8/13/2019 SSRN-id1021070

1/39

i

A Discrete Choice Model of Dividend Reinvestment Plans:

Classification and Prediction

by

Thomas P. Boehm Ramon P. DeGennaroSunTrust Professor of Finance CBA Professor of FinanceThe University of Tennessee The University of TennesseeKnoxville, TN 37996-0540 Knoxville, TN 37996-0540865-974-1723 865-974-1726tboehm@utk.edu rdegenna@utk.edu

Abstract

We use a discrete choice recursive model to classify 629 companies with dividendreinvestment plans (DRIPs) in 1999 and 514 companies without them. Our modelclassifies 72.0% of companies correctly. We interpret misclassified companies as beinglikely to switch their plan status. For example, if financial data erroneously suggest that acompany should have had a DRIP in 1999, then we expect that it would be more likely toinstitute a plan than other companies in the sample. Data from 2004 support this conjecturefor firms that add DRIPs. Companies that add DRIPs tend to have higher levels of

variables that control for management entrenchment except for common shares outstandingwhich tend to be lower, higher levels of variables that control for the ability to paydividends and higher payout ratios. Other factors that impact the likelihood of a companyhaving a DRIP include industry effects and the exchange on which the companys sharesare traded.

August 2010

Keywords: Dividend Reinvestment, Discrete Choice, Recursive.

Not for quotation without permission.

8/13/2019 SSRN-id1021070

2/39

- 1 -

1. Introduction

Dividend reinvestmentplans (DRIPs) allow investors to reinvest their dividends in

additional shares of the same stock that paid the dividend.1

Previous research suggests that

firms that offer such plans differ from those that do not in systematic ways (DeGennaro,

2003). Is it possible to use financial data to determine whether a firm will or will not offer

a plan? And is it possible to take the next step and predict which firms will or will not offer

plans in later years? The answer to both is yes. The financial characteristics of companies

that offer DRIPs dodiffer from those that do not offer them. In addition, financial

statements foreshadow changes in plan status up to five years before management adds a

DRIP. If our model makes a correct prediction in 1999 then the chances are quite high that

the prediction will hold for five years. If our model makes an incorrect prediction that a

company should offer a DRIP, though, then the odds are much higher that there will be a

reversal (in the model's favor) in five years such companies are much more likely to offer

a DRIP. We interpret this as evidence that financial statements foreshadow changes in plan

status.

Our paper is therefore different in spirit from previous research on DRIPs. For

example, some researchers have studied the value of specific plan terms to investors. Two

key examples are Dammon and Spatt (1992), who calculate the value of an option implicit

in share-purchase terms of certain DRIPs, and Scholes and Wolfson (1989), who analyze

and report the success of their efforts to exploit a price discount provision incorporated in

other plans. Another strand of research studies stock prices of companies that announce

1For a thorough description of these plans and a discussion of why firms offer them, see DeGennaro (2003)and the references therein.

8/13/2019 SSRN-id1021070

3/39

- 2 -

plans (e.g. Dubofsky and Bierman (1988), Perumpral et al.(1991) and Dhillon et al.,

(1992)). Still others have explored the interaction of DRIPs and the tax code. Chang and

Nichols (1992), for example, investigate whether Internal Revenue Code Section 305(e)

affects qualifying utilities. They study the cost of equity capital, leverage ratios and stock

price reactions for firms offering DRIPs around the time of the changes in the tax code.

Todd and Domian (1997) conduct a survey to relate plan characteristics to shareholder

participation rates. To our knowledge, though, none has attempted to predict whether or

not a company will have a DRIP in the future.

These other lines of research are a perfect complement to our paper. We show that

predictions of plan changes are possible using financial statements; previous papers show

why our results are important. Our paper has useful managerial implications: Companies

that administer direct investment plans, such as The Bank of New York Mellon

Corporation, can identify firms most likely to be interested in becoming customers. The

results of Dubofsky and Bierman (1988), Perumpral et al. (1991) and Dhillon et al., (1992)

suggest that investors using our approach might be able to develop profitable trading

strategies. Thus, our paper has potential wealth effects for investors as well as immediate

managerial implications.

We find that companies that have DRIPs tend to have higher payout ratios, higher

returns on assets, higher earnings per share, more common shareholders, more common

shares outstanding, but fewer common shares traded. These financial variables are shown

to increase the likelihood that companies pay dividends and that they pay higher levels of

dividends. The more likely it is that a company pays a dividend and the larger the dividend

the greater the probability that the company will have a DRIP. In addition, we find that the

8/13/2019 SSRN-id1021070

4/39

- 3 -

exchange in which a companys shares are traded and the industry in which it operates

influence the likelihood that it will have a DRIP both indirectly through its impact on

dividends and directly on the probability of a company having a DRIP. This is the first

empirical evidence that dividend reinvestment plans might serve to insulate management

from outside control.

2. Data and Methodology

Data are from The Guide to Dividend Reinvestment Plans(1999, 2004) and the

Compustat and CRSP databases. We begin with 852 firms with available data in 1999 that

offered DRIPs. Because DRIP firms tend to be much larger in terms of total assets than

companies without such plans (DeGennaro, 2003), we match these 852 DRIP companies to

a sample of firms without such plans, for a total of 1704 companies. We use total assets in

1999 as our matching variable.

Because the dependent variable in our analysis is discrete (1 = company had a DRIP

at a particular time, 0 otherwise) Ordinary Least Squares regression is inappropriate for two

reasons. First, because the regression analysis is linear, it is quite possible to estimate

probabilities in the sample that are outside the (0,1) interval. In addition, the error terms in

such a regression would be heteroskedastic. To avoid these problems we use a maximum

likelihood logit model. This is the standard way to handle these problems. The logit model

has the following form:

PDRIP= 1/ (1 + e-Xii), (1)

where: PDRIP = the probability the firm has a DRIP,

Xi= a set of variables hypothesized to influence PDRIP,

8/13/2019 SSRN-id1021070

5/39

- 4 -

i= a set of coefficients which represent the estimated impact of X ion PDRIP,

Xii= 0+ X11+ X22+ + Xnn.

Classification methods have a long history of productive uses in business and

finance. These methods include models that use continuous variables as well as those that

use discrete choice variables. Discrete choice models are probably more common. One

form of discrete choice model is cluster analysis. Shaffer (1991), for example, studies

federal deposit insurance funding and considers its influence on taxpayers. Multinomial

logit, another discrete choice approach, has been used at least as far back as Holman and

Marley (in Luce and Suppes, 1965). The development of multinomial logit estimation for

use in the social sciences can be traced to the work of Dominic and McFadden (1975).

3. Sample Statistics

Table 1 presents the number of firms in each category partitioned by year and by

plan status. In 1999 we have a size-matched sample of 852 companies, one of each pair

offering a DRIP and one not offering a DRIP, for a total of 1704 companies. By 2004 the

sample has shrunk considerably for typical reasons. Mergers or other corporate

combinations along with delisting due to financial distress head the list. Only 916 firms

with useable data remain in the sample. Of these, 465 have a plan and 451 do not. Of

course, this masks movement across groups. Table 2 shows that most companies maintain

their plan status, either not offering a plan in either year (387 of 916) or retaining a plan

five years later (394 of 916). A moderate number do change their plan status, though. A

total of 71 companies, or about 7.75 percent, add a plan within the five years and 64 of the

916, or 6.99 percent, drop their plans.

8/13/2019 SSRN-id1021070

6/39

- 5 -

Factors Affecting the Likelihood of Having a DRIP

Table 3 lists the independent variables included in our analysis. It also shows

whether we expect companies that offer DRIPs to have higher or lower values in univariate

tests (left column) and multivariate tests such as our recursive logit model (right column).

Excluding total assets (the matching variable) and the categorical industry variables that we

include as controls, our variables fall into four categories. These rely on DeGennaro

(2003). We call variables in the first categoryfundamental variables, not in the sense of

fundamental economic value, but rather because their economic meanings are

fundamentally changed by a reinvestment plan. These are the payout ratio and the dividend

yield. Consider two companies which are identical except that one has a plan and one does

not. Suppose that the optimal dividend yield is 4 percent for both. The company without a

DRIP simply pays a 4 percent yield. The DRIP company, though, cannot expect plan

participants to retain all of their dividends. The DRIP company must offer a higher explicit

yield to have an effective yield of 4 percent. The same reasoning applies to the payout

ratio. All else equal, we expect DRIP companies to have higher explicit payout ratios and

dividend yields.

We call variables in the second group ability variablesbecause they control for the

ability to pay dividends all else equal, firms that earn more can pay more . These

variables are after-tax return on total assets (ROA) and earnings per share (EPS). Although

the ability to pay dividends may not be directly related to the probability of having a DRIP,

we expect the dividend yield itself to be a strong predictor of having a plan. These ability

variables offer something that the yield itself does not. Specifically, to the extent that the

ability to pay dividends affects the dividend yield in the future, these variables contain

8/13/2019 SSRN-id1021070

7/39

- 6 -

information about future dividends that is not captured by the current yield. Therefore, the

ability variables could provide indirect information about the likelihood of having a DRIP

in the future.

The third group controls for managerial entrenchment. DeGennaro (2003)

speculates that one reason for the existence of DRIPs is that they can insulate management

from threats to its control. Four variables fit this category: The number of common

shareholders, the number of common shares outstanding, the number of common shares

traded, and the number of employees. First, if management worries about retaining its

control, then it prefers a diffuse shareholder base with many small shareholders. Because

voting a proxy takes time, small shareholders are less likely to vote their shares. Second, as

long as small shareholders do not get a large enough position to become activist

shareholders, management wants them to have more voting shares. Third, our evidence

shows that these small shareholders who participate in DRIPs tend to trade less frequently.

Fourth, because employees jobs are often at risk during corporate reorganizations,

employees have incentives to support current management in takeover battles. The

anecdotal evidence in Rosen and Young (1991) supports the intuitive result that employees

probably tend to vote with management in takeover battles. This means that management

wants employees to be shareholders, too, so companies with many employees are more

likely to institute a DRIP (in fact, sometimes plans have features that are quite favorable to

employees and sometimes even restrict participation to employees). DRIP companies,

therefore, should tend to have higher values for all of these variables except for the number

of shares traded, which should be lower for DRIP firms.

8/13/2019 SSRN-id1021070

8/39

- 7 -

Finally, we control for the stocks trading venue. NASDAQ stocks are known to

have higher trading volumes because they are traded differently than on exchanges such as

the New York Stock Exchange (NYSE) or the American Stock Exchange (AMEX). This

could affect the estimated coefficient on the variable that controls for the number of

common shares traded. Tests (not reported for space reasons) show that the influence of

the NYSE and AMEX are comparable but differ from the NASDAQ and Other markets. In

our logistic regression we control forNASDAQand Otherwhile the NYSE and AMEX

serve as the excluded categories to prevent high collinearity.2 BecauseNASDAQand Other

firms tend to be younger we expect the coefficient on these categories to be negative.

Table 4 presents sample statistics and univariate tests. The first line for each

variable is for all companies (up to 1704 non-missing values). The second line presents

two numbers for each statistic; the first of each pair is for DRIP companies and the second

is for No-DRIP companies. As is to be expected, most observations have missing

observations for certain variables. Still, for the full sample of 1704 companies in 1999, we

have upwards of 1300 observations for all variables. Most have more than 1550

observations. Almost all observations on all variables lie within a reasonable range.

Exceptions occur for certain ratios with denominators near zero. For example, Compustat

defines the Payout Ratioas essentially the dollar amount of dividends paid to common

shareholders divided by earnings. Because earnings can be near zero, ratios can be large in

absolute value. Even these cases, though, are relatively rare. To overcome this problem in

our tests below we report median tests as well as t-tests.

2As a robustness check we experimented with a set of maturity variables comprising net sales, the debt ratio,the market-to-book ratio, and the price/earnings ratio (measured at fiscal year-end). These did not havestatistically reliable explanatory power.

8/13/2019 SSRN-id1021070

9/39

- 8 -

The second line for each variable lists the statistics for the DRIP sample and the No-

DRIP sample. A superscripted asterisk on the No-DRIP mean indicates that the values

differ at least at the 5 percent level according to a t-test and a superscripted pounds sign

means that they differ at least at the 5 percent level according to a two-sample medians test.

Ten t-tests are significant. Two of these are the fundamental variables, the Payout Ratio

and theDividend Yield. This is no surprise; after all, we call these fundamental variables

because the economic meaning of these variables is economically different for DRIP

companies compared to No-DRIP companies. As expected given Table 3, the DRIP values

are statistically and economically much higher. DRIP companies have an average payout

ratio that is almost three times higher than No-DRIP companies (almost 52 percent

compared to 18 percent). The average dividend yield for DRIP companies is about 2.5

times higher (3.5 to 1.4). The effective payout ratios and dividend yields of DRIP

companies, of course, are lower because some investors immediately reinvest a portion of

the dividends in the companies stock.

Both ability variables conform to the predictions of Table 3. After-Tax Return on

Assetsis significantly higher for DRIP firms than for No-DRIP firms according to both the

t-test and the medians test. The point estimate ofEarnings Per Shareis higher for DRIP

firms, and the difference is significant according to the medians test.

All four of the managerial entrenchment variables have the relationship predicted in

Table 3, and three of the four are statistically significant according to the t-tests. DRIP

companies have significantly more common shareholders, significantly fewer common

shares traded, and significantly more employees. These can help entrench management if

the additional shareholders tend to hold smaller stakes and trade less. These are well-

8/13/2019 SSRN-id1021070

10/39

- 9 -

known traits of DRIP investors. This result also supports Rosen and Youngs (1991)

suggestion that employees probably tend to vote with management in takeover battles.

The variables that control for the trading venue for the stocks are also statistically

different between DRIP and No-DRIP firms. Stocks which trade on NASDAQ or fit in the

Othercategory (that is, stocks not on NASDAQ, the New York Stock Exchange, or the

American Stock Exchange) are statistically less likely to have a DRIP.

Finally, DeGennaro (2003) shows that DRIP firms cluster by industry. To control

for this we include variables that control for SIC code. Table 4 contains sample statistics

for the following eight one-digit SIC categories: (1) Mining, oil production and

construction, (2) Materials processing, (3) Manufacturing, (4) Transportation utilities and

waste disposal, (5) Wholesale and retail activity, (6) Financial services, (7) Other services,

and (8) Other miscellaneous. It is important to note that the dummy variable that we

exclude from the regression, Other services, includes medical legal, social and

accounting services.3 We include these industry variables in our subsequent multivariate

analysis as control variables.

The medians tests generally agree with the t-tests. They differ for three variables

plus one industry control. Medians tests find a significant difference for earnings per share.

They fail to find significance in the number of common shares traded and the number of

employees. Among the industry controls, the t-test and medians test disagree only for

Manufacturing. Manufacturing firms have significantly higher incidences of DRIPS

according to the medians test.

3There are so few in this category that they would not have provided meaningful results.

8/13/2019 SSRN-id1021070

11/39

- 10 -

Table 5 contains Pearson correlations for the variables. Statistically significant

correlations are pervasive for most variables. The Payout Ratiohas the fewest significant

correlations, with two. Finding that many variables are highly correlated is not surprising

given that many variables are ratios which share either a numerator or denominator with

other variables. For example, after-tax earnings are in the denominator of the payout ratio

as well as in the numerator ofAfter-Tax ROA. This could produce multicollinearity in our

regressions, which would inflate the standard errors. Fortunately, this is a question of

power, not consistency. Our multivariate logit estimates will still be unbiased and

consistent.

4. Logit Results

Table 4 shows that certain firm-specific variables systematically differ between

DRIP firms and no-DRIP firms. DeGennaro (2003) shows that DRIP firms cluster by

industry. Based on this information it might seem as if a reasonable logit specification

would be:

0 1 2 3 4 5 6 7

18

8 9 10 11 12

i i i i i i i i

i i i i j i

DRIP TA PR DY ROA EPS CS CSout

CST Emp NASDAQ Other Industry

(1)

where the subscript isignifies the company and:

DRIPi= 1 if company ihas a DRIP;else 0

CSouti= number of common sharesoutstanding (000,000)

TAi= total assets ($000,000) CSTi= number of common shares traded

(000,000/yr.)PRi= payout ratio (%) Empi= number of employees (000)DYi= dividend yield (%) NASDAQi= unity if traded on NASDAQ; else

zeroROAi= after-tax return on total assets (%) Otheri= unity if not traded on NYSE, AMEX

or NASDAQ; else zeroEPSi= earnings per share ($) 22

16 jIndustry = seven one-digit SIC codes

8/13/2019 SSRN-id1021070

12/39

- 11 -

(see text)CSi= number of common shareholders(000)

i = logistically i.i.d.error term

This specification is problematic, though. Whether or not a firm pays a dividend

(i.e., has some positive dividend yield as opposed to having a dividend yield of zero) might

be expected to be a primary determinant of whether or not a firm would be expected to

have a DRIP. In addition, the remaining variables in equation (1) above might be expected

to be predictors of the likelihood that a firm pays a dividend and the magnitude of the

dividends they pay. Finally, the relationship between a firms dividend yield and the

likelihood of having a DRIP might be non-linear, that is, the difference between a yield of 0

percent and 0.1 percent is probably much more influential than the difference between, say,

2.0 percent and 2.1 percent.4 Consequently, for the purpose of this analysis we represent

the dividend yield as a categorical variable based on its distribution in our sample. Each

firm falls into one of four dividend yield categories: 1) zero dividend yield, 2) dividend

yield greater than zero, but less than or equal to 1.5 percent, 3) dividend yield greater than

1.5 percent, but less than or equal to 3.3 percent, and 4) dividend yield greater than 3.3

percent. We choose these three categories of dividend yield to allocate an equal number of

the dividend-paying stocks to each category. Our model then is specified to be recursive in

nature. Specifically, initially the probability of a firm being in one of the dividend yield

categories is estimated to be a function of the variables other than dividend yield in

equation (1) above. Subsequently, the likelihood of a firm having a DRIP is estimated to be

4Fourteen companies had a DRIP in 1999, dropped it by 2004 and did not pay a dividend in 2004. Forcomparison, only one company that had a zero yield in 1999 dropped its plan and had a positive yield in 2004.Three companies had a zero yield in 1999, added a plan and had a positive yield in 2004. Finally, twocompanies had a positive yield in 1999, added plan and had a zero yield in 2004.

8/13/2019 SSRN-id1021070

13/39

- 12 -

a function of their dividend yield category and a subset of the financial variables in

equation (1) above.5 A recursive structure implies that one decision precedes another. In

this case, it seems reasonable that a firms managers would only consider instituting a

DRIP if the company were paying dividends, or had made a decision to do so. For an

excellent discussion of recursive models with qualitative endogenous variables, see

Maddala and Lee (1976).

This structure proves to predict whether or not firms have plans quite well and also

allows us to compute the marginal influence of variables on the decision. Formally, our

logit model that predicts whether a company offers a DRIP is:

0 1 2 3 4

12

5 6,

L M H

i i i i i

i j i i

DRIP DIV DIV DIV NASDAQ

Other Industry

where:

L

iDIV = 1 if 0 < dividend yield of stock i1.5%, else 0;M

iDIV = 1 if 1.5% < dividend yield of stock i3.3%, else 0;

HiDIV = 1 if 3.3% < dividend yield of stock i, else 0;

iNASDAQ = 1 if stock itrades on NASDAQ, else 0;

iOther=1 if stock itrades does not trade on NASDAQ, NYSE or AMEX, else 0;

iIndustry = 1 if stock iis a member of Mining, Oil and Construction; Materials

Processing; Manufacturing; Transportation, Utilities and Waste Disposal;Wholesale and Retail; Financial Services; Other Miscellaneous;respectively, else 0.

The second part of the recursive structure models the three categories of dividend

yield. This logit model writes the dividend yield as a function of the variables in Table 3.

The model is:

5Initially, all variables were included in the estimation and only those were retained that were demonstratedto have a statistically significant direct impact on DRIP when the dividend yield variables were included inthe equation.

8/13/2019 SSRN-id1021070

14/39

- 13 -

0 1 2 3 4 5 6

16

7 8 9 10

i i i i i i i

i i i j i

DividendYield PR ROA EPS CS CSout CST

Emp NASDAQ Other Industry

Intuitively, the recursive structure models the decision to offer a DRIP as a function

of variables including the dividend yield category; the dividend yield category is in turn

modeled as a function of the fundamental variables, ability variables, managerial

entrenchment variables, the stocks trading venue, and industry effects.

Companies with a dividend yield of zero comprise the excluded category for the

logit model. The first column of Table 6A contains the results using 1999 data. The

middle column contains the logit coefficient estimate and the third column contains

asymptotic t-statistics. This portion of the recursive model produces anR2of over 50

percent and the coefficients on the dividend yield categories and exchange have the

predicted signs (we have no predicted signs for industry variables). In addition, the higher

the dividend yield category, the larger the coefficient. Stocks that pay higher dividends are

more likely to have DRIPs. The gradual increase in the size of the coefficients as the

magnitude of the dividend yield increases is exactly what one would expect (3.103, 3.583,

and 4.334). Three of the seven industry effects are significant at least at the 5 percent level

and another is significant at the 10 percent level. Both exchange variables have negative

signs and Otheris significant at the 1% level.

Table 6B reports the results of the model that predicts the dividend yield category

for each stock. Again, stocks with a dividend yield of zero comprise the excluded category.

The model produces anR2of almost 48%. Excluding Industry Effect variables, all

variables are significant at least at the 5 percent level for at least one dividend category. All

8/13/2019 SSRN-id1021070

15/39

- 14 -

except the number of employees are significant for at least two dividend categories.

Overall, 22 of 27 coefficients are significant at least at the 5 percent level. Because all

dividend categories in Table 6A have positive coefficients, the coefficients in Table 6B

should have the signs predicted in Table 3. For example, the Payout Ratio has a positive

influence on the dividend yield, which in turn has a positive influence on the likelihood of

having a DRIP. This means that the Payout Ratio has a positive influence on the likelihood

of having a DRIP. Table 6B shows that all but one of the 27 coefficients have the predicted

sign.

How can we best interpret these results? Analyzing each group of variables is a

good starting point. All four of the variable groups conform quite well to our predictions in

Table 3. For example, the Payout Ratio is correctly signed for all three categories and the

statistical reliability is very strong in the two higher dividend yield categories (we address

economic significance in Table 7). Five of the six coefficients on the ability variables are

correctly signed, with EPS for the highest dividend yield the only coefficient signed

incorrectly in the entire table. The single incorrect coefficient is economically small and

statistically insignificant. A possible explanation is that EPS is highly correlated with

another variable or combination of variables, making it difficult to separate their

contributions to explaining variation. All nine of the management entrenchment variables

dealing with ownership (the number of common shareholders, the number of common

shares outstanding and the number of common shares traded) are correctly signed and

significant, most at well beyond the one-percent critical value. The fourth variable in this

category, the number of employees, is correctly signed for all three dividend yield

categories and is significant for the highest dividend yield category. The Exchange

8/13/2019 SSRN-id1021070

16/39

- 15 -

variables are always correctly signed and statistically significant. Companies that trade on

NASDAQ or Other (not on NASDAQ, NYSE or AMEX) are less likely to have DRIPs.

The recursive model in Tables 6A and 6B correctly classifies companies with

DRIPs slightly better than companies without them. For DRIP firms the correct

classification rate is 69.2 percent (435 out of 629) and for No-DRIP firms the rate is 75.5

percent (388 out of 514). Overall, the rate of correct classifications is 72.0 percent (823 out

of 1143).

Which Groups of Independent Variables Drive the Result?

Table 7 uses the recursive model in Tables 6A and 6B and 1999 data to present two

other ways to gain insight about the implications of our model. Panel A reports results for

companies with no DRIP in 1999 and Panel B reports results for companies with a DRIP in

1999. The top part of each panel reports the mean predicted probability of having a DRIP

for companies that had no plan five years later, in 2004 (second column), and for

companies that did have one in 2004 (third column).

The 464 companies that did not have a DRIP in 2004 had a predicted mean

probability of having a DRIP of 30.06 percent, whereas the 50 companies that did have a

DRIP in 2004 had a mean probability of having a DRIP of 56.25 percent. The t-statistic for

the difference in these means is 3.84. and the difference is also an economically significant

26.2 percentage points. This means that on average, misclassified firms are much closer to

being correctly classified than the correct classifications are to being misclassified. This is

evidence that the model is picking up factors in the financial statements that foreshadow the

change in DRIP status.

8/13/2019 SSRN-id1021070

17/39

- 16 -

The second parts of Panel A and Panel B report the contribution of each of the five

variable subgroups (fundamental, ability, managerial entrenchment, exchange, and industry

effects) to the difference between those mean predicted probabilities. This difference in

these predicted probabilities of having a DRIP derives from the differences in mean values

for the independent variables rather than the estimated coefficients in the model. For

example, given the positive coefficient on the payout ratio, a higher payout ratio implies

that there is a higher probability of having a DRIP. Thus we can calculate the impact of

each variable group on this difference between the 30.06 percent average likelihood of not

having a DRIP and the 56.25 percent average likelihood of having a DRIP. In the second

part of Panel A in Table 7,we see that differences in Managerial Entrenchment variables

(11.49 percentage points) are the largest component of this differential. This implies that if

the mean values for the Fundamental variables for No-DRIP firms in 2004 are increased to

the level of firms that have a DRIP in 2004, their predicted likelihood of having a DRIP

increases by 11.49 percentage points. This would increase the predicted probability of

having a DRIP for this group by more than a third (an 11.49 percentage point increase over

the 30.06 percent probability for firms adding a DRIP by 2004). The changes associated

with the other classes of variables are smaller although still important, ranging from 3.37

percentage points for the Exchange variables to 5.64 percentage points for the Industry

Effects variables.

The results for firms that did have a DRIP in 1999 are in Panel B of Table 7. The

mean predicted probability of having a DRIP is lower for firms that drop their DRIP

(second column) compared to those that retain a DRIP in 2004 (third column), although the

difference is statistically insignificant. The difference is 63.86 percent versus 67.91

8/13/2019 SSRN-id1021070

18/39

- 17 -

percent, with a t-statistic of 0.14. Although the recursive model misclassifies almost a

quarter of the DRIP companies in 2004 using 1999 data (150 out of 629), those companies

do have a lower likelihood of having a DRIP.

As is true for the companies that have a DRIP in 1999, the variables controlling for

Managerial Entrenchment are the most important, although the contributions are much

smaller. The 1.65 percentage point influence accounts for 40.72 percent of the difference

between the two means. The second most influential set of variables are those that control

for the ability to pay a dividend, which accounts for 1.34 percentage points, or 33.09

percent, of the differential.

The importance of managerial entrenchment variables in the decision to add or drop

a DRIP merits mention because entrenched management has been linked to lower firm

value. Bebchuk and Cohen (2005), for example, show that staggered boards (a common

governance arrangement that insulates managers from dismissal) are associated with lower

corporate values. Ryngaert (1988) finds similar (though weaker) results for poison pills.

Futureresearch would do well to explore whether companies that institute a DRIP to

entrench management suffer stock price declines while those that do so for other reasons do

not. This might explain the conflicting evidence researchers have found concerning the

stock price reaction around the announcement that a company will institute a DRIP. For

example, Dubofsky and Bierman (1988) and Perumpral et al.(1991) find evidence of

positive abnormal returns when companies announce that they will institute a DRIP while

Dhillon et al.(1992) find evidence of losses. Peterson et al.(1987) find mixed results.

Perhaps these latter studies had higher proportions of firms that instituted plans for reasons

8/13/2019 SSRN-id1021070

19/39

- 18 -

of corporate control. Estimating abnormal returns after controlling for the reason for

instituting the DRIP is likely to be fruitful.

Our analysis demonstrates that a firms payout ratio, its ability to pay dividends, the

degree of a firms managerial entrenchment, the trading venue of a firms equity shares, and

the industry in which it operates influence whether or not dividends are paid and the

magnitude of those dividends. Thus, they indirectly influence the likelihood that a firm has

a DRIP. In the case of the firms stock trading venue and industry there is also a direct

impact on the likelihood that a firm will have a DRIP holding dividends constant. Given

these results, the question that remains is how well can we predict changes in plan status?

Predicting Changes in DRIP Status

Consider companies that the logistic model misclassifies; either the model says that

they should have a plan and they do not, or it says that they should not have a plan and they

do. How do we interpret this? One way is to conclude that the model simply fails in such

cases. An alternative interpretation is that financial statements contain information about

futureplan status as well as currentplan status. Under this interpretation, misclassified

companies are more likely to switch their plan status. That is, if the firms financial data

suggest that a company should have had a dividend reinvestment plan in 1999 but it did

not, then we expect that it would be more likely to institute a plan than the other companies

in the sample. Conversely, if it did have a plan but the financial data suggest that it should

not, then we expect that the company would be more likely to drop the plan. Put

differently, misclassifications in 1999 include predictions of changes in plan status as well

as classification errors.

8/13/2019 SSRN-id1021070

20/39

- 19 -

Do the data support this interpretation? For No-DRIP companies the answer is yes.

Figure 1 illustrates the result. A total of 641 companies have data available in both 1999

and 2004. Figure 1 shows the transition matrix of these 641 firms. First, consider the left

side of Figure 1.

Our model uses data from financial statements available in 1999 to classify 641

firms that survived and have data through 2004. These include 288 No-DRIP firms and

353 DRIP firms in 1999. Of these, a total of 438 (203 No-DRIP firms plus 235 DRIP

firms) are classified correctly and 203 (118 DRIP firms plus 85 No-DRIP firms) are

classified incorrectly, for an overall classification accuracy of 68.33 percent, which is

almost exactly the same as the classification rate of 66.50 percent for all firms in 1999.

One interpretation of this is that the model has simply made 203 mistakes, putting No-

DRIP firms in the DRIP category and vice versa. Another interpretation, though, is that the

1999 financial statements of these companies already foreshadow changes in the

companies characteristics changes that would place the firms in the other DRIP category

by 2004. This would be the case if some firms gradually adopt the economic traits that

characterize the other class and if the financial statements reflect this drift before

management decides to change plan status and implements the decision.

The right side of Figure 1 shows that this interpretation has merit. For No-DRIP

firms, 21.2 percent of those predicted to have a DRIP in 1999 establish one by 2004 while

only 9.9 percent of firms correctly classified as No-DRIP in 1999 establish a DRIP by

2004. This means that if the model incorrectly says that a company looks like a DRIP firm,

then that firm is more than twice as likely to establish a DRIP as a company that was

8/13/2019 SSRN-id1021070

21/39

- 20 -

correctly classified as a No-DRIP firm. Apparently, the firms financial statements begin to

foreshadow the transition to a DRIP firm before management institutes a plan.

For DRIP firms, the model correctly classifies 235 of the 353 surviving DRIP firms,

leaving 118 misclassifications in 1999, for a success rate of 66.6 percent. Almost 6 percent

of the incorrectly classified DRIP firms discontinue their DRIP programs by 2004 while

10.6 percent of the correct predictions on DRIP firms in 1999 no longer have a DRIP by

2004. This means that if the model incorrectly says that a company looks like a No-DRIP

firm, then it is less likely to eliminate its DRIP than a company that was correctly classified

as a DRIP firm. The model cannot predict which companies will eliminate DRIPS.

Overall, 25 of 203 (12.3 percent) misclassifications in 1999 self-correct by 2004,

which is two percentage points higher than the percentage of classifications that go bad.

Only 45 of 438 (10.3 percent) of correct classifications go bad during the same period.

Figure 1 shows that if our model makes a correct prediction in 1999 then the chances are

quite high that the prediction will hold for five years. If our model makes an incorrect

prediction, though, the odds are much higher that there will be a reversal (in the model's

favor) in five years. Financial statements apparently do foreshadow changes in plan status.

The ability to predict changes in plan status is concentrated entirely on companies that add

DRIPs.

To investigate this formally we conduct a two-pronged experiment. In the first part,

we consider the companies that do not have a DRIP in 1999 and explore whether they add

one by 2004. In the second part, we consider companies that have a DRIP in 1999 and

explore whether they drop it by 2004. These two groups of firms should have different

financial characteristics.

8/13/2019 SSRN-id1021070

22/39

- 21 -

In the first part of this experiment we partition the 288 surviving companies that did

not have a DRIP in 1999 into those that our model correctly classifies and those that it

misclassifies. We compute the proportion of each group that added a plan by 2004 and test

the difference between them. If the 1999 financial statements contain information about

future plan status, then the companies that the model misclassifies as having a plan in 1999

should add plans significantly more often. To perform the second part of this experiment

we partition the 353 surviving companies that had a DRIP in 1999 into those that the model

correctly classifies and those that it misclassifies. We then compute the proportion of each

group that dropped their plans by 2004 and test the difference between them. If the 1999

financial statements contain information about future plan status, then the companies that

the model misclassifies as not having a plan in 1999 should drop their plans significantly

more often.

Table 8 reports the results of this experiment. If our model does a good job of

prediction, then we would expect that firms that actually had a DRIP in 2004 would be

more likely to be predicted to have a DRIP by our model. That clearly is the case for the

No-DRIP sub-sample (Panel A). Specifically, 18 of the 38 companies that did not have a

DRIP in 1999 but were predicted to have one initiated a DRIP by 2004 (47.37%); while

only 67 of 250 firms that did not have a DRIP by 2004 were incorrectly predicted to have

one (26.80%). This difference in proportions is statistically significant (t-ratio = -2.59). In

contrast, Panel B shows that 25 of the 32 firms that the model predicted to have a DRIP did

not have one in 2004 even though they had a DRIP in 1999 (78.13 percent); whereas 210 of

321 firms that that had a DRIP in 1999 and 2004 were predicted to have a DRIP (65.42

percent). This means that the model fails to improve predictions of which firms will drop

8/13/2019 SSRN-id1021070

23/39

- 22 -

their DRIPs. Our model obviously does a much better job of differentiating between those

firms that might be expected to establish a DRIP program versus those that would be

predicted to drop an existing DRIP.

These results have an immediate managerial implication. Suppose that a firm which

manages DRIPs wants to solicit new business. Can our model help? The answer is yes.

To demonstrate this, suppose that we identify a large number of firms that currently do not

have a DRIP program, but which (because of their financial position) we expect to survive

for up to five years. We would like to identify potential clients because such solicitation is

a costly process. If we were to contact allof the 288 firms we identified that currently do

not have a DRIP, we would find that only about 38 (13.2 percent) would establish a DRIP

over the next 5 years. However, if we used our model to determine which firms to target,

we would contact only 85 firms (about 29.5 percent) of the 288 No-DRIP firms. Of those

85 firms, 18 (21.1 percent) would actually add a DRIP. These 18 firms would represent

about 47.4 percent (18 of 38) of the entire population of No-DRIP firms that might be

expected to actually establish a DRIP program over the next several five years. In short, we

would uncover almost half of the companies that do eventually adopt a DRIP despite

contacting fewer than 30 percent of the likely customers. This would represent a huge

reduction in search costs and seems to be an effective way to minimize marketing expenses.

5. Summary

Most research on dividend reinvestment plans has focused on stock-price responses

to the announcements of such plans or has attempted to value certain plan features. To

date, little research has attempted to determine which type of firm adopts a plan, and none

8/13/2019 SSRN-id1021070

24/39

- 23 -

has attempted to predict whether companies will have a plan in the future. This paper fills

that void. We use discrete choice methods to predict the classification of 1704 companies:

A sample of 852 companies with dividend reinvestment plans in 1999 matched by total

assets to a sample of 852 companies without such plans. We develop a logit model that

successfully classifies over 72 percent (823 of 1143) of these companies yet uses only

readily available contemporaneous financial data. Our analysis demonstrates that a firms

Payout Ratio, a set of variables measuring the firms ability to pay dividends, a set of

variables measuring the degree of managerial entrenchment for a given firm, the trading

venue of a firms equity, and the industry in which the firm operates play a role in

determining whether or not the firm has a DRIP. The channel of influence is through

dividends -- these variables influence whether the firm pays a dividend and the magnitude

of the dividend paid (Table 6B). In turn, these categorical dividend-yield variables are of

primary importance in explaining the presence of a DRIP (Table 6A). In addition to these

indirect effects through variables that control for the dividend yield, variables that control

for the trading venue of the firms equity and industry differentials have a significant direct

influence on the likelihood that a firm has a DRIP as well.

In addition, by interpreting the misclassified companies as those being likely to

switch their plan status, we can then test whether the model can predict changes in plan

status. The underlying premise is that a companys current financial data contain

information not only about whether a company currently has a dividend reinvestment plan,

but also that they contain information about future plan status. We use data from 1999 and

2004 to explore this conjecture. We find that our model can predict plan adoption over the

8/13/2019 SSRN-id1021070

25/39

- 24 -

next five years. This implies that financial statements contain information about changes in

plan status at least five year in the future.

8/13/2019 SSRN-id1021070

26/39

- 25 -

References

Chang, Otto H. and Donald R. Nichols. 1992. Tax Incentives and Capital Structures: The

Case of the Dividend Reinvestment Plan.Journal of Accounting Research30: 109-125.

Dammon, Robert M. and Chester S. Spatt. 1992. An Option-Theoretic Approach to theValuation of Dividend Reinvestment and Voluntary Purchase Plans. The Journal ofFinance47: 331-347.

DeGennaro, Ramon P. 2003. Direct Investments: A Primer.Economic Review88: 1-14.

Dhillon, Upinder S., Dennis J. Lasser and Gabriel G. Ramirez. 1992. DividendReinvestment Plans: An Empirical Analysis.Review of Quantitative Finance and

Accounting : 205-213.

Dubofsky, D. and Bierman, L. 1988. The Effect of Discount Dividend Reinvestment PlanAnnouncements on Equity Value.Akron Business and Economic Review 1 :58-68.

Dominic, T. and McFadden, D. 1975. Urban Travel Demand: A Behavioral Analysis. InContributions to Economic AnalysisJ. Tinbergen, D. W. Jorgenson, and J. Waelbroeck(ed.) North-Holland / American Elsevier: New York.

Guide to Dividend Reinvestment Plans (1999). Temper of the Times Communications, Inc.New York.

Guide to Dividend Reinvestment Plans (2004). Temper of the Times Communications, Inc.New York.

Luce, R. D. and Suppes, P. 1965. Preference, Utility and Subjective Probability. InHandbook of Mathematical PsychologyR. D. Luce, R. R. Bush, and E. Galanter (ed).J. Wiley and Sons: New York.

Maddala G.S. and Lung-Fei Lee. 1976. Recursive Models with Qualitative EndogenousVariables. Annals of Economic and Social Measurement 5: 525-546.

Perumpral, Shalini, Arthur J. Keown and John Pinkerton. 1991. Market Reaction to theFormulation of Automatic Dividend Reinvestment Plans.Review of Business andEconomic Research 26 :48-58.

Peterson, Pamela P., David R. Peterson and Norman H. Moore. 1987. The Adoption ofNew-Issue Dividend Reinvestment Plans and Shareholder Wealth.TheFinancialReview22 : 221-232.

8/13/2019 SSRN-id1021070

27/39

- 26 -

Rosen, Corey M. and Karen M. Young. Understanding employee ownership. CornellUniversity, Ithaca, NY. 1991.

Ryngaert, Michael. 1988. The Effect of Poison Pill Securities on Shareholder Wealth.Journal of Financial Economics 20: 377-417.

Scholes, Myron S. and Mark A. Wolfson. 1989. Decentralized Investment Banking: theCase of Discount Dividend-Reinvestment and Stock-Purchase Plans.Journal ofFinancial Economics 23: 7-35.

Shaffer, Sherrill. 1991. Aggregate Deposit Insurance Funding and Taxpayer Bailouts.Journal of Banking and Finance 15 : 1019-1037.

Smith, Clifford W., Jr. and Ross L. Watts. 1992. The Investment Opportunity Set andCorporate Financing, Dividend, and Compensation Policies.Journal of FinancialEconomics 32 : 263-292.

Todd, Janet M. and Dale L. Domian. 1997. Participations Rates of Dividend ReinvestmentPlans: Differences Between Utility and Nonutility Firms.Review of FinancialEconomics 6: 121-135.

8/13/2019 SSRN-id1021070

28/39

- 27 -

288 No-DRIP firms

641 firms with data for the logit model in both years

203 correct

classifications

85 incorrect

classifications

Status in 2004

183 correct

classifications

Status in 1999

20 incorrect classifications

67 incorrect classifications

18 correct classifications

Establish DRIP

(Go bad): 9.9%

Establish DRIP

(Self-correcting): 21.2%

353 DRIP firms

235 correct

classifications

118 incorrect

classifications

Eliminate DRIP

(Go bad): 10.6%

Eliminate DRIP

(Self-correcting): 5.9%

210 correct

classifications

25 incorrect classifications

118 incorrect classifications7 correct classifications

Figure 1

Apparent model error, or forecasting?

Among firms with complete data for the logit model, a total of 288 firms do not have a

DRIP in 1999 and survive through 2004. Of these, the model correctly classifies 203 firmsas not having a DRIP in 1999. Only 9.9 percent of these add a DRIP by 2004. In contrast,the model incorrectly classifies 85 firms as having a DRIP in 1999. Of these, 21.2 percentestablish a DRIP by 2004. This means that if the model incorrectly says that a companylooks like a DRIP firm, then that firm is more than twice as likely to add a DRIP as acompany that was correctly classified as a No-DRIP firm.

Among firms with complete data for the logit model, a total of 353 firms have a DRIP in1999 and survive through 2004. Of these, the model correctly classifies 235 firms ashaving a DRIP in 1999. Only 10.6 percent of these eliminate their DRIP by 2004. Incontrast, the model incorrectly classifies 118 firms as not having a DRIP in 1999. Of these,5.9 percent eliminate their DRIP by 2004. This means that if the model incorrectly saysthat a company looks like a No-DRIP firm, then that firm is less likely to eliminate itsDRIP than a company that was correctly classified as a DRIP firm.

8/13/2019 SSRN-id1021070

29/39

- 28 -

Table 1

Number of firms by year and plan status

Number of Companies

Year

with Dividend

Reinvestment Plans

without Dividend

Reinvestment Plans

Total

1999 852 852 1704

2004 465 451 916

Table 2

Plan status of the 916 surviving firms in 1999 and 2004

Plan Status in

Years 1999 and 2004

Number of

companies

Percent of

total surviving

firms

Neither 1999 nor 2004 387 42.25%

Not 1999 but 2004 71 7.75%

1999 but not 2004 64 6.99%

Both 1999 and 2004 394 43.01%

Total Surviving Companies 916 100%

8/13/2019 SSRN-id1021070

30/39

- 29 -

Table 3

Expected relation between financial statement data for companies with DRIPs

compared to companies without DRIPs in univariate and multivariate tests

Variable

Expected relation between variable and

the likelihood of having a DRIP

Univariate Tests Multivariate Tests

TotalAssets N/A (matching variable)

Fundamental Variables

Payout Ratio Higher Higher

Dividend Yield Higher Higher

Ability Variables

After Tax Return on Total Assets Higher Higher

Earnings Per Share Higher Higher

Managerial Entrenchment Variables

Number of Common Shareholders Higher Higher

Number of Common Shares Outstanding Higher Higher

Number of Common Shares Traded Lower Lower

Number of Employees Higher Higher

Trading Venue Variables

NASDAQ Lower Lower

Other Lower Lower

8/13/2019 SSRN-id1021070

31/39

- 30 -

Table 4Sample Statistics, 1999 Data, 1704 Firms in Operation in 1999

Variable N Mean

TotalAssets ($MM) Full Sample 1703 14,648

DRIP,No DRIP 852 851 14,353 14,944 45,27Fundamental Variables

Payout Ratio (%) Full Sample 1649 35.02

DRIP,No DRIP 827 822 51.81 18.12*# 112.7Dividend Yield (%) Full Sample 1576 2.55

DRIP,No DRIP 851 725 3.53 1.40*# 3.60

Ability VariablesAfter Tax Return on Total Assets (%) Full Sample 1704 2.65

DRIP,No DRIP 852 852 3.87 1.43*# 7.21Earnings Per Share ($) Full Sample 1621 1.69

DRIP,No DRIP 851 770 1.73 1.64# 1.97

Managerial Entrenchment Variables

Common Shareholders (000) Full Sample 1303 38.14 DRIP,No DRIP 666 637 51.44 24.23*# 202.2

Common Shares Outstanding (000,000)Full Sample

1657 198.80

DRIP,No DRIP 849 808 211.86 185.09 507.8Common Shares Traded (000,000/yr.)Full Sample 1574 168.98

DRIP,No DRIP 852 722 144.93 197.36* 389.4Employees (000) Full Sample 1526 21.91

DRIP,No DRIP 771 755 24.69 19.07* 66.14

Exchange VariablesNASDAQ Full Sample 1704 0.26

DRIP,No DRIP 852 852 0.23 0.30*# 0.42Other Full Sample 1704 0.09

DRIP,No DRIP 852 852 0.00 0.18*# 0.05

Industry Effects Variables Mining, Oil, & Construction Full Sample 1704 0.03

DRIP,No DRIP 852 852 0.03 0.04 0.16

Table continues on the next page

8/13/2019 SSRN-id1021070

32/39

- 31 -

Continued from the previous page

Materials Processing Full Sample 1704 0.15

DRIP,No DRIP 852 852 0.19 0.12*# 0.39Manufacturing Full Sample 1704 0.17

DRIP,No DRIP 852 852 0.18 0.15#

0.39Transportation, Utilities, and Waste Disposal

Full Sample1704 0.17

DRIP,No DRIP 852 852 0.17 0.17 0.38Wholesale and Retail Full Sample 1704 0.07

DRIP,No DRIP 852 852 0.07 0.07 0.25Financial Services Full Sample 1704 0.31

DRIP,No DRIP 852 852 0.31 0.31 0.46Other Services Full Sample 1704 0.06

DRIP,No DRIP 852 852 0.02 0.10*# 0.16Other Miscellaneous Full Sample 1704 0.01

DRIP,No DRIP 852 852 0.01 0.01 0.08

The top line of each variable contains values for the full sample of DRIP and No-DRIP firms. The second line The first of the pair pertains to DRIP firms and the second to No-DRIP firms.*indicates that the difference between the DRIP and No-DRIP means is significant at least at the 5% level using a t-test.#indicates that the difference between the DRIP and No-DRIP medians is significant at least at the 5% level using a two-s

8/13/2019 SSRN-id1021070

33/39

- 32 -

Table 5

Pearson Correlations

TotalAssets

PayoutRatio

DividendYield

After-TaxROA

EPS# Common

Shareholders

# CommonShares

Outstanding

# CommonSharesTraded

Number ofEmployees NAS

TotalAssets 1

Payout Ratio 0.010 1

DividendYield

-0.039 0.138* 1

After Tax ROA -0.008 0.032 0.063* 1

EPS 0.046 0.010 0.011 0.156* 1

# CommonShareholders

0.260* 0.021 -0.016 0.058* 0.021 1

# CommonShares

Outstanding0.373* 0.014 -0.010* 0.167* -0.006 0.528* 1

# CommonSharesTraded

0.171* -0.023 -0.133* 0.155* -0.011 0.285* 0.668* 1

# Employees 0.308* 0.006 -0.078* 0.093* 0.013 0.300* 0.482* 0.247* 1

NASDAQ -0.140* -0.035 -0.127* -0.140* -0.068* -0.079* -0.115* 0.046 -0.148*

Other 0.097* -0.051* -0.101* -0.138* 0.134* -0.035 0.000 -0.028 0.035 -0.1

*indicates significance at the 5% level.

8/13/2019 SSRN-id1021070

34/39

- 33 -

Table 6A12

0 1 2 3 4 5 6,L M Hi i i i i i j i i

DRIP DIV DIV DIV NASDAQ Other Industry Logit Coefficients and

Asymptot ic t-statistics

Pr(Drip=1)1999 Data

Variable Coefficient t-statistic

Intercept -3.061 -8.347***

Fundamental Variables

0< Dividend Yield.

8/13/2019 SSRN-id1021070

35/39

- 34 -

Table 6B

0 1 2 3 4 5

16

6 7 8 9 10

i i i i i

i i i i j i

DividendYield PR ROA EPS CS CSout

CST Emp NASDAQ Other Industry

Logit Coefficients and Asymptotic t-statistics

Excluded Category: Dividend Yield = 0

1999 Data

0< Div. Yld.

8/13/2019 SSRN-id1021070

36/39

- 35 -

Industry Effects VariablesMining, Oil, and Construction 0.944 1.916* 1.151 1.909* 0.917

Materials Processing 1.001 2.621*** 2.238 4.859*** 1.474

Manufacturing 0.779 2.148** 2.033 4.543*** 1.340

Transportation, Utilities,

and Waste Disposal

0.067 0.154 0.962 1.920* 2.672

Wholesale and Retail 0.071 0.171 0.574 1.111 0.523

Financial Services 1.631 4.185*** 2.447 5.180*** 3.163

Other Miscellaneous 1.665 0.957 2.941 1.732* 2.245

Nagelkerke Pseudo R2 0.478

Number of observations 1,143

8/13/2019 SSRN-id1021070

37/39

- 36 -

Table 7Mean Probabilities of DRIP

and

Impact of Variable Group Changes

1999 DataPanel A: Companies with No DRIP in 1999

No DRIP in 2004 DRIP in 2004

Mean Predicted Probability ofhaving a DRIP

30.06% 56.25%

Number of Observations 464 50

t-statistic for Difference in Means 3.84***

Change in No DRIP versus DRIP in 2004

Pr(DRIP in '04 | No DRIP in '99)

Percentage Points % of Total

Payout Ratio 4.88 18.64%

Ability Variables 4.04 15.41%

Managerial Entrenchment Variables 11.49 43.86%

Exchange Effects Variables 3.37 21.54%

Industry Effects Variables 5.64 12.87%

Panel B: Companies with DRIP in 1999

No DRIP in 2004 DRIP in 2004

Mean Predicted Probability ofhaving a DRIP

63.86% 67.91%

Number of Observations 150 479

t-statistic for Difference in Means 0.14

Change in No DRIP versus DRIP in 2004Pr(DRIP in '04 | DRIP in '99)

Percentage Points % of Total

Payout Ratio 0.76 18.82%

Ability Variables 1.34 33.09%

Managerial Entrenchment Variables 1.65 40.72%

Exchange Effects Variables 1.24 30.61%

Industry Effects Variables -0.44 -10.87%

the individual percentage point changes do not sum exactly to the total probability differential becauseof the non-linear (logit) form of the probability calculations.*** indicates significance at the 1% level.

Panel A reports results for companies with no DRIP in 1999 and Panel B reports results for companies with aDRIP in 1999. The top part of each panel reports the mean predicted probability of having a DRIP forcompanies that had no DRIP in 2004 (second column) and for companies that did have one in 2004 (thirdcolumn). The second part of each panel reports the contribution of each of the five variable subgroups(fundamental, ability, managerial entrenchment, exchange and industry effects) to the difference in the meanpredicted probabilities.

8/13/2019 SSRN-id1021070

38/39

- 37 -

Table 8

Transition Matrix: Using 1999 Data to Predict DRIP Status in 2004Panel A: Companies that Did Not Have a DRIP in 1999

Total Firms

with data in2004

Firms with

Data that HadDRIP in 2004

Firms with Data

that Had NoDRIP in 2004

Non-existentfirms in 2004 TotalFirms

Number Predictedto Have a DRIP

85 18 67 41 126

ProportionPredicted toHave a DRIP

47.37% 26.80%

Number PredictedNotto Have a DRIP

203 20 183 185 388

ProportionPredicted to NotHave a DRIP

52.63% 73.20%

Total Number 288 38 250 226 514

t-statistic for theDifference inProportions

-2.59**

Panel B: Companies that Had a DRIP in 1999

Total Firmswith data in

2004

Firms withData that HadDRIP in 2004

Firms with Datathat Had No

DRIP in 2004

Non-existentfirms in 2004

TotalFirms

Number Predictedto Have a DRIP

235 210 25 200 435

ProportionPredicted toHave a DRIP

65.42% 78.13%

Number PredictedNotto Have a DRIP

118 111 7 76 194

ProportionPredicted to NotHave a DRIP

34.58% 21.88%

Total Number 353 321 32 276 629

t-statistic for theDifference inProportions

1.453

Panel A reports the transition data for 288 companies that did not have a DRIP in 1999. Of these, 85 wereincorrectly predicted by the logit model to have a DRIP, while 203 were correctly predicted not to have aDRIP. By 2004, a significantly higher proportion (t-statistic of -2.59) of firms that had a DRIP in 2004 werepredicted to have one (47.37% versus 26.80%) as compared to firms that had no DRIP in 2004.

Specifically, 18 of the 38 companies that did not have a DRIP in 1999 but were predicted to have one initiateda DRIP by 2004 (47.37%); while only 67 of 250 firms that did not have a DRIP by 2004 were incorrectlypredicted to have one (26.80%). This difference in proportions is statistically significant (t-statistic = -2.59).

8/13/2019 SSRN-id1021070

39/39

Panel B reports the transition data for 353 companies that had a DRIP in 1999. Of these, 235 were predictedby the logit model to have a DRIP and 118 firms were predicted not to have a DRIP. By 2004, aninsignificantly lower proportion (t-statistic of 1.453) of firms that had no DRIP in 2004 were predicted not tohave one (21.88% versus 34.58%) as compared to firms that did have a DRIP in 2004.