Journal of Business Finance & Accounting, 38(1) & (2), 198–224, January/March 2011, 0306-686Xdoi: 10.1111/j.1468-5957.2010.02214.x

Institutional Dividend Clienteles Underan Imputation Tax System

AELEE JUN, DAVID R. GALLAGHER AND GRAHAM H. PARTINGTON∗

Abstract: Shareholdings for a sample of institutional equity funds, operating under the Aus-tralian imputation tax system, show that dividend policy and fund holdings are related. Relativeto market benchmarks and ownership levels across firms, institutional funds are overweightin stocks that pay dividends. Among dividend-paying stocks there is no simple preference forhigh dividend yields, probably because the highest dividend yields are not sustainable. Insteadwe find an inverted U relationship between institutional ownership and dividend yield. Thetax hypothesis dominates the prudent-man hypothesis in explaining ownership by institutionalclienteles. Institutional funds have a higher ownership in stocks which carry full imputation taxcredits compared to stocks which have partial, or zero, imputation tax credits.

Keywords: dividends, clienteles, institutions, funds, portfolio holdings, imputation, tax,prudent-man

1. INTRODUCTION

Do dividend clienteles exist? In their seminal study, Miller and Modigliani (1961)demonstrate that dividend policy is irrelevant to firm value under perfect capitalmarkets. However, allowing market imperfections, M&M argue that investor taxcharacteristics, and the differential taxation of dividends and capital gains, can leadto tax-induced dividend clienteles. This clientele hypothesis suggests a relationshipbetween firms’ dividend payout policies and investor characteristics. For example,firms that pay lower dividends attract investors with higher marginal tax rates, whereasfirms with higher dividends are favoured by tax-advantaged institutional investors andby retail investors with lower marginal tax rates.

∗The first author is from the School of Accounting and Finance, Faculty of Commerce, University ofWollongong, Australia. The second author is from the Faculty of Business, University of Technology, Sydneyand Capital Markets CRC Limited, Sydney. The third author is from the Finance Discipline, Faculty ofEconomics and Business, University of Sydney. They would like to thank the Capital Markets Co-operativeResearch Centre (CMCRC) for supporting this research and the Securities Industry Research Centre of Asia-Pacific (SIRCA) for the provision of the data used in this study. They also thank the investment managersfor providing the data used to construct the Portfolio Analytics Database. The authors are also grateful tothe anonymous referee and Peter Pope (editor) who provided helpful comments and suggestions. (Paperreceived November 2007, revised version accepted April 2010)

Address for correspondence: Graham H. Partington, Finance Discipline, Faculty of Economics andBusiness, University of Sydney, NSW 2006, Australia.e-mail: graham.partington@sydney.edu.au

C© 2010 Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UKand 350 Main Street, Malden, MA 02148, USA. 198

Journal of Business Finance & Accounting

INSTITUTIONAL DIVIDEND CLIENTELES 199

The existence of tax-induced dividend clienteles has important implications forcorporate financial policies, as investors’ tax characteristics may influence financialdecisions. For example, Perez-Gonzalez (2003) argues that a change in the dominantshareholder’s income tax rate has a significant effect on dividend payout policies.Similarly, Desai and Jin (2010) find that firms alter payout policy in response to thetax preferences of their institutional investors.

Institutions are likely to be attracted to dividends for two primary reasons. First,institutions as fiduciaries have an incentive to hold stocks that pay higher dividendsunder a common institutional charter and prudent-man rule restrictions (Brav andHeaton, 1998). Second, institutions prefer dividends because of the relative taxadvantage on dividends (Allen et al., 2000).1 The implication from these theories isthat institutions prefer dividend paying stocks to non-dividend paying stocks and willalso prefer higher dividends.

In this study we provide an examination of dividend clienteles among Australianinstitutional investors. For this purpose we use portfolio holdings data for a widecross-section of the market at the institutional level. To date, most of the empiricalwork on institutional dividend clienteles has been conducted in the US. Investigatinginstitutional dividend clienteles in an Australian context is particularly interestingas Australia operates a full imputation tax system.2 The Australian imputation taxsystem, and the funds that we study, provide a unique opportunity to examine taxesand dividend clienteles. This is because there are cases where dividends are clearlytax advantaged relative to capital gains, and cases where dividends are clearly taxdisadvantaged relative to capital gains. As a consequence, we are able to distinguishbetween the prudent-man hypothesis and the tax advantage hypothesis in explaininginstitutional dividend clienteles. We also provide an explanation for the previouslyunexplained result that while institutional investors may have a preference fordividends, they are underweight in the stocks with the highest dividend yields. Thedata also allows us to investigate the impact of several changes in the tax system overthe period of the study.

Under the imputation system, franked dividends, which are dividends paid fromprofits subject to Australian company tax, carry imputation tax credits. The taxcredits, generally called franking credits, represent a full refund of corporate tax. TheAustralian recipients of franked dividends are able to reduce their income tax liabilityon the dividends by the amount of the imputation tax credits. The mechanics of theimputation tax system is demonstrated and compared to the classical tax system inTable 1.

Dividends may be fully franked, partially franked, or unfranked, depending onwhether the profits from which they have been paid are fully taxed, partially taxed, oruntaxed at the statutory corporate tax rate. Companies are required to keep recordsof their available franking credits by maintaining a franking account. Less than fullfranking arises when income is not taxed fully in Australia because of tax shields, suchas past losses, or because profits are earned in overseas tax jurisdictions.

1 Although US corporations face the same tax rates between dividend income and capital gains, the‘dividend received deduction’ allows US corporations to reduce the effective tax rate on dividends by 70to 100%.2 Other countries that have operated full or partial imputation tax systems include New Zealand, Mexico,Finland, Taiwan, Norway, the UK , Germany, France, Italy, Canada and Ireland. Under the partial imputationsystems, only part of the corporate tax paid can be offset against personal tax liabilities.

C© 2010 Blackwell Publishing Ltd

200 JUN, GALLAGHER AND PARTINGTON

Table 1The Mechanics of the Australian Imputation Tax System

Classical Imputation Classical ImputationMarginal Tax Rate (td) 47% 15%

Corporate levelNet income $1.00 $1.00 $1.00 $1.00Corporate tax (tc = 30%) 0.30 0.30 0.30 0.30Cash dividends(D) 0.70 0.70 0.70 0.70

Shareholder levelCash dividends (100% payout ratio) 0.70 0.70 0.70 0.70Gross-up adjustment NA 0.30 NA 0.30Assessable income(aD/(1 − tc)) 0.70 1.00a 0.70 1.00a

Personal tax liability (=assessable income ∗ td) 0.33 0.47 0.105 0.15Franking credits (b(D/(1 − tc) ∗ tc)) NA 0.30b NA 0.30b

Tax paid (cno tax due to surplus tax credits) 0.33 0.17 0.105 0.00c

Dividends after taxes (d$0.7 + a cash refund of $0.15) 0.37 0.53 0.595 0.85d

Notes:This table compares an imputation tax system with a classical tax system. Under the Australianimputation system, Australian resident investors who receive dividends paid from profits that have beentaxed at the Australian corporate tax are entitled to a tax credit. This credit is known as an imputationcredit (or a franking credit). Suppose an Australian investor with a tax rate, td , receives a dividend, D,distributed from corporate profits fully taxed at the Australian corporate tax rate, tc . The investor’s incometax liability is calculated based on the grossed-up amount of the dividend which is calculated as D/(1 −tc). Therefore, the investor’s tax liability is given by (D/(1 − tc))td . Then, the investor is entitled to animputation tax credit which is the full amount of corporate tax paid on the dividend received, (D/(1 −tc))tc . Thus, the tax the investor is liable to pay is equal to (D/(1 − tc))(td − tc).

Our sample of institutions contain two types of funds: (1) Pooled SuperannuationTrusts (PSTs) which are investments open exclusively to pension funds, where theseinvestments are taxed at concessional rates (relative to the corporate tax rate) of 15%,and (2) wholesale Unit Trusts (UT),3 which are not taxed, but their distributions aretaxed in the hands of the unit holders in each fiscal year. Because the franking creditspass through the unit trust, there is no clear basis to hypothesise whether or not unittrusts will prefer franked dividends. This will depend on the extent to which the trusttakes into account unit holders’ preferences for franked dividends, if any. Whetheror not unit trusts have a preference for franked dividends is, therefore, an empiricalquestion. In contrast, as we explain below, it is possible to strongly hypothesise thatpension funds (PSTs) have a clear preference for franked dividends. Consequently, weprovide separate analyses for unit trusts and pension funds.

Franked dividends are particularly attractive to pension funds as they effectivelyface a negative tax rate (tax benefit) on such dividends. Over the period of the studypension funds enjoyed surplus imputation credits of at least 15% on fully frankeddividends (see Table 1, column 4). The pension fund could use the surplus creditsto offset taxes on other income and capital gains, with any unused credits becomingrefundable in cash from 1 July, 2000.4 The result is a clear tax advantage for fullyfranked dividends over capital gains.

3 Pension funds are permitted to invest in unit trusts.4 Prior to 1 July, 2000, the franking credit was wasted when shareholders had no tax payable on otherincome, subsequent to that date any portion of the tax credit not used by shareholders became refundablefor cash.

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 201

The results show that institutional ownership is higher for dividend paying firmsthan for non-dividend paying firms. This result holds after controlling for firmcharacteristics such as size, risk (beta) and past period abnormal returns. However,institutions are not attracted to the highest dividend yield stocks, which we suggest isbecause the highest dividend yields are not sustainable. Instead there is evidence of aninverted U shape relationship between dividend yield and institutional fund holdings.We find that both pension funds and unit trusts are overweight (relative to themarket index) in stocks with fully franked dividends and underweight in stocks withunfranked dividends. This is consistent with the tax advantage hypothesis dominatingthe prudent-man hypothesis in explaining institutional dividend preferences. Theimpact of tax changes over the study period is modest. A greater impact seems toarise from changes over time in the relative proportions of stocks paying fully franked,partially franked and unfranked dividends.

The remainder of the paper is organised as follows. Section 2 presents an overviewof the literature on dividend clienteles and outlines the development of our hypothe-ses. Section 3 provides a description of the data and the sample. In Section 4, wepresent the evidence on institutional portfolio holdings and dividends. Section 5summarises and concludes the study.

2. LITERATURE AND HYPOTHESES

(i) Indirect Evidence on Dividend Clienteles

The first empirical strand of the dividend clientele literature examines stock pricemovements around the ex-dividend day. In their seminal work, Elton and Gruber(1970) find that the stock price drops by less than the dividend. Elton and Grubersuggest that the observed ex-dividend price decline reflects the preferential treatmentof capital gains over dividends and also reveals the marginal investor’s tax rate. Consis-tent with their tax-clientele hypothesis, they find that stocks with higher dividend yieldshave higher ex-dividend price falls. They interpret this empirical evidence as implyingthat investor tax rates decrease with dividend yields. Later studies show, however, thatthe ex-dividend price movement can also be affected by short-term trading. Factorssuch as transaction costs (Kalay, 1982; and Karpoff and Walkling, 1988), and the risksof dividend capture (Fedina and Grammatikos, 1993) have been proposed as affectingex-dividend trading and inducing a positive correlation between dividend yield andthe ex-dividend price drop.

The second strand of studies, including Bajaj and Vijh (1990) and Denis et al.(1994), examines the association between dividend yields and stock price reactionaround dividend change announcements. The dividend clientele hypothesis predictsthat if high dividend yield stocks are held by investors with a preference for dividends,the price reactions to dividend changes should be greater for higher dividend yieldstocks. Bajaj and Vijh (1990) and Denis et al. (1994) find evidence consistent with thisview. However, Denis et al. (1994) note that such price reactions to dividend changeannouncement are also consistent with the dividend signalling hypothesis.

The literature cited above provides mixed evidence regarding the presence of div-idend clienteles. The ex-dividend literature supports both the tax clientele and short-term trading hypotheses, while the dividend announcement literature is consistentwith both the tax clientele and signalling hypotheses. The difficulties of inference

C© 2010 Blackwell Publishing Ltd

202 JUN, GALLAGHER AND PARTINGTON

in the indirect studies might be overcome if the characteristics of the investorswere known. Recently, as more detailed share ownership data has become available,researchers have been able to test for different types of dividend clienteles directly(Grinstein and Michaely, 2005; and Desai and Jin, 2010).

(ii) Direct Evidence on Dividend Clienteles

(a) Retail Investors

Using data on the portfolios of retail investors, Scholz (1992) finds that investors withhigher marginal tax rates hold a portfolio of lower yield stocks. Brav and Heaton(1998) show that older and low-income investors prefer high yield stocks.5 They alsofind that older and low-income retail investors buy small stocks before the ex-dividenddate. Similarly, Perez-Gonzalez (2003) concludes that when a firm’s large shareholdersare individuals, the firm’s dividend payout policy is influenced by the tax preferencesof those shareholders. Overall, the evidence on dividend clienteles for retail investorsis supportive of the presence of tax clienteles.

(b) Institutional Investors

It has been suggested that institutions’ monitoring roles (Shleifer and Vishny, 1986;and Gillan and Starks, 2000) and their information gathering abilities (Michaely andShaw, 1994) can increase the value of the firm. According to Shleifer and Vishny(1986), firms can increase their profile and signal to the market by altering theirdividend payout policies. Allen et al. (2000) predict that firms will pay dividends toattract institutions. They argue that dividends attract institutions for two reasons: taxesand regulation. First, institutions are relatively tax advantaged in relation to dividendincome (Tax-advantage hypothesis). Second, institutions as fiduciaries are expectedto invest according to ‘prudent-man rules’ and firms that pay higher dividends aregenerally considered more prudent (Prudent-man hypothesis).6

In Australia, legislation imposes the prudent-man rule on both pension funds andunit trusts. The investment covenant for pension funds is in the SuperannuationIndustry (Supervision) Act 1993. Section 52(2) of the legislation requires the trustee‘to exercise, in relation to all matters affecting the entity, the same degree of care,skill and diligence as an ordinary prudent person would exercise in dealing withproperty of another’. A similar covenant also exists under the Managed InvestmentsAct 1998 applying to unit trusts. Section 601FC(1) states that ‘the responsible entity ofa registered scheme must exercise the degree of care and diligence that a reasonableperson would exercise if they were in the responsible entity’s position’.

The application of the prudent-man hypothesis in Australia parallels that in theUS. However, the tax situation is quite different from the US and the tax-advantagehypothesis does not universally apply. There are two issues to consider, namely, theextent to which the dividend carries imputation tax credits (the franking level) andwhether the fund is a pension fund, or a unit trust.

5 They find that retail investors in general prefer non-dividend paying stocks to dividend paying stocks.However, for retail investors who hold dividend paying stocks, their preference is for larger dividends.6 In examining the effect of the prudent-man rule on institutional ownership, Del Guercio (1996) includesdividend yield as one of the prudence indicators.

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 203

Dividends and Pension Funds: In Australia the availability of franking credits substan-tially increases the attractiveness of dividends for pension (superannuation) funds. Asshown in Table 1, with a 15% tax rate for pension funds and a 30% corporate taxrate, pension funds effectively receive a 15% tax refund from fully franked dividends.Effectively the pension fund tax rate on such dividends is negative rather than positive.The Association of Superannuation Funds of Australia (ASFA) provides eloquenttestimony to the strong attachment that superannuation funds have to imputationcredits:

. . . AFSA does not support any measure that would have a detrimental impact on thecurrent system in Australia of dividend imputation. Neither a part nor full exemptionof dividends from tax, nor a partial tax credit, would substitute for the effectivenegative 15% tax rate that presently applies to franked dividends derived by Australiansuperannuation funds ASFA (2002, p.7).

Therefore, we hypothesise that pension funds will be overweight (relative to marketindex) in stocks paying franked dividends. In contrast, we also hypothesise thatpension funds will be underweight in stocks that pay unfranked dividends. Thereasons for this latter hypothesis are as follows: since 21 September, 1999, pensionfunds have only been taxed on 70% of capital gains, where the asset is held formore than twelve months. This makes the effective tax rate that pension funds payon capital gains 10.5%. Thus, capital gains are more attractive to pension fundsthan unfranked dividends taxed at 15%. If, however, taxes do not affect institutionaldividend clienteles, and instead clienteles are only a product of the prudent-man rule,then pension funds will be overweight in dividend paying stocks and it will not matterwhether the stocks pay franked or unfranked dividends.

Dividends and Unit Trusts: In the case of unit trusts, it is the unit holders of the trustthat are taxed, rather than the trust structure itself. The dividend and capital gainpreferences of unit holders will depend on their specific tax circumstances, which willvary across investors.7 There is a trade-off between access to the benefits of imputation,and the potential for lower effective tax rates on capital gains. Since the balance ofthis trade-off can vary substantially across investors in the unit trust, it is not clearwhether there will be a preference for franked dividends over capital gains. However,it is plausible to assume that capital gains will be preferred to unfranked dividends.Statutory tax rates on capital gains and income are the same in Australia, since realisedcapital gains are taxed as income. However, resident individuals and pension fundsare only taxed on part of the capital gain. Other investors can reduce the effectivecapital gains tax rate by deferring the realisation of their assets which exhibit capitalgains. Consequently, unfranked dividends are likely to be tax disadvantaged relative tocapital gains.

Therefore, we hypothesise that unit trusts will be underweight in stocks that payunfranked dividends. In contrast, the prudent-man hypothesis predicts that unit trustswill be overweight in dividend paying stocks irrespective of their franking status. Theunit trusts preference for franked dividends is an empirical question, since as discussedabove it is difficult to establish this from a-priori reasoning. The results of the foregoingdiscussion are summarised in the contingency Table 2.

7 The extent to which such preferences affect unit trust portfolios also depends upon whether fundmanagers cater to particular tax clienteles.

C© 2010 Blackwell Publishing Ltd

204 JUN, GALLAGHER AND PARTINGTON

Table 2Dividend Preferences for Pension Funds (PST) and Unit Trusts (UT)

Pension Fund Unit Trust

Hypotheses Tax Prudent-man Tax Prudent-man

FrankedDividend

A preference forfranked dividendsdue to a taxadvantage relativeto capital gains

A preference fordividends

Uncertain whetherfranked dividendshave a taxadvantage relativeto capital gains

A preference fordividends

UnfrankedDividend

Avoidance ofunfrankeddividends due to atax disadvantagerelative to capitalgains

Avoidance ofunfrankeddividends due to atax disadvantagerelative to capitalgains

Note:This is the contingency table which summarises discussions of the Tax Hypothesis and the Prudent-man Hypothesis to be tested for the two types of funds studied; Pension funds and Unit trusts.

Endogeneity and Franking : Evidence that funds are overweight in dividend payingstocks does not necessarily mean that funds select dividend paying stocks. Managersobserving institutions on their share register may respond by increasing dividendpayouts if they believe that is what institutions prefer. However, such potentialendogeneity is less likely in relation to fund weights by franking levels. This is becausethe extent to which dividends are franked is determined by law. Prior to 1 July, 2001,when considerable simplifications to the administration of the imputation system wereintroduced, there were complex rules governing the extent of franking. Anti-streamingprovisions prevented differential franking of dividends and there was a minimum levelto which dividends had to be franked. That minimum level was designed to exhaustthe franking account at the time of dividend payment, subject to allowance for otherdividends that were still to be paid. In other words, dividends generally had to befranked to the maximum amount possible. Thus, if managers wished to influencefranking levels they could only do so indirectly by engaging in activities to changethe amount of tax the company paid. After 1 July, 2001, these rules were relaxedsomewhat. However, there were still extensive anti-streaming provisions and substantialtax penalties for over-franking the dividend. This left managers with some discretionto frank the dividends to a lesser extent than the maximum amount possible. However,there is little evidence that they made use of this flexibility. In the six months after 1July, 2001, the proportion of fully franked dividends for stocks in the index increasedby the maximum quantum observed over the study period.

If institutional investors are overweight in stocks (relative to the market index) withfranked dividends and underweight in stocks with unfranked dividends, this impliesthat some other classes of investor must be underweight in stocks paying frankeddividends and overweight in stocks paying unfranked dividends. The likely candidateshere are overseas investors. Overseas investors are not able to utilise franking credits,and therefore have limited benefit from their receipt. To the extent that the value offranking credits is priced into Australian stocks, then rather than paying for a benefit

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 205

they do not enjoy, it will be more attractive for overseas investors to hold stocks thatpay unfranked dividends.

Impact of Tax Changes: Tax changes relevant for our study took place on 1 July, 2000and 1 July, 2001. Consequently, we identify three periods when analysing the impact oftax changes on fund’s dividend preferences. Period 1 runs from 1 January, 2000 to 30June, 2000 and predates any changes. Period 2 runs from 1 July, 2000 to 30 June, 2001and covers the first set of changes. Period 3 runs from 1 July, 2001 to 31 December,2001 and covers the second set of changes.

The first set of changes (from 1 July, 2000) involved the refund of excess imputationcredits in cash for individuals and superannuation (i.e., pension) funds, and also acut in corporate tax rates. Previously, if an investor had surplus imputation creditsthat could not be offset against the investor’s tax bill, the credits were wasted. Theavailability of cash refunds, therefore, makes fully franked dividends more attractive.However, the effect is not likely to be very large. Well-run pension funds would havepreviously arranged their affairs to minimise the wastage of credits and individualsare unlikely to be a significant clientele for the funds we study. However, Beggs andSkeels (2006) in an ex-dividend study claimed that the refund of surplus frankingcredits led the market to place a positive value on franking credits. The corporate taxrate reduction from 36% to 34% makes the value of imputation credits lower, but theeffect in terms of yield is quite small. The effect is to reduce the average dividend plusfranking credit yield by about one tenth of one percent. On balance we hypothesisethat the net effect of the above changes is likely to be a small but positive increase inthe attractiveness of stocks paying fully franked dividends.

The date of 1 July, 2001, was the commencement for the following changes. Thecorporate tax rate was cut from 34% to 30%, making imputation credits less valuable,but the effect is again small, changing the yield by about 0.15% for the averagedividend plus franking credit yield. Early refunds of excess imputation credits weregiven for the 2001-2002 tax years. This would make franking credits slightly morevaluable. We hypothesise that the net effect of these changes is likely to be littledifference in the attractiveness of stocks paying franked dividends. A further changeon 1 July, 2001, was the introduction of imputation credits of up to 15% in respect offoreign dividend withholding tax. This has no direct impact on the value of frankingcredits, but would allow more Australian companies to frank their dividends. This maywell have contributed to the continuation of a trend for more firms to pay fully frankeddividends.

The empirical evidence on institutional dividend clienteles to date has providedmixed results. Del Guercio (1996) examines the impact of the prudent-man rule oninstitutional holdings. After controlling for several proxy variables for stock qualitysuch as liquidity, firm size, risk and S&P ranking, Del Guercio finds that dividend yieldplays no significant role in explaining the level of ownership by banks, and negativelyaffects mutual fund ownership. The Del Guercio (1996) study, however, provides noclear measure of the proportion of taxable/tax-exempt investors. Taking into accountthe tax heterogeneity within institutional investor classes, Strickland (1996) findsthat taxable institutional investors prefer low dividend yield stocks while tax-exemptinvestors do not show any preference for either high or low yield stocks.

Grinstein and Michaely (2005) find that institutions prefer dividend paying stocksto non-dividend paying stocks. However, institutions do not exhibit any preference

C© 2010 Blackwell Publishing Ltd

206 JUN, GALLAGHER AND PARTINGTON

for high yield stocks.8 Further, Grinstein and Michaely (2005) find no evidence thatinstitutional holdings affect dividend payout policies. On the other hand, Desai andJin (2010) show that the tax preferences of institutional investors influence dividendpayout policies. They find that firms with higher tax-exempt institutional ownershippay higher dividends.

Studies that examine the behaviour of institutional ownership around dividendevents also provide mixed results. Dhaliwal et al. (1999) find that dividend initiationsresult in higher institutional ownership. Michaely et al. (1995), however, find nosignificant drop in institutional ownership following dividend omissions. Examiningboth dividend initiations and omissions, Binay (2001) finds that dividend initiationslead to an increase in institutional ownership while dividend omissions result in a dropin institutional ownership.

(iii) Evidence on the Value of Dividend Imputation Tax Credits

There is substantial debate on whether imputation credits have any value in themarket. However, there appears to be a gap in the literature with respect to whetherdividend franking influences institutional holdings under the imputation tax system.The advantage of the current study is the direct examination of fund holdings, whereasstudies such as Bell and Jenkinson (2002) and Hodgkinson et al. (2006) have to inferpension fund behaviour from ex-dividend price movements.

Direct evidence on the market value of dividends and franking credits is mixed.Walker and Partington (1999) used data from Australian Securities Exchange (ASX)on cum-dividend trading during the ex-dividend period, which allowed the simultane-ous observation of cum-dividend and ex-dividend prices. Chu and Partington (2001)estimate the value of Australian dividends as the price difference between the existingshares and the new shares issued under a rights issue, in which the new share issue doesnot entitle the shareholder to the next dividend. Both studies show that one dollar ofAustralian dividends is worth significantly more than one dollar for dividends thatcarry full imputation tax credits.

A contrary result is found in Cannavan et al. (2004). Analysing the price differencebetween stocks and futures to infer the value of imputation credits, they suggest thatthe value of imputation tax credits was reduced to zero after the trading of dividendsand imputation tax credits was restricted by the 45-day holding rule introduced in1997. Beggs and Skeels (2006) also suggest that franking credits had no value untilthe Government introduced cash refunds for surplus franking credits in July 2000,after which the franking credits were estimated to be worth 57% of their face value.Using the valuation framework developed in Cannavan et al. (2004), Cummings andFrino (2008) find Australian imputation credits to be worth at least 50% of their facevalue during the period from 2002 to 2005. Beggs and Skeels (2006) and Cummingsand Frino (2008) suggest that the year 2000 tax change which permitted a tax rebateof unused tax credits significantly increased the value of imputation credits to themarginal investor in the Australian market.

8 Grinstein and Michaely (2005) allow for investor heterogeneity among institutions to some extent. Inaddition to institutional holdings at aggregate level, they also examine separate institutional holdings;holdings by bank trusts and pension funds, and holdings by mutual funds, investment advisors, andinsurance companies.

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 207

New Zealand introduced an imputation tax system in April 1988 which led to stockdividends that were either taxable, or non-taxable. Anderson et al. (2001) examine theprice response to the announcement of both taxable and non-taxable stock dividends.They find a substantially higher abnormal return for taxable stock dividends than fornon-taxable stock dividends, consistent with a positive valuation of the imputationcredits attached to the taxable dividends.

Evidence from the UK which operates a partial imputation tax system9 is generallysupportive of the tax hypothesis. Lasfer (1995) finds that the price drop-offs increasedsignificantly after the introduction of the 1988 UK Income and Corporation Taxes Actwhich was designed to increase the value of a dollar of taxable dividends. Bell andJenkinson (2002) examine the impact of the 1997 Finance Act on dividend valuationsin the UK . The reform abolished the right of pension funds, the largest investor groupin the UK, to reclaim imputation tax credits on dividends. They find a significant fallin the drop-off ratios of high yield stocks consistent with pension funds being themarginal investors in such stocks.

Ex-dividend price behaviour, involving the abolition of UK pension funds rights toclaim imputation tax credits, was also investigated by Hodgkinson et al. (2006). Whenthe availability of franking credits from UK stocks was removed, there was still a shortwindow where the refund of imputation credits from Irish stocks remained available.The ex-dividend evidence suggests that over this window, the pension funds were themarginal investors in Irish stocks that had substantial imputation credit yields.

Armitage et al. (2006) employ the method used in Chu and Partington (2001) toestimate the market value of dividends in the UK. Armitage et al. (2006) find themarket value of dividends in the UK to be significantly greater than the face value ofdividends. They also conclude that there was no evidence that the loss of imputationcredits to pension funds under the 1997 Finance Act had a substantial impact.10

3. DATA AND SAMPLE DESCRIPTION

The primary data for this study is drawn from the Portfolio Analytics Database andconsists of the monthly portfolio holdings of a representative sample of Australianinstitutional equity funds which exhibits a broad cross-section of the fund managementpopulation in Australia. The funds are wholesale funds, either superannuation fundsor unit trusts, and open to only high net worth individuals, corporates, institutionalinvestors and pension funds.11 Data on monthly portfolio holdings is very difficult toacquire in Australia since holdings disclosure is not mandated by law, and it can onlybe obtained with the co-operation of fund managers. This co-operation was obtainedwith the support of Mercer Investment Consulting for institutional fund managerswithin the Mercer investment universe. The sample was constructed by requesting

9 From 4 April, 1999, a UK resident individual shareholder who receives a dividend from a UK residentcompany is entitled to a credit of 1/9 of the dividend and pays tax on the grossed up amount of the dividendat a rate of 32.5% (10%) if the person is (not) liable to higher than basic rate tax. The tax credits receivedcan then be used as an offset to the personal tax liability.10 For the impact of imputation tax systems on the cost of capital in the Australian and the UK markets, seeDempsey and Partington (2008) and Ashton (1989) respectively.11 Fund managers are typically remunerated by a fixed percentage of total funds under management. InAustralia, institutional fund management performance surveys are widely reported in the public domain ona pre-expenses and pre-tax basis, however, in terms of superannuation and public funds, reported returnsare also available on an after expenses and taxes basis.

C© 2010 Blackwell Publishing Ltd

208 JUN, GALLAGHER AND PARTINGTON

from each fund management institution the holdings of the largest and/or secondlargest pooled investment fund open to institutional investors. Fund managers alsoreported their investment style. These self reporting classifications of fund style areempirically consistent with the characteristics of portfolio holdings for active managers(Ainsworth et al., 2008).

Based on the size of the funds in terms of funds under management, the samplerepresents five of the top 10 Australian institutional fund managers, four ranked 11-20, five ranked 21-30 and the remainder are outside the largest 30 managers. A moredetailed description of the database is available in Gallagher and Looi (2006) andBrands et al. (2006). Comparison of the return on the funds in our sample and thebroader population of investment funds provided by Mercer Investment Consultingshowed that the two sets of returns were almost identical. Comparison of the returndistributions is presented in Table 3.

Table 3Comparisons of Monthly Returns of Funds from the Mercer Investment Universe

and from the Study Sample

Monthly ReturnsJan 2000 to Dec 2001

Total Population Study Sample Test of Difference

Mean 0.0086 0.0089 t = −0.031 (p = 0.97)Median 0.0069 0.0065 z = −0.041 (p = 0.97)StDev 0.0350 0.0349 F = 0.995 (p = 0.495)

Notes:Mercer Investment Consulting provided mean monthly returns for their universe of funds andmean monthly returns were computed for the sample of funds in the study. The table reports the mean,median and standard deviation (StDev) of these monthly returns over the twenty four months covered bythe study. A two-sample t-test and a Mann Whitney U test are used for the test of difference in means andmedians while test of difference in variance is conducted using a two-sample F -test.

We required complete and continuous monthly portfolio holdings for the fundsin the study, and this data was only available for the calendar years 2000 and 2001.Table 4 reports the characteristics of the sample data that we employ in this study. Thesample contains a total of 49 institutional funds, where the value of total funds undermanagement (FUM) is almost $30 billion in 2001. In 2000, an average fund holdsa portfolio of 76 stocks and the number of stocks under management ranges from22 to 271 stocks. The average semi-annual portfolio dividend yield in 2001 declinedslightly to 1.64% from 1.75% in 2000 while the average portfolio dividend frankingpercentage increased from 65% to 76%.12 In addition, the institutions in our sample

12 Following Graham and Kumar (2006), at the end of month t, we compute the dividend yield of theportfolio i, PDYit as:

PDYit =Nit∑

k=1

wikt DYikt ,

where DYikt is the semi-annual dividend yield of stock k in portfolio i from the most recent ex-dividendmonth prior to month t. The semi-annual dividend yield of stock k is calculated as the dividend amountdivided by the closing price on the last cum-dividend day. Within portfolio i, DY for stock k is weighted bywikt, which is the weight of stock k in portfolio i at the end of month t. The weighted dividend yields are then

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 209

Table 4Sample Characteristics

Aggregate-Level Fund-Level Stock-Level

Total Aggregate Portfolio Portfolio InstitutionalNo.of FUM No. of DY Franking OwnershipFunds ($mil) Stocks FUM ($mil) (%) (%) (%)

2000 49 23787.428Mean 76 506.115 1.75 64.82 3.50Median 53 140.016 1.67 64.40 2.17Minimum 22 10.348 1.36 54.24 0.0004Maximum 271 2883.476 2.68 75.12 23.15

2001 49 28984.378Mean 73 591.518 1.64 75.93 3.37Median 49 205.403 1.62 75.70 2.28Minimum 22 16.333 1.29 67.57 0.002Maximum 265 3331.038 2.10 84.88 25.06

Notes:This table presents the summary statistics for the sample of institutional monthly portfolio holdingswhich is obtained from Portfolio Analytics Database for the years 2000 and 2001. At the aggregate level,we report the total number of funds in the sample and aggregate funds under management (FUM).In addition, we provide the statistics for the following portfolio variables: (i) number of stocks in theportfolio, (ii) the value of FUM, (iii) portfolio semi-annual dividend yield (DY) and (iv) portfolio frankingpercentage. For each fund, all the four variables are calculated monthly and averaged over the year. Forany stock k contained in the portfolio holding at the end of year t, institutional ownership is calculatedas the aggregate number of shares of stock k owned by all funds in the sample at year t divided by totaloutstanding shares of stock k at that time.

own on average 3.5% of the companies they invest in and a maximum of 23.15% in2000. A similar pattern is observed in 2001.

To compare the holdings of dividend paying stocks in institutional portfolios tothose of benchmark market portfolios, we obtained the end-of-month compositionof the S&P/ASX Index from Standard and Poor’s. These indices are constructedto provide an equity performance benchmark for fund managers investing on theAustralian Securities Exchange (ASX). According to Brands et al. (2006), 5% of thefunds in the Portfolio Analytic Database are benchmarked to the S&P/ASX 100, 17%to the S&P/ASX 200 and 78% to the S&P/ASX 300. The study, therefore, uses theS&P/ASX 200 index and the S&P/ASX 300 index as benchmark market portfolios.

For each stock in the sample, we collect daily closing prices, the semi-annualdividend amount (since Australian stocks pay half-yearly dividends), the level offranking,13 the ex-dividend date, monthly shares outstanding, and book-to-marketratio from ASX data (provided by SIRCA). Dividend payout ratios are collectedfrom Aspect Fin Analysis.14 Equity betas were obtained from Reuters. We take dataon dividends, rather than distributions (dividends plus share repurchases) as share

summed across the stocks (Nit is the number of stocks in portfolio i) to obtain PDYi at the end of month t.Computing an average of PDY over time, we obtain the average portfolio dividend yield for portfolio i. Theportfolio dividend franking percentage is computed in a similar manner.13 Companies must by law report the extent to which dividends are franked.14 Aspect Fin Analysis is an Australian financial database which provides a history of detailed accounting andfinancial information for all companies listed on ASX.

C© 2010 Blackwell Publishing Ltd

210 JUN, GALLAGHER AND PARTINGTON

repurchases were not a substantial part of firm’s distributions at the time of ourstudy.15

4. EXISTENCE OF INSTITUTIONAL DIVIDEND CLIENTELES

(i) Institutional Portfolio Holdings versus Market Benchmarks

(a) Holdings of Dividend Paying Stocks

We begin by examining the monthly portfolio holdings of institutional investors todetermine whether dividend paying stocks are overweight relative to the marketbenchmarks (S&P/ASX 200 and S&P/ASX 300 indices). Following Graham andKumar (2006), we compute the excess portfolio weight in dividend paying stocks(EWdiv) as the difference between the actual weight in dividend paying stocks and theexpected weight in dividend paying stocks. To measure the actual weight in dividendpaying stocks, we construct the aggregate institutional portfolio (Ip) at the end of eachmonth t by combining the portfolios of all funds in our sample at time t. Each month,we compute the weight of dividend paying stocks in the aggregate portfolio (w div

Ip ,t).Similarly, the expected weight is calculated as the weights of dividend paying stocks inthe relevant S&P/ASX Index (w div

Mp,t). In month t, a dividend paying stock is definedas a stock that makes a dividend payment in the previous year. The excess weight individend paying stocks in the aggregate portfolio in month t is therefore:

EWdivIp ,t = w div

Ip ,t − w divMp,t .

We next compute an average of EWdiv over time T as:

EWdivIp = 1

T

T∑

t=1

EWdivIp ,t .

The results are presented in Figure 1.The discussion focuses on the results using the S&P/ASX 200 index which has

a higher weight in dividend paying stocks than the S&P/ASX 300 and therefore,provides a more conservative benchmark. However, we would reach substantively thesame conclusions using the S&P/ASX 300 as the source of the expected weight. PanelA of Figure 1, shows that the actual weight in dividend paying stocks (92.10%) is higherthan the expected weight (88.00%). Using both the t-test and the Wilcoxon rank test,the excess weight (EWDiv

Ip = 4.10%) is significantly different from zero (t = 2.07, p <

0.05, z = 2.48, p < 0.05). Considering average aggregate funds under management ofapproximately $26 billion in our sample (see Table 4), the excess weight of 4% wouldamount to an extra investment of over $1 billion in dividend paying stocks.

We also examine whether institutions being overweight in dividend paying stocksvaries across firm size. Size quintiles were formed at the beginning of each yearusing the firm’s market capitalisation at the end of the previous December. Each

15 Brown and O’Day (2006) state: ‘Brown and O’Day (2005), using a sample of 1768 ASX-listed companies,find that ordinary dividends remain the dominant payout method. They find that over the years 1996-2003off-market buybacks returned $12.1bn to shareholders, while on-market repurchases totalled $10.5bn. In thesame period ordinary dividends totalled $220.3bn and dominated repurchases in each year of the sample.’

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 211

Figure 1Institutional Preference for Dividend Paying Stocks: Actual versus Expected Weight

Panel A: Portfolio Weight in Dividend Paying Stocks

0

10

20

30

40

50

60

70

80

90

100

Actual S&P/ASX200

S&P/ASX300

Wei

ght

in d

ivid

end

payi

ng s

tock

s (%

)

Panel B: Portfolio Weight in Dividend Paying Stocks, Sorted by company size (Market Capitalisation)

0

10

20

30

40

50

60

70

80

90

100

lowest Q2 Q3 Q4 highest

Size Quintile

Wei

ght

in d

ivid

end

pay

ing

stoc

ks(%

)

Actual S&P/ASX 200 S&P/ASX 300

Notes:This figure shows the aggregate preference of institutional investors for dividend paying stocks.Panel A shows the actual and expected weight in dividend paying stocks. The actual weight is measuredfrom the sample of aggregate institutional portfolio holdings obtained from Portfolio Analytic Database whilethe expected weight is computed using the market portfolio (S&P/ASX200, S&P/ASX300). In Panel B, theactual and expected weights in dividend paying stocks are shown for each of the five size quintiles wherethe quintiles are formed at the beginning of each year based on the stock’s market capitalisation at the endof the previous December.

month t, we compute the EWDiv in the five size quintile portfolios (EWDivSQi,t , i =

1, . . . , 5).16 Panel B in Figure 1 shows the average actual weight and average expectedweight in dividend paying stocks for each size quintile. Panel B shows that the actualweight in dividend paying stocks is consistently higher than the expected weight,

16 The mean (median) market capitalisation in quintiles 1 to 5 are: Q1 = 40.6 (40.7) million, Q2 = 106.9(106.8) million, Q3 = 242.5 (240.7) million, Q4 = 640.6 (625) million and Q5 = 8.06 (3.04) billion.

C© 2010 Blackwell Publishing Ltd

212 JUN, GALLAGHER AND PARTINGTON

with the greatest difference in the lower size quintiles. A two-way ANOVA revealsthe EWDiv in each quintile to be significantly positive. EWDiv is significantly differentfrom zero at the 1% level for Q1, Q2 and Q4 (EW Div

SQ1 = 19.07%, EW DivSQ2 = 19.36%, and

EW DivSQ4 = 5.27%) and at the 5 percent level for Q3 and Q5 (EWDiv

SQ3 = 3.65%, EWDivSQ5 =

4.04%).

(b) Preference for High or Low Dividends

Given evidence consistent with an institutional preference for dividend paying stocks,do Australian institutions have a particular preference for stocks paying low or highdividends? To examine this issue, the sample of stocks from the portfolio holdingsof institutions is restricted to the stocks that pay dividends. Two variables are usedto measure the relative magnitude of dividends: dividend yield (DY) and dividendpayout ratio (DPR). First, we form dividend yield quintiles at the beginning of eachyear, where DY is computed as total dividend per share (DPS) in the previous yeardivided by stock price at the end of the previous December. Similarly, DPR quintilesare formed based on the ratio of the dividend per share (DPS) to the earnings pershare (EPS) in the previous year.17 Then, at the end of each month t, we compute theactual portfolio weight in the five DY quintile and DPR quintile portfolios. The weightin each quintile is standardised by the total value weight in dividend paying stocks inmonth t (w div

Ip ,t). The effect of this is to determine the value weight of each quintile inthe institutional portfolio. Using the market portfolio, the expected weights in the fivequintiles are computed in an analogous manner. Panels A and B in Figure 2, show theportfolio weights in DY quintiles and DPR quintiles respectively.

A two-way ANOVA confirms that the actual and expected weights vary significantlyover the DY quintiles and the DPR quintiles. The excess weights in the lowest DYand DPR quintiles (EWDYQ1

Ip = −8.03%, EWDPRQ1Ip = −12.42%) are significantly negative.

The excess weight in the second lowest DPR quintile (EWDPRQ2Ip = −10.17%) is also

significantly negative, but is insignificant in the corresponding DY quintile (EWDYQ2Ip =

−1.47%). In contrast, the excess weights in the third DY and DPR quintiles (EWDYQ3Ip =

10.62%, EWDRPQ3Ip = 20.03%) are positive and significant. However, the excess weights

in quintiles four and five are not significantly different to zero.18

In summary, Australian institutions tend to have less than expected fund holdingsin the lower yielding stocks, higher than expected holdings in the middle quintile, butthey are not overweight in the highest yielding stocks. The pattern is suggestive of aninverted U relation between institutional holdings and the magnitude of dividends.This is further investigated in Section 4(iii) where institutional holdings of individualfirms are examined with inclusion of controls for variables such as company size.

The question remains as to why institutions are not attracted to the highest yieldingstocks. Stocks in yield quintiles 4 and 5 have low weights in the index and this

17 We exclude special dividends in computing DY and DPR because there is ambiguity about whetherdividend increases due to special dividends are likely to continue. The mean (median) DYs in quintiles 1 to5 are: Q1 = 0.016 (0.017), Q2 = 0.032(0.032), Q3 = 0.046 (0.046), Q4 = 0.06 (0.06) and Q5 = 0.088(0.084).The mean (median) DPRs in quintiles 1 to 5 are: Q1 = 0.30 (0.32), Q2 = 0.52(0.52), Q3 = 0.65 (0.65),Q4 = 0.79(0.78) and Q5 = 1.4 (1.03).18 A two-way ANOVA shows that the excess weights for dividend yield and dividend payout ratio in quintiles4 and 5 are not significantly different from zero (EWDYQ4

Ip = 1.21%, EWDYQ5Ip = −2.33%, EWDPRQ4

Ip = −0.60%

and EWDPRQ5Ip = 2.16%).

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 213

suggests that they are smaller capitalisation stocks. The stocks in quintiles 4 and 5also exhibit some unusual dividend characteristics. The dividend yields are high byhistorical standards and the mean dividend payout ratio of over 100% in quintile 5 isunsustainable. Perhaps the high yields are unsustainable and are driven by decliningprices, and perhaps the high payouts are driven by declining earnings. If so, the lack ofattractiveness of these stocks by institutions is understandable, and the more so if thestocks are smaller capitalisation stocks with lower representation in the market index,higher trading costs and lower levels of liquidity.

Observing the price and dividend changes from 1999 to 2000, the high dividendyield in quintiles 4 and 5 appears to be initially driven by the combination ofdecreasing price and increasing dividends. In 2001, there was some recovery in prices

Figure 2Institutional Preference for Dividend Size and Franking: Actual versus Expected

Weight

0

5

10

15

20

25

30

35

40

45

lowest Q2 Q3 Q4 highest

Dividend Yield Quintile

Wei

gh

t (%

)

Actual S&P/ASX 200 S&P/ASX 300

Panel A: Portfolio Weight in Dividend Quintiles

0

5

10

15

20

25

30

35

40

45

lowest Q2 Q3 Q4 highest

Payout Ratio Quintile

Wei

ght

(%)

Actual S&P/ASX 200 S&P/ASX 300

Panel B: Portfolio Weight in Payout Ratio Quintiles

C© 2010 Blackwell Publishing Ltd

214 JUN, GALLAGHER AND PARTINGTON

Figure 2 (Continued)

0

10

20

30

40

50

60

FF PF UF

Wei

ght (

%)

Franking StatusActual S&P/ASX200 S&P/ASX300

Panel C: Portfolio Weight in Franking Groups

0

10

20

30

40

50

60

FF PF UF

Wei

gh

t (%

)

Franking Status

PSF UT S&P/ASX200

Panel D: Portfolio Weight by Franking Groups and Fund Type

Notes:This figure shows the portfolio weights of dividend yield (DY) quintiles (in Panel A) and dividendpayout ratio (DPR) quintiles (in Panel B) in the aggregate institutional portfolio. Within dividend payingstocks, at the end of each year, the DY quintiles are formed based on total dividend payments in the previousyear and stock prices at the end of the previous December. The DPR quintiles are based on the previousyear’s DPS and EPS. Similarly, the portfolio weights of franking groups in the aggregate institutionalportfolio are shown in Panel C. The franking percentage for each dividend paying stock is determined asthe average of the previous year’s interim and final dividend franking percentages. Firms then are formedinto three franking groups: fully franked (FF), partially franked (PF) and unfranked (UF) groups. Theportfolio weight allocated to stocks in each quintile (or each franking group) is computed at the end ofeach month and this weight is standardised by the total weight in dividend paying stocks in that month.The averages of these monthly weights are reported. The expected weights in the quintiles and frankinggroups in the market portfolio are computed and reported in a similar manner.

but dividends were cut. The earnings of the stocks in quintiles 4 and 5 declined in both2000 and 2001.19 This pattern is not evident in the other quintiles. If the institutions

19 From 1999 to 2000, the prices of stocks in DYQ4 (DYQ5) decreased by 10.2% (12.8%) while dividendsincreased by 7.2% (9.6%) on average. In 2001, although prices of stocks in DYQs 4 and 5 recovered by

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 215

anticipated the declining earnings and dividend cuts in the high yield quintiles, thiswould explain the absence of a preference for high yields.

(c) Preference for Franking

We next examine whether dividend franking affects institutional portfolio holdings. Atthe beginning of each year, the franking percentage is determined for each dividendpaying stock using the interim and final franking information in the previous year.

The franking percentage (FRK%) for stock k at the beginning of year t is computedas:

FRK%k,t = InterimFRK%k,t−1 + FinalFRK%k,t−1

2.

Then firms are partitioned into three franking classes: fully franked (FF), partiallyfranked (PF) and unfranked (UF).20 At the end of each month, the firms are sortedinto each of the three franking groups. The actual weights (from the institutionalportfolio) and expected weights (from the index portfolio) for each group arecomputed each month and standardised by the total value weight in dividend payingstocks in that month. Panel C of Figure 2 shows the averages of the monthly weightsfor each franking group. Using a two-way ANOVA, the excess weight in the FF group(EWF F

Ip = 10.40%) is significantly positive, the excess weight in the PF group is notsignificantly different from zero (EWPF

Ip = −0.10%) and the excess weight in the UFgroup (EWUF

Ip = −10.30%) is significantly negative.21

Panel D of Figure 2 examines whether unit trusts and pension funds have differentpreferences for franked and unfranked dividends. For both types of fund a preferencefor fully franked dividends is observed and unfranked dividends are avoided. Forboth pension funds (PST) and unit trusts (UT) the excess weight in the FF group issignificantly positive (EWF F

PST = 11.26%, EWF FUT = 10.8%) while the excess weight in the

UF group is significantly negative (EWUFPST = −9.38%, EWUF

UT = −10.39%). The excessweight in the PF group for both PST and UT is not significantly different from zero(EWPF

PST = −1.88%, EWPFUT = −0.41%). Comparing the weight in each group (FF, PF

and UF) between PST and UT, no significant difference in the weight is found. Wewere unable from a-priori reasoning to establish whether unit trusts had a preferencefor franked dividends, but the empirical evidence is clear. Unit trusts prefer frankeddividends and their preference matches that of pension funds.

To summarise our results thus far, the examination of monthly portfolio holdings ofinstitutional investors suggests that relative to the market benchmark, institutions areoverweight in stocks paying dividends. Institutions appear to prefer stocks with mid-range dividend yields. Both pension funds and unit trusts have a preference for stocks

10% and 19% respectively, it was accompanied by dividend decreases of 18.2% and 13.3%. The earnings ofstocks in DYQ4 declined by 4.7% in 2000 and 3% in 2001. Poor earnings performance was more apparentfor stocks in DYQ5 which were least favoured by Australian institutions. The earnings of stocks in DYQ5decreased by 6.8% in 2000 and 20.4% in 2001.20 We note that these are not equal groups. Among 241 dividend paying stocks contained in the sample ofinstitutional portfolio holdings, the proportions of stocks that pay FF, PF and UF dividends are 66%,17%and 17% respectively. The mean (median) franking percentage in the PF group is 54% (57%).21 Australian dividends can be unfranked, for example, when the company profits were earned overseasand did not result in any tax payments to the Australian government. Consequently, imputation creditscannot be attached to the dividends.

C© 2010 Blackwell Publishing Ltd

216 JUN, GALLAGHER AND PARTINGTON

paying fully franked dividends, and an aversion to stocks paying unfranked dividends.These results are consistent with the tax hypothesis dominating the prudent-manhypothesis as an explanation for institutional dividend preferences.

(d) The Impact of Tax Changes

Table 5 provides a comparison of holdings of dividend paying stocks, by tax regime,franking status and fund type. For the market portfolio, Table 5 shows that there hasbeen a substantial growth in the proportion of firms paying fully franked dividendsover the study period, which has been roughly matched by a decline in the proportionof stocks paying unfranked dividends. However, there was little change in firms payingpartially franked dividends. These changes are contemporaneous with the tax changes,and it is tempting to conclude that the tax changes led firms to fully frank theirdividends. However, fully franking dividends was not a discretionary choice, as thelevel of franking was heavily regulated. It seems that firms must have been payingmore corporate tax over this period, which led to an increase in franking. In Period3, the availability of additional franking credits arising from allowances for foreignwithholding tax may also have increased the supply of franking credits.

Funds are consistently overweight in fully franked dividend paying stocks in eachof the three tax regimes and are consistently underweight in unfranked dividends.Irrespective of fund type, funds significantly increased their weight of holdings in fullyfranked dividends in tax regimes 2 and 3. However, the funds’ fully franked dividendweights did not grow as fast as the weight in the index. For partially franked dividendsthe funds’ weights dropped and for unfranked dividends the change in weights wasnot statistically significant despite a sharp drop in the weight of unfranked stocks inthe index. It is not clear whether the changes in the fund’s holdings are explained bythe tax changes, or by the general increase in payment of fully franked dividends.

(ii) Institutional Ownership Across Stocks

(a) Preference for Dividend Paying Stocks

This section considers whether the level of institutional ownership in dividend payingstocks is higher than in non-dividend paying stocks.22 For any stock k contained in theportfolio holding at the end of year t, institutional ownership for stock k is calculatedas the aggregate number of shares of stock k owned by all funds in the sample at yeart divided by total outstanding shares of stock k at that time. In each year t, firms aresorted based on their end-of-year market capitalisation and grouped into annual sizequintiles. The annual size quintiles are then combined to form size quintiles for thewhole period. For instance, the smallest quintiles for 2000 and 2001 are combined toform the smallest quintile for the full period, and so on. For each quintile, observationsare then divided into two groups: dividend paying and non-dividend paying groups. Afirm is classified as a dividend paying stock if it makes a dividend payment during yeart. For example, a firm which pays dividends in 2000, but not in 2001, will have oneobservation in the dividend paying group and one observation in the non-dividend

22 To examine institutional preferences for dividend paying stocks via institutional share ownership, wefollow a non-parametric test used in Grinstein and Michaely (2005).

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 217

Table 5Dividend Preferences for Pension Funds (PST) and Unit Trusts (UT)

Period 1 Period 2 Period 3

Jan 2000 ∼ July 2000 ∼ July 2001 ∼ Period2-Period 1 Period 3-Period 2FF June 2000 June 2001 Dec 2001 (ANOVA) (ANOVA)

w F FIp 49.00% 53.78 60.68 4.78% 6 .91%∗

w F FPST 48.90 54.51 60.67 5.61∗ 6 .16∗

w F FUT 49.04 53.53 60.68 4.49 7 .15∗∗

w F FMp 31.89 41.15 53.63 9 .26%∗∗ 12.48%∗∗

EWF FIp 17 .11%∗∗ 12.62%∗∗ 7 .05%∗

EWF FPST 17 .02∗∗ 13.36∗∗ 7 .04∗

EWF FUT 17 .15∗∗ 12.38∗∗ 7 .05∗

PF

w PFIp 38.92% 36.56 29.89 −2.36% −6 .67%∗∗

w PFPST 36.90 35.49 29.58 −1.41 −5.91∗∗

w PFUT 39.49 36.92 29.99 −2.57 −6 .93∗∗

w PFMp 35.68 36.96 35.09 1.28% −1.87%

EWPFIp 3.24% 0.04% −5.20%∗∗

EWPFPST 1.22 −1.47 −5.51∗∗

EWPFUT 3.81∗∗ −0.04 −5.11∗∗

UF

w UFIp 12.08% 9.66 9.42 −2.42% −0.24%

w UFPST 14.20 10.00 9.75 −4.20 −0.25

w UFUT 11.47 9.55 9.33 −1.92 −0.22

w UFMp 32.43 21.89 11.28 −10.56%∗∗ −10.61%∗∗

EWUFIp −20.35%∗∗ −12.23%∗∗ −1.86%

EWUFPST −18.24∗∗ −11.89∗∗ −1.53∗

EWUFUT −20.96∗∗ −12.34∗∗ −1.95

Notes:This table provides a comparison of holdings of dividend paying stocks, by tax regime, frankingstatus and fund type. For each period, the portfolio weight (w) in fully franked stocks (FF), partially frankedstocks (PF) and unfranked stocks (UF) is computed. The weight is computed for the aggregate institutionalportfolio (Ip), market portfolio, S&P/ASX 200 (Mp), Pension funds (PST) and Unit Trust (UT). For eachof Ip, PST and UT, the excess weight in each franking group (EWFF , EWPF and EWUF ) is computed as thedifference between the actual weight (wFF , wPF and wUF ) in each fund and the expected weight in Mp. Thedifference in the weight between two periods is tested using a two-way ANOVA. ∗,∗∗ denote significance atthe 5% and 1% levels, respectively.

paying group. Table 6 reports means and medians of institutional ownership in eachgroup for every size quintile.

Table 6 shows that the non-dividend paying stocks tend to be smaller-sized andthe dividend paying stocks are concentrated in large firms. Institutional ownershipalso tends to be higher for large firms. Comparing the ownership between dividendpaying and non-dividend paying firms; the dividend paying group has significantly

C© 2010 Blackwell Publishing Ltd

218 JUN, GALLAGHER AND PARTINGTON

Tab

le6

Inst

itutio

nalP

refe

renc

efo

rD

ivid

ends

:Exa

min

atio

nof

Inst

itutio

nalS

hare

Ow

ners

hip

Div

iden

dPa

ying

Non

-Div

iden

dPa

ying

Test

ofD

iffer

ence

inO

wne

rshi

p:Pa

ying

vs.

Non

-Pay

ing

Ave

rage

Mea

nM

edia

nM

ean

Med

ian

Wilc

oxon

Size

Mar

ket

No.

ofIn

stitu

tiona

lIn

stitu

tiona

lIn

stitu

tiona

lIn

stitu

tiona

lR

ank

Qui

ntile

Cap

($M

)O

bsO

wne

rshi

pO

wne

rshi

pN

Ow

ners

hip

Ow

ners

hip

Nt-t

est

Test

Low

est

3513

52.

351.

5046

0.79

0.20

893.

73∗∗

4.77

∗∗

212

013

53.

252.

0071

1.04

0.37

642.

64∗

3.32

∗∗

326

713

52.

481.

1882

1.29

0.52

533.

05∗∗

3.45

∗∗

471

213

53.

572.

4090

1.67

0.95

454.

17∗∗

3.86

∗∗

Hig

hest

8,89

313

54.

293.

6411

52.

902.

2220

1.83

2.08

∗

Tota

l1,

995

675

3.36

2.42

404

1.24

0.39

271

8.22

∗∗11

.01∗∗

Not

es:

Thi

sta

ble

pres

ents

the

com

pari

son

ofin

stitu

tiona

low

ners

hip

betw

een

divi

dend

payi

ngst

ocks

and

non-

divi

dend

payi

ngst

ocks

.Fo

rea

chst

ock

kco

ntai

ned

ina

port

folio

hold

ing

atth

een

dof

year

t,in

stitu

tiona

low

ners

hip

isco

mpu

ted

asth

eag

greg

ate

num

ber

ofsh

ares

ofst

ock

kow

ned

byal

lfun

dsin

the

sam

ple

atye

art

divi

ded

byto

tal

outs

tand

ing

shar

esof

stoc

kk

atth

attim

e.A

tye

art,

firm

sar

eso

rted

into

size

quin

tiles

and

all

firm

-yea

rsob

serv

atio

nsar

eag

greg

ated

into

each

quin

tile.

Firm

sth

enar

edi

vide

din

toth

etw

ogr

oups

:tho

seth

atpa

ydi

vide

nds

and

thos

eth

atdo

not

pay

divi

dend

s.A

nyst

ock

that

mak

esa

divi

dend

paym

ent

duri

ngye

art

iscl

assi

fied

asa

divi

dend

payi

ngst

ock.

For

each

size

quin

tile,

mea

nsan

dm

edia

nsof

inst

itutio

nalo

wne

rshi

pin

each

grou

par

ere

port

ed.T

est

ofdi

ffer

ence

inin

stitu

tiona

low

ners

hip

betw

een

payi

ngan

dno

n-pa

ying

grou

psis

cond

ucte

dus

ing

the

t-tes

t(d

iffer

ence

inm

eans

)an

dW

ilcox

onra

nkte

st(d

iffer

ence

inm

edia

ns).

∗ ,∗∗

deno

tesi

gnifi

canc

eat

the

5%an

d1%

leve

ls,r

espe

ctiv

ely.

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 219

higher institutional ownership than is the case for the non-dividend paying group.The differences in the mean and median ownership levels between the two groups arestatistically significant, and this result holds for every size quintile (with the exceptionof the t-test in the highest quintile).

The analysis thus far only controls for size and does not account for other firmcharacteristics that might affect institutional ownerships. Additional control variablesare added in the regression analysis that follows.

(iii) Dividends and Institutional Ownership of Stocks: Regression Analysis

To examine how dividends affect the percent of institutional ownership on the shareregisters of individual firms separate regressions year-by-year, and a regression withdata pooled across years, were estimated. We note that the observations in the pooledregression may not be independent, but the results were generally consistent for thepooled regressions and year-by-year regressions. The form of the regression equationas given by the pooled regression was:

Institutional ownershipkt = α + β0Log(Market -Capkt) + β1Betak + β2BtoM kt

+ β3AbRETkt + β4DUMMYif PAYINGkt+ β5DYkt(or DPRkt)

+ β6DUMMYi f F Fkt + εkt .

The dependent variable is the extent of institutional ownership for individual firms.Institutional ownership for firm k is defined as the aggregate number of shares ofstock k owned by all funds in the sample at year t divided by total outstanding shares ofstock k at that time. In addition to dividend variables, the regression includes variablesto control for stock characteristics which affect institutional ownership. Subscript kin the regression indexes the firm and subscript t the year. Following Grinstein andMichaely (2005), the control variables include: firm size Market-Capkt (as measured bythe log of market capitalisation), equity beta (Betak), book-to-market ratio (BtoMkt),and annual abnormal return (AbRETkt).23 We use the beta of the stock to adjust forrisk, and the book-to-market ratio to account for growth opportunities. The dividend-related variables include: a dummy variable, DUMMYIf PAYINGkt , which takes a value ofone if a stock pays dividends during year t and zero otherwise, dividend yield (DYkt),and a dummy variable, DUMMYI fFFkt , which is set to one if a stock pays fully franked (FF)dividends during year t and zero otherwise.24 We also repeat the regression analysisusing dividend payout ratio as an explanatory variable (DPR) instead of dividendyield.25 Table 7 reports the results of the regression analysis.

The results of the pooled regression and year-by-year regressions in Panel Asuggest that institutional ownership is higher for larger firms. Given the influence

23 The annual abnormal return for stock k is computed as the annual stock return minus the expectedreturn given by the CAPM.24 This coding means that partially franked, unfranked, and non-dividend paying stocks are all coded aszero. To exclude the effect of non-dividend paying stocks, we re-ran the regression using only dividendpaying stocks and deleting the DUMMYIf PAYING variable. The detailed results are not reported here, but theconclusions of the analysis were qualitatively unchanged.25 In dealing with three observations with negative dividend payout ratios, we first ran the regression with anadditional dummy variable which coded those observations as one. Alternatively, we also ran the regressionafter excluding those observations from the sample. The results are essentially the same and we report theresults from the regression which excluded those observations.

C© 2010 Blackwell Publishing Ltd

220 JUN, GALLAGHER AND PARTINGTON

of stock size in portfolio construction and risk management, this finding is to beexpected (see Del Guercio, 1996; Gompers and Metrick, 2001; Grinstein and Michaely,2005; and Brands et al., 2006). The coefficient on DUMMYIfPAYING is significantlypositive which is consistent with institutional preference for dividend paying stocks.There is no evidence that institutions prefer high dividend yields as the dividendyield coefficient is statistically insignificant. The coefficient on the franking dummy,DUMMYIf FF , is positive and significant. This implies that institutional ownership issignificantly higher for firms that pay fully franked dividends, consistent with the taxhypothesis. The effects of dividends on the ownership percentage are substantial.The combined effect of positive dividend payment and franking dummies is toincrease institutional ownership by just over 3%. This is comparable to the meanlevel of institutional ownership for dividend paying stocks.26 The alternative regressionsubstituting the dividend payout ratio for the dividend yield does not alter any of theforegoing conclusions. The coefficient on DPR is insignificant and the coefficients onDUMMYIfPAYING and DUMMYIfFF are positive and significant.

Table 7Regression Analysis Effect of Dividends on Institutional Ownership

Panel A: Pooled Regression and Year-by-Year RegressionInstitutional Ownership (%)

Year-by-Year Regressions

Pooled Regression 2000 2001

Intercept −3.87 −3.94 −3.03 −3.15 −4.52 −4.55(t =−3.03∗∗) (t = −3.07∗∗) (t =−1.59) (t =−1.66) (t =−2.62∗∗) (t =−2.62∗∗)

Log (Market-Cap) 0.29 0.29 0.24 0.24 0.336 0.337(t = 4.22∗∗) (t = 4.24∗∗) (t = 2.28∗∗) (t = 2.33∗) (t = 3.61∗∗) (t = 3.59∗∗)

Beta 0.25 0.26 0.46 0.46 0.107 0.12(t = 1.37) (t = 1.40) (t = 1.45) (t = 1.46) (t = 0.48) (t = 0.53)

Book to Market −0.03 −0.03 −0.025 −0.03 −0.033 −0.037(t = −0.95) (t = −1.11) (t =−0.54) (t =−0.75) (t =−0.78) (t = −0.87)

Annual abnormal 0.06 0.06 0.014 0.025 0.188 0.205Return (t = 0.37) (t = 0.36) (t = 0.07) (t = 0.12) (t = 0.64) (t = 0.69)

DUMMYif PAYING 2.08 1.98 2.03 1.91 2.18 1.90(t = 5.11∗∗) (t = 5.06∗∗) (t = 3.41∗∗) (t = 3.43∗∗) (t = 3.84∗∗) (t = 3.27∗∗)

DY −7.32 −10.45 −4.65(t = −1.29) (t =−1.20) (t =−0.63)

DPR −0.15 −0.26 0.25(t = 0.42) (t =−1.15) (t = 0.67)

DUMMYif FF 1.14 1.04 1.30 1.17 0.89 0.79(t = 3.32∗∗) (t = 3.11∗∗) (t = 2.47∗) (t = 2.27∗) (t = 1.98∗) (t = 1.79)

Obs 675 672 342 341 333 331R2 17.99% 17.88% 15.06% 13.56% 20.89% 20.87%

To examine whether there is an inverted U relation between institutional ownershipand dividend yield (or dividend payout ratio), the regression is re-estimated after

26 In a univariate analysis in Section 4(ii), we find the mean institutional ownership in dividend payingand non-dividend paying firms to be 3.36% and 1.24%, respectively with the difference being statisticallysignificant at the 1% level.

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 221

Table 7 (Continued)

Panel B: Pooled Regression with Year-by-Year Regression Including Quadratic Terms of DY (or DPR)Institutional Ownership (%)

Year-by-Year Regressions

Pooled Regression 2000 2001

Intercept −4.48 −5.11 −4.01 −4.84 −4.60 −4.94(t = −3.51∗∗) (t =−4.00∗∗) (t =−2.13) (t = −2.55) (t =−2.61∗∗) (t =−2.83∗∗)

Log (Market-Cap) 0.35 0.39 0.32 0.37 0.37 0.38(t = 5.20∗∗) (t = 5.82∗∗) (t = 3.15∗∗) (t = 3.69∗) (t = 3.98∗∗) (t = 4.11∗∗)

Beta 0.27 0.24 0.42 0.37 0.16 0.14(t = 1.45) (t = 1.3) (t = 1.32) (t = 1.15) (t = 0.71) (t = 0.63)

Book to Market −0.03 −0.03 −0.019 −0.013 −0.043 −0.041(t =−0.95) (t =−0.97) (t =−0.40) (t = −0.28) (t =−0.99) (t =−0.95)

Annual abnormal 0.11 0.11 0.048 0.047 0.311 0.28Return (t = 0.67) (t = 0.67) (t = 0.22) (t = 0.22) (t = 1.05) (t = 0.95)

DY 53.71 45.38 64.44(t = 3.71∗∗) (t = 2.27∗) (t = 2.97∗∗)

DY 2 −552.51 −478.76 −664.06(t = −3.66∗∗) (t =−2.36∗) (t = −2.86∗∗)

DPR 0.72 0.002 2.33(t = 1.95) (t = 0.00) (t = 3.14∗∗)

DPR2 −0.07 −0.008 −0.50(t =−1.87) (t = −0.15) (t =−2.42∗)

DUMMYif FF 1.30 1.69 1.49 2.04 1.02 1.12(t = 3.75∗∗) (t = 5.39∗∗) (t = 2.82∗∗) (t = 4.25∗∗) (t = 2.22∗) (t = 2.66∗∗)

Obs 675 672 342 341 333 331R2 16.45% 15.16% 13.54% 12.06% 19.32% 19.70%

Notes:This table reports estimates of the following pooled OLS regression.

Institutional ownershipkt = α + β0Log(Market-Capkt ) + β1Betak + β2BtoMkt + β3AbRETkt

+ β4DUMMYIf PAYINGkt + β5DYkt (or DPRkt ) + β6DUMMYIf FFkt + εkt

The dependent variable of institutional ownershipkt is institutional share ownership as of December 31 ofyear t (t = 2000, 2001) for a stock k, as a percentage of total shares outstanding. The control variablesare (i) Log(Market-Capkt ), (ii) Betak obtained from Reuters, (iii) Book-to-Market ratio, BtoMkt from SIRCAand (iv)Annual abnormal return, AbRETkt calculated as the annual return on the stock in year t minusits expected return given by the CAPM. In addition to the control variables, the regression includes: (v)a dummy variable, DUMMYIf PAYINGkt , which takes a value of one if a stock pays dividends during yeart and zero otherwise, (vi) dividend yield, DYkt (or dividend payout ratio, DPRkt ), and (vii) a dummyvariable, DUMMYIf FFkt , which is set to one if a stock pays fully franked (FF) dividends during year t andzero otherwise. In Panel A, annual regressions were undertaken for 2000 and 2001 and the two years ofdata were also combined in a pooled regression. In Panel B, the above regression is re-estimated afterremoving a dummy variable, DUMMYI fPAYINGkt and including the quadratic terms for DY (or DPR). ∗,∗∗denote significance at the 5% and 1% levels, respectively.

including a quadratic term for DY (or DPR). Adding the quadratic dividend yieldto the regression results in the inclusion of four dividend related variables that havecorrelations in excess of 0.5 (DY, DY2, DUMMYIfPAYING and DUMMYIfFF .) Particularlygiven the high correlation between DY and DY2 of 0.92 it is not surprising to find thatmulticollinearity becomes a problem and the condition index rises above 30. A similarbut less severe problem was also evident on adding DPR2 to the regression including

C© 2010 Blackwell Publishing Ltd

222 JUN, GALLAGHER AND PARTINGTON

payout ratios. The multicollinearity problem was mitigated by dropping DUMMYIfPAYING

out of the regressions which reduced the condition index below 30.27 The results ofthe regression are given in Panel B of Table 7.

Focusing on the coefficients for DY and DY2, a significant positive coefficient on DYand a significant negative coefficient on DY2 confirm an inverted U relation betweeninstitutional ownership of firms and dividend yield. The turning point implied bythe quadratic regression is a 4.86% dividend yield. This is consistent with the resultsof the analysis in Section 4(i)(b) which shows a significant decrease in institutions’actual portfolio weight from the third dividend yield quintile to the fourth dividendyield quintile where the average dividend yields were 4.6% and 6% respectively. Theevidence for a similar inverted U relation in regard to the payout ratio is weaker. Thecorrect signs are observed in all regressions but only the 2001 results are significant.

5. SUMMARY AND CONCLUSION

This study examines whether dividends affect Australian institutional portfolio hold-ings using a sample of institutional equity funds. Dividend clientele theories hy-pothesise that institutions are attracted to dividends for two reasons. First becausedividends are tax advantaged for institutions and second because of ‘prudent-manrule’ restrictions. This study provides some capacity to distinguish between these twohypotheses. Under the Australian imputation tax system unfranked dividends aretax disadvantaged, while franked dividends are clearly tax advantaged for pensionfunds, and possibly tax advantaged for unit trusts. Evidence that funds are attractedto franked dividends and avoid unfranked dividends is clear support for the taxhypothesis. In contrast the prudent-man hypothesis predicts a preference for dividendsindependent of franking status.

In studies of institutional dividend preferences there is often an issue of causality.It is not certain whether the institutions buy the stocks because they pay dividendsor whether the firms pay dividends because they observe institutions on their shareregister. Because the key issue in this study is the preference for franked dividends thecausality issue is less serious. The level of franking of dividends was highly regulatedand therefore managers could not arbitrarily increase the level of franking when theyobserved institutions on their share register. The level of franking depended on theextent to which profits were taxed. Although we cannot rule out managers increasingfranking by paying more taxes this would be costly.

The results show that institutions are overweight in dividend paying stocks, butthat this is driven by the excess weight in stocks paying fully franked dividends. Bothpension funds and unit trusts are overweight in stocks paying fully franked dividendsand are underweight in stocks paying unfranked dividends. These results are robustto the inclusion of controls for other variables, such as firm size, that might affectthe weight funds invest in particular stocks. The evidence therefore supports thetax advantage hypothesis in explaining funds’ dividend choices. It is clear that theprudent-man hypothesis is not operating on its own. However, we cannot rule out itsoperation jointly with the tax hypothesis. Prudent-man rules might lead to a preference

27 DUMMYIfPAYING is chosen for deletion since the DY and DY 2 have to be in the regression to test thehypothesis of an inverted U relation between ownership and dividend yield. This leaves a choice betweenDUMMYIfFF and DUMMYIfPAYING for deletion, but if the firm pays a dividend this will be reflected in adividend yield above zero, consequently the least information is lost by deleting DUMMYIfPAYING .

C© 2010 Blackwell Publishing Ltd

INSTITUTIONAL DIVIDEND CLIENTELES 223

for dividends, which because of the tax advantage manifests itself as a preference forfranked dividends.

There were tax changes over the period of the study. These tax changes wereassociated with changes in the extent to which funds were overweight in fully frankeddividends. However, it is not clear whether the changes in funds’ holding wereexplained by the tax changes or by an increasing trend in the proportion of firmspaying fully franked dividends.

Institutions did not prefer the stocks paying the highest dividends. This result isbroadly consistent with the US empirical evidence (Del Guercio, 1996; and Grinsteinand Michaely, 2005). There was, however, an inverted U relation between institutionalownership of individual firms and dividend yield. We also note that the highest yieldingand highest payout quintiles appeared to have dividend yields that would be difficultto sustain and an average payout ratio that would be impossible to sustain. This mayexplain why the highest dividends do not attract institutions.

The analysis of why institutions were not attracted to the highest yield stocks, andthe estimation of the inverted U relationship between ownership and yield, wereundertaken ex-post. Independent confirmation by an ex-ante analysis on new datawould therefore be desirable when such data becomes available.

REFERENCES

Ainsworth, A., K. Fong and D. R. Gallagher (2008), ‘Style Drift and Portfolio Managementfor Active Australian Equity Funds’, Australian Journal of Management, Vol. 32, No. 3,pp. 387–418.

Allen, F., A. E. Bernardo and I. Welch (2000), ‘A Theory of Dividends Based on Tax Clienteles’,Journal of Finance, Vol. 55, No. 6, pp. 2499–536.

Anderson, H., S. Cahan and L. Rose (2001), ‘Stock Dividend Announcement Effects in anImputation Tax Environment’, Journal of Business Finance & Accounting , Vol. 28, Nos. 5&6,pp. 653–69.

Armitage, S., L. Hodgkinson and G. Partington (2006), ‘The Market Value of U.K. Dividendsfrom Shares with Differing Entitlements’, Journal of Business Finance & Accounting , Vol. 33,Nos. 1&2, pp. 220–44.

Ashton, D. (1989), ‘The Cost of Capital and the Imputation Tax System’, Journal of BusinessFinance & Accounting , Vol. 16, No. 1, pp. 75–88.

Bajaj, M. and A. M. Vijh (1990), ‘Dividend Clienteles and the Information Content of DividendChanges’, Journal of Financial Economics, Vol. 26, No. 2, pp. 193–219.

Beggs, D. and C. Skeels (2006), ‘Market Arbitrage of Cash Dividends and Franking Credits’,Economic Record, Vol. 82, No. 258, pp. 239–52.

Bell, L. and T. Jenkinson (2002), ‘New Evidence on the Impact of Dividend Taxation on theIdentity of the Marginal Investor’, Journal of Finance, Vol. 57, No. 3, pp. 1321–46.

Binay, M. (2001), ‘Do Dividend Clienteles Exist: Institutional Reaction to Dividend Events’,Working Paper (Department of Finance, University of Texas at Austin).

Brands, S., D. R. Gallagher and A. Looi (2006), ‘Active Investment Manager Portfolios andPreferences for Stock Characteristics’, Accounting and Finance, Vol. 46, No. 2, pp. 169–90.

Brav, A. and J. B. Heaton (1998), ‘Did ERISA’s Prudent-man Rule Change the Pricing ofDividend Omitting Firms?’, Working Paper (Duke University).

Brown, C. A. and J. W. O’Day (2005), ‘Payout Policy in Australia’, Working Paper (University ofMelbourne, Parkville).

——— ——— (2006), ‘The Dividend Substitution Hypothesis: Australian Evidence’ (Availableat SSRN: http://ssrn.com/abstract=911508).

Cannavan, D., F. Finn and S. Gray (2004), ‘The Value of Dividend Imputation Tax Credits inAustralia’, Journal of Financial Economics, Vol. 73, No. 1, pp. 167–97.

C© 2010 Blackwell Publishing Ltd

224 JUN, GALLAGHER AND PARTINGTON

Chu, H. and G. Partington (2001), ‘The Market Value of Dividends: Evidence from a NewMethod’, Accounting Association of Australia and New Zealand Conference (Auckland).

Cummings, J. and A. Frino (2008), ‘Tax Effects on the Pricing of Australian Stock Index Futures’,Australian Journal of Management, Vol. 33, No. 2, pp. 391–406.

Del Guercio, D. (1996), ‘The Distorting Effect of the Prudent-man Laws on Institutional EquityInvestment’, Journal of Financial Economics, Vol. 40, No. 1, pp. 31–62.

Dempsey, M. and G. Partington (2008), ‘Cost of Capital Equations under the AustralianImputation Tax System’, Accounting and Finance, Vol. 48, No. 3, pp. 439–60.

Denis, D. J., D. K. Denis and A. Sarin (1994), ‘The Information Content of Dividend Changes:Cash Flow Signaling, Overinvestment and Dividend Clienteles’, Journal of Financial andQuantitative Analysis, Vol. 29, No. 4, pp. 567–87.

Desai, M. A. and L. Jin (2010), ‘Institutional Tax Clienteles and Payout Policy’, Journal of FinancialEconomics (forthcoming).

Dhaliwal, D., M. Erickson and R. Trezevant (1999), ‘A Test of Theory of Tax Clienteles forDividend Policies’, National Tax Journal , Vol. 52, No. 2, pp. 179–94.

Elton, E. J. and M. J. Gruber (1970), ‘Marginal Stockholder Tax Rates and the Clientele Effect’,Review of Economics and Statistics, Vol. 52, No. 1, pp. 68–74.

Fedina, M. and T. Grammatikos (1993), ‘Risk Premia and the Ex-dividend Stock Price Behaviour:Empirical Evidence’, Journal of Banking and Finance, Vol. 17, No. 4, pp. 575–89.

Gallagher, D. R. and A. Looi (2006), ‘Trading Behaviour and the Performance of DailyInstitutional Trades’, Accounting and Finance, Vol. 46, No. 1, pp. 125–47.

Gillan, L. and L. T. Starks (2000), ‘Corporate Governance Proposals and Shareholder Ac-tivism: The Role of Institutional Investors’, Journal of Financial Economics, Vol. 57, No. 2,pp. 275–305.

Gompers, P. and A. Metrick (2001), ‘Institutional Investors and Equity Prices’, Quarterly Journalof Economics, Vol. 116, No. 1, pp. 229–59.

Graham, J. and A. Kumar (2006), ‘Do Dividend Clienteles Exist? Evidence on DividendPreferences of Retail Investors’, Journal of Finance, Vol. 61, No. 3, pp. 1305–36.

Grinstein, Y. and R. Michaely (2005), ‘Institutional Holdings and Payout Policy’, Journal ofFinance, Vol. 60, No. 3, pp. 1389–426.

Hodgkinson, L., K. Holland and R. Jackson (2006), ‘Dividend Valuation, Trading and Transac-tion Costs: The 1997 Partial Abolition of Dividend Tax Repayments’, Accounting and BusinessResearch, Vol. 36, No. 4, pp. 253–70.

Kalay, A. (1982), ‘The Ex-dividend Day Behaviour of Stock Prices: A Re-examination of theClientele Effect’, Journal of Finance, Vol. 37, No. 4, pp. 1059–86.

Karpoff, J. and R. Walkling (1988), ‘Short-Term Trading around Ex-dividend Days: AdditionalEvidence’, Journal of Financial Economics, Vol. 21, No. 2, pp. 292–8.

Lasfer, A. (1995), ‘Ex-Day Behaviour: Tax or Short-term Trading Effects’, Journal of Finance,Vol. 50, No. 3, pp. 875–97.

Michaely, R. and W. H. Shaw (1994), ‘The Pricing of Initial Public Offerings: Tests of the AdverseSelection and Signaling Theories’, Review of Financial Studies, Vol. 7, No. 2, pp. 279–319.

———, R. Thaler and K. L. Womack (1995), ‘Price Reactions to Dividend Omissions: Overreac-tion or Drift?’, Journal of Finance, Vol. 50, No. 2, pp. 573–608.

Miller, M. H. and F. Modigliani (1961), ‘Dividend Policy, Growth and the Valuation of Shares’,Journal of Business, Vol. 34, No. 4, pp. 411–33.

Perez-Gonzalez, F. (2003), ‘Large Shareholders and Dividends: Evidence from U.S. Tax Re-forms’, Working Paper (Graduate School of Business, Columbia University).

Scholz, J. K. (1992), ‘A Direct Examination of the Dividend Clientele Hypothesis’, Journal ofPublic Economics, Vol. 49, No. 3, pp. 261–85.

Shleifer, A. and R. W. Vishny (1986), ‘Large Shareholders and Corporate Control’, Journal ofPolitical Economy, Vol. 94, No. 3(1), pp. 461–88.

Strickland, D. (1996), ‘Determinants of Institutional Ownership: Implications for DividendClientele’, Working Paper (University of North Carolina at Chapel Hill).

Walker, S. and G. Partington (1999), ‘The Value of Dividends: Evidence from Cum-dividendTrading in the Ex-dividend Period’, Accounting and Finance, Vol. 39, No. 3, pp. 275–96.

C© 2010 Blackwell Publishing Ltd