Securities regulation, household equity ownership, and trust in the stock market

Along with EU member countries, we also include Iceland, Liechtenstein, and Norway in our sample, countries which are not in the European Union but belong to the EEA. We include these countries because they have agreed to adopt the EU capital market directives, among other things, in exchange for access to the single market. For simplicity, we refer to all countries in our sample as EU countries.

Giannetti and Koskinen (2010) separate investor protection into its public and private enforcement components and find that the association between investor protection and stock-market participation is primarily driven by investors’ ability to privately enforce their own rights (i.e., through the court system). Our results suggest that public oversight is an important determinant of household equity ownership.

PROSP pertains primarily to the preparation of prospectuses for public securities offerings by issuers and only applies to primary market trading—a relatively limited subset of total trading in the stock market. TPD focuses on enhancing corporate disclosures by establishing new requirements and strengthening the enforcement of existing requirements for periodic financial reports. While these directives likely enhance the credibility of corporate disclosures, they focus on improving transparency and are not directly aimed at enhancing retail investor confidence. Nevertheless, we control for the adoption of both PROSP and TPD in our empirical analyses.

We only disclose the country-specific MiFID adoption years because we obtained the country-specific MiFID adoption dates from Jean-Marie Meier with the agreement that we would not disclose the exact dates in the paper. These dates will be publicly available upon the publication of Meier (2019).

Another observable aspect of the directives is their subsequent enforcement. There is evidence that both MAD and MiFID were actively enforced (e.g., Christensen et al. 2016 document that nine out of the 22 sample countries in our main analysis had taken at least one enforcement action under MAD by the end of 2009). However, at a conceptual level, it not clear that households should delay their reaction to new regulation until an enforcement action is taken, particularly if they trust that the government will implement and enforce the laws on the books. (We examine the role of trust in the government in Table 5 Panel C.)

Studies have also used brokerage account data (e.g., Scularbaum et al. 1978; Barber and Odean 2000) or government tax records (e.g., Blume and Friend 1975, 1978; Kopczuk and Saez 2004). Internationally, research has obtained data from government-centralized share registers available in some countries (Grinblatt and Keloharju 2000, 2001; Calvet et al. 2007). While these data sources are highly accurate, they do not sample from the entire population, do not cover all relevant financial assets, or are only available in a few countries. These limitations prevent us from using these data in this study.

ESA2010 replaced the former reporting framework, ESA95, in September 2014. As of the time of our study, not all European countries have transitioned to the ESA2010 reporting standards. However, because the impact of the change in standards on the financial accounts of interest in our study is minimal, we use ESA95 data for countries where the full ESA2010 data is not available. Furthermore, we are unaware of any aspects of MAD, MiFID, or other Lamfalussy Directives that would have affected the calculation of the values of the financial instruments used in our study.

We separately examine changes in equity ownership within investment funds in Section IA2 of the internet appendix.

Austria, Bulgaria, Cyprus, Iceland, Ireland, Liechtenstein, Netherlands, and Romania have missing or incomplete national accounts data. Croatia joined the European Union in the final year of our sample period (in July 2013). Of the 22 countries in our main sample, eight joined the Union over our sample period. In untabulated analysis, we confirm that our inferences are consistent if we exclude these countries.

In the internet appendix (Section IA4), we provide evidence that the levels of household equity ownership and stock-market participation, although conceptually different, are positively correlated (Pearson correlation of 0.63). However, the correlation in levels does not necessarily imply a similar correlation in changes, which is the focus of our study.

An important assumption is that the control group is not also affected by the treatment (i.e., the stable unit treatment value assumption, “SUTVA”). This assumption is unlikely to hold in our settingbecause we expect that foreign households are also affected by the adoption of the directives in other countries (e.g., through a reduction in home bias). This effect biases against finding an impact of the directives. We assess the magnitude of this bias in Table 4 by explicitly controlling for the effect of domestic adoption on foreign households.

In Section IA5 of the internet appendix, we report results controlling for liquidity. While our sample size is slightly smaller for this analysis because some countries lack liquidity data, controlling for liquidity has little effect on the estimated treatment effect of the directives.

In additional (untabulated) analyses, we find no evidence of significant (positive) abnormal returns in either a short-window (i.e., three days) around the directives’ entry-into-force dates or over the quarters subsequent to the entry-into-force dates, which mitigates the concern that the increase in household equity ownership we document is driven by an increase in the value of households’ existing equity portfolios.

Household Equity Returns are calculated using a household-portfolio specific measure from the national accounts. Because the exact timing of portfolio changes is unknown, this adjustment is measured with some error. Thus we replace extreme (more than three standard deviations from the mean) or missing values of Household Equity Return with country-specific stock-market returns. As an alternative to including Household Equity Returns as a control variable, in Internet Appendix Section IA6, we report results where we directly adjust Equity Ownership for changes in equity values using the national-accounts-based adjustment factor. Results are very similar in this alternative specification. Results are also very similar if we use changes in a country’s stock-market index to control for share-price appreciation (untabulated).

The household finance literature finds that the following factors significantly affect household stock-market participation: wealth, momentum, tax rates, education, financial sophistication, and marital status (Cohn et al. 1975; Campbell 2006; Barber and Odean 2013). We directly control for proxies for wealth, momentum, and tax rates. We do not include controls for education, financial sophistication, and marital status, which are likely slow-moving and therefore are unlikely to be correlated with entry-into-force dates. Additionally, they would (in part) be captured by including country fixed effects.

In a sensitivity test, to address voluntary IFRS adoption as a potential correlated omitted variable, we exclude Germany, where voluntary IFRS adoption was common prior to mandatory adoption. Results are slightly stronger when Germany is excluded from the sample.

We do not cluster standard errors by country, because, given that there are only 22 countries in our sample, this approach is likely to overstate or bias the standard errors. Specifically, there are only 22 countries included in the analysis, and clustering by country-year could understate the standard errors. Therefore, to assess the reasonableness of clustering by country-year, we also calculate standard errors using a Monte Carlo approach where we randomly select adoption dates for each country and assess significance by calculating the fraction of counterfactual treatment effects that exceed our actual estimated treatment effect. Using this method, the statistical significance of our results is higher than reported in the paper (untabulated).

This estimate is based on an average participation rate of 0.15 for our sample countries from the 2005 Eurobarometer Survey.

This finding is consistent with the results of Christensen et al. (2016), who, for one test, rely on variation in MiFID adoption dates and find that MiFID has a positive but insignificant effect on stock-market liquidity. Alternatively, Cumming et al. (2011) find a significant effect of MiFID on liquidity, using a control sample of firms from non-MiFID adopting countries (rather than using variation in the adoption dates among MiFID adopting countries).

Calculated by multiplying the maximum MAD Foreign for Latvia (United Kingdom) of 8.28 (0.13) with the coefficient on MAD Foreign of 0.006.

While there is some overlap between the Low Trust and High Trust Differential countries, the set is not the same. In Finland, Luxembourg, and Sweden, High Trust = 1 and High Trust Differential = 0. In Hungary, Italy, and Poland, High Trust = 0 and High Trust Differential = 1.

The mutual fund data do have two important disadvantages that make it unsuitable for our primary analyses of household equity ownership. First, changes in mutual fund holdings are unlikely to be driven solely by the preferences of households (e.g., because other sectors, besides households, invest in mutual funds and mutual fund managers have some discretion over the allocation of invested funds). Second, we cannot measure households’ other liquid assets and cannot assess the proportion of their total liquid assets that households invest in the stock market (i.e., the focus of our main analysis).

Croatia, Romania, and Lichtenstein are excluded from the analysis because of missing MAD or MiFID dates.

Guiso et al. (2009) also consider several other proxies for bilateral trust, including genetic and somatic similarity and the extent of historical military conflict between countries. However, these proxies are not available for several EU countries, so we do not consider them in our analysis.