Contemporaneous Event Studies in Corporate Finance: Methods, Critiques and Robust Alternative Approaches 1st Ed. Edition Jau-Lian Jeng

Download More ebooks [PDF]. Format PDF ebook download PDF KINDLE.
Full download ebooks at ebookmass.com
Contemporaneous Event Studies in

Corporate Finance: Methods, Critiques
and Robust Alternative Approaches 1st
ed. Edition Jau-Lian Jeng
For dowload this book click BUTTON or LINK below
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
Download More ebooks from https://ebookmass.com

More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Alternative Approaches in Conflict Resolution 1st

Edition Martin Leiner
https://ebookmass.com/product/alternative-approaches-in-conflict-
resolution-1st-edition-martin-leiner/
Analysing, Planning and Valuing Private Firms: New

Approaches to Corporate Finance Federico Beltrame
https://ebookmass.com/product/analysing-planning-and-valuing-
private-firms-new-approaches-to-corporate-finance-federico-
beltrame/
Numerical Methods and Optimization in Finance [2nd ed.]

Manfred Gilli
https://ebookmass.com/product/numerical-methods-and-optimization-
in-finance-2nd-ed-manfred-gilli/
Corporate Governance of Chinese Multinational

Corporations: Case Studies 1st ed. Edition Runhui Lin
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/
Corporate Finance - European Edition David Hillier
https://ebookmass.com/product/corporate-finance-european-edition-
david-hillier/
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-
mcgraw-hill-irwin-series-in-finance-insurance-and-real-
estate-12th-edition-brealey/
Principles of corporate finance Brealey
https://ebookmass.com/product/principles-of-corporate-finance-
brealey/
Fundamentals of Corporate Finance (Berk, DeMarzo &

Harford, The Corporate Finance Series) 4th Edition –
Ebook PDF Version
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
ISBN 978-3-030-53808-8 ISBN 978-3-030-53809-5 (eBook)

https://doi.org/10.1007/978-3-030-53809-5
© 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
The applications of event study methodology for empirical corporate fi-

nance initiate the discussions on the robustness of the methods when
empirical data are used. The skepticism (for likely data manipulation) on
applicability of event study methods such as selection of relevant events,
event periods, necessary robust statistics to apply, all require further
analyses to justify the empirical results.
In presenting the possible alternatives in my first book Analyzing
Event Statistics in Corporate Finance, the essential ambiguity of cur-
rent applications of conventional methods (especially the impacts from
the event windows) are briefly surveyed. In addition, the proposed al-
ternative methods may require further rigorous elaborations on the-
oretical developments (in statistics) and demonstrations on empirical
applicability.
These concerns initiates the need to provide an additional book that
may encompass the necessary elaborations and developments. This cur-
rent book can be considered as a reference book for the event study
methodology where issues in economics, finance, management are of
interest.
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.
Los Angeles, USA Jau-Lian Jeng
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
The purpose of the event studies in corporate finance is to identify the

possible impacts from the occurrence of corporate finance events and
propose the plausible explanation/hypothesis to present the causes from
theoretical background. Initial contribution from Fama et al. (1969) pro-
vides the pioneering analysis for event studies via market efficiency as
considering the adjustment of stock price to new information. For years,
some huge amount of publications in applying the event study methodol-
ogy can be found in many journals of (say) accounting research, financial
economics, and corporate finance. However, few publications and books
that survey and analyze the methodology while providing suggestions
and improvements over the limitations of these methods in empirical
applications.
In particular, with the recent concerns on the systemic risk Notice that
the definition of systemic risk stated here emphasizes the persistence and
prevalence of a single corporate event that my affect the entire capital
market or system. This differs from the classical definition of system-
atic risk which is nondiversifiable from the perspective of some well-
diversified portfolios that consist of a large number of securities in the
ix
x Introduction
capital market. which is considered as a single corporate event may per-

sist strongly to impact the entire capital market, it is necessary to see if
the essence or impact of some corporate events may be serious enough to
produce turmoils in the capital market. For instance, the study provided
by Acharya et al. (2016) considers the possible model and measure for
systemic risk where the difficulty of financial institutions during 2007–
2009 causes a widespread externality and impact onto the entire capital
market. Hence, justification of the corporate event studies should not be
limited to assessment on the existence of significant events only. Instead,
the suitable analysis should also consider the persistence of impacts from
events and their possible consequences.
This book is to discuss (1) the conventional event study methodolo-
gy in corporate finance, and (2) to consider the alternative methods that
may improve the robustness of statistical inferences for the possible im-
pacts and endurance from the corporate events. Although event studies
are performed in many publications of finance literature for the corporate
finance, accounting and management issues, few discussions are shown
to assess the soundness of the existing methodology in some rigorous
settings. In particular, inappropriate handling of data selection (such as
pre-event and post-event windows), calculation of expected (or normal)
returns, and applications of empirical statistics (in favor of certain pre-
ferred hypotheses) may derive wrongful justification for the corporate
finance events.
One reason for this skeptical perspective on conventional event studies
is that, although corporate events may be similar or repetitive in nature
as they seem, the underlying circumstance over different time horizons
or schemes of studies may differ. Therefore, although historical evidence
shows that certain hypotheses may prevail in the past, it is too conclusive
to assert that the existing hypotheses will continue to hold true since
the capital market may or may not interpret the given corporate finance
information identically over time.
As stated, some cautions should be imposed on the existing method-
ology since the purpose of corporate finance event study is to objectively
identify whether there exist significant impacts from the events of interest
or not. It is not to cope with some plausible pre-identified hypotheses (or
Introduction xi
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
The conventional statistics with cumulative abnormal returns may ac-

tually lead to wrongful conclusion. Specifically, let the unknown struc-
tural change happen at t = T ∗ where T ∗ < T , T ∗ is unknown and αi
becomes αi∗ , where αi∗ = αi + δi , and δi = 0 for a sufficient number of
firms of study. Suppose the empiricist does not consider the unknown
structural change at all, it is apparent that the cumulative abnormal
returns can be shown as t > T ∗

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
studies should be based on the clear dichotomy between systematic and

firm-specific components of asset returns.
Accordingly, a prerequisite of robust assessments on abnormal returns
(for event studies) lies in the prevalent efforts to obtain the systematic
components of asset returns. Therefore, to obtain the abnormal returns
for the purpose of event studies, one needs to ensure that the (condi-
tional) expected returns of assets are based solely on the nondiversifi-
able systematic information, even though adding some additional vari-
ables (and/or time-series properties such as (G)ARCH in mean) is attrac-
tive with better in-sample goodness-of-fit statistics. Yet, the methodology
should consider that the more essential the event is, the more time it may
consume the impacts and more frequently the impacts may happen on
the information announced.
For instance, the recent studies in the earthquake modeling, the so
called Epidemic Type Aftershock-Sequences (ETAS) model with Condi-
tional Intensity Function (Ogata [1998] 2013) examines the earthquake’s
aftershock model through the time intervals, magnitudes, and frequen-
cies of the aftershocks. Basically, the investigation is on the strength of
earthquakes trough time and the intensities of magnitudes specified as
∞ t K0
λ (t|Ht ) = μ + eα(M−M0 ) N (ds, d M)
M0 0 (t − s + c) p
where K 0 , α, c, p are constants, and N (ds, d M) = 1 if an infinitesimal

element (ds, d M) includes an event (ti , Mi ) for some i, and otherwise,
N (ds, d M) = 0. And the H t is the past history of the earthquake. The
aftershock activity is not only due to the magnitude of the shock but al-
so on the time intervals that carried over the quake and the frequency
of occurrence of some magnitudes (for the aftershocks) during the time
span. Other than the magnitude of the earthquake, the shorter the af-
tershocks (for certain magnitudes) may happen and more frequently, the
higher the intensity of the quake. The application can also be seen in
Egorova L. and I. Klymyuk (2017) in using the Hawke process to ana-
lyze the currency crash. Let the process that describes the history of oc-
currence be denoted as a sequence Ht = {(ti , m i )}i∈N ∗ , N ∗ = N ∪ {0},
where t i denotes the time m i denotes the depth of decline. The process
Introduction xv

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
More specifically, let

λ (t) = λ0 (t) + ν (t − ti )
ti <t
where λ0 (t) : R → R+ is the base intensity and v : R+ → R+

reflects the impact of pat event t i on the current intensity. In the one
p case as Ht = {(ti ,)}i∈N ∗ and the exponential function as
dimensional
ν (t) = i=1 αi e−βi , the base intensity is assumed to be constant, the
intensity function of Hawke process is given as
t
λ (t) = λ0 + αi e−β(t−s) d Ns = λ0 + αi e−β(t−ti ) .
−∞ ti <t
Hence, the intensity of the crash can be specified as the function of

time path and numbers of earthquakes may happen in the time interval
specified. This gives us the reasoning that the event studies which only
investigate the magnitude of the abnormal returns through statistical in-
ferences is not sufficient enough to verify the incidence, In fact, the time
span and frequency of the impacts after the original shock that incorpo-
rate the information flows should be studied to identify the significance
of the events. Therefore, to investigate the events, one should consider
that the essential events will constitute the impacts or frequency of im-
pacts (similar to a significant earthquake) that may endure and last for a
significantly long period of time—depending on the speed of the market
adjustments. In other words, the intensity of the impacts (of events) is to
verify the significance of event duration in addition to the magnitude of
the information announced. Hence, with similar logic, the event studies
in corporate finance should consider the possible time-frame or duration
xvi Introduction
after the announcement of the events. Either the conventional method

in so-called residual analysis or the structural change test (in the regres-
sional coefficients) is no longer sufficient enough to justify the essence
and impacts of the events. Instead, an additional approach in consider-
ing the time span of any event of interest provides the robust method in
assessing the impacts or consequence of the issue.
References
Acharya, V.V., L.H. Pedersen, T. Philippon, and M. Richardson. 2016. Mea-

suring Systemic Risk. Review of Financial Studies 30: 1–47.
Egorova, L., and I. Klymyuk. 2017. Hawkes Process for Forecasting Currency
Crashes: Evidence from Russia. Procedia Computer Science 122: 1182–1188.
Fama, E.F. 1969. The Adjustment of Stock Prices to New Information.
International Economic Review, 10: 1–21.
MacKinlay, A.G. 1997. Event Studies in Economics and Finance. Journal of
Economic Literature 35: 13–39.
Ogata, Y. 1998. Space-Time Point-Process Models for Earthquake Occurrences.
Annals of Institute of Statistical Mathematics 50: 379–402.
Ogata, Y. 2013. A Prospect of Earthquake Prediction Research. Statistical
Science 28: 521–541.
Contents
Part I The Conventional Approach
1 Popular Methods for Event Studies in Corporate

Finance—Subjectivity Versus Robustness 3
Part II Alternative Approach for the Contemporaneous

Event Studies
2 Assessments of Normal Returns 55
3 Occupation Time Statistics—The Intensity of Events 91
4 Monitoring Tests and the Time of Duration 147
5 Sequential Monitoring for Corporate Events in Using

Occupation Time Statistics 161
xvii
xviii Contents
6 Real-Time Applications of Monitoring and Empirical

Performance 201
Concluding Remark 223
Index 225
List of Tables
Table 6.1 The Test Statistics Z (h) 216

Table 6.2 The Conventional CAR’s tests with conventional OLS
estimates and recursive least squares. The traditional
CAR’s test is based on the estimation period from
−250 days to −120 days prior to the event date to
estimate the market model. The recursive least squares
CAR’s test is based on the recursive least squares
estimates for the market model for the entire sample
period. The CAR’s tests for both methods are based on
the event window −2 days prior to the event date up
to +5 days after the event date 220
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
© The Author(s) 2020 3

J.-L. Jeng, Contemporaneous Event Studies in Corporate Finance,
https://doi.org/10.1007/978-3-030-53809-5_1
4 J.-L. Jeng
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
1.2 Identification of Event

Windows—Timing the Events
As requested in almost all conventional event studies, the identification
of event dates is critical for the findings. The issue is not only to identify
the importance of (similar) events that may prevail over various firms
(or over different countries), the precision of selecting the event dates
over the sampled period is essential since the choices may be sensitive for
the empirical results. Incorrect event dates may cause confusion over the
choices of estimation window and post-event window also. In practice,
the conventional methods simply divide the sampled period into three
arbitrary intervals over a time horizon: the estimation window, the event
window, and the post-event window.
Although precise clarification over these windows is informative and
provides better analyses for the event studies, the arbitrariness of the
choices may either cause the sample selection problem or even worse,
the intentional data manipulation for the desired hypothesis to be con-
firmed with empirical results. For instance, Berkman and Truong (2009)
show that the after-hours earnings announcements may occur during the
event periods. They also show that the event dates can not be adjusted
for after-hours earnings announcements. Hence, incorrect identification
of event dates may cause different results of verification for the impacts of
corporate (finance) events.1
More specifically, the subjective classification for the pre-event period
(or estimation window), event period, and post-event period in a time
horizon of interest may introduce the biases in estimating the normal
returns if the pre-event period is identified too distant from the genuine
initiative impacts of the events of interest (in the event window).
Given that the market may likely adapt to the new information sequen-
tially, if the pre-event period (although sufficiently long enough) is iden-
tified too distant from the initial date of event impacts, the estimates for
normal returns in using the past history of the stock returns may not obtain
insufficient information. The reason is intuitive. If the cutoff point for the
pre-event period is too distant from the initial impact, the information for
the market adjustments on systematic components may not be sufficient.
6 J.-L. Jeng
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.
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
In particular, other than the ad hoc methods in determining the event

period, Brown et al. (1985) applied the maximum likelihood estimation
(for each specific firm) to determine the event period for event studies.
Although the work assumes normality for abnormal returns, the optimal
event period (for each firm of interest) is determined by a data-dependent
framework where log-variance ratio can be applied.
Ball and Torous (1988), in using the switching regression model, discuss
the randomization of event dates under normality assumption to identify
the possible bias in estimation period. Burnett et al. (1995) consider pos-
sible multiple structural changes may occur during the event period for
normal return. Karafiath and Spencer (1991) discuss the multiperiod event
study model for the event studies, while Aktas et al. (2007) consider the
consequence of contaminated estimation period due to possible unrelated
events that may occur in the event period determined in ac hoc scheme.
Notice that the issue of event period determination is not separable with
the possible structural change (of the parameters) in the models for asset
returns as denoted by Brown et al. (1985). With the setting of normality
and consider possible structural change within the event period, their
model explicitly determines the optimal event dates (and hence event
period) for each specific firm of interest. The difficulty however is, the
corporate events are (in their setting) assumed influential enough to cause
structural change of the asset pricing models with presumed factors. It is,
however, hard to see that corporate events (which are usually firm-specific)
will cause the structural change of asset pricing models to bear the impacts.
What is difficult to perceive is that if the impacts of corporate events
are significant enough to cause structural change (in the parameters) of
asset pricing models, and if the asset pricing models are to explain the
normal returns (or systematic components of asset returns), there are two
possible conclusions one can conjecture: (1) the asset pricing models are
either incorrectly specified or of constant coefficients; (2) the corporate
events should be considered as possible additional explanatory variables
(or factors) for the asset pricing models. The first issue is easier. That
is, one can apply more robust adaptive schemes that may accommodate
the varying coefficients to estimate the models for normal returns. The
second issue then requires additional (theoretical) justification to explain
why these corporate events are essential enough to include in the possible
cross-sectional asset pricing models. In particular, the cross-sectional

regression models which may include various dummy variables, need to
consider the specific event dates for each firm.
If one would like to investigate the impacts of events in using the
dichotomy of normal and abnormal returns, the difficulty of the above-
mentioned issue 1 can be alleviated by applying model selection tests and
adaptive/recursive estimations. The reason is that the magnitude of firm-
specific risks or abnormal returns, in response to the corporate events,
should not be influenced by the parameter changes due to the model
applied in estimating the normal returns—given that the models applied
in estimating normal returns are only approximations or “filters” for the
systematic expectations. Determination of event period in event studies
hence must ensure that the impacts of events (on abnormal returns) are
not contaminated by the possible influence of market conditions. Unfor-
tunately, many research articles in this field of event studies seem to incur
the combinations of fitted models with structural changes and the corpo-
rate event issues altogether, particularly in the regression models of asset
returns.
For instance, Klein and Rosenfeld (1987) show that the abnormal
returns are positively (negatively) biased when there is a bull (bear) market
in the constant-coefficient models of normal returns. In other words, given
that the corporate events are idiosyncratic risks, the abnormal returns for
the event studies should be immunized from the influence of market’s
systematic components. These discussions lead to the alternative methods
of adaptive schemes in estimation for normal returns where the determi-
nation of event period can be handled from a data-dependent perspective.
More specifically, for the event studies in corporate finance, the estima-
tion of normal returns should be robust enough to consider every possible
systematic component in the return series. And for the robustness of event
study research, the analyses of abnormal returns therefore, are “purified”
to associate solely with idiosyncratic (or firm-specific) risk of return series
in order to consider the impacts of corporate (finance) events.
10 J.-L. Jeng
1.2.1 Event Dependency and Structural Change

in Security Pricing Models
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
rit = X t γi1 + εit , t < t− , t > t+

(1.1)
rit = X t (γi1 + γio ) + εit , t− ≤ t ≤ t + ,
where X t is the t-th date observations of k explanatory variables (or fac-

tors) and γi are the coefficients associated with the factors with a pos-
sible structural change in the event period t− ≤ t ≤ t+ . Assuming nor-
mality for the error terms in Eq. (1.1), and εit ∼ N (0, σit2 ), ∀t with
E[εit εis ] = 0, ∀t = s, the error terms are possibly heteroscedastic and
serially uncorrelated.
Notice that the setting of Brown et al. (1985) does not specifically
separate the return series into normal returns and abnormal returns where
cumulative abnormal returns can be applied to consider the impacts of
events. Instead, the model assumed in Eq. (1.1) includes all possible factors
that can be applied to explain asset returns. Implicitly, it is assumed that
there exists a structural change for the parameters of the fitted model within
the event period which requires firm-specific identification. The presumed
corporate event is considered as influential if there exists a structural change
in the related explanatory variables. However, there is no justification why
the corporate event issues may cause the parameters of the factors to change
in the event period.
And hence, the null hypothesis as no significant event for the fitted
model is to consider that
Ho : γio = 0,
for all i = 1, 2, . . . , n. By stacking up the i-th security return across time

periods and assuming constant variance for unsystematic risk within each
time period, Eq. (1.1) can be shown as
R1 = X 1 γ1 + ε1 , t ∈
/ [t− , t+ ]
(1.2)
R2 = X 2 γ2 + ε2, t ∈ [t− , t+ ],
where ε p ∼ N (0, σ p2 I p ); p = 1, 2, and I p stands for an identity matrix

with the dimension equal to the observations (of i-th security return) in
each time period. Likewise, the fitted regressions for each i-th security
12 J.-L. Jeng
return can be written as
R1 = X 1 γ̂1 + e1
(1.3)
R2 = X 2 γ̂2 + e2 ,
where e p , p = 1, 2 represents the residual vectors from fitted regressions.

Under the null hypothesis and assumed normality, it is feasible to see that
the statistic
−1 −1
(γ̂1 − γ̂2 ) σ12 X 1 X 1 + σ22 X 2 X 2 (γ̂1 − γ̂2 ) (1.4)
will have a chi-square distribution with degree of freedom equal to the

parameters estimated in either partition of time periods. In addition, the
statistic such as
e1 e1 e2 e2
+ 2 ∼ χ(T 2
1 +T2 −2k)
, (1.5)
σ1 2 σ2
where T1 and T2 are the number of observations of i-th security returns

in non-event period and event period, respectively. Given the normality,
it is also available to see that Eqs. (1.4) and (1.5) are independent. And
hence, the test statistic for structural change is shown as
⎡ −1 −1 ⎤
(γ̂1 − γ̂2 ) σ12 X 1 X 1 + σ22 X 2 X 2 (γ̂1 − γ̂2 ) T1 + T2 − 2k
⎢ ⎥
F =⎣ ⎦ , (1.6)
e1 e1 e2 e2 k
σ 2 + σ 2
1 2
is an F -statistic with degree of freedom (k, T1 + T2 − 2k) . In particular,

when the unbiased estimator σ̂ p2 for σ p2 is introduced such that σ̂ p2 =
ep e p
T p −k , p = 1, 2, the test statistic becomes
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,
as stated earlier, the structural changes can be the consequences of other

unrelated events, the market conditions, and many other possible causes.
Besides, it is easy to observe that these parameters (of asset pricing models)
are rarely stable over time in most cases even though there’s no particular
issue or event.
The major contribution of Brown et al. (1985) is that the framework
depicts the bias when event period is incorrectly defined and allows to
determine the optimal event period in using the maximum likelihood
estimation. In other words, the determination for the event period is
based on the data obtained and hence, is data-dependent and may vary
across firms even for similar events.
For simplicity, denote the event period as [a, b] as in Brown et al.
(1985), which may be considered as parameters to be estimated across
different firms. It is also noticeable that, by substituting the maximum
likelihood estimator of the fitted models of systematic risk and various
event periods, the log-likelihood function can be shown as some variance
ratios based on different possible event period (or namely, various (a, b)).
Specifically, the system of equations for security return rit after stacking
up the observations can be shown as

γ∗
Ri = [X : Z i ] i1 +εi
γio∗
= [X : Z i ] γi∗ + εi ,
(1.8)
γ
Ri = [X : Wi ] i1 +εi
γio
= [X : Wi ] γi + εi ,
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
where ∗ and are diagonal covariance matrices with respective elements

as σ p∗2 , or σ p2 as p = 1, 2, corresponding to the observations in the event
period or non-event period, respectively. Based on the above equations, it
is shown that the estimates for the equations under incorrectly identified
event period are given as
−1 −1
γ̂1i γ1i X −1 M
X X M γ
= + −1 w −1 −1 w o , (1.10)
γ̂oi γoi W Mx W W Mx γo
where Mw = I − W (W −1 W )−1 W −1 , and Mx = I − X

(X X )−1 X −1 , = Z − W .
−1
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
L(σ1 , σ2 , γ1 , γ2 ) ∞ − (T − E)lnσ1 − Elnσ2

1
− 2
(R1 − X 1 γ1 ) (R1 − X 1 γ1 )
2σ1 (1.11)

1
− 2
(R2 − X 2 γ2 ) (R2 − X 2 γ2 )
2σ2
Hence, the maximum likelihood estimators for the parameters will be

given as3
−1
γ̂i = X i X i X i Ri i = 1, 2,

σ̂1 = R1 − X 1 γ̂1 R1 − X 1 γ̂1 / (T − E)
2 (1.12)

σ̂22 = R2 − X 2 γ̂2 R2 − X 2 γ̂2 /E,
where E = (b − a − 1). Upon substitution of these estimators, the log-

likelihood function will become
L(a, b)∞ − (T − E)ln σ̂1 − (E)ln σ̂2 . (1.13)
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.
1.2.2 Random Event Dates (Ball and Torous 1988)
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
θt = 1, if the event occurs on day t,

θt = 0, otherwise.
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
pt = Pr ob[θt = 1, θs = 0, s ∈ (−c, . . . , 0, . . . , c), s = t]. (1.14)
Given the definition of probability of event date, the security return

process can be redefined as follows. If there is no event at day t, let the
conditional probability of return process be given as
rit |θt = 0 ∼ N (μi , σi2 ),
which is identical to the specification of return process in estimation

period.5 On the other hand, if there’s an event that happens at day t
in the event period, it is shown that the conditional probability of return
process is assumed as
rit |θt = 1 ∼ N (μi + Aσi , δ 2 σi2 ),

where A is the abnormal performance introduced by the event and it is

actually the standardized return, and proportional to the standard devia-
tion of the i−th firm’s security return in estimation period. In addition,
since by assuming δ 2 = 1, the event-induced volatility can be considered
in this setting.6 The parameters A and δ 2 are assumed as constant across
all firms.
Hence, the significance of events is assessed through the hypothesis
based on the estimates of A and δ 2 while the estimates
pt = Pr ob[θt = 1, θs = 0, s ∈ (−c, . . . , 0, . . . , c), s = t]
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
Stage 2: Assume that μ = μ̂ and σ 2 = σ̂ 2 , and use the security

returns
in the event period to estimate the 2c + 2 of parameters as
A, δ 2 , p−c , . . . , pc−1 .
Given these estimates, the hypothesis testing is to consider the null
hypothesis as
Ho : A = 0, δ 2 = 1,
and the alternative hypothesis as7
Ha :
Ho
where the notation

Ho denotes that the statement of Ho is not true.
18 J.-L. Jeng
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,
The model setting will consider that
xt |θt = 0 ∼ N (0, 1), xt |θt = 1 ∼ N (A, δ 2 ).
Denote the conditional distribution of x given θ = (θ−c , . . . , θo , . . .

θ+c ) as f (x|θ), which is multivariate normal with mean vector Aθ and the
covariance matrix as (δ 2 − 1)θθ T + I2c+1 . Hence, the joint distribution
of x is shown as
f (x) = f (x|θ)d P(θ),
θ
where P(θ) is the distribution function of θ . Given the above setting, it

can be shown that
c
f (x) = pt gt (x), (1.16)
t=−c
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.
1.2.3 Contaminated Estimation Period (Aktas et al.

2007)
The setting of Aktas et al. (2007) is to consider possible contaminating

events in the estimation period which may lead to bias in estimating
the parameters of return-generating process. Although their work solves
some sorts of non-related event for contamination problem in estimation
period, it only partially relates to the issue of event date uncertainty. In
brief, the setting still considers that the event date is known and there exist
some unrelated events that may happen within the estimation period. The
setting is still of the arbitrariness in separating estimation and event periods
with no explicit rule for the classifications. Following Aktas et al. (2007),
let the return-generating process be defined as
⎡
⎤
αi
Ri = X i bi + εi = [1, Rm , Di ] ⎣ βi ⎦ + εi , (1.17)
γi
where Ri is a vector of returns for firm i, X i = [1, Rm , Di ] and Rm is the

vector of market portfolio returns, Di is a dummy variable that is equal to
1 at the event date and 0 otherwise, bi is a vector of coefficient estimates for
firm i. Assuming homoscedasticity for simplicity,9 the covariance matrix
of OLS estimator for each security return is shown as
Covi (bi |X i ) = σi2 (X i X i )−1 . (1.18)
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
The variance of the residuals for each state is assumed as

E[εi1 εi1 |X i ] = σi1
2
I St = 1,

(1.20)
E[εi2 εi2 |X i ] = σi2
2
I St = 2,
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
and the pmn = Pr (St = m|St−1 = n) is the transitional probability of

changing from state n to state m. Hence, the unconditional probability of
each regime can be expressed as
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
C O V O L S (bi |X i ) = p(St = 1)σi,1

2
(X i X i )−1 + p(St = 2)σi,2
2
(X i X i )−1 .
(1.23)
Given that σi,2

2 > σ 2 ,it is easy to see that
i,1
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.
In other words, if there is a multi-event situation in the event period, there

is a possibility that the OLS estimate may be incorrectly stated and the
empirical results of statistical inferences are not reliable.
1.2.4 Multiperiod Event Studies (Karafiath and

Spencer 1991)
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
where ri1 is a T × 1 vector of observed returns for i-th security in the

estimation period, ri2 is a Ti × 1 vector of returns for i-th security in
the event period, X i1 and X i2 are the T × 2 and Ti × 2 matrices for
a constant term and market returns, respectively. In addition, ITi is the
identity matrix with dimension Ti and i is a 2 × 1 vector for market
model parameters, δi is a Ti × 1 vector of abnormal returns for i-th firm
in the corresponding event window. The error terms εi1 and εi2 are all
zero-mean stochastic disturbances. The system can be denoted in a more
compact form as
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
1.3 Cumulative Abnormal

Returns—Construction and the
Statistical Test
The earlier study of Brown et al. (1985) indicates that the event itself
may in fact be the systematic risk during the period surrounding the firm-
specific event. This shows that the confusion may start from the impacts of
the event toward to the understanding of systematic risk and hence, it may
obscure the event study analysis when using the residuals from the asset
pricing models to consider the impacts and effects of the abnormal returns.
Yet, it confirms that the incorrect imposition of systematic risk models
may alternate the findings thereafter. One difficulty from the theoretical
standpoint is that if the firm-specific event may cause the systematic risk
to change, what causes it to be called “systematic”?
In general, Brown et al. (1985) consider the multi-factor return model
for j -th security with a two-stage switching model as
R = X t γ j1 + u jt , t < t−1 or t > t1

R jt = X t (γ j1 + γ j0 ) + u jt = X t γ j2 + u jt for t−1 ≤ t ≤ t1
where X t is the r -th row of a data matrix containing k explanatory variables

and γ j is the j-th component of the associated k-vector of parameters. They
also assume that

u jt ∼ N (0, σ jt2 ) ∀t E u jt u js = 0 ∀t, s
Given the null hypothesis as
H0 : γ0 = 0
we can rewrite these equations in matrix form as
R1 = X 1 γ1 + u 1 , for t∈/ [t−1 , t1 ]

R2 = X 2 γ2 + u 2 , for t ∈ [t−1 , t1 ]
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,
and with the above assumptions, under the null hypothesis
(g1 − g2 ) ∼ N (0, σ12 (X 1 X 1 )−1 + σ22 (X 2 X 2 )−1 ),
Hence, the random variable

−1
(g1 − g2 ) σ12 (X 1 X 1 )−1 + σ22 (X 2 X 2 )−1 (g1 − g2 ) (1.27)
will have a chi-square distribution of degree of freedom equal to the num-

ber of parameters. Then again,
e1 e1 e2 e2
+ ∼ χ(T
2
1 +T2 −2k)
(1.28)
σ12 σ22
where T1 and T2 are the numbers of observations in each model and k is

the number of parameters estimated. Given the mutual independence of
Another random document with
no related content on Scribd:
as Britomart’s, the beauty of the inward purity of womanhood; but it is a
beauty of pure form.
“And soothly sure she was full fayre of face,
And perfectly well shapt in every lim,
Which she did more augment with modest grace,
And comely carriage of her count’nance trim,
That all the rest like lesser lamps did dim.”
(VI. ix.
And yet as she stands on the little hillock she is encompassed with a
cloud of glory.
“Upon a litle hillocke she was placed
Higher then all the rest, and round about
Environ’d with a girland, goodly graced,
Of lovely lasses, and them all without
The lustie shepheard swaynes sate in a rout;
The which did pype and sing her prayses dew,
And oft rejoyce, and oft for wonder shout,
As if some miracle of heavenly hew
Were downe to them descended in that earthly vew.”
(VI. ix.
They saw in the object before their eyes the idea of beauty in earthly
form. The miracle is no more and no less than this; it is “the privilege of
beauty, that being the loveliest she is also the most palpable to sight.”
BIBLIOGRAPHY
Aristotle. “The Nicomachean Ethics of Aristotle,” translated into

English by Robert Williams. London, 1869.
Ayres, Philip. “Lyric Poems.” London, 1687.
Aytoun, Robert. “Poems,” edited by Charles Rogers. London,
1871.
Bigg, C. “Neo-platonism.” London, 1895.
Boyle, Robert. “A Treatise of Seraphic Love.” Edinburgh, 1825.
Brome, Alexander. “Songs and Other Poems.” Second edition.
1664.
Campion, Thomas. “Works,” edited by A. H. Bullen. London, 1889.
Carew, Thomas. “Poems,” edited by Arthur Vincent. London and
New York, 1899.
Cartwright, William. “Comedies, Tragi-comedies with Other
Poems.” London, 1651.
Charleton, Walter. “The Ephesian and Cimmerian Matrons.”
London, 1668.
Cleveland, John. “Works.” London, 1687.
Cowley, Abraham. “Works.” 2 vols. Eleventh edition. London,
1710.
Craig, Alexander. “Poetical Works,” printed for the Hunterian
Club, 1873.
Crashaw, Richard. “Complete Works,” edited by A. B. Grosart. 2
vols. 1873.
Daniel, George. “Poems,” edited by A. B. Grosart. 4 vols. 1878.
Daniel, Samuel. “Complete Works in Verse and Prose,” edited by
A. B. Grosart. 4 vols. 1885.
D’Avenant, William. “Dramatic Works,” edited by James
Maidment and W. H. Logan. 5 vols. Edinburgh and London,
1872–1874.
De Vere, Aubrey Thomas. “Characteristics of Spenser’s Poetry”
and “Spenser as a Philosophic Poet,” in “Essays, chiefly on
Poetry.” 2 vols. London, 1887.
Dodge, R. E. Neil. “Spenser’s Imitations from Ariosto,” in
“Publications of Modern Language Association of America,”
1897.
Donne, John. “Works,” edited by Henry Alford. 6 vols. London,
1839.
“Poems,” edited by E. K. Chambers, with an introduction by George Saintsbury.

2 vols. London and New York, 1896.
Dowden, Edward. “Puritan and Anglican Studies in Literature.”

New York, 1901.
Drayton, Michael. “Poems.” London, 1613.
“Poemes, Lyrick and Pastorall,” printed for the Spenser Society. 1891.
Drummond, William. “Poems,” edited by William C. Ward. 2 vols.

London and New York, 1894.
“Ben Jonson’s Conversations with William Drummond.” Shakespeare Society

Publications, vol. 8.
Dryden, John. “Works,” edited by Sir Walter Scott, revised by

George Saintsbury. 18 vols. Edinburgh, 1887.
Einstein, Lewis. “Italian Renaissance in England.” New York,
1902.
Ferri, Luigi. “Platonismo di Ficino,” in “La Filosofia delle Scuole
Italiane,” vol. 29, pp. 269–294.
Ficinus, Marsilius. “Commentarium in Convivium,” in “Omnia
Divini Platonis Opera.” Basileae, 1551.
Fletcher, Giles. “Complete Poems,” edited by A. B. Grosart.
London, 1876.
Fletcher, J. B. “Précieuses at the Court of Charles I,” in the
“Journal of Comparative Literature,” vol. 1, pp. 120–153.
Fletcher, Phineas. “Poems,” edited by A. B. Grosart. 4 vols. 1869.
Greville, Fulke. “Works,” edited by A. B. Grosart. 4 vols. 1870.
Habington, William. “Castara,” edited by Charles A. Elton.
Bristol, 1812.
Herbert, Edward, Lord Herbert of Cherbury. “Poems,” edited by
John Churton Collins. London, 1881.
Herbert, George. “Poetical Works,” edited by A. B. Grosart.
London, 1876.
Heywood, Thomas. “Love’s Mistress or the Queen’s Masque,”
edited by Edmund Goldsmid. Edinburgh, 1886.
Howell, James. “Familiar Letters,” edited by Joseph Jacobs. 2
vols. 1892.
“Lustra Ludovicii.” London, 1648.
Inge, William Ralph. “Christian Mysticism.” New York, 1899.

Jonson, Ben. “Works,” with notes by William Gifford, edited by Lt.
Col. Francis Cunningham. 3 vols. London, 1871.
Linche, Richard. “Diella,” edited by A. B. Grosart, in “Occasional
Issues of Unique or very Rare Books,” vol. 4. 1877.
Lovelace, Richard. “Lucasta,” edited by W. Carew Hazlitt.
London, 1864.
Lowell, James Russell. “Spenser,” in “Works,” vol. 4. Riverside
edition, 1898.
Masson, David. “Life of John Milton.” 6 vols. Cambridge, 1859–
1880.
Milton, John. “Poetical Works,” edited by David Masson. 3 vols.
London, 1874.
“Prose Works,” edited by Charles Symmons. 7 vols. London, 1806.
More, Henry. “Psychodia Platonica, or a Platonicall Song of the

Soul consisting of ... Psychozoia, Psychathanasia, Anti-
psychopannychia, Anti-monopsychia.” Cambridge, 1642.
Norris, John. “A Collection of Miscellanies.” London, 1706.
Parker, Samuel. “A Free and Impartial Censure of the Platonick
Philosophie.” Second edition. Oxford, 1667.
Plato. “Dialogues of Plato,” translated into English by B. Jowett. 5
vols. New York, 1892.
Plotinus. “Opera Omnia ... cum Marsilii Ficini Commentariis,”
edited by Frederick Creuzer. 3 vols. Oxford, 1835.
“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.
Randolph, Thomas. “Poetical and Dramatic Works,” edited by W.

Carew Hazlitt. 2 vols. London, 1875.
Sedley, Charles. “Works.” 2 vols. London, 1722.
Shakespeare, William. “Poems,” edited by George Wyndham.
London, 1898.
Sidney, Philip. “Astrophel and Stella,” edited by Ewald Flügel.
Halle, 1889.
“The Countess of Pembroke’s Arcadia,” edited by H. Oskar Sommer. London,

1891.
Spenser, Edmund. “Complete Works,” edited by A. B. Grosart. 10

vols. 1882–1884.
“Faerie Queene.” Book I, edited by H. M. Percival. London, 1894.

“Faerie Queene.” Book II, edited by G. W. Kitchin. London, 1868.
Stanley, Thomas. “Poems.” London, 1814.

Suckling, John. “Poems, Plays, and Other Remains,” edited by W.
Carew Hazlitt. 2 vols. London, 1892.
Tulloch, John. “Rational Theology and Christian Philosophy in
England in the Seventeenth Century.” Second edition. 2 vols.
Edinburgh, 1874.
Vaughan, Henry. “Poems,” edited by E. K. Chambers, with an
introduction by H. C. Beeching. 2 vols. London and New York,
1896.
Walton, Izaak. “Complete Angler and the Lives of Donne, Wotton,
Hooker, Herbert, and Sanderson.” London, 1901.
Ward, Henry. “Life of Henry More.” London, 1710.
Whittaker, Thomas. “The Neo-Platonists: a Study in the History
of Hellenism.” Cambridge, 1901.
INDEX
Absence, effect of, in love, 143, 144, 145, 146, 148.

Acrasia, a type of sensual beauty, 20;
captured by Guyon, 21.
Adam, 43, 44, 83.
Amavia, 24.
“Anatomy of the World, An,” 162, 165.
“Answer to the Platonicks,” 162.
“Anti-monopsychia,” 194, 196.
“Anti-Platonick” (Cleveland’s), 162.
“Anti-Platonick” (Daniel’s), 159.
“Anti-psychopannychia,” 194, 195.
“Apology for Smectymnuus, An,” 47.
“Arcadia,” 66.
Archimago, 14.
ἀρετή, identification of Una with, 2.
Ariosto, 26, 39.
Arthegal, his reverence for Britomart, 35, 37, 38, 40;
his training in justice, 28.
Arthur, as heavenly grace, 3, 62;
his relation to the Red Cross Knight, 62;
his function in scheme of “Faerie Queene,” 62.
Astræa, 27.
“Barriers, The,” 125.
Baxter, 92.
Beauty, in Ficino, 109, 112, 113, 114;
in Plato, 34, 35, 220;
in Milton, 41;
in Spenser, 35,38, 41, 109, 111, 112, 113, 220, 221.
of beloved, its relation to absolute beauty, in George Daniel, 131;
in Ficino, 115;
in Shakespeare, 128, 134, 135;
in Spenser, 32–34, 115, 130, 136.
earthly, 79, 80.
heavenly, a fundamental doctrine of Platonism, 1, 103;
identified as wisdom, 73;
as beauty of intelligible world, 76.
“Beauty,” 86.
Beelzebub, 58.
Being, true, 98.
Belphœbe, 5.
Bower of Bliss, 21.
Boyle, Robert, 158.
Bradamante, 39.
Britomart, 35, 36–38.
“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.
Daniel, George, 131, 158.

Daniel, Samuel, 138.
D’Avenant, William, 156.
Diodati, Charles, 41.
Donne, John, mysticism in, 94;
love in, 141, 144, 145, 149, 51, 152, 153, 154;
his idea of woman, 163, 164, 165.
Drayton, Michael, 125, 138.
“Dream, The,” 153.
Drummond, William, his idea of God, 174, 175, 176, 183;
his idea of love, 82, 88, 132;
his idea of heavenly love, 76, 81;
appeal of Platonism to, 76;
his idea of happiness, 86;
his idea of rest, 87.
Dryden, John, 165.
Duessa, 66.
“Ecstacy, The,” 141.

Elissa, 22.
“Enneads,” see under Plotinus.
“Epithalamion,” 31, 32, 33.
“Epithalamy,” 162.
“Epode,” 151.
Eve, 44, 45.
“Faerie Queene,” Christianity and Platonic idealism in, 1;

its teaching on holiness, 1;
its teaching on temperance, see under Guyon;
Platonic ethics in, 26, 39;
its allegorical scheme, 26;
its unity, 29, 30;
beauty of mind in, 33;
function of grace in, 62, 63.
“Fever, A,” 164.
Ficino, see under “Commentarium in Convivium.”
Fidelia, 3.
Fletcher, Giles, 100, 101, 102, 103.
Fletcher, Phineas, 83, 97.
Florimell, 66.
“Forerunners, The,” 90.
“Friendship in Absence,” 161.
Furor, 16.
“Garden of Adonis,” 213–216.

God, as lover of His own beauty, 68, 69;
as The Good, 70, 86;
union of soul with, 100–103;
as “Idea Beatificall,” 102;
as source of beauty, 109, 180;
as Creator, 110;
and the three Platonian hypostases, 172;
as king, 174, 175, 176;
as philosophical principle, 176–178.
Good, The, 169.
“Gorgias,” on temperance, 24.
Grace, doctrine of, its connection with ideal of holiness, 61–63;
its connection with ideal of chastity, 63, 64;
its connection with ideal of temperance, 62, 63;
represented by Arthur, 62.
Greville, Fulke, 138.
Guyon, his adventures, 13;
his struggles with wrath, 14–18;
his struggles with sensual desire, 18–21;
character of his life, 24;
his praise of beauty of mind, 34;
his relation to Arthur, 62.
Habington, William, 147.

Heaven, 92.
Henrietta, Maria, 156, 157.
Herbert, Edward, Lord Herbert, 146, 159.
Herbert, George, 71, 89, 90, 93.
Heywood, Thomas, 156.
Holiness, Platonism and, 10;
its connection with the doctrine of grace, 62.
Holy Spirit, identified with Psyche, 170.
Howell, James, 155, 156, 157.
Hudibras, 22.
“Hymn of Fairest Fair, An,” 174–179, 183, 184.
“Hymne in Honour of Beautie, An,” 106, 109–117, 118.
“Hymne in Honour of Love, An,” 105, 107, 108, 118, 121.
“Hymne of Heavenly Love,” 68, 69, 73, 74, 75, 76, 95, 96.
“Hymne of Heavenly Beautie,” 185, 186.
“Hymne of True Happiness, An,” 86.
Hypostases, the Plotinian, 167, 176, 177.
Idea, Platonic notion of, 95;

connection with mysticism, 95, 101;
term used as title, 138.
“Idea Beatificall,” 102.
Intellect, The, identified with God, 175;
defined, 170, 180;
identified with Christ, 170.
Intelligible world, 77, 78, 81.
“In the Glorious Epiphanie of Our Lord God,” 97–99.
Ithuriel, 59.
Jonson, 122, 123, 151.

“Jordan,” 91.
Joy, in religious experience, 85;
and the beatific vision, 88.
Justice, Spenser’s conception of, 27, 28;
Plato’s conception, 28;
identical with temperance, 28.
Lady, The, in “Comus,” effect of spells of Comus on, 53;

her response to Comus, 54;
her conception of chastity, 54, 56.
Linche, Richard, 138.
Love, nativity of the god of, 120, 121, 122, 123, 124;
treatment of, in Donne, 149–151, 152–155;
in Drayton, 125, 126;
in Ficino, 107, 108, 115, 116;
in Habington, 147;
in Jonson, 123, 124, 125, 151;
in Milton, 41, 47, 82, 83;
in Plato, 34, 35, 120;
in Spenser, 34, 107–109, 115–118, 120, 122;
in Vaughan, 132, 133.
earthly, 83, 88.
heavenly, defined, 67, 72, 73, 84;
in Spenser, 75;
in Drummond, 81.
Platonic, its rise at court, 155, 156;
defined, 155, 156, 160;
ridiculed, 156, 157;
described Lord Herbert, 159, 160;
in Stanley, 158;
in Vaughan, 158;
its immorality, 158.
“Love,” 71.
“Love Freed from Ignorance and Folly,” 124, 125.
Lovelace, Richard, 161.
“Love’s Growth,” 152.
“Love’s Innocence,” 158.
“Love’s Mistress or the Queen’s Masque,” 156.
“Love’s Triumph through Callipolis,” 123, 124.
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.
“Negative Love,” 153–155.

“Nicomachean Ethics,” 22.
“No Platonique Love,” 162.
Norris, John, 86, 87, 89, 157.
One, The, 153, 169, 179.

“Orlando Furioso,” 38, 39.
Palmer, The, 17, 20, 21, 25.

“Paradise Lost,” 49, 58.
“Paradise Regained,” 42.
Pastorella, 46.
Petrarchism, defined, 105, 126;
influence of Platonism on, 105, 127.
Perissa, 22.
“Phædo,” on the function of philosophy, 8;
on effect of sense knowledge on soul, 48, 49 n. 1, 50, 55;
on the super-sensible world, 77, 78, 81 n. 1.
Phædria, 18.
“Phædrus,” on the beauty of wisdom, 4, 127;
on love, 8, 34, 35;
on beauty of virtue, 4, 127, 220;
on reminiscence, 11, 57, 203, 206, 207;
on sight of true beauty, 57;
on beauty, 220.
“Philebus,” on goodness, 61.
“Piscatorie Eclogues,” 83.
“Platonick, The,” 161.
“Platonic Elegy, A,” 143.
“Platonic Love” (Ayres’s), 161;
(Aytoun’s) 161.
“Platonick Love (Cowley’s), 162;
(Cleveland’s), 162;
(Lord Herbert’s), 159, 160.
“Platonic Lovers,” 156.
Platonism, fundamental principle of, 1, 3, 30;
its relation to ideal of holiness, 10;
its part in religious experience, 12, 71, 72, 85, 91, 92, 181, 183;
its relation to ethics of “Faerie Queene,” 26, 30;
its connection with doctrine of grace, 61;
its relation to doctrine of heavenly love, 67, 68;
its appeal to sense of beauty, 85;
its influence on erotic mysticism, 95–104;
its influence on love poetry, 104;
its relation to morality of love, 136, 137, 138;
its influence on discussion of love, 140, 141;
its three hypostases, 167;
its effect on theology, 167;
its attraction for the religious mind, 183, 193, 194, 201, 202, 216,
218, 219.
Italian, appeal of, to Spenser, 117, 118;
its æsthetic theory, 139;
its debt to Plotinus, 140;
its relation to Plato, 140;
its debt to Ficino, 140.
Plotinus, “Enneads” of, on the intelligible world, 77;
on The Good, 153, 154, 169, 181;
on the hypostases, 167;
on soul, 170, 171;
on intellect, 170;
on The One, 176, 177, 181, 194;
on inter-relation of the hypostases, 177, 180;
on matter, 179, 211;
on immortality, 187.
“Prayer for Mankind,” 184.
“Prospect,” 87.
“Psychathanasia,” idea of creation in, 70;
mysticism in, 99, 194;
immortality in, 187–193;
The One in, 197, 198, 199.
Psyche, 170, 171, 172.
“Psychozoia,” idea of the Trinity in, 168;
religious feeling in, 183.
“Pure Platonicke,” 159.
“Purple Island,” 97.
Pyrochles, 17, 18.
Randolph, Thomas, 143.

Red Cross Knight, his sight of Una’s beauty, 7, 9, 10, 11;
on Mount of Contemplation, 8, 9, 10, 11, 62;
training in House of Holiness, 10;
slandered, 14;
character of, 15;
Arthur’s relation to, 62.
“Republic,” on the good, 8;
on temperance, 13, 14, 28;
on principles within soul, 13;
on justice, 28;
on imitative art, 91 n. 1;
on truth and opinion, 125.
Reminiscence, theory of, in Vaughan, 203, 204–206, 207, 208;
in Plato, 203, 206.
“Retreat, The,” 203, 204, 206.
Ruggiero, 39.
Sans Loy, 23.

Satan, his love of beauty, 42, 43–46;
his sight of Eden, 43;
contemplating Adam and Eve, 43, 44, 46;
his regret for lost beauty, 58, 59, 60.
Satyrane, 3.
Sedley, Charles, 161.
“Seraphick Love,” 89.
Shakespeare, 128, 129, 134, 135.
Sidney, Algernon, 157.
Sidney, Philip, on beauty of virtue, 66;
on heavenly love, 84, 85;
on Stella and virtue, 127;
on Plato, 137.
Song—“If you refuse me once,” 161.
Song II—“It Autumn was, and on our hemisphere,” 76, 77, 79–81, 81
n. 1.
Song, “To a Lady,” 158.
Song, “To Amoret,” 133.
σοφια, Una identified with, 2.
σωφροσύνη, Plato’s idea of, 12.
Soul, three principles in the, 13;
effect of sensuous experience on the, 48, 50, 51;
its self-sufficiency, 61;
its union with God, 89, 100–103;
its formative energy, 113, 114;
union of, in love, 141, 143;
defined, 187, 192, 193;
where found, 188;
a self-moving substance, 188;
immortality of, 189, 190, 191, 192;
its identity after death, 195,196;
universal, identified with woman, 164;
defined, 170, 171.
“Soul’s Joy,” 144.
Spenser, Edmund, Platonism in, 3, 5, 7, 21, 22, 31, 35, 39, 117, 218,
220;
his idea of beauty, 4, 32, 33, 65, 66;
his idea of justice, 27;
his idea of temperance, 23, 24, 25;
his idea of virtue, 27, 29;
his idea of a gentleman, 29;
his idea of love, 31, 108;
his idea of heavenly love, 75;
his æsthetics, 109–117;
identifies beloved with idea of beauty, 130, 136;
on his hymns, 139;
his idea of matter, 212;
his world weariness, 216.
Stanley, Thomas, 158.
Suckling, John, 161.
Sylvanus, 6.
“Symposium,” on wisdom, 8;
its dialectic 8, 75;
on beauty of mind, 31;
on nativity of love, 68, 120;
interpreted by Ficino, 107, 140;
on generation and immortality, 119, 120.
“Teares on the Death of Mœliades,” 87, 88.

Temperance, Plato’s idea of, 12, 13, 14, 23;
Spenser’s idea of, 23, 24, 25;
and justice, 28;
connection with heavenly grace, 62.
“Temple of Love,” 156.
“Theologia Germanica,” 168.
θνμός, 13.
“Timæus,” on creation, 70;
on Creator, 110;
on flux, 211, 212.
“To Amoret. Walking in a Starry Evening,” 132.
“To Cinthia, Converted,” 159.
“To Cinthia, coying it,” 159.
“To Cloris, a Rapture,” 161.
“To Lucasta, Going beyond the Seas,” 161.
“To my Mistress in Absence,” 161.
“To the Countess of Huntingdon,” 149, 151, 163.
“To the Platonicke Pretender,” 159.
“To the World. The Perfection of Love,” 147.
Trinity, The, identified with Plotinian hypostases, 168–174;
its unity, 176.
Una, identified with σοφία, and ἀρετή, 2;

identified with truth, 2, 3;
guides Red Cross Knight to Fidelia’s school, 3;
presented as true beauty, 3–10.
“Urania,” 88.
“Valediction Forbidding Mourning,” 145.

Vaughan, Henry, his idea of love, 132, 133, 148;
his idea of Platonic love, 158;
his idea of preëxistence, 203.
Virtue, Plato’s idea of, 27;
a manifold of graces, 27;

Contemporaneous Event Studies in Corporate Finance: Methods, Critiques and Robust Alternative Approaches 1st Ed. Edition Jau-Lian Jeng

Uploaded by

Copyright:

Available Formats

Contemporaneous Event Studies in Corporate Finance: Methods, Critiques and Robust Alternative Approaches 1st Ed. Edition Jau-Lian Jeng

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Contemporaneous Event Studies in Corporate Finance: Methods, Critiques and Robust Alternative Approaches 1st Ed. Edition Jau-Lian Jeng

Uploaded by

Copyright:

Available Formats

Download More ebooks [PDF]. Format PDF ebook download PDF KINDLE.

Full download ebooks at ebookmass.com

Contemporaneous Event Studies in

Download More ebooks from https://ebookmass.com

Alternative Approaches in Conflict Resolution 1st

Analysing, Planning and Valuing Private Firms: New

Numerical Methods and Optimization in Finance [2nd ed.]

Corporate Governance of Chinese Multinational

Corporate Finance - European Edition David Hillier

Principles of Corporate Finance (Mcgraw-hill/Irwin

Principles of corporate finance Brealey

Fundamentals of Corporate Finance (Berk, DeMarzo &

ISBN 978-3-030-53808-8 ISBN 978-3-030-53809-5 (eBook)

The applications of event study methodology for empirical corporate fi-

Los Angeles, USA Jau-Lian Jeng

The purpose of the event studies in corporate finance is to identify the

capital market. which is considered as a single corporate event may per-

The conventional statistics with cumulative abnormal returns may ac-

studies should be based on the clear dichotomy between systematic and

where K 0 , α, c, p are constants, and N (ds, d M) = 1 if an infinitesimal

More specifically, let

where λ0 (t) : R → R+ is the base intensity and v : R+ → R+

Hence, the intensity of the crash can be specified as the function of

after the announcement of the events. Either the conventional method

Acharya, V.V., L.H. Pedersen, T. Philippon, and M. Richardson. 2016. Mea-

Part I The Conventional Approach

1 Popular Methods for Event Studies in Corporate

Part II Alternative Approach for the Contemporaneous

2 Assessments of Normal Returns 55

3 Occupation Time Statistics—The Intensity of Events 91

4 Monitoring Tests and the Time of Duration 147

5 Sequential Monitoring for Corporate Events in Using

6 Real-Time Applications of Monitoring and Empirical

Concluding Remark 223

Table 6.1 The Test Statistics Z (h) 216

© The Author(s) 2020 3

1.2 Identification of Event

In particular, other than the ad hoc methods in determining the event

cross-sectional asset pricing models. In particular, the cross-sectional

1.2.1 Event Dependency and Structural Change

rit = X t γi1 + εit , t < t− , t > t+

where X t is the t-th date observations of k explanatory variables (or fac-

for all i = 1, 2, . . . , n. By stacking up the i-th security return across time

where ε p ∼ N (0, σ p2 I p ); p = 1, 2, and I p stands for an identity matrix

return can be written as

where e p , p = 1, 2 represents the residual vectors from fitted regressions.

will have a chi-square distribution with degree of freedom equal to the

where T1 and T2 are the number of observations of i-th security returns

is an F -statistic with degree of freedom (k, T1 + T2 − 2k) . In particular,

as stated earlier, the structural changes can be the consequences of other

where ∗ and  are diagonal covariance matrices with respective elements

where Mw = I − W (W  −1 W )−1 W  −1 , and Mx = I − X

L(σ1 , σ2 , γ1 , γ2 ) ∞ − (T − E)lnσ1 − Elnσ2

Hence, the maximum likelihood estimators for the parameters will be

where E = (b − a − 1). Upon substitution of these estimators, the log-

L(a, b)∞ − (T − E)ln σ̂1 − (E)ln σ̂2 . (1.13)

1.2.2 Random Event Dates (Ball and Torous 1988)

θt = 1, if the event occurs on day t,

pt = Pr ob[θt = 1, θs = 0, s ∈ (−c, . . . , 0, . . . , c), s = t]. (1.14)

where ∗ and are diagonal covariance matrices with respective elements

where Mw = I − W (W −1 W )−1 W −1 , and Mx = I − X

pt = Pr ob[θt = 1, θs = 0, s ∈ (−c, . . . , 0, . . . , c), s = t]. (1.14)

pt = Pr ob[θt = 1, θs = 0, s ∈ (−c, . . . , 0, . . . , c), s = t]

where the notation

Covi (bi |X i ) = σi2 (X i X i )−1 . (1.18)

(g1 − g2 ) ∼ N (0, σ12 (X 1 X 1 )−1 + σ22 (X 2 X 2 )−1 ),