Hedge Fund Activism, Voice and Value Creation

Abstract

We construct a novel hand-collected large dataset of 205 U.S. hedge funds and 1025 activist events over the period 2005–2013, which records both the Schedule 13D filing date and the voice date, and explore the role of voice in value creation. We employ alternative inferential statistical approaches, including parametric, non-parametric, and heteroscedasticity-robust tests. We reveal that the voice date is important in creating short-term firm value and provide strong evidence that voice is associated with positive abnormal returns. These findings suggest that voice leads to information revelation, with implications for U.S. stock market arbitrage.

1. Introduction

Hedge fund activism is an investment strategy where a hedge fund intervenes in a target firm with a minority stake and influences the firm’s internal decisions. The main objective of this intervention is to reduce agency costs, increase the firm’s performance and maximize shareholder value. Prior literature suggests that increased management monitoring by shareholder activists reduces agency and incentive costs (Brav et al., 2008a; Gilson & Gordon, 2013), especially when activists gain board representation (Goodwin, 2016). Top activist investors target firms with higher agency costs and are more successful in their activist goals when engaging in monitoring management (Krishnan et al., 2016).

Empirical contributions have documented that the stock market reacts favorably to activism, yielding positive average abnormal returns for target firms around the Schedule 13D1 filing date, which is considered the critical date of an activist event, and suggesting that hedge fund activism creates value (Becht et al., 2008; Brav et al., 2008a, 2008b; Clifford, 2008; Klein & Zur, 2009; Greenwood & Schor, 2009; Boyson & Mooradian, 2011; Gow et al., 2014; Bebchuk et al., 2015; Becht et al., 2017) and that as the event window becomes longer, abnormal returns increase (Brav et al., 2008a; Krishnan et al., 2016). Interestingly, positive abnormal returns are obtained even when hedge funds had previously disclosed a smaller stake at the target firm through the 13F filing (Brav et al., 2008a). Moreover, significant performance improvements are obtained when hedge funds switch from passive (Schedule 13G filing) to active (Schedule 13D filing) with no ownership change (Brav et al., 2015). These performance improvements occurred due to the hedge fund’s decision to switch from passive to active; otherwise, they would not have been implemented. Although short-term effects have been examined around the Schedule 13D filing date, there is hardly any evidence on the short-term abnormal returns surrounding the amendment dates.

The present paper departs from the previous literature in that it considers as critical dates of an activist event not only the Schedule 13D date but also the amendment date that contains voice. We define this date as the voice date. We expect voice to carry at least the same importance as the Schedule 13D.

The motivation for considering the voice date arises from the fact that this is the date when a hedge fund clearly asks, proposes, differentiates or demands operational, strategic, corporate governance or capital allocation changes in the “Item 4” section of the Schedule 13D or Schedule 13D/A filing, where investors state the “Purpose of Transaction”. Through voice, activists specify their objectives by revealing their internal assessment and disclosing specific plans or proposals for the target firm. So, we hypothesize that voice functions as an information revelation mechanism. In contrast, if the activists do not state any specific plans or proposals through the Schedule 13D or Schedule 13D/A filings, we define those interventions as non-voice. In these interventions, activists may either make amendments for beneficial ownership changes or not make amendments at all until exit. So far, the vast majority of existing literature assumes that the activism engagement process is initiated by the Schedule 13D filing and terminates with exit. However, Schedule 13D filings that do not contain voice simply signal an activist intention, which may occur or may not. The stock price of the targeted firm on the Schedule 13D filing date should reflect the expected value of two mutually exclusive events: the specific intervention activist purpose and the lack of such disclosed purpose. On the other hand, voice specifies the activism field since hedge funds disclose specific proposals and/or plans. The stock price of the targeted firm on the voice date should reflect the expected value of a successful, a failed or a settlement outcome, multiplied by the corresponding probabilities of occurrence. Although both dates are considered as dates of the activism engagement process initiation, they reflect different expected future outcomes.

The goal of this paper is to investigate how investors react to hedge fund activism (voice) and to the announcement of possible activism (Schedule 13D filing without voice). These two different initiation engagement dates are also compared in terms of abnormal return behavior.

Further motivation for focusing on the voice date is provided by the following examples of anecdotal evidence. On 22 August 2011, Starboard Value LP filed a Schedule 13D indicating an 8.9% beneficial ownership of Mips Technologies Inc. In the “Item 4” section, the reporting persons expressed their belief that that the shares were undervalued and did not have any plans or proposals2. On 13 September 2011, Starboard Value LP filed a Schedule 13D/A stating that they delivered a letter to the board of the company on 12 September 2011 nominating specific people for election to the company’s Board of Directors, and urging the company to discontinue pursuing acquisitions, focus on improving its operating performance and consider buying back shares. As a result, the stock return of Mips Technologies on that date (12 September) was 6.19%, much higher than that on the Schedule 13D filing date (22 August), which was 0.46%. This echoes the relatively important role of voice in driving market responses. We consider this event as a voice event, and we set 12 September 2011 as the voice date when Starboard Value LP delivered the letter to Mips Technologies Inc.

The second example refers to Atlantic Investment Management Inc. (AIM) filing a Schedule 13D on 14 September 2012, indicating a 5.05% beneficial ownership of Rockwood Holdings, Inc. On that date, the firm’s stock return was +1.84%. In the “Item 4” section, the reporting persons stated that they acquired the stock for investment purposes and did not have any current plans3. On 19 November 2012, AIM filed an amendment where a beneficial owner increase to 6.05% was stated without amending the “Item 4” section. On that date, the stock return was +2%. AIM exited on 29 January 2013, filing a Schedule 13D/A stating a beneficial ownership of 4.89%. On the exit date, the stock return was +0.5% We consider this as a non-voice event since AIM did not state any specific plans in any of its filings’ “Item 4” section, and we use the initial Schedule 13D date as the non-voice date.

We carefully collected a large sample of 205 U.S. hedge funds and 1025 activist events over the period 2005–2013. This dataset is novel since it records both the Schedule 13D and the voice dates and includes 283 voice and 487 non-voice events. Using this sample, we explored whether positive abnormal returns exist surrounding the voice date. Our findings strongly suggest that there exist positive abnormal returns not only surrounding the Schedule 13D date but also the voice date, suggesting value creation around the voice date as well. The abnormal returns around the voice date are approximately 1.11% and are higher than the abnormal returns around the Schedule13D date by approximately 33%. These findings are robust to alternative asset pricing models controlling for size, book-to-market, momentum, and GARCH-in-Mean effects. They are also robust to alternative parametric and non-parametric statistical procedures for inference on the significance of abnormal returns, including the Wilcoxon signed-rank test. The abnormal returns of Schedule13D and voice dates are also statistically compared by means of a battery of heteroscedasticity and skewness-robust tests. We document significant differences in the market inefficiency patterns. Our interpretation is that the Schedule 13D date reveals only partial information, with the remainder of the information being revealed at voice. As we record both the voice and the Schedule 13D dates, we also consider the case of 93 events in which the voice date leads the Schedule 13D filing date by less than 10 days. We find evidence of positive abnormal returns for these voice events, suggesting that voice, regardless of whether it occurs before or after the Schedule 13D date, entails information revelation that triggers market response. These results are in line with Becht et al. (2017), who found that positive short-term abnormal returns exist surrounding the amendment dates when activists disclose outcomes, such as board representation and takeovers, through amendments.

This paper contributes to the literature in several aspects. So far, there is no evidence of voice effect on activism interventions. In related studies, researchers assume that different types of voice in the initial filing reveal information, which in turn leads to short-term value creation. However, the voice impact after interventions, which is publicly available through amendments, is ignored in the previous literature. To our knowledge, this is the first attempt to use a subsample of U.S. voice activism that is based exclusively on interventions during the whole period of the engagement process. Therefore, our analysis differs from related studies in that voice is not classified in terms of the initial Schedule 13D filing, but in terms of both the initial filing and the amendment dates where hedge fund activists reveal their voice. We document that voice announcements yield short-term value creation at a chronologically different date than the initial filing.

Second, our work builds on the previous studies, such as those of Brav et al. (2008a, 2008b), Becht et al. (2017), Bebchuk et al. (2015) and Clifford (2008), investigates the short-term abnormal disclosure returns of target companies. Our analysis confirms the presence of highly significant positive abnormal returns several days prior to the announcement of the initial Schedule 13D filing. More interestingly, we extend the earlier evidence for a recent period that covers the 2007–2009 financial crisis. Our analysis indicates that the findings of previous studies are not affected by the financial crisis. These results hold for different models of abnormal returns, such as the Carhart model or models that explicitly account for leverage effects in volatility. Alternative statistical test procedures have also been implemented to validate our findings.

Third, the abnormal stock behavior of voice engagement is compared with the behavior of Schedule 13D filing interventions, in addition to a comparison of voice with non-voice activism. The differences between the abnormal returns of these announcement dates are tested by applying conventional inferential statistical methods, as well as test procedures that exhibit robustness to conditional heteroscedasticity and skewness. We document that voice induces a more profound effect than the Schedule 13D filing intervention (full sample) on the stock returns of target companies in an economic and statistical sense. The evidence suggests the presence of a two-step activism disclosure mechanism, where investors successively respond to information revelation at the initial filing and the voice. Our results also show that voice constitutes a stronger form of activism than non-voice. These results have not been reported in the literature before.

The remainder of the paper is as follows: Section 2 discusses the dataset and the voice date. Section 3 outlines the methodology for abnormal returns, spells out the hypotheses to be tested and discusses the statistical procedures. Section 4 reports the empirical findings, discusses their implications and makes a regulatory policy recommendation. Section 5 provides several robustness checks, including various GARCH-type models for calculating abnormal returns and non-parametric tests for statistical inference and testing abnormal performance using conditional factor models. Section 6 presents conclusions.

2. The Dataset and Voice Dates

2.1. Compiling the Dataset

We obtained Schedule 13D filings between 1 January 2005 and 31 December 2013 using the Historical SEC Edgar Archives from the Edgar Database of the Securities and Exchange Commission (SEC). Our search for “SC 13D” gave us 11.700 filers from a total of 19.352 filings.

The next step was to identify the hedge funds. We formed our sample of “pure-play” hedge funds (Ben-David et al., 2013) following Brunnermeier and Nagel (2004) and Griffin and Xu (2009). We searched the Investment Adviser Public Disclosure website for each of the “Reporting Persons” of the Schedule 13D filings and included in our sample only firms that were registered as investment advisers with the SEC and, thus, filed the ADV form. Form ADV is the uniform form used by investment advisers to register with both the SEC and state securities authorities. Next, we searched the “Item 5” section (“Information About Your Advisory Business—Employees, Clients, and Compensation”) and included in our sample only those firms that had at least 50% of their clients classified as “Other Pooled Investment Vehicles (e.g., hedge funds)” or “High Net Worth Individuals” and charged performance-based fees.

On the basis of these steps, we identified 321 “pure-play” hedge funds that filed a total of 2.098 Schedule 13D filings. We excluded from our sample financial firms (SIC code 6000 to 6799), private firms and activist events that were still live at the end of December 2013. Our final sample consists of 205 hedge funds and 1025 activist events (283 voice and 487 non-voice events).The stock prices of our sample were downloaded from Thomson DataStream (TDS). Descriptive statistics for all stock returns and for the stock returns that comprise each event-based category are reported in Table 1 and indicate evidence of non-normality.

Table 1. Descriptive statistics of the stock returns (in %).

	All	Schedule 13D			Voice			Non-Voice
		$(-20,{\mathit{T}}_0$)	$({\mathit{T}}_0$, 0)	$({\mathit{T}}_0$, 20)	$(-20,{\mathit{T}}_0$)	$({\mathit{T}}_0$, 0)	$({\mathit{T}}_0$, 20)	$(-20,{\mathit{T}}_0$)	$({\mathit{T}}_0$, 0)	$({\mathit{T}}_0$, 20)
Mean	0.041	0.049	0.056	0.057	0.030	0.033	0.037	0.060	0.069	0.069
Median	0.000	−0.018	−0.016	−0.014	−0.032	−0.029	−0.029	−0.010	−0.007	−0.005
Standard Deviation	2.623	3.129	3.177	3.175	3.058	3.087	3.097	3.199	3.259	3.249
Kurtosis	62.651	17.690	20.680	21.135	15.889	16.571	17.233	19.398	24.431	24.779
Skewness	1.985	0.661	0.870	0.901	0.465	0.479	0.544	0.807	1.183	1.194
Minimum	−24.713	−19.775	−20.038	−20.172	−20.038	−20.584	−20.757	−19.853	−20.014	−20.085
Maximum	38.477	24.036	26.468	26.871	22.043	22.904	23.627	25.647	29.167	29.320
Number of Stocks	1025	770			283				487

Notes: Table 1 reports summary statistics of all stock returns and for the stock returns that comprise each event-based category. Stock returns are computed as the logarithmic first differences of the corresponding stocks, multiplied by 100. The summary statistics for each event-based category are calculated by using three different sample windows; the first covers the period from the first available observation to twenty trading days before the event day (−20,$T_0$), the second from the first available observation to the event day ($T_0$, 0), and the third from the first available observation to twenty trading days after the event day ($T_0$, 20). $T_0$ represents the first day of the estimation sample window. Here, −20 denotes 20 trading days before the event day, 0 denotes the event day and 20 denotes 20 trading days after the event day.

2.2. Voice Dates

Hedge funds follow different strategies in order to maximize shareholder value and increase firm performance. These strategies concern the hedge funds’ plans or proposals for the target firm and the timing of their disclosure. Plans or proposals must be disclosed in the “Item 4” section and contain changes in a firm’s corporate governance (i.e., changes in board structure and composition), strategy (i.e., sale of company, spin-off of a subsidiary), capital allocation (i.e., buying back shares, special dividends) and operational performance. There are many cases, though, where hedge funds exit the target firm without publicly stating any of the above objectives.

The timing of disclosure varies according to the activists’ strategies. Activists may disclose their objectives either in their initial Schedule 13D filing, following a more offensive agenda, or later by filing an amendment. There are some cases, though, where activists state their proposals without filing a Schedule 13D (activism under the 5% threshold), which are excluded from our sample.

We search for voice in the “Item 4” section of the initial Schedule 13D, its amendments and the attached Exhibits. Activist interventions are usually followed by amendments where hedge funds make material changes in facts set forth in the initial filing. If the exact date of voice is stated in these filings, we consider this date as the voice date; otherwise, we use the filing date. We hypothesize that voice events generate positive short-term abnormal returns, since important information becomes available to market participants. We include in our sample only the first voice incident of an activist intervention.

In our voice events, the average difference between the identified voice date and the Schedule 13D date is +40 trading days or 56 calendar days; namely, the voice date lags the Schedule 13D date by 40 trading days. Importantly, we identified 93 events in which voice leads Schedule 13D by 1 to 10 days, namely the voice day is in the space (−1, −10) of the Schedule 13D day. This point is taken into consideration when assessing the statistical significance of abnormal returns of voice by looking not only at all voice events but also at this category of voice events leading the Schedule 13D events by 1 to 10 days.

3. Methodology

3.1. Abnormal Returns of Targeted Firms

Our objective is to explore whether there exist statistically significant abnormal returns for the targeted firm around the hedge fund activism announcement date (Schedule 13D date), the voice date and the non-voice date (ex post). The time period of interest for which we observe the three event types, denoted as the event period, covers 10 pre-event days (day −10 to day −1), the event date (day 0) and 10 post-event days (day 1 to day 10). The event window is expanded by 10 days prior to the event in order to capture possible information leaks and by 10 days after the event to account for possible delayed response of investors to announcements.

For our empirical analysis, we employ the event study methodology, as originally introduced by Ball and Brown (1968) and Brown and Warner (1980, 1985). Abnormal returns are assessed in terms of the realized returns and the returns that would be normally expected by the market. Following Brown and Warner (1980, 1985), an abnormal return $A{R}_{it}$ is defined as the difference between the actual return $R_{it}$ of stock i on the event day t and the expected stock return on the event day t predicted by an estimated asset pricing model:

$$A{R}_{it}=R_{it}-E\left(R_{it}|D_t\right)$$

(1)

where $R_{it}=100\ast log\left(P_{it}/P_{it-1}\right),$ $P_{it}$ is the actual price of stock i at event day t, i = 1, 2, …, N, with N the total number of stocks, while$E\left(R_{it}|D_t\right)$ denotes the expected stock returns given the information set $D_t$ available at time period t. The expected returns represent the “normal” returns, namely the returns that would be anticipated if no event took place. The expected returns are predictions for the event day, generated by an asset pricing model fitted to the actual stock returns over an estimation window. We have also computed the cumulative abnormal returns as $CAR\left(t_1,t_2\right)={\displaystyle \sum _{t=t_1}^{t_2}}{AR}_t$. The estimation window represents the sample proportion of the data that precedes the event period. In this paper, the length of the estimation window is determined in terms of data availability. In particular, the estimation period for each stock starts at the first available stock return observation and ends 21 days before the announcement date. Our results are also verified by using an estimation window of fixed length to calculate the abnormal returns.

Abnormal stock returns of the targeted firms are calculated using the market model and the multi-factor model of Carhart (1997). The market model assumes that the returns of each stock are linearly related to the market portfolio returns. The following linear regression model is estimated:

$$R_{it}=a_0+{\beta }_1\left(R_{mt}-R_{ft}\right)+u_{it}$$

(2)

where $R_{mt}$ denotes the S&P 500 market index returns and $R_{ft}$ denotes the risk-free rate.

In the Carhart (1997) model, the conventional market model is enhanced by the size, value and momentum factors:

$$R_{it}=a_0+{\beta }_1\left(R_{mt}-R_{ft}\right)+{\beta }_{2}SM{B}_t+{\beta }_{3}HM{L}_t+{\beta }_{4}MO{M}_t+u_{it},$$

(3)

where $SM{B}_t$ denotes the size factor—the difference between the returns on portfolios based on stocks with small market capitalization and stocks with big market capitalization; $HM{L}_t$ represents the value factor—the difference between the returns on portfolios based on stocks with high book-to-market ratios and stocks with low book-to-market ratios; and $MO{M}_t$ represents the momentum factor—the difference between the returns on portfolios of the winners and losers of the previous year.

We finally consider the GARCH-in-Mean model, as asset pricing theory suggests that higher risk has to be compensated with a higher expected return. Thus, we include a measure of stock return volatility as a term in the generating mechanism of expected returns. Lundblad (2007) provides evidence supporting the adoption of the GARCH-in-Mean model and shows that the market’s risk premium and conditional volatility are positively related4. Motivated by these empirical findings, we consider the following specification for the returns’ generating mechanism5

$$R_{it}=a_0+{\beta }_1\left(R_{mt}-R_{ft}\right)+{\beta }_{2}SM{B}_t+{\beta }_{3}HM{L}_t+{\beta }_{4}MO{M}_t+{\beta }_5\sqrt{h_{it}}{+u}_{it},$$

(3a)

$$u_{it}={\xi }_{it}{\left(h_{it}\right)}^{1/2},$$

(3b)

$$h_{it}=k_0+{\gamma }_1{u}_{it-1}^2+{\gamma }_2{h}_{it-1},$$

(3c)

where $\left\{{\xi }_{it}\right\}$ is a sequence of random variables that are assumed to be independent and identically distributed as Student’s t with unknown degrees of freedom, while $\left\{h_{it}\right\}$ is the sequence of conditional variances that evolve as a GARCH (1, 1) process.

Figure 1 provides a pictorial representation of the abnormal returns for the Schedule 13D, voice and non-voice events, calculated using the market model for the period (−20, +20) in relation to each event. Based on these abnormal returns, we next proceed to formulating and testing the hypotheses of interest.

Figure 1. Average abnormal returns of Schedule 13D filings, voice and non-voice.

3.2. Market Responses to Schedule 13D, Voice and Non-Voice: Hypotheses and Testing

Our objective is to investigate the stock market response to the announcement of voice activism initiation in relation to other hedge fund activism announcements, such as the initial Schedule 13D filings (full sample) and the ex-post non-voice events. This is accomplished by testing the following null hypotheses:

$$H_{0A}:{\mu }_{St}=0,\mathrm{a}\mathrm{g}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{s}\mathrm{t}{H}_{1A}:{\mu }_{St}\ne 0,$$

(4)

$$H_{0B}:{\mu }_{Vt}=0,\mathrm{against}{H}_{1B}:{\mu }_{Vt}\ne 0,$$

(5)

$$H_{0C}:{\mu }_{NVt}=0,\mathrm{against}{H}_{1C}:{\mu }_{NVt}\ne 0,$$

(6)

where ${\mu }_{jt}=E\left(A{R}_{jt}\right),j=V,S,NV$ denotes the expected value of the abnormal returns of the specific event-based category j at event date t, with V, S and NV denoting the stocks that comprise the voice, Schedule 13D and non-voice event categories, respectively.

The null hypothesis $H_{0A}$ states that the stock market does not respond to the Schedule 13D filing announcements. Similarly, the stock market is hypothesized not to respond to voice and non-voice announcements under null hypotheses $H_{0B}$ and $H_{0C}$, respectively.The null hypotheses (4)–(6) are formulated based on the fact that an event will have no impact on stock returns if the average of the cross-sectional abnormal returns on the particular date is equal to zero. Parametric and non-parametric tests are employed to examine the no-mean-event-effect hypotheses6.

We also consider an additional testing approach that relies on the non-parametric Wilcoxon signed-rank test (Wilcoxon, 1945). According to Kolari and Pynnönen (2010), the Wilcoxon test outperforms the parametric test in terms of finite sample power, especially when it is applied to data that are fat-tailed-distributed7^,8.

3.3. Testing for Different Market Reactions Across Different Hedge Fund Activism Events

Further, we investigate the possibility that the market reacts differently to those events by testing the following hypotheses:

$$H_{0D}:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }=0,\mathrm{against}{H}_{1D}:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }\ne 0,$$

(10)

$$H_{0D}:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }=0,\mathrm{against}{H}_{1DR}:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }>0,$$

(11)

$$H_{0D}:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }=0,\mathrm{against}{H}_{1DL}:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }<0,$$

(12)

$$H_{0E}:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }=0,\mathrm{against}{H}_{1E}:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }\ne 0,$$

(13)

$$H_{0E}:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }=0,\mathrm{against}{H}_{1ER}:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }>0,$$

(14)

$$H_{0E}:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }=0,\mathrm{against}{H}_{1EL}:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }<0,$$

(15)

where ${\mu }_{jt}^{\prime }=E\left(SA{R}_{jt}\right),j=V,S,NV$ denotes the expected value of the standardized abnormal returns of the specific event-based category j at event date t. The sample mean is used to estimate ${\mu }_{jt}^{\prime }$. The standardized abnormal returns for each stock i are defined as

$$SA{R}_{it}={\displaystyle \frac{A{R}_{it}}{\sigma \left(A{R}_{it}\right)}},$$

(16)

where $\sigma \left(A{R}_{it}\right)=\sqrt{{\displaystyle \frac{1}{L_1-k}}{\sum }_{t=1}^{L_1}{u}_{it}^2}$ is the standard deviation of the regression prediction errors of each stock, k denotes the degrees of freedom and $L_1$ is the estimation window length. The parameter k is equal to one, four and five when the market model, the Carhart model and the GARCH-in-Mean models are used, respectively. The standardization of the abnormal returns by their standard deviation allows conducting reliable inference on the difference between their sample means because these samples have unequal lengths. The standard two-sample mean t-test of unequal variances of Welch (1947) that has been widely used in the literature is applied to the standardized abnormal returns to test hypotheses (10)–(15). The simulation results of Ruxton (2006) document that the Welch test is favorably compared to the conventional two-sample t-test of equal variances in terms of empirical size.

3.4. Heteroscedasticity-Robust and Non-Normality-Robust Tests

Several studies on testing the mean equality hypothesis raise concerns about possible inferential biases associated with the application of the existing test procedures under the presence of non-normality and heteroscedasticity. For instance, Algina et al. (1994) show that the Welch test faces size distortions when applied to data that are non-normally distributed. Descriptive statistics of stock returns, presented in Table 1, do indicate departures from normality; moreover, heteroscedasticity is a common feature in stock returns.

To allow for these stock returns characteristics in our testing procedure, we consider Yuen’s (1974) trimmed mean-based test. Wilcox (1997) documented that Yuen’s (1974) test is well sized and demonstrates enhanced power under a sequence of local alternatives. More recently, Keselman et al. (2004) proved that a modified version of Yuen’s test, introduced by Guo and Luh (2000) and implemented in conjunction with bootstrap confidence intervals, performs satisfactorily in finite samples. Their test is also based on trimmed means and it ensures robustness to skewness.

The two-sample trimmed mean tests of Yuen (1974) and Guo and Luh (2000) are applied to the standardized abnormal returns in order to evaluate hypotheses (10)–(15). The test procedures are summarized below. Let $SA{R}_{j\left(1\right)}\le SA{R}_{j\left(2\right)}\le ...\le SA{R}_{j\left(N\right)}$ be the standardized abnormal returns of the event-based category j placed in ascending order. The sample trimmed mean of the category j is defined as

$${\mu }_{jt}^{*}={\displaystyle \frac{1}{h_j}}{\sum }_{k=q_j+1}^{N-q_j}SA{R}_{j\left(k\right)},$$

(17)

where $q_j$ denotes the lower and upper cut-off points of the distribution of the standardized abnormal returns, with $q_j=\left[gN\right]$, where $\gamma$ is the percentage of trimming applied to the tails of the distribution, and $h_j=N-2q_j$. Following the simulation results of Keselman et al. (2004), we set g equal to 10%. The Winsorized variance of the standardized abnormal returns is estimated as the sample variance of the Winsorized standardized abnormal returns:

$${\sigma }_{Wj}^2={\displaystyle \frac{1}{N-1}}{\sum }_{k=1}^N{\left(SA{R}_{jk}^{*}-{\mu }_{Wj}\right)}^2,$$

(18)

where

$$\begin{gathered} SA{R}_{jk}^{*}=\{SA{R}_{j\left(q_j+1\right)},ifSA{R}_{jk}\le SA{R}_{j\left(q_j+1\right)}SA{R}_{jk},ifSA{R}_{j\left(q_j+1\right)}<SA{R}_{jk}< \\ SA{R}_{j\left(N-q_j\right)}SA{R}_{j\left(N-q_j\right)},ifSA{R}_{jk}\ge SA{R}_{j(N-q_j)}, \end{gathered}$$

(19)

and

$${\mu }_{Wj}={\displaystyle \frac{1}{N}}\sum _{k=1}^N\left(SA{R}_{jk}^{*}\right).$$

(20)

Yuen’s (1974) test is defined as

$$t_Y={\displaystyle \frac{{\mu }_{Vt}^{*}-{\mu }_{St}^{*}}{{\left(p_V+p_S\right)}^{1/2}}},$$

(21)

where $p_j={\displaystyle \frac{(N-1){\sigma }_{\backslash Wj}^2}{h_j(h_j-1)}}$.

Guo and Luh (2000) proposed a modified version of Yuen’s test that filters out non-parametrically possible excess skewness by applying a Hall (1992) transformation to the original test statistic. Their test is defined as

$$t_H={\displaystyle \frac{\left\{({\mu }_{Vt}^{*}-{\mu }_{St}^{*})+\left(\frac{{\mu }_W}{6{\sigma }_W^2}\right)+\left(\frac{{\mu }_W}{3{\sigma }_W^4}\right){\left({\mu }_{Vt}^{*}-{\mu }_{St}^{*}\right)}^2+\left(\frac{{\mu }_W^2}{27{\sigma }_W^8}\right){\left({\mu }_{Vt}^{*}-{\mu }_{St}^{*}\right)}^3\right\}}{{\sigma }_W}}$$

(22)

where ${\mu }_W={\displaystyle \frac{{\mu }_{3V}}{h_V^2}}-{\displaystyle \frac{{\mu }_{3S}}{h_S^2}}$, ${\mu }_{3j}={\displaystyle \frac{1}{N}}{\sum }_{k=1}^N{\left(SA{R}_{jk}^{*}-{\mu }_{Wj}\right)}^3$ and ${\sigma }_W^2=p_V+p_S$.

Under the null hypothesis of mean equality, both test statistics are distributed as Student’s t with degrees of freedom equal to

$${\nu }^{*}={\displaystyle \frac{{\left(p_V+p_S\right)}^2}{\left\{\left(\frac{p_V^2}{(h_V-1)}\right)+\left(\frac{p_S^2}{(h_S-1)}\right)\right\}}}$$

(23)

Bootstrap critical values are employed for inference. The bootstrap simulations were conducted on the basis of the following steps: First, calculate the zero-mean series $L_{jt}=SA{R}_{jt}-{\mu }_{jt}^{*}$ for the standardized abnormal returns of each event-based category j. This transformation ensures that the empirical distributions of both samples $L_{jt}$ share a common measure of location. Second, generate bootstrap series ${\tilde{L}}_{jt}$ of length N by randomly sampling with replacement from the original series $L_{jt}.$Third, calculate the bootstrap test ${\tilde{t}}_Y$ (or ${\tilde{t}}_H$) using the bootstrap samples ${\tilde{L}}_{jt}$. Fourth, calculate the test statistic $t_Y$ (or $t_H$) using the actual data $L_{jt}.$ Fifth, repeat the first three steps of the procedure B times. A sequence of B pseudo-test values ${\left\{{\tilde{t}}_{Yb}\right\}}_{b=1}^B$ (or ${\left\{{\tilde{t}}_{Hb}\right\}}_{b=1}^B$) is generated. In our empirical investigation, 2500 bootstrap replications are used. Sixth, the null hypothesis of mean equality is rejected at level α if the condition $t_Y\in \left[{\tilde{t}}_{Y\alpha },{\tilde{t}}_{Y(1-\alpha )}\right]$ is not satisfied, with ${\tilde{t}}_{Y\alpha }$ and ${\tilde{t}}_{Y(1-\alpha )}$ representing the lower α and upper (1 − α) percentile of the distribution of ${\left\{{\tilde{t}}_{Yb}\right\}}_{b=1}^B$, respectively.

4. Empirical Findings and Implications

4.1. Abnormal Returns Around the Schedule 13D Announcement

The results for testing the null hypothesis reflected in (4), namely that the stock market does not respond to the Schedule 13D filing announcements, are reported in Table 2 for the market model, the Carhart (1997) model and the GARCH-in-Mean with t distribution model of abnormal returns. Table 2 reports the abnormal stock returns 10 days before and after the announcement date of the Schedule 13D filings, the p-values on the statistical significance of the abnormal returns based on the parametric t-test in (7) (denoted as “p-val”), and the p-value of the non-parametric Wilcoxon signed-rank test. The last column in each table presents the percentage of the positive abnormal returns for the event date.

Table 2. Abnormal returns of Schedule 13D filings.

Models	Market Model				Carhart Model				GARCH-in-Mean with Student’s t Distribution				% of +ARs
Event Day	CAR	AR	p-Val	p-Val (Wilc.)	CAR	AR	p-Val	p-Val (Wilc.)	CAR	AR	p-Val	p-Val (Wilc.)
−10	0.258	0.258	0.310	0.008 ***	0.304	0.304	0.232	0.026 **	0.313	0.313	0.227	0.025 **	44.545
−9	0.913	0.655	0.013 **	0.435	0.955	0.651	0.013 **	0.670	1.004	0.691	0.012 **	0.678	46.623
−8	1.552	0.639	0.009 ***	0.478	1.614	0.659	0.007 ***	0.255	1.718	0.714	0.004 ***	0.178	50.779
−7	2.139	0.587	0.040 **	0.745	2.225	0.611	0.031 **	0.951	2.382	0.664	0.020 **	0.797	48.831
−6	2.698	0.559	0.001 ***	0.009 ***	2.742	0.517	0.003 ***	0.041 **	2.939	0.557	0.001 ***	0.034 **	53.376
−5	3.14	0.442	0.003 ***	0.024 **	3.143	0.401	0.006 ***	0.147	3.376	0.437	0.004 ***	0.098 *	50.259
−4	3.29	0.150	0.209	0.414	3.256	0.113	0.343	0.748	3.525	0.149	0.225	0.605	50.649
−3	3.753	0.463	0.125	0.557	3.697	0.441	0.145	0.700	3.995	0.470	0.121	0.582	50.389
−2	3.778	0.025	0.897	0.279	3.714	0.017	0.931	0.321	4.04	0.045	0.817	0.505	47.402
−1	4.223	0.445	0.001 ***	0.004 ***	4.166	0.452	0.001 ***	0.003 ***	4.518	0.478	0.000 ***	0.002 ***	52.467
0	5.011	0.788	0.000 ***	0.000 ***	5.033	0.867	0.000 ***	0.000 ***	5.408	0.890	0.000 ***	0.000 ***	53.376
1	5.672	0.661	0.000 ***	0.000 ***	5.69	0.657	0.000 ***	0.000 ***	6.095	0.687	0.000 ***	0.000 ***	53.116
2	5.797	0.125	0.308	0.275	5.813	0.123	0.321	0.402	6.245	0.150	0.234	0.481	44.545
3	5.834	0.037	0.748	0.256	5.827	0.014	0.898	0.163	6.286	0.041	0.716	0.239	45.584
4	6.029	0.195	0.127	0.654	5.984	0.157	0.205	0.424	6.471	0.185	0.144	0.518	47.013
5	6.206	0.177	0.148	0.417	6.095	0.111	0.361	0.846	6.607	0.136	0.270	0.787	49.480
6	6.512	0.306	0.142	0.414	6.392	0.297	0.150	0.428	6.928	0.321	0.121	0.630	46.883
7	6.597	0.085	0.636	0.050 *	6.512	0.120	0.501	0.294	7.073	0.145	0.421	0.302	47.272
8	6.575	−0.022	0.878	0.149	6.46	−0.052	0.713	0.216	7.047	−0.026	0.854	0.226	46.233
9	6.926	0.351	0.119	0.122	6.828	0.368	0.101	0.109	7.449	0.402	0.075 *	0.100	52.207
10	6.928	0.002	0.988	0.523	6.855	0.027	0.831	0.615	7.509	0.060	0.628	0.911	48.181

Notes: Table 2 reports the abnormal stock returns (and the corresponding cumulative abnormal returns) 10 days before and after the Schedule 13D filing date. Three models are used to compute the abnormal returns, namely the market model, the Carhart model and the GARCH-in-mean model. We also report the p-values on the statistical significance of the abnormal returns, in particular the p-value of the parametric test (denoted as “p-val”) and the p-value of the Wilcoxon signed-rank test. The last column presents the percentage of the positive abnormal returns for each event date. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

The results from Table 2 show that over the period (−1, +1), the abnormal returns are statistically different from zero at the 1% level, for all models of abnormal returns. In addition, there is evidence of statistical significance of abnormal returns for a period starting nine trading days prior to the Schedule 13D date, which is justified by the fact that the hedge fund has a deadline of 10 days to disclose the information. The rejection of the null hypothesis in (4), that the stock market does not respond to the Schedule 13D filing announcements, is supported by both the parametric t-test and the non-parametric Wilcoxon test. The statistically significant abnormal returns to Schedule 13D announcements are positive, suggesting short-term value creation.

These findings are in line with those of previous studies on U.S. hedge fund activism, which, however, were obtained on the basis of either shorter or relatively earlier periods. For instance, the results of Brav et al. (2008a, 2008b) cover the period 2001–2006, whilst Becht et al. (2017), Bebchuk et al. (2015), and Clifford (2008) considered the periods 2000–2010, 1994–2007, 2003–2005 and 1998–2005, respectively. According to Klein and Zur (2009), results may differ across samples; therefore, the current results, for a period up to 2013 and including events that occurred within the 2007–2009 financial crisis, can be interpreted as extending the earlier results for a recent period that encompasses the turbulent span of a major financial crisis.

4.2. Abnormal Returns Around Voice

The results from testing the null hypothesis (5), namely that the stock market does not respond to voice announcements, are reported in Table 3. The message from this table is as follows: Statistically significant, in most cases at the 1% level, abnormal returns exist over the period (0, +4) and on date −9, regardless of the abnormal returns model and the test statistic employed. The abnormal returns are positive and are approximately 1.11%. Importantly, compared to the abnormal returns around the Schedule 13D, the voice abnormal returns are higher by approximately 64%. Thus, voice yields short-term value creation9.

Table 3. Abnormal returns of voice.

Models	Market Model				Carhart Model				GARCH-in-Mean with Student’s t Distribution				% of +ARs
Event Day	CAR	AR	p-Val	p-Val (Wilc.)	CAR	AR	p-Val	p-Val (Wilc.)	CAR	AR	p-Val	p-Val (Wilc.)
−10	−0.132	−0.132	0.543	0.396	−0.117	−0.117	0.593	0.505	−0.130	−0.130	0.556	0.512	47.703
−9	−0.584	−0.452	0.036 **	0.012 **	−0.525	−0.408	0.053 *	0.041 **	−0.551	−0.421	0.043 **	0.032 **	44.170
−8	−0.532	0.052	0.840	0.417	−0.544	−0.019	0.941	0.238	−0.582	−0.031	0.903	0.244	47.703
−7	−0.655	−0.123	0.728	0.857	−0.645	−0.101	0.773	0.912	−0.706	−0.124	0.724	0.902	50.177
−6	−0.215	0.440	0.219	0.434	−0.182	0.463	0.195	0.209	−0.252	0.454	0.201	0.195	52.650
−5	−0.04	0.175	0.539	0.776	−0.017	0.165	0.559	0.994	−0.110	0.142	0.612	0.943	48.763
−4	−0.159	−0.119	0.628	0.689	−0.127	−0.110	0.649	0.825	−0.234	−0.124	0.611	0.759	50.883
−3	−0.284	−0.125	0.644	0.244	−0.194	−0.067	0.804	0.109	−0.316	−0.082	0.762	0.120	56.184
−2	−0.28	0.004	0.991	0.892	−0.188	0.006	0.987	0.545	−0.323	−0.007	0.984	0.597	50.177
−1	0.188	0.468	0.110	0.031 **	0.252	0.440	0.137	0.054 *	0.101	0.424	0.148	0.063 *	53.004
0	1.240	1.052	0.004 ***	0.000 ***	1.319	1.067	0.003 ***	0.000 ***	1.166	1.065	0.004 ***	0.000 ***	61.131
1	3.059	1.819	0.000 ***	0.000 ***	3.146	1.827	0.000 ***	0.000 ***	2.984	1.818	0.000 ***	0.000 ***	56.537
2	3.60	0.541	0.026 **	0.293	3.688	0.542	0.024 **	0.255	3.510	0.526	0.031 **	0.264	49.823
3	4.172	0.572	0.023 **	0.037 **	4.218	0.530	0.031 **	0.028 **	4.032	0.522	0.036 **	0.037 **	52.297
4	4.709	0.537	0.011 **	0.014 **	4.663	0.445	0.030 **	0.048 **	4.469	0.437	0.031 **	0.051 *	53.710
5	4.970	0.261	0.357	0.731	4.880	0.217	0.440	0.958	4.676	0.207	0.462	0.973	48.763
6	5.283	0.313	0.123	0.416	5.202	0.322	0.106	0.328	4.979	0.303	0.127	0.378	55.830
7	5.661	0.378	0.160	0.978	5.555	0.353	0.184	0.914	5.322	0.343	0.193	0.969	50.883
8	5.907	0.246	0.254	0.724	5.859	0.304	0.163	0.766	5.612	0.290	0.186	0.803	49.823
9	6.239	0.332	0.339	0.329	6.152	0.293	0.402	0.471	5.891	0.279	0.437	0.554	49.823
10	6.959	0.720	0.026 **	0.175	6.872	0.720	0.025 **	0.180	6.609	0.718	0.031 **	0.181	51.590

Notes: Table 3 reports the abnormal stock returns (and the corresponding cumulative abnormal returns) 10 days before and after the voice date. Three models are used to compute the abnormal returns, namely the market model, the Carhart model and the GARCH-in-Mean model. We also report the p-values on the statistical significance of the abnormal returns, in particular the p-value of the parametric test (denoted as p-val) and the p-value of the Wilcoxon signed-rank test. The last column presents the percentage of the positive abnormal returns for each event date. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

These results show that the news disclosed at voice entails information revelation that is reflected in the stock market by a positive response. According to Suominen (2001), information revelation has empirical implications related to conditional volatility (i.e., GARCH-type), which justifies the adoption of the GARCH-in-Mean model for abnormal returns. Based on this contention, we will address the additional aspect of asymmetric conditional volatility using EGARCH and GARCH-GJR models in the following section for robustness. Furthermore, as the average voice date comes chronologically after the average Schedule 13D date, the anticipation by market participants of subsequent voice, created by the Schedule 13D announcement, can result in arbitrage due to the positive abnormal returns around voice identified in our results. This point is related to the contention that there are many aspects to consider when evaluating the effects of information disclosure and the optimal regulation of the level and form of disclosure (Goldstein & Yang, 2017, p. 122).

4.3. Difference Between Abnormal Returns of Schedule 13D and Voice

We next turn to testing the hypotheses in (10)–(12), namely that the market reacts differently to the Schedule 13D and voice events. Table 4 reports the test results of the difference between the average standardized abnormal returns of the voice and the average standardized abnormal returns of the Schedule 13D filing dates. In hypotheses (10)–(12), ${\mu }_{Vt}^{\prime }$ denotes the average standardized abnormal returns of the voice, while ${\mu }_{St}^{\prime }$ denotes the average abnormal returns of the Schedule 13D filing dates. We report the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }\ne 0$(hypothesis in (10)). In this table, “p-val right” denotes the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }>0$(hypothesis in (11)), and “p-val left” denotes the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }<0$ (hypothesis in (12)).

Table 4. Testing for the difference between the standardized abnormal returns of Schedule 13D filings and voice.

Models	Market Model				Carhart Model				GARCH-in-Mean with Student’s t Distribution
Event Day	t-Test	p-Val	p-Val Right	p-Val Left	t-Test	p-Val	p-Val Right	p-Val Left	t-Test	p-Val	p-Val Right	p-Val Left
−10	−1.185	0.236	0.882	0.118	−1.231	0.219	0.891	0.109	−1.273	0.203	0.898	0.102
−9	−3.467	0.001 ***	1.000	0.000 ***	−3.359	0.001 ***	1.000	0.000 ***	−3.464	0.001 ***	1.000	0.000 ***
−8	−1.654	0.098 *	0.951	0.049 **	−1.914	0.056 *	0.972	0.028 **	−2.047	0.041 **	0.980	0.020 **
−7	−1.466	0.143	0.928	0.072 *	−1.517	0.130	0.935	0.065 *	−1.647	0.100	0.950	0.050 *
−6	−0.706	0.481	0.760	0.240	−0.448	0.654	0.673	0.327	−0.561	0.575	0.712	0.288
−5	−0.531	0.595	0.702	0.298	−0.393	0.695	0.653	0.347	−0.529	0.597	0.702	0.298
−4	−1.206	0.229	0.886	0.114	−1.033	0.302	0.849	0.151	−1.165	0.245	0.878	0.122
−3	−0.854	0.394	0.803	0.197	−0.631	0.529	0.736	0.264	−0.734	0.463	0.768	0.232
−2	−0.782	0.435	0.783	0.217	−0.805	0.421	0.789	0.211	−0.882	0.378	0.811	0.189
−1	0.246	0.806	0.403	0.597	0.141	0.888	0.444	0.556	0.073	0.942	0.471	0.529
0	1.597	0.111	0.056 *	0.944	1.419	0.157	0.078 *	0.922	1.327	0.185	0.093 *	0.907
1	3.230	0.001 ***	0.001 ***	0.999	3.195	0.002 ***	0.001 ***	0.999	3.145	0.002 ***	0.001 ***	0.999
2	1.848	0.065 *	0.033 **	0.967	1.912	0.057 *	0.028 **	0.972	1.704	0.089 *	0.045 **	0.955
3	2.491	0.013 **	0.007 ***	0.993	2.579	0.010 **	0.005 ***	0.995	2.384	0.018 **	0.009 ***	0.991
4	0.967	0.334	0.167	0.833	0.615	0.539	0.269	0.731	0.473	0.637	0.318	0.682
5	0.062	0.950	0.475	0.525	−0.012	0.990	0.505	0.495	−0.123	0.902	0.549	0.451
6	0.060	0.952	0.476	0.524	0.041	0.967	0.484	0.516	−0.075	0.940	0.530	0.470
7	0.854	0.394	0.197	0.803	0.602	0.547	0.274	0.726	0.522	0.602	0.301	0.699
8	0.959	0.338	0.169	0.831	1.413	0.158	0.079 *	0.921	1.293	0.197	0.098 *	0.902
9	0.917	0.360	0.180	0.820	0.759	0.448	0.224	0.776	0.669	0.504	0.252	0.748
10	1.642	0.102	0.051 *	0.949	1.606	0.109	0.055 *	0.945	1.495	0.136	0.068 *	0.932

Notes: Table 4 reports the test results of the difference between the average standardized abnormal returns of voice and the average standardized abnormal returns of the Schedule 13D filings. ${\mu }_{Vt}^{\prime }$ denotes the average standardized abnormal returns of the voice, while ${\mu }_{St}^{\prime }$ denotes the average abnormal returns of the Schedule 13D filing dates. We report the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }\ne 0$. “p-val right” denotes the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }>0$.“p-val left” denotes the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }<0$. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

Based on the “p-val”, results suggest that the difference in the abnormal returns is statistically significant over the period (+1, +3) under all three models of abnormal returns. Based on the “p-val right”, the statistically significant difference reveals that the abnormal returns of voice are significantly higher than the returns of Schedule 13D. These results indicate that the announcement made and information disclosed on the Schedule 13D date does not reveal the full but only partial information, with the remainder of the information being revealed at voice.

Our previous finding that the abnormal returns of voice are significantly higher than the returns of Schedule 13D is supported by the heteroscedasticity-robust tests of Yuen (1974) and Guo and Luh (2000). As shown in Table 5 and Table 6, both the Yuen (1974) and the Guo and Luh (2000) tests indicate that the abnormal returns of voice are different from the Schedule 13D returns (based on the “p-val”); moreover, the former are higher than the latter (based on the “p-val right”).The information disclosed at voice does reveal further news that is important and reflected in the market, supporting the conjecture that voice is related to information revelation in the U.S. stock market. Finally, this statistically significant difference between the voice and the Schedule 13D abnormal returns is approximately 0.4% and is interpreted as an average arbitrage profit that can be obtained between the two dates.

Table 5. Heteroscedasticity-robust two-sample t-test results for the difference between voice and Schedule 13D standardized abnormal returns—the market model.

									Bootstrap 95% Confidence Interval
Test Statistics	Yuen Test Statistic				Guo and Luh Test Statistic				Yuen Test Statistic		Guo and Luh Test Statistic
Event Day	t-Test	p-Val	p-Val Right	p-Val Left	t-Test	p-Val	p-Val Right	p-Val Left	Lower Bound	Upper Bound	Lower Bound	Upper Bound
−10	0.306	0.760	0.380	0.620	0.306	0.760	0.380	0.620	−1.965	2.032	−1.977	2.037
−9	−2.299	0.022 **	0.989	0.011 **	−2.309	0.022 **	0.989	0.011 **	−1.879	2.019	−1.888	2.013
−8	−0.661	0.509	0.746	0.254	−0.658	0.511	0.745	0.255	−1.952	1.924	−1.943	1.938
−7	−0.053	0.958	0.521	0.479	−0.052	0.959	0.521	0.479	−1.937	1.898	−1.938	1.912
−6	−0.171	0.864	0.568	0.432	−0.169	0.866	0.567	0.433	−2.086	2.004	−2.085	2.025
−5	−0.541	0.589	0.706	0.294	−0.540	0.590	0.705	0.295	−1.930	2.052	−1.939	2.066
−4	−1.077	0.282	0.859	0.141	−1.077	0.282	0.859	0.141	−1.983	2.107	−2.003	2.109
−3	0.739	0.460	0.230	0.770	0.738	0.461	0.231	0.769	−1.858	1.952	−1.873	1.956
−2	0.617	0.538	0.269	0.731	0.617	0.538	0.269	0.731	−1.948	1.925	−1.955	1.938
−1	0.557	0.578	0.289	0.711	0.563	0.574	0.287	0.713	−2.156	1.817	−2.125	1.850
0	1.747	0.082 *	0.041 **	0.959	1.768	0.078 *	0.039 **	0.961	−2.002	1.884	−1.993	1.940
1	2.845	0.005 ***	0.002 ***	0.998	2.927	0.004 ***	0.002 ***	0.998	−1.991	1.713	−1.946	1.755
2	1.389	0.166	0.083 *	0.917	1.397	0.163	0.082 *	0.918	−2.082	2.025	−2.066	2.042
3	2.465	0.014 **	0.007 ***	0.993	2.484	0.014 **	0.007 ***	0.993	−1.913	2.064	−1.900	2.073
4	1.816	0.070 *	0.035 **	0.965	1.818	0.070 *	0.035 **	0.965	−1.952	2.035	−1.959	2.035
5	−0.173	0.863	0.569	0.431	−0.172	0.863	0.568	0.432	−2.005	2.037	−2.002	2.044
6	0.833	0.405	0.203	0.797	0.836	0.404	0.202	0.798	−2.025	1.936	−2.021	1.944
7	0.465	0.643	0.321	0.679	0.465	0.642	0.321	0.679	−1.880	2.035	−1.880	2.043
8	0.486	0.627	0.314	0.686	0.488	0.626	0.313	0.687	−1.923	1.878	−1.924	1.896
9	0.440	0.660	0.330	0.670	0.443	0.658	0.329	0.671	−1.971	1.983	−1.969	2.000
10	1.577	0.116	0.058*	0.942	1.587	0.113	0.057	0.943	−1.951	2.051	−1.946	2.070

Notes: This table reports the results of the two-sample tests proposed by Yuen (1974) and Guo and Luh (2000) (denoted as Yuen and Hall test statistics, respectively)on the difference between the standardized abnormal returns of voice and the standardized abnormal returns of the Schedule 13D filing dates. The 95% bootstrap percentile interval is also reported for each test statistic. In total, 2500 replications are used in the bootstrap technique. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively. Results in bold indicate statistical significance at the 5% level using the confidence interval.

Table 6. Heteroscedasticity-robust two-sample t-test results for the difference between voice and Schedule 13D average standardized abnormal returns—the Carhart model.

									Bootstrap 95% Confidence Interval
Test Statistics	Yuen Test Statistic				Guo and Luh Test Statistic				Yuen Test Statistic		Guo and Luh Test Statistic
Event Day	t-Test	p-Val	p-Val Right	p-Val Left	t-Test	p-Val	p-Val Right	p-Val Left	Lower Bound	Upper Bound	Lower Bound	Upper Bound
−10	0.279	0.780	0.390	0.610	0.279	0.780	0.390	0.610	−1.828	1.989	−1.837	1.998
−9	−1.981	0.048 **	0.976	0.024 **	−1.990	0.047 **	0.976	0.024 **	−2.053	1.949	−2.056	1.951
−8	−1.329	0.185	0.908	0.092 *	−1.324	0.186	0.907	0.093 *	−1.946	1.886	−1.944	1.908
−7	−0.056	0.955	0.522	0.478	−0.056	0.956	0.522	0.478	−2.123	1.787	−2.135	1.801
−6	0.485	0.628	0.314	0.686	0.487	0.627	0.313	0.687	−2.056	1.939	−2.046	1.956
−5	−0.292	0.770	0.615	0.385	−0.291	0.772	0.614	0.386	−1.855	2.052	−1.857	2.065
−4	−0.670	0.503	0.748	0.252	−0.670	0.503	0.748	0.252	−1.830	2.012	−1.839	2.020
−3	1.244	0.214	0.107	0.893	1.242	0.215	0.107	0.893	−1.887	2.068	−1.903	2.078
−2	0.972	0.332	0.166	0.834	0.970	0.333	0.166	0.834	−1.873	2.056	−1.888	2.057
−1	0.525	0.600	0.300	0.700	0.530	0.597	0.298	0.702	−2.201	1.851	−2.172	1.884
0	1.455	0.147	0.073 *	0.927	1.466	0.144	0.072 *	0.928	−1.992	1.919	−1.983	1.953
1	2.950	0.003 ***	0.002 ***	0.998	3.039	0.003 ***	0.001 ***	0.999	−2.025	1.791	−1.982	1.836
2	1.321	0.187	0.094 *	0.906	1.328	0.185	0.093 *	0.907	−2.007	1.861	−2.000	1.880
3	2.670	0.008 ***	0.004 ***	0.996	2.689	0.008 ***	0.004 ***	0.996	−1.997	1.960	−1.988	1.974
4	1.584	0.114	0.057 *	0.943	1.586	0.114	0.057 *	0.943	−1.893	1.965	−1.890	1.967
5	−0.005	0.996	0.502	0.498	−0.003	0.998	0.501	0.499	−1.933	1.823	−1.937	1.839
6	0.852	0.395	0.197	0.803	0.853	0.394	0.197	0.803	−1.914	1.966	−1.910	1.972
7	0.094	0.925	0.463	0.537	0.094	0.925	0.462	0.538	−1.933	2.098	−1.938	2.103
8	0.938	0.349	0.174	0.826	0.940	0.348	0.174	0.826	−2.038	2.001	−2.030	2.008
9	0.069	0.945	0.472	0.528	0.070	0.944	0.472	0.528	−1.877	1.899	−1.872	1.903
10	1.605	0.110	0.055 *	0.945	1.615	0.107	0.054 *	0.946	−1.890	1.989	−1.877	2.013

See notes of Table 5.

These findings are in line with the dynamic trading models of Kyle (1985) and Ostrovsky (2012). Information revelation occurs when an informed trader, through his or her actions, takes advantage of his or her information and eventually moves the price of the stock to its correct value (Ostrovsky, 2012). In Kyle’s (1985) dynamic trading model, the informational content of prices is examined along with the value of information to an informed trader, and price innovations are modeled as a consequence of information revelation with the informed trader acting in such a way that his or her information is incorporated into prices. In our framework, the informed trader is the hedge fund, and its actions include the hedge fund activism and information disclosure at both the Schedule 13D and the voice date. Our results are an empirical manifestation of Kyle’s findings, in which a hedge fund (informed trader), through hedge fund activism information revelation actions, moves the security price to a new level.

4.4. Abnormal Returns of Voice When Voice Leads Schedule 13D Filings

As our sample records both the voice and the Schedule 13D dates, we identify 93 voice events that chronologically took place prior to the Schedule 13D filing date by10 trading days or less. The voice abnormal returns over the period (−10, +10) for the events are calculated using the three models, and results are reported in Table 7.

Table 7. Abnormal returns of voice events occurring 1 to 10 days before Schedule 13D events.

Models		Market Model				Carhart Model				GARCH with Student’s t Distribution
Event Day	CAR	AR	p-Val	p-Val (Wilc.)	CAR	AR	p-Val	p-Val (Wilc.)	CAR	AR	p-Val	p-Val (Wilc.)
−10	0.372	0.372	0.422	0.383	0.402	0.402	0.520	0.594	0.348	0.348	0.447	0.562
−9	−0.182	−0.554	0.230	−0.610	0.573	0.171	0.200	0.269	−0.291	−0.639	0.160	0.218
−8	−0.841	−0.659	0.121	−0.600	0.721	0.148	0.481	0.407	−0.910	−0.619	0.128	0.420
−7	−0.026	0.815	0.079 *	0.829	0.788	0.067 *	0.237	0.376	−0.090	0.820	0.071 *	0.267
−6	0.467	0.493	0.408	0.393	1.273	0.485	0.659	0.656	0.308	0.398	0.485	0.684
−5	−0.004	−0.471	0.348	−0.433	1.659	0.386	0.365	0.342	−0.117	−0.425	0.390	0.291
−4	−0.361	−0.357	0.343	−0.330	2.030	0.371	0.859	0.645	−0.444	−0.327	0.375	0.844
−3	−0.588	−0.227	0.617	−0.042	2.959	0.929	0.649	0.876	−0.483	−0.039	0.933	0.678
−2	−0.574	0.014	0.981	0.060	3.880	0.921	0.352	0.418	−0.419	0.064	0.915	0.350
−1	0.545	1.119	0.010 **	0.981	3.908	0.028 **	0.048 **	0.034 **	0.571	0.990	0.026 **	0.037 **
0	0.361	−0.184	0.703	−0.178	4.619	0.711	0.558	0.656	0.385	−0.186	0.697	0.539
1	1.038	0.677	0.145	0.619	4.808	0.189	0.024 **	0.026	1.001	0.616	0.188	0.030 **
2	0.662	−0.376	0.520	−0.380	5.331	0.523	0.187	0.172	0.629	−0.372	0.526	0.192
3	1.452	0.790	0.108	0.926	5.380	0.049 **	0.130	0.325	1.553	0.924	0.054 *	0.159
4	1.646	0.194	0.579	0.130	6.066	0.686	0.283	0.268	1.681	0.128	0.696	0.323
5	2.767	1.121	0.113	0.988	6.230	0.164	0.931	0.575	2.680	0.999	0.157	0.897
6	3.442	0.675	0.078 *	0.655	6.312	0.082 *	0.216	0.234	3.388	0.708	0.064 *	0.141
7	4.086	0.644	0.231	0.754	6.480	0.168	0.618	0.642	4.177	0.789	0.157	0.598
8	4.435	0.349	0.303	0.296	6.873	0.393	0.726	0.676	4.523	0.346	0.311	0.611
9	4.906	0.471	0.218	0.412	7.146	0.273	0.484	0.520	4.967	0.444	0.236	0.552
10	5.360	0.454	0.332	0.389	7.557	0.411	0.133	0.061 *	5.377	0.410	0.379	0.113

Notes: Table 7 reports the abnormal stock returns (and the corresponding cumulative abnormal returns) 10 days before and after the voice date. Three models are used to compute the abnormal returns, namely the market model, the Carhart model, and the GARCH-in-Mean model. We also report the p-values on the statistical significance of the abnormal returns, in particular the p-value of the parametric test (denoted as p-val) and the p-value of the Wilcoxon signed-rank test. The last column presents the percentage of the positive abnormal returns for each event date. * and ** indicate statistical significance at the 10% and5% levels, respectively.

Based on inference from both the parametric and the non-parametric Wilcoxon tests, the results show that, at the 5% level, there exist statistically significant abnormal returns one trading day prior to the voice date (−1), under all three models. These returns are positive, in line with the previously documented results in Table 3. The main message that emerges is that voice on its own creates short-term value, and not necessarily as a result of the Schedule 13D filing. Markets respond to hedge fund voice, regardless of whether this occurs prior to or after the Schedule 13D, which signals significant voice-related information revelation.

4.5. Abnormal Returns of Non-Voice and Testing the Difference Between Voice and Non-Voice

As hedge funds aim to change the strategy, the management, or the governance of a target company, they may come into conflict with the managers or the dominant shareholders who control the target company (Pacces, 2016). The choice between voice and non-voice (i.e., exiting the target firm) can be the outcome of this conflict as a rational decision by the hedge fund activist. This decision is dependent on whether the hedge fund can influence the target, whether the hedge fund needs to influence the target or whether the expected gain of influencing the target exceeds the cost. Parameters that affect this choice include the extent to which the target firm already operates at maximum performance and the degree to which other investors in the target firm are dissatisfied with the management (Admati & Pfleiderer, 2009; Kedia et al., 2017).

The results from testing the null hypothesis in (6), regarding the abnormal returns of non-voice, are reported in Table 8. The results indicate that non-voice is also related to positive abnormal returns over the period (−1, +1). These results are qualitatively similar to those reported in Table 2 for the full sample of Schedule 13D announcements. Positive abnormal returns for non-voice could be interpreted as suggesting that the hedge fund activist, after monitoring the firm’s operations, realizes that there is no scope for improving action. As contended by Brav et al. (2008b, p. 1748), some hedge fund activists hope to facilitate value-enhancing changes in the target company as minority shareholders without taking control of the target firm’s board of directors. Non-voice may signal to the stock market that the target firm’s value is already high enough, so no value-enhancing changes can be made. This can be the case when the target firm is already at an optimum level in terms of operations, strategy, etc., and the hedge fund activist has nothing more to offer (hence the non-voice) and exits the minority stake.

Table 8. Abnormal returns of non-voice.

Models		Market Model				Carhart Model				GARCH-in-Mean with Student’s t Distribution			% of +ARs
Event Day	CAR	AR	p-Val	p-Val (Wilc.)	CAR	AR	p-Val	p-Val (Wilc.)	CAR	AR	p-Val	p-Val (Wilc.)
−10	0.563	0.563	0.141	0.022 **	0.595	0.595	0.121	0.033 **	0.615	0.615	0.116	0.041 **	43.983
−9	1.611	1.048	0.009 ***	0.863	1.613	1.018	0.011 **	0.867	1.696	1.081	0.010 **	0.795	47.718
−8	2.458	0.847	0.017 **	0.630	2.481	0.868	0.014 **	0.383	2.649	0.953	0.008 ***	0.277	49.585
−7	3.603	1.145	0.007 ***	0.610	3.637	1.156	0.006 ***	0.514	3.889	1.240	0.003 ***	0.365	49.585
−6	3.967	0.364	0.031 **	0.210	3.927	0.290	0.085 *	0.678	4.241	0.352	0.036 **	0.561	50.622
−5	4.124	0.157	0.261	0.784	4.051	0.124	0.359	0.865	4.419	0.178	0.218	0.926	48.340
−4	4.211	0.087	0.449	0.179	4.114	0.063	0.582	0.346	4.533	0.114	0.356	0.234	52.282
−3	4.919	0.708	0.122	0.932	4.823	0.709	0.123	0.909	5.290	0.757	0.100	0.875	49.378
−2	5.064	0.145	0.510	0.078 *	4.943	0.120	0.582	0.075 *	5.454	0.164	0.457	0.160	44.191
−1	5.420	0.356	0.027 **	0.188	5.333	0.390	0.014 **	0.089 *	5.881	0.427	0.007 ***	0.042 **	51.660
0	5.952	0.532	0.007 ***	0.977	5.980	0.647	0.001 ***	0.123	6.557	0.676	0.002 ***	0.131	47.510
1	6.326	0.374	0.030 **	0.180	6.369	0.389	0.021 **	0.192	6.988	0.431	0.010 **	0.133	51.037
2	6.312	−0.014	0.921	0.031 **	6.365	−0.004	0.979	0.059 *	7.021	0.033	0.823	0.070 *	42.946
3	6.334	0.022	0.875	0.063 *	6.345	−0.020	0.882	0.024 **	7.036	0.015	0.912	0.053 *	43.776
4	6.362	0.028	0.861	0.061 *	6.339	−0.006	0.968	0.047 **	7.063	0.027	0.859	0.060 *	44.813
5	6.591	0.229	0.114	0.873	6.508	0.169	0.243	0.561	7.261	0.198	0.186	0.598	47.303
6	6.830	0.239	0.444	0.091 *	6.754	0.246	0.428	0.100	7.532	0.271	0.383	0.154	45.436
7	6.794	−0.036	0.883	0.029 **	6.760	0.006	0.979	0.263	7.568	0.036	0.882	0.251	45.851
8	6.809	0.015	0.938	0.691	6.701	−0.059	0.761	0.628	7.539	−0.029	0.881	0.629	47.095
9	7.173	0.364	0.290	0.646	7.113	0.412	0.232	0.472	7.989	0.450	0.193	0.443	50.622
10	6.953	−0.220	0.191	0.021 **	6.959	−0.154	0.362	0.113	7.875	−0.114	0.490	0.217	47.303

Notes: Table 8 reports the abnormal stock returns (and the corresponding cumulative abnormal returns) 10 days before and after the Schedule 13D filing date of the firms that were not subject to voice. Two models are used to compute the abnormal returns, namely the market model and the Carhart model. We also report the p-values on the statistical significance of the abnormal returns, in particular the p-value of the parametric test (denoted as “p-val”) and the p-value of the Wilcoxon signed-rank test. The last column presents the percentage of the positive abnormal returns for each event date. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

Another justification of the positive abnormal returns of non-voice (exit) is provided by Admati and Pfleiderer (2009). Over the period from the date of the Schedule 13D to the date of exit, the threat of exit can be a form of activism (Admati & Pfleiderer, 2009). Palmiter (2001, pp. 1437–1438) suggests that large shareholders may be able to affect managerial decisions through the “threat (actual or implied) of selling their holdings and driving down the price of the targeted company”. If managers’ compensation is linked to share prices and if the exit of a large shareholder has a negative price impact, then the presence of a large shareholder, who is potentially able to trade on private information, may help discipline the management and improve corporate governance (Admati & Pfleiderer, 2009). So, exiting may imply that, through the threat of exit, the objective of improving corporate performance has been accomplished.

Combining the results on positive abnormal returns for voice and non-voice suggests that stock market participants can obtain positive abnormal returns in any state of nature following the Schedule 13D announcement, regardless of whether the hedge fund is eventually engaged with the target (voice) or exits (non-voice). This further indicates that the disclosure of information at the Schedule 13D announcement creates anticipation in market participants that, regardless of the eventual voice–exit decision, arbitrage profits (abnormal returns) are to be generated sometime during the course of the hedge fund’s holding of the minority stake. In other words, information revelation occurs at either the voice or the exit decision.

We next turn to testing the null hypotheses reflected in (13)–(15), namely that the previously identified abnormal returns for voice and non-voice are different (Hypothesis (13)). If we find that they are, we can test whether the voice returns are higher than the non-voice returns (Hypothesis (14)), or the opposite (Hypothesis (15)). The results are reported in Table 9. In line with Table 4, this table reports the test results of the difference between the average standardized abnormal returns of the voice and the average standardized abnormal returns of the non-voice. Denoting the average standardized abnormal returns of the voice by ${\mu }_{Vt}^{\prime }$ and the average abnormal returns of the non-voice with ${\mu }_{NVt}^{\prime }$, Table 9 reports the p-value of the t-statistic when the alternative hypothesis is (“p-val”)$H_1:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }\ne 0$. In addition, it reports the “p-val right”, denoting the p-value of the t-statistic when the alternative hypothesis is$H_1:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }>0$, and the “p-val left”, denoting the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }<0$.

Table 9. Testing for the difference between the standardized abnormal returns of voice and non-voice.

Models	Market Model				Carhart Model				GARCH-in-Mean with Student’s t Distribution
Event Day	t-Test	p-Val	p-Val Right	p-Val Left	t-Test	p-Val	p-Val Right	p-Val Left	t-Test	p-Val	p-Val Right	p-Val Left
−10	−1.671	0.095 *	0.952	0.048 **	−1.650	0.099 *	0.950	0.050 *	−1.684	0.093 *	0.954	0.046 **
−9	−3.713	0.000 ***	1.000	0.000 ***	−3.607	0.000 ***	1.000	0.000 ***	−3.670	0.000 ***	1.000	0.000 ***
−8	−1.847	0.065 *	0.967	0.033 **	−2.068	0.039 **	0.981	0.019 **	−2.207	0.028 **	0.986	0.014 **
−7	−2.209	0.027 **	0.986	0.014 **	−2.242	0.025 **	0.987	0.013 **	−2.385	0.017 **	0.991	0.009 ***
−6	−0.338	0.736	0.632	0.368	0.049	0.961	0.481	0.519	−0.104	0.917	0.541	0.459
−5	0.360	0.719	0.360	0.640	0.474	0.636	0.318	0.682	0.299	0.765	0.382	0.618
−4	−1.053	0.293	0.854	0.146	−0.928	0.354	0.823	0.177	−1.090	0.277	0.862	0.138
−3	−1.122	0.262	0.869	0.131	−0.986	0.324	0.838	0.162	−1.103	0.270	0.865	0.135
−2	−1.120	0.263	0.868	0.132	−1.130	0.259	0.871	0.129	−1.224	0.222	0.889	0.111
−1	0.535	0.593	0.297	0.703	0.379	0.705	0.352	0.648	0.308	0.758	0.379	0.621
0	2.861	0.004 ***	0.002 ***	0.998	2.589	0.010 **	0.005 ***	0.995	2.430	0.016 **	0.008 ***	0.992
1	3.830	0.000 ***	0.000 ***	1.000	3.764	0.000 ***	0.000 ***	1.000	3.696	0.000 ***	0.000 ***	1.000
2	2.452	0.015 **	0.007 ***	0.993	2.459	0.014 **	0.007 ***	0.993	2.225	0.027 **	0.013 **	0.987
3	2.490	0.013 **	0.007 ***	0.993	2.638	0.009 ***	0.004 ***	0.996	2.426	0.016 **	0.008 ***	0.992
4	1.765	0.078 *	0.039 **	0.961	1.434	0.152	0.076 *	0.924	1.297	0.195	0.098 *	0.902
5	−0.010	0.992	0.504	0.496	−0.094	0.925	0.537	0.463	−0.196	0.844	0.578	0.422
6	0.838	0.402	0.201	0.799	0.722	0.471	0.235	0.765	0.629	0.530	0.265	0.735
7	1.239	0.216	0.108	0.892	0.979	0.328	0.164	0.836	0.893	0.373	0.186	0.814
8	0.575	0.566	0.283	0.717	1.171	0.242	0.121	0.879	1.061	0.289	0.145	0.855
9	1.145	0.253	0.126	0.874	0.943	0.346	0.173	0.827	0.848	0.397	0.198	0.802
10	2.097	0.037 **	0.018 **	0.982	1.904	0.058 *	0.029 **	0.971	1.777	0.076 *	0.038 **	0.962

Notes: Table 9 reports the test results of the difference between the average standardized abnormal returns of voice and the average standardized abnormal returns of non-voice. ${\mu }_{Vt}^{\prime }$ denotes the average standardized abnormal returns of voice, while ${\mu }_{NVt}^{\prime }$ denotes the average abnormal returns of non-voice. We report the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }\ne 0$. “p-val right” denotes the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{NVt}^{\prime }>0$.“p-val left” denotes the p-value of the t-statistic when the alternative hypothesis is $H_1:{\mu }_{Vt}^{\prime }-{\mu }_{St}^{\prime }<0$. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

The results suggest that the voice abnormal returns are different from the non-voice returns over the period (0, +3) at the 5% or 1% level, regardless of the abnormal returns model. Thus, Hypothesis (13) is rejected in favor of its alternative. Moving on to identifying which abnormal returns are higher, the “p-val right” indicates that the difference between the voice returns and the non-voice returns is strongly significant over the period (0, +3). Thus, we conclude that the voice average abnormal returns are higher than the non-voice average abnormal returns at the 1% level at or after the event. These findings suggest that voice announcements reveal more influential information than the non-voice announcements. Stock markets respond more aggressively to voice than to Schedule 13D, signaling richer information revelation at voice events.

5. Robustness

5.1. Asymmetric (Leverage) Volatility Effects and GARCH Models for Abnormal Returns

Suominen (2001) has shown that information revelation in stock markets suggests that the expected price variability looks similar to a conditional variance GARCH-type model. In this section, we add another aspect of modeling time-varying volatility, namely allowing for asymmetric or leverage effects in the GARCH model, namely for the fact that negative shocks have a greater impact on stock return volatility than positive shocks. The most well-known GARCH-type models incorporating asymmetric effects are the exponential GARCH (EGARCH) (Nelson, 1991) and the GARCH-GJR (Glosten et al., 1993).

We only discuss the results without reporting them due to space limitations10. The Schedule 13D abnormal returns, based on the EGARCH-in-Mean and the GJR-GARCH-in-Mean models with the Student’s distribution, are statistically significant for the period (−1, +1) at the 1% level of significance. These findings are exactly the same as those reported in Table 2 for the market model, the Carhart model and the simple GARCH-in-Mean model. Thus, the results on the significance of the Schedule 13D abnormal returns are robust to asymmetric conditional variance effects11.

Similar results are obtained for the voice abnormal returns. Under both the EGARCH-in-Mean and the GJR-GARCH-in-Mean models, the voice abnormal returns are different from 0 and positive over the period (0, +4) at either the 5% or the 1% level, based on both the parametric and the non-parametric tests. These findings are very similar to those under the previously examined three models in Table 3.

Finally, for the non-voice abnormal returns, the results indicate that these returns are different from 0 over the period (−1, +1) under both models based on the parametric test. However, when using the non-parametric Wilcoxon test, the abnormal returns are significant only on day −1.

The main message from these robustness checks is as follows: Volatility clustering is the main empirical implication of information revelation (Suominen, 2001). This clustering may carry (especially in stock returns) an additional empirical characteristic, namely asymmetry (or leverage effects) (Nelson, 1991; Glosten et al., 1993). We illustrate that the results on the statistical significance of abnormal returns around primarily the voice events (as well as around the Schedule 13D and non-voice events) are robust to both volatility clustering and leverage effects.

Finally, based on the EGARCH-in-Mean and GJR-GARCH-in-Mean models, we proceed to testing the difference between the Schedule 13D abnormal returns and voice returns and the difference between the voice abnormal returns and the non-voice returns. We document that the voice abnormal returns are higher than the Schedule 13D abnormal returns over the period (+1, +3) at the 1% or 5% level.

Turning to the difference between voice and non-voice abnormal returns, we find that over the period (0, +4) the voice returns are higher than the non-voice returns. These findings are in line with those under the market model, the Carhart model and the simple GARCH-in-Mean model and suggest that our results are robust to models of abnormal returns, allowing for richer empirical characteristics in the time-varying volatility process.

5.2. Further Evidence from Alternative Non-Parametric Statistical Procedures

To assess the robustness of our results to alternative inferential statistical procedures, we employ the rank test procedure of Corrado (1989) and Corrado and Zivney (1992). This is a non-parametric procedure that is based on the transformation of all the combined estimation windows and event period abnormal returns into their respective ranks. The test statistic is defined as

$$t_{R,t}={\displaystyle \frac{\frac{1}{N^{\prime }}{\sum }_{i=1}^{N^{\prime }}\left(S{K}_{it}-0.5\right)}{\sigma \left(SK\right)}}$$

(24)

where $S{K}_{it}={\displaystyle \frac{K\left(A{R}_{it}\right)}{1+W_i}}$, $W_i$ is the number of non-missing returns of both estimation and event window for stock i and $N^{\prime }$ is the number of non-missing returns across stocks. The standard deviation is defined as

$$\sigma \left(SK\right)=\sqrt{{\displaystyle \frac{1}{L}}\sum _{t=T_0}^{T_2}{\left[{\displaystyle \frac{1}{N^{\prime }}}\sum _{i=1}^{N^{\prime }}\left(S{K}_{it}-0.5\right)\right]}^2}$$

(25)

where L is the length of both estimation and event windows, $T_0$ is the first day of the estimation window and $T_2$ is the last day of the event window. Under the null hypothesis that the average abnormal returns at event day t are equal to zero, the test statistic $t_{R,t}$ is distributed as standard normal. A fixed-length estimation window is used to calculate the abnormal returns and the Corrado rank statistics for each event day. Estimation windows of length equal to 198, 212, and 212 observations are selected for the voice, the Schedule 13D and the non-voice event-based stocks, respectively.

The results are reported in Table 10, Table 11 and Table 12, for the Schedule 13D, voice and non-voice abnormal returns, respectively. These tables show that voice is associated with statistically significant (at the 5% level) abnormal returns over the period (−2, +5) irrespective of the abnormal returns model employed. In contrast, for Schedule 13D events, there is limited evidence of abnormal returns at the 10% level at best, and under only the GARCH, EGARCH and GJR-GARCH models. For the non-voice events, there is even scarcer evidence of abnormal returns only under the GARCH model and at the 10% level. These results signal that the voice events carry richer news compared to the other two types of events and yield richer information revelation to which stock market participants are more highly responsive.

Table 10. Robustness—Corrado rank test results of the Schedule 13D abnormal returns.

Models	Market Model	Carhart Model	GARCH	EGARCH	GJR-GARCH
Event Day
−10	0.205	0.370	0.162	0.221	0.225
−9	0.192	0.375	0.160	0.219	0.198
−8	0.107	0.212	0.076 *	0.099 *	0.096 *
−7	0.121	0.202	0.060 *	0.093 *	0.086 *
−6	0.127	0.208	0.070 *	0.114	0.096 *
−5	0.169	0.251	0.092 *	0.126	0.112
−4	0.125	0.184	0.084 *	0.108	0.099 *
−3	0.125	0.169	0.066 *	0.082 *	0.087 *
−2	0.114	0.145	0.060 *	0.061 *	0.076 *
−1	0.124	0.175	0.072 *	0.071 *	0.082 *
0	0.157	0.195	0.086 *	0.091 *	0.100
1	0.140	0.221	0.093 *	0.074 *	0.098 *
2	0.184	0.280	0.117	0.110	0.123
3	0.231	0.323	0.146	0.142	0.162
4	0.274	0.362	0.168	0.167	0.185
5	0.270	0.334	0.160	0.152	0.189
6	0.254	0.327	0.171	0.145	0.189
7	0.245	0.289	0.163	0.124	0.181
8	0.256	0.317	0.182	0.123	0.205
9	0.314	0.382	0.229	0.165	0.256
10	0.349	0.387	0.263	0.183	0.260

Notes: This table reports the p-values of the Corrado rank test on the statistical significance of the abnormal returns. The abnormal returns of each stock are estimated using an estimation window of 212 observations (Schedule 13D and non-voice) and 198 observations (voice). * indicates statistical significance at the 10% level.

Table 11. Robustness—Corrado rank test results of the voice abnormal returns.

Models	Market Model	Carhart Model	GARCH	EGARCH	GJR-GARCH
Event Day
−10	0.246	0.613	0.688	0.587	0.570
−9	0.190	0.489	0.631	0.517	0.411
−8	0.074 *	0.288	0.302	0.230	0.142
−7	0.036 **	0.222	0.218	0.172	0.137
−6	0.041 **	0.193	0.113	0.131	0.129
−5	0.049 **	0.196	0.120	0.148	0.141
−4	0.044 **	0.159	0.094 *	0.111	0.085 *
−3	0.040 **	0.171	0.135	0.143	0.114
−2	0.005 ***	0.014 **	0.015 **	0.012 **	0.013 **
−1	0.004 ***	0.013 **	0.015 **	0.021 **	0.015 **
0	0.007 ***	0.015 **	0.016 **	0.020 **	0.016 **
1	0.011 **	0.033 **	0.017 **	0.022 **	0.024 **
2	0.024 **	0.030 **	0.024 **	0.031 **	0.026 **
3	0.021 **	0.029 **	0.029 **	0.032 **	0.017 **
4	0.033 **	0.044 **	0.041 **	0.038 **	0.032 **
5	0.027 **	0.032 **	0.026 **	0.036 **	0.031 **
6	0.033 **	0.044 **	0.037 **	0.053 *	0.044 **
7	0.054 *	0.065 *	0.040 **	0.055 *	0.068 *
8	0.037 **	0.050 *	0.043 **	0.061 *	0.065 *
9	0.054 *	0.070 *	0.062 *	0.061 *	0.066 *
10	0.043 **	0.074 *	0.056 *	0.069 *	0.062 *

Notes: This table reports the p-values of the Corrado rank test on the statistical significance of the abnormal returns. The abnormal returns of each stock are estimated using an estimation window of 212 observations (Schedule 13D and non-voice) and 198 observations (voice). *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

Table 12. Robustness—Corrado rank test results of the non-voice abnormal returns.

Models	Market Model	Carhart Model	GARCH	EGARCH	GJR-GARCH
Event Day
−10	0.304	0.346	0.110	0.184	0.184
−9	0.288	0.322	0.094 *	0.172	0.140
−8	0.245	0.277	0.075 *	0.133	0.111
−7	0.286	0.285	0.070 *	0.144	0.111
−6	0.310	0.307	0.089 *	0.180	0.132
−5	0.364	0.350	0.123	0.201	0.162
−4	0.328	0.323	0.133	0.201	0.179
−3	0.317	0.281	0.120	0.170	0.159
−2	0.364	0.306	0.149	0.186	0.188
−1	0.408	0.357	0.176	0.224	0.216
0	0.463	0.397	0.208	0.264	0.241
1	0.408	0.392	0.195	0.218	0.220
2	0.465	0.450	0.220	0.262	0.251
3	0.518	0.498	0.257	0.303	0.288
4	0.574	0.560	0.307	0.352	0.350
5	0.585	0.553	0.306	0.342	0.349
6	0.548	0.531	0.293	0.323	0.339
7	0.500	0.441	0.265	0.263	0.296
8	0.538	0.472	0.281	0.261	0.339
9	0.605	0.532	0.335	0.313	0.398
10	0.658	0.590	0.393	0.373	0.443

Notes: This table reports the p-values of the Corrado rank test on the statistical significance of the abnormal returns. The abnormal returns of each stock are estimated using an estimation window of 212 observations (Schedule 13D and non-voice) and 198 observations (voice), respectively. * indicate statistical significance at the 10% level.

5.3. Testing Abnormal Performance Using Conditional Factor Models

The abnormal returns of the voice and the Schedule 13D events are calculated using several additional conditional factor models. First, we used the hedge fund risk factor model proposed by Fung and Hsieh (2004) to calculate the abnormal returns. Their pricing model, which is essentially an APT-like conditional model for evaluation of the hedge fund performance (hereafter denoted as the Fung Hsieh model), is defined as

$$R_{it}=a_0+{\beta }_1{F}_{1t}+{\beta }_2{F}_{2t}+{\beta }_3{F}_{3t}+{\beta }_4{F}_{4t}+{\beta }_5{F}_{5t}+{\beta }_6{F}_{6t}+{\beta }_7{F}_{7t}+u_{it},$$

(26)

where $F_{1t}$ denotes a bond trend-following factor, $F_{2t}$ denotes a currency trend-following factor, $F_{3t}$ denotes a commodity trend-following factor, $F_{4t}$ denotes the S&P 500 market index returns, $F_{5t}$ denotes the size spread factor calculated as the difference between the Russell 2000 market index returns and the S&P 500 market index returns, $F_{6t}$ denotes the monthly change in the 10-year U.S. treasury constant maturity yield, and $F_{7t}$ denotes the credit spread factor calculated as the difference between the Moody’s Baa yield and the 10-year U.S. treasury constant maturity yield. The trend-following factors are extracted common portfolio return components based on trend-following funds12.

The basic problem in the application of the above pricing model for the calculation of the abnormal returns is that data on the three trend-following factors are not available at the daily frequency, while our analysis focuses on the abnormal performance of daily stock returns. Therefore, several modifications had to be performed on the previous pricing equation to overcome this limitation. In particular, two alternative futures indices are used as substitutes for the bond and commodity trend-following factors13.The trend-following factors measure straddle returns, which in turn are priced on the volatility of the underlying asset. Therefore, daily market prices on volatility for commodity and bond futures could serve as instruments for straddle prices on both assets. The commodity trend-following factor is substituted by the logarithmic first differences of the S&P dynamic commodity futures index. We document that the correlation between the monthly observations of the commodity trend-following factor and the returns of the S&P commodity futures index is about 0.20. The bond trend-following factor is substituted by the logarithmic first differences of the S&P treasury note futures 30-year index. We find that the correlation between these two is small (approximately 0.06).

Table 13 presents the abnormal performance of the targeted stocks 10 days prior, during and after the Schedule 13D and the voice dates based on the conditional model of Fung and Hsieh (2004). We also report the test results on the statistical significance of the difference between the voice and the Schedule 13D filing standardized abnormal returns. We find that the abnormal returns of the Schedule 13D filing dates are statistically significant at the 5% level 1, 5, 6 and8 days prior to the event date, as well as during the event day. After the Schedule 13D filing event, we document that on average, the abnormal returns are significant for 1 day at the 5% level. As far as it concerns the voice dates, the abnormal returns are statistically significant at the 5% level,1day prior to the voice date, during the voice date, and1 and 3 days after the voice date. The test results indicate that the standardized abnormal returns of the targeted stocks on the Schedule 13D filing dates are larger than those at the voice events 9 and 14 days before the event date. On the other hand, the voice-based standardized average stock returns outperform the Schedule 13D filing-based standardized average stock returns 1,3 and 10 days after the event date at the 5% level of statistical significance. Compared to our previous findings, we now document that voice induces a more severe effect on the stock performance of the targeted companies mainly after the event date. Although we find a larger number of negative abnormal returns at both dates, these are mostly statistically indistinguishable from zero.

Table 13. Hedge fund risk factor model-based abnormal returns of Schedule 13D filings and voice dates.

		Schedule 13D Filings			Voice		Testing for the Difference Between the Standardized Means of Voice and Schedule 13D Filings
Event Day	CAR	AR	p-Val	CAR	AR	p-Val	p-Val	p-Val Right	p-Val Left
−10	0.003	0.003	0.990	−0.012	−0.012	0.950	0.952	0.524	0.476
−9	0.294	0.291	0.191	−0.457	−0.445	0.011 **	0.004 ***	0.998	0.002 ***
−8	0.737	0.443	0.023 **	−0.458	−0.001	0.996	0.238	0.881	0.119
−7	1.078	0.341	0.114	−0.375	0.083	0.673	0.646	0.677	0.323
−6	1.403	0.325	0.020 **	−0.173	0.202	0.371	0.659	0.670	0.330
−5	1.743	0.340	0.005 ***	0.126	0.299	0.204	0.716	0.642	0.358
−4	1.639	−0.104	0.326	−0.165	−0.291	0.152	0.472	0.764	0.236
−3	1.688	0.049	0.772	−0.223	−0.058	0.760	0.680	0.660	0.340
−2	1.705	0.017	0.907	−0.213	0.010	0.970	0.978	0.511	0.489
−1	1.984	0.279	0.022 **	0.316	0.529	0.014 **	0.354	0.177	0.823
0	2.642	0.658	0.000 ***	1.433	1.117	0.000 ***	0.326	0.163	0.837
1	3.126	0.484	0.001 ***	3.115	1.682	0.000 ***	0.005 ***	0.003 ***	0.997
2	3.272	0.146	0.191	3.477	0.362	0.103	0.470	0.235	0.765
3	3.238	−0.034	0.745	4.229	0.752	0.001 ***	0.003 ***	0.001 ***	0.999
4	3.265	0.027	0.817	4.572	0.343	0.059 *	0.133	0.067 *	0.933
5	3.274	0.009	0.938	4.729	0.157	0.485	0.576	0.288	0.712
6	3.312	0.038	0.761	5.002	0.273	0.124	0.245	0.123	0.877
7	3.176	−0.136	0.251	5.195	0.193	0.360	0.168	0.084 *	0.916
8	3.077	−0.099	0.356	5.384	0.189	0.272	0.156	0.078 *	0.922
9	3.159	0.082	0.448	5.616	0.232	0.188	0.461	0.231	0.769
10	3.029	−0.130	0.197	5.998	0.382	0.090 *	0.034 **	0.017 **	0.983

Notes: This table reports the abnormal stock returns (and the corresponding cumulative abnormal returns) 10 days before and after the Schedule 13D filing and the voice date, alongside statistical tests on the difference in their average performance. The abnormal returns are computed by using the hedge fund risk factor model proposed by Fung and Hsieh (2004). We report the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is different than the average of standardized stock returns on Schedule 13D filing dates. “p-val right” denotes the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is larger than the average of standardized stock returns on Schedule 13D filing dates. “p-val left” denotes the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is smaller than the average of standardized stock returns on Schedule 13D filing dates. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

Our event study analysis is also conducted based on a conditional asset pricing approach that allows for time variation in the beta coefficient in line with Chen and Knez (1996). Chen and Knez (1996) developed a general framework for evaluating the performance of a managed portfolio by specifying a minimum set of conditions that any performance measure must satisfy. This approach modifies existing factor models, such as the CAPM model, by incorporating publicly available information to capture time-varying expectations of investors. In line with Chen and Knez (1996), Ferson and Schadt (1996) developed the following dynamic CAPM:

$$R_{it}=a_0+{\beta }_1\left(R_{mt}-R_{ft}\right)+{\mathit{\Gamma }}_i^{\prime }{Z}_{t-1}\left(R_{mt}-R_{ft}\right)+u_{it},$$

(27)

$$R_{it}=a_0+{\beta }_1\left(R_{mt}-R_{ft}\right)+{\mathit{\Gamma }}_i^{\prime }{Z}_{t-1}\left(R_{mt}-R_{ft}\right)+{\delta }_1{\left(R_{mt}-R_{ft}\right)}^2+u_{it},$$

(28)

where $Z_t$ is a vector of instruments, while the coefficient ${\delta }_1$ represents the market timing ability of the mutual fund manager. The dynamic MH model is specified as

$$R_{it}=a_0+{\beta }_1\left(R_{mt}-R_{ft}\right)+{\mathit{\Gamma }}_i^{\prime }{Z}_{t-1}\left(R_{mt}-R_{ft}\right)+{\delta }_1{R}_{mt}^{*}+{\mathit{\Lambda }}_i^{\prime }{Z}_{t-1}{R}_{mt}^{*}+u_{it},$$

(29)

where $R_{mt}^{*}=\left(R_{mt}-R_{ft}\right)I\left\{\left(R_{mt}-R_{ft}\right)-E\left(R_{mt}-R_{ft}|Z_{t-1}\right)>0\right\}$, with $I\left\{.\right\}$ representing the indicator function assigning the value one when the condition inside the bracket is fulfilled and zero otherwise, while $E\left(R_{mt}-R_{ft}|Z_{t-1}\right)$ is calculated as the fitted values of the dependent variable of the regression of $R_{mt}-R_{ft}$ over time on the set of instruments $Z_{t-1}$. The sum of the coefficients ${\delta }_1$ and ${\mathit{\Lambda }}_i^{\prime }$ measures the overall market timing ability of the mutual fund manager.

Table 14 presents the abnormal returns of the targeted stocks 20 days prior, during and after the Schedule 13D filings and the voice dates based on the conditional CAPM model of Ferson and Schadt (1996), along with the comparison of their standardized average abnormal performance. We observe that the abnormal stock returns of the targeted companies are positive and statistically significant at the 5% and the 10% level 2, 3, 7, 9 and 10 days prior to the Schedule 13D filing date, as well as 1, 3, 4, 6, 8 and 10 days after the Schedule 13D filing date. The patterns of the abnormal behavior of the targeted companies on voice dates are slightly different than those on the Schedule 13D filing dates. We document negative and statistically significant abnormal returns at the 10% level 8 days prior to voice, along with positive and statistically significant abnormal returns at the 5% and 10% levels 2 days prior to voice. Again, our findings indicate that after voice, abnormal returns are statistically significant at the 5% level and positive (1 day after), and statistically significant at the 5% level and negative (5 days after). The standardized average abnormal returns at voice outperform the returns on Schedule 13D filing dates 1 and 2 days after the event at the 5% and the 10% level of statistical significance. On the other hand, the Schedule 13D filing-based standardized average abnormal returns are larger than the voice-based standardized average abnormal returns 7 days before the event and 5days after the event at the 5% and the 10% level of statistical significance. These results confirm our previous findings that Schedule 13D filing dates and voice induce different effects on the stock performance of the targeted companies.

Table 14. Abnormal returns of Schedule 13D filings and voice dates based on the dynamic CAPM-FS.

		Schedule 13D Filings			Voice		Testing for the Difference Between the Standardized Means of Voice and Schedule 13D Filings
Event Day	CAR	AR	p-Val	CAR	AR	p-Val	p-Val	p-Val Right	p-Val Left
−10	18.969	18.969	0.061 *	2.718	2.718	0.299	0.935	0.532	0.468
−9	43.542	24.573	0.065 *	33.891	31.173	0.307	0.936	0.532	0.468
−8	40.385	−3.157	0.374	19.097	−14.794	0.075 *	0.286	0.857	0.143
−7	49.254	8.869	0.040 **	18.603	−0.494	0.269	0.045 **	0.978	0.022 **
−6	53.525	4.271	0.405	21.739	3.136	0.292	0.637	0.318	0.682
−5	54.697	11.718	0.119	23.993	2.254	0.672	0.655	0.673	0.327
−4	57.522	2.825	0.586	39.666	15.673	0.227	0.450	0.225	0.775
−3	61.264	3.742	0.094 *	42.603	2.937	0.270	0.940	0.470	0.530
−2	72.079	10.815	0.054 *	66.107	23.504	0.086 *	0.634	0.317	0.683
−1	74.878	2.799	0.630	70.267	4.160	0.183	0.371	0.186	0.814
0	76.307	1.429	0.819	69.602	−0.665	0.595	0.566	0.717	0.283
1	95.350	19.043	0.040 **	71.104	1.502	0.000 ***	0.036 **	0.018 **	0.982
2	92.228	−3.122	0.217	75.864	4.760	0.325	0.138	0.069 *	0.931
3	107.699	15.471	0.081 *	76.039	0.175	0.744	0.532	0.734	0.266
4	113.929	6.230	0.081 *	89.484	13.445	0.130	0.693	0.347	0.653
5	115.130	1.201	0.866	64.009	−25.475	0.045 **	0.071 *	0.965	0.035 **
6	125.431	10.301	0.093 *	71.685	7.676	0.325	0.978	0.511	0.489
7	125.956	0.525	0.820	72.006	0.321	0.273	0.412	0.206	0.794
8	145.927	19.971	0.032 **	82.912	10.906	0.322	0.790	0.605	0.395
9	146.662	0.735	0.516	83.041	0.129	0.721	0.976	0.512	0.488
10	163.578	16.916	0.035 **	97.690	14.649	0.318	0.811	0.594	0.406

Notes: This table reports the abnormal stock returns (and the corresponding cumulative abnormal returns) 10 days before and after the Schedule 13D filing and the voice date, along with statistical tests on the difference in their average performance. The abnormal returns are computed by using the dynamic CAPM model proposed by Ferson and Schadt (1996). We report the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is different than the average of standardized stock returns on Schedule 13D filing dates. “p-val right” denotes the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is larger than the average of standardized stock returns on Schedule 13D filing dates. “p-val left” denotes the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is smaller than the average of standardized stock returns on Schedule 13D filing dates. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

Table 15 reports the results when the dynamic Treynor–Mazuy model is used to calculate the average abnormal performance. The original Treynor and Mazuy (1966) model is a conditional market-timing model for measuring portfolio performance. In comparison with the results of the conditional CAPM model, we find fewer statistically significant abnormal returns on both Schedule 13D filing and voice samples. We find both positive and negative average abnormal returns of the targeted companies that are statistically significant at the 5% and 10% levels several days before (4, 5, 7 days) and after (5,8 days) the Schedule 13D filing dates. Observing the results of the voiced-based sample, we document that the average abnormal returns of the targeted companies are positive and significant at the 5% and 10% levels 1 day prior to the event date and on the event day, while they are mostly positive and statistically significant 1, 2, 3 and 10 days after the event at the 5% and 10% levels. Significant differences between the standardized voice and the Schedule 13D filing returns are found mainly over the period (−1, +3) at the 5% or 1% level. The results of “p-val right” indicate that the voice average abnormal returns are higher than the Schedule 13D filing average abnormal returns at the 1%, 5%, and 10% levels 1, 2 and 6 days prior to the event, during the event day, and 1, 2, 3, 7 and10 days after the event. Overall, these findings confirm our evidence that the announcement of voice reveals more influential information than Schedule 13D filing announcements.

Table 15. Abnormal returns of Schedule 13D filings and voice dates based on the dynamic Treynor–Mazuy model.

		Schedule 13D Filings			Voice		Testing for the Difference Between the Standardized Means of Voice and Schedule 13D Filings
Event Day	CAR	AR	p-Val	CAR	AR	p-Val	p-Val	p-Val Right	p-Val Left
−10	2.442	2.442	0.144	0.019	0.019	0.928	0.495	0.752	0.248
−9	12.649	10.207	0.139	−0.168	−0.187	0.627	0.236	0.882	0.118
−8	−1.534	−14.183	0.235	−0.501	−0.333	0.146	0.528	0.736	0.264
−7	0.222	1.756	0.085 *	−0.476	0.025	0.899	0.433	0.783	0.217
−6	−1.542	−1.764	0.187	−0.116	0.360	0.143	0.052 *	0.026 **	0.974
−5	−1.176	0.366	0.020 **	0.057	0.173	0.474	0.551	0.724	0.276
−4	−5.423	−4.247	0.075 *	−0.081	−0.138	0.495	0.735	0.367	0.633
−3	−4.928	0.495	0.253	−0.043	0.038	0.875	0.647	0.677	0.323
−2	−7.638	−2.710	0.747	7.016	7.059	0.128	0.141	0.070 *	0.930
−1	−8.327	−0.689	0.896	7.749	0.733	0.014 **	0.029 **	0.014 **	0.986
0	−23.677	−15.350	0.251	8.490	0.741	0.024 **	0.011 **	0.006 ***	0.994
1	−13.877	9.800	0.114	9.957	1.467	0.000 ***	0.000 ***	0.000 ***	1.000
2	−25.816	−11.939	0.189	10.362	0.405	0.090 *	0.033 **	0.016 **	0.984
3	−24.484	1.332	0.739	10.857	0.495	0.032 **	0.096 *	0.048 **	0.952
4	−23.076	1.408	0.374	19.218	8.361	0.157	0.452	0.226	0.774
5	−45.235	−22.159	0.075 *	11.911	−7.307	0.277	0.995	0.502	0.498
6	−47.939	−2.704	0.454	12.142	0.231	0.198	0.136	0.068	0.932
7	−56.345	−8.406	0.228	12.371	0.229	0.257	0.110	0.055 *	0.945
8	−50.648	5.697	0.095 *	12.418	0.047	0.783	0.528	0.736	0.264
9	−52.259	−1.611	0.370	12.561	0.143	0.440	0.260	0.130	0.870
10	−45.939	6.320	0.194	13.319	0.758	0.020 **	0.183	0.092 *	0.908

Notes: This table reports the abnormal stock returns (and the corresponding cumulative abnormal returns) 10 days before and after the Schedule 13D filing and the voice date, along with statistical tests on the difference in their average performance. The abnormal returns are computed by using the dynamic Treynor–Mazuy model proposed by Ferson and Schadt (1996). We report the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is different than the average of standardized stock returns on Schedule 13D filing dates. “p-val right” denotes the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is larger than the average of standardized stock returns on Schedule 13D filing dates. “p-val left” denotes the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is smaller than the average of standardized stock returns on Schedule 13D filing dates. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

Table 16 demonstrates the results when the dynamic Merton–Henriksson model of Ferson and Schadt (1996) is implemented to calculate the abnormal stock returns. The findings are very similar to those of the previous table. Our evidence suggests that voice entails information revelation that is reflected in the stock market by mainly a positive response. Moreover, we document that voice and Schedule 13D filing dates are two chronologically different event dates of hedge fund activism, which create stock market inefficiencies. The information revealed in voice induces a stronger impact on the stock performance of the targeted companies since the abnormal returns of voice are significantly higher than the returns of Schedule 13Ds many days prior to and after the event date. These results confirm our main finding that the information disclosed at the Schedule 13D filing reveals only partial information, with the remainder of the information being revealed at voice14.

Table 16. Abnormal returns of Schedule 13D filings and voice dates based on the dynamic Merton–Henriksson model.

		Schedule 13D Filings			Voice		Testing for the Difference Between the Standardized Means of Voice and Schedule 13D Filings
Event Day	CAR	AR	p-Val	CAR	AR	p-Val	p-Val	p-Val Right	p-Val Left
−10	1.918	1.918	0.206	107.186	107.186	0.320	0.844	0.422	0.578
−9	38.469	36.551	0.084 *	126.251	19.065	0.220	0.878	0.439	0.561
−8	36.218	−2.251	0.089 *	111.882	−14.369	0.149	0.722	0.639	0.361
−7	36.655	0.437	0.080 *	111.708	−0.174	0.666	0.201	0.900	0.100
−6	33.256	−3.399	0.165	112.076	0.368	0.298	0.107	0.054 *	0.946
−5	33.482	0.226	0.165	111.928	−0.148	0.743	0.317	0.841	0.159
−4	32.619	−0.863	0.569	111.129	−0.799	0.113	0.286	0.857	0.143
−3	32.749	0.130	0.676	111.125	−0.004	0.992	0.821	0.589	0.411
−2	43.704	10.955	0.177	166.505	55.380	0.156	0.604	0.302	0.698
−1	50.848	7.144	0.446	166.313	−0.192	0.763	0.514	0.743	0.257
0	46.289	−4.559	0.244	167.386	1.073	0.011 **	0.006 ***	0.003 ***	0.997
1	47.248	0.959	0.767	168.682	1.296	0.000 ***	0.003 ***	0.002 ***	0.998
2	44.890	−2.358	0.172	169.008	0.326	0.232	0.083 *	0.042 **	0.958
3	54.615	9.725	0.293	169.143	0.135	0.761	0.776	0.612	0.388
4	53.545	−1.070	0.192	318.151	149.008	0.297	0.117	0.058 *	0.942
5	41.661	−11.884	0.073 *	287.347	−30.804	0.083 *	0.577	0.711	0.289
6	43.155	1.494	0.228	287.078	−0.269	0.570	0.266	0.867	0.133
7	40.958	−2.197	0.124	287.408	0.330	0.225	0.066	0.033	0.967
8	53.178	12.220	0.159	287.629	0.221	0.541	0.836	0.582	0.418
9	52.122	−1.056	0.030 **	287.813	0.184	0.447	0.076 *	0.038 **	0.962
10	58.239	6.117	0.295	288.135	0.322	0.537	0.989	0.505	0.495

Notes: This table reports the abnormal stock returns (and the corresponding cumulative abnormal returns) 10 days before and after the Schedule 13D filing and the voice date, along with statistical tests on the difference in their average performance. The abnormal returns are computed by using the dynamic Merton–Henriksson model proposed by Ferson and Schadt (1996). We report the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is different than the average of standardized stock returns on Schedule 13D filing dates. “p-val right” denotes the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is larger than the average of standardized stock returns on Schedule 13D filing dates. “p-val left” denotes the p-value of the t-statistic when the alternative hypothesis is that the average of standardized stock returns on the voice date is smaller than the average of standardized stock returns on Schedule 13D filing dates. *, ** and *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.

6. Conclusions

This paper emphasizes the role of voice in hedge fund activism. We reveal that the voice date is important in creating short-term firm value and provide strong evidence that voice in hedge fund activism is associated with positive abnormal returns. The abnormal returns around the voice date are approximately 1.11% and are higher than the abnormal returns around the Schedule13D date by approximately 33%. In addition, positive abnormal returns due to voice are found even when voice leads the Schedule 13D event. Therefore, voice on its own creates short-term value, and not necessarily as a result of the Schedule 13D filing. Markets respond to hedge fund voice, regardless of whether this occurs prior to or after the Schedule 13D, which signals significant information revelation. The results are robust to alternative models of abnormal returns and inferential statistical procedures.

These findings are interpreted as evidence that the U.S. stock market response to Schedule 13D events is smaller than that to voice events. Furthermore, the disclosures on the voice date and Schedule 13D date create information revelation and may form a mechanism for arbitrage.