10. Event Studies
Transcript of 10. Event Studies
-
8/7/2019 10. Event Studies
1/4
10 Event Studies
Reference: Bodie, Kane, and Marcus (2005) 12.3 or Copeland, Weston, and Shastri
(2005) 11
Reference (advanced): Campbell, Lo, and MacKinlay (1997) 4
More advanced material is denoted by a star (). It is not required reading.
10.1 Basic Structure of Event Studies
The idea of an event study is to study the effect (on stock prices or returns) of a special
event by using a cross-section of such events. For instance, what is the effect of a stock
split announcement on the share price? Other events could be debt issues, mergers and
acquisitions, earnings announcements, or monetary policy moves.
The event is typically assumed to be a discrete variable. For instance, it could be a
merger or not or if themonetary policy surprise was positive (lowerinterestthan expected)
or not. The basic approach is then to study what happens to the returns of those assets
that have such an event.Only news should move the asset price, so it is often necessary to explicitly model
the previous expectations to define the event. For earnings, the event is typically taken to
be the earnings announcement minus (some average of) analysts forecast. Similarly, for
monetary policy moves, the event could be specified as the interest rate decision minus
previous forward rates (as a measure of previous expectations).
The abnormal return of asset i in period t is
ui;t D Ri;t Rnormali;t ; (10.1)
where Rit is the actual return and the last term is the normal return (which may differ
across assetsand time). The definition of the normal returnis discussed in detail in Section
10.2. These returns could be nominal returns, but more likely (at least for slightly longer
horizons) real returns or excess returns.
Suppose we have a sample of n such events (assets). To keep the notation (reason-
160
timefirm 1 firm 2
1011 0 1
Figure 10.1: Event days and windows
ably) simple, we normalize the time so period 0 is the time of the event. Clearly the
actual calendar time of the events for assets i and j are likely to differ, but we shift the
time line for each asset individually so the time of the event is normalized to zero for
every asset.
The (cross-sectional) average abnormal return at the event time (time 0) is
Nu0 D .u1;0 C u2;0 C ::: C un;0/ =n DPn
iD1ui;0=n: (10.2)
To control for information leakage and slow price adjustment, the abnormal return is often
calculated for some time before and after the event: the event window (often 20 days
or so). For lead s (that is, s periods after the event time 0), the cross sectional average
abnormal return is
Nus D .u1;s C u2;s C ::: C un;s/ =n DPn
iD1ui;s=n: (10.3)
For instance, Nu2 is the average abnormal return two days after the event, and Nu1 is for
one day before the event.
The cumulative abnormal return (CAR) of asset i is simply the sum of the abnormal
return in (10.1) over some period around the event. It is often calculated from the be-
ginning of the event window. For instance, if the event window starts at w, then the
q-period (day?) car for asset i is
cari;q D ui;w C ui;wC1 C : : : C ui;wCq1: (10.4)
The cross sectional average of the q-period car is
carq D
car1;q C car2;q C ::: C carn;q
=n DPn
iD1cari;q=n: (10.5)
Example 10.1 (Abnormal returns for day around event, two firms) Suppose there are
two firms and the event window contains 1 day around the event day, and that the
161
-
8/7/2019 10. Event Studies
2/4
abnormal returns (in percent) are
Time Firm 1 Firm 2 Cross-sectional Average
1 0:2 0:1 0:05
0 1:0 2:0 1:5
1 0:1 0:3 0:2
We have the following cumulative returns
Time Firm 1 Firm 2 Cross-sectional Average
1 0:2 0:1 0:05
0 1:2 1:9 1:55
1 1:3 2:2 1:75
10.2 Models of Normal Returns
This section summarizes the most common ways of calculating the normal return in
(10.1). The parameters in these models are typically estimated on a recent sample, the
estimation window, that ends before the event window. In this way, the estimated be-
haviour of the normal return should be unaffected by the event. It is almost always as-
sumed that the event is exogenous in the sense that it is not due to the movements in the
asset price during either the estimation window or the event window. This allows us to
get a clean estimate of the normal return.
The constant mean return model assumes that the return of asset i fluctuates randomly
around some mean i
Ri;t D i C i;t with E.i;t / D Cov.i;t ; i;ts/ D 0: (10.6)
This mean is estimated by the sample average (during the estimation window). The nor-
mal return in (10.1) is then the estimated mean. Oi so the abnormal return becomes Oi;t .
The market model is a linear regression of the return of asset i on the market return
Ri;t D iCiRm;tC"it with E."i;t / D Cov."i;t ; "i;ts/ D Cov."i;t ; Rm;t/ D 0: (10.7)
Notice that we typically do not impose the CAPM restrictions on the intercept in (10.7).
The normal return in (10.1) is then calculated by combining the regression coefficients
162
time
event windowestimation window(for normal return)
firm i
Figure 10.2: Event and estimation windows
with the actual market return as O i C OiRm;t , so the the abnormal return becomes O"it .
Recently, the market model has increasingly been replaced by a multi-factor model
which uses several regressors instead of only the market return. For instance, Fama and
French (1993) argue that (10.7) needs to be augmented by a portfolio that captures the
different returns of small and large firms and also by a portfolio that captures the different
returns of firms with high and low book-to-market ratios.
Finally, yet another approach is to construct a normal return as the actual return on
assets which are very similar to the asset with an event. For instance, if asset i is a
small manufacturing firm (with an event), then the normal return could be calculated as
the actual return for other small manufacturing firms (without events). In this case, the
abnormal return becomes the difference between the actual return and the return on the
matching portfolio. This type of matching portfolio is becoming increasingly popular.
All the methods discussed here try to take into account the risk premium on the asset.
It is captured by the mean in the constant mean mode, the beta in the market model, and
by the way the matching portfolio is constructed. However, sometimes there is no data
in the estimation window, for instance at IPOs since there is then no return data before
the event date. The typical approach is then to use the actual market return as the normal
returnthat is, to use (10.7) but assuming that i D 0 and i D 1. Clearly, this does not
account for the risk premium on asset i , and is therefore a fairly rough guide.
Apart from accounting for the risk premium, does the choice of the model of the
normal return matter a lot? Yes, but only if the model produces a higher coefficient ofdetermination (R2) than competing models. In that case, the variance of the abnormal
return is smaller for the market model which the test more precise (see Section 10.3 for
a discussion of how the variance of the abnormal return affects the variance of the test
statistic). To illustrate this, consider the market model (10.7). Under the null hypothesis
that the event has no effect on the return, the abnormal return would be just the residual
163
-
8/7/2019 10. Event Studies
3/4
in the regression (10.7). It has the variance (assuming we know the model parameters)
Var.ui;t / D Var."it / D .1R2/ Var.Ri;t /; (10.8)
where R2 is the coefficient of determination of the regression (10.7).
Proof. (of (10.8)) From (10.7) we have
Var.Ri;t / D 2i Var.Rm;t /CVar."it /:
We therefore get
Var."it / D Var.Ri;t / 2i Var.Rm;t/
D Var.Ri;t / Cov.Ri;t ; Rm;t/2
= Var.Rm;t /
D Var.Ri;t / Corr.Ri;t ; Rm;t/2 Var.Ri;t /
D .1R2/ Var.Ri;t /:
The second equality follows from the fact that i D Cov.Ri;t ; Rm;t/= Var.Rm;t /, the
third equality from multiplying and dividing the last term by Var .Ri;t / and using the
definition of the correlation, and the fourth equality from the fact that the coefficient
of determination in a simple regression equals the squared correlation of the dependent
variable and the regressor.This variance is crucial for testing the hypothesis of no abnormal returns: the smaller
is the variance, the easier it is to reject a false null hypothesis (see Section 10.3). The
constant mean model has R2 D 0, so the market model could potentially give a much
smaller variance. If the market model has R2 D 0:75, then the standard deviation of
the abnormal return is only half that of the constant mean model. More realistically,
R2 might be 0.43 (or less), so the market model gives a 25% decrease in the standard
deviation, which is not a whole lot. Experience with multi-factor models also suggest that
they give relatively small improvements of the R2 compared to the market model. For
these reasons, and for reasons of convenience, the market model is still the dominating
model of normal returns.
High frequency data can be very helpful, provided the time of the event is known.
High frequency data effectively allows us to decrease the volatility of the abnormal return
since it filters out irrelevant (for the event study) shocks to the return while still capturing
the effect of the event.
164
10.3 Testing the Abnormal Return
In testing if the abnormal return is different from zero, there are two sources of sampling
uncertainty. First, the parameters of the normal return are uncertain. Second, even if
we knew the normal return for sure, the actual returns are random variablesand they
will always deviate from their population mean in any finite sample. The first source
of uncertainty is likely to be much smaller than the secondprovided the estimation
window is much longer than the event window. This is the typical situation, so the rest of
the discussion will focus on the second source of uncertainty.
It is typically assumed that the abnormal returns are uncorrelated across time and
across assets. The first assumption is motivated by the very low autocorrelation of returns.
The second assumption makes a lot of sense if the events are not overlapping in time, so
that the event of assets i and j happen at different (calendar) times. In contrast, if the
events happen at the same time, the cross-correlation must be handled somehow. This is,
for instance, the case if the events are macroeconomic announcements or monetary policy
moves. An easy way to handle such synchronized events is to form portfolios of those
assets that share the event timeand then only use portfolios with non-overlapping events
in the cross-sectional study. For the rest of this section we assume no autocorrelation or
cross correlation.
Let 2i D Var.ui;t / be the variance of the abnormal return of asset i . The variance of
the cross-sectional (across the n assets) average, Nus in (10.3), is then
Var. Nus/ D
21 C 22 C ::: C
2n
=n2 D
PniD1
2i =n
2; (10.9)
since all covariances are assumed to be zero. In a large sample (where the asymptotic
normality of a sample average starts to kick in), we can therefore use a t -test since
Nus= Std. Nus/ !d N.0;1/: (10.10)
The cumulative abnormal return over q period, cari;q , can also be tested with a t -test.
Since the returns are assumed to have no autocorrelation the variance of the car i;q
Var.cari;q/ D q2i : (10.11)
This variance is increasing in q since we are considering cumulative returns (not the time
165
-
8/7/2019 10. Event Studies
4/4
average of returns).
The cross-sectional average cari;q is then (similarly to (10.9))
Var.carq/ D
q21 C q22 C ::: C q
2n
=n2 D q
PniD1
2i =n
2; (10.12)
if the abnormal returns are uncorrelated across time and assets.
Figures 4.2ab in Campbell, Lo, and MacKinlay (1997) provide a nice example of an
event study (based on the effect of earnings announcements).
Example 10.2 (Variances of abnormal returns) If the standard deviations of the daily
abnormal returns of the two firms in Example 10.1 are 1 D 0:1 and and2 D 0:2 , then
we have the following variances for the abnormal returns at different days
Time Firm 1 Firm 2 Cross-sectional Average
1 0:12 0:22
0:12 C 0:22
=4
0 0:12 0:22
0:12 C 0:22
=4
1 0:12 0:22
0:12 C 0:22
=4
Similarly, the variances for the cumulative abnormal returns are
Time Firm 1 Firm 2 Cross-sectional Average
1 0:12
0:22
0:12
C 0:22
=40 2 0:12 2 0:22 2
0:12 C 0:22
=4
1 3 0:12 3 0:22 3
0:12 C 0:22
=4
Example 10.3 (Tests of abnormal returns) By dividing the numbers in Example 10.1 by
the square root of the numbers in Example 10.2 (that is, the standard deviations) we get
the test statistics for the abnormal returns
Time Firm 1 Firm 2 Cross-sectional Average
1 2
0:5 0:4
0 10 10 13
1 1 1:5 1:8
166
Similarly, the variances for the cumulative abnormal returns we have
Time Firm 1 Firm 2 Cross-sectional Average
1 2 0:5 0:4
0 8:5 6:7 9:8
1 7:5 6:4 9:0
(I have not double checked the calculations...)
10.4 Quantitative Events
Some events are not easily classified as discrete variables. For instance, the effect of
positive earnings surprise is likely to depend on how large the surprise isnot just if there
was a positive surprise. This can be studied by regressing the abnormal return (typically
the cumulate abnormal return) on the value of the event (xi )
cari;q D a C bxi C i : (10.13)
The slope coefficient is then a measure of how much the cumulative abnormal return react
to a change of one unit of xi .
Bibliography
Bodie, Z., A. Kane, and A. J. Marcus, 2005, Investments, McGraw-Hill, Boston, 6th edn.
Campbell, J. Y., A. W. Lo, and A. C. MacKinlay, 1997, The Econometrics of Financial
Markets, Princeton University Press, Princeton, New Jersey.
Copeland, T. E., J. F. Weston, and K. Shastri, 2005, Financial Theory and Corporate
Policy, Pearson Education, 4 edn.
Fama, E. F., and K. R. French, 1993, Common Risk Factors in the Returns on Stocks
and Bonds, Journal of Financial Economics, 33, 356.
167