The informational content of insider trading
disclosures: Empirical results for the Polish
stock market
Henryk Gurgul *, Paweł Majdosz**
* Department of Applied Mathematics, University of Science and Technology,
ul. Al. Mickiewicza 30, 30-059 Krakow, Poland (e-mail: [email protected]) ** Department of Quantitative Methods, School of Economics and Computer Science,
ul. Św. Filipa 17, 31-150 Kraków, Poland (e-mail: [email protected])
Abstract In this paper we try to answer the question as to whether insider
trading disclosures convey valuable information to market participants, valuable
in the sense of the profitability of an investment strategy that faithfully mirrors
insider behaviour. Our interest in this subject is limited to the case of announce-
ments concerning insider transactions issued over a six-year period on the War-
saw Stock Exchange (WSE). Initially, we use event study methodology to check
whether insider trading disclosures are accompanied by a performance of stock
returns as well as trading volume. Two different models generating expected re-
turns (expected volume) are employed to verify the robustness of our results. The
first of these is the regime switching model, with the results then being recalcu-
lated by using a GARCH-type models which seem to be most useful for dealing
with some of the inconvenient statistical properties of stock return and trading
volume data. Afterwards, a technique based on the reference return strategies is
used to examine whether or not outsiders who imitate insider behaviour are able
to profit from it. The major findings are as follows: Firstly, announcements about
the sale of stocks by insiders convey no information to market participants. Sec-
ondly, a statistically significant market response to insider disclosures of pur-
chases of stocks in their own company can be observed in the three days prior to
the announcement release for both return as well as trading volume series, and
finally, outsiders who purchased stocks previously bought by insiders experience
negative returns whereas outsiders disposing of stocks previously sold by insid-
ers earned a return of 8.57% over the six-month period.
JEL Classification: G14
Keywords: Insider trading, Event study, Switching Regressions, GARCH model
We thank two anonymous referees for many valuable comments and suggestions. All remaining er-
rors are our own responsibility.
1 Introduction
Over recent years, we have been witnessing an increased number of theoretical
discussions devoted to insider activities on the stock market. This topic has been
considered from many different points of view. Economists, lawyers and philos-
ophers carry on disputing insiders’ rights to deal in stocks of their own firms.
Some of them set an important trend in the debate by researching the possible
implications of insider trading for insiders themselves, outsiders and other stock
market observers such as analysts. Many statements on this subject have a moral-
ising tone. It is indisputable that nobody should be able to profit from infor-
mation which is not available to other market participants or, even worse, from
spreading rumours concerning own company which then turn out to be untrue. In
this spirit, an insider trading prohibition is often argued for. On the other hand,
there exist circumstances under which insider trading is accompanied by benefits
for all market participants. For instance, in the case of a market crash (like that of
the autumn of 1987), only insiders fully realize that current stock prices do not
correctly reflect the real value of shares. For this reason, insiders are likely to
buy shares when other investors are, in general, prone to close their long posi-
tions. Without insider trading, panic selling of shares would get out of control.
In standard conditions however, public confidence in the financial market
would be undermined by unscrupulous use of insider knowledge. This is why in
many countries there are stringent government regulations against illegal insider
trading activities. Poland is one of these. According the Act on Trading in Finan-
cial Instruments, any information should be regarded as private concerning a fi-
nancial instrument, its issuer, or trading in a financial instrument which is yet
unpublished if it is possible that this information will influence the prices of a fi-
nancial instrument after its release. A notion of private information is then used
to determine who is an insider. Any person has this status with access to private
information due to being a member of the board as well as supervisory directors,
an employer or a shareholder. The list of insiders also includes auditors, solici-
tors, a brokers and any other person possessing private information, regardless of
its source. Defining the notion of an insider so widely requires so-called primary
insiders to be determined to whom especial legal restrictions are applied in order
to protect the interests of other investors. A primary insider is anyone who has
the closest relationships with the issuer of financial instruments, e.g. members of
the board and supervisory directors, auditors, solicitors, and so forth.
It should be emphasised that trading on the basis of private information is il-
legal in Poland and incurs serious penalties. However, with a few notable excep-
tions, sales and purchases of shares by insiders, even those classified as primary
ones, are generally allowable. There is only a requirement to notify about trans-
actions made by insiders. Such notification should take place no later than five
days after the transaction date if its total value exceeds 5,000 euro. For other
transactions the notification deadline is 31st January of the following year. As
mentioned above, over some periods, so-called ‘closed periods’, primary insiders
are not allowed sell and buy shares in the own companies. These periods are
mainly associated with the dates of publication of annual, 6-monthly as well as
quarterly reports and their lengths are equal to two months, one month and two
weeks immediately before publication, respectively.
In this paper, we do not adopt a judgemental role in such a difficult issue as to
whether insider trading is ethical. We are very reluctant to formulate advice or
suggestions addressed to insiders or any government whose target is to improve
market transparency by changing corporate behaviour and/or the legal system.
Instead, our interest is focused on (i) stock market response to announcements
about buying or selling shares by insiders, (ii) the profitability of investment
strategies based on the imitation of insider behaviour by buying shares which
were previously bought by insiders, and selling shares which were previously
sold by insiders.
The above-mentioned purpose of this contribution is accomplished by means
of an event study framework. This methodology is commonly employed in fi-
nancial and accounting research to investigate abnormal market performance as a
result of an event. Of prime importance as regards the accuracy of obtained re-
sults is the choice of an appropriate model to generate expected levels of any var-
iable under consideration. Therefore, tracking market behaviour in the context of
insider trading disclosures requires different econometric models of expected re-
turns and expected volume to be used, guaranteeing the robustness of findings
and correctness of conclusions. Two models are employed under this study to es-
timate the expected (non-disturbed by event) levels of returns and volume. The
Markov switching regression model enables the variance of variable under con-
sideration to change when the event occurs. This feature may appear to be par-
ticularly useful if it is impossible to eliminate all other (confounding) events tak-
ing place in the period over which stock returns and trading volume are
observed. However, the variance still remains unchanging under each regime. To
model changes in stock return (trading volume) in parallel with changes in its
variance the GARCH model is used.
Our results are as follows: Firstly, announcements of sale of stocks by insiders
bring no information besides that already incorporated in stock prices. Secondly,
a statistically significant market response to insider disclosures of purchases of
stocks their own company can be observed in the three days prior to the an-
nouncement release for both return as well as trading volume series. In parallel
with the analysis of how the market reacts to insider trading disclosures, we also
examine whether or not an investment strategy relying on copying of insider be-
haviour is profit-making. It turned that outsiders who purchased stocks previous-
ly bought by insiders experience negative returns whereas outsiders disposing
stocks previously sold by insiders earned a return of 8.57% over the six-month
period.
The remainder of this article proceeds as follows. In Section 2, a rough survey
of existing contributions which deal with the subject is presented. Section 3 out-
lines our methodology which is then used to identify market responses to an-
nouncements concerning insider trading. A brief data description as well as some
statistics of insider trading on the WSE over the analysed period are reported in
Section 4. Section 5 contains our empirical results and the final section con-
cludes the paper.
2 A brief literature review and main hypotheses
Empirical studies of insider trading can generally be classified into two catego-
ries. Some studies focus on the event of insider trading itself, where the central
issue is insider trading profitability and stock market efficiency. Studies of the
second type usually investigate insider trading behaviour in conjunction with an-
other corporate event, especially buybacks and stock splits.
It is characteristic that the previous contributions do not agree about the prof-
itability of insider trading and the informative content of insider trading. Empiri-
cal evidence provided by Kerr (1980), Lin and Howe (1990) as well as Holder-
ness and Sheehan (1985) seem to support the efficient market hypothesis (EMH)
in its strong form. It is worth remembering here that EMH was formulated by
Fama in 1970 and then extended in the following years (see Fama, 1970, 1991).
The above mentioned strong-form of EMH states that stock prices incorporate all
existing information, i.e. not only public information but also that unpublished,
regardless of information sources. On the other hand, Jaffe (1974), Seyhun
(1986, 1988), Madura and Wiant (1995) showed that insider activity is accompa-
nied by abnormal market performance (a significant positive effect on stock
prices). Raad and Wu (1995) report that insiders increase their buying activities
and decrease their selling activities before stock repurchase offers. While stock-
holders of firms with insider net selling activity earn positive excess returns,
those of firms with insider net buying activities earn larger and more significant
abnormal returns.
More recently, Seyhun (1998) documents the informative value of insider
trading. Seyhun also concludes that transaction volume and the position of the
insider within the company are positively correlated to relative performance. The
results show, on the other hand, that the relative performance negatively depends
on firm market capitalisation. According to Seyhun, this can be attributed to
more efficient pricing for large firms since they are more extensively covered by
analysts.
Ma, Sun and Austin (2000) document a significant increase in insider selling
prior to stock split announcements. They also find indices for a significant corre-
lation between insider sales and stock price run-up prior to stock split an-
nouncements. Their results suggest that portfolio diversification may be the dom-
inant factor behind insider trading activities prior to stock split announcements.
A noteworthy contribution is that of Carter et al. (2002). Based on a sample
obtained from the Washington Service Insider Trade database, the authors pro-
vide empirical evidence that some of the information content of insider transac-
tions leaks out prior to the information becoming public, and that the information
leakage positively depends on the length of the interval between the insider buy-
ing activity and the announcement.
After giving some references to the existing contributions concerning the rela-
tionships between insider trading and stock market performance, we proceed
with formulating the main hypotheses, explicitly. Although the direct identifica-
tion and measurement of impact of insider trading on stock market poses many
difficulties, existing evidence concerning developed stock markets all other the
world seems to indicate that sales and purchases of stocks by insiders induce
stock price changes whose direction and scope depend on many factors, among
which the content of private information possessed by insiders is most promi-
nent. This leads to the first hypothesis:
H1: In the context of insider disclosures about stock purchases (sales), the
prices of this tend to increase (decrease).
If other investors (outsiders) believe that signs derived from announcements
of insider trading are able to help with making profits that exceed those offered
by the market portfolio, they will be prone to buying stocks which insiders
bought first and to sell stocks at the some time as insiders. In this context, it
seems to be particularly interesting to check whether such an imitative invest-
ment strategy is really profitable for outsiders. Therefore, the next hypothesis is
formulated as follows:
H2: Imitating the transactions of insiders gives an extra return which mirrors
the private information underlying insider transactions.
As pointed out above, the standard vehicle for answering this type of question
is the event study approach, under this study extended by the reference return
portfolio approach. The framework of event study analysis is presented in the
next section; a description of the reference return portfolio approach is given in
Section 5.2.
3 Methodology
Since Fama et al. (1969) formalised the event study approach more than thirty
years ago, it has become the standard method used in empirical finance and ac-
counting in order to detect the reaction of certain financial market variables such
as returns and volume to some informative events which are assumed to be unan-
ticipated. The well-known constitutive elements of this methodology, within its
classical framework, are: determination of the event, including its announcement
date, determination of the event window and estimation window, specification of
the return generating model and the estimation of the model parameters based on
the data set from an estimation window, computation of the abnormal return (as
well as the cumulative abnormal return) and finally, construction of a statistical
significance test of the event effect.
A common practice is to define an abnormal return of ith firm for day t (ARi,t)
as the difference between the actual stock return of ith firm for day t (ri,t) and the
expected value of stock return of ith firm for day t (E[ri,t|It-1]) conditional on the
set of information It-1 available one day before the event window. The classical
Market Model, introduced by Sharpe (1963) is, on the other hand, the most fre-
quently applied model to generate the expected value of stock returns required
for calculating abnormal returns. In this model the stock returns of a given firm
depend on market portfolio returns (rm,t):
, , , ,, ~ 0, ,i t i i m t i t i t ir r N (1)
where ε represents the error term.
Following Boehmer et al. (1991), define a standardised abnormal return of ith
firm for day t (SARi,t) as:
,
,
,
.ˆ
i t
i t AR
i t
ARSAR
(2)
The denominator of (2) can be obtained from:
1
0
1 1
0 0
21
, ,
1 0
, 21 1
1 02
, ,
1 0
1
1ˆ ˆ 1 ,
1
t
m t m k
k tAR
i t it t
m k m k
k t k t
r rt t
t tr r
t t
(3)
where ˆi denotes the estimated standard deviation in (1), t0 and t1 denote an indi-
cator of the oldest observation within the estimation window and the event win-
dow, respectively.
Note that by defining a standardised abnormal return as (2), the forecasting
nature of the estimated abnormal returns is taken into account (see Boehmer et
al., 1991 and Salinger, 1992).
For N firms included in the sample, the test statistic takes the following form:
,
1
2
, ,
1 1
1
,
1 1
( 1)
N
i t
it
N N
i t k t
i k
SARN
Z
SAR SARN N N
(4)
and for any sub-period within the event window restricted by tL and tU (tL < tU):
;
.1
U
L
L U
t
k
k t
t t
U L
Z
Zt t
(5)
The statistics (4) and (5) are both normally distributed with mean zero and
unit standard deviation.
As Boehmer et al. (1991) demonstrated, a test based on (4) is still well speci-
fied, even if the cross-sectional variance increases during the event period. A re-
cent study, however, shows that it becomes misspecified when the estimation
window data set is disturbed by confusing events (see Aktas et al., 2003). Aktas
et al. provide a potential remedy for this problem. Using the Markov Switching
Regression approach introduced and developed by Hamilton (1989, 1994) in-
stead of the classical Market Model, the above-mentioned authors were able to
eliminate the test bias.
Following Aktas et al., it is assumed that the return generating process can be
adequately modelled using a two-regime process, one regime with a normal level
of variance and one regime with high variance. On the other hand, the model pa-
rameters are assumed to be the same in both regimes. This can be expressed as:
, , ,1, ,1, ,1 ,
, , ,2, ,2, ,2 ,
,1 ,2
, ~ 0, if 1
, ~ 0, if 2
and ,
i t i i m t i t i t i i t
i t i i m t i t i t i i t
i i
r r N S
r r N S
(6)
where Si,t stands for a not directly observable indicator variable taking value 1 if
we are in the low variance regime, and 2 if we are in the high variance regime.
Regime state variable Si,t is assumed to be described by a first-order Markov pro-
cess (this assumption means that Si,t depends only on its lagged value at lag 1).
For days with no event occurrence variance is assumed to achieve the first
(normal) regime. When an event occurs, there is a shift in the variance from the
lower normal regime to the higher event-induced regime (the second regime in
(6)). Implied in this is the assumption that the extent to which the variance be-
comes higher in the case of event occurrence remains the same in spite of fact
that different events may influence variance in a distinctive manner. However,
even if confounding events lead to shifts in variance of different magnitudes,
model (6) seems still to be superior to the classical Market Model relying on the
assumption of constant variance with respect to time. Moreover, major emphasis
is placed within the event study framework upon guaranteeing the estimated lev-
el of variance of the variable under consideration not to be affected by confound-
ing events (and the event in question as well) since this is necessary if statistical
inferences about the direction and scale of event effects are to be valid. In this
context the promising nature of model (6) is particularly undisputed.
The dates of possible confounding events are unknown. If they were known,
the researcher could make an effort to exclude all confounding events from the
sample in a way other than using the switching regression model. Without in-
formation about when the confounding events take place, we have to assume that
their occurrence can be described as a first-order Markov process. The state of
such a process (1 for low regime and 2 for high regime) on a given day t depends
only on which of the regimes was achieved on the previous day, i.e. in day t–1.
Therefore, one of four different situations may happen, which are as follows: (i)
reaching the low regime from the high regime, (ii) reaching the high regime from
the low regime, (iii) staying in the low regime, and (iv) staying in the high re-
gime. Each of the above-defined situations is assigned the corresponding proba-
bility, pij, referred to as the probability that the process is in ith regime in period t
when in the period t–1 it was in jth regime. Noting that p12=1–p11 and p21=1–p22,
there are six parameters (αi, βi, σi1, σi2, p11, p22) to be estimated. A detailed de-
scription of the estimation procedure of the model parameters can be found in
Hamilton (1989, 1994).
With model (6) used to generate abnormal returns, the standardised abnormal
return form (2) is modified by replacing in (3) ˆi with
,1ˆ
i . Using the standard
deviation estimates for regime 1 (low variance) in (3), we do not automatically
produce a higher cross-sectional Z statistic in the homoscedastic case (see Aktas
et al., 2003).
While the above-presented technique of testing abnormal market performance
deals explicitly with the possibility of increases in variance due to confounding
event occurrences, it does not adjust for the possibility that the variance of the
variable under consideration varies persistently over time. Building on the semi-
nal contribution of Engle (1982) the standard vehicle for this type of analysis is
the (G)ARCH-type model, often the GARCH(1,1) model in empirical imple-
mentation.
An original test for abnormal market performance, proposed by Hilliard and
Savickas (2000), is based on the Market Model and the GARCH(1,1) error term.
In this study we, however, decided to use the generalized ARMA(r, m)-MM-
GARCH(p, q) model given by:
, ,0 , , , , , , ,
1 1
2
, ,0 , ,
1 1
, ~ 0,r m
i t i i j t j i m t i t i j i t j i t i t
j j
q p
i t i i j t j i j t j
j j
R R R h
h h
(7)
The proper length of time-lags in the model is identified using the Akaike In-
formation Criterion. The model parameters are estimated by means of the ML-
method from observations included within the pre-event window.
The test statistic (lt) can be expressed as:
2
,
1
1 ,tt
N
i t t
i
ASRl N
SR ASR
(8)
where , , ,
ˆ1/i t i t i tSR AR h and 1
,
1
.N
t i t
i
ASR N SR
In order to test the implications of insider trading disclosures over any sub-
period of the event window whose boundaries are set as tL and tU (tL < tU), the
standardized cumulative abnormal return can be calculated:
,
, ,
,
.
ˆ
U
L
L UU
L
t
i t
t t
i t tt
i t
t t
AR
SCAR
h
(9)
The corresponding test statistics is given by:
, ,
2
, , ,
1
1,
L U L U
L U L U
CAR
t t t t N
i t t t t
i
N Nl ASCAR
SCAR ASCAR
(10)
where 1
, , ,
1
.L U L U
N
t t i t t
i
ASCAR N SCAR
At the start of this study, we define the pre-event window and the event win-
dow. The symmetrical event window covering eleven days starts the fifth day
prior to the event day (the day when the disclosure took place). The pre-event
window, on the other hand, contains two hundred days prior to the event win-
dow. Transforming price series into returns, we used the continuous form of re-
turns. Market portfolio returns are approximated by returns of the market-
capitalisation weighted stock index called WIG. Then, we generate abnormal re-
turns using model (6), and standardised abnormal returns. Finally, for six differ-
ent sub-periods of the event window, we check the null hypothesis about no sig-
nificant event effect by means of test statistic (5). To verify our results we re-
calculate the tests using model (7) and statistics (10).
As a recent study shows (see Tkac, 2001), a combination of firm-
specific/market adjusted is the best way to generate the measure of firm-specific
normal (non-event related) trading activity. Therefore, analysing abnormal trad-
ing volume, we decided to employ model (6) which, taking into account that
volume series are usually more strong correlated, needs to be now modified by
the inclusion of an autoregressive term in the mean equation, and the same test-
ing procedure as used for returns. Trading activity is measured by the number of
shares traded per day. Next, we use the first differences in the natural logarithm
of firm trading volume (market trading volume), instead of individual stock re-
turns (market portfolio returns). As with returns, model (7) and statistics (10)
were finally employed in order to check whether the testing results for abnormal
trading volume are robust.
4 Sample Description
Our initial data set comprises 406 insider disclosures which took place on the
WSE over the period from June 1998 to June 2004 (six years). The source of
these data is the register of insider transactions which can be accessed free on the
Internet at http://mojeinwestycje.interia.pl. Starting work with the data, we cate-
gorised insider transactions into three groups (member of the board, member of
the supervisory directors and others) and collated the total number of insider dis-
closures over the period under study with the level of WIG (see figures 1 and 2).
Figure 1. The number of insider transactions on the WSE
7/19984/1999
2/200012/2000
9/20017/2002
5/20036/2004
8000
10000
12000
14000
16000
18000
20000
22000
24000
26000
0
1
2
3
4
5
6
7
8
Purchases(R) WIG(L)
7/19984/1999
2/200012/2000
9/20017/2002
5/20036/2004
8000
10000
12000
14000
16000
18000
20000
22000
24000
26000
0
2
4
6
8
10
12
Sales(R) WIG(L)
Figure 2. The proportion of three groups of insiders in dealing
Purchases
28%
44%
28%
Member
of the board
Member
of the supervisory
directors
Others
Sales
31%
39%
30%
Member
of the board
Others
Member
of the supervisory
directors
It is a widely documented fact that the level of insider activity on the stock
market is correlated with market changes (see e.g. Biesta et al., 2003). Our data
also seems to support such a statement. One can find that insiders were prone to
buy shares in their own companies when the market started to increase. The most
striking finding, though in a sense expected by us, is that insider transactions
might protect the market from a strong falling tendency. In the second half of
2000, insiders bought shares in defiance of the general trend of stock prices. This
might have a stabilising impact on the market even if amount of such transac-
tions is small when compared to the whole market. A detailed knowledge of the
dates of insider sales (disclosures about such transactions) may also prove useful
to investors. Note that the announcement about selling shares by insiders pre-
cedes the periods over which stock prices were falling. With regard to the pro-
portion of selected categories of insiders in the total dealing, it can be seen that
the members of the board appear to have some prominence and there is little dif-
ference in this matter between purchases and sales.
To be included in our sample, a given firm has to meet the following selection
requirements. Firstly, it should be reliable and large enough to be quoted on the
primary market of the WSE. Secondly, the information of insider trade in shares
of a given firm should be close enough to identify all circumstances of interest,
particularly, the day in which this transaction indeed took place. We excluded
from the sample all cases where insider transactions are related to (granted) op-
tions and warrants. As we pointed out in Section 1, the date of disclosure of in-
sider transaction is not necessarily identical the date of the real insider activity.
Typically, it is five days from the insider transaction to its disclosure but in some
cases this period may covers even a few months. Therefore, we decided to con-
fine our investigation to these of the insider transactions where the corresponding
disclosure took place no later than on the fifth day after the transaction. These
requirements are fulfilled by 164 disclosures. Next, we divided our sample into
two clusters, namely Purchases (71 events included) and Sales (94 events includ-
ed). For more details see Appendix at the end of this paper.
Daily closing prices, as well as the number of shares traded over a day for
firms included in our sample, are derived from the PARKIET database. In a few
cases a single missing piece of data was filled in (as a mean of its direct neigh-
bours).
5 Empirical Results
5.1 Abnormal market performance
We start our investigation by checking residual series from the respective models
of expected returns and expected volume for autocorrelation, non-normality and
ARCH effect. The Ljung-Box Q statistic is used to make sure that the obtained
results are not being markedly affected by a correlation in the model residuals.
The length of time-lag in the above-mentioned statistic is equal to 15. This re-
flects our view that further autocorrelation coefficients, beyond the fifteenth
time-lag, can all be neglected without any detriment to the investigation. We
found that the ratios of the number of cases where the null hypothesis about ran-
domness was rejected at the 5% level of significance to the number of all cases
tested amount to 20% and almost 40% for return and volume series, respectively.
In order to evaluate to what extent the residual series can be distributed as nor-
mal, the Lilliefors test for normality was here employed. It turned out that the re-
siduals from the model of expected returns exhibit strong non-normality; the null
hypothesis of normality had to be rejected 137 times out of 164 cases. With re-
gard to residuals coming from the model of expected volume it should be
stressed that the corresponding null hypothesis was rejected in less than half of
the cases. Finally, we used the ARCH test, up to the fifteenth time-lag, to check
whether or not the residual series suffer from volatility clustering effects. In this
case the ratio of rejection of the null hypothesis, at a 5% level of significance,
achieved almost 40% for both return and volume series.
Summing up the above-reported results, it should be stated that, as we ex-
pected, the switching regression model is not fully designed for dealing with the
well-documented inconvenient properties of financial time series. Despite this,
the results derived from this model still seem valid, at least as a rough approxi-
mation. However, the drawn conclusions need to be confirmed by using model
(7), which deals explicitly with the possibility of autocorrelation as well as vola-
tility clustering of the variable under consideration.
After controlling for model misspecification, we continued to test whether or
not stock returns exhibited an abnormal pattern over the days of the event win-
dow. We decided to examine the stock returns also in several different sub-
periods of the event window to gain insight into trends mirrored in our sample.
Table 1 summarizes our results. We included the cross sectional average
standardised abnormal returns and the corresponding test statistic, as well as p-
value over the entire event window and the selected sub-periods.
Table 1. Testing for abnormal returns in two clusters
Period
,L Ut t
Purchases Sales
Average
SAR ,L Ut tZ p-value
Average
SAR ,L Ut tZ p-value
{–5,–1} 0.125 0.901 0.184 0.038 0.135 0.446
{–3,–1} 0.530* 2.111 0.017 –0.120 –0.135 0.446
{–3,+3} 0.322* 1.815 0.035 –0.465 –0.437 0.331
{+1,+3} 0.099 0.311 0.378 –1.062 –1.075 0.141
{+1,+5} 0.238 1.236 0.108 –0.269 –0.367 0.643
{–5,+5} 0.198 1.624 0.052 –0.078 0.127 0.449
* significant at 5%.
It can be seen that disclosures about insider purchases induce, on average, a
significant increase in stock prices during the period from the third trading day
prior to the announcement to the third trading day after it. In addition, the aver-
age abnormal return statistically differs from zero in the period from the third to
the first trading day before the insider disclosure. On the other hand, insider sales
are not accompanied by a stock price reaction either before or after the day when
the facts about such transactions are disclosed. What can be concluded from this?
Firstly, our findings provide empirical support for the argument made by
Lakonishok and Lee (1998) that insiders sell their shares for many different mo-
tives, but they buy shares for only one motive – to gain profit. Therefore, a signal
for other market participants is more transparent in the case where an insider
purchases shares. When an insider sells shares in his own company, outsiders
need to solve the problem as to what signal is given to them by the insider and
whether they should follow in the insider’s footsteps or not.
Secondly, the results illustrate the complicated nature of relationships between
an insider and the so-called environment which includes his family, friends, his
broker and others. The significance of the test statistic at a 5% level during the
period from day –3 to day –1 can be seen as evidence supporting the hypothesis
about leakage of information concerning insider transactions prior to the official
disclosure. This problem is discussed in detail in Carter et al. (2002). The above-
mentioned author listed the possible sources of such a leakage. Family, friends
and brokers are, taken together, one of the most important sources of information
leakage before the official disclosure.
An effort has been made to explain the cross sectional differences in the scope
of leakage of information by various factors. Unfortunately, applying the regres-
sion suggested by Carter et al., we find no support for a positive relationship be-
tween the information leakage and the time which elapses between an insider
transaction and on official announcement. Also, none of the other factors em-
ployed by Carter et al. appears to have explanatory power for information leak-
age.
Testing for abnormal volume completes this part of our investigation. Table 2
presents the cross sectional average standardised abnormal volume (SAV) as well
as t-Student statistics and corresponding p-value in two clusters.
Table 2. Testing for abnormal volume in two clusters
Period
,L Ut t
Purchases Sales
Average
SAV ,L Ut tZ p-value
Average
SAV ,L Ut tZ p-value
{–5,–1} 0.921* 1.785 0.037 0.752 0.247 0.402
{–3,–1} 0.985** 2.368 0.009 1.030 1.096 0.136
{–3,+3} 0.321 0.638 0.262 0.471 0.197 0.422
{+1,+3} –0.506 –0.970 0.166 –0.802 –1.029 0.152
{+1,+5} –0.771 –1.339 0.090 –0.660 –0.817 0.207
{–5,+5} –0.009 0.079 0.468 0.334 –0.262 0.397
** significant at 1%
* significant at 5%.
With regard to the context of volume in surroundings of insider trading disclo-
sures, abnormal trading activity occurs during the period of three days prior to
the official announcement only in the first cluster (Purchases). This finding is in
line with those previously given in respect to testing for abnormal returns.
As we pointed out at the beginning of this section, the conclusions drawn
above may appear incorrect due to the fact that the switching regression model
used to generate the expected returns as well as the expected trading volume is
unlikely to give a reliable representation of the variable in question when its var-
iance persistently varies with respect to time. Therefore, to confirm our previous
findings a GARCH-based technique, tracing statistically significant movements
of stock prices and trading volume, is employed. For the sake of comparison
simplicity, the same sub-periods of the event window are applied as when the
expected returns and the expected trading volume were obtained by means of the
switching regression model. The results are reported in the tables 3 and 4.
Table 3. Test results of GARCH-based abnormal returns in two clusters
Period
,L Ut t
Purchases Sales
,L Ut tASCAR ,L U
CAR
t tl p-value ,L Ut tASCAR ,L U
CAR
t tl p-value
{–5,–1} 3.484 1.420 0.161 0.970 1.002 0.319
{–3,–1} 3.607* 2.215 0.031 0.816 1.000 0.320
{–3,+3} 4.999* 2.495 0.015 0.805 0.634 0.528
{+1,+3} 1.225 1.196 0.237 0.667 0.697 0.488
{+1,+5} 1.659 1.617 0.112 0.913 0.784 0.435
{–5,+5} 5.310 1.878 0.065 1.206 0.939 0.350
Note that ASCAR values are expressed in per cent.
* significant at 5%.
Table 4. Test results of GARCH-based abnormal volume in two clusters
Period
,L Ut t
Purchases Sales
,L Ut tASCAR ,L U
CAR
t tl p-value ,L Ut tASCAR ,L U
CAR
t tl p-value
{–5,–1} 0.577** 4.155 0.000 0.716** 3.696 0.000
{–3,–1} 0.418* 2.171 0.034 0.958** 5.180 0.000
{–3,+3} 0.091 0.487 0.628 0.437* 2.033 0.045
{+1,+3} –0.242 –1.707 0.093 –0.331 –1.704 0.092
{+1,+5} –0.429* –2.187 0.033 –0.391* –2.116 0.037
{–5,+5} 0.062 0.360 0.720 0.135 0.917 0.362
** significant at 1%
* significant at 5%.
As one can see from the figures, the test results for the significance of stand-
ardized cumulative abnormal returns fully coincide with those derived from the
switching regression model. Regarding trading volume, some differences be-
tween the results based upon the two applied techniques can be, however, identi-
fied. In addition to significant shifts in the level of market activity during the pe-
riods from the fifth and third trading day before the disclosure release in the case
of purchases of shares by insiders, which corroborates the previous findings,
there are also changes in volume over the second half of the event window for
this cluster. The GARCH-based technique reveals that the days immediately af-
ter the insider disclosure brought a significant offset. The same pattern, i.e. posi-
tive changes in the level of market activity during the first half of the event win-
dow and negative ones over the second half, can be also identified in the case of
disclosures concerning insider sales.
5.2 Is imitation of insider behaviour profitable?
In this section, our attention is focused on the profitability of the investment
principle of copying insider behaviour. In other words – we try to find an answer
as to whether outsiders may profit from signals which are transferred to the stock
market by insiders who buy or sell shares in their own companies. One reason in
favour of the imitation of insider behaviour is based on the argument that insid-
ers such as executives and directors have superior information concerning their
company prospects. They know, for instance, when a new product is going to be
on the market earlier than other people. As a result they are also able to forecast
future stock price movements more accurately than other market participants. Ir-
respective of its theoretical basis, all over the world, investors carefully observe
insider activity because it is still believed that thanks to an imitation of insider
investment decisions they are able to profit from information accessed only by
insiders.
To examine the profitability of investment strategies based on insider disclo-
sures, we use the reference return portfolio approach which is similar to that of
Gervais et al. (2001). Our portfolio formulation procedure can be described as
follows. The test period for each stock position in the portfolio covers 130 trad-
ing days (approximately six months) and starts the day after the insider trading
announcement is released. We form the reference return portfolio by taking a
long (short) position in all stocks which were previously bought (sold) by insid-
ers. One zloty is invested in every stock included in the portfolio. At the same
time, every long (short) position is offset by a short (long) position in a reference
portfolio (here the market portfolio). All positions are held without rebalancing
until the end of the test period. Offsetting each position by the reference portfo-
lio, we are able to test the hypothesis that the average portfolio return is equal to
zero, separately, for all the positions in stocks which were previously bought by
insiders (Purchases) and all the positions in stocks which were previously sold
by insiders (Sales).
Figure 3 contains the average returns of the reference return portfolio formula-
tion strategies over the entire test period in two clusters separately, as well as the
net return for these clusters taken together. Because some of the firms previously
included in our sample had incomplete stock price series over the test period, we
reduced the number of events to one hundred and fifty four (66 purchases and 88
sales).
Figure 3. Average returns of the reference return portfolio formulation strategies
trading day
1 11 21 31 41 51 61 71 81 91 101 111 121 130-8,0%
-6,0%
-4,0%
-2,0%
0,0%
2,0%
4,0%
6,0%
8,0%
10,0%
12,0%
Purchases
Sales
Net
Regarding the profitability of investment strategies based on insider disclo-
sures, it is worth noting that the disaggregated analysis of purchases versus sales
shows that outsiders who copy insider behaviour, resulting in stock purchases,
would receive negative returns. Taking into account the above-mentioned fact
that in the case where an insider buys shares in his own company, the signal is
more transparent for outsiders, this finding seems to be a bit surprising. Howev-
er, at least two other researchers (Brick et al., 1989 and more recently Jeng et al.,
1999) found the same return pattern. Their authors based the investigations upon
data from the U.S. stock market, and in both cases it turned out that outsiders im-
itating insiders who bought stocks in their own companies experienced negative
excess returns.
There are two possible explanations for this phenomenon. Firstly, insiders are
likely to sell shares immediately before negative information concerning their
companies is issued, but they may buy the shares in their own companies long
before the good news becomes known to other investors. Hence, the observed
negative return may be a consequence of the relatively narrow window over
which profitableness of the respective portfolios is examined. Secondly, as
demonstrated by Banz (1981) and Reinganum (1981), among others, there is an
inverse relationship between stock returns and the market value of a given firm
(the so-called size effect). Therefore, it is possible that the negative returns in the
first cluster is a result of the overlapping of two opposing effects. The one asso-
ciated with insider trades is positive, but the other, which is rooted in the high
market value of firms included in the sample, forces returns down, so that the re-
sultant effect is negative.
On the other hand, outsiders who sell stocks which were previously sold by
insiders, profit from it. Note that the average portfolio return in this cluster sys-
tematically increases over time. In spite of the fact that insiders sell their shares
for many different motives, the imitation of their investment decisions appears to
be profitable.
Table 5 summarises test results for the significance of the average portfolio
returns in the respective clusters at six different time-points. The test used here is
based on the t-Student statistic with n – 1 degrees of freedom.
Table 5. Testing for significance of average returns of the reference return portfolio for-
mulation strategies
Test period (in trading days):
1 10 20 50 100 130
Purchases (returns in %)
0.443 (0.806)
–0.767 (–0.735)
–0.312 (–0.187)
–2.982 (–1.243)
–3.960 (–1.218)
–5.528 (–1.341)
Sales (returns in %)
0.331 (0.933)
2.266* (1.748)
3.343* (1.841)
6.200* (2.155)
8.970** (2.356)
8.569* (1.846)
Net (returns in %)
0.379 (1.224)
0.966 (1.109)
1.777 (1.405)
2.265 (1.151)
3.429 (1.305)
2.528 (0.783)
t-Student statistics are shown in the parentheses ** significant at 1%
* significant at 5%.
The results illustrate the promising nature of the investment principle of copy-
ing the insider’s behaviour resulting in stock sales as a tool for gaining profit.
Note that in the second cluster (Sales) the average portfolio return is a positive
and statistically significant event at 1%. Consistent with the conclusions drawn
on the basis of Figure 3, the average portfolio return in the first cluster (Purchas-
es) is negative, though insignificant.
6 Conclusions
The main aim of this paper is to explore the information content of insider trad-
ing disclosures. Many previous studies have dealt with this subject and a distinc-
tive contribution to our knowledge of insider transactions has been made by their
authors. Our study contributes to this by investigating stock prices as well as
trading volume reaction to insider disclosures by using a data set from a small
emerging stock market.
Consistent with empirical evidence presented in the literature on developed
stock markets, we find significant positive abnormal returns (excess volume) in
the context of the disclosure date of insiders who purchase stock in their own
companies. Insider sales, as opposed to purchases, have no informational content
given that the various possible motives for this remain essentially opaque. Our
results seem to support the hypothesis about leakage of information concerning
insider transactions prior to official disclosure. However, there is no evidence
supporting a positive relationship between information leakage and the time
which elapses between insider transactions and official announcements.
We are also interested in the benefits to outsiders from public information
about insider trading. It turned out that outsiders, who copy the insider’s behav-
iour resulting in stock purchases, received negative returns. In contrast outsiders
who sell stocks previously sold by insiders may profit from it, in spite of the fact
that insiders might sell their shares for many different motives.
Our study is rather tentative due to the low number of events included in the
sample. Therefore, future research should reveal whether the inferences docu-
mented in this study will hold true also in the case of a larger sample.
References
[1] Aktas N, Bodt E, Cousin J-G (2003) Event Study under Disturbed Estimation Pe-
riod, NBER Working Paper Series
[2] Banz, R.W. (1981) The Relationship Between Return and Market Value of Com-
mon Stock, Journal oJ Financial Economics 9(1), 3–18
[3] Biesta MA, Doeswijk RQ, Donker HA (2003) The Profitability of Insider Trades
in the Dutch Stock Market, NBER Working Paper Series
[4] Boehmer E, Musumeci J, Poulsen A (1991) Event-Study Methodology under
Conditions of Event Induced Variance, Journal of Financial Econometrics 30,
253–272
[5] Brick IE, Statman M, Weaver DG (1989) Event Studies and Model Misspecifica-
tion: Another Look at the Benefits of Outsiders from Public Information about In-
sider Trading, Journal of Business Finance and Accounting 16, 399–424
[6] Carter ML, Mansi SA, Reeb DM (2002) Evidence on Quasi-Private Information
and Insider Trading, NBER Working Paper Series
[7] Engle R.F. (1982) Autoregressive Conditional Heteroscedasticity with Estimates
of the Variance of the U.K. Inflation, Econometrica 50, 987–1008
[8] Fama EF (1970) Efficient Capital Market: A Review of Theory and Empirical
Work, Journal of Finance 25, 383–417
[9] Fama EF (1991) Efficient Capital Markets II, Journal of Finance 46, 1575–1617
[10] Fama EF, Fisher L, Jensen M, Roll R (1969) The Adjustment of Stock Prices to
New Information, International Economic Review 10, 1–21
[11] Gervais S, Kaniel R, Mingelgrin DH (2001) The High-Volume Return Premium,
The Journal of Finance 3, 877–919
[12] Hamilton JD (1989) A New Approach to the Economic Analysis of Nonstationary
Time Series and the Business Cycle, Econometrica 57, 357–384
[13] Hamilton JD (1994) Time Series Analysis, Princeton University Press
[14] Hilliard J., Savickas R. (2000) On stochastic volatility and more powerful para-
metric tests of event effects on unsystematic returns, Working paper, Terry Col-
lege of Business, University of Georgia
[15] Holderness C, Sheehan DP (1985) Raiders or Saviors? The Evidence on Six Con-
troversial Investors, Journal of Financial Econometrics 14, 555–579
[16] Jaffe J (1974) Special Information and Insider Trading, Journal of Business 47,
410–428
[17] Jeng LA, Metrick A, Zeckhauser R (1999) The Profits to Insider Trading, A Per-
formance-Evaluation Perspective, NBER Working Paper Series
[18] Kerr HS (1980) The Battle of Insider Trading vs Market Efficiency, Journal of
Portfolio Management (Summer), 47–50
[19] Lakonishok J, Lee I (1998) Are Insider’s Trades Informative?, NBER Working
Paper Series
[20] Lin J, Howe JS (1990) Insider Trading in the OTC Market, Journal of Finance 55,
1273–1284
[21] Ma Y, Sun HL, Yur-Austin J (2000) Insider Trading Around Stock Split An-
nouncements, The Journal of Applied Business Research 16, 13-25
[22] Madura J, Wiant KJ (1995) Information Content of Bank Insider Trading, Applied
Financial Economics 5, 219–227
[23] Raad E, Wu HK (1995) Insider Trading Effects on Stock Returns Around Open-
Market Stock Repurchase Announcements: An Empirical Study, The Journal of
Financial Research XVIII, 45–57
[24] Reinganum, M.R., (1981) Misspecification of Capital Asset Pricing: Empirical
Anomalies Based on Earnmgs Yields and Market Values, Jotanal of Financial
Economics 9, 19–46
[25] Salinger M, 1992 Standard Errors in Event Studies, The Journal of Financial and
Quantitave Analysis 27, 39–53
[26] Seyhun HN (1986) Insider Profits, Cost of Trading, and Market Efficiency, Jour-
nal of Financial Economics 16, 189–212
[27] Seyhun HN (1998) Investment Intelligence from Insider Trading, MIT Press,
Cambridge MA
[28] Seyhun HN (1998) The Information Content of Aggregate Insider Trading, Jour-
nal of Business 61, 1–24
[29] Sharpe W (1963) A Simplified Model for Portfolio Analysis, Management Sci-
ence 9, 277–293
[30] Tkac PA (2001) A Trading Volume Benchmark, Theory and Evidence, Journal of
Financial and Quantitative Analysis 34, 89–114
Appendix
Table A. Companies included in the sample
Company /ISIN/ Pur-
chases Sales Company /ISIN/
Pur-chases
Sales
YAWAL /PLYAWAL00058/ 1 1 STALEXPORT /PLSTLEX00019/ 1 0 APATOR /PLAPATR00018/ 3 0 SUWARY /PLSUWAR00014/ 1 0 ATLANTIS /PLATLNT00016/ 2 2 SWARZĘDZ /PLSWRZD00017/ 1 0
BUDIMEX /PLBUDMX00013/ 1 3 TELEKOMUNIKACJA POLSKA /PLTLKPL00017/
1 1
BRE /PLBRE0000012/ 1 3 TRAS /PLTRAST00020/ 1 0 BORYSZEW /PLBRSZW00011/ 1 0 WÓLCZANKA /PLWLCZN00013/ 1 2 CERSANIT /PLCRSNT00011/ 1 0 WAWEL /PLWAWEL00013/ 1 0 DROSED /PLDRSED00012/ 2 0 AGORA /PLAGORA00067/ 0 4 ECHO INVESTMENT /PLECHPS00019/
1 1 AMICA /PLAMICA00010/ 0 1
ELDORADO /PLELDRD00017/ 1 0 ELEKTROCIEPŁOWNIA BĘDZIN /PLECBDZ00013/
0 1
EFEKT /PLEFEKT00018/ 1 1 BEEF-SAN /PLBEFSN00010/ 0 1 ELZAB /PLELZAB00010/ 1 1 BAUMA /PLBAUMA00017/ 0 1 ENERGOMONTAŻ POŁUDNIE /PLENMPD00018/
1 0 COMARCH /PLCOMAR00012/ 0 1
ENERGOMONTAŻ PÓŁNOC /PLENMPN00017/
1 1 COMPUTERLAND /PLCMPLD00016/
0 4
INTER GROCLIN AUTO /PLINTGR00013/
1 2 COMPUTER SERVICE SUPPORT /PLCSSUP00012/
0 2
GANT /PLGANT000014/ 1 1 DĘBICA /PLDEBCA00016/ 0 1 HYDROTOR /PLHDRTR00013/ 1 0 FARMACOL /PLFRMCL00066/ 0 1 HUTMEN /PLHUTMN00017/ 1 0 FERRUM /PLFERUM00014/ 0 1
IB SYSTEM /PLBRSTM00015/ 1 0 GARBARNIA BRZEG /PLGRBRN00012/
0 1
INTERNET GROUP /PLARIEL00046/
3 1 GRUPA ONET.PL /PLOPTMS00012/
0 1
IMPEXMETAL /PLIMPXM00019/ 1 1 HOWELL /PLHOWEL00015/ 0 2
JUTRZENKA /PLJTRZN00011/ 1 2 INSTAL KRAKÓW /PLINSTK00013/
0 1
KROSNO /PLKROSN00015/ 1 1 INSTAL LUBLIN /PLINSTL00011/ 0 1 LUBAWA /PLLUBAW00013/ 1 0 IRENA /PLIRENA00018/ 0 1 LENTEX /PLLENTX00010/ 1 2 KRUK /PLKRUK000019/ 0 1 MCI /PLMCIMG00012/ 1 0 KĘTY /PLKETY000011/ 0 1
MILMET /PLMLMET00015/ 1 0 MANOMETRY KFM /PLKFMAN00012/
0 1
MENNICA PAŃSTWOWA /PLMNNCP00011/
1 0 MIESZKO /PLMSZKO00010/ 0 2
MASTERS /PLELPO000016/ 1 1 NETIA /PLNETIA00014/ 0 1 MOSTOSTAL WARSZAWA /PLMSTWS00019/
1 1 OKOCIM /PLOKOCM00018/ 0 1
MUZA /PLMUZA000019/ 1 1 ORBIS /PLORBIS00014/ 0 2
NOVITA /PLNVITA00018/ 2 1 POLLENA EWA /PLFKPEW00018/
0 1
OBORNIKI WFM /PLWFM0000016/
1 0 POLIGRAFIA /PLPLGRF00011/ 0 1