Who Cuts Dividends First? Theory and Evidence from Dividend Reductions

Transcription

1 Who Cuts Dividends First? Theory and Evidence from Dividend Reductions Tyler Hull * Abstract This paper examines dividend reduction timing at the industry level, asking what firm types choose to reduce their dividends earlier in a dividend reducing cycle than others. Theory is proposed that suggests that higher quality firms will reduce dividends sooner to start rebuilding the firm s profitability, while lower quality firms delay reducing their dividends until they must. This paper contributes to the literature by providing a new theoretical model that suggests dividend reduction timing can be an indicator of firm or management quality. The theory is taken to quarterly dividend paying firm level data where the proposed theoretical predictions are empirically tested. The results show support that early dividend reducers are higher quality firms in that they have higher future returns, a greater increase in total asset growth and higher levels of future profitability. * Tyler Hull is at the International Business School, Brandeis University, 415 South Street, MS 032, Waltham, MA tyler@brandeis.edu.

2 1 Introduction What firm types choose to cut their dividends earlier than later? In moments of financial difficulty patterns of dividend reducing cycles can become apparent. Investors assign a high importance to a reduction of a dividend, but do they also value the timing of the dividend reduction? When a firm is the first in its own industry to reduce dividends does that indicate poor quality, because they had to reduce dividends? Does it indicate high quality as they were the first firm to recognize the need to restrict cash outflows? This paper outlines a theory of dividend reduction timing which indicates firm quality. Namely, the theory purposes that firms with greater future potential will be the first to make unavoidable dividend reductions and so begin the process of restructuring the firm, while firms with poor future prospects will delay dividend reductions as long as possible as, wishing to put off an inevitable fall in their stock price. In the context of a dividend reduction cycle this suggests that when a firm reduces their dividends early on, they are more likely a higher quality firm than those who wait to reduce their dividend later in the cycle. In the literature dividend policy has often been seen as a way for firms to signal their own beliefs about their company s future prospects. This type of signal is important because naturally investors have less information then firm insiders and anything that could indicate a firm s future prospects will benefit the investor s decision. This signal is meaningful as dividends are a voluntary cash distribution and maintaining a cash outflow to stockholders is costly. The more costly it is to maintain the more it signals the quality and future expected profits of the firm. If the persistence of large cash dividends signals good firm profits then reductions or cuts in the level of dividend payments will signal trouble in the firm. Consistent with this, Dielman and 2

3 Openheimer, 1984 and Healy and Palepu 1988 find that dividend reductions lead to a negative market reaction. De Angelo and De Angelo (1990) show that firms are very hesitant to reduce dividends. This suggests that a dividend is usually cut out of necessity and firms have strong incentives to maintain dividends if possible. The DeAngelo, DeAngelo and Skinner (1992) paper shows that dividends reductions usually follow a period of poor operating performance. Moving in another direction, Lie (2004) notes that following dividend omissions firm earnings can be quick to recover. This indicates that despite dividend reductions certain firms can still have positive future expectations. Jensen, Lundstrum and Miller (2010) suggests that dividend reductions are a last resort effort and so are not signals of future outlook but eminent changes in the firm s future method of operating. Bulan and Subramanian (2008) in their paper present the idea of existence of good and bad dividend omissions. Firms are considered good if soon after their dividend omission they return to profitability and reinstate their dividend. The story behind a good firm is that they cut or reduce dividends as part of a retooling or self-improvement process and not necessarily due to poor future outlook. Bulan and Subramanian (2008) shows there are material differences at the time of a dividend omission between firms who later recover ( good firms) and those who do not ( bad firms). Another useful paper related to dividend cuts is Chemmanur and Tian (2007). Chemmanur and Tian (2007) introduce a theoretical model in which firms anticipate a possible future dividend cut, these firms then must choose to either prepare or not prepare the market for this future possibility. This model allows for firms of varying qualities to indicate quality types by their choice of market preparation. 3

4 In this paper I have built upon the model and theory presented by Chemmanur and Tian 1 (2007) to further address the idea of good and bad dividend reductions as seen in Bulan and Subramanian. 2 (2008). The new theory specifically looks at the firm s decision on the timing of dividend reductions as compared to other firms within a particular industry. The proposed model allows rational agents, who represent firms in financial difficulty, that decide on when to commit an inevitable dividend reduction. Consistent with the idea that the good firms reduce dividends to rebuild and restructure, the model predicts that these good firms will not wait until the last minute to lower their dividend. They will realize the need for change and will act quickly to bring the firm back to profitability. While, bad firms will not have incentives to cut and display that they are experiencing financial difficulties as they do not have ideas or good prospects of rebuilding the firm s profitability. These firms will have the incentive to maintain the dividend so as to keep up the false pretense that the firm is financially sound. This suggests that the first firms to reduce dividends in a particular industry will be firms with plans and direction and so be a higher quality firm. This differs from the Chemmanur and Tian (2007) model where all dividend cuts occur in the same period. In this paper I wish to address the timing of dividends within dividend cut cycles as an indicator of firm quality. The model proposes certain testable empirical predictions and the second half of this paper provides the first empirical tests of the outlined model. These test are conducted on US 1 My paper incorporates the Chemmanur and Tian (2007) framework of having multiple time periods and 3firm types. 2 The good and bad firm idea is the notion that certain firms will return to profitability sooner, thus being a good firm and that these firms can in some cases be identified at the time of the dividend reduction. Bulan and Subramanian (2008) identifies these firms by firm characteristics at the time of the dividend cut. 4

5 quarterly dividend paying firms and industries are determined by the 3 digit SIC code. The model predicts that those firms who reduce dividends earlier than other dividend reducers in their industry will have higher short and long term returns, larger increases in total assets, and greater profitability. These model predictions are tested empirically, with supportive empirical results. The model contributes to the literature as it provides a simple model which explains how the timing of a dividend cut can be a useful signal of management quality or future prospects to outside investors and additionally provides the first tests of the model. The paper will proceed as follows: In section 2 the model will be explained in greater detail. Section 3 will outline the model s main equilibrium. Section 4 contains the model s testable predictions. Section 5 addresses the data and empirical results of the model. Section 6 addresses other possible equilibria. Section 7 concludes. 2 Model The model focuses on a single unidentified industry. The model has two time periods; time 1, and 2. In the model we have agents who represent their firm, perhaps a manager and an outside investors. The firms are all mature dividend paying firms, which have a regular dividend that is generated from assets in place. Before time 1 a shock has hit this particular industry which creates three different types of firms: perfect, good, and bad. Perfect firms would be firms who were minimally affect by the industry shock. I assume these firms have had no change in their assets in place so the firm s profitability is not affected. While, good and bad firms suffer a shock which destroys their assets in place but leaves the firm with the cash to do one of the following: pay their usual dividend, keep the cash, or invest this money into the company to try 5

6 to restore the firm to its usually level of assets in place. If investment is successful, assets in place are restored allowing the usual dividend to be resumed. Investment can be immediately viewed by the market. Investments in the first period will not have a chance to restore assets in place until time 2, the final period. The only difference between good and bad types is that they have different probabilities of the investment returning the firm to its regular amount of assets in place. Good (bad) firms have the probability of G ( B ) that investment will be successful. Unsuccessful attempts at restoring assets in place will result in a total loss of the investment; with no assets in place and no cash the firm has no residual value. In the model I assume a risk-free rate of return of zero and that all agents are risk-neutral. Firms at the start of time 1 know which firm type they are. Before the end of each period firms must decide how to spend their cash on hand. They can pay it out as dividends, invest it in restoring their assets in place (investment is only for those who have lost assets in place), or they can hold on to the cash. Due to the time it takes for an investment to mature, no investments will be made after the first period. Outside investors do not know the state of the world nor individual firm types. Outsider investors have the following priors: there is X probability that a firm is a good firm, with Y probability a firm is a bad firm and with (1-X-Y) probability a firm is the perfect type. Outside investors are also aware of the values of G and B. 2.1 Firms Decisions: At each time period the manager makes a decision to cut or maintain the firm s dividend. If a manager cuts his dividend he can choose to invest the cut dividend or he can keep the cash in the firm. Retained cash can be forced out of the firm in the second period, when investment was not seen in the first period. The manager chooses his strategy to maximize: 6

7 Max over a and b η 1 E 1 [V 1 (a,b)] + η 2 E 1 [V 2 (a,b)] a,b η is the firms valuation or weight assigned to the firms value or stock price at that time. V is the value or stock price of the firm for different periods. Where a is the firm s action set, a Є Φ { C, NC}. C stands for cut or reducing the dividend to zero. NC stands for not cutting the dividend. And where b is the firm s action set, b Є Δ { I, NI}. I stands for invest in the firm to attempt to restore the firm to future profitability. NI stands for not invest or to keep the cash within the firm. In order for a firm to make a decision about investing or not a firm must have cut their dividend. 2.2 The timing of the game: Before time 1 nature moves deciding player type. As all firms have cash on hand before time 1 they can choose to either pay a dividend or not. If they do not pay a dividend and do not have assets in place they may choose to reinvest this money in the firm to try to rebuild assets in place. After the firm makes these decisions, the outsider is able to observe the firm s choice to cut dividends or not, and their investment decision. The outside investors then decide firm value or stock price based upon if a firm cut its dividend or not and future expectations. This firm value or stock price enters the firm s objective function. Investment in the first period will only potentially payoff in the second period and success will only be seen by the firm at the start of the second period. This delay in building time makes investment only potentially profitable in the first period. The same play is repeated for 7

8 one more period; where firms first decide dividend policy and then once it is paid out firm value is assigned by the outside investors. 3 Equilibrium Equilibrium strategies in this model are Efficient Bayesian Equilibrium (PBE), where the perfect and good types will engage in pure strategies and the bad type will engage in mixed strategies. 3.1 The perfect type: Due to their assets in place they can always payout a dividend. They have no incentive to ever cut their dividend, because if they ever cut their dividend they will start pooling with the firms who do not have assets in place. For these firms there is no possibility of losing assets in place for the game. The perfect type already has reached the optimal level of assets in place and so investment has no potential benefit as it is only a tool to create previously destroyed assets in place. 3.2 Good and bad type firms: The good and the bad firms can choose to pay the first period dividend or not. Because these firms only have enough cash to either pay the first period dividend or invest in the firm, every firm must make a dividend cut either in the first or second period. They can also choose to invest the money that they save or they can keep the money. Because investment takes one period to mature no firm will invest in the second period. All first period investment can be seen before period one s valuations are determined. This means if a firm cuts its dividend but does not invest the funds it saved, then at the end of the first period investors can see this and will demand 8

9 that these funds be paid out in the second period, the remaining firm is worthless. The good and bad firm types have three pure strategies. The first is where they do not cut today but cut tomorrow. The payoff at time one is: η 1 V 1NC + η 2 V 2C V 2C =0, {1} Where V 1NC stands for the expected firm value of 1 st period non cutting firm and V 2C is the value of a second period cutting firm. Here they can mimic firms who maintain the first period s dividends, specifically the perfect type. Next period their type is discovered as they have no more cash to maintain their dividend. As they made no investment and have no remaining cash the firm has no value (V 2C =0). The next pure strategy is to cut in the first period and then choose to not invest the cash in the firm. This strategy will identify the firm as worthless in the future with the only remaining value being the retained dividend that will be paid out in the second period. The payoff at time 1 is: η 1 V 1C/NI + η 2 (0) {2} Where V 1C/NI is the value of a 1 st period cutting firm that does not invest. This should be equal to the value of D 1, the first period dividend value, as the firm will be forced to payout the remaining cash in the firm before the second valuation period. The firm then has zero remaining value for the rest of the game. The third pure strategy is that of cutting the dividend and investing it in the firm, trying to rebuild assets in place. With probability H [H may be G or B], ( G for good, and B for bad, where we know that G> B ) the firm will successfully rebuild assets in place. The payoff is: 9

10 η 1 V 1C + η 2 [ H (V P )+ (1- H )(V F )] V F =0 {3} Where V P is the value of a second period perfect firm (once assets in place are rebuilt the firm becomes identical to perfect firms in value), and V F is the value of an investment failed firm in the second period which is zero as they have no cash and no assets in place. In this pure strategy the firm gives up the opportunity to mimic the perfect type to invest and try to rebuild their firm s assets in place. Using equations 2 and 3 we can see if: η 1 V 1C + η 2 [ H (V P )+ (1- H )(V F )] > η 1 V 1C/NI {4} Then the pure strategy of cutting in the first but not investing is strictly dominated by the strategy of cutting in the first period to invest. After a successful investment, firms can choose to payout a dividend or not to signal investment success. It is considered a dominant strategy to pay this last dividend so as to not be valued as a worthless firm. 3.3 The difference between good and bad firms The bad type has the same pure strategies as the good firm, but it has a lower probability of being able to invest successfully. If it is equilibrium for the good type not to invest then the bad type will clearly not invest either. The bad type can choose to mimic the perfect type for the first period or it can choose to pool with the good type in trying to rebuild assets in place. 3.4 Equilibrium with cycles As it is the goal of the paper to investigate cycles in dividend cuts I will only look at equilibriums where it is expected that some firms will cut in both the first and second periods. Proposition 1: An equilibrium with dividend cutting cycles, where: 10

11 α <1 3, η 1 V 1NC = η 1 V 1C + η 2 [ B (V P )] > η 1 V 1C/NI 4 The paper s main equilibrium is characterized by: (i) At time one the perfect firms do not cut their dividend, nor will they ever. The good firm type always cuts its dividend in the first period and the bad firm type mixes between cutting with probability of (α) and not cutting his dividends (1- α) in the first period. (ii) Good firms and bad firms who cut their first period dividend will invest in assets in place with a probability of 1. (iii) At time 2, investments results are realized. Successful rebuilding of assets in place allows the firm to resume normal dividend payments. Failed firms are worthless. All current firm values are discovered. What this does is force firms who wish to rebuild to pool with potentially bad early cutting firms if it decides to cut early. The perfect type will never cut their dividend as the cost of the dividend is much less than the cost of pooling with the types that do not have assets in place. 1 st period The good type will cut and invest in assets in place with a probability of 1. The bad type will mix with (α) probability of cutting in the first period to invest. It is assumed that dividends 3 This requires that α, the fraction of bad firms who cut dividends in the first period be less than one, thus requiring that dividend cuts occur in both time periods. 4 This requirement implies that the bad firm type finds it equally profitable to cut the dividend to invest in assets in place as it is to mimic the non-cutting firms for one period, leading to a mixing equilibrium. This also requires that these expected values both exceed the expected value of cutting the dividend and not investing in assets in place. 11

12 are such that there is no incentive for a firm to cut its dividend and not invest. (1- α) bad types with will not cut their dividends in the first period. This means all good types and (α) of the bad types cut in the first period and pool together. This leads to an expected immediate price drop. This leaves the perfect and (1- α) bad types not cutting and so pooling together. Outsiders are aware of the (α) value. Outsiders who see a first period dividend cut will use Bayes rule to update beliefs: PNC = 0 GC G = ( ) G B Y Y ( ) X BC ( ) B = ( ) G B ( ) X Y ( ) X BNC = 0 Notation: λ is the probability of being a certain type from the outside investor s perspective. PNC= perfect type who never cuts GC= good type who cuts in the first period BC=bad type who cuts in the first period BNC= bad type who does not cut in the first period G= good type B= bad type P= perfect type Based on the above information, outsiders value a cutting firm as: V = VP G ( GC ) ( 1C V P B BC ) Outsiders will see a dividend payment (no cut) will have the following beliefs: PNC P (1 X Y) = (1 ) (1 X Y) (1 ) X P B 12

13 GC = 0 BC = 0 BNC (1 ) B = (1 ) P B (1 ) X (1 X Y) (1 ) X Based on the above information, outsiders value a cutting firm as: V V 1NC 2C = V ( 0 P PNC ) V 2C ( BNC ) 2nd period In the second period the bad types who did not cut before must now cut as they do not have any funds to pay a dividend. This cut leads to them being exposed as a bad firm with no residual value, bad firms do this as they can pool with the perfect types for the first period. The good and the bad firms who invested will see if their investments paid off or not. At this time all information is known and each firm either has generated assets in place or not. Investment is a sunk cost and has either yielded assets in place or has no residual value. Firms who are restored to assets in place will payout a dividend before the end of the period to signal that investment was successful, with probability 1. No dividend paid in the second period by a 13

14 firm who made an investment indicates investment failure; the firm loses all value with the probability of 1. Proposition 2: In the equilibrium described above: (i) All Good firms will cut their dividends in the first period and only bad firms (firms with a lower likelihood of returning to profitability) will cut in the second period (ii) Dividend cuts either in the first or second period lead to a drop in the firm s stock price. Because all firms previously had assets in place a drop in price is expected. This price drop should be more severe for second period cuts than first period cuts. (iii) Earlier cutting firms should be increasing or rebuilding assets in place, while late firms will not. We can see that given the setup of the model, if the bad type mixes between cutting and not cutting then it is clearly advantageous for the good type with the higher probability of investment success, to cut their dividend and use it for investment in the first period. As it is required that (α) does not equal 1, this automatically creates the dividend cycle, meaning that not all firms cut simultaneously. This means that only bad firms will find it advantageous to cut in the second period due to their lower likelihood of investment success and being able to pool with perfect firms. The intuition for part (ii) is that as firms cut their dividends they are no longer pooling with the perfect type and firm valuation must decrease. The first period cutter s price drop will be a smaller price drop than those who cut in the second. This is because those cutting in the first have some opportunity for the firm value to be restored. Second period cutters have no opportunity for firm value recuperation. 14

15 The intuition for part (iii) is that 1 st period dividend reducing firms in lieu of paying out their dividend reinvest this money in the firm in an attempt to restore assets in place. On the other hand, firms that reduce dividends in the 2 nd period have paid out all of their cash in the form of a dividend in the first period and so they do not have the potential for assets in place to be rebuilt. 4 Testable predictions: 1. Good firms will cut dividends sooner than other firms, to start rebuilding the firm s value. This predicts that on average firms who reduce dividends in the beginning of an industry s dividend reduction cycle will ultimately have a higher future returns. 2. At the time of a dividend reduction there should be a more severe drop in abnormal returns for later reducing firms than for the early reducers, as this is when they indicate their firm type. 3. Early dividend reducers will be self-investing to rebuild or grow assets in place, while later reducers will not. This predicts higher future levels of total assets for early dividend reducers. 4. Firms who are recognized as the leaders of a dividend reduction cycle will regain profitability faster relative to firms who reduce dividends later in the cycle. 5 Data: I use CRSP dividend data for quarterly dividend paying firms, this data is further supplemented with accounting data from Compustat. To be consistent with the literature utilities and financial firms are excluded from the analysis. An industry is determined by their 3-digit SIC code. The model shows investors reacting to firms eliminating dividend payments or omitting a 15

16 dividend. Dividend omissions are both much less frequent and more difficult to identify then dividend reductions. For these reasons the empirical tests will be performed on economically meaningful dividend reductions instead of omissions. A dividend reduction is said to have occurred when the current period s dividend is less than the period s before. To ensure that we are only looking at economically meaningful dividend reductions, we only count a reduction as a cut if it is a reduction of 12.5% or more. We further limit the classification of dividend cuts by only including cuts by firms who do not have any additional dividend cuts within one year of the initial cut. This is to insure the focus of the analysis is on the main or final dividend cut in the event of multiple dividend cuts within the same year. Furthermore, Using a dummy variable which is one if a reduction has occurred and zero if a reduction has not, I find the percentage of firms who have lowered their dividend at any given quarter by industry. Industries that have less than 10 firms in the sample are dropped, to minimize the effect of any individual firm. The data is also further limited to quarters where 10 firms per industry were available. This sectioning allows us to see industry level dividend reduction cycles, specifically showing what percentage of the industry has reduced their dividend in any given calendar quarter. Then by industry the sample s mean percentage of dividend reductions per quarter is found, this value is used as a dividing line between high and low dividend cutting periods 5. All dividend cuts are then classified as either early or late dividend cuts. If a firm reduces their dividend following a quarter of high or above average percentage of firms reducing their dividends then they are considered to be a late dividend cutting firm. On the other hand, if a 5 Many dividend cuts occur following quarters of no industry dividend reductions and most results hold if the dividing line between high and low cutting periods were to be zero versus positive cutting levels. 16

17 firm lowers their dividend a quarter after a low or below average level of industry wide dividend cuts then the firm is categorized as an early dividend reducer. There are other possible dividing lines between early and late dividend reducers that could be implemented. The chosen method is used for its simplicity and how it matches the proposed theoretical model s goal of clear periods of earlier and later dividend reducers. Although the theoretical model suggests that dividend reduction timing should be measured from a particular industry specific shock, here it is measured from the first dividend reductions. This departure from the theory is necessary as the theory does not provide an indication of the duration or length of the different time periods, which may also be industry specific. By measuring dividend timing relative to the percentage of firms reducing dividends in a particular industry, we can re-create the clear distinction between early and late dividend reducers as seen in the model. Dividend reduction timing could also be measured from the date of particular events or NBER classified recession, but this methodology considerably limits the sample size. The sample is further limited to early and late dividend reductions that have complete data for all the control variables and return data ranging from the days around the dividend reduction announcement to up to 3 years after the dividend reduction date. The final sample ranges from with a total of 811 early and late dividend reducers, table 1 provides the observation s yearly distribution by dividend reducing type. Table 2 provides some summary statistics and tests of equality between the early and late dividend reducing groups one quarter before the date of the dividend reduction, no significance difference can be seen between the groups at the 20% level. With the early and late groups now clearly defined the model s predictions can be tested on the two groups. 17

18 5.1 Predictions and results: Below each of the model s predictions is named and empirical tested for the group of 811 dividend reducers. For the following results the main variable of interest is a dummy variable which is equal to one if the dividend reduction occurred early relative to other firms in their industry and zero if the dividend reduction occurred relatively late. 5.2 Prediction 1 Good firms will cut dividends sooner than other firms, to start rebuilding the firm s value. This predicts that on average firms who reduce dividends in the beginning of an industry s dividend reduction cycle will ultimately have a higher future returns. To test this prediction OLS regressions with fixed effects for the year of the dividend reduction and the 3 digit SIC code were performed. The dependent variables are cumulative abnormal returns (CARs) which are calculated as monthly abnormal returns using the Fama- French factor model. The CARs cover the following time periods as measured from the quarter of the dividend reduction: 1 quarter, 2 quarters, 3 quarters, 1 year, 2 years and 3 years. Industry adjusted control variables are used to isolate the importance of the early dummy variable. As an additional control the percentage of the dividend reduction also enters the regression. The sample for this test only consists of observations that are considered to be early or late dividend reducers. As the early dummy variable is a dummy variable the economic interpretation of the coefficient is simple, this is how much additional CAR is expected for the early type over the CAR of the late type. 18

19 Table 3 shows that 1 to 2 quarters from the date of the dividend reduction the early dummy variable has a positive but insignificant coefficient. As of 3 quarters from the date of the dividend reduction this changes to a positive coefficient of 4.9% of additional return associated with early dummy variable which is significant at the 5% level. By one year from the date of the dividend reduction the coefficient has grown to 6.3%, which is significant at the 5% level. At time 2 years this has become 12.2%, significant at the 1% and at time 3 years there is 13.0% extra return associated with the early dummy, significant at the 1% level. Figure 1 provides a graphical representation of the spread between early and late dividend reducers for the above date ranges. These results show that following the dividend reduction early and late dividend cutters cannot be told apart as for CARs 1 to 2 quarters from the reduction date, but as time passes the difference between them grows, which could have possible implications for trading strategies. The last item this table points to is that there is also a positive CAR for firms with a greater ratio of cash to total assets. This can be seen for the CAR time ranges of 3 quarters and beyond, the same time ranges where we see significance with the early dummy. In predicting the CAR the cash to total assets variable has the greatest coefficient in the 3 year specification, in this regression an inter-quartile increase of the industry adjusted measure of the cash to total asset ratio results in a higher CAR of 2.11%. The cash to total asset variable was further tested and interacted with the early dummy to see if this effect existed in only one or both of the early and late groups, yielding no significant difference in relation to their sensitivity to cash levels between these two groups. These results tell the story that even among early and late dividend reducers firms can be further distinguished by their cash to total assets ratios. Higher cash to total assets ratios before a dividend reduction could be another proxy for firm quality, although the 19

20 theory proposed in this paper is limited in that all good firm types reduce their dividends in the first period; the theory does suggest that those with high cash levels surrounding the dividend reduction will be higher quality firms. 5.3 Prediction 2 At the time of a dividend reduction there should be a more severe drop in return for later reducing firms than for the early reducers, as this is when they indicate their firm type. To test this prediction the same specification and similar controls as the first prediction s test were used, but now for different time frames. Now the dependent variables are cumulative abnormal returns (CAR) for 1, 2, 4 and 7 day windows surrounding the announcement date of the dividend reduction. All control variables are from one quarter prior to the dividend reduction announcement date. The results for this test are found in table 4. As predicted we see a positive coefficient for the early dummy variable. This coefficient is significant at the 10% level for the CAR windows from 3 days before to the day of the announcement (-3:0) and for just day of the announcement (0). The day before to the day of the announcement (-1:0) there is a positive coefficient significant at the 5% level. For the other 3 windows the early dummy has positive coefficients but they are not significant at the 10% level. The coefficient for the day of the announcement date suggests that early dividend reducers have an abnormal return that is 37 basis points higher than late dividend reducer, supporting the model s prediction that at least on the day of the dividend announcement there is an significantly lower abnormal return for late dividend reducers. 5.4 Prediction 3 20

21 Early dividend reducers will be self-investing to rebuild or grow assets in place, while later reducers will not. This predicts higher future levels of total assets for early dividend reducers. To test this, the same framework for the above tests will be utilized but now with a new type of dependent variables. The dependent variables are the percentage of total asset growth as measured from the quarter preceding the dividend reduction to 1 quarter, 1 year, 2 years, or 3 years following the dividend reduction. Results for this test are shown in table 5. For these tests, the early dummy is positive significant at the 10% level for 1 quarter and 2 year total asset growth. Once again as this is a dummy variable it is easy to explain the significance of the early dummy s coefficient; early dividend reducers grow in total assets relative to late dividend reducers, 2.5% within the 1 st quarter, and 7.5% for the first two years following the dividend reduction. Early dummy coefficients for one and three years are not significant at the 10%, but they do have the correct sign. These results indicate some viewable differences between early and late groups both shortly after the dividend reduction and a few years after. As significant differences can be seen, this supports the notion that those reducing their dividends early are doing so for the purpose of rebuilding or increasing revenue generating assets or assets in place. 5.5 Prediction 4 Firms who are recognized as the leaders of a dividend cut cycle will regain profitability faster relative to firms who cut dividends later in the cycle. To test this prediction of the model, the same variables employed previously will be used to predict percentage change in a well know measure of profitability, Tobin s Q. The percentage 21

22 change in Tobin s Q is used to measure the rate of profitability improvement. Improvement is measured from the year prior to the dividend reduction, which should be a less volatile time period for this variable, until one to two years following the dividend reduction in quarterly intervals. These time periods were chosen as there is an expected change in profitability, but this change would not happened immediately following the dividend reduction. Furthermore, it is unlikely that this will have a permanent affect suggesting this result may not be apparent in the longer time horizons. The results for this test are shown in table 6. The early dummy in these regressions suggests that early dividend reducers have 4.1%, 5.0%, and 5.8% higher increases in Tobin s Q over later dividend reducers for the time horizons of 1 year, 1 year and 2 quarters, and 1 year and 3 quarters following the dividend reduction. These results are all significant at the 10% level. Time horizons less than one year and greater than two years were also tested (not included in the paper), the early coefficient for these time horizons tends to be positive but not significant at the 10% level. These tests show some support of greater increases in the profitability measure Tobin s Q, for early dividend reducers. 6 Other equilibrium The main equilibrium that is explained and tested is not the only equilibrium that fulfills the criteria of having dividends cuts which happen at different time periods. There can be a non-pooling equilibrium where (α)=0, η 1 V 1C + η 2 [ G (V P )] > η 1 V 1NC > η 1 V 1C + η 2 [ B(V P )] > η 1 V 1C/NI 22

23 This is a fully separating equilibrium. Here the bad type will always delay the dividend cut to the second period and never invest. The good type will cut in the first period to invest in assets in place. Another pooling equilibrium could be where: (α g )<1 where α g is the fraction of good firms who cut in the 1 st period η 1 V 1C + η 2 [ G (V P )] = η 1 V 1NC > η 1 V 1C + η 2 [ B(V P )] > η 1 V 1C/NI In this scenario, the good type will be indifferent between investing or not. This clearly means that the bad type will not cut its dividend in the first period, as he has a lower probability of success. Both of these other equilibrium specifications actually give similar model results, where those of higher quality reduce dividends sooner than other dividend reducers. In line with this fact the 4 main model predictions remain valid for these equilibria. 7 Conclusion This paper presents a clear simple model which answers the question who cuts dividends first in a dividend reduction cycle? The first empirical tests of this model are also included, showing support of the model and its predictions. In the elaborated equilibrium it is shown that better quality firms will be among the first to reduce dividends in a dividend reduction cycle. This early dividend reduction is due to the firm wishing to begin rebuilding profitability before hitting rock bottom. Firms who have less opportunities or ideas on how to restore profitability will wait until the last moment to reduce their dividend, so that in the 23

24 meantime they can pool with firms expected to not lower their dividend. In addition to this, support was shown that early dividend reducers have better short and long-term returns, which if studied in greater detail could potentially be developed into a possible trading strategy. Lastly, support was given that early dividend reducers both grow more in terms of total assets and operating performance measures. 24

25 Works Cited Bulan, Laarni, and Narayanan Subramanian, 2008, A Closer Look at Dividend Omissions: Payout Policy, Investment and Financial Flexibility? working paper, Brandeis University. Chemmanur, Thomas J., and Xuan Tian, 2007, Preparing the Equity Market for corporate Events: Theory and Evidence from Firms Cutting Dividends, Second Singapore International Conference on Finance, working paper series. DeAngelo, Harry, and Linda DeAngelo, 1990, Dividend Policy and Financial Distress: An Empirical Investigation of Troubled NYSE Firms, Journal of Finance 45, DeAngelo, Harry, Linda DeAngelo, and Douglas J. Skinner, 1992, Dividends and Losses, Journal of Finance 47(5), Dielman, T. E. and H. R. Openheimer, 1984, An examination of investor behavior during periods of large dividend changes, Journal of Financial and Quantitative Analysis 19(20), Healy, Paul M., and Krishna G. Palepu, 1988, Earnings information conveyed by dividend initiations and omissions, Journal of Financial Economics 21, Jensen, Gerald R., Leonard L. Lundstrum, Robert E. Miller, 2010, What do Dividend Reductions Signal? Journal of Corporate Finance (Accepted). Lie, Erik, 2005, Operating Performance Following Dividend Decreases and Omissions, Journal of Corporate Finance 12,

26 26

27 Table 1: Year Distribution by Early and Late Dividend Reducers This table reports the distribution of dividend reductions of more than 12.5%, tabulated on a per year basis for regular quarterly dividend paying firms. The frequencies and respective percentages are tabulated separately for early and late dividend reducers. Early dividend cutters are categorized as firms who reduced their dividend following a quarter with a low percentage of industry dividend reducers. Late dividend reducers cut dividends following a quarter with a high percentage of industry dividend reducers. An industry is at a high dividend reducing period if the current percentage is more than the industry sample s average number of dividend reductions. Early Reducers Late Reducers Year Frequency Percentage Frequency Percentage Total

28 Table 2: Summary Statistics This table reports some summary statistics for early and late dividend reducers and simple tests of equality for the two group s means. All differences below are insignificant even at the 20% level. This table provides support that the early and late groups do not appear to have any meaningful differences between them prior to the dividend cut event. Early dividend cutters are categorized as firms who reduced their dividend following a quarter with a low percentage of industry dividend reducers. Late dividend reducers cut dividends following a quarter with a high percentage of industry dividend reducers. An industry is at a high dividend reducing period if the current percentage is more than the industry sample s average number of dividend reductions. Summary statistic variables: Dividend % change, this is the percentage of the dividend decrease, values are all negative; all other variables are taken the quarter before the dividend reduction. Early Reducers Late Reducers Difference Dividend % change Mean Total assets Mean Debt to cash flow ratio Mean Leverage Mean Cash Mean Cash to total assets ratio Mean Market to book ratio Mean Book equity Mean Total capital Mean Stock turnover Mean Tobin s Q 6 Mean Observations This variable has 20 missing observations, with 452 early dividend reducers and 339 late dividend reducers 28

29 Table 3: Long Run Abnormal Returns This table reports the results of several OLS regressions with differing event windows as the dependent variable. Abnormal monthly returns or alphas are calculated using the Fama-French factors. Cumulative abnormal returns (CARs) are calculated for different windows of time, spanning from the month of the dividend reduction up to 3 years following. The independent variables are, Early dummy, a dummy variable that is 1 if this particular dividend reduction occurred following a quarter where the industry it belongs to had a lower than average percentage of dividend reductions, 0 otherwise, Dividend % change, this is the percentage of the dividend decrease, values are all negative; all other variables are taken the quarter before the dividend reduction. I.A. denotes that instead of the actual level of this variable being inserted an industry adjusted level of this variable was used instead. Industry adjusted variables are simply the observation s value minus the median of this variable by 3 digit SIC code. Fixed effects dummies are included for the year of the dividend reduction and the 3 digit SIC code. Heteroskedasticity corrected robust p-vlaues, which are clustered by the 3 digit SIC code, are in brackets. The regression is estimated with an intercept term. ***, **, and * represent statistical significance at the 1, 5, and 10 percent levels, respectively. VARIABLES (1) (2) (3) (4) (5) (6) 2 Quarters 3 Quarters 1 Year 2 Years CAR CAR CAR CAR 1 Quarter CAR 3 Years CAR Early dummy ** 0.063** 0.122*** 0.130*** [0.411] [0.483] [0.037] [0.021] [0.000] [0.000] Dividend % change [0.122] [0.185] [0.678] [0.784] [0.269] [0.309] Log of total assets I.A [0.767] [0.961] [0.305] [0.418] [0.225] [0.561] Debt to cash flow ratio I.A * [0.770] [0.999] [0.606] [0.462] [0.083] [0.189] Leverage ratio I.A ** 0.164** 0.189** [0.349] [0.038] [0.026] [0.014] [0.233] [0.104] Cash to total assets ratio I.A * 0.348* 0.459** [0.226] [0.170] [0.122] [0.094] [0.094] [0.035] Market to book ratio I.A [0.393] [0.479] [0.903] [0.928] [0.504] [0.581] Observations R-squared Adjusted R-squared

30 Table 4: Short Run Abnormal Returns This table reports the results of several OLS regressions with differing event windows as the dependent variable. Abnormal daily returns or alphas are calculated using the Fama-French factors. Cumulative abnormal returns (CARs) are calculated for different windows of time, where 0 denotes the day of the dividend reduction announcement, negative numbers denote days before the announcement and positive numbers denote the number of days following the announcement. AR 0 denotes the abnormal return or alpha for the day of the dividend reduction announcement. The independent variables are, Early dummy, a dummy variable that is 1 if this particular dividend reduction occurred following a quarter where the industry it belongs to had a lower than average percentage of dividend reductions, 0 otherwise, Dividend % change, this is the percentage of the dividend decrease, values are all negative; all other variables are taken the quarter before the dividend reduction announcement. I.A. denotes that instead of the actual level of this variable being inserted an industry adjusted level of this variable was used instead. Industry adjusted variables are simply the observation s value minus the median of this variable by 3 digit SIC code. Fixed effects dummies are included for the year of the dividend reduction and the 3 digit SIC code. Heteroskedasticity corrected robust p-values, which are clustered by the 3 digit SIC code, are in brackets. The regression is estimated with an intercept term. ***, **, and * represent statistical significance at the 1, 5, and 10 percent levels, respectively. (1) (2) (3) (4) (5) (6) VARIABLES CAR -3:0 CAR -1:0 AR 0 CAR 0:1 CAR 0:3 CAR -3:3 Early dummy 0.007* 0.006** 0.004* [0.099] [0.038] [0.091] [0.479] [0.661] [0.367] Dividend % change 0.038* * 0.042** 0.064** [0.054] [0.144] [0.190] [0.095] [0.039] [0.011] Log of total assets I.A * 0.002* ** [0.050] [0.069] [0.393] [0.883] [0.331] [0.042] Debt to cash flow ratio I.A ** *** *** [0.507] [0.761] [0.047] [0.001] [0.009] [0.653] Leverage ratio I.A * 0.027** [0.580] [0.989] [0.051] [0.040] [0.146] [0.981] Cash to total assets ratio I.A ** 0.044** [0.992] [0.395] [0.177] [0.011] [0.038] [0.362] Market to book ratio I.A * 0.003* [0.998] [0.935] [0.381] [0.173] [0.064] [0.097] Observations R-squared Adjusted R-squared

31 Table 5: Long Run Total Asset Growth This table reports the results of several OLS regressions with total asset growth as the independent variables for four different time spans. Total asset growth is measured as the percentage change in total assets from the quarter prior to the dividend reduction to a quarter, a year, 2 years or 3 years after the dividend reduction. The independent variables are, Early dummy, a dummy variable that is 1 if this particular dividend reduction occurred following a quarter where the industry it belongs to had a lower than average percentage of dividend reductions, 0 otherwise, Dividend % change, this is the percentage of the dividend decrease, values are all negative; all other variables are taken the quarter before the dividend reduction. I.A. denotes that instead of the actual level of this variable being inserted an industry adjusted level of this variable was used instead. Industry adjusted variables are simply the observation s value minus the median of this variable by 3 digit SIC code. Fixed effects dummies are included for the year of the dividend reduction and the 3 digit SIC code. Heteroskedasticity corrected robust p-values, which are clustered by the 3 digit SIC code, are in brackets. The regression is estimated with an intercept term. ***, **, and * represent statistical significance at the 1, 5, and 10 percent levels, respectively. VARIABLES (1) (2) (3) (4) One year Two year total asset total asset growth growth One quarter total asset growth Three year total asset growth Early dummy 0.025* * [0.058] [0.405] [0.077] [0.148] Dividend % change [0.780] [0.950] [0.824] [0.913] Log of total assets I.A * [0.264] [0.065] [0.192] [0.151] Debt to cash flow ratio I.A [0.412] [0.360] [0.302] [0.268] Leverage ratio I.A [0.150] [0.551] [0.447] [0.446] Cash to total assets ratio I.A [0.235] [0.681] [0.661] [0.607] Market to book ratio I.A [0.144] [0.220] [0.264] [0.301] Observations R-squared Adjusted R-squared