Information effects of dividends: Evidence from the Hong Kong market

Claessens et al. (2000) report that nearly one-third (32.1%) of the total value of listed corporate assets are controlled by the largest ten families in Hong Kong.

Chang and Chen (1991) suggest that a test on the information content of earnings and dividends is sensitive to the interval between the two signals.

Claessens et al. (2000) and Fan and Wong (2002) find that the ownership structure in Japanese firms is different from that of other East Asian corporations in terms of the degree of control and cash-vote divergence. Japanese firm shares are widely held by large financial institutions.

Fan and Wong (2002) propose an entrenchment effect of ownership structure (confirmed by Morck et al. (1988), McConnell and Servaes (1990) and a proprietary information effect (Christie et al., 2003) on financial reporting. The controlling owners (large families) have strong opportunistic incentives to hold up minority shareholders and to limit information flows to the public as well as competitors. Hence, the market would expect that the controlling owners, who oversee accounting reporting policies, do not report high-quality accounting information.

Holthausen (1990), Healy and Palepu (1993), and Dechow (1994) argue that accruals help make accounting numbers more informative to better reflect firm performance. An ability of earnings to signal performance depends on the extent to which managers exercise their discretion to manipulate reported earnings.

Since there is a possibility that firms may change the dates of their fiscal year-end, the number of months covered in a fiscal year may be different from 12. Of the 4,637 financial statement records extracted from the PACAP database, in 152 cases, there are other than 12 months covered in a fiscal year. To avoid incorrect classification of "increase" or "decline" subsamples, we exclude events with different numbers of months in consecutive fiscal years.

To calculate the unexpected changes in dividend distribution (increase, decline, and zero) between two years, we use only the amount announced as the regular final dividend. Sometimes, firms declare extra cash distributions as special dividends or bonus dividends in addition to the usual final dividend. These extra cash distributions are usually one-time payment. To avoid misclassification of unexpected changes due to the distribution of irregular payments, we exclude any extra cash distribution amount in our calculation of the unexpected dividend changes. When the previous dividends per share and average dividends per share over the past three years are zero, we scale the magnitude difference between the realized dividend per share and expected dividends per share by the share price.

A sample check on the announcement dates in the dividend announcement section indicates that our announcement dates are either the same day as or one day earlier than the date firms publicize their reported earnings and dividend information in the media. Therefore, we use a three-day interval (−1 to +1) for measuring the abnormal returns for announcement.

The PACAP database includes two types of firms, financial companies and industrial companies. The types of accounting variables used to measure the performance and riskiness of the financial companies are substantially different from those used for industrial companies. Including financial companies in our study may create problems in the control firm analysis and for variable selection in the regression analysis for dominant signals.

Return on assets is the ratio of net income to total assets. Operating cash flow is measured using the indirect method to convert net income from operations to cash flow from operations.

To compute the beta, we use the market model with returns from t = −200 to t = −51.

Table 2 reports the results of the abnormal returns using the control firm approach. For an additional test of robustness, we estimate the abnormal return using the market model. The results are shown in Appendix A. The abnormal returns calculated using the control firm approach and market model are qualitatively similar. The signs of returns for different combinations of unexpected earnings and dividend events are the same but only the magnitudes of returns are different.

To illustrate that our regression models do not suffer from multicollinearity problem, we compute the variance inflation factor (VIF) for our models. The VIF values for all our independent variables are less than 10, indicating that our regression results are not affected by multicollinearity problem (Belsley et al., 1980). In addition to VIF measurement, we also calculate the Pearson Correlation for the independent variables in each model to see if they suffer from potential multicollinearity problem. The correlation coefficient is low and ranges from −0.1211 to 0.3640. In addition, for the independent variables in Eqs. (5) and (6) tested, only 6 out of 29 Pearson correlation coefficients are significant at 0.05 level. Evidence from VIF measurement and correlation analysis shows that our models (Eqs. (5) and (6)) do not suffer from multicollinearity problem.

The regression analyses in Table 3 show that a special dividend distribution is a significant factor explaining share price performance. To provide further evidence that this finding does not impart a bias, we eliminate the subsample of events with special dividend announcements and repeat the regression analyses. The regression results are qualitatively the same. These results show that UDPSChg remains a significant factor determining abnormal returns.

Holthausen et al. (1995) and Hribar and Collins (2002) examine measurement error in accrual estimates. Jones (1991) model and its variations have the ability to decompose accruals into discretionary and non-discretionary accruals. In order to show that our findings are not affected by the way we estimate the proxy for discretionary accruals, we also compute current accruals (Rangan, 1998) and discretionary accruals defined in the Jones model (1991). We find consistently a negative relation between CAR and Accrual, whatever estimation methods we use to compute the discretionary accrual measure. Our major finding that the dividend signal is the stronger factor still holds for different discretionary accrual measures.

In a robustness test, we use another alternative interactive dummy variable, DUDiP*UE, for the unexpected increase in dividends per share (which takes the value of 1 if there is an unexpected increase in dividends per share between year y and the average of years y−1, y−2, and y−3 and 0 otherwise) and UEPSChg. A repeat of the regression analysis using different combinations of DUDiN*UE, DUDiZ*UE, and DUDiP*UE yields similar results.

To show that our findings are not sensitive to the method used to estimate the variables, we repeat the analysis using Expectation Model 2. Similar results are obtained, suggesting that our results are robust to the different measures of dividend signals and variables.

To demonstrate the robustness of our findings, we also use ratio of operating income to total assets as our measure of return on assets. The results are qualitatively the same.