Empirical studies of savings behaviour have focused primarily on the total level of savings rather than its allocation among different types of assets. Although there is a substantial literature on the theory of portfolio allocation there are very few empirical studies of household portfolios using individual data. Those studies that do exist are now rather outdated. In Discussion Paper No. 43 Jonathan Leape and CEPR Research Fellow Mervyn King examine a new survey of 6010 US households and their asset holdings. The survey contains detailed information on household income, as well as portfolio composition and net worth. High income families were "oversampled' in the survey, that is, they were represented more often in the survey than in the population as a whole. Over 40 per cent of the sample reported net worth in excess of $100,000, and the sample contained 204 millionaires. This "over- representation' of the wealthy in the survey allows more reliable conclusions to be drawn about their portfolios.
Leape and King discuss the questions that arise in modelling how the household chooses both the number of assets held and the fraction of its net worth allocated to each asset. The survey data demonstrate that most households own only a small number of assets. Conventional portfolio theory suggests substantial diversification and therefore appears unable to explain this behaviour. Leape and King propose an empirical model in which households' portfolios are incomplete, and they use this model to estimate the responses of households to taxes and to changes in wealth. They find that taxation has significant effects on the composition of portfolios. They also find that holdings of corporate equity and certain other assets are strongly influenced by household wealth; a one per cent increase in wealth appears to increase holdings of these assets by more than one per cent.
Leape and King also find that the observed patterns of asset ownership appear to reflect the costs of acquiring and processing information about different assets. Such costs will vary among households, and the authors' empirical results show that education and occupation are important determinants of the ownership decision for a number of assets. The existence of such household-specific information casts doubts on the traditional assumption of "homogeneous expectations'. It also suggests that both theoretical and empirical work on portfolio behaviour must pay greater attention to the origin and the costs of the information held by households.

Wealth and Portfolio Composition: Theory and Evidence
Mervyn A King and Jonathan I Leape

Discussion Paper No. 43, January 1985 (ATE)