Forecasting financial variables industry-wide
An empirical study to which extend insider are able to forecast earnings and
cash flow in branch related companies.
This thesis has studied publications about the predictability of accounting variables through individual insider purchases industry-wide. Based on this empirically gathered information, this thesis tries to draw conclusions as individual insider purchaser for the predictability of earnings. The analysed models in this thesis show that when a separation between earnings and cash flow is made, the holding returns will behave differently over time. To conclude: insider purchases in the same industry is relevant for the predictability of earnings and less relevant for cash flows.
Suzanne Groen 10003288 Bachelor Thesis
Finance & Organization
Supervisor: Shivesh Rafaél Changoer 22-08-2013
2
Contents
1 Introduction 3 2 Literature 5 3 Research Methodology 9 3.1 Data 9 3.2 Sample 10 3.3 Methodology 11 4 Results 13 4.1 Descriptive Statistics 13 5 Conclusion 22 5.1 Implications 23 6 References 24 7 Appendix A 253 1. Introduction
This thesis has studied publications about the predictability of accounting variables through
individual insider purchases industry-wide. Based on this empirically gathered information,
this thesis tries to draw conclusions as individual insider purchaser for the predictability of
earnings. The analysed models in this thesis show that when a separation between earnings
and cash flow is made, the holding returns will behave differently over time. To conclude:
individual insider purchases in the same industry is relevant for the predictability of earnings
and less for cash flows.
As opportunistic insider trading, you can say something about making abnormal profits on
transactions. Piotroski (2004) describes insiders as the best informed party in a company,
judged by opportunities and risk.1 Seyhun (1988) examines if aggregate trading by corporate insiders in their own firm, predictive is for the market return. Theorized is that all factors that
effects security returns, in response is to trading by corporate insiders. (Company specific,
branch own and economic wide). Therefore analysis about insider trading might uncover the
effects of industrywide factors which are currently not noticedin the prices of securities. This
leads towards the purpose of this thesis; to deliver detailed proof of the predictability of
earnings and cash flow as an industry insider purchaser. In this thesis an analysis is made to
what extend insider purchases can predict the accounting variables earnings and cash flow
from branch related companies. This thesis differs in some ways from previous research. It
takes into account both variables earnings and cash flows. Also it looks only at insider
purchases, so not the sales. Only looking at the insider purchases has a reason which is based
1
Their trading activities are subject to regulations and restrictions. In the stock market, the trades from insiders are the most heavily analyzed activities. Because the insider trades are enormously in the stock market is it not astonishing that there is a huge amounts of empirical literature about insider trading (Cohen, 2012). Trades of insiders are visible to investors through disclosure on the Security and Exchange Commission (SEC), which means that insiders must report their trading activities to the SEC. In the years before 2001 the trades had to be registered before the tenth of the next month. From August 2002 the Sarbanes Oxley (SOX) was established and requires from insiders that they disclose their trading activity in two business days (Veenman, 2012).
4 on prior research. Previous study from Lakonishok (2001) implies namely that insider
purchases consist of more private information than the insider sales.
Three previous studies about insider trading and the predictability of future cash flows
in the case of insider trading are relevant for this thesis. A study by Seyhun (1988) shows that
there is a positive correlation between aggregate trading activities and the return of the market
portfolio. Seyhun (1988) also argues that to a certain extent market returns remain predictable
even if the insider information has become publicly known. Conclusion is that there is not always a difference in effects of industry wide, firm specific and economy wide components.
This thesis shows evidence of predictable market returns in the case of future cash flows. This
study contributes to the study from Seyhun (1988) because this research emphasizes on both
variables earnings and cash flows. The reason for this particular research with also earnings
being tested is that cash flow brings timing and measure problems. This makes it interesting
to look at both variables instead of only cash flows. Another study by Piotroski and Roulstone
(2005), related to this thesis, states that insiders are possessors of superior information and
trade on the basis of contrarian beliefs. They found different relations for insider trades with
respect to some accounting variables. This study does not only checks book-to-market ratio’s
but also size and momentum. This gives another effect on the relation between future cash
flows and insider trades. While the separation between cash flow and earnings might have an
extra impact. The last more or less interesting study is by Veenman (2012) who concludes
that companies with more information uncertainties, insider purchase fillings trigger more
positive market reactions. Furthermore it is suggested that market reactions to purchase filings
are predictably associated with earlier earnings changes. This says something about the
persistence of accrual. Veenman’s study is focussed on all different companies whereas this
5 the effect of future earnings information is not tested with randomly chosen companies, but is
selectively tested with branch related industry.
This paper is organized in five sections. In the second section some existing literature about
the field of this research is discussed. The third section describes the data and research
methodology that will be used. The fourth section analysis the results obtained from the
regression made. Finally the last section is the conclusion combined with some implications for further research in this topic.
2. Literature Review
Over the last decades many studies have been carried out to determine different trading
activities of insiders with private information. A considerable input was delivered by Cohen et
al (2012), who was trying to show the information flow from insiders. According to Cohen et
al (2012) decoding the information flow from insider trades, as well as some insights about
the type of information and environment, is achievable. They classified trading between
opportunistic and routine trading, by doing this they isolate the information that insiders have
about the future of the firm. Cohen et al (2012) show that ‘’routine’’ insiders are not informative about a firm’s future and so the abnormal returns are zero, while opportunistic
traders yield an abnormal return and can predict future news and events. Cohen et al (2012)
argue that insider trades can be informative to investors. This can be used to search for other
insights about insiders activities. This is possible be course insiders have to disclose their
trading activities which means that the trades are visible for all kinds of people. Related with
this was the study from Brochet (2010) which concluded that the information flow from
insider trades was as Cohen et al (2012) argued.Brochet (2010) found evidence that after the
introduction of the SOX the abnormal returns and trading volumes increased and expedited in
the case of insider purchases. Thus the SOX has brought more transparency with regard to
6 A further important contribution to insider trading insights was made by Seyhun
(1988), who provided theoretical evidence that their generally exists a relationship between
aggregate trading activities and the return of the market portfolio. The return to the market
portfolio during the future two months is positively correlated with net aggregate insider
trading activity in a certain month. This means that insiders react with an increase in their
stock purchases prior to increases in the stock market. But the stock purchases decreases by
insiders, following increases in the stock market. Evidence for this conclusion comes from
observed mispricing by insiders and a part of that comes from economy wide changes that are
unanticipated. This explains that even after the information from insiders became public, the
future market returns still were predictable to some extent. Brochet (2010) analyzes around
the SOX regime, the information content of Form 4 filings. He found that after the SOX
regime the trading volumes and abnormal returns are significantly greater around filings of
insider stock purchases. This understanding and achievement in information about insider
activities triggered more research towards the relation between insider activities and future
market returns. For example Lakonishok (2001) who did similar research. Evidence showed
that insiders are able to predict cross-sectional returns and that especially purchases from
insiders reveal more information because selling has no predictive power.
Additionalstudiesfrom Seyhun (1998), Jaffe (1974) and Jeng et al. (2003) were built
on previous studies which proved the positive relation between trading and the returns. Their
focus was on abnormal returns over several periods. Seyhun (1998) shows that there is a
correlation between abnormal returns and the level of trading frequencies. The same applies
for Jaffe (1974); insiders trade on the basis of their extra information and on their ability to
identify mispricing in their own firms. Jeng et al. (2003) base their theory on performance
evaluation and found evidence for significant abnormal returns for insider purchases. This
7 researchers intended to examine more detailed evidence about abnormal returns through
insider trading. They build on Seyhun’s (1988) research who made a huge step forward in the
ongoing insight about insider trading. Several papers concentrate on other aspects of insider
trading namely the type of information and how these different types are reflected into stock
prices. Studies focussed on firm-specific, industry-level and market-level information. As
Seyhun (1988) states, there is not always a difference between the effects of firm-specific and
economy wide factors due to the examined relation between aggregate insider trading and
market movements. According to King (1966) there exists a correlation between market and
industry movements with stock prices. Whereas Roll (1988) argues that a part of the stock
price is not referable to market or industry movements but has something to do with
firm-specific information. These different approaches of looking at market returns are contraire
with the positive relation that Seyhun (1988) found earlier.
Research from Piotroski (2004) extends the investigationsfrom the studies above with
synchronicity but taking the three different types of information into account. He measures
stock return synchronicity through which extent trading activities can influence the stock
prices. When industry and market wide information is in cooperated in the study there will be
more synchronicity, while the opposite will occur when only firm-specific information is
used. Piotroski (2004) found evidence for a negative relation between stock return
synchronicity and insider trades. They also examine the timing of incorporation of
industry-level and firm-specific earnings into prices on the basis of different activities of these
informed parties. Piotroski (2004) found that only the firm-specific component of future
earnings news incorporated in the price is accelerated by insider trading. A year later Piotroski
(2005) looked at various types of firm events which might influence market returns. He
examined whether there is a relation between future cash flow news and insider trades.
8 each case. For example; insider trades are positive to future earnings performance, positive to
the book to market ratio and negative to returns. This implicates that insiders trade on the
information they have about future cash flows but only a small proportion of insider
purchases is explained by superior information about future cash flows. They also found a
stronger relationship between future earnings and insider purchases in weak information
environments. In this research Piotroski (2005) shows that this predicting of accounting
variables also could be important in the case of insider trading. Piotroski (2005) argues that
there is a lot to learn from investigating insider transactions. Not only the relation between insider trades and market reactions can be tested, as Seyhun (1988) did but also the
predictability of accounting variables can be examined (Piotroski, 2005). An example is the
more detailed paper of Ke et al. (2003) about insider trades consistency with cognition of
future earnings over a following period of earnings increases. Ke et al. (2003) argues that
insiders trade to profit because of the anticipated earnings trends. These trends are up to two
years in the future and differ for sales and purchases. Like Seyhun (1998) showed, insiders
much rather buy than sell when positive news comes out, which is described to the difference
of in formativeness.
Predicting future performance
More general research about the predictions of future cash flows and earnings is often
conducted because it seems to be an important question underlying financial reporting. Bowen
et al. (1986) examined the relation between various measures of cash flows and earnings.
They argued that earnings is not exactly a good predictive variable for future cash flows
compared with past cash flows. While Greenberg et al. (1986) had a different conclusion, they
concluded that earnings have more predictive power than cash flows. A later paper from
9 year-end based. This difference makes the predictability of cash flows out of accruals possible
for quarterly based amounts. All of the above research was based on general predicting power
of accounting variables. This,together with taking into account the predicting power in the
situation of insider trades, is relevant for this study. Examiningthe different studies from
above together might give a reason to think of that accounting variables in case of insider
transactions is predictable industry-wide.
Last, Veenman (2012), the study which is mostly related to this paper: in how they
approach their investigation with respect to their information and data. Veenmans (2012)
results show that in situations of uncertainty about the implications of valuation of reported
earnings, the Form 4 filings provide useful and important information to investors. The
persistence of previously reported earnings can be learned by investors from the Form 4
filings. Insider purchases give also signals about private information with respect to the
valuation implications of both previous, as well as future earnings. To conclude: Veenman’s
(2012) study argues that the market reactions to insiders are a combination of future earnings
anticipation and the persistence and implications of valuations of past reported earnings.
Veenman (2012) explains this by the fact that firms with higher information uncertainty,
market reactions to Form 4 purchase filings are far more positive. Veenman (2012) uses the
earnings response coefficient model to provide evidence about resolving the problem of
earnings information uncertainty. The outcome of this model shows that with past earning
changes for purchases that have positive earnings changes, the market reactions are positively
related, while negative market reactions will occur with negative changes. This means that
concerning the implications for future earnings, investors adjust their expectations and that a
part of the reaction is a delayed reaction to previously earning news. The larger the size of
insider purchases or the relative number of shares traded, the stronger the connection. This
10 studies. It includes all the ingredients that is needed for this paper but the contribution of this
paper is that it makes a distinction between accounting variables which might have a different
outcome because of the time problems of cash flow.
Based on the results of prior empirical research that has been done over the last decade, it is
expected that earnings will be a good variable to predict compared to cash flows.Earnings are
the net benefits of a firms operation, which is a given amount. Whereas cash flow is the
movement of money in or out a business. The cash flow amounts are daily changing which
might come from unexpected expenses or incomes. That’s why it makes it hard to predict
such movements in time. Therefore earnings might be a better variable to predict than cash
flows. And so the hypothesis to be tested is that the coefficient of B1 according to earnings
will be greater than that of B2 (B1>B2). This means that is expected that the variable earnings
has predictable influence industry-wide.
3. Data & Research Methodology
This section will formulate how the data required for this research was gathered. After that it
describes the conditions the data has to fulfil and the exact data that was selected. Finally this
section defines the methodology of this thesis.
3.1 data
The used datasets all comes from the database Wharton Research Data Services (WRDS).
From CRSP the data of adjusted returns were found, this is done as follows. Select the daily
stock file and use as firm identifier PERMNO for a later match with Thomson Reuter. Out of
this data the holding period returns will be calculated on basis of size/ book-to-market ratio
and momentum.
Insider trading data will be collected from Thomson Reuter at the insider filing data
11 table 22. Select NCUSIP as the firm identifier which will be merged with the data from CRSP
with identifier PERMNO. This paper only takes into account insider purchases, this means
that all the other transactions will be excluded from the equation.
The dataset of COMPUSTAT will be used to find the data for earnings, cash flows and
the control variables. In this paper four control variables will be used consisting of reporting
lag, loss, tax and R&D expenses. This is done by selecting North America, fundamental
quarterly and searching for the entire database. The output will be extended with quarterly
data items and using thereby CUSIP as the firm identifier. Another remark is made here by
the fact that only the variables earnings and cash flows plus accruals will be examined as
predictable independent variables.
The data collected from those three files will be separately filtered and merged so that
they have the same identifier and are as one document in STATA. All the exact variables
from those three files are defined in Appendix A.
3.2 Sample
The period on which this study focuses is from September 2000 till August 2012. The data
gathered from WRDS was subjected to some adjustments in order to get sufficient
observations. The following insider transactions have been dropped. First the option related
sales were excluded because this would lead to misclassification of purchases and counting it
double (Veenman, 2012). Second, those not related to common stocks observations. Third, the
amended fillings. Fourth, if more than one insider trades per security date. Lastly, security
dates from mixes. The observations left over were summed up and give an insider trading file
with one observation per person per security date. Over the complete sample several filters are
2 Header contains in normal cases the business address, name, phone and position of the insider. Table 1 contains non-derivative transaction information, it reports the number of shares bought/sold, price paid/received, transaction date and the shares remaining after the trade. Table 2 contains derivative
information and includes option exercises and option grants. The type of information, strike price, number of shares, date of expiring, date the options vest and holdings are reported in it.
12 applied to ensure consistency of the data. Those with invalid elements and numerous missing
are deleted, likewise incomplete data where price and return data were not available on the
transaction date. Finally with merging the files together some extra observations were lost.
This procedure results in a final sample of 490,689 observations.
3.3 Methodology
In order to find out if earnings is a better variable to predict industry-wide than cash flows and
indirectly investigate how far in advance you can predict it – this is done by choosing
different periods - an equation is set up. As mentioned earlier, the focus of this thesis, for
what the model concerns, relies on Veenmans (2012) paper. For testing the hypothesis a
measure for the dependent and independent variable has to be chosen. As dependent variable,
the left hand side of the equation, the adjusted holding period return of firm i will be used.
The independent variables, right hand side of the equation, are individual insider purchases
industry-wide multiplied by earnings of firm i and the control variables. Another equation has
to be made to compare the holding period returns, the second equation has the same
dependent variable just as the control variables but another independent variable. This
equations right hand side is individual insider purchases now multiplied by cash flows plus
individual insider purchases multiplied by accruals. The equations looks as follows.
Equation 1:
{ }
{ }
13 Equation 2: { } { }
Where holding period return (HPR) is defined as the total return received on an asset portfolio
by holding on to the portfolio for different periods. This computation is calculated later on in
this thesis. The variable Sum shares security day represents the individual insider purchases.
The sum is measured by the sum of all shares traded in a given firm at a given security date.
Further QEARN is included which is Earnings which comes from compustat described as
income before extraordinary items (IBQ) and is deflated by lagged total assets (ATQ).Cash
flows also come from compustat defined as net cash flows from operating activities
(OANCFY). Accruals were generated and calculated as income before extraordinary items
(IBQ) minus cash flows from operating activities. The paper controls for some factors that are related to insider trading and firm’s information environments. First there is controlled for
reporting delay, under the name of REPORTINGLAG computed as security date minus
transaction date. TAX is included as a dummy and defined by transaction month. TAX is one
if transmonth equals 12. Finally, the variables losses for firms (IBQ) and research and
development expense (XRDQ) were included as dummy variables. This is done be course of Veenman’s (2012) study which says that the use of Earnings for forecasting future cash flows
is complicated by those two variables3.
As mentioned before the holding period returns, which is a proxy for new information,
need to be calculated. WRDS will give the output for the daily returns. Therefore for this file
some calculations had to be made to get the correct output. In order to acquire the daily
abnormal returns, which is the characteristic adjusted returns based on size, book to market
3 All variables are defined in Appendix A
14 and momentum, some computations are made. To get the abnormal returns, first stocks are
assigned to one of the Fama-French 5 x 5 portfolios. This portfolio is the result from the
returns that are adjusted, which is the intersection of five portfolios based on size and five
portfolios based on book-to-market value.4 Then, to obtain daily abnormal returns, the
portfolio returns are subtracted from individual stock returns. From these obtained adjusted
returns the holding period returns were calculated through the formula (1 + return stock i day
t) till (1 + return stock i day t-1) because different periods are used.
4. Descriptive statistics and Results
4.1 Descriptive statistics
Table 1 presents descriptive statistics of observations of insider purchases. The different time
periods for holding returns are summarized together and gives a good view of the returns
further in time. Looking at the mean values for HPR, the values will be negative over the
holding time. The first quarter of insider purchases industry-wide is associated with a negative
holding period return which become even worse over time. Although 25 percent of the
purchases in third quartile has at least a response of 0,11. According to the results of the event
study of Veenman (2012), the numbers are much higher and thus the effect is stronger. But
not all the market reactions are positive (Q1 values of CAR%), 25 percent is associated with
CAR%. The main variables sumsharesecday, oancfy and accruals are highly skewnessed as
single variables. For the regression oancfy, accruals and qearn will be multiplied by
sumsharesecday. While among the main variable qearn as percentage of lagged total assets,
the average is around 25.33 percent. Veenman’s (2012) results show that Qearn is centered
around zero, which means on average zero. This might suggest that seasonally differenced
changes in quarterly earnings plays a role. For the control variables, lossdummy is positive on
15 average with 5.04 percent of the firms reporting a loss. Tax is positive with a percentage of
7.11 and the research and development expense is around 0.1 percent. The mean(median)
reporting delay equals 13.5(1.00) so this means that reporting occurs mostly on time. Loss and
rnd are much higher (rnd 37.6 and loss 36.0) in the study of Veenman (2012). This means
Loss and rnd have influence in the equation because they are both very significant. While in
this thesis (loss 5.04 and rnd 0.1) the positive percentage are very small.
Table 2 show the correlations based on pearson and spearman, which are pairwise and
rank correlations between the variables used in table 1. The different periods for HPR are all
positive related with insider trading industry wide(sumsharesecday), cash flows(oancfy) and
research and development expense(rnd). Tax becomes positive related as HPR reaches 100
days, on the other hand reportinglag changes over time for the different periods of HPR.
Negatively related to HPR are qearn, accruals and lossdummy. The correlations between this
study reasonable correspond with the correlations from prior research (Veenman, 2012).
Further insider trading industry-wide(sumsharesecday) is positive related to cash flows and
research and development expenditures which is consistent with expectations.
Sumsharesecday is also negatively related to the variables qearn, accruals, tax, loss and
reportinglag. Finally, the main variables qearn and cash flow are positive correlated. This is as
16 Table 1
Descriptive statistics for insider purchases sample 2001-2011
Stats Obs. Mean Std. Dev. Q1 Median Q3
Dependent var. HPR 20 145718 .002128 .1163804 -.019321 0 .0116347 HPR 40 144269 .0008772 .1732285 -.032665 0 .0128443 HPR 60 143540 .0005033 .2207013 -.044599 0 .012351 HPR 80 142476 -.0002007 .2630922 -.054648 0 .0081604 HPR 100 140743 -.0009955 .3149383 -.065309 0 .0011011 HPR 120 139812 -.001088 .3763875 -.076148 0 0 HPR 140 138787 -.0014705 .4543483 -.085005 0 0 HPR 160 137277 -.0050226 .4694184 -.093905 0 0 HPR 180 136146 -.0066035 .4583878 -.100723 0 0 HPR 200 135049 -.0072653 .4755926 -.106321 0 0 HPR 220 133546 -.0086347 .5220964 -.112719 0 0 HPR 240 132653 -.0091062 .5781428 -.119453 0 0 Main var. sumsharesec 319160 -30095.7 1773988 -3749.99 0 2000 qearn 149142 .2533054 13.88436 0 .002195 .0152522 oancfy 133934 -69.97375 1378.562 -.949 0 0 accruals 132431 111.4022 1577.195 0 .149 6.354 Control var. lossdummy 163460 .0504099 .2187899 0 0 0 rnd 92653 .000572 .0239104 0 0 0 tax 319160 .0710678 .2569385 0 0 0 reportinglag 319160 13.50342 96.07889 0 1 4
56
5 Bold text indicates that the correlations of pearson and spearman are significant at a level of 0.05 or better 6 The pearson correlations are below the diagonal and spearman correlations are above the diagonal
Correlation. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1HPR20 — .471 .344 .316 .292 .287 .244 .208 .204 .188 .176 .192 .028 -.138 -.137 .092 -.192 - - -.134 2HPR40 .694 — .611 .503 .495 .419 .344 .356 .320 .288 .268 .284 .072 -.166 -.166 .112 -.178 - .063 -.057 3HPR60 .527 .811 — .710 .647 .547 .503 .515 .463 .408 .395 .380 .064 -.214 -.213 .124 -.223 - - -.057 4HPR80 .445 .694 .873 — .746 .647 .555 .559 .523 .475 .432 .408 .036 -.246 -.246 .156 -.236 - - -.032 5HP100 .381 .586 .734 .863 — .766 .690 .655 .571 .531 .527 .519 .052 -.296 -.296 .144 -.286 - - -.070 6HPR120 .319 .482 .606 .723 .916 — .790 .691 .631 .583 .531 .539 .072 -.262 -.261 .132 -.261 - - -.019 7HPR140 .270 .394 .495 .604 .830 .940 — .806 .738 .706 .662 .639 .052 -.273 -.273 .216 -.263 - - -.044 8HPR160 .263 .392 .492 .601 .812 .909 .961 — .782 .687 .643 .635 .056 -.345 -.345 .232 -.339 - - -.044 9HPR180 .265 .399 .497 .601 .788 .874 .914 .960 — .802 .726 .710 .044 -.333 -.333 .204 -.312 - - -.082 10HP200 .256 .401 .500 .602 .768 .838 .867 .916 .962 — .814 .758 .068 -.269 -.269 .148 -.245 - - -.095 11HPR220 .233 .356 .443 .537 .737 .824 .875 .907 .932 .958 — .865 .079 -.293 -.293 .148 -.263 - - -.108 12HPR240 .209 .319 .402 .497 .712 .813 .864 .893 .906 .923 .964 — .056 -.293 -.293 .188 -.263 - - -.083 13sumshar .005 .003 .002 .001 .002 .003 .002 .002 -.000 -.000 -.001 -.000 — -.038 -.038 -.028 -.031 - .063 -.121 14qearn -.018 -.013 -.012 -.004 -.003 -.008 -.011 -.009 -.008 -.007 -.007 -.013 -.000 — - -.227 .893 - .063 .312 15accruals .006 -.014 -.006 -.007 -.008 -.009 -.013 -.007 -.007 -.007 -.005 -.006 -.033 -.001 — -.229 .893 - .063 .312 16oancfy .018 .011 -.000 .002 .004 .004 .006 .003 .003 .004 .005 .005 .030 .001 -.923 — -.181 - -.063 -.108 17reportl. .001 .002 -.002 -.001 -.000 -.001 .000 -.001 .000 .001 -.000 -.001 -.001 .002 .004 .012 — - .071 .277 18lossdum -.004 -.008 -.012 -.011 -.009 -.018 -.019 -.017 -.031 -.029 -.028 -.026 -.002 .006 .003 .004 .012 — - - 19 rnd -.005 .031 .005 -.001 -.002 .017 .008 .022 .032 .027 .032 .024 .016 .004 -.001 -.001 .000 .004 — - 20tax .010 -.006 -.003 -.004 .001 .008 .011 .010 .011 .009 .011 .016 -.003 .002 .038 -.000 .027 .125 -.00 —
Table 3-6 reportthe testing results from the two different regressions. First both equations are
estimated through using the OLS regression function. Second, the option robust is used
because homoscedasticity and normality of the random error terms is not assumed. Finally a
two-tailed significance is used. Table 3 first shows the results of the basic model without the
control variables of equation 1 and table 4 shows the full model with the control variables for
equation 1 and so on for equation 2. All tables contains seven models which means the
different holding period returns as dependent variable. Some different periods are given to illustrate the course of the holding period return. For table 3 and 4 the regression coefficient
should be interpreted as the difference in holding period returns and the predictability of
earnings as industry insider purchaser.
Table 3find a strong association between insider purchases predictability of earnings
and the different periods for holding return. This shows that equation one on his own without
any control variables is significant, looking at the t-statistic. The significance is stronger if the
holding period return increases and starts at HPR80. This findings suggest that the effect of
earnings on the holding period return industry-wide is predicted well. The different period
returns show that earnings in this study will be predicted well for a bit more than eight months
in front. This strong relation between future earnings and insider purchases might indicate a
weak environment as Piotroski (2005) showed in his paper. This study also finds really small
coefficients around 0.001, which tells the slope for insider purchases in case of the
predictability of earnings at the different periods. Table 3 and 4 show similarities between the
basic model and full model but also some differences. From table 4, the full model, there is
only statistical significance from holding period return 160 at the level of 10 percent or better.
Although this model show return differences for qearn, the effect of the holding period returns
on the predictability of earnings industry-wide is predictable but from later on. This evidence
19 earnings is predictable after nine months. Which is noteworthy because Cohen et al (2012)
suggest that the holding returns become less after nine months. If lossdummy and reportinglag
are left out of the regression the significance for the variables is similar. The effect of leaving
out those two control variables work stronger as the holding period returns are larger. Again,
the full model also show small coefficients for insider purchases times earnings around
0.0013 with some outliers of 0.0033. The coefficients of qearn are between 0.0007 and 0.002,
which is also small. The effect of leaving out lossdummy and reportinglag does not change
the coefficients by much. For the different control variables there is some significance, for tax
in the beginning periods and for reportinglag later on the holding returns. The coefficients on
the control variables suggest further that new information is increasing with firms having
research and development expenditures (rnd) which become more if the holding period return
increases. Holding period returns are lower in industry-wide firms with positive tax amounts,
increases till the holding period is at half and becomes negative at 160 days by firms having a
reporting delay.
Overall evidence show that equation one, controlled for the environments, on his own
is statistical significant which means that the effect of earnings on holding period return is
good predictable and can be predicted, as this research shows, in advance till 240 days.
Table 3 basic model equation 17
OLS regression HPR20 HPR40 HPR80 HPR120 HPR160 HPR200 HPR240 Insider x earnings 0.39 -0.63 4.23 4.23 6.53 8.71 3.65 (0.695) (.531) (0.00) (.000) (0.00) (0.00) (.000) qearn -20.31 -0.85 3.39 2.85 4.47 6.40 2.01 (0.757) (0.397) (0.001) (0.004) (0.00) (0.00) (0.045) intercept -.68 -.23 -.42 -1.78 -1.70 -1.23 -1.29 (.497) (.82) (-.674) (.074) (.09) (.219) (.197) N 64011 63615 63264 62813 62511 62285 62059 R-squared .0003 .0001 .0001 .0002 .0002 .0001 .0002
20 Table 4 full model with control variables equation 189
OLS regression HPR20 HPR40 HPR80 HPR120 HPR160 HPR200 HPR240 Insider x earnings 1.34 0.62 1.56 2.53 3.11 7.64 7.70 (0.180) (.536) (.119) (.012) (.002) (0.00) (0.00) Qearn -0.84 -0.94 3.03 1.46 2.68 5.54 -2.64 (0.399) (0.345) (0.002) (0.144) (0.007) (0.000) (0.008) Insider (sumshare) -0.56 -2.24 -2.07 -1.96 -2.65 -1.38 -1.74 (0.577) (0.025) (0.039) (0.050) (0.008) (0.166) (0.082) lossdummy - - - - rnd -.45 2.91 -.23 1.37 1.36 1.41 1.15 (.654) (.004) (.816) (.169) (.175) (.160) (.251) tax -1.95 -1.23 -1.97 -1.99 -0.86 0.04 -0.42 (.051) (.220) (.049) (.046) (.389) (.965) (.676) reportinglag 1.26 1.33 0.86 0.06 -2.08 -2.01 -1.97 (.209) (.185) (.389) (.953) (.037) (.045) (.049) intercept -0.13 0.58 0.45 -0.20 -0.11 0.23 0.22 (.897) (.57) (.652) (.841) (.912) (.817) (.828) N 62023 61868 61744 61592 61475 61375 61288 R-squared .0056 .0017 .0005 .0011 .0009 .0010 .0009 Coeff insider .0013 .0014 .0024 .0012 .0012 .0033 .0010
Table 5 and 6 reports the results from testing equation two, table 5 show the basic model and
6 the full model. For table 5 and 6 the regression coefficient should be interpreted as the
difference in holding period returns and the predictability of cash flows as industry insider
purchaser. This thesis does not find a strong association between insider purchases accounting
for cash flows and the different periods of holding returns. Although at the holding period of
20, 160 and 240 it shows a significance coefficient. Looking at accruals, there is statistical
significance at 10 percent or better between insider purchases and holding period return,
starting from holding period 100 which has a p-value of 0.062. Because these predictions are
in the same line, some values are not reported in the table. OANCFY only show a statistical
significant outcome at the holding period of 240. This findings suggest that the effect of
8
This table does not report the main coefficients of qearn, lossdummy, rnd, tax and reportinglag. To save space.
21 holding period returns on cash flows industry-wide is not predicted well. This statement confirms earlier founding’s. Piotroski (2005) proves that only a small proportion of insider
purchases is explained by superior information about future cash flows. Those different time
periods point to predicting cash flows in advance starting at 240 days at least. The coefficients
of insider purchases accounting for cash flows and accruals are not worth mentioning here
because they are really small, they show positive small changes. If accruals and the
interaction term are excluded from the regression, the significance become even worse for the
predictability of cash flows.
From the full model, table 6,some difference in statistical significance controlling for
the environment now show up. The findings above hold for cash flows after controlling for
other factors related to insider purchases. This indicates that equation two has no predictable
power at significance levels 10 percent or better from the periods of 20 till 240 days. The
coefficient for cash flow with the control variables is around 0.0003 what is really small and
suggest a small change in the dependent variable. This is the same for accruals with an
average coefficient for the different periods of 0.0002. OANCFY coefficient is on average
0.0004, which is again small. The control variables showsome significance with reference to
the different holding period returns. The variable research and development expenditures
(rnd) is significant at holding period 40. Tax is significant at the beginning period returns and
reportinglag is significant starting at the holding period of 160 and further. The coefficients of
the control variables further points to different holding period returns. Firms with research
and development expenses facing lower holding returns in the periods from 20 till 40 and
from then it starts increasing for the different holding period returns. The holding period
returns are lower in industry-wide firms with a positive tax amount, increases till the holding
period is at half and becomes negative at 160 days by firms having a reporting delay.
22 does not give any significance to insider times oancfy. The coefficient is far in minus so it
does not contribute to the regression to do so.
Overall evidence shows that equation two, controlled for the environment and other
factors related to insider purchases, has no statistical significance. Which only means that the
effect of cash flows and accruals on holding returns is not good predictable for the days from
20 till 240 in this paper.
Table 5 basic model equation 210
OLS regression HPR20 HPR40 HPR80 HPR120 HPR160 HPR200 HPR240 Insider x oancfy 2.17 -0.35 1.26 1.40 2.17 0.16 2.71 (.030) (.726) (.209) (.162) (.030) (.873) (.007) Insider x accruals 2.48 -0.06 1.61 3.46 2.48 2.01 3.24 (0.013) (.950) (.108) (0.001) (0.013) (.044) (.001) OANCFY 1.71 1.74 1.17 1.51 1.71 1.20 3.09 (0.087) (0.081) (.241) (.132) (0.087) (.232) (.002) intercept -1.11 0.08 -0.12 -1.22 -1.17 -0.80 -0.73 (.269) (.933) (.902) (.221) (.240) (.424) (.462) N 63452 63135 62844 62480 62224 62031 61853 R-squared .0005 .0001 .0001 .0001 .0001 .0001 .0001
Table 6 full model with control variables equation 21112
10 This table does not report the main coefficients of insider x cash flows, insider x accruals and oancfy. This to save space.
11
Again this table does not report the main coefficients of insider x accruals, oancfy, lossdummy, rnd, tax and reportinglag.
23 OLS regression HPR20 HPR40 HPR80 HPR120 HPR160 HPR200 HPR240 Insider x oancfy -0.48 0.16 0.18 1.08 1.30 1.05 1.68 (0.633) (0.870) (0.857) (0.278) (0.193) (0.294) (0.093) Insider x accruals -0.44 0.13 0.21 1.14 1.36 1.11 1.73 (0.659) (0.896) (0.835) (0.253) (0.175) (0.268) (0.083) OANCFY 0.12 -0.36 -0.78 0.75 1.38 0.94 1.07 (0.904) (0.716) (0.433) (0.454) (0.167) (0.350) (0.286) lossdummy - - - - rnd -0.42 2.94 -0.22 1.40 1.37 1.41 1.16 (.673) (.003) (.827) (.161) (.170) (.157) (.248) tax -1.98 -1.24 -1.98 -2.00 -0.85 0.05 -0.39 (.048) (.217) (.048) (.045) (.398) (.958) (.695) reportinglag 1.25 1.33 0.86 0.06 -2.08 -2.01 -1.97 (.210) (.185) (.389) (.953) (.038) (.045) (.049) intercept -0.18 0.58 0.46 -0.18 -0.10 0.24 0.22 (.854) (.564) (.643) (.858) (.922) (.807) (.825) N 62016 61862 61739 61588 61472 61372 61285 R-squared .0051 .0017 .0004 .0010 .0009 .0010 .0010 Coeff insider -.0001 .0001 .0003 .0002 .0002 .0002 .0004
Finally the results from tables 4 and 6 has to be compared for answering the actual hypothesis
of B1 > B2. There can be seen a lot of similarities when it comes to the control variables in
table 4 and 6 but also differences between the market reactions in case of accounting for
earnings or cash flows.
In both cases it shows that, looking at significance, how longer the holding period is
how stronger the predictability is by insider purchases industry-wide for earnings or cash flow
in each period of time. This suggest that there might be a pattern for insider purchases
predictability as you move on the days of holding period. From equations one it shows that as
an industry insider purchaser the predictability of earnings has the biggest influence on the
regression. While for equation two the different control variables have the most influence
dependent on the holding period return. Remarkableis that the main variable for equation two
is not that strong which probably come from the fact that the environment generate more
uncertainty about future cash flows. The two equations are different in overall significance.
The equation according to earnings suggest a significance, which mean earnings is good to 12 Also the main effects of cash flow and accruals are not reported in this table.
24 predict industry-wide by insider purchasers. The relation becomes even stronger when the
holding period returns is longer. Although equation two do not show any significance. This
does not correspond with prior research of predicting future cash flows in general (Willinger
and Lorek, 1996). They made a distinction between quarterly and year-end based amounts
which leads to predicting cash flows out of accruals for quarterly based amounts. This
difference in outcome might come from predicting in general and predicting in case of insider
purchases because Piotroski (2005) found a different result. Looking at the significance of the
control variables there seems to be not a lot of difference between table 4 and 6, the
coefficients are almost the same in both equations. The influence of the control variable is the
same for both equations but works stronger in equation two. Which means that equation two
is more dependent of the control variables.
The coefficientsofinsider times earnings and cash flows at the different time periods
suggest that even though the changes are really small there is a small positive change of the
dependent variable, holding period return. For equation two there might be a pattern if it
comes to insider purchases accounting for cash flows after 240 days. Whereas equation one
shows another pattern with big outliers in some time periods. Comparing the amounts of the
coefficients for earnings and cash flows it shows that the results are in line with the
hypothesis, suggesting that the value of earnings is bigger than for cash flows. This result
consistent with Greenberg et al. (1986): Theycome with the result of earnings with more
predictive power than cash flows for general accounting not specific in case of insider
purchases, which might suggest that it is related. As can be seen the values will converge at a
certain point in time because for earnings it is going downward whereas for cash flows it
looks like it is growing. Also the coefficient for the control variables are almost the same in
both equations. The two regressions indirect show the predictability of earnings and cash
25 advance predictability. For earnings and cash flows it seems that both can be predicted for a
certain term, this model only shows till 240 days in front.
Based on the compared results there is concluded that the hypothesis is supported by
the results from table 4 and 6, that says that B1>B2 for all the different holding period returns.
5. Conclusion
This paper examines to what extent individual insider’s purchases can forecast the accounting
variables earnings and cash flows from companies in the same industry. For the entire
database and 490,689 observations in the period of September 2000 till August 2012. This
study provides preliminary evidence with respect to the predictive power of insider purchases.
So the hypothesis is not formally tested because it just looks if the coefficient of earnings is
bigger than cash flows. The conclusion is below.
At first as can be seen from regression one, is that the relation found in this research
between the holding period return and the predictability of earnings is significant. Which
means that as an industry insider purchaser, earnings have indeed predictable power.Whereas
the relation between the holding periods return and cash flow predictability is not significant.
A conspicuous remark made here is that the relation between insider purchases accounting for
cash flows and holding period return is not significant and thus is not in line with the
expectations of this paper, while prior theory came to different conclusions. The different
outcome might come fromthe controlled variables in this thesis.Final results show that there
is strong support for the hypothesis which says that as an individual insider purchaser,
earnings is better predictable in the future than cash flows. Another indirect argue that is made
in this thesis is the future predictability for a certain term. This thesis shows future
predictability till at least eight months in front. The main conclusion is B1> B2 and therefore
as an individual insider purchaser, earnings is a better predictable variable compared to cash
26 The major difference between this study and that of Veenman (2012) is that this paper
is focused on firms in the same industry. The event-study of Veenman (2012) is different than
this study because it measures the market reaction while this thesis focusses on holding
periods return as proxy for new information. The time of this paper matched the best of that
with Veenman (2012).
From this paper also some limitations or implications are clear, as the high skewness
of some variables so the natural logarithm had to be taken, the small coefficients are not easy
to compare so standardized coefficients had to be made and the lossdummy is omitted in this
regression which means it was not taking into account. Finally, the control variables
lossdummy and reportinglag must be excluded, it will give a significance different
conclusion.
Although the find result is the one expected, there might be still a lot of research in this area
which can be gathered.
References
- Barth, M. E., D. P. Cram and K. K Nelson. 2001. Accruals and the prediction of future cash
flows. The accounting review.
- Bowen, R. M., D. Burgstahler, and L. A. Daley. 1986. Evidence on the relationship between earnings and various measures of cash flow. The accounting review.
27 - Brochet, F. 2010. Information content of insider trades before and after the Sarbanes-Oxly
Act. The accounting review.
- Cohen, L., Malloy, C., Pomorski, L. 2012. Decoding inside information. The Journal of finance.
- Greenberg, R. R., G. L. Johnson, and K. Ramesh. 1986. Earnings versus cash flow as a predictor of future cash flow measures. Journal of accounting, auditing, and finance.
- Jaffe, Jeffrey F. 1974. Special information and insider trading. The Journal of Business.
- Jeng, Leslie A., A. Metrick, and R. Zeckhauser. 2003. Estimating the returns to insider
trading: A performance evaluation perspective. The review of economics and statistics.
- Ke, B., S. Huddart, and K. Petroni. 2003. What insiders know about future earnings and how they use it: Evidence from insider trades. Journal of Accounting and Economics.
- King, B. 1996. Market and industry factors in stock price behaviour. The journal of business.
- Lakonishok, Josef, and Inmoo Lee. 2001. Are insider trades informative? Review of
financial studies.
- Lorek, K. S., and G. L. Willinger. 1996. A multivariate time-series prediction model for cash flow data. The accounting review.
- Piotroski, Joseph D. and Roulstone, Darren T., 2004.The influence of analysts, institutional investors, and insiders on the incorporation of market, industry and firm-specific information into stock prices. Journal of accounting and economics.
- Piotroski, J., and D. Roulstone. 2005. Do insider trades reflect both contrarian beliefs and superior knowledge about future cash flow realizations? Journal of Accounting and Economics.
- Roll, R. 1988. R-squared. The journal of finance.
- Seyun, H. Nejat. 1988. The information content of aggregate insider trading. The Journal of business
- Seyun, H. Nejat. 1998. Investment intelligence from insider trading. The journal of finance.
- Veenman, D. 2012. Disclosures of insider purchases and the valuation implications of past earnings signals. The accounting review.
Appendix A Variable definitions
HPR = holding period return, (1 + return stock i day t) till (1 + return stock i day t-1) – 1 for days 20, 40, 80,120,160,200 and 240.
SUMSHARESECDAY = individual insider purchases, measured by the sum of all shares traded in a given firm at a given security date.
QEARN = quarterly earnings (compustat’s data item IBQ) as percentage of lagged total assets (compustat’s data item ATQ).
OANCFY = net cash flows from operating activities on year base by compustat’s database. ACCRUALS = adding together different investments or interest over a period of time, measured as income before extraordinary items (IBQ) minus cash flows from operating activities (OANCFY).
LOSS = losses of firms using the level of cash flow from operations as proxy for losses, by compustat’s database IBQ. Variable equals one if it is negative and zero otherwise.
XRDQ = research and development expense by compustat’s database. Indicator variable equal to 1 when firm reports research and developments expenditures, zero otherwise.
LAG = reporting lag, measured as the difference between the security date and the transaction date of the insider.