تقاضا برای دارایی های پر ریسک : انتخاب نمونه و اوراق بهادار خانگی

We estimate a microeconomic model of household asset demands that allows for the fact that households typically have zero holdings of most assets. The adjustments for non-observed heterogeneity generalize methods developed by Dubin and McFadden (1984. Econometrica 52, 345–362). Simulating our model using a random sample of US households, we examine distributional and demographic effects on macroeconomic demands for money, stocks and bonds.

This paper applies discrete-continuous econometric techniques to microeconomic data on household portfolio composition in order to estimate individual asset demands from the 1983 Survey of Consumer Finances (SCF). Since the data set we study includes a random sample of the US population, by simulating our model, we obtain estimates of economy-wide money, stock and bond demand elasticities that fully allow for the non-linearities in asset demand generated by the combination of a discrete choice of asset portfolio with a continuous choice of quantities demanded for given portfolio composition. Early research using cross-sectional, household-level data to investigate portfolio choices includes Uhler and Cragg (1971), Friend and Blume (1975) and King and Leape (1984). Recently, research in this field has become very active and several studies directly complement our own. Ioannides (1992) utilizes the panel data structure of the 1983 and 1986 SCFs, focusing particularly on changes in individual portfolios between these two dates. Hochguertel and van Soest (1996) examine how housing wealth affects total liquid financial wealth,1 and Poterba and Samwick (1997) use several waves of the SCF but focus exclusively on age versus cohort effects. Hochguertel et al. (1997) examine tax effects, Guiso et al. (1996) consider the relation between labor income risk and portfolio choice, and Heaton and Lucas (1997) study the effect of entrepreneurial risk on portfolio composition. Agell and Edin (1990) estimate a model resembling that of King and Leape on Swedish data, stressing tax effects in particular. Bertaut (1998) estimates bivariate probits on the 1983 and 1989 SCFs to investigate what determines whether households directly own stocks.2 All the above studies, however, adopt reduced-form specifications and do not link discrete and continuous choices through utility maximization as is done in the present paper. To cope with the fact that a large number of individuals hold only a subset of the assets available, we make the plausible assumption that households face monitoring costs, either in terms of time or money, of holding various portfolios. Such an assumption is consistent with the observation that hardly anyone holds, say, $1's worth of stocks. Past attempts at modeling incomplete portfolios have employed the assumption of non-negativity constraints that prevent investors from short selling particular assets as they would ideally choose to do. See, for example, Auerbach and King (1983). The assumption of fixed portfolio monitoring costs adopted in the present study provides a simple and flexible alternative. Similar assumptions might well prove useful in coping with zero consumption levels in studies of deterministic consumer demand.3 In the presence of fixed monitoring costs, one may decompose a household's optimal portfolio choice into (i) the choice of which assets to include in the portfolio and, (ii) the decision of how much of those assets to hold. The first stage involves comparing the maximum amount of utility attainable given different incomplete portfolios, while the second stage may be viewed as a ‘continuous’ decision as to how much of each asset or liability to demand. Just as in standard consumer theory, Roy's identity implies demand functions for risky assets that are ratios of partial derivatives of the indirect utility function. Sandmo (1977) seems to have been the first to apply this technique in portfolio theory, but the present paper is the first to combine such an approach with discrete-continuous econometric methods. The techniques applied in this paper, and in particular the selectivity adjustments, may be viewed as an extension of the methods developed for different purposes by Dubin and McFadden (1984) (building on previous work by Heckman (1978)). Dubin and McFadden use discrete-continuous econometric methods to examine household consumption of different forms of energy. While Dubin and McFadden only estimate a single demand function using limited information procedures, our study estimates the whole system simultaneously, imposing the cross-equation restrictions implied by the theoretical model. Extending Dubin and McFadden's selectivity adjustments to a multivariate system is non-trivial and may be regarded as one of the contributions of this paper. The SCF provides detailed information on the asset holdings, income and demographic characteristics of a sample of 4262 US families. To estimate the model, we aggregate asset and liability holdings into stocks (S), bonds (B) and money (M). These categories are more highly aggregated than one might wish but including more assets is difficult given the computational requirements of the estimation. Agents are all assumed to hold some quantity of money so investors may choose to hold any one of four different sets of assets (SBM,SM,BM,M). Households in the SM category comprised less than 1% of the sample and were not included in the study since it would not be feasible to obtain estimates of the parameters relating to this regime with any precision. Our model sheds light on two sets of issues. The first is the impact of demographic effects and non-traded assets on financial markets.4 Our study allows an explicit role for such factors by permitting the parameters of the indirect utility and demand equations to depend linearly on dummy variables describing demographics and on continuous variables such as housing equity, mortgage debt, etc. Second, our model provides evidence on the validity of the standard aggregation assumptions frequently made in financial economics. Commonly adopted assumptions such as Hyperbolic Absolute Risk Aversion (HARA) preferences5 or normally distributed rates of return imply individual and aggregate asset demands that are linear functions of initial wealth. To assess the importance of wealth distribution for aggregate portfolio demand, one may examine the extent to which individual demand functions are non-linear in wealth.6 We simulate our model for the actual dataset and for various perturbations to wealth distribution and demographics. The simulations allow for endogenous regime switching as households not only change their asset demands but also switch from holding one basket of assets to another. Our general finding is that some forms of distributional shocks are very significant. For example, redistributing wealth from poor to rich significantly raises the demand for bonds. A proportionate rise in the wealth of all households, however, raises the demand for stocks. Some demographic changes are also important for aggregate asset demands. For example, a shift from married to single household heads raises the demand for bonds at the expense of that for stocks. The paper is arranged as follows. Section 2 reviews our basic model of portfolio choice and describes its econometric implementation. Section 3 gives a brief account of the data used. Section 4 presents the results and reports on some tests of model specification. Section 5 concludes.

This paper has employed discrete-continuous econometric methods, allowing for sample selectivity, to model US households' portfolio decisions. Our major conclusions are: 1. Asset Engel curves are non-linear, generating signi"cant increases in demands for stocks and bonds as wealth rises. In a simulation designed to replicate the impact on the portfolio choices of the US population, we "nd that a 10% proportional rise in wealth of all households leads to a 24% and a 25% increase in stock and bond demand, respectively. A 10% rise in the absolute dispersion of log wealth leads to a 11% rise in stock demand but only a 2% rise in the demand for bonds. 2. Household characteristics apart from wealth have important e!ects on portfolio decisions. For individual households, family size, sex of household head, education and attitudes to risk and liquidity signi"cantly in#uence the basket of assets households end up possessing. Race, marital status, and occupation are less important. 3. Simulations of the impact of changes in household characteristics on the asset demands of the population as a whole suggest that shifts from married to single household heads, changes in educational level, occupation, liquidity and risk preference can lead to big percentage changes in bond demands, often with o!setting changes in both stock and money demands. 4. Selectivity adjustments are highly statistically signi"cant. This suggests that the results of past studies which employed discrete-continuous methods but did not allow for selectivity may be hard to interpret.