Content uploaded by Arnold R. Cowan
Author content
All content in this area was uploaded by Arnold R. Cowan on Mar 02, 2020
Content may be subject to copyright.
1
Goodbye to the Patell Test in Eventus
Arnold R. Cowan
CEO, Cowan Research LC
Wells Fargo Professor of Finance, Iowa State University
This draft: March 1, 20200
It’s not really goodbye. You still can run the Patell (1976) test in Eventus® software if you need it.
The longtime default parametric test for basic two-step event studies is only moving out of the
spotlight. The new default now in place for most users is the standardized cross-sectional test by
Boehmer, Musumeci and Poulsen (BMP; 1991); for a smaller number of users, the change may
await the next update or version upgrade. The BMP standardized cross-sectional test is an en-
hanced version of the Patell test.1
This note reviews some of the research that informed our decision to take the rare step of chang-
ing a default setting and provides details about the change.
Relevant Research
The standardized abnormal return test of the null hypothesis that the mean abnormal return is
zero, derived by Patell (1976), has important strengths. Brown and Warner (BW; 1980, 1985)
report simulation evidence that the test is well specified in random samples of actual security
returns. Further, they show that the Patell test greatly improves power to detect an abnormal
return (artificially induced for the simulations) by making use of firm-specific variance estimates.
However, BW also report that a variance increase on the event date can seriously bias the Patell
test. Increased return variances around events have been reported at least since Beaver (1968).
BMP develop an extension of the Patell test, the standardized cross-sectional test, which brings
in cross-sectional variance information to correct for variance increases. BMP provide BW-type
simulation evidence that the standardized cross-sectional test is robust to event-date variance
increases. Focusing on the potential power-degrading effects of outlying estimation-period ob-
servations with high trading volume, Graham, Pirie and Powell (1996) report simulation
evidence that the standardized cross-sectional test provides the best power of three tests studied,
while holding the correct test size (true-null rejection rate). Higgins and Peterson (1998) conduct
simulations and use empirical distribution functions to equalize power under the null hypothesis
across tests. Their results support the overall superiority of the BMP standardized cross-sectional
test.
1 The Patell test receives mentions in the literature under other names such as standardized abnormal return test
and test assuming cross-sectional independence, and sometimes is attributed to various other authors including for
example Brown and Warner (1980, 1985), Dodd and Warner (1983) and Mikkelson and Partch (1988).
2
Harrington and Shrider (2007) show analytically that under plausible assumptions, cross-sec-
tional differences in event-induced abnormal returns cause the event-day variance to exceed the
variance of estimation-period residuals. This effect is separate from a potential transient or long-
term shock to the variance of the normal return-generating process. Harrington and Shrider also
provide a more rigorous analytical foundation for the BMP standardized cross-sectional test. The
authors report additional simulation evidence of heteroskedasticity-related biases of the Patell
test that the standardized cross-sectional test avoids.
Campbell, Cowan and Salotti (2015) conduct BW-type simulations of multi-country event stud-
ies and report that the standardized cross-sectional test performs well under a variety of
conditions in global samples. However, it is less powerful than nonparametric tests examined,
and it tends to reject a true null hypothesis too often in single-country non-U.S. samples and
when the stock experiencing a firm-specific event is a dominant member of a country index. The
authors recommended using the standardized cross-sectional test in conjunction with a nonpar-
ametric test for international event studies.
Marks and Musumeci (2017) argue that due to slow convergence of the sample mean to a normal
distribution, an event-induced variance shift is not necessary for the Patell test to be biased. They
conduct simulations with larger samples and far more replications than previous studies and find
over-rejection of true null hypotheses by the Patell test but. In contrast, the standardized cross-
sectional test performs nominally. Marks and Musumeci endorse Harrington and Shrider’s rec-
ommendation that researchers always use the BMP standardized cross-sectional test BMP in
preference to the Patell test.
The Patell and Standardized Cross-Sectional Tests in Eventus
Starting with the first commercial version in 1989, Eventus software provided the Patell test by
default. In 2017, we decided that changing the default parametric test was the responsible deci-
sion in light of the weight of research evidence published over more than a quarter century, the
admonition of Karolyi (2011) that the academic finance profession has room for improvement in
taking the “con out of econometrics”2, the warnings of Harvey (2017) concerning “p-hacking”,
and our observation from support interactions that it is not uncommon for users to accept the
default test as standard.
The standardized cross-sectional test has been an available option in Eventus for approximately
two decades; users have been able to select it by adding the StdCSect option to the EvtStudy
statement. Going forward, if one does not select a specific parametric test or a non-two-step
method (such as calendar-time portfolio regression or Ibbotson RATS), Eventus will provide the
standardized cross-sectional test.
2 The phrase originated with Leamer (1983).
3
The Eventus implementation of the BMP test applied to multi-day or multi-month windows in-
corporates a correction for the serial correlation that occurs when using the same coefficient
estimates for each daily or monthly abnormal return (Cowan, 1993) .This correction also has
been available for the Patell test in Eventus for many years through the use of the EvtStudy state-
ment option Serial. Serial has been, and still is, activated by default for the BMP standardized
cross-sectional test from the first release of Eventus that offered the test. It always has been, and
continues to be, necessary to specify Serial explicitly if one wants to use the correction with the
original Patell test.
Users often recycle and adapt old programs of their own or those of others for new studies. We
have seen many Eventus user programs that specify the Patell option explicitly, typically to select
this test along with other parametric tests in the same run. Third-party tools to assist Eventus
uses may also specify Patell. To ensure that our change of default is not rendered moot by un-
conscious habit or code inertia, we have taken another rare step and changed the meaning of the
Patell option. Instead of invoking the original Patell test, the simple EvtStudy statement option
specification Patell now selects (or for some users will select) the standardized cross-sectional
test, which one could think of as Patell 2.0. Users still can select the Patell 1.0 test by using the
new EvtStudy statement option specification Patell=Original (case does not matter).
A side benefit of the changes is that some users will encounter less confusion when seeking a
Patell-type test with a multi-factor benchmark model such as the Fama-French three-, four- and
five-factor models in two-step mode. Eventus offers the standardized cross-sectional test but not
the original Patell test with multi-factor models.
Users also may want to be aware of another available EvtStudy statement option specification,
StdCSect=Adjusted, which selects the adjusted standardized cross-sectional test of Kolari and
Pynnönen (2010). This test (perhaps Patell 3.0) retains the benefits of the BMP standardized
cross-sectional test while further accounting for cross-correlation, which especially arises when
all firms experience the event on the same calendar date. 3
The original Patell, the standardized cross-sectional test and the adjusted standardized cross-
sectional test must be run separately; it is not possible to select two or more in the same run.
A specific suggestion of Karolyi (2011) for improving the credibility of conclusions from empiri-
cal financial economics is the use of simulation methods, of which bootstraps are an example. In
Eventus, use the EvtStudy statement option Bootstrap (or just Boot) to compute nonparametric-
bootstrap p-values of parametric tests, or Bootstrap=Wild to compute wild bootstrap p-values.
3 WRDS web queries that formerly defaulted to the Patell option now default to the Patell=Original option. This
design choice was made by WRDS. We recommend that users of these query pages select the StdCSect option or
add a line containing only ?*#! StdCSect=Adjusted to the beginning of the request file to select the adjusted
standardized cross-sectional test.
4
Abnormal returns and standardized abnormal returns at the security-event (“firm”) level do not
vary across the three tests. The tests differ only in how they aggregate security-event results to
produce an overall test statistic.
Summary
Recent and older research shows that the Boehmer-Musumeci-Poulsen (BMP) standardized
cross-sectional test produces more reliable inferences than the original Patell test on which it is
built. Accordingly, we decided to make the BMP standardized cross-sectional test the new de-
fault parametric test in Eventus for two-step event studies. Users are free to select the original
Patell test or the newer Kolari-Pynnönen (KP) adjusted standardized cross-sectional test using
appropriate option specifications. Optional bootstrap p-values also can be selected for the vari-
ous tests.
The following table shows how to select various test choices in Eventus if the new defaults are
active, as they now are for the majority of users.
To Select This:
A
dd This (in Upper, Lower or Mixed
Case) Anywhere Among the Options of
the EvtStudy Statement, or Specify
One Option Per Line at the Beginning of
the Request File on Lines Starting with
the Four Characters ?*#!:
Standardized cross-sectional test (BMP) None or StdCSect or Patell
A
djusted standardized cross-sectional test (KP) StdCSect=Adjusted
Original Patell test Patell=Original
Original Patell test plus correction for serial correla-
tion in multi-day or multi-month windows
Patell=Original Serial
Standardized cross-sectional test (BMP) for the
Fama-French five-factor model (as an example of se-
lecting the test for any multi-factor model)
FamaFrench5 TwoStep or
FamaFrench5 TwoStep StdCSect
A
dd nonparametric bootstrap p-values for any of the
above tests
Bootstrap
A
dd wild bootstrap p-values for any of the above tests Bootstrap=Wild
Questions or comments on this note are welcome; you can contact us through our support web-
site, support.eventstudy.com. Detailed usage help is limited to verified authorized users.
Registering with your institutional e-mail address will help us quickly verify your eligibility.
5
References
Beaver, W.H., 1968. The information content of annual earnings announcements. Journal of Ac-
counting Research, pp.67-92. http://www.jstor.org/stable/2490070
Boehmer, E., Musumeci, J. and Poulsen, A.B., 1991. Event-study methodology under conditions
of event-induced variance. Journal of Financial Economics, 30(2), pp.253-272.
https://doi.org/10.1016/0304-405X(91)90032-F
Brown, S.J. and Warner, J.B., 1980. Measuring security price performance. Journal of Financial
Economics, 8(3), pp.205-258. https://doi.org/10.1016/0304-405X(80)90002-1
Brown, S.J. and Warner, J.B., 1985. Using daily stock returns: The case of event studies. Journal
of Financial Economics, 14(1), pp.3-31. https://doi.org/10.1016/0304-405X(85)90042-X
Campbell, C.J., Cowan, A.R. and Salotti, V., 2010. Multi-country event-study methods. Journal
of Banking & Finance, 34(12), pp.3078-3090. https://doi.org/10.1016/j.jbankfin.2010.07.016
Cowan, A.R., 1993. Tests for cumulative abnormal returns over long periods: Simulation evi-
dence. International Review of Financial Analysis, 2(1), pp.51-68. https://doi.org/10.1016/1057-
5219(93)90006-4
Dodd, P. and Warner, J.B., 1983. On corporate governance: A study of proxy contests. Journal of
Financial Economics, 11(1-4), pp.401-438. https://doi.org/10.1016/0304-405X(83)90018-1
Graham, A.S., Pirie, W.L. and Powell, W.A., 1996. Detecting abnormal returns using the market
model with pre‐tested data. Journal of Financial Research, 19(1), pp.21-40.
http://dx.doi.org/10.1111/j.1475-6803.1996.tb00582.x
Harrington, S.E. and Shrider, D.G., 2007. All events induce variance: Analyzing abnormal re-
turns when effects vary across firms. Journal of Financial and Quantitative Analysis, 42(1), pp.229-
256. https://doi.org/10.1017/S002210900000226X
Harvey, Campbell R., The scientific outlook in financial economics (July 17, 2017). Duke I&E
Research Paper No. 2017-05. Available at SSRN: https://ssrn.com/abstract=2893930 or
http://dx.doi.org/10.2139/ssrn.2893930
Higgins, E.J. and Peterson, D.R., 1998. The power of one and two sample t-statistics given event-
induced variance increases and nonnormal stock returns: A comparative study. Quarterly Journal
of Business and Economics, pp.27-49. http://www.jstor.org/stable/40473233
Karolyi, G.A., 2011. The ultimate irrelevance proposition in finance? Financial Review, 46(4),
pp.485-512. http://dx.doi.org/10.1111/j.1540-6288.2011.00309.x
6
Kolari, J.W. and Pynnönen, S., 2010. Event study testing with cross-sectional correlation of ab-
normal returns. Review of Financial Studies, 23(11), pp.3996-4025.
https://doi.org/10.1093/rfs/hhq072
Leamer, E.E., 1983. Let's take the con out of econometrics. American Economic Review, 73(1),
pp.31-43. http://www.jstor.org/stable/1803924
Mikkelson, W.H. and Partch, M.M., 1988. Withdrawn security offerings. Journal of Financial and
Quantitative Analysis, 23(2), pp.119-133. https://doi.org/10.2307/2330876
Patell, J.M., 1976. Corporate forecasts of earnings per share and stock price behavior: Empirical
test. Journal of Accounting Research, pp.246-276. http://www.jstor.org/stable/2490543
This document was last updated March 1, 2020.
