Managerial FinanceEarnings per share versus cash flow per share as predictor of dividends per shareJohn Consler Greg M. Lepak Susan F. Havranek

Article information:To cite this document:John Consler Greg M. Lepak Susan F. Havranek, (2011),"Earnings per share versus cash flow per share aspredictor of dividends per share", Managerial Finance, Vol. 37 Iss 5 pp. 482 - 488Permanent link to this document:http://dx.doi.org/10.1108/03074351111126960

Downloaded on: 18 October 2014, At: 03:30 (PT)References: this document contains references to 12 other documents.To copy this document: permissions@emeraldinsight.comThe fulltext of this document has been downloaded 2110 times since 2011*

Users who downloaded this article also downloaded:Wael Mostafa, (2014),"The relative information content of cash flows and earnings affected by theirextremity: UK evidence", Managerial Finance, Vol. 40 Iss 7 pp. 646-661 http://dx.doi.org/10.1108/MF-06-2013-0128Michael R. Powers, Abdulrahman Ali Al#Twaijry, (2007),"Dividend policy and payout ratio: evidence fromthe Kuala Lumpur stock exchange", The Journal of Risk Finance, Vol. 8 Iss 4 pp. 349-363Kamran Ahmed, Muhammad Jahangir Ali, (2013),"Determinants and usefulness of analysts' cash flowforecasts: evidence from Australia", International Journal of Accounting & Information Management,Vol. 21 Iss 1 pp. 4-21

Access to this document was granted through an Emerald subscription provided by 198285 []

For AuthorsIf you would like to write for this, or any other Emerald publication, then please use our Emerald forAuthors service information about how to choose which publication to write for and submission guidelinesare available for all. Please visit www.emeraldinsight.com/authors for more information.

About Emerald www.emeraldinsight.comEmerald is a global publisher linking research and practice to the benefit of society. The companymanages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well asproviding an extensive range of online products and additional customer resources and services.

Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committeeon Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archivepreservation.

*Related content and download information correct at time of download.

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity A

t 03:

30 1

8 O

ctob

er 2

014

(PT

)

Earnings per share versuscash flow per share as predictor

of dividends per shareJohn Consler and Greg M. LepakDepartment of Business Administration,

Le Moyne College, Syracuse, New York, USA, and

Susan F. HavranekDepartment of Accounting, Le Moyne College, Syracuse, New York, USA

Abstract

Purpose – The purpose of this paper is to compare the relative power of operating cash flow andearnings in the prediction of dividends.

Design/methodology/approach – A linear mixed effects model is used in terms of selected modelfit criteria.

Findings – Based on the selected model fit criteria, cash flow per share is shown to produce a betterfit than earnings per share, but it cannot be said how much better.

Research limitations/implications – Quarterly CRSP and Compustat data from 2000 to 2006 for1,902 dividend-paying firms are analyzed. Future work would need a different methodology todetermine how much better cash flow is as a predictor of dividends.

Practical implications – Both earnings per share and cash flow per share are found to bereasonable dividend predictors.

Social implications – Additional insight is provided on modeling factors that contribute to a firm’sdecision to engage or disengage in a dividend payment policy.

Originality/value – The study described in this paper continues work on predicting dividendsper share. Results show cash flow per share is a better predictor than earnings per share. Investors andanalysts predict dividends as part of their stock valuation work. This study suggests focusingattention on using cash flow per share as the predictor of dividends.

Keywords Dividends, Cash flow, Earnings per share, Modelling

Paper type Research paper

I. IntroductionCash flow might be expected to be a better predictor of dividends than earnings,because cash flow is less subject to accounting manipulation than earnings. It can beargued, cash flow is necessary for earnings, which are necessary for cash dividends.While the link between earnings and dividends has been well established in theliterature, little work has been done linking cash flow directly to dividends. This studyinvestigates this issue to see if cash flow is the better predictor of dividends.

Quarterly CRSP and Compustat data from 1,902 dividend-paying firms between 2000and 2006 are used in the study. Linear mixed effects models are used to test whichvariable, earnings per share or cash flow per share, is the better predictor of dividendsper share. Models are tested both without and with control variables of total assets, debtratio, market to book value ratio, current liquidity ratio, quarterly beta, and fourthquarter indicator. These control variables have been shown to have significantrelationships to dividends in prior studies.

The current issue and full text archive of this journal is available at

www.emeraldinsight.com/0307-4358.htm

MF37,5

482

Managerial FinanceVol. 37 No. 5, 2011pp. 482-488q Emerald Group Publishing Limited0307-4358DOI 10.1108/03074351111126960

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity A

t 03:

30 1

8 O

ctob

er 2

014

(PT

)

Cash flow per share is shown to be the better predictor of dividends per share, bothwithout and with control variables, in terms of selected model fit criteria. However, giventhe model used, it is not clear if cash flow is a significantly better predictor.

II. Literature reviewArticles since 1985 provided the relevant literature review. Miller and Rock (1985) findearnings determine firm value with asymmetric information. This would argue thatearnings per share might be a good predictor of dividends per share with minimal timelags. This provides a base upon which to build the literature review.

Cambell and Shiller (1988) establish a link between the moving average of a firm’searnings and the present value of all future dividends. This predictive ability may holdtrue for earnings per share and dividends per share, as well.

McCann and Olson (1994) provide further support for the relationship between afirm’s earnings and dividends. The article supports the paying of dividends regardlessof the perfection of the capital markets.

Benartzi et al. (1997) find that dividends convey information on current and pastearnings. Therefore, the reverse may be true also. Support for concurrent changes inearnings and dividends is provided. This again argues for minimal time lags in currentwork.

Lamont (1998) uses quarterly earnings to help predict short-term returnssuccessfully. This work supports the minimal time lag concept.

Fama and French (2001) support profitability as one characteristic that affects afirm’s decision to pay cash dividends. Strongly negative earnings cause termination ofcash dividend payments. This work suggests a relationship between earnings and cashdividends.

Koch and Sun (2004) find that the market reaction to dividend changes is a delayedreaction to previous changes in earnings. Support for a relationship between earningsand dividends is implied.

Liu et al. (2007) investigate if operating cash flows or accounting earnings are betterat explaining equity valuations. In all cases, earnings forecasts were a better summarymeasure of value. This article was critical in stimulating the current work. The questionof earnings versus cash flow can be applied to predicting dividends, as well as valuation.

Recent work provides further support for the link between earnings and dividends.Bali et al. (2008) find a strong positive relationship between earnings and expectedreturns, which includes dividends.

The constant dividend growth model for valuing stocks suggests that value investorsare attracted and retained when firms are able to evidence a long history of stabledividend payments. However, Fama and French (2001) document that fewer firms arepaying dividends and that strongly negative earnings cause termination of cashdividend payments. While most of the literature supports a strong relationship betweenearnings and dividends, this study will investigate if the relationship between cash flowand dividends is even stronger.

III. Sample and dataFirms that declared cash dividends, excluding payments made as part of liquidations,acquisitions or reorganizations, during the period of January 1, 2000 and March 31, 2006were identified in CRSP. It was assumed that dividends declarations made during

Predictorof dividends

per share

483

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity A

t 03:

30 1

8 O

ctob

er 2

014

(PT

)

the last 15 days of a quarter and anytime during the following quarter prior to the last15 days were dependent on the quarter of interest. For example, if the first quarter runsJanuary 1-March 31, dividends declared between March 16 and June 15 would beassumed to be dependent on financial activity during the first quarter.

Some industries were observed to have monthly dividend payments or multiple typesof cash dividends as coded by CRSP. When this was the case, the multiple dividendamounts were summed and reported as a single observation for the quarter. In order to usea panel data methodology, time identifications based on calendar quarter were assigned.Firms with fiscal quarters ending between January 1, 2000 and March 31, 2000 are labeledas time period 0. All fiscal year and quarter-ends were retained in the sample. Quarterlybeta was calculated for fiscal quarters using daily prices and NYSE equal-weightedmarket index data from CRSP. Monthly high and low market prices, quarterly balancesheet and income statement values and outstanding share data were collected fromCompustat. Observations with missing balance sheet or income statement data weredeleted. Missing data for high and low market price and outstanding shared were handcollected where possible; otherwise, the observations were deleted. In addition, firms withthe term “trust” in the company name were deleted. The authors noted that cash flow datafrom the statement of cash flows were missing across and within firms over the timeperiod in question. We opted to use a commonly calculated value for operating cash flow(net income plus depreciation and amortization plus net working capital) to increase thesample size and avoid any unknown biases reflected by the missing cash flow data.

The main objective of this study is to assess which of two variables, earnings per shareor cash flow per share, does better in predicting dividends. The response variable isdefined as quarterly cash dividends per share (DPS, $ per share) from the first quarter of2000 to the first quarter of 2006 (25 periods). It is important to note that quarterly dividendswere not obtained for all firms at all 25 time points. There were 23,334 observationsinvolving 1,902 firms in all industries. To reduce skewness, the analyses reported here arebased on the natural log transformed DPS values: log (DPS values þ 1).

Preliminary analyses show that the log DPS values increase from the first quarter of2000 to the first quarter of 2006. A linear curve provided a reasonable approximation tomodel the increase in log DPS values. For this panel study, a model for the mean responseis used that allows the rates of change in the DPS values to differ between and withinfirms in the time span under investigation. The response pattern for each firm includesan intercept at baseline and a slope, where the intercepts and slopes are allowed to varyrandomly. It is assumed that the random intercepts and slopes account for thewithin-firm correlation of the response measurements over time.

Models are discussed that relate dividends to earnings per share and cash flow pershare, while controlling for other time-varying financial variables. Quarterly data fromthe first quarter of 2000 to the first quarter of 2006 were obtained on the followingvariables at all measurement occasions at which a DPS value was available:

. Earnings per share (EPSH), calculated as net income divided by a weightedaverage of common shares outstanding for the year.

. Operating cash flow per share (CFPSH), calculated as net income plusdepreciation plus net working capital divided by the weighted average ofcommon shares outstanding used to obtain earnings per share (Liu et al., 2007).

. Total assets (ASSETS, millions $).

MF37,5

484

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity A

t 03:

30 1

8 O

ctob

er 2

014

(PT

)

. Debt ratio (DEBT, total liabilities divided by total assets).

. Market to book value ratio (MARKET, average price per share divided by bookvalue per share).

. Current liquidity measure (LIQUIDITY, total current assets divided by totalcurrent liabilities).

. Quarterly beta (QTRBETA).

. Indicator variable (IQTR), representing the fourth fiscal quarter.

To address skewness, the data on ASSETS and DEBT were log transformed:log(ASSETS) and log(DEBT þ 1), respectively. In addition, the calculated values forearnings per share, operating cash flow per share, market to book, current liquidity, andquarterly beta were assigned to centiles. To simplify notation, these transformed variablesare denoted EPSH, CFPSH, MARKET, LIQUIDITY, and QTRBETA, respectively.

IV. Analysis and resultsWe began the investigation by fitting a linear mixed effects model (Pinheiro and Bates,2000) to the dividends and earnings per share data. This flexible approach can be used infinancial analysis to model population characteristics that are common to all firms, aswell as random response patterns that correspond to individual firms over time. Thistype of model accommodates inherently unbalanced longitudinal data, i.e. the number ofmeasurements on each firm can be different and the measurements need not be collectedat the same set of measurement occasions.

Equations (1a) and (1b) present linear mixed effects models for log DPS and EPSH;equation (1b) includes control variables:

EðlogðDPSÞijjbiÞ ¼ ðb1 þ b1iÞ þ ðb2 þ b2iÞtij þ b3EPSHij þ 1ij;

i ¼ 1; . . . ;N; j ¼ 1; . . . ; ni

ð1aÞ

EðlogðDPSÞijjbiÞ ¼ ðb1 þ b1iÞ þ ðb2 þ b2iÞtij þ b3EPSHij þ b4MARKETij

þ b5logðASSETSÞij þ b6logðDEBTÞij þ b7LIQUIDITYij

þ b8QTRBETAij

þ b9IQTRij þ 1ij; i ¼ 1; . . . ;N; j ¼ 1; . . . ; ni;

ð1bÞ

Log(DPS)ij represents the log DPS value for the ith firm at the jth measurement occasion, tij

is the time since baseline (tij ¼ 0 in the first quarter of 2000). In each model, the vectors ofrandom coefficients bi ¼ (b1i, b2i) are independent and identically distributed with amultivariate distribution N(0, G), and the eij are within-firm errors, which are independentand identically distributed with a N(0,s2) distribution, independent of the random effects.The random effects corresponding to the intercepts and slopes induce covariance amongthe repeated measures. Note that IQTRij is a fourth-quarter indicator, i.e. IQTRij ¼ 1 if thejth measurement occasion for the ith firm is in the fourth quarter and 0 otherwise. Thefourth quarter DPS are expected to be larger than the other three quarter DPS figures.

The models in equations (1a) and (1b) were estimated using maximum likelihood withdata from the first quarter of 2000 to the first quarter of 2006. The Hausman (1978)specification test confirmed the need for random intercepts and slopes in both estimatedmodels. Results are given in Table I.

Predictorof dividends

per share

485

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity A

t 03:

30 1

8 O

ctob

er 2

014

(PT

)

Estimate SE p-value

Estimated model in equation (1a) – EPSHConstant 0.15735 0.00518 ,0.0001tij 0.00253 0.00024 ,0.0001EPSH 0.00026 0.00003 ,0.0001MARKETLog(ASSETS)Log(DEBT)LIQUIDITYBETAIQTRVar(b1i) 0.04956Var(b2i) 0.00007Cov(b1i,b2i) 20.00096Var(1ij) 0.00697Log likelihood 19,808.8AIC 239,603.6Estimated model in equation (1b) – EPSHConstant 0.05536 0.01818 0.0023tij 0.00212 0.00025 ,0.0001EPSH 0.00022 0.00003 ,0.0001MARKET 0.00011 0.00005 0.0407Log(ASSETS) 0.02170 0.00238 ,0.0001Log(DEBT) 20.13421 0.01856 ,0.0001LIQUIDITY 0.00019 0.00006 0.0023BETA 20.00008 0.00003 0.0038IQTR 0.00310 0.00132 0.0192Var(b1i) 0.05144Var(b2i) 0.00007Cov(b1i,b2i) 20.00099Var(1ij) 0.00692Log likelihood 19,877.4AIC 239,728.8Estimated model in equation (2a) – CFPSHConstant 0.14371 0.00570 ,0.0001tij 0.00241 0.00024 ,0.0001CFPSH 0.00055 0.00005 ,0.0001MARKETLog(ASSETS)Log(DEBT)LIQUIDITYBETAIQTRVar(b1i) 0.05269Var(b2i) 0.00007Cov(b1i,b2i) 20.00098Var(1ij) 0.00694Log likelihood 19,820.7AIC 239,627.3Estimated model in equation (2b) – CFPSHConstant 0.08222 0.01845 ,0.0001tij 0.00202 0.00024 ,0.0001CFPSH 0.00072 0.00007 ,0.0001

(continued )

Table I.Estimation resultsfor models in equation(1a), (1b), (2a) and (2b)

MF37,5

486

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity A

t 03:

30 1

8 O

ctob

er 2

014

(PT

)

There is clear evidence that changes in mean log(DPS) are related to EPSH in the estimatedmodel without controls. The significant positive relationship between dividends andearnings per share also holds when control variables are present in the model.

Similar model building procedures were used to fit the models in equations (2a) and(2b) for log (DPS) and CFPSH using maximum likelihood estimation:

EðlogðDPSÞijjbiÞ ¼ ðb1 þ b1iÞ þ ðb2 þ b2iÞtij þ b3CFPSHij þ 1ij;

i ¼ 1; . . . ;N; j ¼ 1; . . . ; ni

ð2aÞ

EðlogðDPSÞijjbiÞ ¼ ðb1 þ b1iÞ þ ðb2 þ b2iÞtij þ b3CFPSHij þ b4MARKETij

þ b5logðASSETSÞij þ b6logðDEBTÞij þ b7LIQUIDITYij

þ b8QTRBETAij þ b9IQTRij þ 1ij;

i ¼ 1; . . . ;N; j ¼ 1; . . . ; ni;

ð2bÞ

Table I indicates a significant positive relationship between dividends and cash flow pershare in models with and without the financial control variables. Interestingly, all thecontrol variables in the model for dividends and earnings per share are significant, andremain so in the estimated model for dividends and computed cash flow per share.

Although the model with cash flow in equation (2a) and the model with earnings inequation (1a) are not nested, they can be compared directly in terms of their maximized loglikelihoods since they have the same number of parameters. The higher the log likelihoodnumber, the better the model fits the data. Comparing the fitted models without controls,the maximized log likelihood for the cash flow model is 19,820.7; the corresponding valuefor the earnings per share model is 1,908.8. Thus, the model with cash flow per shareprovides the better fit. An alternative measure for comparing non-nested models for thesame data is the Akaike information criterion (AIC, AIC ¼ 22 log likelihood þ 2 (numberof parameters)) (Sakamoto et al., 1986). The smaller the AIC, the better the model fit. TheAIC values for the cash flow model in equation (2a) and the earnings per share model inequation (1a) are239,627.3 and239,603.6, respectively. Hence, the model with cash flowswithout controls provides the better fit.

In like manner, the models with control variables in equations (2b) and (1b) can becompared in terms of their maximized log likelihoods and AIC values. The estimated loglikelihoods and AIC values indicate that the model with cash flow per share, rather thanthe model with earnings per share, is the better fitting model for the data at hand.

Estimate SE p-value

MARKET 0.00017 0.00005 0.0014Log(ASSETS) 0.02008 0.00241 ,0.0001Log(DEBT) 20.16456 0.01862 ,0.0001LIQUIDITY 20.00038 0.00009 ,0.0001BETA 20.00008 0.00003 0.0057IQTR 0.00282 0.00132 0.0325Var(b1i) 0.05227Var(b2i) 0.00007Cov(b1i,b2i) 20.00099Var(1ij) 0.00690Log likelihood 19,895.0AIC 239,763.9 Table I.

Predictorof dividends

per share

487

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity A

t 03:

30 1

8 O

ctob

er 2

014

(PT

)

Hence, we conclude that the better predictor of dividends per share is cash flow pershare rather than earnings per share.

V. ConclusionUsing quarterly data from 2000 to 2006 for dividend paying companies from CRSP andCompustat data of 1,902 firms, the study finds cash flow per share to be a better predictorfor dividends per share than is earnings per share, at least in terms of selected model fitcriteria. This is the case with or without control variables of total assets, debt ratio,market to book value ratio, liquidity measure of current ratio, quarterly beta, and fourthquarter indicator. These control variables have been shown to have a significantrelationship to dividends in prior studies.

The linear mixed effects models used are flexible enough to meet the needs of thestudy. These models are able to represent covariance structure with relatively fewparameters and can handle data that become available at irregular times.

Study results suggest that cash flow per share is the better predictor of dividendsper share, but it cannot be said how much better. In future work, a different approachwould be necessary to address the issue of magnitude raised here.

References

Bali, T.G., Ozgur, D. and Tehranian, H. (2008), “Aggregate earnings, firm-level earnings, and expectedstock returns”, Journal of Financial and Quantitative Analysis, Vol. 43 No. 3, pp. 657-84.

Benartzi, S., Michaely, R. and Thaler, R. (1997), “Do changes in dividends signal the future or thepast?”, The Journal of Finance, Vol. 52 No. 3, pp. 1007-34.

Cambell, J.Y. and Shiller, R.J. (1988), “Stock prices, earnings, and expected dividends”,The Journal of Finance, Vol. 43 No. 3, pp. 661-76.

Fama, E.F. and French, K.R. (2001), “Disappearing dividends: changing firm characteristics orlower propensity to pay?”, Journal of Financial Economics, Vol. 60, pp. 3-44.

Hausman, J. (1978), “Specification tests in econometrics”, Econometrica, Vol. 46, pp. 1251-71.

Koch, A.S. and Sun, A.X. (2004), “Dividend changes and the persistence of past earningschanges”, The Journal of Finance, Vol. 5, pp. 2093-116.

Lamont, O. (1998), “Earnings and expected returns”,The Journal of Finance, Vol. 53 No. 5, pp. 1563-87.

Liu, J., Nissim, D. and Thomas, J. (2007), “Is cash flow king in valuations?”, Financial AnalystsJournal, Vol. 63 No. 2, pp. 56-68.

McCann, D. and Olson, G.T. (1994), “The linkage between dividends and earnings”,The Financial Review, Vol. 29 No. 1, pp. 1-14.

Miller, M.H. and Rock, K. (1985), “Dividend policy under asymmetric information”, The Journalof Finance, Vol. 40 No. 4, pp. 1031-51.

Pinheiro, J.C. and Bates, D.M. (2000), Mixed-effectsModels in S and S-Plus, Springer, New York, NY.

Sakamoto, Y., Ishiguro, M. and Kitagawa, G. (1986), Akaike Information Criterion Statistics,Reidel, Dordrecht.

Corresponding authorJohn Consler can be contacted at: consleje@lemoyne.edu

MF37,5

488

To purchase reprints of this article please e-mail: reprints@emeraldinsight.comOr visit our web site for further details: www.emeraldinsight.com/reprints

Dow

nloa

ded

by N

ew Y

ork

Uni

vers

ity A

t 03:

30 1

8 O

ctob

er 2

014

(PT

)