Definitions and data source

472 J. Tin International Review of Economics and Finance 8 1999 467–478

3. Definitions and data source

With the exception of the price level, all data used in this study are obtained from waves 4 and 7 of the 1993 panel of the SIPP conducted by the U.S. Bureau of the Census. Data in wave 4 of the 1993 SIPP panel were collected by interviewers from February to May of 1994 and contains the information on the lagged variables. Data in wave 7 were gathered from February to May of 1995 and contain current information on socio-economic characteristics of the individuals. The price level is represented by the Consumer Price Index CPI published by the Bureau of Labor Statistics and is used to deflate all nominal quantities in the model. SIPP is a longitudinal survey that collects monthly data on income and program participation of individuals in the U.S. The 1993 SIPP panel contains 10 waves of data on about 52,000 persons and was conducted by interviewers from February 1993 to May 1996. It covers the civilian noninstitutional population of the U.S. and members of the Armed Forces living off post or with their families on post. The primary focus of SIPP is persons 15 years of age and over. Data on financial assets were collected by interviewers from individuals who own financial assets in their own names or jointly with spouses. 2 Financial assets covered in the survey include 1 regular or passbook savings accounts in banks, savings and loan, or credit unions; 2 money market deposit accounts; 3 certificates of deposit or other savings certificates; 4 NOW or Super-NOW accounts or other interest- earning checking accounts; 5 money market funds; 6 U.S. government securities; 7 municipal or corporate bonds; 8 other interest-earning assets; 9 stocks and mutual fund shares; and 10 non–interest-earning checking accounts. In this article, the amount of wealth is defined as the sum of current income, current values of financial and nonfinancial assets estimated by an individual in his or her name or jointly owned with the spouse, current values of homes, amount due from sale of business or property, IRA, Keogh, U.S. savings bond, and proceeds from sales of motor vehicles. A detailed list of the sources of income for each individual is given in the Appendix. The risk-adjusted user cost of an asset is estimated in the following manner. The nominal rate of return on each financial asset is obtained from the SIPP by dividing the amount of return by the total amount of asset reported by the respondents. Since the price level at time t 1 1 is not observable at time t, the expected relative prices, P t P t 1 1 , is approximated by P t 2 1 P t , which is constant across individuals in a cross- sectional analysis. Inserting these values into Eq. 3 yields estimates of the real actual rate of excess return on each financial asset, including the benchmark asset. Estimates of the risk-premium adjustment can be calculated from the definition frequently used in conventional consumption-based capital asset pricing model Mankiw Shapiro, 1986; Breeden et al., 1989 [Eq. 12]: φ t 5 1 varc t Covr t ,c t 12 where c t is the expected growth rate of consumption, which is approximated by the J. Tin International Review of Economics and Finance 8 1999 467–478 473 actual growth rate in this study. Regressing the actual rate of excess return on the risk-premium adjustment, φ t , yields predicted values that can be used as the expected rate of excess return in Eq. 2 to obtain estimates of the risk-adjusted user cost for each financial asset. 3 Since the benchmark asset and the number of cross prices may vary among individu- als with different utility functions and wealth constraints, cross prices are omitted from analysis in this study. To include a cross price in a regression, an individual must have at least three types of financial assets with different expected rates of return in the portfolio, and those who possess only two types of financial assets can no longer be included in analysis. This large decline in sample size as a result of using cross prices may bias the coefficient estimates, especially when a large number of cross prices is included in the regression. In addition to wealth and risk-adjusted user cost, age, education, race African-American 5 1, 0 otherwise, and gender female 5 1, male 5 0 are also included in each regression to examine if demographic variables have any impact on asset demand. All regression results are obtained with maximum likelihood under the assumption that the error term possesses the first-order serial correlation, Eq. 13, e t 5 r e t 2 1 1 e t 13 where r is the first-order serial correlation coefficient and e t is an error term with 0 mean and constant variance.

4. Empirical results