Contemporaneous Event Studies in Corporate Finance: Methods, Critiques and Robust Alternative Approaches 1st Ed. Edition Jau-Lian Jeng
Contemporaneous Event Studies in Corporate Finance: Methods, Critiques and Robust Alternative Approaches 1st Ed. Edition Jau-Lian Jeng
Contemporaneous Event Studies in Corporate Finance: Methods, Critiques and Robust Alternative Approaches 1st Ed. Edition Jau-Lian Jeng
https://ebookmass.com/product/contemporaneous-
event-studies-in-corporate-finance-methods-
critiques-and-robust-alternative-approaches-1st-
ed-edition-jau-lian-jeng/
OR CLICK BUTTON
DOWLOAD NOW
https://ebookmass.com/product/alternative-approaches-in-conflict-
resolution-1st-edition-martin-leiner/
https://ebookmass.com/product/analysing-planning-and-valuing-
private-firms-new-approaches-to-corporate-finance-federico-
beltrame/
https://ebookmass.com/product/numerical-methods-and-optimization-
in-finance-2nd-ed-manfred-gilli/
https://ebookmass.com/product/corporate-governance-of-chinese-
multinational-corporations-case-studies-1st-ed-edition-runhui-
lin/
eTextbook 978-0133507676 Fundamentals of Corporate
Finance (Pearson Series in Finance)
https://ebookmass.com/product/etextbook-978-0133507676-
fundamentals-of-corporate-finance-pearson-series-in-finance/
https://ebookmass.com/product/corporate-finance-european-edition-
david-hillier/
https://ebookmass.com/product/principles-of-corporate-finance-
mcgraw-hill-irwin-series-in-finance-insurance-and-real-
estate-12th-edition-brealey/
https://ebookmass.com/product/principles-of-corporate-finance-
brealey/
https://ebookmass.com/product/fundamentals-of-corporate-finance-
berk-demarzo-harford-the-corporate-finance-series-4th-edition-
ebook-pdf-version/
Jau-Lian Jeng
Contemporaneous
Event Studies in
Corporate Finance
Methods, Critiques
and Robust Alternative
Approaches
Contemporaneous Event Studies in Corporate
Finance
Jau-Lian Jeng
Contemporaneous
Event Studies
in Corporate Finance
Methods, Critiques and Robust
Alternative Approaches
Jau-Lian Jeng
School of Business and Management
Azusa Pacific University
Los Angeles, CA, USA
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2020
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher,
whether the whole or part of the material is concerned, specifically the rights of translation, reprinting,
reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical
way, and transmission or information storage and retrieval, electronic adaptation, computer software,
or by similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with
regard to jurisdictional claims in published maps and institutional affiliations.
This Palgrave Macmillan imprint is published by the registered company Springer Nature
Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
v
vi Preface
For the technicality applied, readers are suggested to have some read-
ings in mathematical statistics, econometrics, and/or probability theory.
Understanding of diffusion processes, asymptotic theory of mathematical
statistics are certainly of advantage. Although the contents of the book
are mainly for the applications of event study methodology in social sci-
ence or business management, the technical elaborations with mathe-
matical reasoning in stochastic processes and econometrics are essential
to provide rigors in various arguments.
Acknowledgments The writing of this book will not be feasible without the
assistance of the editorial board of Palgrave Macmillan publishing, the fruitful
reviews of Dr. Hou and Dr. Phillips, and the spiritual supports of my family.
Appreciation
Especially, this book is done during the period of stroke where my wife’s
hearted support is cordially appreciated. In particular, during the edit-
ing period, my mom passed away due to a stage 4 lung cancer and my
dad passed away later on in which the Coronavirus consumes the glob-
al economy entirely. Yet, the project continued. Many thanks are given
to the one who said “I am!” in granting me the strength and the heal-
ing to carry on. Hopefully, this work will commemorate my parents for
their hardship in supporting my education and provide the foundation
for anyone who’s interested in the event studies in corporate finance or
else and will benefit from the discussions.
vii
Introduction
ix
x Introduction
findings) for the events that may interest the readers or, simply to match
the trend of existing literature.
For instance, the contents of this book discuss the possible similarities
between the applications of cumulative sum (CUSUM) statistics with
structural change tests, and the applications of cumulative abnormal re-
turns (CAR’s) in event studies of some corporate finance issues. In partic-
ular, it is easy to verify that the applications of CAR’s are sensitive to the
presumption of pre- and post-event windows where manipulation can be
easily introduced if some intended hypotheses are of interest.
Namely, if the pre-event window is set to accommodate the intent
to prove the significance of events of interest, and if there is a possible
significant parameter change in the systematic component of asset re-
turns, there is a high likelihood that the resulting CAR’s may show that
their (cross-sectional) mean (within the event window) will be statistical-
ly different from zero. In other words, the empirical statistical results may
show that there’s a significant impact from the events of interest. In fact,
it is a parameter change in the systematic components of asset returns
that provides the cause, and the result is not related with the corporate
events of concern.
More specifically, it is easy to see the usual CUSUM tests and the
cumulative abnormal returns (CAR’s) can be related to each other and
that, the results may lead to incorrect verification for the hypotheses of
interest. For instance, let the specification of normal rate of return be
given as the following simplified univariate model such that
rit = αi + βi xt + εit ,
where i = 1,2, …,n is the index for firm’s rates of return, t = 1, 2,…, T
is the time index over the entire time horizon of interest, x t is the com-
mon factor (such as market index return) for the return series. Suppose
there is an unknown structural change over the time period of interest,
and there is an interest of study to verify if there is a significant corpo-
rate event during the time periods by using the conventional cumulative
abnormal returns according to MacKinlay (1997).
xii Introduction
t
∼
t
∗
C A Ri = εis = t − T δi + εis ,
s=0 s=0
where ts=0 εis is the genuine cumulative abnormal returns. It is easy
to see that if the cross-sectional average of these cumulative abnormal
returns are used to form the statistics, the cross-sectional average of these
CAR’s may reject the null hypothesis such that
1 ∼ 1
n t n n t
AC A R = εis = t − T ∗ δi + εis .
n n
i=1 s=0 i=1 i=1 s=0
The statistic ACAR may become significantly different from zero even
n t
though n1 i=1 s=0 εis may actually converge to zero. In other words,
if there exists an unknown structural change of the parameter(s) in the
fitted model of normal returns, it is likely that the conventional CAR’s
(and hence, the ACAR’s across all sampled firms) may become significant
statistically when the event period is extensive enough to encompass the
time of possible structural change in parameter(s). The statistics may even
become more significant when the excedance of time (after structural
change) t − T ∗ expands.
Alternatively, following the same equation as above, if there is a sig-
nificant impact from the corporate events of some firms, this gives the
condition that for some significant number of firms, the ts=0 εis may
not be of zero mean. Using the conventional CUSUM (cumulative sum
of residuals) from the above regressions (even with recursive estimation)
Introduction xiii
may exceed certain pre-specified boundaries and reject the null hypoth-
esis as there is no structural change in the parameters α i or β i . In other
words, the applications of CUSUM statistics may wrongfully conclude
that there is a significant structural change in parameters of regressions
during the sampled period of time, while the so-called structural change
is actually the consequence of the idiosyncratic corporate events.
Particularly, the usual structural change tests or the so-called monitor-
ing tests rely on whether the test statistics exceed the boundaries (pre-
specified according to the assumption of convergence in distribution to
a certain diffusion process) or not. The approach ignores two issues: epi-
demic change and dependent sequential test statistics due to dependence
in sequential multi-hypotheses.
For instance, when there is an epidemic change, the parameters may
change for a tentative short period and end with no change afterward.
The usual structural change test may simply reject the null hypothesis
of no change when the statistics cross the boundaries when observations
within the epidemic-change time period are applied. In other words, the
tests usually assume permanent change on the parameters. This result,
in turns, is not often observed in corporate finance events. On the oth-
er hand, the tests statistics applied in the sequential monitoring test are
usually sequentially dependent where the control of type-1 errors in test
should take this possibility into account.
Furthermore, there are research articles that proclaim the usage of
firm-specific information may help explaining the forecasts of asset re-
turns. It is not surprising to state that, although these applications of
firm-specific variables assist contemporaneous explanation on asset re-
turns, the contribution is usually short-lived and lack of explanation
from the theoretical standpoints. However, applications of these addi-
tional firm-specific variables (for conditional expectations of asset re-
turns) may actually dilute the specification of abnormal returns which
are the essential variables for the event studies.
More specifically, it is dubious to accept those claims since event stud-
ies should focus on the possible impacts on the so-called abnormal re-
turns, where abnormal returns should be the additional returns in ex-
cess of the expected returns merely based on systematic (nondiversifiable)
economic variables. In other words, the prerequisite for the robust event
xiv Introduction
N (t) = Nt = i∈N ∗ δ(ti <t) as a counting process with intensity as
N (t + t)
λ (t|Ht ) = lim t→0 E |Ht .
t
References
xvii
xviii Contents
Index 225
List of Tables
xix
Part I
The Conventional Approach
1
Popular Methods for Event Studies
in Corporate Finance—Subjectivity Versus
Robustness
1.1 Introduction
This chapter is to survey the conventional methods in event studies of
corporate events. Intuitively, since the corporate (finance) events are con-
sidered as the newly available information concerning the specific insti-
tutions of interest, the information regarding the institutions alone is
considered as especially firm-specific. Therefore, the methods usually con-
sider the approximation of normal returns in using either the asset pricing
models or time series modeling to filter them. The residuals from these
methods are then, considered the abnormal returns since they are the devi-
ations of stock returns from the expected returns (based on the systematic
information).
In other words, the abnormal returns are considered as firm-specific
since the contents of information regarding corporate events are not sys-
tematic. Therefore, the event studies of corporate (finance) issues typically
focus on the statistics and statistical inferences of abnormal returns. This
also leads to the understanding that confusion over systematic versus firm-
specific information may misguide the event studies even when robust
statistical methods are applied. Specifically, the systematic components of
stock returns must be filtered out properly so that the event study (based
on abnormal returns) can be implemented.
Therefore, the dichotomy of normal returns versus abnormal returns
with correct separation is essential for the robustness and justifiability
of the empirical results in event studies. It is easy to see that possible
manipulations of the empirical results are feasible if one is simply eager to
prove the desired hypotheses of interest by considering various alternative
methods to obtain the abnormal returns, the event windows, and the
estimates of normal returns in order to confirm the ideal results.
To pursue the event studies (of corporate finance), the robustness of the
estimates for normal returns, the acquisitions of abnormal returns, and
the statistics that may be robust to outliers will determine the profundity
of the analyses. In the following, Section 1.1 discusses the arbitrariness
of conventional methods in cutting the time horizon into the pre-event,
in-the-event, and post-event periods. When lack of some proper criteria in
determining the separation of these time periods, the empirical results will
become relatively similar to the attempts to forge some stories from the
mess. Section 1.2 is to consider the construction of cumulative abnormal
returns where the asymptotic distribution and properties are discussed.
Specifically, the generalization of the statistics under dependence and het-
eroscedasticity is provided.
Section 1.3 considers the similarity between the usual CUSUM tests
for structural changes in parameters (of expected rates of return) and
the test statistics based on cumulative abnormal returns. In particular, the
similarity between the two may incur incorrect verification of the empirical
results in event studies when the event period is determined arbitrarily with
any ad hoc scheme. Section 1.4 contains various econometric issues when
the conventional event study methods are applied, while the intuitive
explanations on the alternative occupation time statistics will be derived
in Chapters 3 and 4.
Similar surveys are provided in the text of corporate finance by Eckbo
(2007). However, the discussions in the said text do not cover the details
of sample selection, the specification of estimation and event periods, and
the similarity of CUSUM tests (with residuals) in structural changes in
parameters (of expected rates of return) versus the cumulative sums of
abnormal returns.
1 Popular Methods for Event Studies in Corporate Finance . . . 5
Namely, the estimates for coefficients from either time series modeling
or market model for instance, may not be informative enough to generate
the normal returns in terms of predictive regressions. In other words,
the predictive returns (based on the coefficients of fitted models from
observations of pre-event period) may not be updated enough to represent
the correctly specified normal returns. As a result, some non-related events
may be introduced in the estimation window that will also cause the bias
of estimation of normal returns. This is the so-called contamination issue
for the estimation period in the event studies.
Similar arguments can be found in McWilliams and Siegel (1997) who
argue that applications of event study method in management research
may require cautious research designs such as the length of event window,
outlines, and possible confounding effects when other events may also
happen in the event period. The empirical results of event studies can
only be justifiable if the research designs are well-stated to consider all
these issues in using the conventional approach.
On the other hand, if the cutoff date for the pre-event period (or esti-
mation period) is too close to the genuine initial date of event period
(or even, may surpass the initial date of event’s impacts), although the
estimates for the normal returns are more updated, the event period will
be shortened so that the test statistics on the impacts of event(s) may be
subject to limited number of observations. This instead, will cause the test
statistics for the events’ impacts to be less informative. In addition, even
if the event period is correctly identified, there are still several issues such
as the possible event dependence of structural change in security pricing
models (in forming both the normal and abnormal returns by Brown et
al. (1985), and the event-induced increase of volatility in event period of
Boehmer et al. (1991) and Brown and Warner (1985).
Another issue is that given the correct specification of event period,
the precise event date (which may be different from announcement date)
is usually unknown. Ball and Torous (1988) assume the event date as
a random variable (in the event period) and devise maximum likelihood
method to perform event studies. Their analyses consider that the random
event date may affect both the conditional mean and conditional variance
of security returns. Hence, the possible event-induced increase of variance
of security returns is also included in the analysis.
1 Popular Methods for Event Studies in Corporate Finance . . . 7
Krivin et al. (2003) study the possible ways to identify the length of
event period. Their conclusion states that there are three rules to determine
the length of event period: (1) significant abnormal returns- that is, to
identify the length of event period until the first day when the abnormal
return becomes statistical insignificant; (2) significant abnormal trading
activity—to determine the length of event period if there is persistent
abnormal trading activity after the disclosure; (3) significant abnormal
intraday volatility—to determine the length of event period based on the
persistence of significant abnormal intraday volatility.
Although these rules sound plausible, it appears that they are actually
creating the post facto methods for event period determination after learn-
ing the existence of significant events already. The purpose of event studies
is to identify if the abnormal returns are significant enough to verify the
hypotheses of interest based on the impacts of events. However, the said
article instead, is to determine the length of event period after already real-
izing the significance of the possible impacts of the events. The issue is,
if one knows the events are significant enough already, the determination
of event period is no longer a matter of concerns unless one is willing to
identify the duration of impacts from events.
Aktas et al. (2007) instead, assume the stock returns follow a state-
dependent process by combining the market model with a Markov switch-
ing regression model. They apply the simulations with contaminated
events in estimation period to verify that when there is no event-induced
increase of volatility, the rank test of Corrado (1998) performs best in
terms of power. However, when the event-induced increase of volatility is
introduced, the setting will destroy the robustness of conventional meth-
ods including the rank test. Nevertheless, these works still do not solve
the issues for optimal separation date of estimation period versus event
period, and the length of event period.
Another difficulty is that the known or presumed event date may not
be correctly identified since possible leakage of inside information may
dampen the possible impacts the events may cause. Furthermore, the gen-
uine impact date may either be earlier or later than the presumed initial
date of event period (where the event date is usually denoted as day zero in
the event period by the conventional methods). In the following sections,
several issues that may concern the decisions of event period are discussed.
8 J.-L. Jeng
Brown et al. (1985) consider the possible structural change in the fitted
models during the event period by using switching regression for the asset
return series. Given the possible impacts from the events, the uncertainty
may change the structural relationship in both systematic and unsystem-
atic risks in the regression models of asset returns. If the change is not
taken into account, the statistical results may contain upward biases on
the parameters of regression models in the event period. However, the
estimation bias can also be induced from the specification error of inap-
propriate event period.
In other words, estimation errors may invalidate the statistical inferences
if imposing universal (ad hoc) event period for many firms in the event
studies (Lee and Varela 1997). Brown et al. (1985) argue that incorrect
identification of event period may result in estimation errors for param-
eters in both systematic and unsystematic risk when the dummy variable
(which identifies the security returns are in the event period) is incorrectly
specified.
The results then, provide the rationale that the optimal or nearly correct
event intervals must be determined on a case-by-case basis. In addition, the
optimal event period can be obtained through the application of maximum
likelihood estimation for each security return individually. Although the
approach considers the possible impact from events on the fitted models,
it is still of subjectivity that the said structural change only occurs in the
event period without theoretical justification that the change only occurs
in the event period and not in the other time periods prior to or after
the event period. In other words, the setting assumes the conditions that
impacts of events are the causes of structural changes in parameters, which
could be due to other causes such as market conditions.
The setting indicates that the event period should be determined by the
data, instead of any ad hoc method. This perspective also shows that appli-
cations of data-dependent adaptive methods in identifying and estimat-
ing both the systematic and unsystematic risk for security returns are far
more robust in the verification of event studies. Let the security return be
specified as the switching regression for multi-factor pricing model such as
1 Popular Methods for Event Studies in Corporate Finance . . . 11
Ho : γio = 0,
R1 = X 1 γ1 + ε1 , t ∈
/ [t− , t+ ]
(1.2)
R2 = X 2 γ2 + ε2, t ∈ [t− , t+ ],
R1 = X 1 γ̂1 + e1
(1.3)
R2 = X 2 γ̂2 + e2 ,
1 −1 −1
(γ̂1 − γ̂2 ) σ̂12 X 1 X 1 + σ̂22 X 2 X 2 (γ̂1 − γ̂2 ), (1.7)
k
which is an F -statistic with degree of freedom (k, T1 + T2 − 2k) . In other
words, the statistical inferences for event studies are considered in terms
of the structural change of the presumed asset pricing models. However,
1 Popular Methods for Event Studies in Corporate Finance . . . 13
where Z i and Wi contain different dummy variables Dit∗ and Dit that
identify different event periods , respectively, where Dit∗ = 1 when the
observation t is in the true event period, 0 otherwise; and Dit = 1 when
the observation t is in the arbitrarily imposed event period, and 0 other-
wise. In other words, the setting is to introduce possible incorrect identi-
fication of event period. The first set of equations represents the correct
identification of event period, while the second one represents the incorrect
identification of event period.2
14 J.-L. Jeng
Given the above setting, and let Ri stand for the vector of stacked-up
observations of i-th security return it is feasible to obtain the best linear
unbiased estimators for γi∗ and γi in the system of equations such that
−1
γ̂i∗ = (X : Z i ) ∗−1 (X : Z i ) (X : Z i ) ∗−1 Ri
−1 , (1.9)
−1 −1
γ̂i = (X : Wi ) (X : Wi ) (X : Wi ) Ri
Hence, the second term in Eq. (1.10) represents the bias of estimated
coefficients in estimation period due to incorrect identification of event
period. It is feasible to see that the bias increases when the identification
error increases. Following the system of equations (1.8) and denoting
the event period as [a, b], it is shown that the log-likelihood function for
the entire sample will be
Hence, the optimal event period (a, b) can be obtained through the
maximum of log-likelihood function, which depends on the maximum
likelihood estimates of coefficients for the fitted models of assumed
explanatory variables under the null hypothesis. This shows that the deter-
mination of event period can be based on data-dependent procedures on
a case-by-case basis across all firms. This also points out that the ad hoc
universal procedure that sets an identical event period for each firm of
interest is not representative enough to verify the impacts of events. The
other thing is that the volatility only increases in the event period when
the event is acknowledged. The problem is that when the new information
is correcting the noises and rumors in the event announcement, it may
actually reduce the uncertainty in the markets. During the event period
the volatility increases when the new information kicks in. However, it
is short of the explanation why the arrival of new information makes the
volatility increases. In particular, the arrival of new information which
may not necessarily increase the volatility increases, could end up with
reduction of volatility.
Furthermore, one thing to notice is that the determination of optimal
event period can be expressed as the estimated variance ratio. In that case,
the event-induced volatility could be as a way to detect the possible event
period of interest. However, the setting of Brown et al. (1985) is based on
the normality assumption and is actually a model of structural change in
the event period for asset return series. In other words, the setting actually
16 J.-L. Jeng
combines the parameter changes in the asset returns as the impacts of the
events.
Assuming normality, Ball and Torous (1988) set the i-th security return
for each day in estimation period as rit ∼ N (μi , σi2 ) where μi and σi2
are the mean and variance, respectively. For the event period, the return-
generating process is formed with the introduction of a random event date
such that, for a symmetric event period (−c, . . . , 0, . . . , c), the actual
event date (which assumes only one single event may happen) can lie
between the date −c and c, c is an integer, where the presumed event date
is denoted as day 0.4 With an additional indicator θt , the random event
date in the event period is defined as
In addition, assuming that there’s only one event occurs in the event
period, define pt as the probability of an event that happens at day t such
that for t = −c, . . . , c
will consider the probability of events on the alternate days in the event
period (−c, . . . , c − 1). The estimation of the model is shown as a two-
stage procedure such as
Stage 1: Under the estimation period with m security returns, com-
pute the maximum-likelihood estimates as
1 1
m m
μ̂ = rt , σ̂ =
2
(rt − μ̂)2 (1.15)
m t m−1 t
Ho : A = 0, δ 2 = 1,
Ha :
Ho
Following the setting in Ball and Torous (1988), let x be the (2c +
1) × 1 vector of standardized security return for a single firm where for
each date8
ˆ t = −c, . . . , c.
xt = (rt − μ̂)/σ,
2 +c
where gt (x) = f (x|θt = 1) = √ 1 ex p(− (xt −A)
2 ) s=−c,s=t
2πδ 2 2δ
x 2
√1 ex p(− s ).
2π 2
Notice that although the system allows possible difference of condi-
tional distribution of f (x|θ ), it assumes homoscedasticity for either state
as θt = 0 or θt = 1. In addition, there is no discussion as to why the vari-
ance (even though it was increased during the event period) will remain
on the same level (for all firms of interest) within the event period such as
t = −c, . . . , c. Besides, there is no discussion that the information may
differ (in timing) for the actions taken by the institution.
1 Popular Methods for Event Studies in Corporate Finance . . . 19
Let the error terms be state dependent where St is the state variable
St = 1 for the low variance regime, and St = 2 for the high variance
regime. In other words, there are two states in the estimation period with
some unrelated events that may be considered as random disturbances
originally. Explicitly it can be denoted as
Ri = X i bi + εi1 St = 1,
(1.19)
Ri = X i bi + εi2 St = 2.
20 J.-L. Jeng
2 > σ 2 . Aktas et al. (2007) state that the setting of higher variance
where σi2 i1
regime allows the possibility of unrelated events in the sample, which leads
to the case of event-induced increase of volatility.10
The transition between these two regimes is specified as a Markov chain
of order 1, where the transition matrix can be expressed as
p11 1 − p22
P= , (1.21)
1 − p11 p22
1 − p22 1 − p11
p(St = 1) = ; p(St = 2) = .
2 − p11 − p22 2 − p11 − p22
(1.22)
Assuming strict exogeneity and transitional probabilities are given, it is
shown that the covariance matrix of the OLS estimator will become
σi,1
2
(X i X i )−1 ≤ p(St = 1)σi,1
2
(X i X i )−1 + p(St = 2)σi,2
2
(X i X i )−1 .
(1.24)
In other words, the standard error of OLS estimates is overestimated if
the return process is a two-state process. Hence, given the event-induced
volatility, the conventional event study methodology may suffer loss of
power when the estimation period is contaminated by unrelated events.
1 Popular Methods for Event Studies in Corporate Finance . . . 21
Karafiath and Spencer (1991) consider the event study for abnormal per-
formance during a multiperiod event window. They demonstrate that the
usual test statistics applied in event studies do not follow the standardized
normal and hence, the results are thus biased. Furthermore, the bias may
increase with the length of event window and lead to excessive rejection
of the null hypothesis of no events.
Let the observed returns over the estimation period of equal length T
be denoted as rit , where i = 1, 2, . . . n, t = 1, 2, . . . T. For each firm,
there is a multiperiod event window of length Ti . In other words, the event
windows for each firm may be of different lengths and different calendar
dates. Assuming the event windows occur following the estimation period,
it can be shown that
ri1 X i1 0 i ε
= + i1 , (1.25)
ri2 X i2 ITi δi εi2
Ri = X i i + εi , i = 1, 2, . . . , n, (1.26)
22 J.-L. Jeng
ri1 X i1 0 i
where Ri = , Xi = , and i = accordingly.11
ri2 X i2 ITi δi
For the regressor Rmt in market model, it is assumed that they are contem-
poraneously uncorrelated with the disturbance such that E (Rmt εit ) = 0,
for all t = 1, . . . , T + Ti , i = 1, . . . , n, which gives plim X i εi = 0.
T →∞
In addition, let E εi εi = σi2 IT +Ti , where {εit }t=1,...T +Ti are serially
uncorrelated, and contemporaneously uncorrelated across firms for all
i = 1, 2, . . . , n.12
H0 : γ0 = 0
and
u i ∼ N (0, σi2 Ii ); i = 1, 2.
Using the above equation systems, the least squares estimates for the equa-
tions can be shown as
R1 = X 1 g1 + e1 ,
R2 = X 2 g2 + e2,
e1 e1 e2 e2
+ ∼ χ(T
2
1 +T2 −2k)
(1.28)
σ12 σ22
“Poemes, Lyrick and Pastorall,” printed for the Spenser Society. 1891.
“On Suicide,” “Two Books on Truly Existing Being,” “Extracts from Treatise on
the Manner in which the Multitude of Ideas subsist, and concerning the
Good.” Translations by Thomas Taylor. London, 1834.
“Cælica,” 138.
Calidore, 46.
“Cantos of Mutabilitie,” 217.
Carew, Thomas, 158.
Cartwright, William, 162.
Charles I, 156.
Charleton, Walter, 157.
Chastity, Milton’s idea of, 47, 48, 54, 55, 56.
Christ, mystical love of, 92, 93, 94, 95, 96, 97, 99;
as true being, 98;
Milton’s idea of, 180, 181.
“Christ’s Triumph after Death,” 100–103.
Cleveland, John, 161, 162.
“Colin Clouts Come Home Againe,” 122.
“Commentarium in Convivium,” on love, 107, 108, 115, 116, 121;
on beauty, 109, 110, 112, 113, 114;
its interpretation of Plato, 140.
“Comus,” effect of sensuality on soul taught in, 49;
chastity in, 48, 49, 56;
doctrine of grace in, 63, 64;
parallelism with “Phædo,” 49, n. 1.
Comus, his attempts on The Lady, 51–54;
his character, 53.
“Corruption,” 205, 206.
“Court-Platonicke,” 159.
Cowley, Abraham, 161, 162.
Craig, Alexander, 138.
Crashaw, Richard, 97, 99, 138.
Mammon, 19.
Margaret of Valois, 156, 157.
“Masque of Beauty, The,” 122, 123.
Matter, in Plato, 211, 215;
in Plotinus, 210, 211, 215;
in Spenser, 212.
Mean, the Aristotelean doctrine of the, described, 21, 22;
in Spenser, 22, 23.
Milton, John, his notion of woman, 40, 41, 44;
his treatment of Eve, 44, 45;
his love of beauty, 41, 44, 64, 65;
his debt to Platonic philosophy, 47;
his idealism, 47, 48, 55, 57, 61;
his conception of sin, 49, 57, 58;
hold of Platonism on, 40, 47, 55, 56, 57, 61, 64, 65;
his idea of chastity, 47, 48, 54, 55;
doctrine of grace in, 63, 64;
his idea of beauty, 64;
his idea of love, 82, 83;
his idea of God, 180;
his idea of Christ, 180.
More, Henry, mysticism in, 99, 196, 199;
his idea of the Trinity, 168, 173;
his idea of soul, 187–193;
his idea of Christ, 170;
his idea of the Holy Spirit, 170–174;
religious feeling in, 182, 201;
his argument for immortality, 187–193;
hold of Platonism on, 193, 202;
on innate ideas, 195.
Mysticism, erotic, defined, 92, 93;
in George Herbert, 93;
in Donne, 94;
its connection with Platonism, 95;
relation of love of Christ to, 95–103.