Dividend signalling and investor protection: An international comparison

This study examines whether dividend changes signal future earnings growth in non-US markets following the Ham et al. (2020) methodology and whether the strength of the earnings signal varies with the level of investor protection. Based on the notion that weak investor protection reduces the cost of cutting dividends and as such increases managers ’ discretion to change div-idends, we expect that the strength of the earnings signal in dividend increases becomes weaker as investor protection decreases. In a sample drawn from 38 different markets, our results indicate while dividends can signal future earnings in non-US markets, the strength of the signal is weaker than that in the US. In line with our predictions, we find that for firms in a strong investor protection environment, dividend changes are correlated more strongly with subsequent earning changes than is the case for firms in weak investor protection environments.


Introduction
In the seminal study, Lintner (1956) presents interview evidence that US managers only increase dividend payouts when they believe growth in future earnings can sustain the dividend increase.However, subsequent empirical studies find little or no evidence to support this dividend signaling theory, which predicts that dividend increases signal subsequent earnings increases (Watts, 1973;Grullon et al., 2005;Hasan, 2021;Benartzi et al., 1997;Fukuda, 2000;Ali et al., 2017).Ham et al. (2020), after refining the methodology compared to prior studies, reexamines the signaling role of dividends in the US market.Their findings suggest that dividends can signal future earnings changes.The signaling role hinges on the premise that dividends represent a long-term commitment.Specifically, if investors assume that the company will keep paying at least the same level of dividends in the future, firms only increase dividend payments if they are sure there are sufficient future earnings to maintain that level of payments.
Given that dividends in the markets outside the US contain very limited commitment (Pathak and Gupta, 2021), whether the findings from the US market can be generalized to other markets remains an open question.Therefore, this study first extends Ham et al. (2020)'s re-examination into a cross-country setting.Then by taking advantage of the heterogeneity investor protection across markets, we explore whether varying investor protection environments are associated with the potential signal strength of dividends about future earnings.
Prior empirical studies employ a fiscal year approach to investigate whether dividends signal future earnings (Watts, 1973;Grullon et al., 2005;Hasan, 2021;Benartzi et al., 1997;Fukuda, 2000;Ali et al., 2017).The fiscal year approach first identifies the year in which a dividend change occurs and sets that year as period 0, then computes subsequent earnings changes as the difference between annual earnings and the earnings in the prior period.Ham et al. (2020) points out that such an approach may misclassify earnings after dividend changes into period 0, and therefore underestimate the calculation of earnings increases after dividend changes.To circumvent the bias introduced by the fiscal year approach, Ham et al. (2020) proposes an event window approach, which first identifies quarter 0 in which a manager declares a dividend change and truncates quarters surrounding the event quarter to conduct the analysis.Using this approach, Ham et al. (2020) finds a positive relation between dividends and future earnings changes in the US, consistent with the hypothesis that dividends convey insiders' private information about future earnings.
The dividend payment in the US is characterized by its embedded long-term commitment of dividends or dividend stickiness (Guttman et al., 2010).There are two implications arising from dividend stickiness.First, dividend decreases are associated with pronounced declines in stock prices (Ghosh and Woolridge, 1988;Christie, 1994;Brav et al., 2005).Second, because of the large costs associated with dividend decreases, managers who expect earnings to decline or not increase will not increase dividends ex ante. 1 As a result, there will be a separating equilibrium where firms which announce dividend increases primarily comprise those foreseeing future earnings increases.As Lintner (1956) indicates, the long-term commitment embedded in dividends can be a prerequisite condition for dividends to signal earnings.In contrast, non-US firms exhibit considerable flexibility in terms of dividend changes (Pathak and Gupta, 2021).Lack of commitment in dividends paid by non-US firms suggests it is worthwhile re-examining the signaling role of dividends in those countries by employing the event window approach.
Furthermore, we explore whether the extent to which dividends signal earnings depends on country-level investor protection in a cross-country setting.Prior research shows that investor protection is a key institutional factor shaping dividend policies (Mitton, 2004;Bae et al., 2012).In a weak investor protection setting, insiders may indulge in excessive consumption of benefits, including the transfer of assets and investment in negative net present value projects.To enable this rent seeking behavior, managers desire to retain cash by reducing dividend payouts.From the perspective of investors, particularly in environments with weak corporate governance, there's a preference for high, immediate dividends over potential future capital gains.This preference leads managers to distribute free cash flows accumulated within a firm as dividends to mitigate outside shareholders' concerns about insiders' rent seeking behaviour.As a result, the changes in dividend polices are more determined by free cash flows on hand, rather than by the growth in earnings in subsequent periods.Therefore, we expect that changes in dividends in low investor protection environments convey less information about future earnings than when investor protection is strong.To examine the above expectations, we follow the research design used by Ham et al. (2020) and retrieve dividend declarations by non-financial firms from CRSP for the US market and Compustat Global for non-US markets.Our sample consists of 253,238 firm-year observations from 38 markets from 1971 to 2021.Observations from non-US countries are mainly from 2000 to 2021.
We first successfully replicate the Ham et al. (2020) findings in the US context, and expand their results to show that dividend changes are positively associated with earnings changes in each of the years subsequent to dividend changes in non-US markets, albeit the association is less strong than that in the US market.The coefficient magnitude declines with the increasing horizon of earnings to be forecasted.These findings suggest that even though the long-term commitment to dividend increases is not as strong in non-US markets, dividend changes in these markets are still positively associated with future earnings changes.
Furthermore, we examine how investor protection affects the association between dividend and earnings changes.We employ a dichotomous variable measuring the legal system (code vs common law) to capture varying levels of investor protection.When we first partition observations into code and common law subsamples, there is almost no association between dividend and earnings changes in the code law subsample.Instead, dividend changes only have a positive and significant association with future earnings growth for companies operating in a common law setting.We find positive associations between dividend changes and future earnings changes up to three years subsequent to the change in dividends in the common law subsample In contrast, there is no significant predictive relationship between dividend changes and future earnings in the long term for countries with code law traditions.In addition, we test the association between dividend changes and future earnings changes using alternative measures of investor protection developed by La Porta et al. (1998).Across different dimensions of investor protection-Disclosure index, Liability index, Investor Protection index-the results show that the interaction between dividend changes and investor protection generally indicates a significant positive relationship with future earnings changes.All these findings are consistent with our expectation that the signaling strength of dividends regarding future earnings varies with investor protection.The estimated results are robust to the inclusion or exclusion of US-based observations.
In additional analyses, we measure the impact of institutional ownership on dividend signaling across countries and find that dividends provide a stronger signal of earnings in firms with stronger investor protection, even when accounting for variations in institutional ownership.Moreover, we also examine the effect of legal enforcement on our findings and the result indicates that the strength of law enforcement does not notably modify the signalling power of dividends.
This study contributes to the literature in several ways.First, the study is the first to provide international evidence on the impact of investor protection on dividend signaling.We offer a plausible explanation for why dividend changes do not convey future earning information for firms from countries with weak investor protection.Prior international studies that fail to support dividend signaling theory speculate the no-result finding is likely caused by inadequate investor protection in non-US countries (Fairchild et al., 2014;Ali et al., 2017).However, this speculation has not been explicitly examined as most studies only focus on a single country or several countries with similar characteristics (developed countries or emerging markets only).We contribute to this literature by investigating 1 Managers have several incentives to increase dividends.For example, dividend increases can be driven by industry peers (Grennan 2019).
this in a cross-country setting and taking advantage of the varying levels of investor protection in the sample.We provide much needed empirical evidence supporting the common but untested assumption that investor protection is an important moderator for the strength of the association between dividend changes and future earnings changes (Fairchild et al., 2014;Ali et al., 2017).With this, we contribute to the literature investigating the dividend signaling theory and provide compelling evidence explaining the lack of consistent results in earlier studies.
Second, we employ the refined research design in Ham et al. (2020) to examine the dividend signaling hypothesis in a non-US setting.Given that most previous studies report mixed results have used different sample compositions and different research methodologies to examine the dividend signaling hypothesis.This study adds to the literature by showing that dividend signaling theory holds for non-US markets.
The rest of the paper is organized as follows.Section 2 discusses the literature review and develops the hypothesis.Section 3 describes the data, the variables measures, and the methodology.Section 4 provides descriptive statistics and empirical results.Section 5 reports additional tests and robustness checks.Section 6 concludes.

Literature review and hypotheses development
The dividend signaling hypothesis posits that dividend payment conveys a signal for a firm's future earnings.In a seminal study, Lintner (1956) interviews managers of 28 US firms from different industries.He finds that most managers believe shareholders prefer a stable or steady increase in dividend payout ratios.To cater for such a preference, managers tend to smooth dividend streams by partially adjusting dividends until finally the target dividend payout ratio is reached.On the other hand, managers only increase dividends if they believe an increase in earnings is permanent so that the firm can sustain dividend increases.The information content of dividend' hypothesis is further developed by Miller and Modigliani (1961).They document that dividend changes trigger revisions in stock prices in the same direction because investors are likely to interpret a change in dividend payment as a change in managers' view of future earnings for the firm.
Several analytical models have been developed to explain why dividends can reveal information about future earnings.Bhattacharya (1979) proposes that the dividend is a signal indicating a firm's future cash flow in an imperfect-information setting.In particular, since outside investors cannot distinguish the profitability of productive assets held by a cross-section of firms, managers have incentives to pay a high level of dividends if they possess more productive assets.Miller and Rock (1985) show dividends can emerge as a separating equilibrium for managers to signal the good state of firms.Managers in bad-state firms cannot mimic the dividend payment as the cost (forgoing investment opportunities) of doing so is too high.
Despite dividends allowing insiders to signal their firms' future profitability to outside investors being widely recognized in prior studies, empirical studies find little evidence supporting this signaling hypothesis.Watts (1973) finds that dividend changes do convey information about future earnings changes, but the effect is small.Moreover, stock price changes in response to dividend announcements are smaller than transaction costs, which indicates the information contained is trivial.DeAngelo et al. (1996) finds no support for the argument that dividend increases are a useful signal of improved future earnings performance.They suggest the lack of informativeness of dividend changes is because managers are overly optimistic about future profitability and only make limited cash commitments when they increase dividends.These no-result findings may be driven by the small sample size used (310 firms in Watts (1973) and 145 firms in DeAngelo et al. (1996)), which leads to low power of their tests.
After increasing the test power by using a larger sample, Benartzi et al. (1997) documents a positive correlation between dividend changes and earnings changes.Specifically, firms that increase dividend payouts experience significant earnings increases in the prior year and the current year but not in subsequent years.The result that dividend increases piggyback past and current earnings increases is further supported by Fukuda (2000) using Japanese data.Furthermore, Fukuda (2000) finds that the earnings in the subsequent year decline back to the level before the increase, which means that the changes in earnings are temporary instead of permanent.Findings from the Japanese setting suggest that managers' excessive optimism about future earnings or their tendency to use dividends to distribute temporary earnings might jeopardize the signaling value of dividends.
In addition, another stand of studies explains the non-result findings by arguing that the models employed do not control for expected earnings changes appropriately.Nissim and Ziv (2001) reports that previous research failed to capture the true relation between two variables due to omitted correlated variables and measurement error in the independent variable.Specifically, after control for earning expectations Nissim and Ziv (2001) show that current dividend changes are positively associated with unexpected future earnings changes.However, the linear pattern in earning expectation models is challenged by Grullon et al. (2005) and Fama and French (2000), showing that large changes in earnings revert faster than small changes, and negative changes revert faster than positive changes.After considering the nonlinearities in earnings patterns, Grullon et al. (2005) documents dividends do not signal future earnings.Ham et al. (2020) further refined the empirical models employed by previous studies to reexamine the signaling value of dividends in the US.They point out that the 'fiscal year approach' used by previous research attenuates the estimated association between dividends and subsequent earnings changes.Specifically, the 'fiscal year' approach compares earnings in the current year when dividend changes are announced with the earnings announced in subsequent years to calculate earnings growth.Under this approach, all the quarterly earnings announced after dividend changes but before the current year end will be misclassified into past or current years.As a result, the past or current earnings will be overestimated for firms increasing dividends, which accordingly under-estimates the subsequent earnings growth.To address this limitation, Ham et al. (2020) adopts the 'event window' approach.This approach first identifies the event quarter in which a firm changes its dividends and then truncates a certain number of quarters surrounding this event quarter, which ensures that no future earnings will be classified into either past or current earnings.By exploring this 'event window' approach, Ham et al. (2020) presents strong evidence that firms with dividend increases are associated with higher future earnings growth.
There is widespread interest in examining dividend signaling theory in non-US markets.For instance, Choi et al. (2011), Kadioglu and Ocal (2015) and Hasan (2021) investigate whether dividends signal earnings growth in the Korea, Turkey, and UK markets, respectively, but do not find consistent evidence.On the other hand, another stream of studies documents some evidence that dividends, at best, signal earnings in the short term (Ali et al., 2017;Al-Shattarat et al., 2018).Given the mixed findings from the crosscountry studies, which may result from inappropriate research design as highlighted by Ham et al. (2020), it is worthwhile examining the signaling role of dividends in non-US markets by employing the event window approach developed by Ham et al. (2020).
The finding that dividend increases are associated with higher future earnings growth (Ham et al., 2020) may not be generalized to non-US countries.As pointed out by Lintner (1956), a key pre-requisite for dividend changes to signal earnings is that investors prefer stable dividends.Stated differently, investors will penalize a firm if they observe the firm cuts/omits dividends.Consistent with this statement, Ghosh and Woolridge (1988) and Christie (1994) document a 7 % drop in share price in a two-day window surrounding a dividend cut/omission in the US.In a survey study, Brav et al. (2005) show that managers have a strong desire to avoid dividend cuts.The large costs associated with dividend cuts/omissions have two implications.First, managers who expect declines in future earnings will not pool with managers who expect growth in future earnings to increase dividends.Moreover, firms which declare dividend increases primarily consist of those anticipating future earnings growth.Second, most firms tend to keep their dividends unchanged (Ham et al., 2020;Guttman et al., 2010).In contrast, in non-US countries, managers are more flexible in determining dividend policies (Javakhadze et al., 2014), which implies, even without promising earnings growth, managers can still increase dividends for certain short-term benefits and later drop the dividends without incurring substantial costs.Therefore, firms which declare dividend increases in non-US countries may contain varying earning change profiles, and it is unclear to predict that dividend changes can signal earnings growth.Based on the argument, we propose the following hypothesis in the null form: H1: In non-US markets, dividend changes are not associated with future earnings changes.
Varying levels of investor protection can influence the strength of the earnings signal in dividends.There are two streams of literature debating the relation between investor protection and dividend payout policies.Both predict reduced managerial discretion in dividend policy, limiting the ability of managers to signal future earnings through dividend changes.On the one hand, the outcome model suggests that strong shareholder protection can result in a high level of dividend payout as outside investors rely on the good investor protection mechanism to extract dividends (Francis et al., 2011;Jiraporn et al., 2011, Atanassov andMandell, 2018).For instance, La Porta et al. (2000a) study dividend polies in 33 countries and show that firms in common law countries, which typically exhibit stronger investor protection, tend to distribute higher dividends.Similarly, Mitton (2004) finds a positive relation between firm-level corporate governance and dividends payouts in a sample of 365 firms from 19 countries, while Athari (2022) investigates 517 listed firms across 5 Asian countries, finding that stronger country-level investor protection mechanisms are associated with increased dividend payouts.On the other hand, the substitution model argues that, in markets with weak investor protection, managers have incentives to alleviate investors' concerns about insiders' misappropriation of retained cash flows by paying dividends (Jensen, 1976).In other words, in those markets, dividends serve as a mechanism to mitigate potential agency conflicts (Zhao, 2000;Setiawan and Kee Phua, 2013).
Despite the outcome and substitution models providing different predictions regarding the association between investor protection and dividend payouts, we postulate that in countries with weak investor protection, managers may have more opportunities to engage in activities that benefit themselves at the expense of shareholders.Therefore, external investors might prefer to receive an immediate dividend rather than wait for the potentially unreliable capital gain from future investment.From the managers' perspective, to attract external investment, there may be a pressing need to distribute any free cash flows accumulated within the firm.In environments characterized by weak investor protection, the distribution of dividends may rely more on previously realized cash flows rather than future expected earnings growth.As a result, the signaling value of dividends about future earnings should be weaker in low investor protection environments than in strong investor protection environments.The second hypothesis proceeds as follows: H2: When investor protection is weak, there should be a weaker association between dividend changes and earnings growth than when investor protection is strong.

Model specification
To examine H1, we specify the following model in accordance with Ham et al. (2020) to employ the event window approach: where ΔE t+i equals the annual earnings growth in the ith year after a dividend change in period t.To compute ΔE t+i, we first identify the consecutive reporting periods surrounding the observed dividend change date.Unlike in the US where listed firms typically file quarterly reports, most non-US markets only require firms to file earnings semi-annually.To establish the baseline period that categories these consecutive reporting periods into pre-and post-dividend change phases, we identify the most recent reporting period end following the dividend declaration date and designate this period as t.Then the annual earnings for the first year post-dividend change is the sum of earnings before extraordinary items (IBQ) from quarters (or semiannual intervals) t + 1 (first subsequent semiannual interval) to t + 4 (second subsequent semiannual interval) in markets with quarterly (semi-annual) reporting.Similarly, annual earnings for the second and third year are calculated based on IBQs from quarters t + 5 to t + 8 (corresponding to semiannual intervals t + 3 to t + 4) and IBQs from quarters t + 9 to t + 12 (corresponding to semiannual intervals t + 5 to t + 6) in markets with quarterly (semi-annual) reporting, respectively.We extract the 4 (2) consecutive quarters (semiannual intervals) before the period t and aggregate them into annual earnings before period t.ΔE t+i is the difference between annual earnings in year t + i and the annual earnings before period t, divided by the market capitalization one year before the dividend announcement.If the currency of the financial reporting is different from that of the share price, we convert both currencies into USDs and compute the dependent variables.ΔDIV t is a percentage change in dividends that equals the dividend announced in the current period minus the most recent dividend announced previously within the prior 365 days, divided by the previously announced dividend.X collects a set of control variables, including annual earnings and earnings changes before period t and cumulative abnormal market reactions preceding each dividend announcement over several time windows.In particular, we add two consecutive annual earnings (IBQ(t-i) with i = 1,2) just before the dividend announcement into the model.For each of these annual earnings, we subtract the earnings for the previous t year to calculate the earnings change (ΔIBQ(t-i) with i = 1,2).We create five cumulative market reactions over different windows, namely [-20,-3], [-40,-21], [-60,-41], [-120,-61], and [-240,-121], with dividend announcement date as 0, to capture the information embedded in stock prices in relation to the dividend announcement and future earnings growth.Furthermore, as Grullon et al. (2005) points out the non-linear effect of control variables may change the inferences from our model, we also introduce a battery of interactions including: 1) the product of a variable indicating a control variable is negative and the variable, 2) the product of a variable indicating a control variable is negative and the variable squared, and 3) the product of a variable indicating a control variable is positive and the variable squared.All variable definitions are included in Appendix.We set the maximum value of ΔDIV t as 2. For other continuous variables, we winsorize the top and bottom 1 % to mitigate the effects of outliers.Unless stated otherwise in this study, standard errors are clustered by the year of dividend declarations.In addition, we control for year, industry and country fixed effects to remove the effect of year, industry and country invariant factors.β 1 is the coefficient on the variable of interest.H1 predicts that estimated β 1 is not significantly different from 0.
H2 proceeds to examine how variation in investor protection affects the information conveyed by dividend changes across different countries.We expand Equation ( 1) by adding the interaction between investor protection and dividend changes (ΔDIV t × IP): We use is an indicator variable measuring the legal origin of a country to proxy the strength of investor protection in a country.La Porta et al. (2000b) suggests that common law countries have the strongest form of investor protection.In subsequent cross-country studies, common law has been widely portrayed as a proxy of good investment protection (La Porta et al., 2002;He, 2008;Francis et al., 2001).To be consistent with H2, the estimated β 2 should be positive.

Sample selection
We retrieve our financial data from Compustat Global and Compustat North America.The stock prices data are obtained from Center for Research in Security Prices (CRSP), covering 38 countries from the period 1971 to 2021.For the US observations, we apply the filters in prior studies (Ham et al., 2020;Benartzi et al., 1997;Nissim and Ziv, 2001) in the selection process.The initial sample excludes financial firms (SIC code 6000-6999), which consists of firms issuing common shares (share codes 10 and 11) listed in the three main exchanges (NYSE, AMEX and NASDAQ).An announced dividend must be regular and quarterly (distribution code 1232), preceded by another dividend announced in the prior 180 days.In addition, we drop dividend announcements that might be contaminated by other distribution events since the prior announcement (e.g., other types of dividends or stock splits).For the non-US observations, for each GVKEY, we require the affiliated issues to be the primary issue, which should carry the right to participate in profit distribution.After this requirement, if there are still multiple issues associated with a particular GVKEY, we keep the issue with the longest history.In non-US countries, most firms pay either semi-annual or annual dividends, with a small proportion paying quarterly dividends and very few pay dividends more frequently than quarterly.Thus, we only delete observations paying dividends more than 4 times a year, rather than restrict our sample to only quarterly dividend payers.For both US and non-US dividend announcements, once we identify the dividend announcement quarter, surrounding which we require the availability of adjacent quarterly earnings over the time window of [-8,+12].
To calculate stock price reactions over varying time windows before dividend announcements, we extract stock price level data from CRSP for US shares and Compustat Global for non-US shares.We delete observations with either missing stock return or firm fundamental data.
For firms listed on multiple exchanges, we only keep the observations if the country code of headquarters is aligned with the country code of stock exchanges to avoid double counting.To enter the sample, a country-year must have at least 50 GVKEYs.

Descriptive statistics
Table 1 presents an overview of dividend changes for the sample.There are large differences in the distribution of dividends between US and non-US firms.The dividend payment in the US is extremely stable, in 82.54 % of observations dividend payments remain unchanged, and dividend decreases are extremely rare (1.43 %).The remarkable stability of dividend payout distribution in the  absolute value of dividend decreases (37.75 %), suggesting that non-US firms increase dividend much less conservatively.Table 2 presents the magnitude of the earnings changes following dividend changes.Table 2 Panel A reports the mean and median earning changes for US firms.For firms with dividend increases, the greater the dividend increase, the higher the earnings growth is in all the subsequent three years.The firms with the largest dividend increase also see their earnings grow faster than the smallest dividend increase group.However, the picture of dividend decreases is unexpected.The firms with dividend decreases experience a great positive change in earnings, compared with those with no dividend changes.Furthermore, the median value of earning changes with dividend decreases is even higher than those with the fourth quartile dividend increases, especially in years 2 and 3.The result in Panel A is quite different from the result inBenartzi et al. (1997). 2he difference may be caused by two reasons.First, the sample composition is different.Benartzi et al. (1997)'s sample contains 7186 company-year observations from 1979 to 1991 while the US data in this study has 176,271 observations.The second is that the change in earnings is computed differently.Benartzi et al. (1997) uses the fiscal year approach while we adopt the event window approach advocated by Ham et al. (2020).
Panel B, Table 2, shows the results of firms in countries with weak investor protection (code law countries).For year 1, earning changes are associated with the changes in dividend payment.Earnings changes are constant for the unchanged dividend group and monotonically increase with the dividend increase.However, the earning changes in year 2 and year 3 are different from the first year.In the following two years, none of the groups of increasing dividends firms shows significantly faster earnings growth than the nochange or decrease group, nor does the fourth quartile increase group grow faster than any low dividend increase group.The findings suggest that in countries with weak investor protection, although dividends can signal earnings growth, this signaling is very limited.
In comparison, descriptive statistics for firms in countries with strong investor protection (common law countries) are more in line with dividend signaling theory.Firms with higher dividend increases experience greater earnings increases for the next three years.However, most earnings changes in the lowest quartiles have a negative sign.The first quartile group experiences a greater earning decrease than the no-change group.In sum, the results in Table 2 indicate that, across the 38 markets, dividends signal earnings.In countries with strong investor protection, the signaling effect of dividend changes regarding earnings changes seems stronger.

Do dividend changes predict future earnings changes for global markets?
The results of Eq. ( 1) are presented in Table 3.For the US market, there are strong significant positive coefficients on the dividend changes (0.015 with a t-statistic of 9.54 in year 1, 0.013 with a t-statistic of 5.45 in year 2 and 0.009 with a t-statistic of 2.82 in year 3).The coefficients are similar to the results of Ham et al. (2020), but the coefficients are lower.The sign of the coefficients for most of the control variables are in the same direction as in Ham et al. (2020), deviations may be caused by the different sample period. 3The coefficient in Column (1) indicates that a one standard deviation change in dividends paid (ΔDIV) is associated with a 0.0275 standard deviation change in earnings growth in the first year after the dividend change.We calculate the marginal effect of E(t-1) a one standard deviation change in E(t-1) results in a 0.0725 standard deviation change in earnings growth in year 1, to assess the economic significance of dividend signaling.This comparison shows the effect of dividend signaling is about 38 % of that of the level of earnings in the half year before the earnings announcement.The results indicate that dividends are adjusted based on long-term earnings prospects.That is, when managers forecast a change in earnings one to three years from now, they signal this information to investors by changing the levels of dividends.
In column (4) − (6), we apply Ham et al. (2020)'s model in a global context to examine whether dividend signaling theory is applicable in other countries.For these markets, there are strong significant positive coefficients on the dividend changes in all three years (0.0087 with a t-statistic of 13.43 in year 1, 0.0069 with a t-statistic of 8.09 in year 2 and 0.0070 with a t-statistic of 5.17 in year 3).The results are consistent with the US findings, although the regression coefficients for ΔDIV are lower in the non-US setting.Thus, the dividend signaling theory is valid for non-US firms even if investor expectations are different compared to their USA counterparts and their dividends change more frequently compared to US firms.

The effect of investor protection on the signaling strength of dividend changes
Table 4 reports the results of Eq. ( 2) which examines the relation between dividend signaling and investor protection.In Columns (1)-( 3), using the whole sample, we measure the investor protection employing the legal system (code vs common law).The coefficients on ΔDIV now capture the effect of dividend signaling in countries with a code law background, while the coefficients on ΔDIV×Common denote the incremental effect of dividend signaling in countries with a common law background.The sum of two coefficients (e.g., in Column (1)) gauges the dividend signaling in common law countries.The coefficient on ΔDIV is only statistically significant when the year 1 earnings growth is the dependent variable, showing that dividend signaling in code law countries is only available in the short term, while an insignificant interaction coefficient for Year 1 suggests that there is no significant difference in the short-term impact of dividend changes on earnings growth between common and code law countries.However, the positive and significant joint coefficients in subsequent years (Year 2 and Year 3) indicates that the longer-term effects of dividend signaling vary between these two legal systems, with firms located in common law countries showing more effective dividend signaling in the long

Table 3
Information content of dividend changes for the US and non-US markets.This table reports the regression result of Eq(1) using the Ham et al. (2020)'s model for the year 1970 to 2021.The dependent variable is the annual change in earnings before extraordinary items.The dividend change is defined as the current dividend less the prior dividend, scaled by the prior dividend.The controls for past earnings change, past earning levels, as well as the nonlinear controls for earning changes and levels are included.Six return variables capturing stock prices changes over 240 trading days around the dividend declaration date are also included.Clustered standard errors by dividend declaration year.The t-statistics are reported in parentheses.***, **, * Denote statistical significance at the 1 percent, 5 percent, and 10 percent levels of two-tailed tests, respectively.

Table 4
Effect of investor protection on the information content of dividend changes.Table 4 reports the regression results of Eq(2).In Column (1)-( 3), the variable IP in Eq(2) is measured by Common.Common is a dummy variable that equals to 1 for countries with common law tradition, equals to 0 for countries with code law tradition.F-statistics tests the null hypothesis that ΔDIV+ΔDIV×Common is equal to 0. The remaining variables are defined as in Table 3. Clustered standard errors by dividend declaration year.The t-statistics are reported in parentheses.***, **, * Denote statistical significance at the 1 percent, 5 percent, and 10 percent levels of two-tailed tests, respectively.run compared to firms located in code law countries.The results show that firms from code law countries exhibit a limited, short term, ability to signal earnings through dividends, while firms from common law countries can use dividends more effectively to signal earnings long-term.This finding is consistent with previous discussions suggesting that, in countries with weaker investor protections, dividends are more likely to correlate with current earnings changes rather than future earnings changes.Stronger investor protections in common law countries allow managers more scope to signal earnings through dividends, enabling dividends as a more reliable indicator of future financial performance.When we restrict our analysis to the sample excluding the US firms, the results (in Columns ( 4)-( 6)) are very similar.The main difference being that in the non-US sample firms based in code law countries seem to be able to signal some medium term earnings through their dividends.However, the earnings signal contained in dividends is significantly stronger in firms based in common law countries.
Overall, the results in Table 4 suggest that firms in countries with stronger investor protection show a considerably stronger association between dividend changes and future earnings, which supports our second hypothesis.

Impact of investor protection on dividend signaling with alternative measures
Table 5 examines on how different measures of investor protection influence on dividend signaling, using a range of indicators of investor protection as outlined by McLean et al. (2012) and La Porta et al. (1998): the disclosure index, measuring the quality of disclosures around new stock issues; the liability index, measuring the ease with which an investor can pursue an issuer and their related parties; the investor protection index, which is a comprehensive measure based on the liability and disclosure index measuring the protection of investors; and the anti-self-dealing index, measuring restrictions on self dealing between a firm and a related party.Each of these components plays a critical role in shaping the transparency and accountability of corporate behavior, which in turn may affect how dividends are functioned by managers.
Consistent with the observations in Table 4, the strength of the earnings signal contained in dividend changes is significantly stronger as investor protection increases when we measure investor protection using the disclosure index, the liability index and the investor protection index.The interaction terms ΔDIV×Disclose, ΔDIV×Liability, and ΔDIV×Protect generally exhibit positive and (marginally) significant effects across three years horizon in the full sample (panel A) and the sample excluding the USA (panel B).We do not find significant evidence of dividends more strongly signaling earnings when we use the anti-self-dealing index.These results highlight the importance of investor protection in strengthening the informational role of dividends regarding future earnings, particularly in countries with strict regulatory environment.

Additional analysis
The evidence presented thus far establishes that investor protection moderates the strength of the earnings signal contained in dividend changes.In this section, we provide additional analyses to examine whether the above relation varies with the varied levels of institutional ownership and law enforcement.

Impact of legal origin and institutional ownership on dividend signaling
In this section, we assess whether the main result is robust to the concentration of institutional ownership across firms.This analysis is crucial as institutional ownership can significantly affect the interpretation and impact of financial regulations and corporate behaviors (Shleifer and Vishny, 1997;Ferreira and Matos, 2008).On the one hand, when dividends serve as a signal to the market about the firm's quality and its prospects.Institutional investors, with their demand for regular dividends, reinforce this signal, demonstrating their confidence in the firm's profitability and stability.(Short et al., 2002).On the other hand, the influence of institutional ownership on dividend signaling can be negative if institutional investors have a shorter-term focus, they might pressure the company for higher dividends irrespective of long-term growth plans.By exercising their voting rights and influencing board decisions, they can pressure management to adopt policies that are in their own interest (Brav et al., 2005).
We focus on the variable High_Institution, which measures the level of institutional ownership.We collect institutional ownership information from factset, and calculate the total number of shares owned by institutional investors at the end of each quarter.To compare across firms we scale total shares owned by institutional owners by total shares outstanding.We mark observations as having high institutional ownership if institutional owners own a proportion of shares outstanding that is at or above the median for their country and year.We then add this indicator to model (2), as well as an interaction term with the change in dividends, to test is the signaling strength of dividends changes significantly based on the share ownership of institutional investors.
Our analysis shows that high institutional ownership consistently exerts a positive and significant impact on future earnings growth across all time periods.This suggests that institutions play a crucial role in enhancing financial performance.In contrast, the interaction between dividend changes and high institutional ownership (ΔDIV×High_Institution) generally does not show significant results.This insignificant result suggests that while institutional ownership positively affects company earnings, its interaction with dividend changes does not significantly influence future earnings growth.In addition, our main result remains unaffected, indicating that the earnings signal contained in dividends is still higher in firms located in common law countries, even after controlling for their relative institutional ownership.Overall, these results indicate that the signaling power of dividends is not enhanced or diminished by the level of institutional ownership, highlighting the complexities in how dividends and ownership structures affect company earnings.

Table 5
Other measurement of investor protection.Table 5 reports the regression results of Eq(2), with the variable IP in Eq(2) is substituted by other measurement of investor protection in Mclean et al. (2012).Disclosure refers to the requirements for transparency in financial reporting which is quantified as the arithmetic mean of six subindices.Liability measures the ease with which investors can sue various parties involved in issuing securities due to misleading statements in a prospectus and is calculated as the arithmetic mean of three subindices.Protect measures the written law and its enforcement.Antiself measures the regulation of transactions between two firms controlled by the same person.The remaining variables are defined as in Table 3. Clustered standard errors by dividend declaration year.The t-statistics are reported in parentheses.***, **, * Denote statistical significance at the 1 percent, 5 percent, and 10 percent levels of two-tailed tests, respectively.Impact of legal origin and legal enforcement on dividend signaling Our measures of investor protection are generally based on the legal framework in which firms operate.However, the strength of the law is separate from how diligently and effectively it is enforced.Davidson and Pirinsky (2022) find that enforcement is effective in restraining insider trading even when there are a few regulations for firms.Nguyen (2021) also finds that even if the legal system remains the same, a decrease in legal enforcement can lead to a rise in white-collar crime.The prior literature suggests that although the law is well established, violations will still occur when there is limited legal enforcement.
In Table 7 we test the influence of the strength of enforcement of the law on our results.We create Enforcement, which indicates if a country has a public or private enforcement strength that is at or above the median for the sample.To determine the strength of public and private enforcement we make use of the enforcement indices developed in La Porta et al. (1998).The interaction term ΔDIV×Enforcement is added to model (2) to assess how variations in enforcement levels affect the relationship between dividend changes and future earnings.Contrary to previous literature, our findings indicate that the level of enforcement does not significantly alter the main results.The interaction terms involving enforcement show minimal impact across all periods.This result suggests that while enforcement practices are crucial for maintaining market discipline and regulatory compliance, they may not necessarily enhance or impair the signaling power of dividends.

Conclusions
Linter (1956) shows that managers believe investors prefer stable dividends, and they cater to such a preference by keeping dividends level unless certain increased dividend payouts can be maintained.Dividend stickiness implies cutting dividends is costly for firms, which ensures the presence of a separating equilibrium where managers increase dividends only if future earnings growth can sustain dividend increases.Prior empirical studies fail to document evidence to support the above signaling theory of dividends.As pointed out by Ham et al. (2020), this lack of evidence can be attributed to poor identification of pre and post dividend increase earnings.By using a refined method, Ham et al. (2020) provides evidence in support of the dividend signaling theory.
As dividends in non-US countries are not as sticky as those in the US, we suspect dividend increases may not signal earnings increases outside of the US.We extend Ham et al. (2020) into a non-US setting which consists of 38 markets to investigate whether dividend increases predict earnings increases.Our results suggest that dividends signal earnings growth in this broad sample, although the signaling strength is lower in non-US markets than in the US market.
Investor protection can be a factor which affects the cost of cutting dividends for firms.We predict that in countries featured by weak (high) investor protection, the signal value of dividend increases regarding future earnings is low (high).Using the crosssectional country-level variation in investor protection, we show that dividends effectively signal future earnings growth in countries with strong investor protection, while changes in dividends do not show a sustainable significant association with future earnings in countries with weak investor protection.Our study contributes to literature examining the signaling theory of dividends in non-US markets and documents some necessary conditions for dividend changes to signal future earnings changes.

Table 6
Impact of legal origin and institutional ownership on dividend signalling.Table 6 reports the influence of Dividend Changes on Future Earnings Changes by Legal Origin and Institutional Ownership.High_Institution a measure of institutional investment intensity, a higher value of the Enforcement variable indicates a more robust legal enforcement.The remaining variables are defined as in Table 4. Clustered standard errors by dividend declaration year.The t-statistics are reported in parentheses.***, **, * Denote statistical significance at the 1 percent, 5 percent, and 10 percent levels of two-tailed tests, respectively.Difference between annual earnings in year t + i and the annual earnings before quarter t, divided by the market capitalization one year before the dividend announcement.E t+i is the annual earnings growth in the ith year after a dividend change in quarter t

High_Institution
Equal to one the firm has above median institutional ownership by country year, as measured by total shares owned by institutional investors scaled by total shares outstanding, and zero otherwise.

Table 1
Distributions of dividend changes across 38 countries.This table reports the distribution of dividend events across 38 countries over the 1970-2021 sample period.The sample contains 253,238 dividend announcements made by listed firms during the period 1971-2021.Dividends changes are defined as current dividend less the prior dividend divided by the prior dividend.
US market is consistent with the evidence inLinter (1956)that the dividend payment is a pre-commitment device to investors, which might be necessary for dividends to signal earnings.In contrast, among all non-US firms, most dividend increases are much more common (49.94 %), while in 22.13 % of observations the dividend payment decreases, and only in 27.93 % of observations the dividend payment remains unchanged.The unstable dividend payments in non-US markets suggest that the long-term commitment function of dividends is likely to be weakened or even nullified in other markets.In addition, although the frequency of dividend decreases in US firms is low, the decreases are greater.The absolute value of dividend decreases (43.92 %) is almost double that of dividend increases (19.96 %), implying that US firms are conservative in increasing dividends (small magnitude) but very aggressive in dropping dividends once determined.While the average dividend increase (43.64 %) in non-US firms is almost 10 % higher than the Table 1 (continued )

Table 2
Earnings changes following dividend changes.This table presents changes in earnings in the year of and the years following dividend changes.Each firm-year observation in the sample is categorized into either one of the dividends increasing quintiles, no change in dividends, or a dividend decrease.Panel A reports the distribution of earnings changes for firms listed on the NYSE, Amex, or Nasdaq exchanges.Panel B and C describes the distribution of earnings changes based on investor protection.The degrees of investor protection are classified as countries with common-or code-law tradition.

Table 7
Impact of legal origin and enforcement level on dividend signalling.Table7reports the influence of dividend changes on future earnings across different legal origin and levels of legel enforcement.The interaction terms ΔDIV×Enforcement examines how the level of law enforcement might modify the impact of dividend changes on earnings changes.The remaining variables are defined as in Table4.Clustered standard errors by dividend declaration year.The t-statistics are reported in parentheses.***, **, * Denote statistical significance at the 1 percent, 5 percent, and 10 percent levels of two-tailed tests, respectively.