Do Changes in Dividends Have Larger Influence on Market Value in Emerging Markets than in Developed Markets? Evidence from Germany and China.

(1)

Do Changes in Dividends Have Larger Influence on Market Value

in Emerging Markets than in Developed Markets?

Evidence from Germany and China.

Master Thesis Ruiheng Yang 28th November, 2012

Abstract

This paper investigates the long-run economic meaning of dividends and the influence of dividends on market value for Germany and China, which differ much in asymmetric information and agency problems. Because dividends may assist in reducing asymmetric information and agency problems and because these problems are likely to be larger in China, I assume the mitigating effects of dividends will be larger in China. Panel regression analysis shows that changes in dividends have indeed a greater influence on market value of Chinese firms.

JEL classification

G35

Keywords

Dividends, market value, asymmetric information, agency problems

Msc International Financial Management Msc Business and Economics

Faculty of Economics and Business Faculty of Social Sciences

(2)

1. Introduction

The discussion on dividend and its influence on market value has been an important topic in the domain of corporate finance for a long time. As noted by Miller and Modigliani (1961), “the effect

of a firm's dividend policy on the current price of its shares is a matter of considerable importance, not only to the corporate officials who must set the policy, but to investors planning portfolios and to economists seeking to understand and appraise the functioning of the capital markets”.

Assuming perfect market and investor rationality, they posit that a firm‟s market value is independent of its dividend policy (Miller & Modigliani, 1961).

Miller and Modigliani‟s (1961) work lays the foundation for future studies, in which researchers have always been attempting to see whether dividend is truly irrelevant after introducing market imperfections in the model. Tax is the first imperfection added and directly complicates the discussion since firms pay out dividends even when taxation is biased in favor of capital gains – the famous “Dividend Puzzle” (Black, 1976). Later, researchers develop the signaling theory (Bhattacharya, 1979; John & Williams, 1985; Miller & Rock, 1985) and the agency approach (Easterbrook, 1984; Jensen, 1986) in order to explain why firms pay out dividends even in the presence of non-neutral taxation. Both of the theories view dividends as a tool to mitigate problems between managers and investors, either asymmetric information or agency problems.

(3)

Although the signaling theory and the agency approach provide a theoretical framework as how dividend will influence market value in the presence of asymmetric information and agency problems, the contest between them makes it hard to distinguish between either one of them. Furthermore, they do not tell whether the influence will vary given the different level of asymmetric information and agency problems across country. Similarly, researchers using event studies do not explain why the results vary over time and across country and they only capture market return around the announcement day of changes in dividends. La Porta et al. (2000a), Faccio et al. (2001) and Avizian et al. (2003) do show that dividend policy is different around the world because of varying determinants but whether the influence of dividends on market value is also different remains unanswered.

Noticing all the problems unaddressed, I set out to investigate the long-run economic meaning of dividends. According to the signaling theory and the agency approach, dividends matter, because they mitigate problems between managers and investors, either asymmetric information or agency problems. It is also quite often the case that emerging markets have more asymmetric information and agency problems. Therefore, I attempt to answer the following research question in this paper:

Do changes in dividends have larger influence on market value in emerging markets than in developed markets?

Important to be mentioned in advance is that the scope of this paper will be discussing the meaning of dividends in the long-run. Also, my discussion on changes in dividends focuses on changes in dividends which are not fully expected by investors. In addition, I mainly refer to asymmetric information and agency problems between managers and investors in order to avoid over-complication. Moreover, I focus on dividends only rather than including the alternatives of them as well (e.g. repurchases). Finally, I use China as the representative of an emerging market and Germany as the representative of a developed market.

(4)

comparing the results from the segmented samples and by interpreting those from the combined sample and whether tax effects are excluded or included. Robustness tests also show that the results do not vary much when cross-section and time-period random effects are used and that the finding stands whether the market is advancing or declining.

After doing so, I add to the knowledge on dividend relevance by proposing cross-country variation of the influence of changes in dividends. I also seek explanation from asymmetric information and agency problems, standing in line with the signaling theory and the agency approach. Moreover, my work adds to the work done by researchers using event study techniques and researchers like La Porta et al. (2000a), Faccio et al. (2001) and Avizian et al. (2003). Implications for managers are that in emerging markets (as learnt from China in comparison with Germany), market value increases relatively more by paying out dividends, highlighting the importance of dividends for managers from these markets.

The organization of this paper is as follow: Section 2 provides in-depth description of the literatures. Section 3 presents the reasoning which underlies this paper‟s argument and the hypothesis. Section 4 introduces the methodology of this paper. Section 5 introduces the sample and data. Section 6 presents the results from panel regression and the results of 2 robustness tests. Section 7 concludes the whole paper.

2. Literature review

2.1. Theories

(5)

always cancel each other out, given the firm‟s investment decision. Therefore, dividend policy does not affect market value, which, is solely determined by the firm‟s investment policy. In this line of reasoning, changes in a firm‟s dividend policy should also not cause changes in its market value (Miller & Modigliani, 1961).

Although Miller and Modigliani‟s (1961) work lays the foundation for later studies on dividend, it is built on very strict conditions which contradict the fact that the market is not perfect. Among all the imperfections, tax catches researchers‟ early attention. As noted by Miller and Modigliani (1961), tax is “undoubtedly the major systematic imperfection in the market.” Brennan (1970) shows that securities with high dividend yields would have higher expected returns (before tax) than those with low dividend yields, given that taxes on dividends are higher than those on capital gains. Miller (1976) and Black (1976) also address this differential tax issue.1 To date, it is common for researchers to account for differential tax issue when studying topics related to dividend (e.g. La Porta et al., 2000a).

Tax is indeed an important issue, while the discussion of it in the domain of the influence of dividend on market value is often associated with theories rooted in asymmetric information and agency problems. The signaling theory relaxes Miller and Modigliani‟s (1961) assumption that everybody will have equal and costless access to information. It is developed by a series of researchers, whose studies concentrate on explaining why firms pay dividends even in the presence of non-neutral taxation (Bhattacharya, 1979; John & Williams, 1985; Miller & Rock, 1985; Bernheim, 1991). The main reasoning behind the signaling theory lies in that managers possess more information about firms‟ profitability than investors, in which case they use dividends as costly signals to convey information not previously known to investors; as a result, managers can change investors‟ perceptions about firms‟ future earnings prospects, thus influencing market value.

As mentioned earlier, another strand of researchers incorporate agency problems into the study on

(6)

dividends and market value. Embedding agency problems, conflicts of interests between managers and investors enter the study on corporate finance quite early and agency problems are regarded as the center of analysis of the modern corporation (Berle & Means, 1931; Jensen & Meckling, 1976); however, it is until much later that researchers take the agency approach to look at issues regarding dividends. Easterbrook (1984) first suggests that paying out dividends forces managers to raise capital in financial markets, as a result of which firms are subject to monitoring by outside investors. Putting forward the free cash flow hypothesis, Jensen (1986), on the other hand, argues that dividends can reduce cash available to managers, preventing them from spending excess cash on investment projects with low returns and unnecessary perks. Later, more researchers have developed theories in support of Easterbrook (1984) and Jensen (1986), who also view dividends as the tool to alleviate agency problems (Fluck, 1998 & 1999; Myers, 2000; Gomes, 2000). Of all these theories following the agency approach, although they differ on how investors force managers to pay out dividends, their essence is the same; that is, failure to disgorge cash harms investors‟ interests because it leads to the diversion or waste of cash (La Porta et al., 2000a). Therefore, in the logic of these theories, investors should prefer dividends over retained earnings.

The agency approach differs from the signaling theory in that the former roots in conflicts of interests between managers and investors while the latter implicitly assumes managers will act in investors‟ best interests in the situation of asymmetric information (Liang et al., 2010). However, both the signaling theory and the agency approach view dividends as a tool to mitigate problems between managers and investors, either asymmetric information or agency problems. They both move away from the assumptions of Miller and Modigliani (1961), implying that changes in dividends will evoke stock price reactions and thus may not be truly irrelevant to firms‟ market values (Blau & Fuller, 2010).

2.2. Empirical evidence

(7)

Later, empirical studies on the signaling theory produce more mixed outcomes. Michaely et al. (1995) and Grullon et al. (2002) report significant relations between stock price returns and changes in dividends (or initiations and omissions). However, Benartzi et al. (1997) find that changes in dividends contain information of past earnings rather than future earnings changes, which contradicts the logic of the signaling theory. Furthermore, they discover that a decrease in dividends is followed by a clear pattern of earnings increase (Benartzi et al. 1997), in line with Healy and Palepu‟s (1988) finding. Grullon et al. (2002) also confirm these researchers‟ results. In a word, changes in dividends do impact market value (stock prices) but it may not necessarily be the case that dividends contain positive information, as argued in the signaling theory.

Compared with the signaling theory, it is more difficult to derive precise empirical implications from the agency approach since it is not as structured as the signaling models (Allen & Michaely, 2003). Studies following the free cash flow hypothesis have relatively clear implications for the relationship between overinvestment problems and the impact of changes in dividends on market value (Allen & Michaely, 2003). Lang and Litzenberger (1989) find out that changes in dividends have greater effects for firms with higher likelihood of overinvesting (measured by growth opportunities using Tobin‟s Q), consistent with the free cash flow hypothesis suggested by Jensen (1986). Through investigating the relationship between excess funds and firms‟ payout policies, Lie (2000) not only shows that firms increasing their dividends have more cash than their peers but also concludes that the market reacts to the announcements of special dividends and that these reactions are positively related to firms‟ amount of excess cash and negatively related to investment opportunities. Lie‟s (2000) findings provide more direct support for the arguments in the agency approach since they imply that investors prefer dividends because paying out dividends prevents cash from being wasted by managers. On the other hand, Yoon and Starks (1995) employ a sample covering a longer time period and report findings inconsistent with the free cash flow hypothesis. Similar to empirical studies done in light of the signaling theory, the results reported in above researches provide mixed support for the argument that paying out dividends avoids the waste of cash, although market reactions to dividends are evident.

(8)

in dividends do cause market reactions although the explanations for these reactions from the theories receive mixed support. Among all the empirical studies related to changes in dividends, a lot of researchers use event study techniques and measure abnormal returns, in order to capture market reactions to dividends announcements (announcement-day-effect). Regardless of what theory they use to explain these reactions, the signs reported in their studies are uniform (positive abnormal returns for increases and the other way around), while the sizes of reactions (values of abnormal returns) vary for studies done in different countries and over different time periods. For example, Bajaj and Vijh (1990) report average cumulative abnormal returns of 1.04% for dividend increase and -0.53% for dividend decrease for a sample of US firms during the period 1962 to 1987; while Alangar et al. (1999)‟s research on US firms from 1976 to 1990 finds 1.03% and -4.20% respectively. Lonie et al. (1996) present average cumulative abnormal returns of 2.03% and -2.15% respectively for increases and decreases, for UK firms in 1991. Gurgul et al. (2003) find average cumulative abnormal returns of 1.50% and -1.78% respectively for increase and decrease with a 10-year analysis on Australian firms from 1992-2002. Capstaff et al. (2004) show smaller reactions in Norway, with average cumulative abnormal returns being 1.00% and -0.59% respectively for increases and decreases. Apart from the variation of outcomes over country and time, another point which could be observed from this line of researches is that when firms omit dividends, the impacts will be much larger than those caused by announcements of dividend increases or decreases (Healy & Palepu, 1988; Michaely et al., 1995; Alangar et al., 1999).

(9)

2.3. Variation

In line with the observation that empirical outcomes vary over time, some researchers have implied that investors‟ demand for dividends is time-varying (Baker & Wurgler, 2004). The catering theory of dividends proposed by Baker and Wurgler (2004) argue that firms pay out dividends when investors put a stock premium on dividends payers, which implicitly suggests that market reactions to dividends announcement are different over time because of time-varying demand, and that firms counter this effect by taking demand into consideration. However, they further argue that the catering theory has more explanatory power than the other theories of dividends and their finding that firms omit dividends when demand is low is inconsistent with the observation that market reacts more aggressively to omission, as reported by Healy and Palepu (1988), Michaely et al. (1995) and Alangar et al. (1999). This problem is addressed by Ali and Urcan (2012), who propose a synthesis between the catering theory and the signaling theory by arguing that firms signal future earnings prospects rather than catering to investors‟ needs when demand for dividends is low.

Recently, Fuller and Goldstein (2011) find results similar to Baker and Wurgler (2004). They model firms‟ market returns as a function of a dummy indicating whether a firm pays dividend or not and state that dividend payers outperform non-dividend payers more in declining market, which gives indirect relevance for the relation between dividends and market value. Also in their paper, it is inconclusive which theories of dividends explain the results better (Fuller & Goldstein, 2011). On the other hand, their work does not directly focus on changes in dividends and their influence on market value and their use of a dummy variable does not capture the magnitude of dividends.

(10)

have performed cross-country analysis on dividend policy; these researchers base their arguments on asymmetric information and agency problems, which underlie the reasoning in the signaling theory and the free cash flow hypothesis, instead of going for either one of them for explanation (La Porta et al., 2000a; Faccio et al., 2001; Avizian et al., 2003). Among them, La Porta et al. (2000a) confirm the positive relationship between dividend payout ratio and the level of investor protection while Faccio et al. (2001) find out that firms with group-affiliation feature or multiple large shareholders pay higher dividends in Europe than in East Asia. Moreover, Avizian et al. (2003) find that firms in emerging markets have less stable dividends compared to their US counterparts. Unfortunately, these studies focus on answering how much dividends firms pay and to whom in different countries but do not elaborate on the relationship between dividends and market value in different countries.

3. Hypothesis

3.1. Some elaborations

(11)

than event study techniques. In line with such studies, the influence of changes in dividends on market value will be focused on long-term effects. Moreover, investors may not fully expect changes in dividends because dividends are sticky (Lintner, 1956), although regularly-paid dividends are generally already expected and will not cause change in market value (given the market is at least semi-strong, they are already reflected in market value). To avoid confusion and complication, I assume all the changes in dividends investigated in this paper are not fully expected by investors, as a result of which they may cause change in market value.

As summarized in the literature review, empirical outcomes reported by researchers using event study techniques to capture announcement-day-effect tend to vary across countries, which provides this paper with a good starting point to investigate whether changes in dividends on market value differ between developed and emerging markets. Furthermore, researchers like La Porta et al. (2000), Faccio et al. (2001) and Avizian et al. (2003) have provided a theoretical basis for viewing and explaining this cross-country difference from the perspectives of asymmetric information and agency problems.

(12)

corporate finance literature, thus detangling them may lead to over-complication.2

3.2. Developed vs. Emerging

It is well acknowledged that, compared with emerging markets, developed markets have less asymmetric information between managers and investors (more transparent) because they have undergone a long history of development and have stricter laws and regulations regarding disclosure. An overview of this issue can be developed from the Transparency and Disclosure scores (T&D score) compiled by Standard & Poor (S&P), which are good indicators of the level of transparency and are commonly used by corporate researchers (e.g. Patel et al., 2002; Aksu & Kosedag, 2006). This score analyzes disclosure from three perspectives: disclosure of ownership structure and investor relation (OwnStr), financial transparency and disclosure (FinDisc) and disclosure of the board and management structure and processes (BrdMgmt).3 Table 1 presents the T&D scores in different geographical regions reported by S&P (2002). The higher the level of transparency and information disclosure, the lower the level of asymmetric information. Therefore, it can be learnt from Table 1 that emerging markets (in Latin America and Emerging Asia) show

2

In corporate finance literature, asymmetric information and agency problems are inter-related. Asymmetric information can be viewed as a source of agency problems (Barnea et al., 1985; Chowdhury, 2004; Marnet, 2008) and increasing disclosure of information may help protect investors and can consequently mitigate agency problems (Klapper & Love, 2004).

3

In light of the signaling theory, the most relevant score to this paper‟s topic is financial transparency and disclosure (FinDisc).

Table 1: Transparency and Disclosure scores (T&D scores) in different geographical regions

Transparency and Disclosure scores (T&D scores) are computed by S&P and analyze disclosure from three perspectives: disclosure of ownership structure and investor relation (OwnStr), financial transparency and disclosure (FinDisc) and disclosure of the board and management structure and processes (BrdMgmt).

(in %) Composite score OwnStr DinDIsc BrdMgmt Number of firms

Europe 58 46 73 51 351

UK 70 54 81 70 124

Non-UK 51 41 69 41 227

US (annuals) 42 25 66 31 500

US (combined data sources) 70 52 77 78 500

Japan 61 70 76 37 150

Asia-Pacific 48 41 60 42 99

Latin America 31 28 58 18 89

Emerging Asia 40 39 54 27 253

(13)

lower levels of transparency and disclosure than those in developed markets (in Europe, UK, US, Japan.), which indicates that problems with respect to asymmetric information are more severe in emerging markets.

One source of agency problems arises from the fact that managers are better able to pursue their self-interest driven goals when investors know less about the state of affairs of the firm, (Barnea et al., 1985; Chowdhury, 2004; Marnet, 2008), including the managers‟ performance (Marnet, 2008). Therefore, agency problems arising from it should be more severe in emerging markets, which, as discussed already, have lower level of transparency and disclosure. Despite of this, agency problems are essentially problems from the contractual relationship between the agent and the principals (in this paper, managers and investors) (Jensen & Meckling, 1976), thus the severity of agency problems can be viewed by how well this relationship (or “contract”) is protected. This leads to the discussion on the strength of the legal system, the enforcement of law, investor protection and ownership structure (e.g. La Porta et al., 1998, 1999 & 2000b; Claessens et al., 2000; Berkowitz et al., 2003), on which emerging markets and developed markets also differ. Because of weak judicial systems, poor enforcement of law, the “contract” is poorly protected in emerging markets and the situation is worsened by commonly found ownership concentration or even pyramidal ownership structure, cross-ownership and dual class shares (OECD, 2007). In these circumstances, managers are not only less monitored or less punished by law but also have more reasons to act outside all investors‟ best interests (e.g. when managers satisfy only the controlling parties‟ needs). Therefore, emerging markets have more agency problems than developed markets.

3.3. Changes in dividends & Markets value: Developed vs. Emerging

(14)

Compared with those in developed markets, investors in emerging markets know less about a firm‟s future prospects because the market is less transparent and less disclosed information is available to them, in which circumstances a change in dividend carries important information about the firm‟s future prospects and helps ease the information asymmetry between managers and investors. However, thinking the other way around, changes in dividends play less important roles in easing this kind of asymmetric information in developed markets. The idea is that changes in dividends‟ ability in mitigating asymmetric information matters more in places where more information asymmetry is observed. Therefore, it is proposed that changes in dividends have larger influence on market value in emerging markets.

On the other hand, compared with developed markets, managers in emerging markets face less constrains and are more likely to act out of investors‟ best interests either because of more information asymmetry and weaker protection of the “contract” or greater conflicts of interests rooted in problems from ownership structure. Therefore, investors in emerging markets should prefer increases in dividends more than those in developed markets because less cash is now at the hand of managers; and the opposite effects apply to decreases in dividends. Again, this also leads to the proposition that changes in dividends have larger influence on market value in emerging markets.

The final point relates asymmetric information and agency problems to my assumption that changes in dividends are not fully expected by investors. In emerging markets, investors‟ abilities in predicting changes in dividends may be more limited (compared with those in developed markets) either because they are less informed or because there are more conflicts of interests between managers and them. As a result, the influence of changes in dividends on market value should again be larger in emerging markets because changes in dividends are less expected by investors there.

Summarizing the above points, the hypothesis is proposed as follows:

H0: Changes in dividends do not have a different influence on market value in emerging markets

(15)

H1: Changes in dividends have larger influence on market value in emerging markets than in

developed markets.

4. Methodology

4.1. Estimation

I use multivariate panel regression to investigate the influence of changes in dividends on market value. In the models, I design market value in the form of returns and allow multiple variables into the models for explanation. Most variables are designed similar to Faulkender and Wang (2006), who use 1-year lagged market value (market value of previous year) as the standardization for most variables in their analysis of the impact of changes in cash holdings. The purpose for doing so is to interpret the estimated coefficients as the change in market value for a 1-unit change in dividend, as well as for a 1-unit change in other corresponding independent variables (Faulkender & Wang, 2006). Paying out dividends is a way for firms to distribute cash and causes changes in cash holdings; therefore, it makes good sense to use the same strategy to examine the influence of changes in dividends on market value.

(16)

agency approach to explain the results (e.g. Bajaj & Vijh, 1990; Alangar et al., 1999; Lonie et al., 1996; Gurgul et al., 2003; Capstaff et al., 2004). These results implicitly show that 1-unit increase in dividend should cause a decrease in market value which is smaller than 1 (increase can be viewed as negative decrease), including the effects of asymmetric information and agency problems. Again, the opposite applies to decreases in dividends. As a result, I expect that the coefficient measuring the influence of changes in dividends on market value will be above -1 in both developed and emerging markets.

In line with the hypothesis proposed, I also expect that the influence of changes in dividends on market value will be larger in emerging markets than in developed markets because of higher levels of asymmetric information and agency problems in emerging markets. To be able to find the difference, I will use a country dummy (to be discussed in Section 4.2) to distinguish between developed markets and emerging markets. Alternatively, I will also compare the regression results from the segmented samples (which consist of firms from developed markets only and firms from emerging markets only). Moreover, I will estimate the influence of changes in dividends on market value both excluding tax effects and including tax effects. Since investors prefer dividends more when dividends are taxed less and less when dividends are taxed more (Black, 1976), 1-unit change in dividend should have a different influence on market value in places where the tax laws are different. Therefore, I expect that the difference of the influence of changes in dividends on market value between a developed market and an emerging market will get even larger when tax effects are included than when they are excluded, if tax laws in the emerging market favor dividends more. This expectation goes the other way around (the difference will get smaller) if tax laws in the developed market favor dividends more.

(17)

dropped out but dropping such a well-defined variable in finance is inappropriate.

Summarizing the above points, I provide all the listed steps here. First of all, I will run panel regressions with cross-section fixed effects for the segmented samples (which consist of firms from developed markets only and firms from emerging markets only) and then for the combined sample (with the country dummy), both when tax effects are excluded and included. After doing so, I will repeat the first step but with cross-section random effects and time-period random effects (with random effects, the model is estimable even if overall market return stays in the model), as a robustness test. Finally, I will employ another robustness test, where a year dummy (to be discussed in Section 4.2) is constructed to distinguish between market declination and advance, to see if the finding holds in both situations, in light of Fuller and Goldstein (2011). The remaining part of Section 4 will describe all the variables used in the regressions (including robustness tests) and the models used in each step.

4.2. Variables

To examine the influence of changes in dividends on market value, I use changes in market value of firm i in year t (ΔMVi,t) against changes in dividends of firm i in year t (ΔDivi,t); they are then

both standardized by 1-year lagged market value (MVi,t-1). This approach is similar to Faulkender

and Wang (2006)‟s when investigating the relationship between changes in cash holdings and market value. The two variables are thus expressed as follows, respectively: ΔMVi,t/ MVi,t-1 and

ΔDivi,t/ MVi,t-1. In this way, the estimated coefficients can be interpreted as the change in market

value caused by a 1-unit change in dividend. Also to be noticed is that stock return is the difference between MVi,t and MVi,t-1, divided by MVi,t-1 (Faulkender & Wang, 2006), the dependent variable

(ΔMVi,t/ MVi,t-1) is thus a form of return. This designation is similar with those in studies

(18)

My purpose in this paper is to test whether changes in dividends have larger influence on market value in emerging markets than in developed markets; therefore, a dummy variable is needed. The dummy variable is expressed as CDUMi, which takes 1 if firm i is listed in an emerging market

(China) and 0 if otherwise (Germany). On the other hand, I argue for asymmetric information and agency problems as the explanation for the different influence of changes in dividends on market value. That is, I only intend to capture the difference caused by the difference in asymmetric information and agency problems. Emerging markets also differ from developed ones in that they grow faster and firms there face more investment opportunities, which may influence how market value is influenced by changes in dividends since it makes more sense to retain the money for appreciation when there are plenty investment opportunities. Therefore, to account for this, a control variable for investment opportunities is constructed and is designed in the same logic with dependent variable (ΔMVi,t/ MVi,t-1), where changes in total assets for firm i in year t (ΔTAi,t) is

used to measure investment opportunities and is then divided by 1-year lagged market value of firm i (MVi,t-1). It is then expressed as ΔTAi,t/ MVi,t-1.

Size and overall market return are also controlled in this paper since it is well acknowledged that they will influence market value. Therefore, it is necessary to isolate the influence of size and overall market return on market value from that of changes in dividends. Besides, in the dividend literature, Fama and French (2001) find that dividend payout ratios are higher for larger firms while researchers studying announcement-day-effect of dividends report higher price reactions for small firms (Bajaj & Vijh, 1990; Mitra & Owers, 1995; Jin, 2000), which gives more reasons to control for size. The control variable for size is designed as the natural log of firm i‟s sales in year t, expressed as ln(SALESi,t). In this paper, overall market return is proxied by the return on the stock

market index in year t (the market in which firmi is listed), expressed as RMKTi,t.

(19)

traditional leverage ratio. It is thus expressed asΔDEBTi,t/ MVi,t-1, where changes in firm i‟s total

debt in year t (ΔDEBTi,t) is divided by 1-year lagged market value of firm i (MVi,t-1).

As is mentioned already, I will account for tax effects when estimating the influence of changes in dividends on market value as non-neutral taxation may also influence how changes in dividends influence market value (Black, 1976; Bernheim, 1991). I take tax effects into account in the same way as La Porta et al. (2000a), where the effects are measured by the tax advantage of dividends (AdvTAXi,t). The specification of the calculation of the tax advantage of dividend is available in

Appendix 1.

Finally, I construct a year dummy to distinguish between market declination and advance as a robustness test. Fuller and Goldstein (2011) find out that dividend payers outperform (in terms of market return) non-dividend payers by more when the market is declining. Following similar logic, it could be possible that the influence of changes in dividends on market value will also vary over different market situations. It is thus important to know if the hypothesis holds in both market situations (advancing and declining) or not. If the hypothesis still holds, it would then be interesting to see whether the difference in changes in dividends on market value between developed and emerging markets gets larger when the market is declining. To do so, a dummy variable YDUMi,t is constructed, which equals to 1 if the market is declining (RMKTi,t<0) and 0 if

the market if advancing (RMKTi,t>0). A summary of all the variables can be found in Table 2.

Table 2: Summary of all variables

Variable Meaning Description

ΔMVi,t/MVi,t-1 Changes in market value divided by 1-year lagged market value Dependent variable ΔDivi,t/MVi,t-1 Changes in dividend divided by 1-year lagged market value Measure for dividend ΔTAi,t/ MVi,t-1 Changes in total asset divided by 1-year lagged market value Control for investment opportunities

ΔDEBTi,t/ MVi,t-1 Changes in total asset divided by 1-year lagged market value Control for debt

RMKTi,t Return on stock market index Control for market return

ln(SALESi,t) Natural log of sales Control for size

CDUMi Country dummy, equals to 1 if a firm is in emerging market Interaction term

YDUMi,t Year dummy, equals to 1 if the market is declining ( RMKTi,t<0) Interaction term for robust test

(20)

4.3. Regression models

When tax effects are excluded, I use the following model to estimate the influence of changes in dividends on market value for the segmented sample:

ΔMVi,t/MVi,t-1 = α+ β1 * (ΔDivi,t/MVi,t-1) + β2*ΔTAi,t/ MVi,t-1

+ β3*ΔDEBTi,t/ MVi,t-1 + β4 * RMKTi,t + β5 * ln(SALESi,t) + εit, (1)

where all variables are designed as specified in Section 4.2 (also summarized as in Table2). CDUMi is not necessarybecause the sample is already segmented. As is discussed, I expect that β1

will be above -1, either for developed or emerging markets.

For the combined sample, I use another equation:

ΔMVi,t/MVi,t-1 = α+ β1 * (ΔDivi,t/MVi,t-1) + β2* CDUMi * (ΔDivi,t/MVi,t-1)

+ β3*ΔTAi,t/ MVi,t-1 + β4*ΔDEBTi,t/ MVi,t-1 + β5 * RMKTi,t

+ β6 * ln(SALESi,t) + εit, (2)

where all variables are designed the same as equation (1). It is important to notice that CDUMi

equals to 1 if the firm is listed in emerging markets and 0 if otherwise, the influence of 1-unit change in dividend on firm‟s market value is thus estimated by (β1 + β2) in emerging markets and

β1 in developed markets (excluding tax effects). It is expected by this paper that β2 will be

significant and larger than 0 as a result of the differences in asymmetric information and agency problems between emerging markets and developed markets. The other thing to be noticed is that only a slope interaction is assumed for CDUMi because adding country intercept effects in will

make the model inestimable with cross-section fixed effects.

To account for tax effects, I multiply ΔDivi,t/MVi,t-1 by AdvTAXi,t. Therefore, the model for

estimating the influence of changes in dividends on market value for the segmented sample is restructured as follows:

ΔMVi,t/MVi,t-1 = α+ β1 * AdvTaxi,t * (ΔDivi,t/MVi,t-1) + β2*ΔTAi,t/ MVi,t-1

where all variables are designed the same as equation (1) except that ΔDivi,t/MVi,t-1 is multiplied

(21)

estimated in equation (1).

The model used to estimate the combined sample when tax effects are included is then as follows: ΔMVi,t/MVi,t-1 = α+ β1* AdvTAXi,t * (ΔDivi,t/MVi,t-1)

+ β2* CDUMi * AdvTAXi,t * (ΔDivi,t/MVi,t-1) + β3*ΔTAi,t/ MVi,t-1

where all variables are the same as equation (2) except that ΔDivi,t/MVi,t-1 is multiplied by

AdvTAXi,t. Similar to equation (2), the influence of 1-unit change in dividend on market value is

thus estimated by (β1 + β2) and β1 respectively for emerging markets and developed markets

(including tax effects). I expect that β2 estimated here should be different from that estimated in

equation (2) (either larger or smaller, depending on how the tax laws differ).

Equation (1), (2), (3) and (4) will be estimated using cross-section fixed effects. Then as a robustness test, I will re-estimate these four equations with cross-section random effects and time-period random effects to see whether the estimated coefficients differ a lot or not. With random effects, the intercept effects of CDUMi can be added in. I therefore restructure equation (2)

and (4) as (5) and (6) for estimation of the combined sample when tax effects are excluded and included respectively:

ΔMVi,t/MVi,t-1 = α+ β1 * (ΔDivi,t/MVi,t-1) +β2* CDUMi

+ β3* CDUMi * (ΔDivi,t/MVi,t-1) + β4*ΔTAi,t/ MVi,t-1

ΔMVi,t/MVi,t-1 = α+ β1* AdvTAXi,t * (ΔDivi,t/MVi,t-1) + β2* CDUMi

+ β3* CDUMi * AdvTAXi,t * (ΔDivi,t/MVi,t-1) + β4*ΔTAi,t/ MVi,t-1

where all variables are designed the same as equation (2) and (4) except that the intercept effects of CDUMi are added. Therefore, with random effects, the influence of 1-unit change in dividend

on market value is thus estimated by (β1 + β3) and β1 respectively for emerging markets and

(22)

For the second robustness test, I add YDUMi,t into the equations (3) and (4) and use cross-section

fixed effects. For the segmented sample, I use the following model: ΔMVi,t/MVi,t-1 = α+ β1* AdvTAXi,t * (ΔDivi,t/MVi,t-1)

+ β2* YDUMi,t * AdvTAXi,t * (ΔDivi,t/MVi,t-1) + β3*ΔTAi,t/ MVi,t-1

where all the variables are designed the same as equation (3) except the year dummy is added. The intercept effects of YDUMi,t is not added here because the model would become inestimable with

cross-section fixed effects.

For the combined sample, I use the following model: ΔMVi,t/MVi,t-1 = α+ β1* AdvTAXi,t * (ΔDivi,t/MVi,t-1)

+ β2* CDUMi * AdvTAXi,t * (ΔDivi,t/MVi,t-1)

+ β3* YDUMi,t * AdvTAXi,t * (ΔDivi,t/MVi,t-1)

+ β4* YDUMi,t * CDUMi * AdvTAXi,t * (ΔDivi,t/MVi,t-1) + β5*ΔTAi,t/ MVi,t-1

where all the variables are designed the same as equation (4) except that the year dummy is added. It is important to notice that the intercept effects of CDUMi and YDUMi,t are not included because

the model would become inestimable with cross-section fixed effects. Another thing to be noticed is that YDUMi,t is also set to interact with CDUMi * AdvTAXi,t * (ΔDivi,t/MVi,t-1).

5. Sample & Data

5.1. Germany & China

(23)

firms in Germany and China, the requirement for a minimum sample size could easily be satisfied. On the other hand, both Germany and China satisfy, at least, the weak form of market efficiency (Higgs & Worthington, 2003; Lima & Tabak, 2004), which justifies the use of market value in the models and the aim of this paper to focus on the long-run economic meaning of changes in dividends. Furthermore, both Germany and China commit to International Financial Reporting Standards (IFRS), so accounting information on firms in the two countries should be comparable. This is important as many variables in this paper‟s models will require the use of accounting information (e.g. total assets, debt and sales).

From the perspectives of asymmetric information and agency problems, Germany and China have large differences in their institutional set-up. German capital market has undergone a long history of development with a well-established market system.4 China did not start its transition from a planned economy into a market economy until late 1970s and the first effective change took place as late in 1991 when the first stock exchange (Shanghai Stock Exchange) was established5; the liberalization of capital markets was also lagged till World Trade Center (WTO) accepted China as its member in 2001. Compared with Germany, China is still catching up in many perspectives. As noted in the research conducted by S&P (2008), the level of transparency and disclosure is low in China, compared with many other countries. The T&D score computed for China in the research is 46% while that computed for Germany in 2003 is 68%. Additionally, compared to Germany as a traditional civil law country, China‟s current legal system is only slightly more than a half-century old; the enforcement of law is also a problem in China as “Rule of Law” has not yet fully replaced of “Rule of Person”; in particular, investor protection is still weak although amendments were made in 2006 to Securities Law which emphasized the interests of investors (S&P, 2009). At last, although Germany is characterized by concentrated ownership (Clark & Wójcik, 2005), China is characterized by concentrated ownership mixed with state ownership, where the state holds large amount of non-tradable shares and institutional investors hold the majority of floating shares; in this case, managers act more in line with the government and even large institutional investors

4_{The history of Frankfurt Stock Exchange traces back to as early as the 16}th_{century and was established as the leading stock} exchange in Germany after World War II.

5

(24)

could suffer from it6 (S&P, 2009).

During the time period covered in this paper, both Germany and China witnessed booming in capital markets from 2006 to 2007 and experienced difficulties after the crisis starting late 2007. Several reforms also took place during 2000 to 2010 for the two countries, which are rather interesting to this paper‟s topic. In Germany, 2 major tax reforms took place in 2000 and 2007 respectively and the latter one made taxation on dividends and capital gains neutral (effective from 2009, and both of them are now taxed at a flat rate of 25%). In China, an amendment was made to the tax law in 2005 which cut tax on dividends to half and a major tax reform took place in 2007 which united the corporate tax rate for domestic and foreign firms; however, to date, there is still no tax law covering the buying and selling of shares in the secondary market, so capital gains made from it are currently taxed at 0%. During the time period, China has also been trying hard to close in the gap between itself and developed countries; amendments to Company Law and Securities Law were made in 2006 to strengthen the legal system and to better protect investors and new regulations were released in 2007 by China Securities Regulatory Committee (CSRC) to enhance information disclosure. However, as contended by S&P (2009), the gap is still huge between China and developed countries despite the tremendous efforts it has made. Therefore, Germany and China, during the time period 2000 to 2010, give me a good background to conduct research as they are dramatically different with respect to asymmetric information and agency problems but they are not so different that the results may be incomparable.

5.2. Selection criteria

This paper studies Germany and China during the period 2000 to 2010. Therefore, first of all, firms chosen should either be listed in Germany or China and should be active during 2000 to 2010. Second, only domestic firms listed in Germany and China are chosen since foreign firms are usually subject to different laws and regulations. Third, the stock exchanges chosen are Frankfurt

6

(25)

Stock Exchange in Germany and Shanghai Stock Exchange, Shenzhen Stock Exchange and Hong Kong Stock Exchange in China. Only Frankfurt Stock Exchange is chosen for Germany as it accounts for the majority of turnover in Germany. Both Shanghai Stock Exchange and Shenzhen Stock Exchange are chosen for China as Chinese firms can only be listed in either one of them. Hong Kong Stock Exchange is also chosen for China because many large Chinese firms choose only to list themselves there. The four exchanges are chosen so that the sample covers as many firms as possible while keeping out unnecessary repeats and complication. Fourth, financial and utility firms are excluded from the sample (using FTSE industry classification).7 After applying the four selection criteria, the sample consists of 915 German firms and 2634 Chinese firms, with 39039 firm-year observations in total. A full description on industries included in the sample and the number of firms in each industry is shown in Table 3.

7

Financial and utility firms include Bank, Electricity, Equity investment instruments, Equity warrants, Financial services, Gas, water and multiutilities, Life insurance, Non-equity investment instruments, Non-life insurance and Oil and gas.

Table 3: Industries selected and the number of firms after applying the 4 selection criteria (Germany and China, 2000 – 2010)

Firms presented here satisfy 4 selection criteria: 1. listed German or Chinese firms, which are active during 2000 to 2010; 2. domestic firms; 3. the firms should be listed in Frankfurt, Shanghai, Shenzhen or Hong Kong Stock Exchange; 4. financial and utility firms should not be included. As a result of the 4 selection criteria, 915 German firms and 2634 Chinese firms are selected, amounting to 3549 firms in total.

Industries Number of firms Industries Number of firms

GE CH Total GE CH Total

Automobile and parts 24 113 137 Industrial engineering 60 280 340

Aerospace and defense 2 8 10 Industrial metals and mining 8 118 126

Alternative Energy 40 24 64 Industrial transportation 17 107 124

Beverages 12 39 51 Leisure goods 13 42 55

Chemicals 35 244 279 Media 67 35 102

Construction and Materials 33 165 198 Mining 15 59 74

Electronic and electrical equipment 39 274 313 Mobile communication 9 5 14

Fixed line telecommunications 2 3 5 Personal goods 28 116 144

Food and drug retailer 8 19 27 Pharmaceuticals and biotechnology 31 183 214

Food producer 17 106 123 Software and computer service 112 108 220

Forestry and paper 4 25 29 Support services 40 51 91

General industry 7 33 40 Technology hardware and equipment 34 123 157

General retailer 30 89 119 Tobacco 0 0 0

Healthcare equipment and service 29 25 54 Travel and leisure 24 69 93

(26)

Finally, firm-year observation with missing values during 1999 to 2010 is excluded. 1 extra year is required here because the regression models use delta values and the standardization of 1-year lagged market value. As a result, only firms with at least 1 firm-year observation and a 1-year lag during 2000 to 2010 remain in the sample. This criterion reduces the sample to 552 German firms and 1017 Chinese firms, with 8713 firm-year observations having no missing values (unbalanced panel). A full description on the final sample is available in Table 4. As can be seen in Table 4, more Chinese firms are included in the final sample but German firms, on average, provide more years of observations. Therefore, the final sample is quite evenly distributed between German firms and Chinese firms, in terms of the number of firm-year observations.

Table 4: Final sample (Germany and China, 2000 – 2010)

Firm-year observations with missing values during 1999 to 2010 are excluded from the sample. A 1-year lag is required because of the designation of the models. As a result, only firms with at least 1 firm-year observation and a 1-year lag remain. Panel A shows that there are 552 German firms and 1017 Chinese firms with at least 1 firm-year observation and a 1-year lag and there are in total 8713 firm-year observations with no missing values. These firms (firm-year observations) are used for the regressions. Panel B shows the break-up of firms (firm-year observations) employed in the regressions by year.

Panel A: Number of firms and firm-year observations used for the regressions

Number of firms Number of firm-year observations

Germany 552 4384

China 1017 4329

Panel B: Break-up of firms (firm-year observations) employed in the regressions by year

Year Number of firms (firm-year observations) employed

Germany China Total

(27)

5.3. Data

Data on market value (MVi,t), total cash dividends paid (Divi,t), total assets (TAi,t), total debt

(DEBTi,t) and sales (SALESi,t) are collected from Datastream for all German and Chinese firms

throughout the time period 1999 to 2000 (as mentioned, 1 extra year required). Then, the retrieved data are used to calculate ΔMVi,t/MVi,t-1, ΔDivi,t/MVi,t-1, ΔTAi,t/ MVi,t-1, ΔDEBTi,t/ MVi,t-1 and

ln(SALESi,t). The description of these 5 variables is available in Table 5. The dummy variable

(CDUMi) is constructed as described in methodology, which equals to 1 if the firm is listed in

China (as an emerging market) and 0 if the firm is listed in Germany (as a developed market). In addition, market index is also collected from Datastream in order to calculate overall market return (RMKTi,t). DAX 200 and Shangzheng are used as proxies for Germany and China respectively.

Description on the 2 indexes and calculated overall market returns can be found in Table 6. The

Table 5: Descriptive statistics (Germany and China, 2000 – 2010)

This table presents descriptive statistics on dependent and independent variables (except for RMKTi,t and dummies). It should be noticed here that ΔDivi,t/MVi,t-1 and AdvTaxi,t * ΔDivi,t/MVi,t-1 will be used separately (in different models).

Germany China Total

Mean SD Mean SD Mean SD

ΔMVi,t/MVi,t-1 0.1993 1.1277 0.2880 0.9605 0.2434 1.0488

ΔDivi,t/MVi,t-1 -0.0017 0.0034 0.0044 0.0013 0.0014 0.0026

ΔTAi,t/ MVi,t-1 0.0113 0.0249 0.6013 0.0556 0.3044 0.0431

ΔDEBTi,t/ MVi,t-1 -0.0260 0.0174 0.1582 0.0312 0.0656 0.0253

ln(SALESi,t) 11.8416 2.3772 13.9244 1.7103 12.8764 2.3197

AdvTaxi,t 0.7916 0.1062 0.8376 0.0484 0.8145 0.0858

AdvTaxi,t * ΔDivi,t/MVi,t-1 -0.0018 0.0025 0.0036 0.0011 0.0009 0.0020

Table 6: Market index and overall market return (Germany and China, 2000 – 2010)

This table presents the market index and overall market returns in Germany and China during 2000 to 2010. DAX 200 is used to proxy overall market movements in Germany and Shangzheng is used to proxy overall movements in China.

(28)

other dummy variable (YDUMi,t) is constructed based on the data on overall market return

(RMKTi,t), which equals to 1 if the market is declining (RMKTi,t<0) and 0 if it is advancing

(RMKTi,t>0). As for data on tax, I use combined sources from Jones Day (2002), Von Bröckel

(2008) and Ernst & Young (2012).8 The advantage of tax (AdvTaxi,t) is calculated using retrieved

data on tax. It is then multiplied by ΔDivi,t/MVi,t-1 to derive AdvTaxi,t * ΔDivi,t/MVi,t-1 so that tax

effects can be taken into account. The description of AdvTaxi,t and AdvTaxi,t * ΔDivi,t/MVi,t-1 is

available in Table 5.

Table 5 shows that the mean of ΔTAi,t/ MVi,t-1 is much lower in Germany (0.0113) than in China

(0.6013), which indicates that German firms face less investment opportunities. This is consistent with the reason to control for investment opportunities in the models (emerging market face more investment opportunities than developed markets). Table 5 also shows that taxation in Germany and China is, in general, biased in favor of capital gains (dividends are taxed more than capital gains). Moreover, AdvTaxi,t is larger in China (0.8376) than in Germany (0.7916), which means

that taxation in China favors dividends more than taxation in Germany or that taxation in China is less biased in favor of capital gains. Table 6 captures another perspective of the difference between

8

Data on tax and the calculation of the tax advantage of dividends is available in Appendix 2.

Table 7: Number and percentage of dividend –paying firms employed in the regressions (Germany and China, 2000 – 2010)

This table describes the number and percentage of dividend-paying firms employed in the regressions by year.

Year Number of dividend-paying firms

employed in the regressions

Number of firms employed in the regressions

Dividend-paying firms as a percentage of firms employed in the regressions

Germany China Total Germany China Total Germany China Total

(29)

Germany and China, where it is shown that the Chinese market (overall market return ranging from -3.60% to 223.50%) is much more volatile than the German market.

Through further studying the collected data, it is learnt that, in terms of the number of firm-year observations, comparable number of dividend-paying firms and non-dividend paying firms are included (Table 7). Furthermore, firms that increase (ΔDivi,t>0), maintain (ΔDivi,t=0) and decrease

(ΔDivi,t<0) dividends account for considerable number of firm-year observations respectively

(Table 8). Therefore, the data set is of good quality in that it is not heavily biased against any kind of behavior of dividends (i.e. pay vs. not pay; increase vs. maintain vs. decrease).

6. Results

6.1. Panel regression results

Results of panel regressions with cross-section fixed effects for the segmented sample of German firms and Chinese firms and the combined sample are reported in Table 9. Panel A in Table 9 reports the results when tax effects are excluded, where equation (1) and (2) are used respectively

Table 8: Number and percentage of firms employed in the regressions which increase, maintain and decrease dividends (Germany and China, 2000 – 2010)

This table describes the number and percentage (in parentheses) of firms employed in the regressions which increase (ΔDivi,t>0), maintain (ΔDivi,t=0) and decrease (ΔDivi,t<0) dividends.

Year Increase Maintain Decrease

Germany China Total Germany China Total Germany China Total

(30)

to produce results for the segmented and the combined sample. Panel B in Table 9 reports the results when tax effects are included, where equation (3) and (4) are used respectively to produce results for the segmented and the combined sample.

Table 9: Results of panel regressions with cross-section fixed effects for the segmented sample and the combined sample (excluding and including tax effects) (Germany and China, 2000 – 2010)

The segmented samples consist of 552 German firms with 4384 observations and 1017 Chinese firms with 4329 observations. The combined sample consists of 1569 German and Chinese firms with 8713 observations. Equation (1) and (2) are used to produce results for the segmented sample and the combined sample respectively when tax effects are excluded. When tax effects are included, equation (3) and (4) are used. All the equations are estimated using cross-section fixed effects with ΔMVi,t/MVi,t-1 as dependent variable.

Panel A: Results of panel regressions with cross-section fixed effects for the segmented sample and the combined sample

(excluding tax effects)

Germany China Combined sample

Coefficient p-value Coefficient p-value Coefficient p-value

CDUMi*(ΔDivi,t/ MVi,t-1) 0.5363 0.0000

ΔDivi,t/MVi,t-1 -0.0133 0.7786 0.4253 0.0000 -0.0132 0.7653

ΔTAi,t/ MVi,t-1 0.0680 0.0000 0.0262 0.0000 0.0379 0.0000

ΔDEBTi,t/ MVi,t-1 -0.1680 0.0000 -0.0092 0.1278 -0.0607 0.0000

RMKTi,t 0.4204 0.0000 0.8686 0.0000 0.8011 0.0000 ln(SALESi,t) -0.1189 0.0001 -0.0256 0.2560 -0.0778 0.0000 C 1.5854 0.0000 0.4883 0.1118 1.1537 0.0000 Adjusted R-squared 0.1018 0.3232 0.1555 F-statistic 1.8930 3.0246 2.0194 p-value (F-statistic) 0.0000 0.0000 0.0000

Panel B: Results of panel regressions with cross-section fixed effects for the segmented sample and the combined sample

(including tax effects)

CDUMi*AdvTAXi,t*(ΔDivi,t/MVi,t-1) 0.6397 0.0000

AdvTAXi,t*(ΔDivi,t/MVi,t-1) -0.0300 0.6390 0.4904 0.0000 -0.0312 0.5874

ΔTAi,t/ MVi,t-1 0.0680 0.0000 0.0262 0.0000 0.0379 0.0000

ΔDEBTi,t/ MVi,t-1 -0.1680 0.0000 -0.0090 0.1370 -0.0605 0.0000

(31)

As shown by Panel A in Table 9, the estimated coefficient of ΔDivi,t/MVi,t-1 for the sample of

German firms stands at -0.0133 (p-value = 0.7786) while that for the sample of Chinese firms is 0.4253 (p-value = 0.0000). The coefficient for Chinese firms is significant from 0 and is larger than that estimated for German firms, which is not significantly different from 0. Consistent results are reported for the combine sample, where a dummy is used to identify between German and Chinese firms. The estimated coefficient of ΔDivi,t/MVi,t-1 is -0.0132 (p-value = 0.7653) and that of

CDUMi*(ΔDivi,t/ MVi,t-1) is 0.5363 (p-value = 0.0000). This indicates that the coefficient of

ΔDivi,t/MVi,t-1 for China is significantly larger by 0.5363 than that for Germany. Either by

comparing the segmented samples or by interpreting the combined sample, the conclusion is the same. In the absence of tax, 1-unit change in dividend corresponds to larger change in market value in China than in Germany; this finding is statistically significant.

When tax effects are included, similar results are reported. In Panel B in Table9, the estimated coefficient ofAdvTAXi,t*(ΔDivi,t/MVi,t-1) for Germany is -0.0300 (p-value = 0.6390) compared

with 0.4904 (p-value = 0.0000) for China. Again, the estimated coefficient for China is larger than the coefficient for Germany. As for the combined sample, the estimated coefficient of AdvTAXi,t*(ΔDivi,t/MVi,t-1) is -0.0312 (p-value = 0.5874) and that of CDUMi* AdvTAXi,t*

(ΔDivi,t/ MVi,t-1) is 0.6397 (p-value = 0.0000). Similarly, it is indicated here that the coefficient of

AdvTAXi,t*(ΔDivi,t/MVi,t-1) for China is significantly larger by 0.6397 than the coefficient for

Germany. This means that, in the presence of tax, 1-unit change in dividend has larger influence on market value in China than in Germany; this finding is also statistically significant.

(32)

between Germany and China is not only statistically significant, but also economically significant (roughly 0.5-0.6). It is argued in this paper that the difference is attributable to the difference with respect to asymmetric information and agency problems between Germany (as a developed market) and China (as an emerging market). Since dividends can be viewed as the tool to mitigate asymmetric information (Bhattacharya, 1979; John & Williams, 1985; Miller & Rock, 1985) and agency problems (Easterbrook, 1984; Jensen, 1986), dividends should matter more in places where there are more information asymmetry and agency problems, holding other factors constant.

Furthermore, in all cases, the estimated coefficients for Germany stand significantly at 0 while those for China are all significant from 0. Nonetheless, all the estimated coefficients are significant from -1, whether for Germany or for China. This means that 1-unit increase in dividend will not cause 1-unit decrease in market value in both Germany and China, instead the corresponding decrease is significantly smaller than 1 (increase can be viewed as negative decrease). That is, 1-unit increase will not decrease market value in Germany and it will even increase market value in China. This finding is consistent with my expectation that the influence of 1-unit increase in dividend corresponds to less than 1-unit decrease in market value because of asymmetric information and agency problems. In this way, the explanation supplements the argument proposed above and is also consistent with the signaling theory (Bhattacharya, 1979; John & Williams, 1985; Miller & Rock, 1985) and the agency approach (Easterbrook, 1984; Jensen, 1986).

It can also be noticed that, when tax is included, the estimated coefficient of CDUMi* AdvTAXi*

(33)

6.2. Robustness tests

As the first robustness test, I re-do all the panel regressions with cross-section random effects and

Table 10: Results of panel regressions with cross-section random effects for the segmented sample and the combined sample (excluding and including tax effects) (Germany and China, 2000 – 2010)

The segmented samples consist of 552 German firms with 4384 observations and 1017 Chinese firms with 4329 observations. The combined sample consists of 1569 German and Chinese firms with 8713 observations. Equation (1) and (5) are used to produce results for the segmented sample and the combined sample respectively when tax effects are excluded. When tax effects are included, equation (3) and (6) are used. It is important to know that the intercept effects of CDUMi are added in equation (5) and (6). All the equations are estimated using cross-section random effects with ΔMVi,t/MVi,t-1 as dependent variable.

Panel A: Results of panel regressions with cross-section random effects for the segmented and the combined sample (excluding tax

effects)

CDUMi -0.0432 0.0632

ΔDivi,t/MVi,t-1 -0.0224 0.6354 0.3917 0.0000 -0.0220 0.6042

ΔTAi,t/ MVi,t-1 0.0704 0.0000 0.0239 0.0000 0.0381 0.0000

ΔDEBTi,t/ MVi,t-1 -0.1587 0.0000 -0.0151 0.0057 -0.0556 0.0000

RMKTi,t 0.4300 0.0000 0.9123 0.0000 0.8431 0.0000 ln(SALESi,t) -0.0081 0.3203 0.0327 0.0000 0.0068 0.1726 C 0.2738 0.0052 -0.3274 0.0010 0.0843 0.1665 Adjusted R-squared 0.0468 0.3692 0.1621 F-statistic 44.0853 507.6451 241.7740 p-value (F-statistic) 0.0000 0.0000 0.0000

Panel B: Results of panel regressions with cross-section random effects for the segmented and the combined sample (including tax

effects)

CDUMi -0.0431 0.0693

ΔTAi,t/ MVi,t-1 0.0704 0.0000 0.0239 0.0000 0.0382 0.0000

ΔDEBTi,t/ MVi,t-1 -0.1588 0.0000 -0.0150 0.0061 -0.0555 0.0000

(34)

Table 11: Results of panel regressions with time-period random effects for the segmented sample and the combined sample (excluding and including tax effects) (Germany and China, 2000 – 2010)

The segmented samples consist of 552 German firms with 4384 observations and 1017 Chinese firms with 4329 observations. The combined sample consists of 1569 German and Chinese firms with 8713 observations. Equation (1) and (5) are used to produce results for the segmented sample and the combined sample respectively when tax effects are excluded. When tax effects are included, equation (3) and (6) are used. It is important to know that the intercept effects of CDUMi are added in equation (5) and (6). All the equations are estimated using time-period random effects with ΔMVi,t/MVi,t-1 as dependent variable.

Panel A: Results of panel regressions with time-period random effects for the segmented sample and the combined sample

(excluding tax effects)

CDUMi -0.0182 0.4397

ΔDivi,t/MVi,t-1 -0.0179 0.7006 0.4223 0.0000 -0.0133 0.7450

ΔTAi,t/ MVi,t-1 0.0701 0.0000 0.0209 0.0000 0.0359 0.0000

ΔDEBTi,t/ MVi,t-1 -0.1500 0.0000 -0.0137 0.0056 -0.0532 0.0000

Panel B: Results of panel regressions with time-period random effects for the segmented sample and the combined sample

(including tax effects)

CDUMi -0.0182 0.4393

ΔTAi,t/ MVi,t-1 0.0701 0.0000 0.0209 0.0000 0.0359 0.0000

ΔDEBTi,t/ MVi,t-1 -0.1500 0.0000 -0.0136 0.0059 -0.0531 0.0000