Dividend preference of tradable-share and non-tradable-share holders in Mainland China
doi: 10.1111/j.1467-629X.2008.00284.x
Louis Cheng acknowledges the financial support (grant number: G-YF41) from the Hong Kong Polytechnic University. We want to thank the two anonymous referees and the Editor for their valuable inputs and comments.
Abstract
Comprehensive data on corporate announcements of Chinese firms allows us to examine the preference for, and determinants of, cash and stock dividends. The results indicate that Chinese public investors prefer stock dividends over cash dividends, which are preferred by large state and legal person shareholders generally. Stock dividends, which do not require an explicit cash outflow from a firm, are found to be positively related to higher earnings, supporting the signalling hypothesis of dividend policy. In an imperfect market, these results have some implications for government regulation of financial markets.
1. Introduction
In many emerging markets, firms typically pay stock dividends rather than cash dividends. China, one of the fastest growing emerging markets, opened the Shanghai Stock Exchange in December 1989 and the Shenzhen Stock Exchange in April 1991. Since then, we have learned that Chinese investors appear to favour stock dividends over cash dividends (Chen et al., 2002). There are many hypotheses explaining why firms pay out stock dividends. The signalling and retained earnings hypotheses, which are closely linked and relate stock dividends to a firm's good growth or investment potential, appear to be the leading contenders in explaining stock dividend policies of firms in the USA (Baker et al., 1995).
Under the current Chinese tax system, cash dividend is subject to tax. However, not all cash dividend income is taxable. The cash dividend income is exempt from tax if the cash dividend income is less than the 1 year saving deposit rate declared by China's central bank (People's Bank of China). If the cash dividend income is higher than that above-mentioned amount, a flat tax rate of 20 per cent would be charged on the excess amount. For stock dividend, since the gains on stock dividend are not yet realized, there is no immediate tax effect. Dividend yields for Chinese firms are generally low, and the declared saving rates in China have been around 2–10 per cent during the past decades; therefore, the tax effect differential for cash and stock dividends would be insignificant. In short, both stock and cash dividends are not exempt from tax. Practically speaking, there is no tax preference issue in China's securities market nowadays.
Some particular characteristics of the Chinese financial market likely influence the dividend policy of the Chinese firms. First, the stock market watchdog, the China Securities Regulatory Commission (CSRC), requires minimal disclosure requirements for listing firms, clearly different from the detailed Securities and Exchange Commission requirements in the USA. Chinese firms also seldom disclose voluntary information to market participants. The opaqueness of firms and their limited information dissemination to the public increase investment risk. Therefore, Chinese investors have to rely on major firm policies such as stock dividends to convey signals about a firm's prospects. The signalling role of dividends is corroborated by Haw et al. (2000), who show that good-news Chinese firms release annual reports earlier than bad-news firms. This result clearly supports the signalling hypothesis in China.
Second, China's financial markets are new and still developing. There are limited investment opportunities because Chinese bond and derivatives markets are underdeveloped (Fung and Leung, 2001). Therefore, reinvestment by plowing back earnings should be viewed positively. If firms indeed have good investment prospects, shareholders prefer stock dividends in order to preserve cash for investments; seasoned equity financing is not readily available for future funding needs because of regulatory constraints. Therefore, the underdevelopment of China's market implies that rational Chinese stockholders would prefer stock dividends to cash dividends.
Why would a Chinese firm pay cash dividend in light of such market impediments? The explanation lies primarily in China's corporate ownership in the listed Chinese firms that issue three types of shares. A-shares, denominated in the Chinese currency, are tradable on the Shanghai and Shenzhen Stock Exchanges and are available to the domestic public only.1 B-shares, denominated in Hong Kong dollars or US dollars, are also tradable and are issued to both foreign investors and Chinese residents who have the required foreign currencies (US dollar for Shanghai B-shares and HK dollar for Shenzhen B-shares). They account for about one-third of all shares. The remaining two-thirds on the exchanges are called state shares (shares owned by the state government) or legal person shares (shares owned by financial institutions).
In April 2005, the Mainland Chinese government launched an economic reform on split share structure by converting non-tradable shares (state and legal person shares) into tradable shares. However, after the reform, only a limited number of converted shares have been sold to the public for two main reasons. First, the CSRC requires that no shares can be converted within the first year. In addition, the amount of non-tradable shares can be converted must be less than 5 per cent after 1 year and less than 10 per cent after 2 years. Second, the central government aims to maintain a healthy equity market and does not want to see the huge amount of converted shares flooding the market and pushing down the market prices. Hence, although there might be little or no non-tradable shares existed legally by 2007, trade limitation of the converted shares (i.e. de facto non-tradable shares) would exist for the next several years. Consequently, the agency problems created by non-tradable shares persist and the empirical issue we investigate here would still be important and relevant.
Both tradable-share holders and non-tradable-share holders have equal voting rights and cash flow claims on firms, but non-tradable-share holders represent the controlling stockholders, because they own the majority of shares. Controlling stockholders prefer cash dividends to stock dividends, given the ‘me-first’ rule. Although non-tradable shares are not traded on the exchanges, they can be sold with government permission. The prices of non-tradable shares range from about 70 to 130 per cent of book values, substantially lower than the market price (Huang and Fung, 2004). If dividend policy serves as a signal to the market, firm value (price) will change as a result. Price appreciation will not translate into financial gains for the controlling stockholders whose shares cannot be traded through the stock exchanges. Therefore, they would prefer cash dividends to realize an immediate financial gain.
Our analysis of stock dividends in a logit model suggests that firms with a higher percentage of A-share have a greater chance of issuing stock dividends. The results further support our contention that public investors indeed prefer stock dividends, which serves as a signal to the market of higher profitability.
The rest of the present paper is organized as follows. First, we present the literature and our hypotheses in the second section. The data and methodology are explained in the third section. We explain our empirical results in the fourth section. The last section is the conclusion.
2. Literature review and hypotheses
2.1. Literature review
It is well documented in the accounting and finance literature that dividend policies convey information. Miller and Modigliani (1961), John and Williams (1985) and Miller and Rock (1985) posit a dividend information hypothesis that the market reacts to the information content of cash dividends, interpreting dividend as a credible signal for the prediction of future earnings and permanent increase in the future cash flows of a firm. A voluminous literature since then explores the informativeness of cash dividends empirically, yielding mixed results.
Kalay and Loewenstein (1985) and Nissim and Ziv (2001) find a strong positive relation between dividend changes and a firm's ability to generate future earnings and cash. Contrary studies provide evidence of an insignificant relation between dividends and future earnings (DeAngelo et al., 1992; Benartzi et al., 1997). Other studies show that dividend loses its information content in explaining firm's future performance when earnings and earnings-related variables (such as earnings forecasts) are released simultaneously (Conroy et al., 2000; Mikhail et al., 2003). A new view is the tunnelling perspective, which argues that cash dividends might be used as a tool to re-direct firm resources to benefit large shareholders and top management at the expense of minority shareholders (Faccio et al., 2001).
A number of studies explore the reason for and the impact of issuing stock dividends. Foster and Vickrey (1978) report that stock dividend issues generate positive abnormal returns on the declaration date rather than on the ex-date, supporting a signalling function of stock dividends. Other studies providing evidence of significant announcement effects and support for the signalling hypothesis include Grinblatt et al. (1984), Lakonishok and Lev (1987), McNichols and Dravid (1990) and Banker et al. (1993).
The trading range hypothesis argues that issue of stock dividends can bring share prices into a price range that makes a stock more affordable for investors (McNichols and Dravid, 1990). Grinblatt et al. (1984) and Lakonishok and Lev (1987) provide support in favour of the signalling rather than the trading range hypothesis. Cash dividends and stock dividends have been argued to be substitutes for one another. Ghosh and Woolridge (1988) find that the issue of stock dividends can mitigate the negative market reaction due to reduction or omission of cash dividends, which provides evidence for the cash substitution hypothesis.
The American Institute of Certified Public Accountants requires that stock issues of less than 25 per cent of authorized equity should be treated as stock dividends to reduce the retained earnings account. Therefore, the retained earnings hypothesis argues that stock distributions of less than 25 per cent are a signal of future earnings as the stock dividend-paying firms are expected to replenish the retained earnings account with future earnings. Empirical evidence for the retained earnings hypothesis is mixed (Woolridge, 1983; Banker et al., 1993). Survey results also provide some evidence for various hypotheses on stock dividend issue. Eisemann and Moses (1978) survey chief financial controllers and show support for the signalling, liquidity, cash substitution and retained earnings hypotheses.
Several studies relate ownership structure of Chinese firms to different aspects of firm performance. Sun et al. (2002) find that there is a positive relation between government ownership (measured by state share or legal person share) and firm performance. Cull and Xu (2005) find that ownership structure affects firms’ reinvestment decisions as the private ownership has a positive effect on profit reinvestment rates. Chen et al. (2002) find stock dividends rather than cash dividends perform a better signalling role in the Chinese market. In short, the existence of various classes of stock in China provides ample research opportunities in examining the effect of ownership types on corporate behaviour and performance.
2.2. Hypotheses
The signalling hypothesis suggests that the stock dividend or cash dividend is a signal to the market of profitability. However, in China, we argue that due to the existence of controlling shareholders holding non-tradable shares, investors with tradable shares prefer stock dividends over cash dividends. Therefore, these investors react to these dividend announcements in the stock market according to their preference. As public shareholders prefer stock dividends over cash dividends in light of the limited investment opportunities available in China, the stock market is expected to react positively to stock dividends, but not cash dividends. We call this assertion ‘Dividend Preference Hypothesis’. The hypothesis can be tested in three ways as follows.
Test 1: A stock dividend announcement has a positive effect on share prices, and a reduction of a cash dividend is viewed favourably by the market.
Second, cash dividends are preferred by controlling stockholders with non-tradable shares because capital gains (from an increase in share price) cannot be translated into a financial gain for them as their shares are not tradable and have no market value. Cash dividends are actual financial benefits paid directly to the controllable shareholders. The me-first rule for non-tradable-share holders can be tested as:
Test 2: Cash dividends are favoured by holders of non-tradable shares.
By the same token, stock dividends are preferred by public stockholders (either A-shareholders or B-share stockholders) and serve as signals to the market of higher profitability. Hence, a third test is proposed as follows.
Test 3: Stock dividends are favoured by public shareholders, but not by holders of non-tradable shares.
3. Data and methodology
Our data on announcement dates, dividend distribution, financial statements, company returns and market returns come from the China Stock Market and Accounting Research (CSMAR) database and the Taiwan Economic Journal database. We first collect earnings and dividend data for all firms (except firms in finance industry) listed on the Shanghai and Shenzhen Stock Exchanges. This database contains 14 066 contemporaneous announcements of final earnings and dividends from January 1990 to December 2005.
As the CSMAR database provides complete data for announcement dates and earnings and dividends only from 1993 to 2005, the sample is reduced to include 12 179 simultaneous earnings and dividend announcements. Unavailability of data for the measurement of various variables further reduces the number of observations for the determinants of cash dividends and stock dividends to 7560.2 In China, the fiscal year for all listed firms is the calendar January–December year. According to the ‘Detailed Rules on Information Disclosure by Listed Companies’ published by the CSRC on 10 June 1993, firms must make final simultaneous announcement of earnings and dividends for the fiscal year within 120 days after the end of the fiscal year (i.e. before the end of April of the next fiscal year) absent permission to the contrary. Therefore, we exclude events with announcements later than the end of April to avoid measurement bias arising from delayed announcements of earnings and dividends. We also eliminate events if firms have made a seasoned equity offering during the announcement period to avoid confounding effects on the abnormal return measurement.
The final sample for event study of the announcement effects includes 3826 simultaneous announcements of dividends and earnings. Of the 3826 observations, 565 observations pay a stock dividend (14.77 per cent); and 3261 observations do not pay a stock dividend. There are 1841 observations paying a cash dividend (48.12 per cent). The availability of data for the measurement of variables in the regression model on the information content (abnormal market reaction) of earnings and dividends reduces the number of observations for regression analysis to 3048.
3.1. Construction of expectation models for event study
In general, for markets where the Institutional Brokers Estimate System (I/B/E/S) provides long and broad enough coverage, it is logical to use analyst's forecasts as the expected figures to generate earnings and dividend estimates. Unfortunately, I/B/E/S's coverage for mainland China did not start until a few years ago and very few firms are actually being covered by I/B/E/S.3 Consequently, we have no choice but to use our own model. In the early days when I/B/E/S data were not popular in the USA, some papers use their own expectation model to generate earnings and dividend forecasts. Our model is based on this literature and we estimate the unexpected earnings using the random walk model with a growth component. Ball and Brown (1968), Ball and Watts (1972) and Watts and Leftwich (1977) suggest that annual earnings follow a random walk which implies that market expects a firm to have the same earnings as the previous year if there is no additional information. Therefore, the previous year's earnings can be a good reference to estimate the current year's earnings. Bamber (1987) uses previous year's earnings to estimate expected earnings.
Besides using previous year's earnings as a proxy of expected earnings, the market might also expect the firm to have improvement in annual earnings. Therefore, some studies also include a growth component in the measurement of unexpected earnings. Chen et al. (2005) use the previous year's earnings growth rate of own firm to estimate unexpected earnings. Cheng et al. (2007) incorporate an industry adjustment, which is the growth rate or the percentage change of the mean earnings for all firms in the expectation model.
We need to construct an expectation proxy to model market expectations of earnings and cash dividends, according to the levels of prior year.4 We then expand the market expectations model by adding an industry adjustment factor, which is the growth rate or the percentage change of the mean earnings (dividends) for all firms in industry k of current year (year y) from the previous year (year y – 1).5 The expectations models for earnings and dividends are:
()
()where Ē(Eiy) is the expected earnings per share of firm i in year y; Ē(Diy) is the expected cash dividend per share of firm i in year y; 
is the industry adjustment factor for earnings per share (the percentage change of the mean earnings per share for all firms of industry k in year y and year y – 1); Eiy is the actual earnings per share for firm i in year y; Diy is the actual cash dividend per share of firm i in year y; 
is the industry adjustment factor for dividend per share (the percentage change of the mean cash dividend per share for all firms of industry k in year y and year y – 1); y represents the fiscal year, from 1996 to 2005; and k represents the industry category. There are in total 104 industry sectors in our sample.
3.2. Measurement of earnings changes, cash dividend changes and stock dividend changes
We use the expectations model to estimate the unexpected earnings [
(Eiy)] which is our proxy for earnings surprise. (Eiy) is the difference between actual earnings per share (Eiy) and expected earnings per share [Ē(Eiy)] scaled by the expected earnings per share:
()Complementary to the method used to estimate earnings surprise, the unexpected cash dividend per share [(Diy)] is the difference between the actual cash dividend per share (Diy) and the expected cash dividend per share (Ē(Diy)) scaled by the share price (SPit):6
()where SPit is the share price of firm i on day t.
We use the sign of the unexpected change in earnings per share to divide the earnings announcement events into an ‘unexpected earnings increase’ subsample for a positive change in earnings per share and an ‘unexpected earnings decline’ subsample for a negative change in earnings per share. We have three groups of dividend changes: ‘unexpected dividend decline’, ‘unexpected dividend increase’ and ‘dividend zero’. The unexpected increase (decline) subsample includes events of non-zero and higher (lower) dividend per share than the expected dividend per share measured using the expectations model. All events where the dividend payouts are zero are in the dividend zero subsample. Our measurement of the impact on stock dividend is the percentage change on stock dividend per share scaled by the share price.
3.3. Measurement of abnormal share price reaction by market model
We use an event study methodology to evaluate share price performance upon earnings and dividend announcements. The event date, t = 0, is the announcement date recorded in the Financial PRC General of the Taiwan Economic Journal database. The estimation period covers 150 days from t = –200 to t = –51. We use the market model to estimate the abnormal market reaction.7 The abnormal return (ARit) is the difference between the actual returns of firm i and the expected return estimated on day t. To examine the impact of the earnings and dividend surprises around the announcement over various event windows, we accumulate the abnormal returns for four time intervals (–10 ≤ t ≤ –2; –1 ≤ t ≤ +1; –3 ≤ t ≤ +3; and –10 ≤ t ≤ +10). We use the method outlined in Brown and Warner (1985) as the test statistic for the significance of the abnormal return.
3.4. Information content of cash dividends and stock dividends on abnormal return
We measure the relative information effect of cash and stock dividends on abnormal returns in a regression model:
CAR = α0 + β1UCDPS + β2SDPSChg + β3UEPS + β4 LnTiming + β5IdioRisk + β6DBShare + β7MJAccrual + β8 LnFirmSize.
()CAR is the cumulative abnormal return over the different event windows. For the independent variables in this regression model, we use three explanatory variables related to dividends and earnings. They are UCDPS, SDPSChg and UEPS. In addition, five control variables (LnTiming, IdioRisk, DBShare, MJAccrual and LnFirmSize) are used as well.
UCDPS is the unexpected change in cash dividend per share between the cash dividend per share in year y and the cash dividend per share in year y – 1 with an industry adjustment factor scaled by share price on day t = 0. We use the unexpected cash dividend because the market anticipates regular cash payment, which might not have any new information to the market. In equation (5), SDPSChg is the percentage change in stock dividend per share.8 UEPS is the unexpected change in earnings per share between earnings per share in year y and earnings per year in year y – 1 with an industry adjustment factor.
Chen et al. (2005) and Haw et al. (2000) find that Chinese firms announce good news (positive abnormal earnings) earlier and bad news later using a timing variable. Based on this logic, we believe that using timing as a control variable is appropriate. LnTiming is the log value of the number of trading days from fiscal year-end (December 31) to announcement date. IdioRisk is the standard deviation of residual between actual return and estimated return from the market model over 150 days in the estimation period from –200 to –51 relative to the announcement date. This variable is used to proxy for idiosyncratic risk or firm-specific characteristics. Morck et al. (2000) demonstrate that stock markets in low-income countries (like China) experience rather synchronous price movements because of poor protection of private property rights and less protection for public investors. The beta (market) risk and stock price synchronicity for Chinese firms are relatively high, implying that investors might not be able to extract useful information from the firm-specific component of risks for risk arbitrage. DBShare is a dummy variable that takes the value of 1 if the firm also issues B-shares.
MJAccrual, which measures earnings management, is the value of abnormal accrual estimated using a modified version of the Jones (1991) model. Earnings management is common in emerging markets where there are large private benefits of control and weak outside investor protection (Leuz et al., 2003). The accounting literature has documented the relation between accruals and future returns and the impact of abnormal accruals on earnings quality (Sloan, 1996; Xie, 2002; Kothari et al., 2005). Kasanen et al. (1996) provide evidence that earnings management can be used to smooth dividends. With reference to ‘Circular on Issuing Supplementary Provisions on Implementation Measures for Suspending and Terminating the Listing of Loss-making Listed Companies’ announced by the CSRC on 4 December 2001 and enforced since 1 January 2002, stock transactions of firms with 2 years of successive negative earnings would be terminated when stock listing is suspended. Therefore, it is possible that Chinese firms might engage in earnings management to avoid suspension. Hence, we adopt the methodology of Dechow et al. (1995) to use the level of abnormal accruals as a measure of earnings management to control for the potential effect of earnings management in our model.
Finally, LnFirmSize is the log value of the total assets. We use equation (5) to examine our Test 1 as to whether a stock dividend or a cash dividend has any effect on share price.
3.5. Determinants of cash dividends and stock dividends
We examine the determinants of cash dividends in the firm through the regression model:
Cash Div-to-Assets = α0 + β1 Ownership + β2 Tobin's Q + β3 Cash Equivalent-to-Assets + β4 EPS + β5 LnTiming + β6 IdioRisk + β7MJAccrual
()We use equation (6) to examine Test 2 that cash dividends are preferred over non-tradable shares. Equation (6) also enables us to test the signalling hypothesis. Cash Div-to-Assets is the ratio of a cash dividend to total assets, which is the dependant variable. We develop four measures of Ownership, NonTPer, DNonTPer, APer and DBShare, to help us shed light on different hypotheses of Ownership. NonTPer is the ratio of non-tradable shares to total shares. DNonTPer is a dummy variable that takes the value of 1 if the ratio of non-tradable shares to total shares is greater than the sample median, and 0 otherwise. APer is the ratio of A-shares to total shares. DBShare is a dummy variable that takes the value of 1 if the firm also issues B-shares.
Tobin's Q is the ratio of market value of equity and debt to replacement cost (Smirlock et al., 1984). Cash Equivalent-to-Assets is the ratio of cash equivalent assets (cash and investment holding) to total assets, which is a measure of firm liquidity. EPS is earnings per share, which is a measure of firm profitability.9
To test our Test 3, we use equation (7):
Stock Dividend = α0 + β1Ownership + β2Cash Div-to-Assets + β3 Tobin's Q + β4Cash Equivalent-to-Assets + β5EPS + β6 LnTiming + β7IdioRisk + β8MJAccrual.
()The dependent variable, Stock Dividend, can be measured in two ways. The first is to use a dummy (1, 0) variable, DSDividend. In this regard, we use a logit model to estimate equation (7) because it results in a more realistic assumption of the residual term resulting from the (1, 0) dummy dependent variable. This logit procedure can avoid the potential estimation problem created by zero values in the data due to the existence of numerous non-dividend payers. Second, we use a continuous variable that measures the dollar value of adjusted stock dividend per share as the dependent variable, ASDPS.10 In both versions, we can also examine the dividend preferential effect for cash and stock dividends and the signalling effect of stock dividends.
4. Empirical results
4.1. Measurement of abnormal share price reaction by market model
Table 1 shows the market model abnormal returns for four separate intervals, the pre-event window (–10 ≤ t ≤ –2) and three announcement windows (–1 ≤ t ≤ +1, –3 ≤ t ≤ +3, –10 ≤ t ≤ +10). Announcement effects are divided into two categories: ‘increase in unexpected earnings’ (Panel A) and ‘decline in unexpected earnings’ (Panel B).
| Sample size | Cumulative abnormal return (t-statistic) | ||||
|---|---|---|---|---|---|
| −10 ≤ t ≤ −2 | −1 ≤ t ≤ +1 | −3 ≤ t ≤ +3 | −10 ≤ t ≤ +10 | ||
| Panel A: Unexpected earnings increase | |||||
| Unexpected cash dividend increase | 665a | 0.0010 | 0.0012 | 0.0024 | –0.0051 |
| (0.27) | (0.60) | (0.75) | (–0.93) | ||
| Unexpected cash dividend decline | 302b | 0.0102 | –0.0034 | –0.0012 | –0.0012 |
| (3.55)*** | (–2.08)** | (–0.48) | (–0.28) | ||
| Zero cash and positive stock dividend | 122c,f | 0.0347 | 0.0082 | 0.0137 | 0.0286 |
| (5.78)*** | (2.36)** | (2.58)*** | (3.12)*** | ||
| Zero cash and zero stock dividend | 660 | –0.0008 | –0.0046 | –0.0081 | –0.0053 |
| (–0.29) | (–2.93)*** | (–3.38)*** | (–1.27) | ||
| Positive stock dividend | 310 | 0.0258 | 0.0091 | 0.0191 | 0.0239 |
| (5.28)*** | (3.21)*** | (4.42)*** | (3.20)*** | ||
| Zero stock dividend | 1439 | –0.0004 | –0.0035 | –0.0058 | –0.0078 |
| (–0.20) | (–3.08)*** | (–3.33)*** | (–2.59)*** | ||
| Panel B: Unexpected earnings decline | |||||
| Unexpected cash dividend increase | 485d | 0.0038 | –0.0013 | –0.0041 | –0.0037 |
| (1.33) | (–0.79) | (–1.61) | (–0.85) | ||
| Unexpected cash dividend increase and zero stock dividend | 406d,g | 0.0010 | –0.0051 | –0.0092 | –0.0102 |
| (0.33) | (–2.76)*** | (–3.29)*** | (–2.10)** | ||
| Unexpected cash dividend decline | 388 | 0.0018 | –0.0089 | –0.0104 | –0.0112 |
| (0.78) | (–6.54)*** | (–5.04)*** | (–3.13)*** | ||
| Zero cash and positive stock dividend | 104e | 0.0147 | 0.0034 | 0.0032 | 0.0072 |
| (2.37)** | (0.95) | (0.59) | (0.76) | ||
| Zero cash and zero stock dividend | 1099e | –0.0084 | –0.0100 | –0.0125 | –0.0111 |
| (–3.72)*** | (–7.64)*** | (–6.26)*** | (–3.20)*** | ||
| Positive stock dividend | 255 | 0.0138 | 0.0091 | 0.0101 | 0.0115 |
| (3.75)*** | (4.28)*** | (3.13)*** | (2.05)** | ||
| Zero stock dividend | 1822 | –0.0048 | –0.0093 | –0.0121 | –0.0112 |
| (–2.82)*** | (–9.56)*** | (–8.11)*** | (–4.36)*** | ||
- ** and *** denote significance at 0.05 and 0.01 levels, respectively.
- a In the ‘Unexpected Cash Dividend Increase’ category, 114 events of 665 have positive stock dividends. As the result is the same after controlling for stock dividends, we do not further split the ‘Unexpected Cash Dividend Increase’ category by positive stock dividend and zero stock dividend.
- b In the ‘Unexpected Cash Dividend Decline’ category, 74 events of 302 have positive stock dividends; we do not further split the sample due to the small sample size.
- c Of the total 782 event of ‘Zero Cash Dividend’ category in Panel A, we focus on the 122 events of ‘Zero Cash and Positive Stock Dividend’ because the combined signal of unexpected earnings increase, zero cash dividend and positive stock dividend is the best combination of a signal as there is good earnings performance, no self-interest to pay cash dividends, and positive stock dividend distribution.
- d We further split the 485 events of ‘Unexpected Cash Dividend Increase’ category into subsamples of positive stock dividend and zero stock dividend. We show the result of 406 events of ‘Unexpected Increase in Cash Dividend and Zero Stock Dividend’ because it is the predicted worst signal. We do not show the 80 events of positive stock dividend as the favourable signal of a positive stock dividend might confound the result of an unfavourable signal of unexpected increase in cash dividend for a small sample.
- e We divide the 1113 events of the ‘Zero Cash Dividend’ subsample in Panel B into positive stock dividend and zero stock dividend to show the difference in the market reaction between positive stock dividend and zero stock dividend.
- f f Predicted best signal.
- g g Predicted worst signal.
We show in Panel A that an unexpected earnings increase with an unexpected cash dividend reduction or no dividends results in increased pre-event abnormal returns (statistically significant). Of the 782 events of zero cash dividends (Zero Cash and Positive Stock Dividend subsample and Zero Cash and Zero Stock Dividend subsample), the 122 events of zero cash and positive stock dividend, which is the combined signal of an unexpected earnings increase, a zero cash dividend, and a positive stock dividend, represent the best combination of signal: good earnings performance, no self-interest payment of cash dividend, and positive stock dividend.
In Panel B, we split the 485 events of unexpected cash dividend increase into subsamples of positive stock dividends and zero stock dividends. The result of 406 events of unexpected increase in cash dividend and zero stock dividend is the predicted worst signal. According to Test 2, the unexpected increase in cash dividend is an indication that controlling shareholders are applying the me-first rule, which the holders of tradable shares might view that as a bad signal. We exclude the 80 events of positive stock dividends in the unexpected cash dividend increase category, as a favourable signal of a positive stock dividend might confound the result of the unfavourable signal of an unexpected increase in the cash dividend.
We also divide the 1113 events of zero cash dividend subsample in Panel B into positive stock dividend (104 events) and zero stock dividend (1099 events) in order to show the difference in the market reaction between a positive stock dividend and a zero stock dividend. The significantly negative abnormal returns for zero cash and zero stock dividend, as compared to the insignificant returns for zero cash and positive stock dividend, suggest that the market reacts favourably to stock dividend distributions. Although the distribution of cash dividend evidences negative and significant abnormal returns, we find that stock dividends can generate positive and significant stock price reaction in firms. In general, the market views the distribution of stock dividends positively and the distribution of cash dividends negatively. The results reported in Table 1 clearly support our Test 1.
We conjecture that for good news (an unexpected earnings increase), there is some information leakage, and the market reaction takes place before the official announcement of earnings and dividends. For bad news, there is no information leakage, and the reaction takes place during the short event window. For robustness purposes, we also measure abnormal returns using a control firm approach, and find similar results.11
4.2. Determinants of cash dividends and stock dividends
We show the results for equation (5) in Table 2 and those for equations (6) and (7) in Table 3. All variables in Tables 2 and 3 have been checked for multicollinearity. Pearson correlations are also checked for all independent variables. If a significant correlation is found for any pair of independent variables in the same model, we conduct a robustness check by running the model using one of the paired variables at a time to make sure that the significance of these variables has not been affected by the presence of the other related variables. The t-statistics are adjusted for heteroscedacity using White's procedure (1980).
|
Panel A: Descriptive statistics |
|||||
|---|---|---|---|---|---|
| Mean | Median | Maximum | Minimum | Standard deviation | |
| CAR (–10 ≤ t ≤ −2) | 0.0000 | –0.0026 | 0.4003 | –0.3360 | 0.0647 |
| CAR (–1 ≤ t ≤ +1) | –0.0047 | –0.0053 | 0.2277 | –0.1975 | 0.0449 |
| CAR (–3 ≤ t ≤ +3) | –0.0059 | –0.0064 | 0.3730 | –0.3039 | 0.0637 |
| CAR (–10 ≤ t ≤ +10) | –0.0066 | –0.0098 | 0.5627 | –0.5639 | 0.1050 |
| UCDPS | –0.0016 | 0.0000 | 0.3982 | –0.4004 | 0.0206 |
| UEPS | –0.1163 | –0.1251 | 0.9984 | –0.9990 | 0.4900 |
| SDPSChg | –0.0197 | 0.0000 | 1.0000 | –1.4901 | 0.3093 |
| Timing (days) | 50.5128 | 52.0000 | 78.0000 | 7.0000 | 15.7860 |
| LnTiming (days) | 3.8599 | 3.9512 | 4.3567 | 1.9459 | 0.3813 |
| IdioRisk | 0.0187 | 0.0173 | 0.2821 | 0.0032 | 0.0124 |
| MJAccrual (RMB) | 0.0147 | 0.0112 | 0.9355 | –1.2707 | 0.1077 |
| LnFirmSize (RMB) | 21.0969 | 20.9841 | 26.9782 | 18.6508 | 0.9172 |
|
Panel B: Regression result |
||||
|---|---|---|---|---|
| Time Interval for CAR (equation 5) | ||||
| −10 ≤ t ≤ −2 | −1 ≤ t ≤ +1 | − 3 ≤ t ≤ +3 | −10 ≤ t ≤ +10 | |
| Beta coefficient (t-statistic) | ||||
| Intercept | 0.0986 (3.33) | –0.0306 (–1.50) | 0.0091 (0.28) | 0.1306 (2.60) |
| UCDPS | 0.0422 (0.76) | 0.0534 (1.65) | 0.0671 (0.88) | 0.0139 (0.13) |
| SDPSChg | 0.0217 (5.21)*** | 0.0096 (3.38)*** | 0.0151 (4.01)*** | 0.0234 (3.79)*** |
| UEPS | 0.0023 (0.90) | 0.0045 (2.60)*** | 0.0057 (2.38)** | 0.0052 (1.29) |
| LnTiming | –0.0160 (–4.69)*** | –0.0042 (–1.77) | –0.0088 (–2.62)*** | –0.0108 (–2.12)** |
| IdioRisk | –0.9009 (–6.57)*** | –0.2575 (–4.25)*** | –0.5890 (–6.75)*** | –1.9212 (–10.01)*** |
| DBShare | 0.0043 (0.92) | 0.0001 (0.02) | 0.0055 (1.07) | 0.0183 (2.36)** |
| MJAccrual | 0.0179 (1.72) | –0.0081 (–1.12) | –0.0011 (–0.11) | –0.0005 (–0.03) |
| LnFirmSize | –0.0009 (–0.73) | 0.0023 (2.58)*** | 0.0015 (1.04) | –0.0028 (–1.29) |
| Adjusted R2 | 0.0477 | 0.0157 | 0.0235 | 0.0562 |
| F | 20.0838 | 7.0850 | 10.1461 | 23.6769 |
| p-value | 0.00 | 0.00 | 0.00 | 0.00 |
| Number of observations | 3048 | 3048 | 3048 | 3048 |
- ** and *** denote significance at the 0.05 and 0.01 levels, respectively. Cumulative abnormal returns (CAR) is the cumulative abnormal return over the different event windows (–10 ≤ t ≤ –2, –1 ≤ t ≤ +1, –3 ≤ t ≤ +3, and –10 ≤ t ≤ +10). UCDPS is the unexpected change in the cash dividends per share between cashdividend per share in year y and cash dividend per share in year y – 1 with an industry adjustment factor scaled by share price on day t = 0. Of the 3048 observations of UCDPS, 1623 observations pay positive cash dividend (53.25 per cent) and 1425 observations pay zero cash dividend (46.75 per cent). Of the 3048 observations, 488 observations pay stock dividend (16.01 per cent) and 2560 observations do not pay stock dividend (83.99 per cent). SDPSChg is the percentage change in stock dividend per share. UEPS is the unexpected change in the earnings per share between earnings per share in year y and earnings per year in year y – 1 with an industry adjustment factor. Timing is the number of days from the fiscal year-end date (31 December) to the announcement date. LnTiming is the log value of the number of trading days from the fiscal year-end date (31 December) to the announcement date. IdioRisk is the standard deviation of the residual between actual return and estimated return from the market model over the 150 days in the estimation period from –200 to –51 relative to the announcement date. DBShare is a dummy variable which takes the value of 1 if the firm also issues B-shares. MJAccrual is the value of the abnormal accrual estimated using the modified version of the Jones (1991) model. LnFirmSize is the log value of the total assets. t-statistics are adjusted for heteroscedacity using White's procedure (1980).
|
Panel A: Descriptive statistics |
|||||
|---|---|---|---|---|---|
| Mean | Median | Maximum | Minimum | Standard deviation | |
| NonTPer | 0.6310 | 0.6400 | 0.9966 | 0.0000 | 0.1274 |
| APer | 0.3513 | 0.3462 | 1.0000 | 0.0034 | 0.1338 |
| CDPS (RMB) | 0.0606 | 0.0000 | 1.8684 | 0.0000 | 0.0970 |
| Cash Div-to-Assets (RMB) | 0.0109 | 0.0000 | 0.1988 | 0.0000 | 0.0176 |
| ASDPS (RMB) | 0.7424 | 0.1540 | 24.3609 | 0.0000 | 1.9891 |
| Price (RMB) | 8.8340 | 7.9800 | 53.9500 | 0.4260 | 4.6883 |
| Tobin's Q (RMB) | 2.6051 | 2.2322 | 9.9925 | 0.5201 | 1.4106 |
| Cash Equivalent-to-Assets (RMB) | 0.1514 | 0.1241 | 0.8783 | 0.0001 | 0.1145 |
| EPS (RMB) | 0.1747 | 0.1828 | 2.7082 | –2.3259 | 0.3073 |
| LnTiming (days) | 3.9571 | 4.0073 | 5.8551 | 1.9459 | 0.4971 |
| IdioRisk | 0.0199 | 0.0177 | 0.3339 | 0.0003 | 0.0167 |
| MJAccrual | 0.0051 | 0.0051 | 1.2928 | –1.2707 | 0.1101 |
|
Panel B: Regression results |
|||||||||
|---|---|---|---|---|---|---|---|---|---|
| Dependent variable | Equation (6) Cash Div-to-Asset | Equation (7) DSDividend | Equation (7) ASDPS | ||||||
| Independent variable | Beta coefficient (t-statistic) | Beta coefficient (z-statistic) | Beta coefficient (t-statistic) | ||||||
| (a) | (b) | (c) | (a) | (b) | (c) | (a) | (b) | (c) | |
| Intercept | 0.0033 | 0.0111 | 0.0191 | –1.3593 | –2.2190 | –3.1149 | 1.1275 | 0.8842 | 0.6346 |
| (1.63) | (6.09) | (10.08) | (–3.40) | (–6.28) | (–8.39) | (4.54) | (3.91) | (2.70) | |
| NonTPer | 0.0154 | –1.3752 | –0.3631 | ||||||
| (11.04)*** | (–4.65)*** | (–2.19)** | |||||||
| DNonTPer | 0.0037 | –0.0834 | 0.0118 | ||||||
| (9.86)*** | (–1.14) | (0.26) | |||||||
| APer | –0.0147 | 1.9194 | 0.6448 | ||||||
| (–11.10)*** | (6.47)*** | (3.88)*** | |||||||
| DBshare | –0.0022 | –0.0102 | 0.0839 | ||||||
| (–2.42)** | (–0.05) | (0.93) | |||||||
| Cash Div-to-Assets | –20.6400 | –21.6084 | –20.0895 | –9.0240 | –9.3982 | –8.7964 | |||
| (–7.40)*** | (–7.74)*** | (–7.25)*** | (–5.32)*** | (–5.48)*** | (–5.20)*** | ||||
| Tobin's Q | 0.0001 | 0.0001 | 0.0001 | –0.0895 | –0.1016 | –0.0787 | 0.0310 | 0.0274 | 0.0344 |
| (0.71) | (0.80) | (0.62) | (–3.09)*** | (–3.52)*** | (–2.73)*** | (1.77) | (1.57) | (1.95) | |
| Cash Equivalent-to-Assets | 0.0228 | 0.0223 | 0.0230 | 0.3108 | 0.3944 | 0.2283 | 0.4043 | 0.4322 | 0.3710 |
| (12.20)*** | (12.00)*** | (12.28)*** | (0.95) | (1.21) | (0.70) | (1.79) | (1.91) | (1.65) | |
| EPS | 0.0172 | 0.0172 | 0.0169 | 3.5147 | 3.5385 | 3.6438 | 1.7211 | 1.7220 | 1.7390 |
| (20.62)*** | (20.58)*** | (20.06)*** | (14.74)*** | (14.78)*** | (14.97)*** | (13.93)*** | (13.94)*** | (14.11)*** | |
| LnTiming | –0.0021 | –0.0021 | –0.0023 | –0.0625 | –0.0439 | –0.0244 | –0.1373 | –0.1323 | –0.1319 |
| (–5.00)*** | (–4.97)*** | (–5.43)*** | (–0.77) | (–0.54) | (–0.30) | (–2.66)*** | (–2.55)** | (–2.55)** | |
| IdioRisk | –0.0197 | –0.0188 | –0.0197 | 3.5801 | 3.2464 | 3.6968 | 2.2382 | 2.1139 | 2.3328 |
| (–1.87) | (–1.80) | (–1.87) | (2.26)** | (2.05)** | (2.34)** | (1.28) | (1.21) | (1.33) | |
| MJAccrual | –0.0073 | –0.0073 | –0.0069 | 0.5349 | 0.5648 | 0.4535 | –0.2912 | –0.2829 | –0.3184 |
| (–4.10)*** | (–4.05)*** | (–3.87)*** | (1.47) | (1.56) | (1.26) | (–0.90) | (–0.88) | (–0.99) | |
| Adjusted R2 | 0.1536 | 0.1523 | 0.1523 | 0.0686 | 0.0681 | 0.0696 | |||
| F | 197.0069 | 195.0575 | 170.8165 | 70.5797 | 70.0172 | 63.8180 | |||
| p-value | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | |||
| LR statistic | 640.1375 | 617.5002 | 666.0651 | ||||||
| p-value | 0.00 | 0.00 | 0.00 | ||||||
| Number of observations | 7560 | 7560 | 7560 | 7560 | 7560 | 7560 | 7560 | 7560 | 7560 |
- ** and *** denote significance at 0.05 and 0.01 levels, respectively. NonTPer is the ratio of non-tradable shares to total shares. DNonTPer is a dummy variable that takes a value of 1 if the ratio of non-tradable shares to total shares is greater than the sample median, and 0 otherwise. APer is the ratio of A-shares to total shares. DBShare is a dummy variable that takes the value of 1 if the firm also issues B-shares. CDPS is the value of cash dividend. Of the 7560 observations of CDPS, 3392 observations pay positive cash dividend (44.87 per cent) and 4168 observations pay zero cash dividend (55.13 per cent). Cash Div-to-Asset is the ratio of cash dividend to total asset. DSDividend is a dummy variable that takes the value of 1 if the event is with a distribution of stock dividends, and 0 otherwise. Of the 7560 observations of DSDividend, 981 observations pay stock dividends (12.98 per cent) and 6579 observations do not pay a stock dividend (87.02 per cent). ASDPS is the value of the adjusted stock dividend. Price is the share price. Tobin's Q is the market value of equity and debt to replacement cost. Cash Equivalent-to-Assets is the ratio of cash equivalent assets (cash and investment holding) to total asset. EPS is earnings per share. LnTiming is the log value of the number of trading days from the fiscal year-end date (31 December) to the announcement date. IdioRisk is the standard deviation of the residual between actual return and estimated return from the market model over the 150 days in the estimation period from –200 to –51 relative to the announcement date. MJAccrual is the value of the abnormal accrual estimated using a modified version of the Jones (1991) model. t-statistics are adjusted for heteroscedacity using White's procedure (1980).
Panel A of Table 2 provides the descriptive statistics for all variables in equation (5). There are negative mean and median values for CAR, particularly for the periods –1 ≤ t ≤ +1 and –3 ≤ t ≤ +3. The mean and median number of trading days between fiscal year-end and announcement day are 50.51 and 52, respectively. Assuming there are about 22 trading days in 1 month, the mean of 50.51 days indicates that firms announce earnings, on average, in March or later.
The results in Panel B of Table 2 echo the results for the announcement effect in Table 1 that cash dividends convey little information to the market. Unexpected earnings per share (UEPS), on the other hand, show a significant and positive relation with the CAR over –1 ≤ t ≤ +1 and –3 ≤ t ≤ +3.12 In addition, the variable for a stock dividend change, SDPSChg, exerts a strong positive effect on CAR, indicating that paying a stock dividend sends a favourable signal to the market.
Control variables for earnings management (MJAccrual) is not significant statistically, and for firm size (LnFirmSize) is significant only in the –1 ≤ t ≤ +1 period. LnTiming is negatively significant, suggesting that the market reacts positively to early announcement than to late announcement. IdioRisk, a proxy for the idiosyncratic risk of the firm, is negatively significant. It is used to measure firm uncertainty or firm-specific characteristics. After controlling for firm size, IdioRisk is still highly significant. It seems clear that the effect of IdioRisk is not due to firm size or market volatility.13 DBShare is significant for the –10 ≤ t ≤ +10 period, indicating that the market slightly favours firms with outstanding B-shares.
We examine the different hypotheses on the determinants of cash and stock dividends, using Cash Div-to-Assets in equation (6) and two measures of stock dividends, DSDividend and ASDPS, in equation (7).14 Table 3 documents the results. Panel A provides the summary statistics of variables of equations (6) and (7). The mean value for the ratio of non-tradable shares to total shares is 0.6310, indicating a majority of shares are non-tradable shares.
The results on the determinants of cash dividends in equation (6) are interesting. For the separate runs of regressions including the ownership variables (NonTPer,15 DNonTPer, APer and DBShare), the results of positive coefficients on NonTPer and DNonTPer indicate that the larger the non-tradable shares, the higher the cash dividend payout. The coefficient on NonTPer is 0.0154, which is significant at the 1 per cent level. Similarly, we show that a lower cash dividend is associated with a higher percentage of A-shares (APer) and with B-shares listing (DBShare).
The variable of LnTiming is negative, implying that firms paying cash dividends typically announce early.16 The coefficients on cash equivalent-to-assets are all positive and highly significant. That is, firms with a higher cash equivalent-to-total assets ratio pay more cash dividends, implying that these firms have sufficient cash for cash dividend payments and do not necessarily expropriate wealth from the public shareholders. That controlling shareholders prefer cash dividends to capital gains makes sense, because their shares are not tradable and, therefore, they cannot benefit directly from price appreciations of the firm.
The effect of MJAccrual on cash dividends is negative and statistically significant across the three regressions.17 That is, the more earnings management (i.e. earnings inflation), the lower the cash dividend payout. As controlling shareholders cannot benefit from price appreciation of firm, they should be less likely to manage earnings. The negative relation between Cash Div-to-Assets and MJAccrual supports the assertion that the cash dividend-paying firms are not playing the game of earnings inflation to trick investors.
The logit results for DSDividend of equation (7) indicate that firms pay a stock dividend because of more A-share stockholders; this factor has a positive effect on the stock dividend. Coefficient on APer is statistically positive, while the coefficient on NonTPer is statistically negative. The results imply that public or A-share shareholders, and not controlling shareholders, prefer stock dividends. Cash Div-to-Assets is negatively related to DSDividend, indicating that stock and cash dividends are inversely related. We see this negative relation as an indirect consequence of the dividend preference between tradable-share holders and non-tradable-share holders.
Tobin's Q is negatively and significantly related to DSDividend. That is, the higher the Tobin's Q, the lower the probability firms will pay stock dividend. This result shows that a higher (lower) growth firm, reflected by a higher (lower) Tobin's Q actually has a greater chance to pay cash (stock) dividend. This result is consistent with our findings that firms with more non-tradable shares (owned by government and state-owned institutions) pay more cash dividend. In fact, it is highly possible that when the Chinese government selectively privatizes its state-owned enterprises, it has more privileged information on the growth potential of all these institutions. Therefore, it is reasonable for the government and its related agencies to maintain a bigger portion of the share holdings in firms with a better growth potential. Indeed, many firms with a stronger government influence (higher non-tradable shares) are those in protected industries and/or in product (including services) markets enjoying high growth. Consequently, firms with stronger growth and higher cash dividend payouts are simply characteristics of high government-controlled firms.
The earnings measure, EPS, has a positive and significant effect on DSDividend. This result supports the hypothesis that a stock dividend serves a signalling purpose. As firms want to conserve cash for investment projects, a stock dividend policy signals to the market that they are doing well. IdioRisk has a positive and significant effect on DSDividend. If we follow the conjecture that idiosyncratic risk (measured by IdioRisk) reflects the amount of firm-specific information available (Morck et al., 2000), then our result shows that firms with higher firm level information have a greater probability to pay stock dividend.
When we compare the results of the cash dividend regression and the stock dividend dummy logit model, we find a positive relation between EPS and dividend payout (cash dividend or stock dividend), indicating that, whatever the form of dividend distribution, firms with stronger earnings tend to distribute more. That is, profitability itself (earnings) might not be enough to allow us to differentiate whether firms pay cash or stock dividends.
When the Chinese firms choose to pay a cash or stock dividend, there are discretionary and circumstantial characteristics. Our results clearly indicate that the level of cash dividend payout is a positive function of the percentage of non-tradable shares, whereas a stock dividend is positively related to the percentage of tradable A-shares. Such a dividend pattern reflects the unique capital market behaviour in mainland China. Firms have to have cash before they can choose to pay a cash dividend. The positively significant relation between cash balances and cash dividend is consistent with this requirement.18 This also explains why we do not observe a positive relation between cash balances (Cash Equivalent-to-Assets) and stock dividend.
Results of the regression using the dollar value of stock dividend per share (ASDPS) are similar to those using DSDividend in testing Tests 2 and 3 for the ownership variables. However, LnTiming affects the size of stock dividends negatively, implying that an earlier dividend announcement implies a higher amount of stock dividend.
5. Conclusion
We examine data for Chinese firms listed on the Shanghai and Shenzhen Stock Exchanges to explore reasons why firms pay out stock or cash dividends. We find several interesting results. First, firms show positively significant pre-event abnormal returns if there is unexpected earnings increase but with an unexpected cash dividend reduction or no cash dividend. Firms with unexpected earnings declines and zero cash dividends display significantly negative abnormal returns, but significantly positive returns in the case of stock dividends. Stock dividends typically generate positive market reactions, despite earnings increases or declines. Our results indicate that the market views stock dividends positively and cash dividends negatively.
Higher non-tradable share ownership implies more cash dividends, a result supporting the me-first rule effect that non-tradable-share holders want an immediate financial gain in the form of a cash dividend. Cash dividends are not driven by profitability as measured by return on assets, but are significantly related to higher earnings per share. To a certain extent, cash dividends provide some signalling effects to the market. The distribution of stock dividends in a logit model is positively related to the number of public shares (a result consistent with the market impetus hypothesis) and higher earnings, a proxy for profitable investment opportunities.
Our results shed light on the dividend behaviour of Chinese firms and enable us to better understand the rationale of paying stock or cash dividends in these firms. There is clear evidence to support the three tests proposed. The results might explain why firms pay stock dividends in other emerging markets as well. There is a clear preferential effect between stock and cash dividends in the logit model, a result supporting our hypothesis. China represents an imperfect market. Our results support the agency theory problem, and suggest some policy implications for the Chinese government formulating rules on dividend policies.
