning risky assets. 15 References Altonji, J. G., Ulrich, D., & Segal, L. (2000). Black/White differences in wealth. Economic Perspectives, 38-50. Barsky, R., Bound, J., Charles, K., & Lupton, J. (2002). Accounting for the Black-White wealth gap: A nonparametric approach. Journal of the American Statistical Association, 97 (459), 663-673. Blau, F. D. & Graham, J. W. (1990). Black-White difference in wealth and asset composition. Quarterly Journal of Economics, 105(2), 321-339. Bucks, B. K., Kennickell, A. B., & Moore, K. B. (2006). Recent changes in U.S. family finances: Evidence from the 2001 and 2004 Survey of Consumer Finances. Federal Reserve Bulletin, 92, 1-38. Coleman, S. (2003). Risk tolerance and the investment behavior of Black and Hispanic heads of household. Financial Counseling and Planning, 14(2) 43-52. Deaton, A. (1997). The analysis of household surveys: A microeconometric approach to development policy. Baltimore, MD: Johns Hopkins University Press. Fairlie, R. W. (2005). An extension of the Blinder-Oaxaca decomposition technique to logit and probit models. Journal of Economic and Social Measurement, 30 (4), 305-316. Racial Differences in Risky Asset Ownership: A Two-Stage Model of the Investment Decision-Making Process Gutter, M. S. &Fontes, A. (2006). Racial differences in risky asset ownership: A twostage model of the investment decision-making process. Financial Counseling and Planning, 17(2) 64-78. Gutter, M. S., Fox, J. J., & Montalto, C. P. (1999). Racial differences in investor decision making. Financial Services Review, 8, 149-162. Ibbotson Associates (2005). Yearbook of stocks, bonds, bills, and inflation, 2005. Lai, C. W. & Hanna, S. D. (2004). Are the portfolios of older worker households efficient? Journal of Personal Finance, 3(4), 101-117. Montalto, C. P. & Sung, J. (1996). Multiple imputation in the 1992 Survey of Consumer Finances. Financial Counseling and Planning, 7, 133-146. National Association of Real Estate Investment Trusts (2007). Why Invest in REITs? http://www.investinreits.com/reasons/performance.cfm, retrieved March 16, 2007. Oaxaca, R. (1973). Male-female wage differentials in urban labor markets. International Economic Review, 14 (3), 693-709. Oaxaca, R. & Ransom, M. (1994). On discrimination and the decomposition of wage differentials. Journal of Econometrics, 61, 5-21. Schooley, D. K. & Worden, D. D. (1996). Risk aversion measures: Comparing attitudes and asset allocation. Financial Services Review, 5(2), 87-99. Sharmila, C. (2002). Racial and ethnic differences in wealth and asset choices. Social Security Bulletin, 64 (4), 1-16. Smith, J. P. (1995), Racial and ethnic differences in wealth in the Health and Retirement Study. Journal of Human Resources, 30, Supplement S158-S183. Wang, C. & Hanna, S. D. (2006). The risk tolerance and stock-ownership of businessowning households. Consumer Interests Annual, 52. Wolff, E. N. (1998), Recent trends in the size distribution of household wealth. Journal of Economic Perspectives, 12 (3), 131-150. 16 Zhong, L. X. & Xiao, J. J. (1995). Determinants of family bond and stock holdings. Financial Counseling and Planning, 6, 107-114. 17 Table 1 Means and Proportions of Household Characteristics by Racial/Ethnic Groups, 2004 Variable White Black Hispanic Own stocks directly and/or indirectly 0.57 0.26 0.19 Own investment real estate 0.20 0.12 0.12 Own business assets 0.14 0.05 0.04 Own one or more risky assets 0.64 0.33 0.27 Age 50.39 47.2 41.82 Income ($) 76,958 37,781 38,395 Net worth ($) 553,363 109,718 126,129 Education of respondent Less than high school degree 0.17 0.26 0.50 High school degree 0.20 0.24 0.20 >12 years education without degree 0.22 0.25 0.18 2 years degree 0.06 0.06 0.02 Bachelor degree 0.21 0.12 0.06 Post bachelor degree 0.13 0.08 0.03 Health of respondent Poor health 0.06 0.10 0.09 Fair health 0.16 0.21 0.28 Good health 0.48 0.43 0.41 Excellent health 0.30 0.27 0.23 Risk tolerance level Not willing to take risk 0.37 0.57 0.65 Average risk tolerance 0.43 0.28 0.20 Above average risk tolerance 0.18 0.09 0.10 Substantial risk tolerance 0.03 0.05 0.05 Couple household 0.55 0.26 0.50 Own home 0.76 0.50 0.48 Presence of child aged under 19 at home 0.40 0.48 0.62 Female respondent 0.54 0.68 0.52 Actual (unweighted) number of households 3,511 482 347 Weighted percent of sample 73.61% 13.56% 9.18% Calculated by authors, weighted analysis of the 2004 SCF, with 13 households deleted that had different responses to racial/ethnic group in different implicates. Households with respondents coded as “Other racial/ethnic group” are included in the overall analyses but the results are not presented here. 18 Table 2 Risky Asset Ownership by Race/Ethnic Category, 2004 Percent Distribution Original SCF Race/Ethnic Categorization White 73.60% Black 13.56% Hispanic 9.17% Other, including Asian 3.67% New SCF Race/Ethnic Categorization White Non-Hispanic 71.82% Black Non-Hispanic 13.41% Hispanic 11.17% Other, including Asian 3.61% Own Stocks Directly and/or Indirectly Own Any Risky Asset 56.69%BC 25.51%A_ 18.65%AD 45.92%C_ 63.66%BC_ 32.63%AD_ 26.74%AD_ 60.55%BC_ 57.32% BC_ 25.62%A__ 20.53% AD 46.66%C__ 64.33%BC 32.67%AD 28.21%AD 61.52%BC Based on authors’ calculations, weighted analyses of 2004 Survey of Consumer Finances datasets, using RII technique with all implicates. A Significantly different from White at 0.05 level or better B Significantly different from Black at 0.05 level or better C Significantly different from Hispanic at 0.05 level or better D Significantly different from Asian/Other at 0.05 level or better 19 Table 3 Logistic results of racial difference in risky asset ownership in different samples Samples Full Sample Black Parameters coefficient p-value coefficient Intercept -8.6597 .000 -7.1268 Racial/ethnic groups: reference category= White Black -0.6184 .000 Hispanic -0.6684 .000 Asian/other groups -0.1709 .4726 Age of respondent/100 7.6228 .000 10.1196 Age/100 squared -7.0924 .000 -9.3816 Education of respondent: reference category=less than high school High school degree 0.3991 .0082 0.2034 >12 years but no degree 0.8065 .000 1.1657 2 year degree 0.8713 .000 1.4644 Bachelor degree 1.3638 .000 0.7319 Post BS 1.5029 .000 2.0730 Health condition of the respondent: reference category=poor health Excellent health 0.6568 .0040 0.2404 Good health 0.8182 .0002 0.7488 Fair health 0.1866 .4193 -0.2345 Expect to inherit wealth 0.3187 .0310 0.4775 Risk tolerance: reference category=not willing to take any risk Average 1.3170 .000 1.2835 Above average 1.7521 .000 1.1265 Substantial 1.7828 .000 1.0763 Log of income 0.2656 .000 0.0612 Log of net worth 0.2152 .000 0.1910 Couple (versus noncouple) 0.3568 .0009 0.0812 Homeowner (versus rent) 0.2469 .0429 0.1163 Child<19 -0.2838 .0085 -0.1335 Gender of respondent: reference category=male respondent Female respondent -0.0470 .6462 -0.1247 Likelihood ratio 2749.6882 .000 207.4501 Concordance 92.5 86.6 p-value .000 White coefficient p-value - 8.7103 .000 Hispanic coefficient p-value -21.5262 .000 .0480 .0701 7.2218 -6.6944 .000 .000 4.8335 -4.7065 .5736 .6123 .6319 0.3507 .0528 0.8626 .0889 .0038 .0124 .1207 .0004 0. 6726 0. 7092 1.5795 1.4273 .000 .0057 .000 .000 1.9793 1.1491 1.4726 0.6649 .000 .3784 .0358 .5074 .6808 .1715 .6983 .3166 0.6743 0.7865 0.1963 0.2994 .0108 .0017 .4631 .0675 0.7256 1.2158 1.0717 0.2352 .4933 .2285 .2978 .8018 .000 .0062 .0519 .5053 .000 1.2981 1.9475 2.1087 0.2629 0.2260 .000 .000 .000 .000 .000 1.3824 1.3146 1.5843 1.5074 0.1457 .0015 .0209 .0148 .000 .0133 .7880 0.3696 .0035 0.5641 .1802 .7031 .6253 0.2879 -0.2758 .0529 .0333 0. 5162 -1.0112 .2728 .0182 .6465 .000 0.00684 1806.6933 92.1 .9555 .000 -0.1952 221.7992 93.3 .6318 .000 20 Table 4: Decompositions of risky asset ownership of Black versus White Households Specifications based on the coefficients of three types of samples (1) (2) (3) Full sample Black White White risky asset ownership 0.6370831 0.6370831 0.6370831 Black risky asset ownership 0.3266508 0.3266508 0.3266508 Black-White gap 0.3104322 0.3104322 0.3104322 Contributions of individual variable age 0.0027733 0.0082362 0.0027766 percentage 0.89% 2.65% 0.89% Standard error 0.0018351 0.0093460 0.0021691 Education of respondent percentage Standard error 0.0276147 8.90% 0.0029920 0.0447952 14.43% 0.0137480 0.0291764 9.40% 0.0034564 Health condition of respondent percentage Standard error 0.0111366 3.59% 0.0023057 0.0090538 2.92% 0.0073665 0.0110680 3.57% 0.0027328 Expectation of inheritance percentage Standard error 0.0035342 1.14% 0.0015886 0.0065488 2.11% 0.0062766 0.0031479 1.01% 0.0016731 Risk tolerance percentage Standard error 0.0576447 18.57% 0.0043576 0.0576961 18.59% 0.0134937 0.0592377 19.08% 0.0051424 Income percentage Standard error 0.0349686 11.26% 0.0048381 0.0121048 3.90% 0.0179528 0.0335846 10.82% 0.0051597 Net worth percentage Standard error 0.1540481 49.62% 0.0087262 0.1487488 47.92% 0.0219598 0.1554906 50.09% 0.0100102 Couple households percentage Standard error 0.0198953 6.41% 0.0060243 0.0045715 1.47% 0.0170822 0.0203280 6.55% 0.0070221 Home ownership percentage 0.0149084 4.80% 0.0066333 2.14% 0.0170365 5.5% 21 Specifications based on the coefficients of three types of samples (1) (2) (3) Full sample Black White Standard error 0.0074987 0.0175929 0.0089652 Presence of child aged under 19 percentage Standard error 0.0027312 0.88% 0.0010406 0.0013111 0.42% 0.0027014 0.0026247 0.85% 0.0012289 Female respondent percentage Standard error 0.0015048 0.48% 0.0032787 0.0045265 1.46% 0.0098935 -0.000215550 -0.07% 0.0038631 Overall explained difference percentage 0.3307599 106.54% 0.3042261 98.01% 0.3342555 107.63% Unexplained difference percentage -0.0203277 -6.54% -0.0062061 -0.0199% -0.0238233 -7.63% 22 Table 5: Decompositions of risky asset ownership of Hispanic versus White Households Specifications based on the coefficients of three types of samples (1) (2) (3) Full sample Hispanic White White risky asset ownership 0.6370831 0.6370831 0.6370831 Hispanic risky asset ownership 0.2676478 0.2676478 0.2676478 Hispanic-White gap 0.3694353 0.3694353 0.3694353 Contributions of individual variable age 0.0042616 0.000473932 0.0043972 percentage 1.15% 0.13% 1.19% Standard error 0.0037545 0.0162458 0.0044477 Education of head percentage Standard error 0.0526597 14.25% 0.0065400 0.0610312 16.52% 0.0198262 0.0510064 13.81% 0.0075833 Health condition of respondent percentage Standard error 0.0138003 3.74% 0.0029418 -0.000313840 -0.08% 0.0074368 0.0123291 3.34% 0.0032969 Expectation of inheritance percentage Standard error 0.0042915 1.16% 0.0019285 0.0024777 0.67% 0.0097057 0.0037292 1.01% 0.0019750 Risk tolerance percentage Standard error 0.0727156 19.68% 0.0052961 0.0543840 14.72% 0.0160104 0.0750290 20.31% 0.0062484 Income percentage Standard error 0.0287046 7.77% 0.0041315 0.1230444 33.31% 0.0251098 0.0285574 7.73% 0.0044918 Net worth percentage Standard error 0.1108654 30.01% 0.0068908 0.0568416 15.39% 0.0211263 0.1124444 30.44% 0.0079513 Couple households percentage Standard error 0.0062095 1.68% 0.0018733 0.0071903 1.95% 0.0053581 0.0065067 1.76% 0.0022014 Home ownership percentage 0.0114741 3.11% 0.0173686 4.70% 0.0129651 3.51% 23 Specifications based on the coefficients of three types of samples (1) (2) (3) Full sample Hispanic White Standard error 0.0057310 0.0162501 0.0067769 Presence of child aged under 19 percentage Standard error 0.0108484 2.94% 0.0041112 0.0262617 7.11% 0.0114747 0.0104848 2.84% 0.0049212 Female respondent percentage Standard error 0.000488929 0.13% 0.0010657 0.0020071 0.54% 0.0042118 -0.000069568 -0.02% 0.0012464 Overall explained difference percentage 0.316319629 85.62% 0.350766692 94.96% 0.31737932 85.92% Unexplained difference percentage 0.05311567 14.38% 0.018668608 0.04% 0.052055568 14.08% 24 Figure 1 Risky Investment Assets, Household Allocation, All Households, 2004 SCF. real estate other than primary residence, 25% (24.94%) stock, 38% (37.88%) business ownership, 37% (37.18%) 25 Figure 2 Risky Investment Assets, Household Allocation, Among Households with Risky Assets < $1,000,000, 2004 SCF real estate other than primary residence, 29% (29.02%) stock, 52% (51.92%) (19.06%) business ownership, 19% 26 Figure 3 Risky Ownership Rates for Blacks, Whites, Hispanics, and Others, 2004 Survey of Consumer Finances. 70% Risky Investment Ownership Rate 60.5% 60% 50% 59.6% 63.7% 55.5% 44.3% 43.1% 40% 32.7% 26.8% 30% 20% 10% 0% Black Hispanic Other Race/Ethnic Group of Respondent predicted White actual Calculated by authors, weighted analysis of the 2004 Survey of Consumer Finances. Predicted rates are calculated from the logistic regression results for the full sample in Table 3, at the mean values of other variables, and adjusted so that at the mean of all variables, the predicted ownership rate equals the overall sample ownership rate. 27 Figure 4 Decomposition of Black-White Risky Asset Ownership Differences in 2004 Based on Overall Sample. Factors Related to Lower Risky Ownership Rate of Blacks. expect inheritance, 1% female, inher income, 11% Income female child<19, 1% Child age, 1% Age health, Health couple, 6% Couple net worth, 50% NW own home, 5% Own Hm education,Educ 9% risk tolerance,Risk 19%Tol Calculated by authors, based on Table 4, Overall Sample 28 Figure 5 Decomposition of Hispanic-White Risky Asset Ownership Differences Based on Overall Sample. Factors Related to Lower Risky Asset Ownership Rate of Hispanics. income, 8% Income female, inher 0.1% female expect inheritance, 1% child<19, 3% Child age, Age 1% net worth, 30% health, Health NW couple, 2% Couple own home, 3% Own Hm education, Educ 14% Risk Tol risk tolerance, 20% Calculated by authors, based on Table 5, Overall Sample 29 Appendix 1: Description of Explanatory Variables Variables Description Demographic variables Age Age of respondent Age-squared Squared age of respondent Race/ethnic self-identification of the respondent (based on variable X6809) White Black Hispanic Other (including Asian) Respondent’s education Less than high school 1 if years of education <12, 0 otherwise High school 1 if years of education =12, 0 otherwise More than 12 years’ education without degree 1 if years of education>12 and no degree 2 year degree 1 if get a 2 year’s degree Bachelor degree 1 if get a bachelor degree Post bachelor degree 1 if get a degree higher than bachelor Respondent self-assessment of health Poor health 1 if the respondent said she or he had a generally poor health; 0 otherwise Fair health 1 if the respondent said she or he had a generally fair health; 0 otherwise Good health Excellent health Respondent’s investment risk tolerance Not willing Average risk tolerance Above average risk tolerance Substantial risk tolerance Business ownership Home ownership Married couple Female respondent Presence of child aged under 19 Economic variables Log of Income Log of Net worth 1 if the respondent said she or he had a generally good health; 0 otherwise 1 if the respondent said she or he had a generally excellent health; 0 otherwise 1 if not willing to take any risk, 0 otherwise 1 if not willing to take average risk, 0 otherwise 1 if not willing to take above average risk, 0 otherwise 1 if not willing to take substantial risk, 0 otherwise 1 if the respondent said the household owned business, 0 otherwise 1 if the respondent said the household owned home, 0 otherwise 1 if the respondent is married to household member, 0 otherwise 1 if the sex of the respondent is female, 0 otherwise 1 if there is a child aged under 19 living in the household, 0 otherwise Log of household’ annual income (if income ≤ 0, use ln (0.01) Log of household’s net worth (if net worth ≤ 0, use ln (0.01) 30