Gazdasági Ismeretek | Befektetés, Tőzsde » Chaudhry-Boldin - Dividend Policy and Earnings, A Study of Short and Long-Term Causality

Source: http://www.doksinet Dividend Policy and Earnings: A Study of Short- and Long-Term Causality Dr. Mukesh K Chaudhry Department of Finance and Legal Studies Indiana University of Pennsylvania Indiana, PA 15705 (USA) E-Mail: chaudhry@iup.edu *Dr. Robert J Boldin Department of Finance and Legal Studies Indiana University of Pennsylvania Indiana, PA 15705 (USA) E-Mail: rboldin@iup.edu *Contact person Keywords: dividend policy, earnings per share, cointegration, causality Dividend Policy and Earnings: A Study of Short- and Long-Term Causality Source: http://www.doksinet Abstract This study examines whether the earnings per share (EPS) and the dividend per share (DPS) exhibit a long-run equilibrium relationship. The data employed in this study consist of monthly EPS and DPS for 28 of the DJIA companies obtained from Bloomberg over the past 10 years. The companies under investigation have the EPS and the DPS available over the period studied. Dividends are generally paid out

of earnings. The amount and timing of the dividend paid is a function of the respective company’s dividend policy. Therefore, the EPS t can be expressed in terms of the DPS t as follows: EPS t = αDPS t where α is a non-negative constant. The equation suggests that there is a linear relationship between the EPS t and the DPS t . I. Introduction Source: http://www.doksinet This research examines the short- and long-term relationship between dividends and earnings per share across large companies from different industries. Dividend policy, when implemented by publicly listed corporations, constitutes a long-term managerial commitment to shareholders (Faccio, Lang & Young, 2001). While dividends are typically paid from earnings, the amount provided could come from any figure between net income after taxes but before extraordinary items to comprehensive income. Cash dividends can be paid from past earnings as long as there is a sufficient amount in the retained earnings account

and adequate cash in the cash accounts. For many shareholders, their quarterly dividend check is used as their primary source of income. For this reason, management often aims to provide a smoother dividend payment stream than that delivered by the pattern of earnings. This is evidenced by the fact that, even in years of a loss, corporations still pay a dividend that is similar to that paid in the prior profitable year. Market reactions surrounding dividend changes are well documented. For example, Koch and Sun (2004) found that when a dividend change is preceded by an earnings change of the same direction, the market reaction will be in the same direction. The effect is apparent within approximately three days of the dividend announcement. Charitou, Lambertides and Theodoulou (2010) also investigated market reaction to dividend announcements. Their investigation, however, focused on firms that had a loss. In their research design, they segmented their sample into two cohorts;

established firms and less-established firms. They found that market reaction is significantly more negative for established firms that declare dividend reductions after record earnings and dividend payments than for less-established firms. DeAngelo, Deangelo, and Skinner (1992) documented that established firms that incur a first-time loss but do not reduce dividends, manage to recover to positive profits. In contrast, Source: http://www.doksinet established firms that reduced dividends continued to incur losses. These findings indicate that dividend policy might provide unofficial insight into the expected earnings pattern of the firm. That is, if the dividend payment is not reduced in a loss year, then management is reasonably certain that the firm will return to profit in the subsequent year. This may be due to new management, which adopts the “big bath” approach. Significant write-downs and write-offs are implemented so that the new management team can start with a clean

slate. In contrast, if the loss year is accompanied by a reduction in the dividend amount, this might indicate that the loss in earnings is probably a longer-term problem. There is supporting evidence showing that significant changes to the dividend paid appears to be driven by changes in permanent earnings (Hsu, Wang, & Wu, 1998). Stock prices tend to reflect market expectations of the future direction of earnings (Marsh & Merton, 1987). Using stock price as a proxy for the firm’s intrinsic value, Marsh and Merton found that dividend changes are significantly related to past stock price. Thus, managers are reluctant to reduce dividends knowing that a reduction will most likely lead to a drop in the market price of the stock. They are especially unwilling to reduce the dividend amount if they know in advance that the earnings reduction is most likely temporary. Charitou, Lambertides, and Theodoulou (2011) show that dividend reduction announcements by established firms are

accompanied by larger negative market reactions than for similar announcements by less-established firms. This finding indicates that investors do take notice of dividend directions and show their viewpoint through the change in the market price of the stock. Nissim and Ziv (2001) focused on “the information content of dividends hypothesis.” Using a modified regression model, they found that in each of the two years after the dividend change, there was a positive association with earnings changes. To some extent, dividend Source: http://www.doksinet announcements could provide a more useful indicator of management’s view as to the future earnings of the company than the earnings figure itself. This notion is supported by Healy and Palepu’s (1988) finding that dividend changes signal a manager’s private information about future earnings changes. They found that investors regard extreme changes in dividends as indicators of management’s forecasts of future earnings change.

As an aside, between 1978 and 1999, the percentage of US industrial firms paying dividends has declined from 66.5% to 208% (see Fama & French, 2001) It is worthy to note that two stocks that received the most publicity over the past 10 years were Apple and Google; they do not pay dividends. Any return that the shareholders of those companies earn is derived from selling their stock at a gain. This paper builds on prior research by hypothesizing that a change in dividend policy signals to the market that management is expecting earnings to continue in the same direction for the long-term. Several issues are addressed: First, what is regarded as a dividend policy change? For example, is there a percentage benchmark (say, 10%) that the market regards as significant? Second, how many years constitute the long-term? Is it three, five, seven or ten years? General Motors continued to pay dividends for a number of years despite long-term losses which eventually led to its filing for

bankruptcy. Large US banks such as Bank of America and Citibank have been in similar positions since 2008. How has the market reacted to those losses and the reduction in dividends? This research addresses the above questions by investigating corporations listed in the Dow Jones Industrial Average (DOW) index that have paid dividends from 1992 to 2011. The paper is organized as follows: Part II explains the design of the empirical tests and sample selection. Part III outlines the results using descriptive statistics The last part presents the conclusions. Source: http://www.doksinet II. Research Design and Data Since dividends are generally paid out of current earnings, the two are related. The dividend payment, however, depends on the dividend policy of the firm. Hence, the two variables are related as follows: EPS t = α DPS t The value of α is dependent on the dividend policy of the firm. But, to determine the long-term dividend policy followed by the firms, the time series

behavior of different companies that comprise the DOW were examined. Most time-series are nonstationary and the use of cointegrated methodologies account for this characteristic.1 Engle and Granger (1987) suggest that if a system of variables is cointegrated, then economic forces interact to bind these variables together in a long-run equilibrium relationship. They suggest that an error correction model (ECM) can represent the cointegrated variables.2 In general, the ECM shows the dependence of this period’s price change on the last period’s price change, thus providing a measure of how far the system is out of its long-run equilibrium. There are preliminaries a researcher has to observe before applying the methods of cointegration. Specifically, before testing for cointegration between two or more series, it is necessary to test whether the different time series are integrated to the same order.3 This is done by applying conventional unit roots tests, described below. Stationary

(Unit Root) Tests for Individual Time Series Source: http://www.doksinet In general, most texts on stationarity of a time series (TS t ) will probably begin with an estimation of the following regression equation, if no linear trend is considered. ρ ∆TS =α + α TS + ∑ γ ∆TS +ε j t− j t 0 1 t −1 t j =1 (1) and by equation (2) when linear trend and a parameter for drift are considered: ρ ∆TS =α + α TS + α t + ∑ γ ∆TS +ε j t− j t 0 1 t −1 2 t j =1 (2) where ∆ represents differences (first difference unless otherwise noted), α 0 represents the term for drift in the series, α 1 allows testing for a unit root, and α 2 verifies the presence of a trend. The error-correcting mechanism is represented by ∆TS t− j in the model. If the hypothesis α 1 = 0 cannot be rejected, then the series is said to have a unit root and is nonstationary. Conversely, if the hypothesis, α 1 = 0 , is rejected, it is concluded that the series does not contain a

unit root and is stationary. Tests involving parameters α 0 and α 2 verify the presence of drift and trend, respectively. Inclusion of the p lagged values ensures a white-noise series The number of lags is determined by a test of significances, such as the Akaike information criterion (AIC) (Akaike, 1973).5 Importantly, the distribution of the ordinary t and F statistics computed for the regressions do not have their expected distribution. Thus, in order to test various hypotheses, critical values have been computed using Monte Carlo techniques and are tabulated in various references (see, e.g, Davidson and MacKinnon, 1993) Tests for stationarity and Source: http://www.doksinet cointegration use the Philips-Perron (P&P) non-parametric testing procedure. The P&P procedure is used since the crucial iid error assumption is not needed.6 Integration/Segmentation Tests For these tests, the methodology developed by Johansen (1988) was used. This method is preferred to

alternatives since it enables testing for the presence of more than one cointegrating vector. The description that follows draws from Johansen (1988, 1991, 1994) and Johansen and Juselius (1990, 1991). The Johansen method provides some distinct advantages. For example, identification of the number of cointegrating vectors is possible with the Johansen test. Such inferences are based on the number of significant eigenvalues. Also, many argue that the statistical properties and power for Johansen’s test are generally higher than for alternative procedures. To check for stationarity arising from a linear combination of variables, the following AR representation for a vector VTS made up of n variables is used, k s −1 VTS = c + ∑ φ Q + ∑ π VTS +ε i it i t −i t t i =1 i =1 (3) where VTS is at most I(1), Q it are seasonal dummies (i.e, a vector of non-stochastic variables) and c is a constant. It is not necessary that all variables that make up VTS be I(1) To find

cointegration in the system, only two variables in the system need be I(1). If only two time Source: http://www.doksinet series are examined (bivariate representation), however, then both have to be I(1). If an errorcorrection term is appended, then: s −1 k −1 ∆VTSt= c + ∑ φiQit + ∑ Γ I ∆VTSt −i +ΠVTSt −k +ε t =i 1 =i 1 (4) which is basically a vector representation of equation (1) with seasonal dummies added. All long-run information is contained in the levels terms, ΠVTS t − k , and short-run information in the differences ∆VTS t −i . The above equation would have the same degree of integration on both sides only if Π = 0 (the series are not cointegrated) or ΠVTS t − k is (0), which infers cointegration. In order to test for cointegration, the validity of H 1 (r), shown below, is tested as: H ( r ) : ∏ =γβ ′ 1 (5) where β is a matrix of cointegrating vectors and γ represents a matrix of error correction coefficients. The hypothesis H 1

(r) implies that the process ∆VTS t is stationary, VTS t is nonstationary, and β ′ VTS t is stationary (Johansen, 1991). The Johansen method yields the Trace and the λ max statistics that enable determination of the number of cointegrating vectors. Description of Data Quarterly data for earnings per share (EPS) and dividends per share (DPS) were obtained from Bloomberg for the DOW 30 stocks. The data for two companies could not be utilized due Source: http://www.doksinet to lack of data. Hence, the remaining 28 companies were used in the analysis The sample spans the period March 31, 1992 until September 30, 2011. One important issue regarding the data requires some discussion. The use of quarterly data, and the lesser frequency of observations versus the use of weekly or daily data, may create misgivings about the power of unit-root and cointegration tests against an alternative hypothesis, such as a trend stationary model. Hakkio and Rush (1991), from their Monte Carlo

simulations, however, showed that the frequency of information plays a minor role; rather, it is the length of data series or span, which is more important in discerning whether the time series are cointegrated or not. This finding is also supported by Shiller and Perron (1985) and Perron (1989) who found that changing the frequency of observations, while keeping the sample length fixed, is not helpful when testing for cointegration because it is mainly a long-run phenomenon. III. Results The list of companies included in our analysis along with their ticker symbol and SIC Code have been provided in the Appendix. Descriptive statistics for dividend payout ratio of the firms in our sample are shown in Table I. From the mean values it can be seen that most of the firms displayed in Table I that firms pay a significant amount of dividends from earnings. For instance, mean value of dividend payout ratio ranges from 75% for Pfizer Inc. (PFE) to 19% for International Business Machines (IBM).

The period under examination was very challenging for these companies as they faced a recession in 2000/2001 and followed by the financial crisis in 2008 in conjunction with the European crisis which in-turn significantly impacted their earnings. Despite the negative conditions, these firms managed to pay consistently high dividends. From the descriptive statistics it is evident that companies such as, General Electric (GE), Johnson and Johnson (JNJ), J.P Morgan Chase and Company (JPM), Kraft Foods Inc (KFT), 3 M Company Source: http://www.doksinet (MMM), Merck and Company (MRK), Pfizer Inc. (PFE), Procter and Gamble Co (PG), AT &T (T), Verizon (VZ), and Exxon Mobil (XOM) repeatedly paid very high dividends from their earnings over long periods. Even when these firms suffered losses or had low earnings, they still paid dividends. This can be seen from the maximum and minimum value of the dividend payout ratio which clearly indicates that almost all firms in the sample continue

to pay dividends even when they have low earnings or suffer losses for some of the time periods in the sample. This is indicative of the fact that the dividend policy decisions by the firms is a long-term criteria and has only a minimal effect on the basis of short-term fluctuation in earnings. To obtain a deeper understanding of the dividend policy of the firms, the time series behavior of both dividends and earnings per share were examined. In order to eliminate autocorrelations in the time-series for both EPS and DPS, the appropriate lag length was found using the Akaike information criterion (AIC). The lag length is selected by minimizing the AIC over different choices for the length of the lag. The values of AIC are formulated by computing the value of the equation T log (RSS) +2 K, where K is the number of regressors, T is the number of observations and RSS is the residual sum of squares. Tests for Stationarity of Each Time Series Using the ADF and P&P Test The tests of

stationarity are shown in Table II using both Augmented Dickey Fuller (ADF) and Philips-Perron (P&P) Test. The time series were tested for a unit root using the ADF and P&P tests. These tests suggest that most of the time series for EPS and DPS are nonstationary Stationary time series for EPS include, Alcoa (AA), Boeing (BA), Du-Pont (DD), KFT, MRK, PFE, T, TRV, and VZ whereas, stationary time series for DPS include Caterpillar (CAT), KFT, PG, XOM. While it is reassuring to note the non-rejection of nonstationarity, this is not Source: http://www.doksinet altogether surprising since many other studies find nonstationary in time series (Phillips and Perron, 1988; Brenner and Kroner, 1995; and, Doukas and Rahman, 1987). Johansen Bivariate Tests for Cointegration Rank for EPS and DPS The results for systems (composed of the EPS and DPS for each company) using Johansen’s method are presented in Table III. Tests using the Trace statistic are reported in Table III along with

the critical and the ‘p’ values. These are basically likelihood ratio tests, where the null hypothesis is L r+1 =L r+2 =.=L p =0, indicating that the system has p-r unit roots, where r is the number of cointegrating vectors. The rank is then determined using a sequential approach starting with the hypothesis of p unit roots. If this is rejected then the next hypothesis L 2 =L 3 =.=L p =0 is tested and so on From Table III it is evident that there is a close relationship between EPS and DPS as all bivariate systems are cointegrated. Hence, the EPS and DPS are closely related to each other in a long-term equilibrium relationship. Error Correction Model for EPS and DPS From Table IV, the closeness of the relationships between EPS and DPS using the error correction methodology is shown. In addition, using the lagged values of change in EPS and DPS one can establish the lead and lags between the two variables. For AA for example, coefficients for the error correction cointegration

equation are significant and negative indicating mean reversion of the long-term dividend policy. Also, from the lagged differenced values it is evident that ΔEPS(-1) leads in a significant way the ΔDPS(-1) indicating that changes in EPS affects the level of dividends paid by this company. For American Express (AXP), the error correction coefficients for both ΔEPS and ΔDPS are significant. The fact that ΔEPS does not lead or lag ΔDPS for both lags in a significant way may indicate that AXP’s dividend policy may be independent of the level of earnings in the short-term for this company. For both BA and Source: http://www.doksinet Bank of America (BAC), the coefficients for the error correction equation are significant. Also, similar to AA, earnings lead dividends per share with a lag of two periods. CAT exhibits a mean reverting error correction cointegrating relationship. But, there is no clear cut causality that exists in the short-term between ΔEPS and ΔDPS. This may

imply that the change in earnings may not affect the amount of dividend payment by this firm in the shortrun. There is a long-term relationship between dividends and earnings for CAT, however, as discussed previously. Similar to CAT, Chevron Corporation (CVX), DD, GE, Hewlett Packard (HPQ), JPM, KFT, MRK, PFE, T, TRV, and VZ do not indicate any short-term relationship between earnings and dividends, but do exhibit long-term mean-reverting behavior between ΔEPS and ΔDPS. This may imply that these companies, in the short-run, may not follow dividend payments based on current earnings but may rely on the earnings from the previous years. As pointed out in several studies, many firms are reluctant to reduce the dividend payments even when they incur losses as the investors consider a dividend reduction as a negative signal about the future prospects of the company in the long-term. Several firms also smooth out the dividend payments on the basis of their long-term earnings potential

rather than short-term profits. This finding is similar to the findings by Charitou, Lambertides, and Thedoulou (2010, 2011), Healy and Palepu (1988), Hsu, Wang, and Wu (1998) and Marsh and Merton (1987). On the other hand, Disney (DIS), Home Depot (HD), McDonald Corporation (MCD), and United Technologies Corporation (UTX), the earnings lead the dividends with lags. These companies therefore, display contemporaneous dividend policy directly related to earnings in the short-term. It should be noted that these companies are very well established and have consistent earnings. The consistent dividend payers, however, have declined in recent times as the Source: http://www.doksinet percentage of US industrial firms paying dividends has shrunk from 66.5% to 208% from 1978 to1999 (Fama and French, 2001). After the financial crisis of 2007/08 the amount of dividend payment has gone up slightly as compared to 1999. Bi-directional causality exists for companies such as IBM, JNJ, MMM, PG,

Walmart (WMT) and XOM. This indicates that the earnings affect dividend policy as well as dividend policy also has an impact on the earnings of these companies. It may also mean that if companies retain a larger amount and make investments in projects that generate large net present values for the company, the future growth of earnings is likely to increase. These are very progressive companies with large investments globally. Hence, these companies follow a dividend policy based on the opportunities to grow in the global markets. IV. Conclusions V. Bibliography Appendix Source: http://www.doksinet TICKER SIC NAME AA AXP BA BAC CAT CVX DD DIS GE HD HPQ IBM INTC JNJ JPM KFT KO MCD MMM MRK PFE PG T TRV UTX VZ WMT XOM 3334 6199 3721 6021 3531 2911 3087 7990 3621 5211 3571 3570 3674 2834 6021 2000 2080 5812 3699 2834 2834 2844 4813 6712 3724 4813 5331 1311 Alcoa Inc. American Express Co. Boeing Co. Bank Of America Co. Caterpillar, Inc. Chevron Corporation Du Pont Walt Disney Co.

General Electric Co. Home Depot Inc. Hewlett-Packard Co. International Business Machines Intel Corporation Johnson and Johnson JP Morgan Chase and Co. Kraft Foods Inc. Coca Cola Company McDonald’s Corporation 3 M Company Merck and Company Pfizer Inc. Procter and Gamble Co. AT&T Inc. The Travelers Companies United Technologies Corporation Verizon Communications Wal-Mart Stores Exxon Mobil Corporation Source: http://www.doksinet Table I Descriptive Statistics for Dividends Payout Ratio (DPS/EPS) for DOW Companies Company Mean Median Max Min Company Mean Median Max Min AA 0.28 0.29 1.29 -0.88 KFT 0.54 0.53 2.42 0.00 AXP 0.28 0.20 2.00 -0.56 KO 0.24 0.45 1.70 -8.50 BA 0.23 0.28 2.80 -3.50 MCD 0.23 0.13 1.45 -0.87 BAC 0.71 0.42 12.8 -0.67 MMM 0.47 0.49 1.25 -1.96 CAT 0.25 0.25 2.97 -7.52 MRK 0.61 0.47 3.80 -2.24 CVX 0.47 0.38 5.83 -1.97 PFE 0.75 0.44 9.33 -3.00 DD 0.22 0.43 2.69 -15.75 PG 0.42 0.39 0.98

-0.15 DIS 0.35 0 10.5 -2.26 T 0.57 0.61 2.39 -4.50 GE 0.54 0.45 1.85 0.33 TRV 0.28 0.26 1.38 -0.52 HD 0.12 0.14 1.13 -7.50 UTX 0.33 0.29 2.00 -0.28 HPQ 0.36 0.19 4.00 -0.24 VZ 0.54 0.68 3.21 -7.18 IBM 0.19 0.15 5.00 -2.08 WMT 0.21 0.19 0.37 0.18 INTC 0.26 0.06 3.50 0.00 XOM 0.45 0.43 1.03 0.13 JNJ 0.42 0.36 3.57 -0.91 JPM 0.48 0.37 5.43 -1.89 Source: http://www.doksinet Table II Augmented Dickey-Fuller (ADF) and Phillips Perron (PP) tests for unit root in the of the quarterly DPS and the EPS for twenty eight companies for the time period March, 1992 through September 2011. EPS Ticker: AA AXP BA BAC CAT CVX DD DIS GE HD HPQ IBM INTC JNJ JPM KFT KO MCD MMM MRK PFE PG T TRV UTX VZ WMT XOM ADF -6.40* -2.45 -7.88* 0.45 0.27 -1.88 -11.49* -0.54 -1.86 -1.09 -0.10 -1.39 -2.49 -0.20 -2.92* -5.97* 1.21 0.02 -1.18 -6.65* -6.87* -0.39 -7.63* -5.08* -0.31 -4.47* 4.31* -1.21 DPS PP -6.41* -1.55 -7.96* 0.45 -11.23* -1.88

-11.42* -3.03* -3.03* -3.24* -2.21 -3.28* -2.44 -3.09* -2.99* -5.97* -4.81* -2.52 -1.41 -6.62* -6.79* -0.32 -7.63* -5.13* -2.57 -7.64* -2.90* -1.19 ADF -2.39 -0.33 1.05 -2.06 -6.54* 2.45 -1.69 -10.45* -1.73 0.53 -10.39* 1.80 -8.73* 4.05* -2.93* -4.26* 3.80* -6.00* 0.24 -2.54 -1.24 7.55* 0.12 -2.75* 5.25* 0.19 7.36* 4.30* Critical Value (10%) PP -2.12 -0.22 -0.51 -1.34 -6.97* 1.23 -1.64 -1.77 -1.72 1.25 -0.79 1.29 0.76 -0.82 -1.89 -3.15* 1.23 0.06 1.33 -2.17 -1.24 3.65* -0.12 -1.79 1.74 0.23 2.71* 3.01* -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 -2.59 Critical values are taken from MacKinnon (1991). *,, implies significance at 10%, 5%, and 1% levels respectively. Source: http://www.doksinet Table III Long-Run Cointegrating Relationship Between Dividends and Earnings: Johansen’s Methodology Group DPS and EPS r= Eigen Value Trace Statistic Critical Value (5%)

Probability 0 1 0.324 0.065 34.41* 3.02 15.49 3.84 0.00 0.12 AXP 0 1 0.190 0.001 15.89* 0.07 15.49 3.84 0.04 0.80 BA 0 1 0.319 0.018 30.22* 1.35 15.49 3.84 0.00 0.24 BAC 0 1 0.269 0.065 17.95* 2.02 15.49 2.85 0.02 0.12 CAT 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 CVX 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 DD 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 DIS 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 GE 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 HD 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 HPQ 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 IBM 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 INTC 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 JNJ 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 JPM 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 Ticker AA Source: http://www.doksinet Table III Continued Group DPS and EPS r= Eigen Value Trace

Statistic Critical Value (5%) Probability 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 KO 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 MCD 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 MMM 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 MRK 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 PFE 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 PG 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 T 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 TRV 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 UTX 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 VZ 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 WMT 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 XOM 0 1 0.324 0.065 34.41* 2.02 15.49 2.85 0.00 0.12 Ticker KFT Source: http://www.doksinet Table IV Estimates of the Parameters from the Error Correction Model for Dividends versus Earnings per Share Model for EPSt Regressor Ticker: AA

Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient SE 0.09263 1.00000 -3.21517* 0.92723 Error Correction Coint Equation1 ΔEPS -0.28236 * (-2.06) ΔDPS 0.05595* (5.23) ΔEPS(-1) -0.12722 (-0.86) -0.81439 (-0.71) -1.59281 (-1.25) 0.24455 (0.19) -0.05262* (-4.57) -0.00312 (-0.35) -0.41158* (-4.14) -0.20121* (-2.04) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: AXP Cointegrating Equation Constant EPS(-1) DPS(-1) -0.10007 1.00000 -3.58970* 1.13260 Error Correction Coint Equation1 ΔEPS -0.21280* (-3.09) ΔDPS 0.01604* (2.02) ΔEPS(-1) -0.04777 -0.00988 (-0.43) (-0.77) -0.00928 0.39540* (3.75) (-0.76) -0.87374 -0.50320* (-0.88) (-4.39) -1.03562 -0.23023* (-1.04) (-2.01) *,, represent significance at 1%, 5% and 10% levels. ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: BA Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient SE -0.26269 1.00000 -1.35911* 0.58856

Error Correction Coint Equation1 ΔEPS -0.099091* (-4.74) ΔDPS 0.00856* (2.13) ΔEPS(-1) -0.12768 (-0.16) 0.03255 (0.27) 8.79246 (-1.25) 5.31067 (0.85) -0.004955* (-1.51) -0.00286 (-1.26) -0.16597 (-1.40) -0.18428* (-1.53) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: BAC Cointegrating Equation Constant EPS(-1) DPS(-1) 0.06427 1.00000 -2.04811* 0.2786 Error Correction Coint Equation1 ΔEPS 0.27159* (2.02) ΔDPS 0.11367* (4.74) ΔEPS(-1) -0.56752* (-3.90) -0.86320* (-6.72) -0.75057 (-1.14) 0.60073 (0.91) -0.02904 (-1.12) -0.07369* (-3.22) 0.51730* (4.38) -0.14646 (-1.26) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: CAT Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient -0.10748 1.00000 -2.71415* SE 0.64276 Error Correction Coint Equation1 ΔEPS -0.218666 * (-2.59) ΔDPS 0.13636* (4.01) ΔEPS(-1) -0.06446

(-0.46) 0.02924 (0.41) -0.46351* (-1.57) 0.04109 (0.14) -0.17585* (-3.12) 0.04340* (1.51) -1.12586* (-9.48) -0.20121* (-3.48) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: CVX Cointegrating Equation Constant EPS(-1) DPS(-1) 0.60589 1.00000 -4.25379* 0.74611 Error Correction Coint Equation1 ΔEPS -0.34498* (-2.40) ΔDPS 0.01477* (1.79) ΔEPS(-1) -0.00536 (-0.04) -0.09349 (-0.73) -0.91547 (-0.44) -2.08454 (-0.99) -0.00286 (-0.35) 0.00050 (0.07) -0.56543* (-4.71) -0.26579* (-2.20) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: DD Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient 0.05502 1.00000 -2.06192 SE 1.79546 Error Correction Coint Equation1 ΔEPS -0.110650* (-4.60) ΔDPS 0.00105 (0.53) ΔEPS(-1) -0.16117 (-0.87) -0.20474* (-1.88) -5.75288 (-0.39) 11.8349 (0.81) -0.00054 (-0.36) -0.00018 (-0.20) -0.11046

(-0.92) -0.10731 (-0.90) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: DIS Cointegrating Equation Constant EPS(-1) DPS(-1) 0.21206 1.00000 -8.37803* 0.62047 Error Correction Coint Equation1 ΔEPS -0.02181 (-0.46) ΔDPS 0.31688* (10.86) ΔEPS(-1) -0.67768* (-5.86) -0.1145 (-0.93) -0.24743 (-0.92) 0.13986 (0.75) -0.30823* (-4.27) -0.02247 (-0.29) 1.16282* (6.94) 0.53148* (4.60) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: GE Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient -0.04304 1.00000 -1.65166* SE 0.07587 Error Correction Coint Equation1 ΔEPS -0.10787 (-0.43) ΔEPS(-1) -0.56746* (-2.30) -0.03830 (-0.22) -0.18236 (-0.50) 0.03446 (0.09) -0.22208* (-3.36) -0.04958 (-1.09) 0.07419 (0.76) 0.05914 (0.63) -0.23962 1.00000 -1.09588* 0.38893 Error Correction Coint Equation1 ΔEPS -0.04731 (-0.58) ΔDPS

0.04063* (4.29) ΔEPS(-1) -0.13118* (-1.57) -0.80318* (-9.86) 0.29365 (030) -0.76723 (-0.77) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: HD Cointegrating Equation Constant EPS(-1) DPS(-1) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and ΔDPS 0.37677* (5.65) -0.01967* (-2.01) -0.03718* (-3.88) -0.70362* (-6.17) -0.05232 (-0.45) *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: HPQ Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient SE 0.13818 1.00000 -7.14974* 3.89941 Error Correction Coint Equation1 ΔEPS -0.03040 (-0.45) ΔDPS 0.04144* (2.81) ΔEPS(-1) -0.17289 (-1.43) -0.46185* (-3.91) 0.37049 (0.60) -0.51822 (-0.97) -0.02215 (-0.83) 0.00572 (0.21) -0.74086* (-5.47) 0.02182 (0.18) -0.54337 1.00000 -2.42055* 0.42263 Error Correction Coint Equation1 ΔEPS -0.68979* (-5.60) ΔDPS 0.03522* (6.46) ΔEPS(-1) -0.10011 (-0.70) 0.31580* (2.60) 11.3263* (4.37)

11.1606* (4.54) -0.03001* (-4.73) -0.00656 (-1.22) -0.37022* (-3.22) -0.11696 (-1.07) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: IBM Cointegrating Equation Constant EPS(-1) DPS(-1) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: INTC Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient SE -0.17533 1.00000 -1.05203* 0.48773 Error Correction Coint Equation1 ΔEPS -0.21353 * (-2.41) ΔDPS 0.11946* (3.15) ΔEPS(-1) 0.03523 (0.28) 0.11188 (0.93) -0.93291* (-3.67) -0.63228* (-2.42) 0.07323 (0.02) -0.00088 (-0.35) -1.19204* (-10.98) -0.33485* (-3.00) -0.26992 1.00000 -1.37747* 0.13002 ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: JNJ Cointegrating Equation Constant EPS(-1) DPS(-1) Error Correction Coint Equation1 ΔEPS 0.22194* (2.01) ΔEPS(-1) -1.00620* (-9.33) -0.39328* (-3.18) -2.78467* (-1.57) -11.3680* (-9.79) ΔEPS(-2)

ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and ΔDPS 0.04287* (6.81) 0.01573* (2.55) -0.00571 (-0.80) -0.36998* (-3.66) -0.48121* (-7.25) *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: JPM Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient SE -0.81771 1.00000 0.50469 1.09814 Error Correction Coint Equation1 ΔEPS -0.34246 * (-2.16) ΔDPS 0.04916* (1.75) ΔEPS(-1) -0.05872 (-0.34) -0.02211 (-0.14) 0.69838 (0.77) -1.12276 (-1.08) -0.01578 (-0.51) 0.00772 (0.28) -0.10459 (-0.65) -0.07235 (-0.39) -0.48007 1.00000 0.05189 0.44105 Error Correction Coint Equation1 ΔEPS -1.14863* (-3.92) ΔDPS - 0.01969 (-1.26) ΔEPS(-1) 0.13185 (0.56) 0.09244 (0.55) -4.48729 (-1.42) -1.30377 (-0.98) 0.01440 (1.14) 0.00828 (0.92) -0.26995* (-1.60) -0.10811* (-1.52) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: KFT Cointegrating Equation Constant EPS(-1) DPS(-1) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2)

(t-values are in parentheses and *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: KO Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient SE -0.35946 1.00000 -0.45392* 0.24801 Error Correction Coint Equation1 ΔEPS -0.24485 (-1.47) ΔDPS 0.03302* (6.81) ΔEPS(-1) -0.57342* (-2.96) -0.29646 (-1.49) 8.61242* (2.90) 6.35806* (1.99) -0.03725* (-6.62) -0.04208* (-7.31) -0.53955* (-6.26) -0.24534* (-2.65) -0.17682 1.00000 -1.74372* 0.07853 Error Correction Coint Equation1 ΔEPS -0.37247* (-2.02) ΔDPS 1.06304* (5.98) ΔEPS(-1) -0.34479* (-2.41) -0.60829* (-5.61) -0.10815 (-0.45) 0.13522 (0.97) -0.26156* (-1.90) -0.34420* (-3.29) 0.66663* (2.58) 0.34858* (2.58) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: MCD Cointegrating Equation Constant EPS(-1) DPS(-1) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and *,, represent significance at 1%, 5% and 10% levels. Source:

http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: MMM Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient SE 0.34412 1.00000 -3.25414* 0.26216 Error Correction Coint Equation1 ΔEPS -0.63401* (-3.96) ΔDPS 0.01198 (1.39) ΔEPS(-1) 0.01267 (0.09) 0.02523 (0.20) 0.47703 (0.22) 3.30108* (1.51) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: MRK Cointegrating Equation Constant EPS(-1) DPS(-1) -0.01483* (-1.90) -0.00038 (-0.06) -0.20679* (-1.78) -0.23839* (-2.03) 0.00132 1.00000 -1.91686* 0.48240 Error Correction Coint Equation1 ΔEPS -0.87511* (-4.39) ΔDPS 0.00632 (1.23) ΔEPS(-1) 0.03485 (0.21) -0.01765 (-0.14) 2.23995 (0.48) 1.58511 (0.34) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and -0.00323 (-0.75) -0.00209 (-0.64) -0.16572 (-1.39) -0.10906 (-0.91) *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: PFE Cointegrating Equation Constant

EPS(-1) DPS(-1) Coefficient SE -0.07956 1.00000 -1.21193* 0.21558 Error Correction Coint Equation1 ΔEPS -0.98308 * (-4.59) ΔDPS 0.042991* (1.74) ΔEPS(-1) 0.03748 (0.21) 0.04115 (0.33) -0.98143 (-0.95) 0.55811 (0.53) -0.03677* (-1.83) -0.01485 (-1.03) -0.03977 (-0.33) -0.03003 (-0.25) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: PG Cointegrating Equation Constant EPS(-1) DPS(-1) -0.26489 1.00000 -1.19042* 0.22525 Error Correction Coint Equation1 ΔEPS -0.39378* (-4.24) ΔDPS 0.02718* (4.88) ΔEPS(-1) -0.40253* (-4.29) -0.39345* (-3.97) 5.21015* (2.77) 11.1871* (6.80) -0.04056* (-7.21) -0.02132* (-3.58) -0.46254* (-4.09) -0.24446* (-2.47) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: T Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient -0.21999 1.00000 -0.76531 SE 0.52926 Error Correction Coint Equation1

ΔEPS -0.99820* (-26.97) ΔDPS 0.00279 (0.42) ΔEPS(-1) 0.00795 (0.06) 0.00538 (0.07) -2.36669 (-0.87) -0.50948 (-0.19) 0.00188 (0.35) 0.000406 (1.08) 0.16091 (1.28) -0.28533* (-2.26) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: TRV Cointegrating Equation Constant EPS(-1) DPS(-1) -0.07621 1.00000 -2.59975* 4.25334 Error Correction Coint Equation1 ΔEPS -0.48003* (-3.32) ΔDPS 0.00369 (0.60) ΔEPS(-1) -0.07133 (-0.49) -0.0922 (-0.72) -1.46537 (-0.53) 0.15521 (0.06) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and -0.00582 (-0.94) -0.00102 (-0.19) -0.58721* (-4.97) -0.28361* (-2.42) *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: UTX Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient -0.22722 1.00000 -2.11859* SE 0.33210 Error Correction Coint Equation1 ΔEPS -0.22502* (-1.69) ΔDPS 0.03994* (3.87) ΔEPS(-1) -0.33182* (-2.47) -0.44003* (-3.90) -0.43777 (-0.28)

0.39640 (0.26) -0.02488* (-2.39) -0.01433* (-1.63) -0.32275* (-2.72) -0.23137* (-1.93) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: VZ Cointegrating Equation Constant EPS(-1) DPS(-1) -0.06465 1.00000 -1.05295 2.14194 Error Correction Coint Equation1 ΔEPS -0.73358* (-4.25) ΔDPS -0.00056 (-0.29) ΔEPS(-1) -0.14258 (-0.90) 0.09064 (0.76) -11.3827 (-1.06) -0.11924 (-0.01) 0.00071 (0.40) 0.00085 (0.63) -0.15042 (-1.26) -0.10351 (-0.87) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and *,, represent significance at 1%, 5% and 10% levels. Source: http://www.doksinet Table IV Continued Model for EPSt Regressor Ticker: WMT Cointegrating Equation Constant EPS(-1) DPS(-1) Coefficient SE 0.19865 1.00000 -6.09809* 0.53490 Error Correction Coint Equation1 ΔEPS -0.26068* (-4.81) ΔDPS - 0.00457 (-1.45) ΔEPS(-1) -1.01037* (-9.48) -0.32576* (-1.84) -9.54996* (-5.00) -6.65585* (-3.27) 0.07780* (12.58) 0.04011* (3.91) 0.17826* (1.61) -0.01042 (-0.09) -0.17005

1.00000 -2.59725* 0.93605 Error Correction Coint Equation1 ΔEPS -0.13928* (-1.86) ΔDPS 0.01162* (4.79) ΔEPS(-1) 0.02240 (0.18) -0.06413 (-0.52) 7.34860* (2.11) 0.20499 (0.06) -0.00631* (-1.59) 0.00044 (0.10) -0.40434* (-3.58) -0.42014* (-3.66) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) Ticker: XOM Cointegrating Equation Constant EPS(-1) DPS(-1) ΔEPS(-2) ΔDPS(-1) ΔDPS(-2) (t-values are in parentheses and *,, represent significance at 1%, 5% and 10% levels