Savings selectivity bias, subjective expectations and stock market participation

Abstract

Studies of household stock market participation report low participation rates. The explanations cited are that the fixed costs associated with participation and high risk aversion discourage households from buying stocks. However, the low participation rate findings are unchallenged. We argue that because prior studies fail to recognize that not all households save, there exists a selection bias when estimating the household participation rate. After correcting for this selection bias, as well as accounting for the influence of subjective expectations on market participation, we show that the unconditional probability of participating in the stock market would increase twofold.

Notes

¹ For more details about the impact of stock market participation and the equity premium puzzle, see Madsen and Dzhumashev (2009), Roche (2006), and Jagd and Madsen (2009).

² In addition, the literature has sought other behavioural explanations: need for financial education, narrow framing of risks and nonstandard preferences (Gomes and Michaelides, 2005).

³ The geography information is critical in our study since it is used as an instrument to correct for bias in the structural model.

⁴ The imputation procedure is described in Kennickell et al. (2000). According to the SCF codebook, ‘The overwhelming majority of variables that originally contained missing values are imputed five times by drawing repeatedly from an estimate of the conditional distribution of the data. These imputations are stored as five successive replicates (“implicates”) of each data record. Thus, the number of observations in the full dataset (21 545) is five times the actual number of respondents (4309)’.

⁵ See Kennickell et al. (2000) for more details.

⁶ Money in transaction accounts is not considered savings but rather cash in advance for consumption. If we label a transaction account as savings, then we have to consider credit card debt and other short-run payments such as mortgage payments. Those debt payments overwhelmingly exceed the money in transaction accounts.

⁷ These findings about saving rates are consistent with SCF surveys for other years.

⁸ A Union Bank of Switzerland (UBS)/Gallup poll conducted in April 2000 asked investors about what specific information they use when purchasing or selling any stocks or other financial holdings. The poll found that 28.4% employ only informal searches, 14.1% use only professional searches and 40.1% use a combination of informal and professional searches.

⁹ Bertaut and Starr-McCluer (2001) report that only 19.2% of households owned stocks for the year 1998 (see their Table 3).

¹⁰ See Giannetti (2005) for long-run participations with mean-reverting stock return.

¹¹ See Vissing-Jørgensen (2002, p. 829).

¹² We thank Daniel Hamermesh for suggesting the two instruments used in this study.

¹³ Although the SCF does not provide data about an investor's state, data about the region is provided. In correspondence with the SCF Senior Economist and Project Director, we were told that ‘The SCF is not designed to be representative at the state level … the selection of respondents within these areas in the SCF sample is made to balance population groups across regions, but not within states’.

¹⁴ There is an econometrics issue that is related to the wealth variable. In our sample, 10.47% of households have negative wealth levels. Consequently, we cannot use the log of wealth. One possible solution for dealing with negative wealth is to use the inverse hyperbolic sine function (see Pence (2006) for details). However, estimating the model with this function does not fit as well as when we employ the wealth variable.

¹⁵ Notably, the impact of the race variable on stock market participation also has dropped significantly when we account for savings selectivity.

¹⁶ Grinblatt et al. (2009) provide interesting results on the relationship between Intelligence Quotient (IQ) and stock market participation.