The Effect of Past Earnings and Dividend Patterns on the Information Content of Dividends When Earnings Are Reduced

Dataset

The sample comprises all firms on Compustat for the sample period 1986–2005 that meet the following criteria: (a) industrial firms,¹⁶ (b) non-missing values for dividends and earnings before extraordinary items, and (c) availability of at least one annual loss or earnings reduction preceded by positive annual earnings and an annual dividend payment. Consistent with the literature, we use annual data.¹⁷ All variables are winsorized at the 1st and 99th percentiles.

The resulting 4,873 firm-year observations are classified into two subsamples according to the number of annual earnings and dividend payments prior to their first annual loss. The first subsample includes 3,194 less-established firms, which are those firms that incurred positive annual earnings and dividends for at most three consecutive years, then suffered a loss or a reduction in earnings. The second subsample includes 1,679 more established firms, those that had been producing positive annual earnings and dividends for at least seven consecutive years, and in the subsequent year incurred a loss or an earnings reduction.¹⁸

Out of the 3,194 firms in the less-established subsample, 1,236 firms (39%) announced dividend reductions or omissions. For the established subsample, 515 firms (30%) reduced or suspended dividend payments, while the remaining 1,164 firms (70%) sustained or increased dividends. Table 1 reports the percentage of dividend reduction relative to dividend omission announcements for the two subsamples. Some 888 (72%) of the 1,236 dividend reductions in the less-established sample of firms result in a still positive level, whereas the remaining 348 reductions represent suspensions of dividend payments. For the established sample of firms, 487 (95%) of the 515 dividend reductions represent dividend decreases, while only the remaining 28 firms (5%) proceeded with complete omissions of dividend payments, indicating that managers of established dividend-paying firms are more reluctant to suspend dividend payments.

Table 2 describes the distribution of the ‘Time’ subsamples. A firm belongs in sample Time i (where i= 1, 2, 3 for the less-established sample and i= 7, 8, 9, 10 for the established sample) if it reports a loss or an earnings reduction in year t (where t is the event year), after having positive earnings and dividends for i years prior to the first annual loss or earnings reduction. For example, in year t, Time 2 firms incurred a loss or an earnings reduction and had positive earnings and dividends for years t − 1 and t − 2.

Descriptive Statistics

Table 3 reports descriptive evidence for the main variables used in our regression analysis. By and large, the empirical literature proxies profitability using return on assets, ROA_t=IB_t / TA_t−1 where IB_t is income before extraordinary items (annual Compustat data item #18) and TA_t−1 is lag total asset (annual Compustat data item #6).¹⁹IB_t includes special items that in most cases comprise negative expenses related to restructuring charges, write-offs, impairments and so forth (Collins et al., 1997; Bradshaw and Sloan, 2002). Hence, low earnings realizations due principally to special items are more likely to be transitory (Burgstahler et al., 2002). However, future earnings and dividend payouts are predominately affected by the permanent component of earnings, and thus, special items by their nature would be expected to have little or no impact both on future earnings and on dividend decisions (e.g., DeAngelo et al., 1992; Kormendi and Zarowin, 1996; Penman and Sougiannis, 1997; Skinner, 2004). Accordingly, we posit that in examining managers’ adjustments of dividends with respect to information about low earnings realizations should be based on earnings excluding special items. ‘Sustainable’ earnings are proxied as return on assets net of special items, that is, ROA_t= (IB_t−SPI_t) / TA_t−1 where SPI_t represents special items (annual Compustat data item #17).

We also include net cash flows from operating activities (CFO_t), scaled by lag total assets (annual Compustat data item #308 divided by TA_t−1), since the literature suggests that cash flows from operations contain information that explains dividend changes (Charitou and Vafeas, 1998; Charitou, 2000; Joos and Plesko, 2004).²⁰ Following Fama and French (2001), there are controls for firm size (SIZE_t) and market-to-book (MTB_t) ratio. Firm size is a commonly used proxy for the firm's information environment, as larger firms institute better mechanisms for periodic information releases (Zeghal, 1984; Atiase, 1985; Donnelly and Walker, 1995). SIZE_t is proxied by the natural logarithm of the firm's total assets.²¹MTB_t is defined as the market value of equity (i.e., the closing price [annual Compustat data item #199] times shares outstanding at fiscal year end [annual Compustat data item #25], scaled by book value of equity [annual Compustat data item #60]), and is used as a proxy of the investment opportunity set of the firm. Consistent with DeAngelo et al. (2006), we employ firm sales growth rate (SALEGR_t) as a proxy for growth, and retained earnings to total common equity (RETTE_t).²² Finally, we include the debt–equity ratio, DEBTEQ_t, to control for the possibility that the firm is close to its covenant restrictions, which may influence its dividend-paying behaviour (Duke and Hunt, 1990; Press and Weintrop, 1990; Aivazian et al., 2006). DEBTEQ_t is defined as long-term debt plus debt in current liabilities (annual Compustat data item #9 plus annual Compustat data item #34), all scaled by total equity (annual Compustat data item #60).

The results presented in Table 3 show that the mean and median ROA_t, CFO_t, SIZE_t and RETTE_t are greater for the established subsample (Panel A) versus the less-established subsample (Panel B). They are in line with the conventional finding that established dividend payers are larger and more profitable (Fama and French, 2001; DeAngelo et al., 2004; DeAngelo et al., 2006). Furthermore, SALEGR_t and DEBTEQ_t are greater for the less-established sample of firms, showing that firms with greater growth rates have higher debt covenants and a stronger incentive to retain cash, and thus exhibit less-established patterns of past dividend payments.

Finally, Table 4 shows correlation coefficients between the main variables. Panel A reports correlation coefficients for the less-established sample, Panel B for the established sample, and Panel C for the merged sample. DIV_REDUCTION is a qualitative variable that takes the value 1 if dividend payments are reduced or omitted, and zero if dividend payments are either increased or sustained. According to all panels of Table 4, SALEGR_t is positively correlated with ROA_t, indicating that earnings increase with sales. Moreover, ROA_t is positively correlated with CFO_t, MTB_t, SPI_t and RETTE_t, whereas ROA_t is negatively correlated with SIZE_t and DEBTEQ_t. The correlation coefficient between ROA_t and DIV_REDUCTION is more negative for the established sample firms, indicating that the association between the incidence of dividend reductions or omissions and current profitability is higher for the established firms than for the less-established firms.

The Information Content of Dividend Changes in Explaining Future Earnings

The first hypothesis posits that, in the event of an earnings reduction, dividend policy changes have more information content in explaining future earnings, the longer are past earnings and dividend patterns. Information content relates to the predictive ability of current earnings with respect to future earnings. That is, we conjecture that knowledge that a firm has reduced its dividends improves the ability of current earnings to predict future earnings—the stronger the historic-patterns effect.

To formally test this hypothesis, we use the following OLS model:

()

where DIV_REDUCTION= 1 if dividend payments are reduced or omitted, and zero if dividend payments are either increased or sustained;

ESTAB= 1 if the firm belongs in the established sample of firms, and zero if it belongs in the less-established sample of firms; and

NON-LOSS= 1 if in the event year the firm incurred an earnings reduction, and zero if it incurred a loss.

All the remaining variables in the model are either defined as before, or represent interactions between the variables already described (e.g., ROA_t*DIV_REDUCTION is the interaction between ROA_t and DIV_REDUCTION).

Table 5 reports the OLS regression results for model (1). Column (1), sets forth the results for a variation of model (1) that excludes some of the control variables, while columns (2) and (3) present variations of the full model. All tests of statistical significance are based on the White (1980) heteroscedasticity-consistent standard errors and control for firm fixed-effects.

Regression results with respect to model specification (1) yield a higher adjusted R² (46.4%) compared to the model presented in column (1) (R² = 23.4%). In fact, the adjusted R² reported in columns (2) and (3) is much higher compared to similar model specifications reported in prior studies. For example, DeAngelo et al., 1992, in their OLS regressions of future earnings on current earnings, report an adjusted R² of at most 29.7%, whereas Charitou (2000) reports an adjusted R² of 17.43%. This difference may be attributed to several factors. On the one hand, these studies consider only loss firms with established track records. In contrast, our sample contains both loss firms and earnings-reducing firms with varying degrees of past earnings and dividends records. Moreover, the OLS regression models employed in the aforementioned studies fail to incorporate the alternative control variables considered here. However, controlling for a broad variety of firm characteristics that may influence dividend policy decisions (i.e., beyond earnings and cash flows) is crucial since the dividends literature offers only rough guidelines on the key determinants of the decision to pay dividends and on the best ways to capture those determinants empirically (DeAngelo et al., 2006). Thus, employing explanatory variables that potentially affect dividend policy decisions results in a better specified model yielding a substantially higher adjusted R². Consequently, the relevant regression coefficients allow more reliable inferences.

In all model specifications, ROA_t is positive and statistically significant, indicating the importance of current earnings as a predictor of subsequent-year earnings. DIV_REDUCTION is negative and statistically significant, revealing that dividend policy has information content in explaining subsequent earnings, regardless of past earnings and dividend patterns. The negative sign denotes that firms that reduce or eliminate their dividend payments during the event year have lower earnings in the subsequent year. Moreover, using the interaction variable ROA_t*DIV_REDUCTION, we investigate whether dividend policy changes improve the ability of current earnings to predict future earnings. Lack of statistical significance indicates that the predictive ability of ROA_t does not depend on dividend policy for the less-established subsample.

The qualitative variable ESTAB is included to test whether the patterns of prior earnings and dividends explain future earnings. Although ESTAB does not exhibit statistical significance, we investigate the information content of past earnings and dividend patterns when ESTAB is interacted with current earnings and dividend policy changes. First, the interaction variable ESTAB*ROA_t is employed to test whether past earnings and dividend patterns improve the predictive ability of current earnings with respect to future earnings. The positive coefficient on ESTAB*ROA_t (0.257 with p-value = 0.000) illustrates that among earnings-reducing or loss firms, the magnitude of earnings conveys incrementally more information about earnings prospects the longer are the historic earnings/dividend patterns.

Nonetheless, the main variable of interest in testing whether past earnings and dividend patterns, along with dividend reductions, are significantly associated with the predictive ability of ROA_t, is the interaction variable ESTAB*ROA_t* DIV_REDUCTION. In accord with H1, the estimated coefficient on ESTAB*ROA_t*DIV_REDUCTION is negative and statistically significant at the 1% level (−0.187 with p-value = 0.000), beyond ESTAB*ROA_t and ROA_t*DIV_REDUCTION. The negative sign denotes that dividend cuts significantly burden the overall positive relationship between ROA_t and ROA_t+1, the longer the string of prior earnings and dividends. In other words, established firms that proceed with dividend cuts have lower earnings in the subsequent year vis-à-vis less-established firms. Thus, evidence herein supports that dividend reductions signify that low earnings will persist in the future the more established the earnings patterns.²³

Most of the control variables exhibit statistical significance below the 5% level (with the exception of MTB_t). The NON-LOSS dummy and the interaction variable NON-LOSS*ROA_t are used to assess the informativeness of losses vis-à-vis earnings reductions with respect to future profitability. Other than DeAngelo et al. (1992) and Charitou (2000), the lack of statistical significance on NON-LOSS*ROA_t illustrates that among earnings-reducing or loss firms, dividend reductions significantly improve the ability of current sustainable earnings to predict future earnings, regardless of whether current earnings constitute positive (while reduced) earnings or losses.²⁴

Consistent with our expectations, the estimated coefficients on cash flows (CFO_t) and retained earnings (RETTE_t) are positive, indicating that firms with greater cash holdings exhibit higher subsequent earnings. Size (SIZE_t), sales growth rate (SALEGR_t) and debt to equity (DEBTEQ_t) each exhibit a negative coefficient, showing that larger firms with higher sales growth and higher debt covenants have lower earnings in the following year. The negative coefficient on SALEGR_t and DEBTEQ_t can be justified to the extent that these two variables proxy for future growth prospects. Firms with higher growth opportunities exhibit higher sales growth rates and higher debt covenants that impair short-run profitability (Grullon et al., 2002; DeAngelo et al., 2006). The negative sign on SIZE_t is probably due to the fact that larger firms report larger losses (or earnings reductions) and thus, lower year-ahead earnings. The negative sign on SPI_t seems reasonable given that evidence in Table 3 shows that, on average, our sample firms incur negative special items.

In summary, the important implication of the regression analysis set forth in this section is that beyond the aforementioned control variables, the coefficient of the interaction variable ESTAB*ROA_t*DIV_REDUCTION remains negative and statistically significant. Thus, in line with H1, the information content of dividends is significantly associated with past earnings and dividend patterns, as dividend changes strengthen the ability of current sustainable earnings to predict subsequent earnings the more established past earnings and dividend patterns have been prior to the first annual earnings reduction.

Association Between Magnitude of Earnings Reduction and Dividend Policy Decisions Conditional on Past Earnings and Dividend Patterns

Thus far, our results support that among earnings-reducing or loss firms, past earnings and dividend patterns strengthen the information content of dividends. In particular, we show that dividend reductions or omissions enhance the ability of current earnings to predict future earnings the more established past earnings and dividends records have been prior to the first annual loss or earnings reduction. Essentially, our results describe the effect of historic earnings/dividend patterns and dividend reductions in signalling the persistence of low earnings realizations. In this respect, dividend reductions are a stronger signal that current low sustainable-earnings realizations will persist into the future for established versus less-established firms.

Thus, two inferences can be drawn. First, current sustainable earnings significantly affect the dividend-payment decision. For instance, the literature supports that losses or low earnings realizations on the part of dividend-paying firms are often short-lived and reflect the capitalized effect of future negative cash flows (e.g., Basu, 1997; Burgstahler et al., 2002; Skinner, 2004; Joos and Plesko, 2005). Accordingly, given management conservatism with respect to dividend policy decisions, if losses or reductions in earnings are transitory, then dividends should not be cut (Brav et al., 2005). However, if losses or earnings reductions are the outcome of a fundamentally ‘bad’ performance (such as lower revenues or structural inefficiencies), then sustainable earnings (i.e., earnings before extraordinary and special items) will make reduced dividends more likely. Second, given that managers are keener to avoid dividend cuts the more established are past earnings/dividend patterns, dividend reductions or omissions for established firms are more likely to be the outcome of sufficiently serious earnings difficulties with persisting effects. Under this rationale, established firms that reduce or omit dividends should incur losses or earnings reductions of greater magnitude compared to less-established firms. Accordingly, the importance of sustainable earnings in generating dividend reductions should become higher the longer are past earnings/dividend patterns.²⁵

Before moving on to test formally the relationship between earnings and dividend patterns, earnings magnitudes and the decision to pay dividends, we present an outline of the earnings performance of our sample firms that proceed with dividend reductions or omissions. The purpose of doing so is to acquire an initial insight into the relationship between past earnings/dividend patterns and earnings magnitude for dividend-reducing firms. Thus, we first assess whether established firms that cut dividends suffer earnings reductions of greater magnitude vis-à-vis less-established firms, by isolating and comparing the level of and changes in income before extraordinary and special items, NIB_t (=IB_t−SPI_t), and net income (NI_t), for the subsamples of firms that reduced or omitted dividend payments.

Table 6 presents descriptive statistics on the earnings performance of the subsample of firms that, for the event year, incurred a loss, and in the loss year reduced or suspended dividend payments. Similar information is reported in Table 7, but for the subsample of firms that reported earnings reductions. We report mean and median values for NIB_t, NI_t, ΔNIB, ΔNI, TA_t−1, ROA_t (= (IB_t−SPI_t) / TA_t−1) and ΔROA, as well as standard deviations. NI_t is annual Compustat data item #172.ΔNIB is NIB_t−NIB_t−1 (similarly, ΔNI=NI_t−NI_t−1 and ΔROA=ROA_t−ROA_t−1). Panel A and Panel B (in both tables) provide descriptive evidence for our subsamples of firms that belong in the less-established and established subsamples, respectively. In Panel C, we report tests of differences between the means (two-sided t-tests) and medians (two-sided Kolmogorov-Smirnov and Mann-Whitney U tests).

In accordance with our expectations, both loss- and earnings-reducing firms in the established subsample suffer significantly greater earnings reductions than those in the less-established subsample. Results shown in Panels A and B of Table 6 indicate that the mean and median NIB_t and NI_t are substantially more negative for the subsample of established dividend-reducing loss firms. The same holds for the subsample of firms that experienced earnings reductions and reduced or omitted dividend payments. The results in Table 7 show that the mean and median changes in profitability captured by ΔNIB and ΔNI are almost twice as negative for the established subsample. Nevertheless, ROA_t and ΔROA are shown to be more negative for the less-established subsample of firms; this is due to much lower lag total assets (e.g., mean TA_t−1 is two times smaller for the less-established subsample in Table 6, and two-and-a-half times smaller in Table 7). Panel C reveals that the means and medians of the variables under focus are statistically different from each other.

In summary, the descriptive evidence on the relationship between earnings magnitude and historic earnings and dividend patterns reveals that established dividends-reducing firms suffer lower earnings realizations (i.e., greater losses or earnings reductions) vis-à-vis the less-established dividends-reducing sample of firms.

Formally testing the relationship between historic earnings and dividend patterns, and earnings magnitude and the likelihood of dividend reductions, entails estimating logistic regressions, controlling for other firm characteristics that may influence dividend policy decisions. Hence, we estimate the following specification:

()

According to H2, among earnings-reducing or loss firms, the magnitude of sustainable earnings (as captured by earnings excluding extraordinary and special items) must have a greater impact on the likelihood of dividend reductions for established versus less-established firms.

Table 8 reports logistic regression analysis results for model (2). In the first column, the model is run using only the less-established subsample, while in column (2) we consider only the established subsample. Column (3) presents the full model when both samples are pooled in a single logistic regression. As expected, the coefficient of ROA_t is negative and statistically significant in all model specifications, indicating that the depth of earnings reductions significantly affects the decision to pay dividends. Accordingly, firms with higher earnings reductions (i.e., less positive earnings) or with more negative earnings (i.e., deeper losses), show a higher likelihood to proceed with dividend cuts. This result holds for both the less-established sample of firms (model specification (1)) and the established sample of firms (model specification (2)). Thus, beyond DeAngelo et al. (1992) and Charitou (2000), evidence thus far supports the notion that dividend reductions are more likely given lower earnings realizations, regardless of past earnings and dividends records.

More importantly, with respect to model specification (3) (where both samples are pooled together), the estimated logistic regression coefficient of the variable of interest, that is, ESTAB*ROA_t, is also negative and statistically significant (−7.636, p-value = 0.000) beyond ROA_t. These results can be interpreted as follows. First, in the case of earnings-reducing firms (i.e., those firms that exhibit a positive, while reduced, ROA_t), the historic-patterns effect dominates the earnings-reduction effect: Established firms (i.e., ESTAB= 1) exhibit a significantly more negative association between the probability of dividend cuts and ROA_t than less-established firms. In other words, among earnings-reducing firms, the longer past earnings/dividend patterns are, the more reluctant managers are to deviate from an ongoing commitment to paying regular dividends to shareholders. Second, in the case of loss-firms (i.e., when ROA_t is negative), the historic-patterns effect dominates the loss effect: A loss increases the probability of reducing or omitting dividends significantly more for established versus less-established firms. Thus, the longer the string of past earnings and dividends prior to the first annual loss, the more a dividend reduction signals that losses are not short-lived, but rather result from a substantial deterioration in profitability that is expected to persist, and accordingly, dividend cuts become more likely.²⁶

Viewed collectively, the negative and statistically significant coefficient on ROA_t indicates that a firm's current earnings problems, as captured by the magnitude of earnings (which represents either positive, albeit reduced, earnings or losses), significantly affects the probability of a dividend reduction. However, the statistical significance of ESTAB*ROA_t beyond ROA_t supports that a given magnitude of sustainable earnings in established firms has a significantly greater negative impact on the likelihood of dividend cuts compared to less-established firms. In other words, one dollar of sustainable earnings in established firms has much greater information content with respect to reducing or omitting dividend payments than one dollar of sustainable earnings in less-established firms.²⁷

The results in Table 8 also show that the coefficient on the interaction variable NON-LOSS*ROA_t is positive and statistically significant in all model specifications. Thus, the overall negative relationship between the probability of dividend reductions and positive earnings for the earnings-reduction firms (i.e., when NON-LOSS= 1) is smaller (i.e., the overall negative coefficient on ROA_t is lower) compared to the overall negative coefficient in the case of loss firms (i.e., when NON-LOSS= 0). This result is not surprising, as it illustrates that losses provide greater information content in explaining dividend reductions compared to (reduced but) positive earnings.

The negative and significant coefficients on SIZE_t, CFO_t and SALEGR_t indicate that larger firms with greater cash holdings and increasing sales revenues show a greater likelihood of keeping up with their dividend policies. The statistically insignificant coefficient on SPI_t is consistent with special items providing no incremental information on dividend policy decisions. Also, the estimated coefficients on MTB_t, DEBTEQ_t and RETTE_t are not significant.

In summary, the evidence provided in Table 6 and Table 7, taken together with the logistic regression results set forth in Table 8, provides support for H2. The results show that established firms suffer greater losses or earnings reductions, and this results in a higher likelihood of dividend reductions or omissions. Consequently, the importance of the magnitude of earnings in explaining dividend changes is higher the longer the historic earnings/dividend patterns precede the event year. Thus, losses or earnings reductions have more information content for reducing or omitting dividends for established than for less-established firms.

Robustness Tests: Information Content of More Severe Versus Less Severe Dividend Reductions in Explaining Future Earnings

Another robustness test entails examining the effect of dividend reductions on the predictive ability of earnings by distinguishing between less severe versus more severe dividend cuts. To this end, DIV_REDUCTION is modified as follows:

DIV_OMISSION= 1 if the firm omitted or reduced its annual dividend payment by more than 50% compared to previous year's level, and zero otherwise.

The rationale is that, given management reluctance to deviate from an ongoing commitment to pay regular dividends, complete omissions of dividend payments should constitute much more dramatic changes in corporate policy as opposed to dividend reductions. Evidence from the dividend literature is generally consistent with this reasoning (Allen and Michaely, 2003). For example, studies document firms suspending dividends experience significantly lower negative abnormal returns than do firms that announce dividend reductions (e.g., Charest, 1978; Healy and Palepu, 1988; Christie, 1990; Michaely et al., 1995; Benartzi et al., 1997). More importantly, Michaely et al. (1995) find firms omitting dividends continue to underperform the market for the next three years. Conversely, Benartzi et al. (1997) show that firms that reduce dividends exhibit a negative excess returns drift only for the first year following the dividend reduction event.²⁸ Thus, contrary to dividend omissions, dividend reductions are not correlated with significant long-run excess returns, and as Benartzi et al. (1997) argue, this is because dividend decreases are much less dramatic events than dividend omissions.

However, we make a distinction between slashing the annual dividend payment by more than or less than 50%, mainly because the literature supports the notion that, at least over the last three decades, the mean reduction in dividends for U.S.-listed firms has been approximately 50%. Namely, Grullon et al. (2002), studying dividend changes in firms listed on the NYSE and AMEX for the sample period 1967–93, report a mean dividend decrease of −44.8%. More recently, Chen et al. (2007) examine dividend changes for firms listed on NYSE, AMEX and NASDAQ for the sample period 1974–99 and document mean and median dividend decreases of −46.07% and −50%, respectively. With respect to the sample period 1986–2005, untabulated results reveal mean and median dividend decreases of −48.9% and −50%, respectively. Thus, the critical threshold of −50% seems to be appropriate for classifying dividend decreases as less or more ‘severe’.

Accordingly, in light of the results reported in Table 5 (and discussed above) on the information content of dividends for established versus less-established firms, dividend omissions (or reductions by more than 50%), coupled with earnings reductions, should act as a much more powerful signal on future performance, especially for established firms. This supposition is tested by employing regression model (1) but incorporating the DIV_OMISSION dummy variable as described above.

In line with this conjecture, the results reported in Table 9 show that ESTAB*ROA_t*DIV_OMISSION exhibits a much more negative and statistically significant coefficient compared to that reported in Table 5 (i.e., the estimated coefficient in Table 5 is −0.043 with p-value = 0.042, whereas the coefficient reported in Table 9 is −0.720 with p-value = 0.004). Thus, more severe dividend reductions entail higher information content, as dividend suspensions or significant reductions in dividends are more reliable signals that managers expect profitability problems to persist, compared to dividend reductions of smaller magnitude. Yet more importantly, results hereto document that beyond the ‘severe dividends reduction effect’, a dividend cut will send a more reliable signal about the persistence of earnings difficulties the more established are past earnings and dividend patterns.

Finally, since dividend omissions or reductions of more than 50% constitute a powerful signal regarding future firm performance, we conjecture that sustainable earnings of established firms should provide much greater information content in explaining dividend policy changes when distinguishing between less severe versus more severe dividend cuts. As a direct test of this view, Table 10 reports logistic regression results of the relation between the likelihood of severe dividend reductions and the magnitude of sustainable earnings for established vis-à-vis less-established firms, incorporating the same control variables as described in equation (2). Thus, the dependent variable is now DIV_OMISSION. Results in Table 10 show that the coefficients on ROA_t and on ESTAB*ROA_t are negative and statistically significant, but more importantly, are much greater in magnitude than the corresponding coefficients observed in Table 8.

In summary, the robustness tests provided in Tables 9 and 10 are consistent with our hypotheses. Logit analysis reveals that among earnings-reducing or loss firms, the magnitude of a firm's current earnings affects the likelihood of a dividend cut significantly more the longer the historic earnings/dividend patterns persist. This result, coupled with the stylized fact that managers’ reluctance to reduce dividends is higher the more established is the historic dividend policy, implies that dividend cuts are more likely to reflect persistent earnings difficulties the stronger the historic-patterns effect (Hypothesis 2). In other words, because management is less flexible to deviate from an implicit commitment to pay dividends the more consistently dividends have been distributed in the past, earnings difficulties need to be serious enough to warrant a reduction in regular dividend payouts. Thus, dividend cuts are a stronger signal that current low sustainable-earnings realizations will persist in the future for established versus less-established firms. In this respect, the logistic regression findings also provide corroborative evidence in favour of Hypothesis 1, supporting the notion that adverse dividend policy changes are more informative with respect to earnings prospects the longer the history of past earnings/dividends.