Ownership structure and dividend policy: Evidence from Italian firms

Abstract

This paper reports on empirical investigations into the relationship between dividend policy and ownership structure of firms, using a sample of 139 listed Italian companies. Ownership structure in Italy is highly concentrated and hence the relevant agency problem to analyse seems to be the one that arises from the conflicting interests of large shareholders and minority shareholders. This paper therefore attempts to test the rent extraction hypothesis by relating the firm’s dividend payout ratio to various ownership variables, which measure the degree of concentration in terms of the voting rights of large shareholders. The hypothesis that other non-controlling large shareholders may have incentives to monitor the largest shareholder is also tested. The results of the empirical analysis reveal that firms make lower dividend payouts as the voting rights of the largest shareholder increase. Results also suggest that the presence of agreements among large shareholders might explain the limited monitoring power of other ‘strong’ non-controlling shareholders.

Keywords:

• Dividend policy
• ownership structure
• agency costs

Acknowledgements

The authors are grateful to Andrea Pilati, Mario Quagliariello and Andrea Resti for helpful comments and suggestions. Opinions expressed in this paper are exclusively the author’s and do not necessarily reflect those of Banca d’Italia.

Notes

1. A pyramidal structure occurs when the controlling shareholder owns one corporation through another which he does not totally own. It enables individuals to control a wide range of assets and activities with a limited amount of investment, by spreading the voting rights of minority shareholders over a large number of firms and concentrating their own in the company at the top of the pyramid.

2. The separation between ownership and control decreased between 1990 and 2000 (CONSOB, annual report 2000). Not only firms making use of non-voting shares has fallen from 97 in 1990 to 60 in 2000, but also the number of hierarchical levels within pyramidal group structures has decreased. However, in 2001 the leverage obtained by pyramids has risen back to the levels of the early 90s.

3. Before the reform the ‘antidirector rights index’, which is used as a summary measure of legal protection for investors (see La Porta, 1999 for details), was equal to one out of six for Italy; after the reform the index has risen to five (CONSOB, 2001), the average for civil law countries being 2.33.

4. Since the recently listed firms do not normally pay dividends, the exclusion of these firms resulted in a reduction in the number of firms that are identified as non-payers of dividends in our sample. However, recent IPO’s are likely not to pay dividends because of poor cash flows and revenues. Therefore, the lack of dividends is less likely to reflect agency conflicts among shareholders.

5. In order to check if this last non-stochastic selection of observations in the sample might have caused a bias in the regressions, we used Heckman’s two-step estimator (Heckit). The results suggest that sample selection bias does not seem to affect our estimation results.

6. Obviously, the representation ratio is lower partly because some of the largest firms in terms of market capitalization were excluded. Specifically, 4 of the 10 largest Italian listed firms are banks or insurance companies, which are not included in the sample. Furthermore, Telecom Italia Mobile SPA, a large telecommunication corporation, was excluded since it was a typical outlier. Finally, the largest automotive firm (Fiat SpA) is not in the sample as its earnings were negative in 2001.

7. The complete set of variables (with the description of the sources where they have been collected from and of the year they refer to) is reported in the Appendix.

8. Recent papers on dividend policy measure the dividend rate by: Dividend/Cash flow, Dividend/Earnings, Dividend/Sales (La Porta et al., 2000). Faccio et al. (2001) also use Dividends/Market Capitalization. Gugler and Yurtoglu (2001) only use the classical Dividend/Earnings ratio.

9. The ownership variables come from the Consob database which contains the list of direct stakes—expressed in terms of voting rights above the 2% threshold for each firm. It also contains the name of the ‘declarant’ of each stake, i.e. the owner that holds the direct stake. Therefore, when a firm belongs to a pyramidal group, it is possible to know the ultimate owner, which is the individual or the company at the top of the pyramid. For instance, 56.7% of the firm ‘Autogrill’ in 2001 is owned by ‘Edizione holding SpA’ that, in turn, is owned by ‘Ragione di G. Benetton & C SAPA’. This firm is expression of the Benetton family and hence Autogrill is controlled by a family with a stake of 56.7%.

10. For total assets and leverage an average value over the period 1997 to 2000 is considered. In a few cases the average is only based on two (1999 to 2000) or three (1998 to 2000) years because of the limited availability of data. Similarly, the average value for the market-to-book ratio is determined for the whole sample using the period 1999 to 2000. We computed a two year average for the market-to-book-value variable because data on the market capitalization of each listed firm is not available for 1997 and 1998 on the Milan Stock Exchange website.

11. This is consistent with Faccio and Lang’s (2002) findings, which report that 72.33% of Italian non-financial firms were family owned in 1996. It also confirms that ownership structures tend to be rather stable over time, as noted in La Porta et al. (1999).

12. Faccio et al. (2001) find that in 1996 the average percentage of control rights of the first shareholder was 48.3%.

13. The Pearson correlation matrix, which is not reported, shows that the correlations between the independent variables are either low or moderate degree, which possibly points to the absence of multicollinearity.

14. The TOBIT specifications can be estimated with the MLE. However, in the presence of heteroscedasticity the standard TOBIT model results in inconsistent coefficient estimates. Therefore, in our estimation, we use Huber/White/sandwich estimator of variance in place of the conventional MLE variance estimator by employing the ‘robust’ option of the ‘intreg’ command in STATA 8.1.

15. It is likely that one of the explanations as to why firms with higher level of ownership concentration pay lower dividends is related to the taxation of shareholders’ dividend income. To the extent that large shareholders will be subject to a higher tax bracket, and the dividend income is taxed at a higher rate than the capital gains income, controlling shareholders will be less willing to receive income in terms of dividend payments. However, we do not have data on the taxation of controlling shareholders to test if this explanation holds.

16. We also explored the possibility that the exact nature of the relation between the voting rights of the largest shareholder and the firm’s dividend policy may depend on the existence of coalitions between large shareholders. Accordingly, we introduced an interactive dummy (Voting_rights1*Agreements) that allows us to test both the main impact on the firm’s dividend policy and the conditional impact of the voting rights of the largest shareholder when there is a coalition between shareholders. The estimated coefficient of the interaction variable (the results are not reported for brevity) was positive for both definitions of dividend payout ratio but insignificant, though marginally. Thus, we cannot provide strong evidence for our prediction that the negative effect of the largest shareholder’s voting rights on the firm’s dividend payouts is reduced in the presence of a voting syndicate.

17. We find that the average stake of the largest shareholder is 51.87% in the presence of coalitions, whereas it is 30.92% if there are no agreements among shareholders.

18. It is worth mentioning that our results are not perfectly comparable to those of Faccio et al. ’s (2001). This is mainly because we do not distinguish between group affiliated firms and independent firms. Furthermore, as far as the significance of the estimates is concerned, it is important to note that Faccio et al. ’s estimated coefficient of the O/C ratio for Italy is statistically significant when the dividend rate is measured as the ratio of dividends on sales, whereas it is not significant when the dividend rate measures we use in this work—dividend/earning and dividend/market capitalization ratios—are considered.

19. In the binary choice model the value of the dependent variable is limited to two values, 1 if the decision is to pay dividend and 0 if the decision is not to pay. Thus the dependent variable is the probability of the firm deciding to pay dividends conditional on the information set specified by the regressors.

20. Although the model predicts correctly the vast majority of payers (about 90%), about half non-payers are predicted as payers. However a check on the 22 non-payers incorrectly predicted as payers in specification 1 has shown that 12 were former payers, whose profitability has become negative in recent years.

21. One notable exception is the finding regarding the Voting_rights_all variable in specification 5, which measures ownership concentration in terms of voting rights of all disclosed shareholders. Its estimated coefficient positive, though insignificant.