Post on 24-Jun-2020
†Graduate School of Business, Columbia University, 3022 Broadway, Uris Hall, NY, NY 10027. Tel: 212.854.7631, fax: 212.316.9219, email mc2797@columbia.edu. ‡Owen Graduate School of Management, Vanderbilt University, 401 21st Ave S., Nashville, TN 37203-2422. Phone: 615.343.9387, fax: 615.343.7177, email: Jacob.Sagi@owen.vanderbilt.edu. *College of Business, University of Illinois at Urbana-Champaign, 1206 S. 6th Street, Champaign, IL 61820. Phone: 217.265.6598, email: zhiwang@illinois.edu. JEL Classification: G12 Keywords: Closed-end funds, CEFs, managed distribution plans, MDPs.
A Neoclassical Model of Managed Distribution Plans:
Theory and Evidence
Martin Cherkes†, Jacob S. Sagi‡, and Jay Wang*
November, 2009
A Neoclassical Model of Managed Distribution Plans:
Theory and Evidence
Abstract
Jensen (1986) identifies the need to motivate managers to distribute funds that earn a ‘below-market’ rate of
return as a major problem in corporate finance. Equity closed-end funds (CEFs) provide an example of how
capital markets perform this function. CEFs exist to provide investors with portfolio services that investors
cannot easily obtain on their own (e.g., liquidity or superior stock picking ability). When a fund does not
convincingly provide these services, it trades at a discount to its net asset value (NAV) because, through their
fees, managers split the total value of the fund with investors. A Managed Distribution Plan (MDP), where
investments might be partially liquidated to increase investors’ cash flows, lowers the value of the manager’s
claim on the assets of the fund. This is a direct transfer of wealth from the manager to the shareholders á la
Jensen, and will be adopted by managers who fear an eventual liquidation of the fund via a proxy vote. We
model the threat of such liquidation through the intermediation of an activist shareholder. Among other things,
our model predicts that MDPs are more likely to be adopted by funds that appear to be less effective in
providing portfolio services to their investors and that are relatively easy to liquidate or ‘attack’. We test the
model on a panel of 236 CEFs and find good agreement with our model .
2
1. Introduction
Closed-end funds1 (CEFs) attract the attention of financial economists because they are
among the simplest corporate organizations that offer easy data availability and puzzling presence
of unexplained regularities. A recent puzzling regularity, mainly appearing among equity closed-
end funds, is the presence of a minimum dividend commitment: the so-called managed distribution
plan (MDP).2 Under an MDP, the fund’s management commits to a minimum annual payout of,
say, 10% of the net asset value (NAV). Should the fund’s annual earnings be lower than 10%, the
shortfall is funded by partial asset liquidation. Funds with MDPs experience an average post-MDP
increase in price resulting in a smaller discount relative to NAV. Wang and Nanda (2008)
document that during the 1990-2006 period MDP funds had a median monthly discount of 0.86%,
while funds without any MDP had a discount of 10.19%.
In a Modigliani-Miller setting (i.e., absent market frictions), dividend policy is irrelevant.
What are the market frictions that explain the impact of a minimum dividend policy on the
closed-end fund’s value? Johnson, Lin and Song (2006) explain the MDP impact via an
asymmetric information argument: The promised dividends are a costly signal by “good”
managers to differentiate themselves from “bad” managers. Wang and Nanda (2008), however,
show that a signaling explanation is unlikely to be the complete story. Post-MDP CEF
performance is stronger only for funds with moderate payout targets (MDPs below 10%), even
though the improvement in discount is most pronounced for aggressive payout funds (MDPs at
1 As with open-end mutual funds, closed-end funds typically invest in publicly traded securities and manage their holdings for income and capital appreciation. Unlike open-end funds, which continuously sell and redeem at the net asset value (NAV) per share, closed-end funds raise capital by selling publicly traded shares in an IPO. Subsequently, the shareholders of closed-end funds can sell their shares on the secondary market at prices that may vary significantly from NAV – this is the so-called closed-end fund puzzle. 2 Out of 188 equity CEFs in existence in 2006, 75 had in place a managed distribution plan (see Table 1 in Wang and Nanda, 2008).
3
10% or higher). For aggressive payout funds, where the signaling effect is supposed to be the
strongest, there is no discernible improvement in NAV performance. Moreover, funds with
aggressive MDPs shrink in size, further undermining a signaling explanation.
In this paper, we argue that MDPs can induce a wealth transfer between the management
and its shareholders, and (absent an adverse selection motive with respect to managerial ability)
will only be adopted under the threat of shareholder action.3 To provide heuristic support for this
view, consider that a CEF management team providing no value to its shareholders can only be
dismissed through costly shareholder action. Suppose the CEF pays shareholders a perpetual
dividend of 1% and pays the management a perpetual fee of 1% (both a percentage of NAV). Then
the management has a claim to half the value of the assets. On the other hand, if assets are
liquidated or reinvested so that the dividend is increased to 2% of NAV, while keeping the
management compensation at 1%, the management’s ownership is reduced to one third of NAV.4
Why would managers agree to such a wealth transfer in practice? One answer is that the
management may fear a more drastic wealth transfer. Funds usually trade at a discount prior to the
MDP adoption, and the adoption of an MDP is often preceded by an attempt from a large
shareholder to open-end or liquidate the CEF.5 Because open-ending or liquidation would remove
the discount, a large shareholder may have both the profit-motive and the means to press for such
an outcome through a costly proxy fight. Instead of risking full liquidation, managers may
rationally give up part of their rents to influence the probability of the proxy fight success. An
3 The Johnson et. al. (2006) signaling hypothesis may well be applicable to a subset of MDPs to which our model of MDP adoption is not immediately applicable. Nearly half of the MDPs adopted by CEFs over our observation period correspond to those adopted at inception, when the fund trades at a premium. We are careful to separate these types of MDPs in our empirical work, although our results appear robust to their inclusion. 4 The impact of payout policy on CEF valuation is also documented by Pontiff (1996) who finds that closed-end funds paying smaller dividends exhibit larger discounts. 5 Bradley at al. (2008) documents 127 such attempts during year 1988-2002 (see their Table 3), although they do not relate the attempts or their eventual success to the adoption of MDPs. During the years 1999-2002 close to one third of equity CEFs experienced such “attacks.” We thank Wei Jiang for generously providing us with their data.
4
MDP reduces the economic incentive to mount an activist’s challenge (in the case of preemption)
or a successful proxy fight to liquidate the fund’s assets.
To better understand the tradeoffs and their empirical implications, we develop a simple
sequential game representing the interaction between a CEF manager, an activist, and the
shareholders. The cost of a large share acquisition, the cost of mounting a proxy fight, and the cost
of fund liquidation are stochastic parameters with known distribution functions whose values are
realized just before a decision has to be made. The manager first decides whether to establish a
preemptive MDP plan to discourage the emergence of an activist. Next, a prospective activist
decides whether to acquire a position in the fund based on information acquired about the costs of
obtaining a toehold. If the activist enters the game, the manager may respond by changing the size
of the MDP; in turn, the activist (having meantime learned more information) decides whether to
initiate a proxy fight. Ultimately, if a proxy fight is initiated, the shareholders vote to liquidated the
fund and the game is over. Shareholders also play a crucial role in the initial decision of the activist
to mount an “attack” by setting the market price (i.e., the discount) of the CEF. If an attack and
subsequent liquidation is very likely, then shareholders will set the CEF price closer to its NAV,
thereby reducing the benefits to the attacker. If shareholders believe that an attack is unlikely, then
the CEF discount will be large, thereby enticing an activist to initiate a challenge. This feedback,
also considered by Edmans, Goldstein, and Jiang (2009) in the context of corporate takeovers,
injects a nontrivial component into the analysis of the equilibrium decision to adopt an MDP, and
motivates the use of a formal model.
Several equilibrium outcomes are possible, depending on the fund’s characteristics. If the
cost of acquiring a position in the fund is low relative to waging a proxy battle, the management
will choose a preemptive MDP to deny activists a cheap option to mount a proxy battle. Otherwise,
5
the management will only adopt an MDP in reaction to a challenge by an activist, and this only if
the economic incentives for mounting a successful proxy battle are sufficiently high. These
economic incentives increase with management fees and the size of the activist’s position, and
decrease with the fund’s payout policy, managerial value-added, and costs of liquidating the funds’
assets. The fund discount is an endogenous function of the same parameters influencing the MDP
and the decision to adopt it.
Guided by the model insights, we analyze an exhaustive hand-collected data set of all
equity CEFs in existence between 1990 and 2006 and find strong support for the theory. While
managerial fees do not appear to substantially change after an MDP adoption, the growth rates of
assets under management significantly decline in the three years following the adoption of an
MDP (no such decline takes place for non-MDP funds). This confirms that MDPs reduce
managerial compensation and lends support to the wealth transfer effect we conjectured earlier.
In our panel, CEFs with pre-emptive MDPs are attacked less often by activists and have a lower
probability to be terminated.6 A Probit regression confirms that the likelihood of an attack
increases significantly with managerial fees, and is reduced significantly by the presence of an
MDP, share illiquidity, asset illiquidity, and managerial value-added. The same set of variables,
postulated in the model, have significant explanatory power in predicting both the likelihood of
MDP adoption and the amount of capital returned to shareholders. In particular, our empirical
evidence suggests that funds with more liquid shares and assets, higher management fees, lower
leverage ratio, and lower managerial ownership are more likely to adopt an MDP. We find
support for the model’s prediction that the tradeoff between the illiquidity of fund shares versus
the illiquidity of fund assets determined whether an MDP is more likely to be adopted 6 Johnson, et. al. (2006) also find that firms with a minimum dividend policy survive longer and interpret this as an incentive to signal. In our model this observation is also predicted, but instead of signaling quality the manager adopts the MDP to reduce the incentive for shareholder action.
6
preemptively or after an activist attack. Moreover, the size of an activist’s toehold is strongly
related to the MDP target payouts following an attack.
Beyond the obvious relation to the existing literature on CEF MDPs (Johnson, et. al.,
2006; and Wang and Nanda, 2008), our paper explains the impact of CEF payout policy by
linking the literature that deals with rational explanations of CEF discounts with the activist
investor literature. As in Ross (2002), Berk and Stanton (2007), and Cherkes, Sagi, and Stanton
(2009), we assume that the value of a CEF is a function of the tradeoffs between the fees
managers extract and the value they create for shareholders. While Berk and Stanton (2007)
assume fund termination has an exogenous trigger, we model the process explicitly and tie it to
economic fundamentals and empirical observables.
Shleifer and Vishny (1986) show that a blockholder has an economic incentive to
influence corporate policy even if doing so is costly to the blockholder. In this regard, our paper
is closest to Bradley et al. (2008) who investigate activist attempts to open-end CEFs when they
are at a deep discount, although they do not tie this to the adoption of MDPs. Our model predicts
the decline in the discount that they empirically find and that is also found in our data. In
addition, their paper identifies a deep discount relative to other funds (in the same asset class) as
the most important variable in predicting an attack. As mentioned earlier, we view the discount
as endogenously determined and find empirical support for more fundamental variables in
determining the discount, the MDP, and the likelihood of an activist attack.
The article proceeds as follows. We develop a formal model in Section 2, Section 3
describes data and presents the empirical tests of our model. We conclude in Section 4.
7
2. Model
2.1 Preliminaries
The closed-end fund (CEF) owns an asset whose date-t market value (net asset value or
NAV) is Ct and which provides a continuous dividend stream Δ×Ct. All market valuations are
performed on a risk-adjusted basis and interest rates are normalized to zero.7 For capital-markets
to be in equilibrium, it must be that the asset depreciates at a rate of Δ per unit time (i.e., Ct+τ =
Ct e−τΔ ) so that
Ct = PV(dividends) = ∫ Δ×Ct+τ dτ = Δ×Ct ∫ e−τΔ dτ . (1)
By law, the dividend Δ must be distributed to shareholders by the CEF, net of any fees. Consider
now a CEF with the following attributes: Shareholders receive an additional liquidating dividend
of δ ≥ 0, the manager has the ability to enhance the asset’s growth by an additional rate of α ≤ Δ,
and the manager charges a fee of k times the NAV (i.e., k × Ct). The variable, Δ+δ, can be
viewed as a managed distribution plan (MDP). If these conditions are maintained in perpetuity,
the total value of the asset, including the contribution of active management and without
deducting fees, is
Vt = PV(dividends) = ∫ (Δ+δ)Ct+τ dτ (2)
= (Δ+δ)Ct ∫ e− τ(Δ−α+δ) dτ = Ct × (Δ+δ)/(Δ−α+δ).
Through the stream of fees, the manager effectively owns a share of Vt. Assuming that these
policies are constant through time and there is no threat of liquidation, the market value of the
manager’s share is
7 These assumptions allow us to abstract from asset performance uncertainty which is not particularly germane to the points we wish to make.
8
Mt = ∫ kCt+τ dτ = kCt ∫ e− τ(Δ−α+δ) dτ = Ct × k/(Δ−α+δ), (3)
so the shareholders’ value is
Pt = Vt − Mt = Ct × (Δ−k+δ)/(Δ−α+δ). (4)
Consequently, the CEF premium is given by
premt = Pt / Ct − 1 = (α−k)/(Δ−α+δ). (5)
If k – α > 0 then a permanent increase in δ will lead to an increase in Pt and in the premium, as
well as a decrease in Mt. Thus, whenever the CEF is at a discount, a liquidating dividend can
serve as a method of transferring ownership from the fund manager to the shareholders, thereby
shrinking the discount. This simple comparative static illustrates the wealth transfer induced by
an MDP.8 What the simple calculation does not answer is why a manager of a CEF would agree
to a wealth transfer via an MDP, and how the market’s anticipation of such an event affects the
price of the CEF before the MDP is adopted. The model we next develop addresses these
questions.
2.2 Modeling Shareholder Activism
Suppose that k – α > 0, meaning that in the absence of a policy change, the fund would
trade at a discount according to Eq. (5).9 We are now going to consider the possibility of
shareholder action: Forced liquidation by an activist shareholder, together with the possibility of
a response by management via a change in the liquidating dividend. For tractability, we’ll
assume the following timeline:
8 Pontiff (2006) notes that dividends reduce the duration of an asset, thus reducing the holding costs for a potential arbitrageur. In our setting, the same effect is at work in reducing the value of the manager’s position. 9 Because they IPO at a premium, from the point of view of investors, it must be that k – αt < 0 for an IPO taking place at date t. Here, we only consider funds that are seasoned, so that the value contributed by the management has settled to some kind of a steady state, α, which is assumed to be below the fees charged.
9
There are five key dates, denoted by t = 0,1,2,3, and 4. While we continue to assume that
dividends and fees are paid continuously, for the sake of analytic tractability, we assume that no
payoffs are made between dates 0 and 4 (and that this is reflected in the NAV of the underlying
asset). This is tantamount to assuming that the flow of payoffs from the asset under management
between the date of a pre-emptive MDP announcement and a proxy battle is small relative to the
total asset value.
t = 0: The CEF management inherits a fund with parameters k, Δ, and α. It can then change δ
(the baseline liquidation policy) such that δ0 ≥ 0. The CEF’s market value is set to reflect the
subsequent possibility of a hostile attempt to liquidate the fund or force the management to
change its policies.10
t = 1 assumptions: A single activist can decide to pay χ1 × C1 to initiate an attack and acquire γ
shares of the fund, up to a limit of 1γ ≤ . The proportional cost, 1 1[0, ]χ χ∈ , is distributed
uniformly and revealed to the activist at date 1 prior to deciding whether or not to initiate an
attack; χ1 represents the cost of information acquisition, opportunity costs, and setting up the
minimal necessary infrastructure in preparation for a decision on whether to initiate a proxy
fight.11
t = 2 assumptions: The manager can react to the attack (should it take place) by selecting a
liquidation dividend, δ2, to maximize the value of his or her contingent cash flows.12 If an attack
10 The requirement that δ0 ≥ 0 follows from the fact that δ is a liquidating dividend and that, by law, the fund cannot retain payouts or realized capital gains from its underlying assets. 11 We assume that the probability of initiating an attack is one-half if the NPV of doing so is zero. 12 One can also consider a lump-sum liquidation of the fund or a decrease in the management fee. If α is sufficiently small and the manager discounts his undiversfiable CEF cash flows at a rate sufficiently greater than the market rate of 0, then one can show that the manager will find it optimal to transfer value to shareholders exclusively via a flow of liquidation dividend (i.e., δ>0). The intuition is as follows: If the manager discounts cash flow at a rate higher than the market, then he or she will always prefer to transfer to shareholders a long- rather than a short-duration cash
10
does not take place, then the CEF policies determined at date 0 are made permanent and no
further attack can take place.
t = 3 assumptions: The activist decides whether to proceed with a proxy fight. Initiating a proxy
fight entails an expenditure of where C3 is the value of the asset at date-3, and 3 3[0, ]χ χ∈
is revealed to the attacker at date-3 prior to determining whether or not to initiate a proxy fight.
From the point of view of dates earlier than date-3, is distributed uniformly. If the attacker
backs down, the fund continues indefinitely under the policies determined at date-2.
t = 4 assumptions: The shareholders vote to liquidate the fund (if a proxy fight was initiated at
date-3) and receive a liquidating dividend of 1 , where C4 is the value of the asset at
date-4, or continue with the fund indefinitely under the policies adopted by the management at
date-2. The liquidation cost, F is known to the shareholders prior to voting, but it is uniformly
distributed on 0, (with 1 from the point of view of dates prior to date-4. We make the
following assumptions about the parameters:
3 32 .Fχ χγ γ
< < (6)
This set of inequalities ensures that the probabilities of initiating a proxy fight and the
subsequent probability of liquidation are interior.
While the model is not fully dynamic, in that we only budget for a single opportunity at
activism, our sense is that the economic tradeoffs can be equally well illustrated in our simpler
flow, when the two flows have the same present value. Thus, when transferring wealth to shareholders, the manager prefers a liquidating dividend to a lump-sum payment or to a cut in fees (assuming all three transfer the same value from the manager to shareholders). When α is large, this intuition breaks down because a liquidating dividend is more inefficient than a fee reduction, as it both shifts ownership and reduces the overall value the manager brings to the asset through active management. CEFs do not exhibit a change in management fee subsequent to the adoption of an MDP, nor are they prone to unforced large-scale redemption of capital. Thus, by electing to only model the transfer of value through a liquidation dividend we are not sacrificing a great deal of realism.
11
setting. We solve the model via backward induction. We summarize the results below, leaving
the proofs of the more involved arguments to the Appendix.
t = 4: Using the expression for Pt in (4), with the policy δ2, the shareholders liquidate the
fund if and only if the proceeds from liquidation exceed the continuation value of the fund; i.e., if
and only if 1 , where δ2 is determined at date 2. Notice that shareholders will
never liquidate a fund unless k > α, which is assumed. Moreover, liquidation takes place if and
only if the realized cost is less than the CEF discount, (assuming the CEF is continued).
Thus, the probability of liquidation before F is revealed is
ℓ , 1 . (7)
Where , is the smaller of a and b. If the discount is higher than then liquidation takes
place for certain. Because , there is no finite value of that can rule out liquidation in
case of a shareholder vote.
t = 3: The activist will proceed with a proxy fight if the present value of such a fight, as
calculated by the activist, is higher than the present value of accepting the policies adopted by
the manager at date-2. A proxy fight will therefore take place if and only if
ℓ proceeds from liquidation|the fund is liquidated
1 ℓ CEF continuation value CEF continuation value,
Where C3 is the date-3 NAV. The first term on the left side of the inequality is the benefit to the
activist from liquidation, weighed by the probability that shareholders vote for liquidation; the
second term is the weighted benefit in case of a failed vote. The third term is the activist’s cost of
proceeding with a proxy battle. The right side of the inequality is the benefit to the activist from
taking no action. Conditional on liquidation, the liquidation cost is distributed uniformly between
12
0 and , with an expected value of . Thus, the expected proceeds from liquidation,
conditional on liquidation, are given by 1 . Plugging in, a proxy fight will take
place if and only if,
ℓ 2 . (8)
If 3χ is sufficiently low, a proxy fight will take place for sure. Moreover, because ℓ 0
and , there is no finite that rules out a proxy fight. We note that the inequality in (8) can
be recast in terms of the discount of the fund, should it be allowed to continue without a contest,
as ℓ 2 . I.e., an attack takes place if the future discount and probability of
liquidation are sufficiently high. Both of these quantities, however, are endogenous and
determined by the management’s choice of . Before is revealed, the probability of a proxy
fight can be calculated from (8) as
ℓ , 1 . (9)
The second inequality in (6) implies that the likelihood of drawing a high (cost of initiating a
proxy fight) is high enough to guarantee that ℓ .
t = 2: The manager sets by maximizing the value of his cash flows as follows
max 1 ℓ . (10)
The term in the square bracket equals the probability of not having the fund liquidated (one less
the probability of a proxy battle and subsequent liquidation). The total payoffs to the manager in
the event of liquidation is zero because we assume no payoffs are made until date 4.
13
Proposition 1: At the optimum, ℓ , 1 . Let and .
Then
δ 1 Δ α , 0 , (11)
where , is the larger of a and b. If 1 then ℓ . Finally, the CEF
discount at date 2 is
1 if 1
otherwise. (12)
The parameter is a measure of the strength of the MDP response to an activist’s attack; from
the activist’s point of view, it measures the benefit of liquidation relative to the cost of a proxy
fight. The parameter is a measure of the costly frictions preventing liquidation, and is
therefore related to the discount subsequent to an optimal MDP response. The optimal MDP
response to an initial attack is increasing in k-α and the holdings of the attacker (i.e., γ), while it
is decreasing in the costs of initiating a proxy fight, in the cost of liquidation, and in the
mandatory level of dividends, Δ.13 Eq. (12) can be interpreted to say that the management
reduces the discount to the point where it is in line with the liquidation frictions ( can be
viewed as proportional to a harmonic mean of and , the frictions preventing liquidation).
One can also see that, given an interior solution for the MDP (i.e., 1), the discount is not a
function of the managerial expenses (i.e., k), asset payoffs (i.e., Δ), or managerial ability – this is
because, as is evident from (11), the optimal MDP acts to offset the negative impact of k on
shareholders and the positive impacts of Δ and α. We refer to and whenever . 13 Johnson et. al. (2006) found no instance of MDPs in the bond funds they examined. The large values of and Δ in bond CEFs (see Cherkes, et. al., 2009) may explain the relative absence of MDPs in this category of funds.
14
t = 1: The prospective attacker anticipates the manager’s reaction (i.e., δ ), and decides
whether to pay χ1 and initiate an attack by acquiring γ shares. Assuming shares are acquired,
γ can be between 0 and , the maximum ownership allowed before the SEC requires disclosure
of intent (e.g., 5%). The activist maximizes the following objective function:
δ ℓ δ proceeds from liquidation | the fund is liquidated
1 δ ℓ δ CEF continuation valueδ
2 .
P1 is the price of a share bought, while C1 is the NAV per share at date-1. The price is ‘pre-
attack’, reflecting the fact that the activist’s ‘attack’ is not observed prior to the purchase of
shares, and thus the value of the CEF doesn’t reflect that an attack has occurred until after the
shares are purchased (i.e., at date 2).14 The expected cost of a proxy-battle being initiated at date-
3 is (the probability of a proxy-battle being initiated times the expected cost conditional
on initiation). Using (11), one can write as,
1 if δ
1 1 if δ 0. (13)
In either case, is convex in , meaning that its maximum in 0, is at a corner. Setting
and :
, if 1 and 1 11
16 ,
, if 1 and 11516
,
0, otherwise.
(14)
14 Naturally, the P1 incorporates the market’s anticipation of a possible attack. This is analyzed in the t = 0 case.
15
One can now revisit Eq. (11) of Proposition 1 and substitute for in the various expressions.
Corollary to Proposition 1:
δ 1 Δ α , 0 , (15)
and the CEF discount at date 2 is
1 if 1
otherwise. (16)
In examining (14), it is important to note that if is sufficiently high, the activist will
not initiate an ‘attack’ even if 0. This is a crucial difference from the activist’s date-3
decision which always depends on the realized cost of a proxy battle. If 0 at date-3 then
initiating a proxy battle is a free valuable option for the activist, and a proxy battle will take
place. On the other hand, if 0 at date-1, then initiating an attack is not costless in
expectations as long as 0. In particular, because market participants free-ride on the activist
who pays the future costs of a proxy battle, and this is reflected in , the manager can set an
initial MDP policy to rule out an attack. Next, we analyze this in greater detail.
t = 0: Let δ be the probability that an activist will acquire shares at date 1. The
manager’s objective function, as a function of the liquidation policy, δ , is given by:
δ1 δ
Δ α δδ 1 ℓ
Δ α
If there is no attack at date-1, the manager’s capitalized fees are given by the first term. If there is
an attack, then the manager’s value function corresponds to the expression in (10). The following
result describes the optimal policy at date-0.
Proposition 2: Let 8 1. The optimal liquidation dividend policy is given by
16
δ 16
Δ α if 1 and
1615
1 Δ α if 1 and 1
0 otherwise.
(17)
The subsequent probability of attack is
δ
12 if δ 0 and 1
12
16 158 7 if δ 0 and 1
0 otherwise.
(18)
Thus, if δ 0 then δ 0 (i.e., preemption at date zero is decisive in the sense that it
eliminates the possibility of a future attack). Finally, the date-0 discount is given by
41 0 if δ0 0 and 1
11
163 if δ0 0 1
78
1 0 if δ0 0 and 115
16if δ0 0 1
(19)
The quantity measures the costs of initiating an attack relative to the future costs of a
proxy battle and eventual liquidation. If it is low, then according to Eqs. (17) and (18), the
manager will select a pre-emptive liquidation dividend that will ward off any future attack. On
the other hand, if is not low, the manager will choose to distribute the minimum amount of
payouts (i.e., 0).15 In the latter case, the probability of a date-1 attack by an activist is
strictly positive but also strictly less than one. The reason that the probability of attack must be 15 Managers will not adopt a pre-emptive MDP in the case of a high cost of attack alongside low costs of a proxy battle and eventual liquidation. This is because the manager, given the high costs of attack, will prefer to adopt the MDP after an attack is initiated.
17
strictly less than one is that the market prices it in, thus making it less profitable for the activist to
attack in the first place. This feedback also plays an important role in Edmans, Goldstein, and
Jiang (2009).
If an attack is precluded by a pre-emptive MDP (i.e., if δ 0), then the discount is
permanently set at the level indicated by Eq. (19). If an attack is possible, then the date-0
discount is strictly higher than the date-2 discount. Thus, an attack will be followed by a decline
in the discount even if the management does not adopt an MDP. Figure 1 summarizes the overall
MDP strategy and its impact on the discount, probability of attack, and subsequent probability of
a proxy battle and liquidation.
2.3 A Summary of the Model Predictions
From the previous analysis, and supported by examination of Figure 1, one can derive a
number of empirical predictions which we investigate in the next section.
A central premise of the model is that an MDP, if adopted when the fund is at a discount,
amounts to a wealth transfer between the management and the shareholders. This leads to the
following empirical prediction:
Prediction 1: An MDP should lead to a relative decline in managerial compensation and in the
fund’s discount.
The model also predicts that pre-emptive MDPs are effective in reducing the likelihood
of a subsequent attack and liquidation. In fact, in our simplistic setting, a pre-emptive MDP
completely rules out an eventual attack and liquidation. In practice, such corner solutions are
unlikely, but the intuition should be robust. The model also predicts that managers optimally
electing to not respond to an attack should experience fewer liquidations than reactive MDPs.
18
Prediction 2: Pre-emptive MDPs should be associated with a lower probability of attack and a
lower probability of liquidation. Reactive MDPs should be liquidated more often than funds that
optimally elect to not adopt an MDP in response to an activist attack.
A common feature of all the MDPs (whether preemptive or reactive) in Figure 1, is that
they weakly increase with Δ α α. Hence,
Prediction 3: The probability and magnitude of MDPs increase with managerial fees and activist
toe-hold. They decrease with asset liquidation costs and managerial value-added.
In Figure 1, 1 separates preemptive and reactive MDPs. Thus,
Prediction 4: Restricting attention to funds that adopt MDPs, those in which the CEF share
illiquidity is low relative to the underlying asset illiquidity and the cost of a proxy battle will be
more likely to adopt preemptive MDPs.
The Corollary to Proposition 1 derives the MDP target as a function of the model
parameters indicating that the activist toehold is an important variable.
Prediction 5: A ‘reactive’ MDP target (i.e., Δ δ ) should increase with the activist toehold.
Turning to more predictions concerned with activist attacks, one has:
Prediction 6: The likelihood of an activist attack increases with managerial fees, and declines
with managerial value-added, asset liquidation costs, acquisition costs, and proxy battle costs.
Moreover, the likelihood of an attack is lower in the presence of an MDP.
As Eq. (19) of Proposition 2 indicates, an attack should result in a decline in the discount.
Prior literature already documents the decline in the discount subsequent to the adoption of an
MDP. The proposition also implies a decline in the discount even for funds that do not
implement a ‘reactive’ MDP (see Figure 1).
19
Prediction 7: An activist attack is followed by a relative decline in the CEF discount whether or
not the fund counters with a reactive MDP.
3. Empirical Analysis 3.1 Sample Description
Our empirical analysis employs the CEF data collected by Wang and Nanda (2008),
consisting of 236 closed-end equity funds spanning the years 1990 to 2006. Monthly share
prices, distribution amount, and number of shares outstanding are from CRSP. Monthly discount
data are from Lipper and ETF Connect, and used to calculate monthly NAVs. We created a list
of activists by searching through proxy statements of funds that adopted MDPs to identify
shareholders that requested an MDP prior to adoption. We then created a database of “activist
attacks” by augmenting the data generously provided by Bradley, Brav, Goldstein and Jiang
(2008) with 13D/G SEC filings by any activist on our list. We also hand collect, from annual
reports, information on distribution, total assets, total liabilities, management fee, management
ownership, expense ratios, and unrealized capital gains. Where available, we obtain funds’
quarterly stock holdings from the CDA/Spectrum Database. Finally, we use Joel Hasbrouck’s
Gibbs measures for share illiquidity.16
If an MDP is adopted in a year prior to any activist attack year, we then define it as a
preemptive MDP. Otherwise it is a reactive MDP. Under this classification, there are 70
preemptive MDPs and 22 reactive MDPs in our sample.17 Of the former, 43 MDPs were adopted
at inception, meaning that the fund was technically trading at a premium when the MDP was
16 http://pages.stern.nyu.edu/~jhasbrou/Research/GibbsEstimates2006/Liquidity%20estimates%202006.htm 17 We also observe 16 cases of MDP terminations: 8 associated with fund discontinuation due to either liquidation, change of legal structure, or merger with other funds; the other 8 funds terminated the MDP and continue to operate as closed-end funds. However, 3 of them were later forced to restore the MDP under shareholder pressure.
20
adopted. Because our model is not technically applicable to such MDPs we are careful to
distinguish them in our empirical analysis. By contrast, only 3 out of the 49 seasoned or ‘non-
inception’ MDPs adopted the plan after a year that the fund traded at a premium. Consistent with
industry norms and with the prior work of Johnson et al. (2006) and Wang and Nanda (2008), we
use the 10% payout target as the cutoff for classifying moderate versus aggressive payout
policies. For MDP funds in our sample, the payout target ranges from a low of 5% to a high of
20%, with a median around 10% in most years. Table 1 presents the year-end statistics on the
number of CEFs in our sample, the number of MDP funds (preemptive and reactive), and the
median payout target for MDP funds during the period from 1990 to 2006. The Table illustrates
the rapid growth in MDP adoptions experienced since the late 1990s, and the shift from
predominantly aggressive to predominantly moderate MDPs. Domestic equity funds dominate
the sample and thus their attributes are reported separately.
3.2 Key Variable Definitions
The model identifies fundamental measures that ought to affect MDP adoption. Among these
are the fund’s distributions, the cost of share acquisition, the cost of liquidating the fund’s assets,
the value added by fund managers, and the costs imposed by the management.
Distribution Yields: We estimate the annual Distribution Yield as the total dollar amount of
annual distribution per share normalized by the year-beginning NAV. We further decompose the
total distribution into three sources: Net Investment Income, Realized Capital Gains, and Return
of Capital (partial liquidation).
21
Share Illiquidity: Using daily volume and return data, we calculate the Amihud Illiquidity
Ratio (Amihud, 2002) averaged over each year to measure the illiquidity of fund shares. We
require at least 100 trading days data available for computing the annual measure. The Amihud
Illiquidity Ratio measures the absolute change in share price due to a $1 increase in trading
volume – a higher ratio reflecting a larger price impact and thus less liquid fund shares.18
Asset Illiquidity: Our first measure is based on the First-Order Serial Autocorrelation
(AR1) coefficients estimated from monthly NAV returns in the rolling three-year periods. This
measure was first developed by Getmansky, Lo and Makarov (2004) to proxy for asset illiquidity
in the hedge fund industry. The idea is that funds holding less liquid (less frequently traded)
assets should exhibit higher serial autocorrelation in NAV returns. One advantage of using this
return-based asset illiquidity measure is that it can be readily estimated for all CEFs with a
sufficiently long NAV return series. A disadvantage is that it only indirectly proxies for the
underlying asset illiquidity.
We also construct a holdings-based measure by downloading from the CDA/Spectrum
Database all available asset holdings reported by the CEFs in our sample. For each fund that
reports holdings in any given year, we calculate the annual value-weighted Gibbs BMA and
Gibbs LCF illiquidity measures (see Hasbrouck, 2006) based on the report that accounts for the
highest percentage of total net assets (TNA) during that year. We require funds in any given year
to report holdings accounting for at least 50% of TNA, meaning that these more direct measures
are missing for a substantial number of fund/years.
18 We choose the Amihud Illiquidity Ratio over bid-ask spread for two reasons: (1) The bid/ask quotes are highly unreliable in the CRSP Stock Database; and (2) The Amihud measure captures how share price changes in response to the trading size, and is therefore more appropriate from the point of view of an activist attempting to gain a significant toehold in the CEF.
22
Managerial Skill As argued in Cherkes, Sagi and Stanton (2009), one managerial value-
adding channel is through pooling money from retail investors and investing in illiquid assets.
Hence, the various asset illiquidity measures discussed above not only reflect the costs of proxy
contests but may also be interpreted as managerial contribution to shareholder value.19 In
addition, fund managers can add value through superior stock picking. To measure managerial
stock-picking skills, we estimate the Carhart (1997) Four-Factor Adjusted Alphas based on the
monthly NAV returns observed during the previous three years. Finally, leverage at
advantageous rates, relative to individual borrowing rates, enhances the benefits that accrue to
shareholders. However, it also allows the management to make higher average payouts.20 We
construct the annual Leverage Ratio by computing the year-end ratio of Total Liability relative to
Total Assets applicable to common shareholders.
Managerial Entrenchment: We use annually reported Management Fees to proxy for the same
quantity in the model and obtain managerial ownership of fund shares from the annual proxy
statements filed with the SEC. Specifically, we compute on an annual basis the total number of
common shares beneficially owned by all board members and executive officers as a percentage
of the total number of common shares outstanding.
Other Control Variables: When investigating the determinants of MDP adoption, we also
control for fund characteristics such as year-end size, fund age, and year-end accumulated
unrealized capital gains. The threat from activist shareholders may decrease with size (as
measured by TNA) because it would be harder for individual investors to accumulate enough of
a toe-hold to mount a shareholders’ action. Older funds (as measured by age) may be more likely
19 Ramadorai (2009) finds strong support for the Cherkes, Sagi and Stanton (2009) model in hedge funds, whose organizational form closely mimics that of CEFs. 20 If an unlevered asset pays an average dividend of 5% to the CEF shareholders, then the same asset, leveraged at a debt-equity ratio of 1:1 with a debt service of 3%, would pay an average dividend of 7%. Thus leverage can make it easier for a CEF to commit to an MDP.
23
to implement MDPs or be attacked because, consistent with Berk and Stanton (2007), investors
require time to credibly conclude that the management’s added value is less than the managers’
fees. Finally, the accumulation of a large amount of unrealized capital gains may contribute to
the discount due to potential tax liabilities (Malkiel, 1977) and may also impact a fund’s MDP
decision.21 We normalize the year-end accumulated unrealized gains as reported in the annual
reports by the year-end TNA.
Table 2 presents summary statistics for the key variables as defined in the previous
subsection – both for the full sample and for the sub-sample of domestic equity funds. We
provide the mean (median) statistics over all fund and year observations for non-MDP, pre-
emptive MDP and reactive MDP funds. Preemptive MDP funds are further segmented into those
that adopted at inception versus those that adopted later.
3.3 Testing Prediction 1 – Managerial Compensation and MDP Adoption
Prediction 1 states that an MDP is effective I n reducing the fund discount because it
transfers to the shareholders some of the present value of managerial claims on fund assets.
Table 3A reports average management fees, TNA growth and discounts during the three-year
period before vs. the three-year period after an MDP adoption. For each MDP fund, we identify a
control group of non-MDP funds that have the same investment objective and compute the
averages during the same pre-MDP and post-MDP periods. We report the mean (median)
differences between the MDP group and the non-MDP control group as well as the difference-in-
differences. Panel A suggests that the proportion of NAV charged as a fee by management is
significantly higher for MDP funds both the pre- and post-MDP periods. However, there is no
21 The capital gains overhang may have two conflicting effects. On the one hand, it increases the liquidation costs relative to NAV, thereby discouraging an MDP or an activist’s attack. On the other hand, it may also discourage a manager from effectively using his or her stock-picking skills for fear of forcing a capital gains distribution and reducing the amount of assets under management.
24
evidence suggesting that MDP funds significantly change the management fees following the
MDP adoption. This is consistent with our model assumptions that the proportional fees do not
change (see also the discussion in Footnote 11).
Panels B and C show that compared to the non-MDP control group the average MDP
fund has significantly higher annual TNA growth rate and larger fund size before the MDP
adoption but not afterwards. The difference-in-differences test suggests that the relative mean
(median) TNA growth rate and fund size for the MDP group significantly decrease by 9.38%
(5.04%) and $65.56 million ($18.96 million) following the MDP adoption. The impact in dollar
terms on managers is a relative decline of $0.46 million ($0.33 million) per annum – significant
at the 10% level (see Panel D). Consistent with other studies, Panel E confirms that the mean
(median) drop in relative discount for the MDP group, following the MDP adoption, is 5.29%
(5.21%) – significant at the 1% level. In summary, we find no material change in percentage
management fee but a significant drop in asset growth following the MDP adoption, which
contributes to a relative decline in managerial compensation.
One concern with the above analysis is that the relative drop in managerial compensation
may be simply a size effect. In Table 3B, we use a matching algorithm based on fund assets
during the period from 2000 to 2006. For each of the 26 MDP funds that had MDPs in place at
the end of 2000 (some adopted prior to 2000), we identify a matched non-MDP fund with the
same investment objective and the closest TNA. We then examine the absolute and relative
change in TNA, managerial compensation, and discount during the subsequent 6-year period.
As shown in Panel A of Table 3B, the average TNA for the MDP funds is significantly
larger than the matched non-MDP funds in 2000. Strikingly, the mean (median) fund assets for
the MDP group shrank by $53 million ($40 million) per annum over the next 6 years. In contrast,
25
there is no discernible change in fund assets for the matched non-MDP group. The difference-in-
differences test suggests that, relative to the matched non-MDP group, the mean (median) TNA
for the MDP group declines by $54 million ($65 million) per year – significant at the 5% level or
better. Simultaneously, managerial compensation for the MDP funds declined significantly both
in absolute and relative terms. This loss, when capitalized, is significant for both the managers
and the shareholders. Panel C in Table 3B confirms the earlier finding about the discount..
Hence, the adoption of MDPs, while effective in reducing fund discounts, appears to take
a significant toll on managerial compensation. This supports the view of our theoretical model
(Prediction 1) that MDPs effectively reduce the managerial claim on fund assets and transfer the
managerial rent back to shareholders.
3.3 Testing Prediction 2 – MDP adoption, attacks and terminations
While in the model termination is abstracted to correspond only to asset liquidation
instigated by activist shareholders, in practice terminations occur in various forms. These include
liquidation, open-ending, or a merger with another CEF (usually in the same fund family). Each
of these cases may be deemed undesirable for managers, but less clearly so for mergers.22 Open-
ending is typically resisted by fund managers and may be accompanied by a takeover of the
board, while mergers can imply a de-facto change in management.23 For more discussion of
these various changes in control over a fund’s assets, see Bradley et. al. (2008). However,
terminations can occur for reasons not related to a contest between managers and shareholders.
To ensure consistency with the model, we only examine “hostile terminations” – those for which
22 Mergers sometimes take place when the TNA of the fund is too low to justify the fixed costs of management. On the other hand, aggressive MDPs can lead to the kind of decline in TNA that would ultimately justify a merger. 23 Even without the prospect of firing managers, open-ending transfers power to shareholders through redemption rights. In a neoclassical setting, funds will be withdrawn until the value added by managers equals their compensation (see Berk and Green, 2004). For a fund at a discount, this entails a loss in assets under management and thus a loss to the management.
26
there is evidence of a prior contest (i.e., terminations by funds that were “attacked” at some
point). Reasons for fund termination are obtained from the CRSP termination code (or obtained
by checking SEC filings). We only count fund terminations that can be classified as liquidations,
open-ending, or a merger with another CEF. We further restrict our analysis to the sub-sample of
CEFs with inception date prior to 2004. For MDP funds, we also require the MDP adoption year
to be prior to 2004. We then follow these funds until the end of 2006 to observe the presence of
activist attacks and fund terminations.24 Using this set of terminations, we define “hostile
terminations” as those in which the fund previously experienced an activist attack (as defined
earlier).25
Table 4 reports hostile termination rates for various definitions of “termination”.26
Because the model distinguishes between funds that were attacked and responded with an MDP
(or did not respond) and funds employing preemptive MDPs, we report termination statistics for
these groups separately. We also separately include funds that adopted MDPs at inception for
reference. The results are similar across definitions of “hostile termination”. The next-to-last row
indicates that the hypothesis that preemptive MDPs experience hostile termination at the same
rate as other funds is rejected in favor of the model’s prediction (i.e., that preemptive MDPs
reduce the likelihood of hostile termination). The hypothesis that reactive MDPs terminate at the
same rate as funds that optimally choose to not respond to attacks is just shy of being rejected
(see the last row in Table 4), but in a direction contrary to the model’s prediction. One possible
24 Termination takes time, and the two-year cutoff is there to prevent biasing the statistics for funds that incepted or adopted an MDP close to the end of our observation period. 25 Using our definition, we observed 36 hostile terminations and 6 non-hostile terminations in our restricted subset. Among the six non-hostile terminations, three were liquidated (apparently because their TNA fell to very low values), one was open-ended as required by its charter, and the remaining two were merged into other funds. 26 Bradley et. al. (2008) essentially view all terminations involving liquidations, open-ending, or mergers to be in the interest of activists and against the interest of managers. In Table 4 we also examine more restrictive definitions of terminations.
27
explanation (outside the model), is that managers do not always act optimally in response to
activist attacks, leading to a “suboptimal” frequency of contests.
Finally, and not reported in the Table, five of the 16 (31%) post-inception preemptive
MDP funds experienced attacks, as compared with 84 of the 160 (53%) funds that did not adopt
MDPs preemptively. The difference in rates of attack is significant at the 10% level and supports
the model hypothesis that preemption reduces the likelihood of attack.
3.4 Testing Prediction 3 – the cross-sectional determinants of MDP adoption
For any CEF f in year t, let MDPf,t be an indicator variable taking on the value 0 if the
CEF has no MDP in place in year t, the value1 if it has a moderate MDP in place, and the value 2
if it has an aggressive MDP in place. We begin by estimating the following ordered Probit model
with style fixed-effects:
MDP , I I ∑ β X , ε , . (20)
For firm f in year t, MDP*f,t is a normally distributed latent variable determining the value of
MDPf,t. The cutoffs (intercepts) are determined by I1 and I2. The explanatory variables (the
X , ’s) are lagged by one year unless otherwise stated and (except for the indicators) are
standardized annually by subtracting the mean and scaling by the standard deviation of all CEFs
in the sample. The model largely guides our choice of explanatory variables, defined in Section
3.2: year-end TNA; fund age; share illiquidity as measured by the Amihud Illiquidity ratio; year-
end leverage ratio; asset illiquidity measured by the first-order autocorrelation based on the
monthly NAV returns in the previous three years; management fee; an indicator variable that
equals one if management beneficial ownership exceeds 10%; year-end accumulated unrealized
capital gains; and the four-factor alpha based on the monthly NAV returns in the previous three
years. In some specifications, we also include an indicator variable for the presence of activist
28
attack in earlier years. While an endogenous variable that depends on the other variables on the
right side of the regression, the presence of an activist partly proxies for the presence of low
proxy battle costs, which we do not directly observe.
Table 5 presents the estimation results for Eq. (20), excluding funds that adopted an MDP
at inception.27 Standard errors are heteroskedasticity robust and clustered by fund. Results are
shown with and without the activist attack variable. In the regressions using “All Funds”, the
coefficient estimates are largely consistent with our model prediction. The presence of activist
attack significantly increases the likelihood of observing an aggressive (moderate) adoption by
21% (9%). Moreover, both share illiquidity and asset illiquidity have strong negative impact on
the probability of observing an MDP with statistical significance at the 1% level. A one standard
deviation increase in share illiquidity measure reduces the likelihood of observing both
aggressive and moderate MDPs by about 4%. Similarly, a one standard deviation in asset
illiquidity measure reduces the probability of observing both aggressive and moderate MDPs by
2%. These percentage changes are economically significant because the base probability for
observing an aggressive (moderate) MDP is about 7% (9%). The coefficient estimates for
leverage, management fee, management ownership, unrealized capital gains, and four-factor
alpha also have signs consistent with our model, although lack statistical significance.
The regression analysis for the sub-sample of Domestic Equity Funds produces even
stronger results. As shown in the last column of Table 5, the coefficient estimates for share
illiquidity, leverage ratio, asset illiquidity, management fee, and management ownership all have
signs consistent with our model predictions and statistically significant at the 5% level or better.
In terms of economic significance, we find that: a one standard deviation increase in share
illiquidity decreases the probability of observing aggressive and moderate MDPs by about 6% 27 The results presented in Tables 5 and 6 are robust to the inclusion of funds that adopted MDPs at inception.
29
each; a one standard deviation increase in leverage ratio decreases the likelihood by 5% each; a
one standard deviation increase in asset illiquidity decreases the probability by 4% each; a one
standard deviation increase in management fee increases the chance by 9% and 5%; and
beneficial ownership of more than 10% implies a drop in the probability of 10% and 14%.
Again, these percentage changes are economically significant because the base probabilities for
observing aggressive and moderate MDPs in Domestic Equity Funds are 11% and 18%,
respectively.
In summary, consistent with Prediction 3, we find that the likelihood of observing MDPs
decreases in Share Illiquidity, Asset Illiquidity, Leverage Ratio, and Management Ownership and
increases in the presence of Activist Attack and Management Fee.
Robustness: Instead of using the first-order autocorrelation based on NAV returns to
proxy for Asset Illiquidity, we use either Gibbs measure based on the reported asset holdings
(Hasbrouck, 2006). Because only domestic equity funds regularly report their portfolio holdings,
we restrict the analysis that uses a Gibbs measure to this sub-sample. The results, available upon
request, are remarkably similar to those in Table 5 for all variables of interests. Regardless which
Gibbs measure is used, the coefficient estimates is negative and statistically significant at the 1%
level. Given that the CEF sample used in our analysis is relatively small, we also investigate
whether our empirical results are sensitive to alternative standardization method. There is no
qualitative difference if instead of standardizing variables using the mean we use the median in
the Probit regression (results are available upon request).
Next we empirically examine the magnitude of MDP distribution. The model assumes
that the MDP dividend (i.e., δ) corresponds to discretionary capital returned to investors. Thus in
testing the model it seems reasonable to focus on discretionary distributions. Under US tax law, a
30
closed-end fund must distribute to shareholders all the net investment income (from dividend and
interest income) and at least 95% of the realized capital gains in order to maintain its corporate
tax exempt status. While managers have discretion over the type of assets they manage, and
therefore exert some influence over the fund’s investment income and realized capital gains, this
influence is either indirect or confounded by other objectives (e.g., portfolio rebalancing). On the
other hand, the return of capital through a dividend distribution is likely to be less related to the
fund’s overall investment strategy and more related to the effect we directly model. Indeed,
Table 2 confirms that funds without an MDP in place rarely return capital through dividend
distributions. Table 6 reports the results of a Tobit panel regression using funds’ Return of
Capital as a dependent variable and the same explanatory variables used in Probit regression of
Table 5. Here too, we estimate the regression separately for all funds in the sample and for
Domestic Equity Funds only. The results are largely consistent with Table 5 and Prediction 3. As
shown in the “All Funds” column, the presence of an Activist Attack is associated with a 4.04%
increase in annual return of capital. A one standard deviation increase in each of Share
Illiquidity, Asset Illiquidity, and Leverage Ratio is associated with a 3.28%, 2.20%, and 1.33%
decrease in the return of capital, respectively. Finally, a one standard deviation increase in the
accumulated Unrealized Capital Gains decreases the annual return of capital by about 2%,
further confirming our interpretation of the return of capital as a “last resort” in meeting payout
targets.
3.5 Testing Prediction 4 – Preemptive versus Reactive MDPs
According to Figure 1, the cost of obtaining a toehold versus the cost of waging a
successful proxy battle determines whether a fund is more likely to adopt an MDP reactively
31
versus preemptively. We use the Amihud share illiquidity measure to proxy for the cost of
obtaining a toehold, and we use leverage and the Getmansky, et. al. (2004) AR1 measure to
proxy for the cost of liquidation. We also calculate a measure of Relative Illiquidity, capturing
the tradeoff between fund-share and underlying-asset illiquidity: We first convert both share
illiquidity and asset illiquidity measures into ranks (between 0 and 1) relative to other funds and
then compute the rank differences. In addition, we employ a fund’s TNA as an alternative proxy
for share liquidity and its four-factor alpha to control for the dual role played by leverage.28 All
variables are measured as the average over the 5-year window from 3 years before to 2 years
after the MDP adoption year.
Table 7 reports estimation results from multivariate Probit regressions using only those
funds that preemptively adopted MDPs after inception. For consistency, variables other than the
Relative Illiquidity measure are also employed as percentage rank. The Amihud ratio and AR1
measures of illiquidity are not significantly related to MDP type, whether individually or
differenced in the form of the Relative Illiquidity measure. However, TNA is significant and has
the correct sign. Moreover, controlling for performance, leverage is positively related to
preemption. This is consistent with the view that leverage increases the cost of liquidation (and
eventual success of a proxy battle). One can interpret this as weakly supportive of the model.
3.6 Testing Prediction 5 – Reactive MDPs and toeholds
The model clearly implies that the magnitude of a reactive MDP is increasing in the stake
of the activist attacking the fund. We obtain MDP targets from Wang and Nanda (2008) and
toehold information from 13D/G filings. The scatter plot in Figure 2 clearly shows a positive
28 Leverage can both increase the value added by managers and the costs of liquidating the fund (i.e., waging a successful proxy battle). In particular, because debt-holders must be paid in full upon liquidation of the fund’s assets, leverage effectively amplifies the illiquidity costs to the CEF shareholders.
32
relationship. A simple univariate OLS regression of the target on the size of the toehold yields a
slope coefficient of 0.157, which is highly significant (t = 4.55) and strongly supportive of the
model. The slope would be much steeper if one were to only consider activist shares below 20%.
3.7 Testing Prediction 6 – Cross-sectional Determinants of MDP Attacks
In our model, the same set of economic fundamentals used to predict MDP adoptions are
also the ones that drive activist shareholders’ decision on whether and when to launch liquidation
or open-ending attacks. To test this, we estimate a Probit regression for the presence of an attack
using the same explanatory variables in Table 5 (also with style fixed effects). Here, the
dependent variable is an indicator variable that equals one if an attacker is present for fund f at
year t, and zero otherwise. In some specifications we also employ an MDP indicator variable that
equals one if the fund has an MDP in place in the previous year and zero otherwise.29 The
results are presented in Table 8. The estimation results are broadly consistent with our model
predictions. Overall, share illiquidity, leverage, and asset illiquidity are all negatively related to
the likelihood of being attacked. A higher management fee appears to increase the probability of
being attacked. The results are robust for the full sample and the sub-sample of domestic equity
funds, and provide further evidence that the determinants of shareholder activism are similar to
those behind MDP adoption.
3.8 Testing Prediction 7 – Shareholder Activism and the Discount
Bradley et al. (2008) document that the presence of activist attack leads to a decline in
fund discount. Our model generates similar predictions. Furthermore, our model predicts that an
29 The MDP indicator variable is an endogenous variable of the remaining explanatory variables. However, to the extent that the latter are imperfect proxies (and we do not have a proxy for the direct costs borne by the activist of waging a proxy battle) the MDP indicator can have explanatory power.
33
activist attack should be followed by a decline in the fund’s discount whether or not the fund
counters with a reactive MDP. To test this, we employ an event study similar to Bradley et al.
(2008) but distinguish between reactive MDP funds and non-MDP funds. Specifically, we
examine the change in discount during the 7-year window around the year when activists’ attack
occurred. We have a total of 122 documented attacks. For each fund under attack, we identify a
matched fund that has the same investment objective, has the most similar discount level 2 years
before the attack (Year -2), but has never been subject to activists’ attack during the event
window. We report the difference in discount of the attacked fund relative to the matched fund
for each year during the event window. We also separately report the relative discount changes
for the sub-sample of funds that reacted to an attack by adopting an MDP (46 cases) and the sub-
sample of funds that did not adopt MDPs (76 cases) as the response during the event window.
Table 9 presents the estimation results. During the three years prior to activist attack, the
attacked funds have a higher discount relative to the matched non-MDP funds. By the end of
year 3 following the attack, the mean and median discounts for the attacked funds are
significantly lower than the matched non-MDP funds. For the “All Attacks” sample, the mean
discount for the attacked funds is 4.6% lower than the matched funds. This is consistent with the
finding in Bradley et al. (2008), although the magnitude is smaller. For the “MDP” sample, the
difference in mean discount is more striking – around 7.4% lower than the matched funds. For
the “Non-MDP” sample (funds that were attacked but did not respond with MDP adoptions), we
also observe a decline in the mean discount but with smaller magnitude – around 2.8% lower
than the matched funds. In a broad sense, these results are consistent with Prediction 7. However,
the model (see Figure 1) suggests that the funds that did not respond to an attack with an MDP
ought to fare better (under the assumption that not responding is optimal). This appears related to
34
the observation in Section 3.3 that attacked funds that do not respond to an attack with an MDP
are terminated more often than those that respond with an MDP. In other words, in the narrow
sense of the model, not enough funds respond to an attack with an MDP adoption.
4. Conclusions
A managed distribution plan (MDP) where investments are liquidated to increase
investors’ cash flows lowers the value of the manager’s claim on the NAV of the fund. This is a
direct transfer of wealth from the manager to the shareholders á la Jensen, and will be adopted by
managers who fear an eventual liquidation of the fund via a proxy vote. We model the threat of
such liquidation through the intermediation of an activist shareholder. Our model predicts that
MDPs are more likely to be adopted by funds that appear to be less effective in providing
portfolio services to their investors and that are relatively easy to liquidate or ‘attack’. We test
the model on a panel of CEFs and find support overall. It indeed appears that MDPs are
associated with a relative loss in managerial compensation (as well as a relative increase in
shareholder value). Preemptive MDPs appear to be associated with fewer attacks and fewer
terminations. We find that funds with more liquid shares (proxies for lower trading costs) and
higher fees are more likely to adopt MDP; and funds with high asset illiquidity (proxies for
liquidation costs), high leverage ratios (a proxy for value added to shareholders), and high
managerial ownership are less likely to adopt MDP. The cross-sectional determinants of MDP
adoption coincide with those that invite activist attack and governing the magnitude of capital
returned to investors – both in the model and the data. Consistent with the model, we find that
preemptive MDPs are more likely than reactive MDPs when the cost of acquiring a toehold is
low in relation to the cost of waging a proxy battle. The only aspect of the model that does not
35
receive strong supported in the data concerns the decision to not adopt an MDP in response to an
activist attack. The model predicts that such funds, because the decision not to adopt is
presumably optimal, ought to exhibit fewer terminations and higher post-attack premia while the
data exhibits the opposite trend. Given the overwhelming support for the numerous other
predictions made by the model, one can be led to interpret the data to suggest that some
managers do not respond optimally to activist attacks (by adopting MDPs), instead choosing an
inefficient contest that results in higher discounts and more terminations than necessary.
Appendix
Proof of Proposition 1 It should be clear that the manager will set so that ℓ 1.
One need therefore consider the case in which ℓ 1 and the case
ℓ 1. As increases, it passes from the former region into the latter. In the first case, the
manager optimizes
max Δ α δ 1 2 Δ , (21)
for which the solution to the first order condition is δ Δ α , meaning that the
probability of liquidation is ℓ δ 1, by (6). Thus the optimum in the first case must be a
corner solution: the largest for which 1, and corresponding to
⁄Δ α . In the second case, the manager optimizes
max Δ α δ 12 Δ , (22)
36
and the maximand is given by in Eq. (11). Because is attainable in (22), is the optimum
and both ℓ and are interior at the optimum.
The value of the CEF (i.e., P2), per unit of NAV (i.e., C2), is given by
1 ℓΔΔ
ℓ
2 1ΔΔ , (23)
where the first term corresponds to the continuation value of the CEF in case of no eventual
liquidation, and the second term is the contribution from possible liquidation. Plugging in the
value for from Eq. (11), one obtains the desired result for the discount, 1 .
■
Proof of Proposition 2: The manager optimizes his objective function:
max1 δ
Δ α δδ 1 ℓ
Δ α , (24)
Where the δ is the probability of an initiated attack at date-1. δ solves
δ
115 ⁄
16 2 , 1 if
1 Δ 18 Δ , 1 if 0
(25)
where is zero if a < 0 and a otherwise. The date-1 CEF price is set by the market’s
anticipation of an attack to
37
1 δΔ δΔ α δ
δ 1α
Δ α δ 1 ℓ
2
(26)
The first term is the payoff if there is no attack, while the second is the expected payoff
conditional on an attack (and includes the expected payoff from liquidation). So the date-1 CEF
discount (prior to an attack) reflects the possibility of a sequence of attacks, as well as the
continuation value of the fund. Let Δ . Because and ℓ are interior,
one can use Proposition 1 to rewrite this as follows,
1 δ 1 δ if
1 δ 1 1 δ if 0. (27)
It’s straight forward to check that δ 1 is not a consistent solution, because by plugging
that into Eq. (27) and substituting the results back into (25) one obtains a negative probability
for δ . After some algebra, one can combine Eqs. (27) and (25) to solve for δ :
δ
if
if 0
. (28)
If δ 0 then the manager’s objective function is monotonically decreasing in δ . If
δ 0, then in each case in Eq. (28) the manager’s objective function can be written in the
form which is also monotonic in δ and admits only corner solutions. The corners are
determined by the smallest δ such that δ 0 and by δ 0. In the former case, to
38
completely rule out an attack, the manager has to consider the activist’s calculation in Eq. (14)
and set the dividend policy so that
11516 2 if δ
1 Δ 18 Δ if δ 0
.
If an attack has been completely ruled out, then , giving δ δ where
δΔ α if
1 Δ if 0. (29)
To see which solution dominates, consider the manager’s objective function, , at the two
corners:
δ 1516 α 2 if δ
Δ α 18 Δ if δ 0
.
whereas
0
Δ α 1 034 0 α 2 if δ
Δ α 1 00 1 2 Δ
Δ α if δ 0
.
39
with 0 given by Eq. (28) . Comparing the two managerial policies when δ 0, one obtains
that 0 δ if and only if 0 . This, however, is true only when . For the
case where an attack will be met by a response (i.e., δ 0), it is possible that 0
δ . The boundary is determined by 0. In particular, as 0, the δ 0 solution
prevails. On the other hand, as ∞, the δ δ solution dominates.
Finally, we note that under the assumption that Δ α , required
to ensure δ 0, it must be that and thus the probability of attack
0 is strictly positive.
■
References
Berk, Jonathan and Richard Green (2004). Mutual Fund Flows and Performance in Rational Markets, Journal of Political Economy, 112 (6),1269-1295. Berk, Jonathan and Richard Stanton (2007). Managerial Ability, Compensation, and the Closed-End Fund Discount, Journal of Finance 62 (2), 529-556. Bradley, Michael, Alon Brav, Itay Goldstein and Wei Jiang (2008). Costly Communication, Shareholder Activism, and Limits to Arbitrage, forthcoming, Journal of Financial Economics. Carhart, Mark (1997). On Persistence in Mutual Fund Performance, Journal of Finance 52, 57-82. Cherkes, Martin, Jacob Sagi, and Richard Stanton (2009). A Liquidity-Based Theory of Closed-End Funds, Review of Financial Studies 22, 257-297. Edmans, Alex, Itay Godlstein, and Wei Jiang (2009). Takeover Activity and Target Valuations: Feedback Loops in Financial Markets, Working Paper.
40
Getmansky, Mila, Andrew Lo, and Igor Makarov (2004). An Econometric Model of Serial Correlation and Illiquidity in Hedge-Fund Returns, Journal of Financial Economics 74, 1054-1090. Hasbrouck, Joel (2006). Trading Costs and Returns for US Equities: Estimating Effective Costs from Daily Data, working paper, Stern School of Business, New York University. Jensen, Michael C. (1986). Agency Costs of Free Cash Flow, Corporate Finance, and Takeovers, American Economic Review 76 (2), 323-329 Johnson, Shane, Ji-Chai Lin, and Kyojik Song (2006). Dividend Policy, Signaling, and Discounts on Closed-End Funds, Journal of Financial Economics 81, 539-562. Pontiff, Jeff (1996). Costly Arbitrage: Evidence from Closed-End Funds, Quarterly Journal of Economics 111 (4), 1135-115 Pontiff, Jeff (2006). Costly Arbitrage and the myth of idiosyncratic risk, Journal of Accounting and Economics, 42, 35-52. Ramadorai, Tarun (2009), The Secondary Market for Hedge Funds and the Closed-Hedge Fund Premium, Oxford University Working Paper. Ross, Stephen (2002). Neoclassical Finance, Alternative Finance and the Closed End Fund Puzzle, European Financial Management 8 (2), 129-137. Shleifer, Andrei and Robert Vishny (1986). Large Shareholders and Corporate Control, Journal of Political Economy 94 (3), 461-488. Wang, Z. Jay and Vikram Nanda (2008). Why Do Aggressive Payout Policies Reduce FundDiscounts – Is It Signaling, Agency Costs, or Dividend Preferences? Working paper, University of Illinois at Urbana-Champaign.
41
Figure 1: The figure depicts the various date-0 and date-2 MDP policies. There are four crucial quantities governing MDP policies and their economic consequences. is a combined measure of the cost of a proxy battle and eventual liquidation; measures the relative advantage of mounting a proxy battle (it is a ratio of the discount absent an MDP policy to ); measures the cost of an attack by an activist relative to the eventual costs of a proxy battle and liquidation; finally, Δ α is a measure of the payout of the managed asset net of managerial ability. A pre-emptive MDP with payout policy of δ (which is in addition to Δ) is possible in the hatched region, and rules out any future attacks by activists. The date-0 discount is . If a pre-emptive MDP is not adopted, then the probability of an attack at date-1 is given by . If an attack takes place, the date-2 MDP is denoted by δ and the new discount is (the latter declines relative to ). Once an attack takes place and management responds, the probability that a proxy battle will take place and result in a liquidation is given by δ ℓ δ .
Adoption of post-attack MDP (‘Reactive MDP’)
Decisive preemption
1516
δ 1 Δ α , 0
11
16
δ Δ α , 0
=1
1
δ 0,12
16 158 7 ,
78 1 0
78
δ 1 Δ α ,78 , δ ℓ δ
14
If an attack takes place:
δ 0,12 , 1
18
δ 0, 118 ,
δ ℓ δ 4
If an attack takes place:
Non-Adoption of post-attack MDP
No preemption
42
Figure 2: A scatter plot of MDP targets announced by funds subsequent to activist attacks. The horizontal axis indicates the percentage share of the activist in the fund. The vertical axis indicates the target MDP as a percentage of NAV.
43
Table 1. Number of CEFs with Managed Distribution Policy (MDP): 1990 – 2006 This table reports the total number of CEFs in our sample, the number of MDP funds, and the median payout target for the MDP group at each year-end from 1990 to 2006. We report summary statistics for all funds and domestic equity funds, respectively. The all funds group consists of CEFs with investment styles classified by the Wall Street Journal as Convertible Funds, General Equity Funds, Preferred Equity Funds, Special Equity Funds, and World Equity Funds. The domestic equity group consists only of General Equity Funds and Special Equity Funds. An MDP fund is defined as pre-emptive (reactive) if the MDP adoption year is prior to (after) any activist attack year.
All Funds Domestic Equity Funds
Total MDP Funds
Preemptive Reactive Payout Target
(%)
Total MDP Funds
Preemptive Reactive Payout Target
(%) 1990 84 12 11 1 10.00 31 8 7 1 10.00 1991 88 12 11 1 10.00 31 8 7 1 10.00 1992 99 12 11 1 10.00 32 8 7 1 10.00 1993 109 12 11 1 10.00 35 8 7 1 10.00 1994 131 13 12 1 10.00 42 9 8 1 10.00 1995 134 15 14 1 10.00 43 9 8 1 10.00 1996 139 15 14 1 10.00 46 9 8 1 10.00 1997 138 21 18 3 10.00 46 13 11 2 10.00 1998 139 26 21 5 10.00 46 14 12 2 10.00 1999 135 28 21 7 10.00 48 16 12 4 10.00 2000 128 28 21 7 10.00 48 16 12 4 10.00 2001 124 26 21 5 10.00 49 16 12 4 10.00 2002 133 27 19 8 10.00 55 22 14 8 10.00 2003 152 33 23 10 9.00 62 25 16 9 9.93 2004 180 50 37 13 7.91 81 36 26 10 8.00 2005 192 68 55 13 7.74 94 49 38 11 7.61 2006 189 73 58 15 8.00 93 52 39 13 8.00
44
Table 2. Summary Statistics
This table reports the mean (median) statistics for the following fund characteristics: annual distribution yield relative to year-beginning NAV (including total distribution and sources of distribution from net investment income, realized capital gains, and return of capital) ; average monthly discount during the year; share illiquidity measured by the average daily Amihud (2002) Illiquidity Ratio during the year; asset illiquidity measured by the first-order autocorrelation coefficient based on the monthly NAV returns during the previous three years and two Gibbs measures (multiplied by 100) based on either the basic market-adjusted model (BMA) or the latent common factor model (LCF); the year-end leverage ratio defined as total liabilities relative to year-end TNA; the Carhart (1997) four-factor alpha based on the monthly NAV returns during the previous three years; the annual management fee; the annual managerial ownership as a percentage of common shares outstanding; Year-end fund TNA (applicable to common shareholders); fund age; and the accumulated unrealized capital gains normalized by year-end TNA. For each variable, we first calculate the time series average over years for each fund and then average across funds for non-MDP group, pre-emptive (inception and post-inception) MDP group and reactive MDP group, respectively. The sample period is from 1995 to 2006.
All Funds Domestic Equity Funds Non-
MDP Pre-emptive
MDP Reactive
MDP Non-MDP
Pre-emptive MDP
Reactive MDP
Inception Post-inception
Inception Post-inception
Number of Funds 195 43 27 22 78 31 14 16 Distribution Yield (%)
Total Distribution 6.07 (6.57)
7.85 (7.49)
10.95 (10.04)
10.82 (9.98)
6.95 (7.47)
7.84 (7.63)
11.42 (10.54)
11.25 (10.63)
Investment Income 2.80 (1.39)
3.68 (3.69)
3.47 (3.40)
1.52 (0.72)
2.43 (1.34)
3.00 (2.57)
1.90 (1.23)
0.99 (0.70)
Realized CapGain 2.99 (2.02)
2.71 (2.34)
6.13 (5.48)
5.40 (3.66)
3.95 (3.84)
2.94 (2.63)
7.96 (8.78)
5.35 (3.66)
Return of Capital 0.28 (0.00)
1.45 (0.30)
1.34 (1.11)
3.90 (1.91)
0.57 (0.00)
1.90 (0.74)
1.57 (1.40)
4.91 (3.12)
Discount (%) 8.55 (10.37)
3.73 (5.34)
3.94 (5.15)
5.01 (7.49)
8.68 (9.85)
4.33 (5.42)
6.08 (5.58)
2.39 (3.31)
Share Illiquidity
8.70 (6.83)
3.46 (3.03)
6.87 (5.27)
6.04 (4.70)
9.38 (5.07)
3.24 (2.94)
8.54 (5.66)
5.68 (4.53)
Asset Illiquidity
1st Order Autocorrelation 0.05 (0.06)
-0.05 (-0.04)
0.02 (0.06)
0.04 (0.05)
-0.01 (0.01)
-0.06 (-0.05)
-0.05 (-0.04)
0.03 (0.02)
Gibbs Measure: BMA 0.37 (0.30)
0.20 (0.17)
0.39 (0.37)
0.33 (0.24)
0.36 (0.30)
0.20 (0.15)
0.39 (0.39)
0.31 (0.24)
Gibbs Measure: LCF 0.35 (0.27)
0.21 (0.18)
0.33 (0.32)
0.31 (0.21)
0.33 (0.27)
0.20 (0.17)
0.34 (0.33)
0.31 (0.21)
Managerial Skill
Leverage 0.19 (0.06)
0.27 (0.35)
0.30 (0.32)
0.10 (0.06)
0.21 (0.09)
0.22 (0.10)
0.21 (0.19)
0.13 (0.09)
Four-Factor Alpha -0.17 (-0.16)
-0.00 (-0.10)
-0.20 (0.07)
-0.64 (-0.16)
0.00 (-0.02)
-0.03 (-0.11)
-0.56 (-0.10)
-0.59 (-0.09)
Managerial Entrenchment (%)
Management Fee 0.92 (0.88)
0.78 (0.85)
0.83 (0.80)
1.00 (1.00)
0.88 (0.85)
0.79 (0.85)
0.83 (0.80)
1.00 (1.00)
Management Ownership 1.89 (0.00)
0.65 (0.00)
2.54 (0.00)
2.70 (0.00)
3.95 (0.00)
0.81 (0.00)
3.06 (0.00)
3.75 (0.00)
TNA ($mil) 375.57 (156.66)
438.74 (274.45)
671.17 (618.04)
244.01 (186.85)
422.24 (260.32)
471.36 (288.79)
565.31 (659.74)
243.31 (159.34)
Age (Years) 10.25 (7.50)
3.80 (2.00)
10.94 (6.00)
15.73 (15.50)
11.65 (4.50)
3.40 (2.00)
14.57 (14.50)
16.22 (15.50)
Unrealized CapGain (%) 11.47 (9.08)
12.44 (12.41)
21.46 (24.58)
16.11 (13.85)
18.60 (14.53)
13.41 (12.58)
32.46 (31.56)
17.80 (13.86)
Table 3A. Change of Management Fee and Fund Size around MDP Adoption This table examines the change in fund assets and managerial compensation during the three-year periods before vs. after the MDP adoption. For MDP funds, we compute the mean (median) statistics for the following variables: average management fee (%), average annual TNA growth rate (%), average year-end TNA ($ million), average annual managerial pay (average monthly net assets in million dollars * percentage management fee), and average discount (%). We report the statistics for both the pre-MDP and post-MDP periods and also report the changes. For each MDP fund, we identify a control group of non-MDP funds that have the same investment objective and compute the mean (median) statistics during the same pre-MDP and post-MDP periods. We report the mean (median) difference between the MDP group and the non-MDP control group as well as the difference-in-difference results around MDP adoption. We include in Panel A a total of 37 MDP funds with both pre- and post-MDP percentage management fees available. We include in Panels B to E a total of 25 MDP funds with both pre- and post-MDP fund TNA data available. Statistical significance for the mean (median) tests of 1%, 5%, and 10% are indicated by ***, **, and * respectively. Pre-MDP Post-MDP Change
Panel A: Management Fee (%) MDP Funds 0.963 (1.000) 0.952 (1.000) -0.011 (0.000) Non-MDP Funds 0.815 (0.783) 0.824 (0.814) 0.009 (-0.011) Difference 0.148*** (0.119***) 0.128*** (0.094***) -0.020 (0.011)
Panel B: TNA Growth Rate (%)
MDP Funds 13.853 (11.372) 3.576 (7.014) -10.277** (-4.113*) Non-MDP Funds 6.019 (5.482) 5.127 (4.628) -0.892 (-1.329) Difference 7.834* (4.564**) -1.551 (-0.959) -9.384* (-5.036**)
Panel C: TNA ($millions)
MDP Funds 284.466 (153.188) 340.146 (190.252) 55.680 (40.075) Non-MDP Funds 172.558 (141.189) 293.797 (195.742) 121.239*** (100.651***) Difference 111.908** (21.706) 46.349 (40.486) -65.559** (-18.957*)
Panel D: Management Fee ($millions)
MDP Funds 2.810 (1.509) 3.210 (1.825) 0.400 (0.375**) Non-MDP Funds 1.225 (1.418) 2.090 (1.468) 0.865*** (0.751***) Difference 1.585** (0.371**) 1.120** (0.348*) -0.464* (-0.328*)
Panel E: Discount (%)
MDP Funds 10.433 (13.233) 8.491 (10.051) -1.942 (0.500) Non-MDP Funds 9.481 (8.641) 12.828 (12.238) 3.347*** (3.589**) Difference 0.952 (2.219) -4.338** (-1.735**) -5.290*** (-5.207***)
Table 3B. Change of Management Fee and Fund Size for MDP Funds: 2000-2006 This table examines the change of fund assets, managerial compensation, and fund discount for MDP and non-MDP funds during the period 2000-2006. For MDP funds, we report the mean (median) year-end TNA, annual managerial pay (average monthly net assets in million dollars * percentage management fee) and average monthly discount in 2000. We also report the mean (median) statistics for the annual averages during the subsequent 5-year period (2001-2006) and the changes relative to year 2000. For each MDP funds, we identify a matched non-MDP fund that have the same investment objective and the closest fund TNA in 2000. We report the mean (median) difference in fund TNA, management fee and discount between the MDP group and the non-MDP matched group as well as the difference-in-difference results. The analysis includes 26 MDP funds in 2000 that maintained the closed-end status for the next three years or longer. Statistical significance for the mean (median) tests of 1%, 5%, and 10% are indicated by ***, **, and * respectively. 2000 2001-2006 Change
Panel A: TNA ($millions) MDP Funds 406.934 (206.659) 354.009 (149.394) -52.926** (-39.590***) Non-MDP Funds 285.445 (154.669) 286.578 (207.593) 1.133 (-0.357) Difference 121.490* (8.335**) 67.431 (-17.973) -54.059** (-64.855***)
Panel B: Management Fee ($millions)
MDP Funds 3.530 (1.950) 3.065 (1.387) -0.465** (-0.460***) Non-MDP Funds 2.136 (1.217) 2.359 (1.572) 0.223 (-0.000) Difference 1.394*** (0.705***) 0.706 (0.004) -0.688** (-0.653***)
Panel C: Discount (%)
MDP Funds 11.384 (12.672) -2.575 (-1.489) -13.959*** (-12.505***) Non-MDP Funds 18.056 (17.468) 9.706 (9.967) -8.350*** (-5.910***) Difference -6.672*** (-4.604***) -12.281*** (-10.748***) -5.609** (-7.587**)
Table 4. MDP and Hostile Fund Termination This table reports summary statistics for hostile fund terminations (terminations associated with activist attacks) during the period from 1990 to 2006. We restrict our analysis to the sub-sample of CEFs with inception year prior to 2004. For MDP funds, we also require the MDP adoption year to be prior to 2004. We then follow the funds until the end of 2006 to observe fund terminations and activist attacks. We consider funds that were attacked before adopting an MDP and subsequently did not respond with an MDP (Attacked, no response) or did respond with an MDP (Attacked, MDP response). We also consider funds that employed MDPs preemptively both after and at inception (Preemptive MDP post- or at inception). We report the total number of funds with inception year prior to 2004 and the percentage of funds that were terminated before the end of 2006. In the third column, we consider terminations involving liquidations, open-endings, and mergers with another CEF. The fourth column considers terminations involving only liquidations and open-endings, while the last considers only liquidations. The last two rows calculate the p-values for the hypotheses that preemptive MDPs are terminated as frequently as attacked MDPs, and whether funds that respond to attacks with an MDP are terminated as frequently than those that do not respond.
Definitions of “hostile termination”
Types of funds N Liquidated, Open-ended or Merged
L1
Liquidated or Open-ended L2
LiquidatedL3
Attacked, no response 65 46% 40% 29% Attacked, MDP response 19 26% 16% 11%
Attacked, total 84 42% 35% 25%
Preemptive MDP, post-inception 16 6% 6% 6% Preemptive MDP, at inception 32 3% 3% 0%
Prby( Li(Preemptive MDP, post-inception) = Prby( Li(Attacked, total) )
0.000 0.001 0.022
Prby( Li(Attacked, response) = Prby( Li(Attacked, no response) )
0.126 0.052 0.100
Table 5. Factors Affecting the MDP Adoption: Probit Regressions (without Inception MDP Funds)
This table investigates how various fund characteristics affect its MDP adoption using ordered Probit regressions. The dependent variable is the MDP Event (0=No-MDP, 1=Moderate MDP, 2=Aggressive MDP). The regressors include: the fund TNA at the end of previous year; the fund age at the end of previous year; an indicator variable (Activist Attack) that equals one if the fund was ever attacked by activist shareholders in the previous three years; the share illiquidity measured by the average daily Amihud illiquidity ratio of fund shares in the previous year; the leverage ratio in the previous year; the asset illiquidity measured by the first-order serial correlation of monthly NAV returns in the previous three years; the management fee in the previous year; an indicator variable (Mgt Ownership) that equals one if more than 10% of common shares were beneficially controlled by board members and executive officers in the previous year; the accumulated unrealized capital gains as a percentage of year-end TNA in the previous year; and the four-factor alpha based on the monthly NAV returns in the previous three years. All variables except the indicator variables are standardized by subtracting the mean and scaling by the standard deviation of all funds in any give year. We also control for style fixed effects based on the Wall Street Journal style classification. We separately report regression results for all funds in our sample and domestic equity funds only. The standard errors are heteroskedasticity robust and clustered by fund. The Chi-statistics are reported in parentheses. Statistical significance of 1%, 5%, and 10% are indicated by ***, **, and * respectively. MDP Event {0=Non-MDP, 1=Moderate MDP, 2=Aggressive MDP} All Funds Domestic Equity Funds Intercept 2 -1.514***
(60.98) -1.268***
(61.52) -1.205***
(17.84) -0.943***
(17.35) Intercept 1 -1.023***
(50.58) -0.812***
(40.86) -0.539 (6.06)
-0.325 (2.86)
TNA 0.027 (0.06)
-0.001 (0.00)
0.002 (0.00)
-0.093 (0.22)
Age 0.002 (0.00)
0.054 (0.11)
0.116 (0.21)
0.245 (1.01)
Activist Attack 0.920*** (19.84)
-- 0.962** (7.69)
--
Share Illiquidity -0.414*** (8.84)
-0.435*** (10.74)
-0.319* (2.99)
-0.388** (4.60)
Leverage -0.138 (1.06)
-0.178 (2.19)
-0.321* (3.25)
-0.353** (6.00)
Asset Illiquidity -0.196*** (7.09)
-0.194*** (9.46)
-0.228** (3.84)
-0.258** (5.75)
Mgt Fee 0.103 (0.93)
0.170* (2.75)
0.279* (2.81)
0.375** (5.42)
Mgt Ownership -0.128 (0.07)
-0.221 (0.24)
-0.979** (4.59)
-1.040** (6.05)
Unrealized Capgain -0.014 (0.02)
-0.012 (0.02)
0.015 (0.01)
0.018 (0.02)
Alpha -0.008 (0.01)
-0.032 (0.22)
0.091 (1.37)
0.059 (0.50)
Style Fixed-Effects Included Included Included IncludedObservations 1130 1130 359 359
49
Table 6. Factors Affecting the Return of Capital: Tobit Regressions (without Inception MDP Funds)
This table investigates how various fund characteristics affect its decision to distributing return of capital using Tobit regressions. The dependent variable is annual return of capital scaled by year-beginning NAV. The regressors include: the fund TNA at the end of previous year; the fund age at the end of previous year; an indicator variable (Activist Attack) that equals one if the fund was ever attacked by activist shareholders in the previous three years; the share illiquidity measured by the average daily Amihud illiquidity ratio of fund shares in the previous year; the leverage ratio in the previous year; the asset illiquidity measured by the first-order serial correlation of monthly NAV returns in the previous three years; the management fee in the previous year; an indicator variable (Mgt Ownership) that equals one if more than 10% of common shares were beneficially controlled by board members and executive officers in the previous year; the accumulated unrealized capital gains as a percentage of year-end TNA in the previous year; and the four-factor alpha based on the monthly NAV returns in the previous three years. All variables except the indicator variables are standardized by subtracting the mean and scaling by the standard deviation of all funds in any give year. We also control for style fixed effects based on the Wall Street Journal style classification. We separately report regression results for all funds in our sample and domestic equity funds only. The t-statistics are reported in parentheses. Statistical significance of 1%, 5%, and 10% are indicated by ***, **, and * respectively. Return of Capital (%) All Funds Domestic Equity Funds Intercept -19.875***
(-10.03) -19.025***
(-9.87) -7.516***
(-5.75) -6.995***
(-5.41) TNA -1.336**
(-2.07) -1.853***
(-2.74) -1.404** (-1.97)
-1.909*** (-2.65)
Age -0.822 (-1.22)
-0.413 (-0.61)
-1.037 (-1.02)
-0.401 (-0.40)
Activist Attack 4.838*** (4.16)
-- 3.439** (2.48)
--
Share Illiquidity -3.281*** (-3.53)
-3.713*** (-3.84)
-1.844** (-2.00)
-2.134** (-2.24)
Leverage -1.326** (-2.12)
-1.540** (-2.34)
-1.497** (-2.05)
-1.575** (-2.08)
Asset Illiquidity -2.200*** (-3.96)
-2.328*** (-4.09)
-1.419** (-2.02)
-1.670** (-2.33)
Mgt Fee 0.011 (0.02)
0.364 (0.64)
-0.032 (-0.04)
0.328 (0.44)
Mgt Ownership 2.758 (1.54)
2.045 (1.11)
-1.111 (-0.57)
-1.659 (-0.82)
Unrealized Capgain
-2.081*** (-3.70)
-2.074*** (-3.63)
-2.282*** (-3.37)
-2.356*** (-3.37)
Alpha -0.044 (-0.09)
-0.120 (-0.24)
0.631 (1.07)
0.574 (0.94)
Style Fixed-Effects
Included Included Included Included
Observations 1122 1122 354 354
50
Table 7. Preemptive (post-inception) MDP vs. Reactive MDP This table investigates the determinants of preemptive vs. reactive MDP adoptions using probit regressions. The probability is calculated for preemptive MDP adoption. The explanatory variables include share illiquidity measured by the Amihud illiquidity ratio, asset illiquidity measured by the AR1 coefficient based on the NAV returns, leverage ratio, and four-factor alpha. All variables are measured as the annual averages during the 5-year window from 3 years before and 2 years after the MDP adoption year. We convert all variables into ranks between 0 and 1. We also construct a relative illiquidity measure (share illiquidity relative to asset illiquidity) by calculating the difference between the share illiquidity rank and the asset illiquidity rank. The data sample includes 47 pre-emptive MDP adoptions and 22 reactive MDP adoptions. The Chi-statistics are reported in parentheses. Statistical significance of 1%, 5%, and 10% are indicated by ***, **, and * respectively.
Event {1=Preemptive MDP; 0=Reactive MDP} (4) (5) (6) Intercept 0.129
(0.52) -1.759***
(7.81) -1.258 (1.46)
TNA 2.098** (5.78)
1.811*
(3.30) Share Illiquidity
0.370 (0.16)
Asset Illiquidity
-1.152 (1.89)
Relative Illiquidity
-0.246 (0.29)
0.810 (1.78)
Leverage 2.108** (6.22)
2.263** (6.31)
Alpha -0.334 (0.19)
-0.369 (0.22)
Num of Obs 49 49 49
51
Table 8. Factors Affecting the Activist Attack: Probit Regressions This table investigates how various fund characteristics affect the likelihood of being attacked by activists using Probit regressions. The dependent variable is theAttack Event (0=No-Attack, 1=Attack). The regressors include: the fund TNA at the end of previous year; the fund age at the end of previous year; an indicator variable (MDP) that equals one if the fund has MDP in place in the previous year; the share illiquidity measured by the average daily Amihud illiquidity ratio of fund shares in the previous year; the leverage ratio in the previous year; the asset illiquidity measured by the first-order serial correlation of monthly NAV returns in the previous three years; the management fee in the previous year; an indicator variable (Mgt Ownership) that equals one if more than 10% of common shares were beneficially controlled by board members and executive officers in the previous year; the accumulated unrealized capital gains as a percentage of year-end TNA in the previous year; and the four-factor alpha based on the monthly NAV returns in the previous three years. All variables except the indicator variables are standardized by subtracting the mean and scaling by the standard deviation of all funds in any give year. We also control for style fixed effects based on the Wall Street Journal style classification. We separately report regression results for all funds in our sample and domestic equity funds only. The standard errors are heteroskedasticity robust and clustered by fund. The Chi-statistics are reported in parentheses. Statistical significance of 1%, 5%, and 10% are indicated by ***, **, and * respectively. Attack Event {0=No-Attack, 1=Attack} All Funds Domestic Equity Funds Intercept -1.378***
(118.88) -1.343***
(98.40) -1.383***
(73.30) -1.137***
(44.61) TNA -0.121
(1.47) -0.120 (1.47)
-0.561*** (7.83)
-0.610*** (10.78)
Age 0.196** (5.74)
0.198** (5.48)
0.395*** (7.00)
0.446*** (7.96)
MDP -0.135 (0.40)
-0.740** (5.20)
Share Illiquidity -0.011 (0.03)
-0.020 (0.08)
-0.306*** (6.66)
-0.386*** (8.24)
Leverage -0.288*** (7.05)
-0.292** (6.60)
-0.208 (2.20)
-0.250 (1.91)
Asset Illiquidity -0.167*** (7.89)
-0.171*** (8.52)
-0.274** (5.25)
-0.316*** (7.26)
Mgt Fee 0.127* (2.83)
0.135* (3.11)
0.075 (0.45)
0.116 (0.64)
Mgt Ownership -0.204 (0.31)
-0.221 (0.37)
0.172 (0.20)
-0.008 (0.00)
Unrealized Capgain
0.015 (0.08)
0.014 (0.06)
0.054 (0.36)
0.054 (0.33)
Alpha -0.012 (0.06)
-0.017 (0.12)
-0.135 (1.65)
-0.148 (2.29)
Style Fixed-Effects
Included Included Included Included
Observations 1202 1202 407 407
52
Table 9. Activists’ Attacks and Fund Discounts This table examines the change of discounts during the 7-year window around the year when activists’ attack occurred. We have a total of 122 cases of attacks. For each fund under attack, we identify a matched fund that has the same investment objective, has the closest discount level 2 years before the attack (Year -2), but has never been subject to activists’ attack during the event window. We report the mean (median) difference in discount of the attacked fund relative to the matched fund for each year during the event window. We also separately report the relative discount changes for the sub-sample of funds that reacted with adopting MDPs (46 cases) and the sub-sample of funds that did not adopt MDPs (76 cases) as the response during the event window. Statistical significance of 1%, 5%, and 10% are indicated by ***, **, and * respectively. All Attacks (122) MDP (46) Non-MDP (76) Year -3 3.470***
(1.620***) 1.378
(1.348) 4.885***
(2.223***) Year -2 3.614***
(1.058***) 2.311**
(0.389**) 4.470***
(1.491***) Year -1 4.504***
(2.990***) 3.547***
(3.632***) 5.133***
(2.347***) Year 0 2.724***
(1.588***) 2.518*
(2.007*) 2.860**
(1.372**) Year +1 -0.241
(-0.640) -0.438
(-0.396) -0.112
(-0.878) Year +2 -2.669**
(-3.960**) -5.219**
(-4.994**) -0.994
(-3.490) Year +3 -4.636***
(-3.879***) -7.364***
(-6.351***) -2.843*
(-2.512*)