ceo overcon dence and the long-term ... - 國立臺灣大學
TRANSCRIPT
CEO Overconfidence and the Long-TermPerformance Following Unexpected R&D Increases
Sheng-Syan Chena, Keng-Yu Hoa, and Po-Hsin Ho∗,a
aDepartment of Finance, National Taiwan University, Taipei, Taiwan.
October 25, 2010
Abstract
In this paper, we examine the relationship between unexpected increase inresearch and development (R&D) expenditure and CEO overconfidence. Pre-vious studies show a positively significant market reaction to the increase inR&D expenditure both in the short run and long run. However, our empiricalresults show that the positive long-run stock performance is found only for firmswith non-overconfident CEOs. In addition, we also provide evidence that un-expected increase of R&D expenditure is a beneficial investment decision onlyfor firms without overconfident CEOs. We suggest that the empirical findingsmay be due to overestimation of future cash flow and overinvestment by over-confident CEOs. Our study indicates that overinvestment behavior often leadsto value-destroying R&D investment projects.
Keywords: R&D Expenditure, CEO overconfidence, Long-Run PerformanceJEL Classification: G00; G14; G30
∗Corresponding author.Tel: +886-2-33663834; E-mail address: [email protected]; Fax: +886-2-23660764; Address:Department of Finance, National Taiwan University, No. 1, Sec. 4, Roosevelt Road, Taipei, 10617Taiwan.
1
1 Introduction
Many studies have documented that how the overconfident CEOs affect the firms’
investment decisions. The story of CEO overconfidence stems from a prominent styl-
ized fact, the “better-than-average” effect, in the psychology literature. Prior research
(Larwood and Whittaker (1977); Svenson (1981); Alicke (1985)) suggest that people
tend to overestimate their wisdom or skill relative to the average. People are more
likely to attribute good outcomes to their actions and bad outcomes to bad luck or
external factors. Previous studies consider that attribute the evidence of overconfi-
dence to three main factors: the illusion of control, a high degree of commitment to
nice outcomes, and abstract reference points which make it hard to compare perfor-
mance between individuals (Weinstein (1980); Alicke, Klotz, Breitenbecher, Yurak,
and Vredenburg (1995)). First, a CEO tends to be overconfident when he believes
that the outcome is under his control. Second, a CEO more likely to be overconfi-
dent when he is highly committed to firm’s good performance because his personal
wealth varies with the company’s stock price and he invests his human capital in the
company. Lastly, a CEO overestimates his ability of choosing profitable investment
projects when the reference point of comparing his skill relative to the average is
abstract.
Using the overestimation of future cash flow as the source of managerial overconfi-
dence, Malmendier and Tate (2005) propose a simple model and the empirically show
that managerial overconfidence can cause corporate investment distortions. They find
that the sensitivity of investment to cash flow is strongest in the presence of overcon-
fidence. In addition, Malmendier and Tate (2008) point out that CEO overconfidence
can help to explain merger decisions. Overconfident CEOs overestimate their skill to
generate returns. As a result, they pursuit value-destroying mergers and overpay for
target companies. Malmendier and Tate (2008) also suggest that, in general, over-
confident CEOs are more likely to engage in mergers and acquisitions. They find
that the possibility of making an acquisition are 65% higher if the CEO is classified
as overconfident, the magnitude is larger if the merger is diversifying and does not
require external financing. The market reaction at merger announcement done by
2
overconfident CEOs is significantly more negative than those by non-overconfident
CEOs.
Being different from Malmendier and Tate’s analytic framework, Ben-David, Gra-
ham, and Harvey (2007) provide a theoretic framework to analyze the relations be-
tween managerial overconfidence and corporate policies. However, they define the
CEO overconfidence as miscalibration. Based on their definitions, overconfident man-
agers who either underestimate the volatility of their firms’ future cash flows or use
lower discount rates than unbiased managers do. Nevertheless, they propose that
overconfident managers invest more than less confident managers. They empirically
use the narrowness of probability distributions for predicted future stock market re-
turns which are surveyed from hundreds of U.S. Chief Financial Officer (CFOs) as a
proxy for each respondent’s confidence, and the findings support their prediction.
Geol and Thakor (2009) also propose a theoretical model on the relations among
overconfidence, CEO selection, and corporate governance. In their paper, there are
three types of CEOs: excessively overconfident CEOs, moderate overconfident CEOs,
and excessively diffident CEOs. They predict that excessively overconfident CEOs
face a greater likelihood of forced turnover than CEOs with moderate overconfidence.
They further demonstrate excessively overconfident CEOs reduce firm value due to
overinvestment. Campbell, Gallmeyer, Johnson, Rutherford, and Stanley (2010) em-
pirically examine the prediction of Geol and Thakor (2009) and find strong support
to their model.
Other empirical studies focusing on the relevant research topics include Mal-
mendier, Tate, and Yan (2010) who examine the relation between overconfidence
and major financial decisions. In addition, Liu and Taffler (2008) study the relation
between overconfidence and M&A decisions, while Billett and Qian (2009) examine
the linkage between overconfidence and M&A frequencies. Finally, Hribar and Yang
(2010) examines whether overconfidence increases the issuance of overly optimistic
management earnings forecasts and greater earnings management.
Although, as mentioned above, there are many prior studies on various financial
decisions and CEO overconfidence, few studies investigate the related research issue
from R&D expenditures. To the best of our knowledge, the only two related studies
3
are Galasso and Simco (2010) and Hirshleifer, Low, and Teoh (2010), who study
the relationship between CEO overconfidence and the innovation behaviour. They
find that overconfident CEOs are more likely to make their firms towards innovation.
They also examine relation between CEO overconfidence and patent citations. The
results suggest that overconfident CEOs obtain more patents and patent citations.
Our study, on the other hand, deviates from their research in two important ways.
First, our study investigates the relation between CEO overconfidence and the effect
of unexpected increase in R&D expenditure. Second, we examine the long-term
abnormal stock return and operating performance following unexpected increase of
R&D from the perspective of CEO overconfidence. Both Galasso and Simco (2010)
and Hirshleifer, Low, and Teoh (2010) do not examine the effect of overconfidence on
firm performance following unexpected R&D increase.
R&D decisions of overconfident CEOs are an interesting research issue at least
for three reasons. First, R&D increases represent a managerial decision made by
CEOs and Eberhart, Maxwell, and Siddique (2004) argue that R&D increases are
beneficial investment decisions in the long run. From this perspective, we wonder
whether CEOs with overconfidence could also make profitable R&D investments, since
previous studies point out that managerial overconfidence result in overinvestment
(Malmendier and Tate (2005; 2008); Ben-David, Graham, and Harvey (2007); Geol
and Thakor (2009)). We thus conjecture that only CEOs without overconfidence
could make beneficial R&D decisions.
Second, past research (Chan, Martin, Kensinger (1990); Zantout and Tsetsekos
(1994); Szewczyk, Tsetsekos, and Zantout (1996); Eberhart, Maxwell, and Siddique
(2004)) suggest that market reactions of increase in R&D expenditure are positively
significant. However, the market investor reactions at merger announcement by CEOs
with overconfidence are lower than firms without overconfident CEOs in Malmendier
and Tate (2008). Hence, based on previous discussion on CEO overconfidence, we
conjecture that investors could identify the overconfident level of CEOs, and only
firms with non-overconfident CEOs could earn long-term abnormal stock returns.
Third, although R&D investment is one type of firm’s investment decision, it has
unique features that are different from other long-term investments. The information
4
contain of an R&D increase not only conveys tangible information but also reflects
intangible information about the prospect of future cash flow. Since overconfident
CEOs may overestimate future cash flow of their investment projects, it is intriguing to
further study R&D expenditure from the aspect of CEO overconfidence. Furthermore,
Simon and Houghton (2003) argue that CEO overconfidence is positively related to the
introduction of pioneering and risky products using survey data. They also indicate
that those products are less likely to achieve success. Because R&D expenditure may
provide a new technology development or create a new product, the influence of CEO
overconfidence might play an important role in the innovation decision of company.
We use various methodologies in detecting the long-term abnormal stock returns
and operating performance. For the abnormal stock returns, we find consistent evi-
dence that non-overconfident CEOs exhibit significantly positive long-term abnormal
stock returns following R&D increases. On the other hand, CEOs with overconfidence
could not earn significantly positive long-term abnormal stock returns following R&D
increases. For the aspect of operating performance, our findings indicate that R&D
investment are beneficial only when a firm’s CEO is not overconfident. In addition,
both the stock and operating performance results are robust for high-tech samples,
but weaker for low-tech samples.
Our research contributes the literature by combining CEO overconfidence and
R&D investment decisions. To the best of our knowledge, we are the first paper
to examine the CEO overconfidence from the aspect of unexpected R&D increase.
Second, comparing with previous research which indicates that investors could earn
positively abnormal stock returns in R&D increases decisions, we show that investor
could gain significantly abnormal stock returns for firms with non-overconfident CEOs
from the effect of increase in R&D expenditure. The phenomenons are due to overcon-
fident CEO overestimates the future prospect of investment project. Finally, we find
that R&D increases are beneficial investments only for firms without overconfident
CEOs.
The rest of the paper is organized as follows. In section 2, we describe the data
used in our paper. Section 3 introduces the methodology used in this paper. Section
4 analyzes the empirical results, and section 5 concludes the paper.
5
2 Data
2.1 Measuring Overconfidence
Our measures of CEO overconfidence are the same as a series paper of Malmendier
and Tate (2005; 2008). They employ the panel data of the CEO’s personal portfolios
in executive options exercise to identify whether CEO is overconfidence. CEO obtains
the right to purchase their company’s share from executive options, and the exercise
price of the options are usually at the stock price on the grant date. The duration of
most executive options is ten years, and the vesting period in which the options are
unexercisable are usually four years.
Merton (1973) derives that investors should not exercise their options early be-
cause the time value of options is non-negative for European call options on non-
dividend-paying stocks. However, this logic may not apply to executive options,
because executive options are non-tradeable and CEOs cannot short sell their hold-
ings to hedge the idiosyncratic risk of their company’s stocks. In addition, CEOs
also highly exposed their wealth to company risk since a large part of their compen-
sation is equity-based and they invest their human capital in the firms. As a result,
risk-averse CEOs should exercise option grants early if the stock price is sufficiently
high. Lambert, Larcker, and Verrecchia (1991) propose a theoretical framework to
show that risk averse CEOs should exercise in-the-money executive stock options be-
fore expiration to avoid their exposure to company-specific risk. Hall and Murphy
(2002) further derive that overconfident CEOs delay exercise their option grants until
expiration, even when the underlying stock price exceeds rational exercise thresholds.
In our sample, CEOs persistently deny to exercise highly in-the-money vested op-
tions until the year of expiration. The interpretation of this delay exercise behaviour
is overconfidence, i.e., overestimation of the firm’s future returns. Other alternative
explanations of this behavior such as positive inside information, signaling, board
pressure, risk tolerance, and taxes have demonstrated fail to interpret the delay exer-
cise behavior in Malmendier and Tate (2008). They provide strong evidence that the
empirical results are consistent with fact that overconfident CEOs think that they
can make profitable merger and acquisition activities. We adopt the same criterions
6
in Malmendier and Tate (2008) as the CEO overconfidence indicator and also name
the variable as Longholder. CEOs who, at least once during their tenure, keep an op-
tion until the year of expiration, even though the option is at least 40% in-the-money
during its final year. The exercise threshold of 40% is calibrated from the model of
Hall and Murphy (2002) which assumes a constant relative risk aversion (CRRA) of
three and 67% of his or her wealth in the company stock.1 We apply this measure as
a managerial fixed effect and it reflects that overconfidence is a persistent trait.
2.2 Sample Construction
The initial sample of our overconfident CEO dataset is identical in Malmendier and
Tate (2008),2 and this sample is constructed from Hall and Liebman (1998) and
Yermack (1995). The empirical sample covers 477 large publicly traded US firms from
the years 1980 to 1994. A firm must be included at least four times on the lists of the
largest U.S. companies constructed by Forbes magazine from 1984 to 1994. These
data are constructed by complete information on CEOs stock and option holdings
and provide an entirely detailed picture of the CEO’s portfolio rebalancing over his
tenure.
Due to the sample limitation of CEO overconfidence data, our research period cov-
ers from 1980 to 1994. The stock price data of our samples are obtained from Center
for Research on Security Prices (CRSP). We include only ordinary common equities
whose share type codes are “10” or “11” in CRSP to estimate the abnormal stock
return. Moreover, we retrieve the accounting data from COMPUSTAT database to
compute abnormal operating performance. We impose three major requirements fol-
lowing Eberhart, Maxwell, and Siddique (2004) to measure the unexpected increase
of R&D expenditures. First, an unexpected R&D increase is defined as a firm’s ra-
tio of R&D to assets increase. Second, we concentrate on firms whose R&D intensity
(defined as ratios of R&D to assets and ratios of R&D to sales) are greater than 5 per-
cent. Third, we only focus on an economically significant R&D increase. As a result,
1The particular choice of parameter values is not important for our results: according to Mal-mendier and Tate (2008), the median percentage in-the-money entering the final year for optionsheld to expiration is 253%.
2We are grateful to Professor Ulrike Malmendier who provides the CEO overconfidence data.
7
the firm must also increases its dollar R&D at least 5 percent and its ratio of R&D
to assets at least 5 percent. Finally, we yield 54 overconfident CEOs with significant
R&D increase and 97 non-overconfident CEOs with significant R&D increase.
We use annual accounting data and follow Eberhart, Maxwell, and Siddique (2004)
which use a three-month lag to allow the market to be informed of the accounting
data. Hence, we begin measuring these returns from the fourth month following the
fiscal year-end in which the firm increases its R&D.3 The definition of all variables
used in this paper are described in the Appendix.
3 Methodology
Although there is still much debate on the empirical estimation of long-term abnormal
stock returns, there are two popular empirical methods on computing long-run abnor-
mal stock returns: buy-and-hold abnormal returns (BHARs hereafter) of the event
firm relative to a benchmark and calendar-time abnormal returns (CTARs hereafter)
using a factor model to estimate the risk adjusted returns.
Barber and Lyon (1997) show that the arithmetic summation of returns do not
precisely reflect the returns of investors. Furthermore, Lyon, Barber, and Tsai (1999)
argue that the calendar-time return method would bias in nonrandom sample. How-
ever, Fama (1998) argues that event-time returns exhibit a cross-sectional dependence
problem that causes the downward bias of standard error.
Additionally, there is also a debate regarding the use of value-weighted calendar-
time returns versus equal-weighted calendar-time returns. On one hand, Loughran
and Ritter (2000) argue that equal-weighted portfolio is better, since the mispricing is
more likely to occur with smaller firms. On the other hand, Fama (1998) documents
that value-weighted portfolio is more appropriate because it considers the total wealth
effects experienced by investors.
3In addition, in order to rule out the repeated event of R&D increase, we also examine the samplewithout overlapping. In other words, a firm can only be included in the sample once during ourresearch period.
8
3.1 Buy-And-Hold Abnormal Return (BHAR)
We compute the one-, three-, and five-year BHARs for all sample firms with a three-
month lag following the unexpected R&D increase, respectively. For the ith sample
firm from month T1 to T 2, BHARi,t is expressed as follows:
BHARi,t =T2∏t=T1
(1 +Ri,t)−T2∏t=T1
(1 +Rb,t), (1)
where Ri,t is the return of the sample firm in event month t, and Rb,t is the return of
the benchmark over the same period. The test statistics used is the following:
t = BHAR×√n/σ(BHAR), (2)
where BHAR is the average across firms’ BHARs, σ(BHAR) is the cross-sectional
standard deviation of BHARs for all sample firms, and n is the number of firms.
Barber and Lyon (1997) document that the control firms approach eliminate the
new listing bias, rebalancing bias, and the skewness problem. Hence, we use the
matching firm approach as the benchmark. Following Barber and Lyon (1997), we
first identify all firms with a market value of equity within 30% of the market value
of equity of the sample firm at the beginning of the year in which R&D increase
significantly. From this set of firms, we choose the firm which has the closest book-
to-market equity ratio to the sample firm.
3.2 Calendar-Time Abnormal Return (CTAR)
Since Fama (1998) point out that the problem of cross-sectional dependence of return
is severe when the sample event dates are close, we use a calendar-time approach that
could mitigate such bias.
We use a Fama-French three-factor model and estimate the average monthly ab-
normal returns, αi , by using the following equation:
Rpt −Rft = αi + βi(Rmt −Rft) + siSMBt + hiHMLt + εpt, (3)
where Rpt is the monthly return on the equal- or value-weighted portfolio in calendar
month t (where a sample stock is included if month t is within the 60-month period
9
following its R&D increase) , Rft is the 1-month treasury bill return, Rmt is the
CRSP value-weighted market index return, SMBt is the difference in the returns on
the value-weighted portfolios of small and big stocks, and HMLt is the difference in
the returns on the value-weighted portfolios of high and low book-to-market stocks.
We also use the Carhart four-factor4 model to measure the monthly abnormal re-
turn following R&D increase by estimating the intercept from the following equation:
Rpt −Rft = αi + βi(Rmt −Rft) + siSMBt + hiHMLt +miUMDt + εpt, (4)
where UMDt is the difference in the return on a value-weighted portfolio of high and
low momentum stocks. Additionally, we impose each calendar month to consist of at
least three observations in calculating the monthly average returns of the calendar-
time return series.
In addition, Berk, Green, and Naik (2000) argue that a firm’s systematic risk may
change because of an investment in R&D. To confirm this argument, we also estimate
each factor loading based on a rolling regression approach in Equation (3) and (4). For
instance, we use the first 60 months to estimate the factor loadings in the Equations
(3) and (4). We then estimate the abnormal return in the 61st month as the difference
between the actual portfolio return and the expected portfolio return. The expected
portfolio return is computed as the estimated factor loadings times their respective
factor returns in the 61st month. We repeat this step every month and then average
the time series of the abnormal returns. Finally, we use the volatility of the time
series abnormal returns to estimate the standard errors of their respective averages.
3.3 Operating Performance Measures
We compute the sample firm’s operating performance from the first year to the fifth
year following the year in which they unexpected increase R&D by an economically
significant amount. The abnormal operating performance is defined as a sample firm’s
raw operating performance minus its control firm’s operating performance. We first
screen control firms that do not have the same corporate event as the sample firms
4We obtain the data of Rm, Rf , SMB, HML, and UMD from Wharton Research Data Services,WRDS, website.
10
in the event year. In this study, we use two different criteria to select the matched
firm. The first principle of the matched firm is based on the industry and pre-event
performance. We choose a group of control firms which are in the same two-digit SIC
code as the sample firm that do not unexpectedly significant increase its R&D during
the sample firm’s R&D increase year. From those screened firms, we select a firm as
the matched control firm that has the closest operating performance measures (OPMs
hereafter) with sample firm’s OPMs prior the sample firm’s R&D increase year.
For the second matching criterion, we form another group of control firms based
on the sample firm’s characteristics, i.e. size, book-to-market equity ratio, and mo-
mentum (defined as the prior 12 months returns). Ikenberry and Ramnath (2002) and
Eberhart, Maxwell, Siddique (2004) also use this matching criteria. In the beginning
of a sample firm’s R&D increase year, we choose a matched firm that the market
equity value is within 30% of the market equity value of the sample firm and then we
choose the jointly lowest value absolute of the difference in these characteristics. Fi-
nally, compare the relative operating performance improvements between the sample
firm and control firm.
4 Empirical Results
4.1 Descriptive Statistics
Panel A of Table 1 shows the sample selection procedure in this study. We finally
obtain 151 firm-year observations with 39 firms. There are 54 (15) firm-year (firm)
observations for overconfident CEOs and 97 (30) firm-year (firm) observations for
non-overconfident CEOs.
Panel B of Table 1 reports the descriptive statistics, and all variables in dollar
are adjusted in the 1994 dollars. On average, for the year in which sample firms
increase their R&D, firms with overconfident CEOs have $8,264 million in annual sales
and sample firms with non-overconfident CEOs have $3,648 million in annual sales.
The average level of book value of total assets and market value of common stock
for firms with overconfident CEOs are 7,142 million and 8,628 million, respectively.
The average level of book value of total assets and market value of common stock
11
for firms with non-overconfident CEOs are 3,091 million, 4,045 million, respectively.
The average and median book-to-market equity ratio for firms with overconfident
CEOs are significantly greater than non-overconfident CEOs. We compute the book-
to-market equity as the ratio of book equity to market equity and apply the same
procedure in Daniel and Titman (1997) to define book equity. All the mean difference
tests of firm characteristics between overconfident CEOs and non-overconfident CEOs
are at least significant under 5% level. In addition, for all the variables, the median
test of the difference between overconfident CEOs and non-overconfident CEOs are
statistically significant at 1% level, except for median difference test of the book-to-
market equity which are significant under 5% level.
Moreover, the median difference tests of the R&D intensity measure and the per-
centage of increase in dollar R&D between overconfident CEOs and non-overconfident
CEOs are significant under 5% level. To sum up, based on the mean and median dif-
ference tests, Table 1 suggests that the characteristics of firms with and without
overconfident CEOs who unexpected increase R&D may be different. Accordingly, it
is important to control the firm characteristics in the following cross-sectional regres-
sion analysis.
[Insert Table 1 about here]
4.2 Buy-and-Hold Abnormal Returns
Table 2 reports the one-, three-, and five-year BHARs following unexpected signifi-
cant R&D increase,5 respectively. The median test results of BHARs in Table 2 are
similar to those of the mean test. Hence, we concentrate our discussion on the results
of the mean test. On average, firms with non-overconfident CEOs exhibit positive
abnormal stock returns following R&D increase. In Panel A, the result shows that
non-overconfident CEOs earn almost 14% (54% or 102%) for investing one (three or
five) years. Moreover, the positive BHARs for firms with non-overconfident CEOs are
5The results of non-overlapping samples are qualitatively similar with full samples. Moreover, wefurther examine our results by trimming and winsorization the 5th and 95th of our full samples andnon-overlapping samples, and the results are also qualitatively similar with the original full samples.For brevity, we do not report here.
12
positively significant during all investment periods. On the other hand, the one- and
three-year BHARs of firms with overconfident CEOs do not significant differ from
zero, and the five-year BHARs for overconfident CEOs are even negatively signifi-
cant. Furthermore, for the full samples, the BHARs difference between firms with
non-overconfident CEOs and overconfident CEOs are positively significant. In other
words, the average BHARs of firms with non-overconfident CEOs are significantly
greater than firms with overconfident CEOs.
Although Eberhart, Maxwell, and Siddique (2004) suggest that R&D expenditure
usually conveys intangible information so that investors under-react their stock price,
our finding suggests that such argument exists only for non-overconfident CEOs. CEO
with overconfidence may feel that they are “better-than-average” and overestimate the
future cash flow after investing the R&D project. The R&D increase of overconfident
CEOs may suggest another form of overinvestment, and investors may identify that
overconfident CEOs overestimate future cash.
Furthermore, previous studies (Chan, Martin, and Kensinger (1990); Eberhart,
Maxwell, and Siddique (2004); Brown, Fazzari, and Petersen (2009)) point out that
R&D expenditure may have difference effects on high- and low-tech industries because
the attribute of R&D investment for high-tech industries are more complicated than
low-tech industries. Accordingly, we separate our samples into high- and low-tech
industry categories by Brown, Fazzari, and Petersen (2009)6 in Panel B and Panel
C of Table 2, respectively. In Panel B, all BHARs are positively significant for non-
overconfident CEOs and are insignificant different from zero for overconfident CEOs.
Except for one-year BHAR, the average BHAR difference between non-overconfident
CEOs and overconfident CEOs are also significantly positive at 1% level for three- and
five-year BHARs. For the low-tech firms in Panel C of Table 2, only one-year BHAR
for non-overconfident CEOs are positively significant. The BHARs for overconfident
CEOs are still insignificant different from zero. However, the average difference be-
tween non-overconfident CEOs and overconfident CEOs are still positively significant
for one- and five-year BHARs.
6Brown, Fazzari, and Petersen (2009) classifies the industries with SIC codes 283, 357, 366, 367,383, 384, and 737 as high-tech industries.
13
The results for high-tech firms are consistent with that firms with non-overconfident
CEOs could earn significantly positive abnormal returns than firm with overconfident
CEOs following their R&D increase.
[Insert Table 2 about here]
Table 3 provides the empirical results from the cross-sectional regressions. We
examine the relationship between the one-, three-, and five-year BHAR and CEO
overconfidence after firm significantly increase their R&D, respectively. Except for
Model (3) of Table 3, θ1 in Model (1) and Model (5) indicate that the BHARs following
R&D increase for firms with overconfident CEOs are significantly lower than non-
overconfident CEOs after controlling other factors. Furthermore, for Model (2), Model
(4), and Model (6) of Table 3, we add high-tech industry dummies and the interaction
term between CEO overconfidence and high-tech industry. Contrast with the results
of θ1 in Model (1), Model (3), and Model (5) of Table 3, not only the coefficients of θ1
are not negatively significant anymore, but the coefficients of θ3 are not significant,
except for model (4). However, it does not mean that CEO overconfidence plays no
role in high-tech industry. In fact, we need to test the sum of θ1 and θ3 to examine
the importance of CEO overconfidence in high-tech industry. The bottom of Table
3 reports the results of the Wald test for θ1 + θ3 = 0. The results show that all the
statistics of the Wald test in Table 3 are negatively significant. These findings are
consistent with our previous results which indicate the BHARs of overconfident CEOs
are significantly lower than non-overconfident CEOs.
[Insert Table 3 about here]
To conclude, investors may underreact the information content of R&D increase
only for firms with non-overconfident CEOs. A possible explanation could be that
overconfident CEOs may invest in value-destroying projects resulting from their over-
estimation of future cash flow or underestimation of the risk from R&D investment.
However, this phenomenon is more pronounced for high-tech firms.
14
4.3 Calendar-Time Abnormal Returns
Table 4 reports the long-term abnormal stock returns following R&D increase by
calendar-time approach.7 In Panel A of Table 4, for the full sample, we estimate
the risk adjusted abnormal returns using the Fama-French three-factor and Carhart
four-factor model. For both the equal- and value-weighted schemes, the intercepts
of non-overconfident CEOs are significantly positive with Fama-French three-factor
model (0.61 percent and 0.65 percent) and with the Carhart four-factor model (0.67
percent and 0.61 percent). However, the abnormal returns of overconfident CEOs
do not significantly differ from zero for Fama-French three-factor model and Carhart
four-factor model under equal- or value-weighted cases. Furthermore, the difference of
abnormal returns between non-overconfident CEOs and overconfident CEOs are posi-
tively significant for equal-weighted calendar-time portfolios when using Fama-French
three-factor model. In addition, when estimating by Carhart four-factor model, the
difference of abnormal returns between non-overconfident CEOs and overconfident
CEOs are positively significant for value-weighted calendar-time portfolios.
We report the rolling regression results in Panel B. With the Fama-French three-
factor model and the Carhart four-factor model, the differences of the equal- and
value-weighted abnormal returns between non-overconfident CEOs and overconfident
CEOs are all positively significant at least under 5 percent significance level. Fur-
thermore, the abnormal stock returns of value-weighted calendar time portfolio for
firms with overconfident CEOs are significantly negative for the Fama-French three
factor mode and Carhart four factor model. The results suggest that our findings are
robust when take into the consideration of the changes in risk of the portfolio over
time.
[Insert Table 4 about here]
Table 5 provides the subsample tests of abnormal stock returns for calendar time
portfolio. Our sample firms are divided into high- and low-tech firms according to
the definition in Brown, Fazzari, and Petersen (2009). With the high-tech samples,
7We also estimate the abnormal return by the purged factor portfolios which eliminate the samplefirms and non-overlap samples, and the results are qualitatively similar.
15
we obtain positively significant abnormal return estimates across all categories for
non-overconfident CEOs. The abnormal returns computed from the Carhart four-
factor model for overconfident CEOs become positively significant with equal- and
value-weighted return. The abnormal return difference between non-overconfident
CEOs and overconfident CEOs are significantly positive with value weighting. For
the low-tech firms, however, both the equal- and value-weighted abnormal returns are
insignificantly different from zero, using the Fama-French three factor model or the
Carhart four-factor model, for non-overconfident and overconfident CEOs. Therefore,
the abnormal return difference between non-overconfident CEOs and overconfident
CEOs are insignificantly different from zero for low-tech firms. In summary, the
results of calendar-time abnormal returns are consistent with BHARs.
[Insert Table 5 about here]
4.4 Operating Performance
We report the abnormal operating performance of our sample firms in Table 6.8 The
empirical results of median tests of abnormal operating performance are similar to
the results of mean tests. Hence, we only concentrate the results on the mean tests.
In panel A of Table 6, we show the changes in the abnormal operating performance of
OPM1 following R&D increase. The results based on the industry and performance
matching firm criterion, on average, reveal that firms with non-overconfident CEOs
significantly improve their operating performance than their matched firm. This
finding is consistent with previous literature that R&D investments benefit company’s
operating performance in the long run.
On the other hand, overconfident CEOs do not operate better than their bench-
mark following their R&D increase. This result consists with our inference that
overconfident CEOs exhibit overinvestment. As a result, their abnormal operating
performance reveal flat. These results may suggest that R&D is a beneficial invest-
ment only when firm’s CEO is not overconfident. In addition, we compare the relative
8In order to avoiding the extreme value affect out results, we also compute the abnormal operatingperformance by trimming and winsorization the 5th and the 95th of our smaples, and the results aresimilar. Also, we estimate the abnormal operating performance for non-overlapping samples, andthe results are still qualitatively similar to the full samples.
16
changes in abnormal operating performance for non-overconfident CEOs with over-
confident CEOs and the difference between them are significantly positive each year.
Similarly, non-overconfident CEOs also outperform their benchmark following the
R&D increase year when we use the characteristics as the matching firm criteria. For
overconfident CEOs, their abnormal operating performance are still insignificantly dif-
ferent from zero. Being consistent with the industry and performance match method,
the difference of abnormal operating performance between CEOs with overconfidence
and without overconfidence are significantly positive except for the first year following
significantly increase in R&D.
In Panel B of Table 6, we report the changes in the abnormal operating per-
formance of OPM2 following R&D increase. On average, the results of industry
and performance match also reveal that non-overconfident CEOs significantly im-
prove their profit margin following R&D increase. Although CEOs with overconfi-
dence also significant improve their operating performance than their control firm
after the third year following R&D increase, the magnitude of abnormal operat-
ing performance for firms with non-overconfident CEOs are greater than firms with
overconfident CEOs. Therefore, the difference of abnormal operating performance
between non-overconfident CEOs and overconfident CEOs are still significantly pos-
itive. The results are qualitatively similar to the results by using performance and
industry matched firm criterion when choosing the characteristic as the alternative
matched method. The difference of abnormal operating performance between non-
overconfident CEOs and overconfident CEOs are still significantly positive after the
second year following the R&D increase.
The implication of Table 6 suggests that R&D increase may benefit firms only
when their CEOs are not overconfident. Eberhart, Maxwell, and Siddique (2004)
point out that, in general, R&D investment are beneficial. We further examine this
argument from the viewpoint of CEO overconfidence. When considering CEO over-
confidence, R&D investment may benefit firms whose CEO are not overconfidence.
Otherwise, the R&D investments may be inefficiency if CEO is overconfident.
[Insert Table 6 about here]
17
Table 7 presents the subsample tests for the abnormal operating performance.9
For the high-tech firms in Panel A of Table 7, the result of abnormal operating
performance for non-overconfident CEOs are significantly positive and overconfident
CEOs are insignificantly different from zero. In addition, the difference between them
are also significantly positive from the second year following R&D increase both for the
results of industry and performance match method and characteristic match method.
For the low-tech sample in Panel B of Table 7, although the abnormal operat-
ing performance of OPM1 for overconfident CEOs are still insignificantly different
from zero, the abnormal OPM1 for non-overconfident CEOs are only significantly
positive in the first two year following R&D increase for industry and performance
match. Additionally, the difference for OPM1 between non-overconfident CEOs and
overconfident CEOs are significantly positive only in the first three years following
R&D increase. The results of abnormal operating performance of OPM2 for non-
overconfident CEOs are significantly positive in the second, fourth, and fifth year fol-
lowing R&D increase. The difference of abnormal OPM2 between non-overconfident
CEOs and overconfident CEOs are still significantly positive every year except for the
third year following R&D increase.
For the case of characteristic matching firm approach, however, the non-overconfident
CEOs even perform significantly worse than their benchmark for OPM1 in the third
and fourth year following significantly increase in R&D. Besides, the non-overconfident
CEOs also significantly underperform their benchmark for OPM2 in the third year
following R&D increase. However, the difference of abnormal operating performance
between non-overconfident CEOs and overconfident CEOs are insignificantly different
from zero each year. To summarize, the results in Table 6 are also consistent with our
previous results that the effect of R&D increases are significantly positive in high-tech
industries, and Eberhart, Maxwell, and Siddique (2004) also obtain similar empirical
evidence.
[Insert Table 7 about here]
9For the sake of brevity, we only report the results of mean test. The results of median test aresimilar to the results of mean tests.
18
Table 8 provides the cross-sectional regression of long-term abnormal operating
performance. We further examine the relationship between the five-year abnormal
operating performance and CEO overconfidence. Except for Model (5) of Table 8,
θ1 in Model (1) and Model (3) suggest that the abnormal operating performance
following significantly increase in R&D for overconfident CEOs are significantly lower
than non-overconfident CEOs. The result is consistent with our previous empirical
finding.
For Model (2) and Model (4) the results of Wald test in the bottom of Table 8
are negatively significant except for Model (6), although the coefficients of θ1 are not
positively significant anymore. To sum up, strongly reinforcing evidence about the
effect of CEO overconfidence on R&D expenditure are provided by Table 8 that over-
confidence is associated with a substantially lower abnormal operating performance
following unexpected significantly R&D increase, especially in high-tech industries.
[Insert Table 8 about here]
5 Conclusion
In this paper, we investigate the relationship between unexpected R&D increase and
CEO overconfidence. Previous studies document that R&D increase is a beneficial in-
vestment decisions that could improve a firm’s operating performance following R&D
increase. In addition, investors usually underreact to intangible information, such as
the prospect of future cash flow of R&D investment. Therefore, the long-term abnor-
mal stock returns following R&D increase are significantly positive. CEOs may feel
themselves “better-than-average” and they often result in the overestimation of the
future return from their investment or underestimation the likelihood of failure. Since
CEO overconfidence often leads to overinvestment behavior or raise the investment-
cash flow sensitivity, we examine whether the unexpected R&D increase made by
overconfident CEOs still benefit for firms and whether investor exhibit underreaction
for firms with overconfident CEOs following unexpected R&D increase.
We examine both the long-term abnormal stock and operating performance by
varies methodologies. The results show that investors earn abnormal stock returns
19
for firms with non-overconfident CEOs following R&D increase. In addition, the long-
term abnormal stock returns for non-overconfident CEOs are significantly larger than
overconfident CEOs. We also find consistently strong evidence that R&D increase is a
beneficial investment decisions only for firms with non-overconfident CEOs. In other
words, the long-term abnormal operating performance for non-overconfident CEOs
are significantly greater than overconfident CEOs. Cross-sectional regression analyses
also show that overconfidence is associated with a substantially lower abnormal stock
and operating performance following unexpected significantly R&D increase. Further-
more, our findings are stronger for high-tech industries because R&D expenditures
are usually complicated and essential for high-tech industries.
Our results contribute to the link between behavior corporate finance and R&D
investment decisions. A large of recent studies investigates how CEO overconfidence
affects a firm’s corporate decisions. Past research demonstrates that increases in R&D
expenditure are beneficial for shareholders and firms. Our work complements these
gaps by using option based overconfidence measure to deeply examine the impact
of R&D increase. Our finding also provide empirically evidence to support previous
research which propose overconfident manager are more like to be the pioneer to
introduce risky products and these products are less likely to achieve success. Since
R&D expenditure may bring a new technology development or create a new product,
the effect of increase in R&D expenditure for overconfident CEOs may be inefficiency.
20
Appendix: Variable Definitions
Variables Definitions
Panel A: CEO Overconfidence Measure and Characteristics*
Longholder Dummy variable: 1 if the CEO at some point during his tenure heldan option grants until the last year before expiration, in case thatthe option grants was at least 40% in-the-money entering its lastyear
CEO Stock Ownership Percentage of common stocks owned by CEO and his immediatefamily
CEO Vested Options CEO’s holdings of options that are exercisable within six monthsdivided by the common shares outstanding, and multiplied by ten
Panel B: Firm Characteristics
Market Capitalization Fiscal year end price (Compustat Item 25) multiplied by outstandingshares (Compustat Item 199)
Tobin q Market value of assets to book value of assets (Compustat Item 6)
Market Value of Assets Book value of assets (Compustat Item 6), minus the book value ofequity plus the market value of common equity (Compustat Item25 × Compustat Item 199)
Book Value of CommonEquity
Stockholder’s equity (Compustat Item 216) plus any deferred tax(Compustat Item 74) and any investment tax credit (CompustatItem 208), minus any preferred stock
Preferred Stock Redemption value (Compustat Item 56) if it is available, otherwiseuse liquidating value (Compustat Item 10) if it is available, and ifnot available use carrying value (Compustat Item 130)
Book-to-Market equity Book value of common equity to market value of common equity
Sales Growth Sum of sales (Compustat Item 12) minus prior year sales and dividedby prior year sales
Cash Ratio of cash and short-term Investments (Compustat Item 1) tobook assets
PPE/Emp Ratio of property, plant, and equipment (Compustat Item 7) to theemployees (Compustat Item 29)
Panel C: R&D Intensity Measure
R&D/Assets R&D (Compustat Item 46) divided by assets (Compustat Item 6)
R&D/Sales R&D divided by sales
Panel D: Profit Margin Measure
OPM1 EBIT (Compustat Item 178) divided by sales
OPM2 Sum of EBIT and after-tax R&D (taxes (Compustat Item 16) totaxable income (Compustat Item 170) multiplied by R&D) dividedby sales
* The data in Panel A are provided by Professor Ulrike Malmendier.
21
References
Alicke, Mark, 1985, Global self-evaluation as determined by the desirability andcontrollability of trait adjectives, Journal of Personality and Social Psychology49, 1621–1630.
Alicke, Mark, M. Klotz, David Breitenbecher, Tricia Yurak, and Debbie Vreden-burg. 1995, Personal contact, individuation, and the better-than-average effect,Journal of Personality and Social Psychology 68, 804–825.
Barber, B. and J. Lyon, 1997, Detecting long-run abnormal stock returns: the em-pirical power and specification of test statistics, Journal of Financial Economics54, 341–372.
Berk, J., Green, R,, and V. Naik, 2000, Valuation and return dynamics of newventures, Working paper, University of California Berkeley.
Billett, Matthew and Yiming Qian, 2007, Are Overconfident CEOs Born or Made?Evidence of Self-Attribution Bias from Frequent Acquirers, Working Paper,University of Iowa.
Brown, James, Steven Fazzari, and Bruce Petersen, 2009, Financing Innovation andGrowth: Cash Flow, External Equity, and the 1990s R&D Boom, Journal ofFinance 74,151–185.
Campbell, T., Michael Gallmeyer, Shane Johnson, Jessica Rutherford and BrookeStanley, 2010, CEO Confidence and Forced Turnover, Working Paper, TexasA&M University.
Chan, Su Han, John Martin, and John Kensinger, 1990, Corporate research anddevelopment expenditures and share value, Journal of Financial Economics 26,255–276.
Daniel, Kent and Sheridan Titman, 1997, Evidence on the Characteristics of CrossSectional Variation in Stock Returns, Journal of Finance 52, 1–33.
Daniel, Kent and Sheridan Titman, 2003, Market reactions to tangible and intangibleinformation, Working paper, Northwestern University.
Eberhart, Allen, William Maxwell, and Akhtar Siddique, An Examination of Long-Term Abnormal Stock Returns and Operating Performance Following R&DIncreases, Journal of Finance 69, 623–650.
Fama, E., 1998, Market efficiency, long-term returns, and behavioral finance, Journalof Financial Economics 49, 283–306.
22
Galasso, Alberto and Timothy Simco, 2010, CEO Overconfidence and Innovation,Working Paper, University of Toronto.
Goel, A. and A. Thakor, 2008, Overconfidence, CEO selection, and Corporate Gov-ernance, Journal of Finance 63, 2737–2784.
Hall, B. and J. Liebman, 1998, Are CEOs really paid like bureaucrats? QuarterlyJournal of Finance 113, 653–691.
Hall, B. and K. Murphy, 2002, Stock options for undiversified executives, Journal ofAccounting Economics 33, 3–42.
Hirshleifer, David, Angir Low, and Siew Teoh, 2010, Are Overconfident CEOs betterInnovators?, Working paper, University of California Irvine.
Hribar, Paul and Holly Yang, 2010, Does CEO Overconfidence Affect ManagementForecasting and Subsequent Earnings Management?, Working Paper, Universityof Iowa.
Ikenberry, David and Sundaresh Ramnath, 2002, Underreaction to self-selected news:The case of stock splits, Review of Financial Studies 15, 489–526.
Lambert, Richard, David Larcker, and Robert Verrecchia, 1991, Portfolio Consider-ations in Valuing Executive Compensation, Journal of Accounting Research 29,129–149.
Larwood, Laurie, and William Whittaker, 1977, Managerial myopia: Self-servingbiases in organizational planning, Journal of Applied Psychology 62, 94–198.
Liu, Yue, and Richard Taffler, 2008, CEO Overconfidence in M&A Decision Makingand its Impact on Firm Performance, Working Paper, University of Edinburgh.
Longhran, T. and J. Ritter, 2000, Uniformly least powerful tests of market efficiency,Journal of Financial Economics 55, 361–389.
Lyon, J. D., Barber, B. M. and T. C.-L., 1999, Improved methods for tests forlong-run abnormal stock returns, Journal of Finance 54, 165–201.
Malmendier, Ulrike and Geoffrey Tate, 2005, CEO overconfidence and corporateinvestment, Journal of Finance 60, 2661–2700.
Malmendier, Ulrike and Geoffrey Tate, 2008, Who makes acquisitions? CEO over-confidence and the market’s reaction, Journal of Financial Economics 89, 20–43.
Malmendier, Ulrike and Geoffrey Tate, and Jon Yan, 2010, Managerial Beliefs andCorporate Financial Policies, Working Paper, University of California Berkeley.
23
Newey, W. K. and K. D. West, 1987, A Simple, Positive Semi-definite, Heteroskedas-ticity and Autocorrelation Consistent Covariance Matrix, Econometrica 55,703–708.
Simon, Mark and Susan M. Houghton, 2003, The Relationship between Overconfi-dence and the Introduction of Risky Products: Evidence from a Field Study,Academy of Management Journal 46, 139–149.
Svenson, Ola, 1981, Are we all less risky and more skillful than our fellow drivers?Acta Psychologica 47, 143–148.
Szewczyl, Samuel, George Tsetsekos, and Zaher Zantout, 1996, The valuation ofcorporate R&D expenditures: Evidence from investment opportunities and freecash flow, Financial Management 25, 105–110.
Weinstein, N., 1980, Unrealistic optimism about future life events, Journal of Per-sonality and Social Psychology 39, 806–820.
Yermack, D., 1995, Do corporate award CEO stock options effectively? Journal ofFinancial Economics 39, 237–269.
Zantout, Zaher and George Tsetsekos, 1994, The wealth effects of announcementsof R&D expenditures increases, Journal of Financial Research 17, 205–216.
24
Tab
le1:
Des
crip
tive
and
Sum
mar
ySta
tist
ics
Th
ista
ble
pro
vid
essa
mp
lese
lect
ion
pro
ced
ure
inP
an
elA
,an
dP
an
elB
show
sth
esu
mm
ary
stati
stic
sfo
rth
esa
mp
leof
over
con
fid
ent
CE
Os
an
dn
on
-over
con
fid
ent
CE
Os
wit
h(u
nex
pec
ted
an
dec
on
om
ically
sign
ifica
nt)
rese
arc
han
dd
evel
op
men
tin
crea
ses.
Th
esa
mp
lep
erio
dco
ver
sfr
om
1980
to1994.
Lon
gh
old
eris
ab
inary
vari
ab
lew
her
e1
refe
rsth
at
the
CE
Oat
som
ep
oin
td
uri
ng
his
tenu
reh
eld
an
op
tion
gra
nts
unti
lth
ela
styea
rb
efore
exp
irati
on
,in
case
that
the
op
tion
gra
nts
was
at
least
40%
in-t
he-
mon
eyen
teri
ng
its
last
yea
r.T
he
vari
ab
les
sale
s,to
tal
ass
ets,
mark
etca
pit
aliza
tion
,an
db
ook-t
o-m
ark
eteq
uit
yare
mea
sure
das
the
beg
inn
ing
of
the
sam
ple
firm
’sR
&D
incr
ease
yea
r,an
dare
ad
just
edby
the
CP
Ito
refl
ect
1994
dollars
.T
he
vari
ab
led
efin
itio
nof
sale
s,to
tal
ass
ets,
mark
etca
pit
ali
zati
on
,b
ook-t
o-m
ark
eteq
uit
yare
des
crib
edin
Ap
pen
dix
.T
he
R&
Din
ten
sity
rati
ois
als
om
easu
red
as
of
the
beg
inn
ing
of
the
R&
Din
crea
seyea
r.T
he
per
centa
ge
incr
ease
ind
ollar
R&
Dis
mea
sure
dover
the
R&
Din
crea
seyea
r.T
-tes
t(K
rusk
al-
Wall
iste
st)
are
emp
loyed
tote
stfo
rd
iffer
ence
bet
wee
nth
em
ean
s(m
edia
ns)
for
the
firm
sw
ith
over
con
fid
ent
CE
Os
an
dfi
rms
wit
hn
on
-over
con
fid
ent
CE
Os
an
dth
eP
-valu
eare
rep
ort
edin
the
last
two
colu
mn
sof
Pan
elB
.
Pan
elA
:S
am
ple
Sel
ecti
on
Pro
ced
ure
Fir
m-y
ear
Fir
m
Nu
mb
erof
ob
serv
ati
on
sin
Com
pu
stat
from
1980
to1994
148,9
73
18,8
04
Nu
mb
erof
ob
serv
ati
on
sh
ave
un
exp
ecte
dsi
gn
ifica
ntl
yin
crea
seR
&D
5,2
47
2,1
90
Nu
mb
erof
ob
serv
ati
on
sh
aveLongh
older
5,3
92
445
Nu
mb
erof
ob
serv
ati
on
sh
aveLongh
older
an
du
nex
pec
ted
sign
ifica
ntl
yin
crea
seR
&D
151
39
Pan
elB
:S
um
mary
Sta
tist
ics
Non
-over
con
fid
ent
CE
Os
(N=
97)
Over
con
fid
ent
CE
Os
(N=
54)
Diff
eren
ce
Mea
nM
edia
nS
td.
Dev
.M
ean
Med
ian
Std
.D
ev.
Mea
nM
edia
n
Sale
s($
MM
)3,6
48.0
81,2
12.9
38,0
95.6
98,2
64.1
55,5
42.0
011,9
55.5
40.0
133**
<.0
001***
Tota
lA
sset
s($
MM
)3,0
91.0
71,2
85.9
25,7
68.7
87,1
41.9
25,1
51.7
09,4
53.2
60.0
054***
<.0
001***
Mark
etC
ap
italiza
tion
($M
M)
4,0
45.1
22,1
02.6
65,8
71.7
28,6
27.9
86,0
26.6
98,9
03.9
20.0
011***
<.0
001***
Book-t
o-m
ark
eteq
uit
y0.4
00.3
60.2
30.4
80.4
60.2
60.0
482**
0.0
486**
R&
Din
ten
sity
mea
sure
(%)
R&
D/A
sset
s9.5
88.7
74.2
28.5
07.2
33.5
50.1
113
0.0
177**
R&
D/S
ale
s12.1
78.7
614.0
37.3
96.5
82.9
00.0
016***
0.0
014***
Incr
ease
ind
ollar
R&
D(%
)20.7
712.7
227.5
112.4
610.0
410.1
90.0
087***
0.0
048***
Th
esy
mb
ols
***,
**,
*in
dic
ate
stati
stic
al
sign
ifica
nce
at
the
1,
5,
10
per
cent
level
s,re
spec
tivel
y.
25
Table 2: BHARs following R&D Increase
This table provides one-, three-, and five-year BHARs following sample firms’ R&D increase, respectively. For theith sample firm from month T1 to T2, BHARi,t express as follows:
BHARi,t =
T2∏t=T1
(1 +Ri,t)−T2∏t=T1
(1 +Rb,t),
where Ri,tis the return of the sample firm in event month t, and Rb,t is the return of the benchmark over the sameperiod. The test statistics used is the following:
t = BHAR×√n/σ(BHAR),
where BHAR is the average across firms’ BHARs, σ(BHAR) is the cross-sectional standard deviation of BHARsfor all sample firms, and n is the number of firms. For each event firm, we compute corresponding one-, three-and five-year BHAR following R&D increase year with a three month lag. We use control firm approach as thebenchmark. We first identify all firms with a market value of equity within 30% of the market value of equity of thesample firm in the year of R&D increase. From this set of firms, we choose the firm which has the closest book-to-market equity ratio to the sample firm. Then, we compute and report the mean and median BHARs of firms withoverconfident CEOs and firms with non-overconfident CEOs, respectively. Non-OC CEOs refers to the firms withnon-overconfident CEOs, and OC CEOs refers to the firms with overconfident CEOs. We use the Kruskal-Wallistest to test the median difference. Firms are classified into high-tech and low-tech firms using the definition inBrown, Fazzari, and Peterson (2009).
Longholder1-Year BHAR 3-Year BHAR 5-Year BHAR
N Mean Median N Mean Median N Mean Median
Panel A: Full Sample
Non-OC CEOs 97 0.1445** 0.1494*** 97 0.5431*** 0.3016*** 97 1.0202*** 0.4264**OC CEOs 54 −0.0218 −0.0533 54 −0.1568 −0.1631 54 −0.3277* −0.4739*Difference 0.1663** 0.2070** 0.6999*** 0.4647*** 1.3478*** 0.9003***
Panel B: High-Tech Sample
Non-OC CEOs 77 0.1381** 0.1580** 77 0.6464*** 0.3318*** 77 1.2134*** 0.4331**OC CEOs 39 −0.0043 −0.0434 39 −0.1387 −0.1073 39 −0.2371 −0.0369Difference 0.1424 0.2014* 0.7851*** 0.4391*** 1.4505*** 0.4700**
Panel C: Low-Tech Sample
Non-OC CEOs 20 0.1691** 0.1069** 20 0.1451 0.2412 20 0.2761 0.3359OC CEOs 15 −0.0673 −0.1144 15 −0.2040 −0.2800 15 −0.5633 −1.1538Difference 0.2364** 0.2213** 0.3491 0.5212 0.8394* 1.4897**
The symbols ***, **, * indicate statistical significance at the 1, 5, 10 percent levels, respectively.
26
Table 3: Cross-sectional Regression of BHARs
In this table, we estimate the equation as follows:
BHARit = θ0 + θ1OCit + xit−1β + εit
where BHARit refers to the 1-, 3-, and 5-year BHARs following R&D significantly increases for firm i in yeart. OCit refers to the CEO overconfidence indicator, Longholder, for firm i in year t, and xit−1 is a vector ofcontrol variables of firm i in year t − 1. The definitions of control variables, x, are described in Appendix.In addition, we measure the following equation by adding the high-tech indicator as the explanatory variableand its interaction term between CEO overconfidence indicator:
BHARit = θ0 + θ1OCit + θ2HTit + θ3OCit ×HTit + xit−1β + εit
where BHARit refers to the 1-, 3-, and 5-year BHARs following R&D significantly increases for firm i inyear t. OCit refers to the CEO overconfidence indicator, Longholder, for firm i in year t. HTit refers tothe high-tech industries indicator, and xit−1 is a vector of control variables of firm i in year t− 1. Standarderrors are clustered at firm level, heteroskedasticity consistent, and reported in the parentheses. The bottompresents the results of Wald test for θ1 + θ3 = 0, and P -values are reported in the parentheses.
1-Year BHAR 3-Year BHAR 5-Year BHAR
(1) (2) (3) (4) (5) (6)
Intercept −0.814 −0.978 −2.067 −2.655 −8.275 −9.646(−1.47) (−1.71)* (−1.49) (−1.81)* (−2.32)** (−2.52)**
Longholder −0.205 −0.034 −0.532 0.385 −1.060 0.806(−1.72)* (−0.18) (−1.64) (0.72) (−1.80)* (0.66)
High Tech 0.267 0.778 1.861(1.25) (1.30) (1.50)
Longholder×High-Tech −0.228 −1.236 −2.216(−0.98) (−1.87)* (−1.60)
Size 0.000 0.000 0.000 0.000 0.000 0.000(−1.08) (−1.20) (−1.10) (−1.31) (−0.55) (−0.69)
BM 0.789 0.740 1.649 1.446 2.953 2.509(1.37) (1.34) (1.38) (1.29) (1.27) (1.13)
Tobin q 0.009 0.010 0.381 0.370 −0.154 −0.170(0.08) (0.08) (1.16) (1.13) (−0.46) (−0.51)
R&D/Asset 0.773 −0.065 −3.738 −5.500 −2.789 −8.091(0.49) (−0.04) (−0.80) (−1.39) (−0.27) (−0.89)
ROA 3.115 3.158 3.983 4.859 15.436 13.646(2.41)** (2.54)** (1.08) (1.38) (2.32)** (2.29)**
Sales Growth 0.162 0.139 0.086 −0.091 1.163 1.144(0.32) (0.28) (0.09) (−0.09) (0.62) (0.56)
Cash 0.727 0.826 1.776 2.081 3.664 5.659(1.07) (1.29) (0.87) (1.11) (0.76) (1.21)
PPE/Emp −0.002 0.000 −0.004 0.000 0.001 0.004(−0.56) (−0.11) (−0.40) (−0.04) (0.04) (0.26)
Stock Ownership 0.585 0.334 6.703 5.666 29.233 25.910(0.59) (0.32) (3.00)*** (2.31)** (6.26)*** (6.13)***
Vested Options 3.950 2.312 5.431 0.873 44.802 14.388(0.45) (0.24) (0.20) (0.03) (0.98) (0.29)
(Vested Options)2 −91.419 −72.609 −120.193 −52.363 −390.702 124.745(−0.62) (−0.46) (−0.28) (−0.12) (−0.61) (0.16)
Year Effects Yes Yes Yes Yes Yes YesSIC 2-digit Effects Yes Yes Yes Yes Yes Yes
Obs 136 136 136 136 136 136Adjusted R2 0.095 0.094 0.213 0.231 0.264 0.323
Wald Tests: θ1 + θ3 = 0 −0.262 −0.852 −1.410P -value (0.070)* (0.033)** (0.028)**
The symbols ***, **, * indicate statistical significance at the 1, 5, 10 percent levels, respectively.
27
Tab
le4:
Lon
g-T
erm
Abnor
mal
Ret
urn
sfo
rC
alen
dar
-Tim
eP
ortf
olio
Th
ista
ble
pro
vid
esab
norm
al
stock
retu
rns
for
the
sam
ple
incr
easi
ng
thei
rR
&D
from
1980
to1994.
We
use
the
Fam
a-F
ren
chth
ree-
fact
or
mod
elto
esti
mate
the
ab
norm
al
retu
rnby
the
follow
ing
equ
ati
on
:
Rpt−Rft
=αi
+βi(Rmt−Rft)
+s iSMBt
+hiHMLt
+ε pt
wh
ereRpt
isth
em
onth
lyre
turn
on
the
equ
al-
or
valu
e-w
eighte
dp
ort
folio
inca
len
dar
montht
(wh
ere
asa
mp
lest
ock
isin
clu
ded
ifm
ontht
isw
ith
inth
e60-m
onth
per
iod
follow
ing
its
R&
Din
crea
se)
,Rft
isth
e1-m
onth
trea
sury
bill
retu
rn,Rmt
isth
eC
RS
Pvalu
e-w
eighte
dm
ark
etin
dex
retu
rn,SMBt
isth
ed
iffer
ence
inth
ere
turn
son
the
valu
e-w
eighte
dp
ort
folios
of
small
an
db
igst
ock
s,an
dHMLt
isth
ed
iffer
ence
inth
ere
turn
son
the
valu
e-w
eighte
dp
ort
folios
of
hig
han
dlo
wb
ook-t
o-m
ark
etst
ock
s.W
eals
ou
seth
eC
arh
art
four-
fact
or
mod
elto
mea
sure
the
month
lyab
norm
al
retu
rnfo
llow
ing
R&
Din
crea
seby
esti
mati
ng
the
inte
rcep
tfr
om
the
follow
ing
equ
ati
on
:
Rpt−Rft
=αi
+βi(Rmt−Rft)
+s iSMBt
+hiHMLt
+miUMDt
+ε pt,
wh
ereUMDt
isth
ed
iffer
ence
inth
ere
turn
on
avalu
e-w
eighte
dp
ort
folio
of
hig
han
dlo
wm
om
entu
mst
ock
s.A
dd
itio
nally,
we
imp
ose
that
each
cale
nd
ar
month
at
least
have
thre
eob
serv
ati
on
s.In
pan
elB
,w
eu
seth
efi
rst
60
month
sto
esti
mate
the
fact
or
load
ings
for
equ
ati
on
(3)
an
d(4
).T
hen
,w
ees
tim
ate
the
ab
norm
al
retu
rnin
month
61
as
the
diff
eren
ceb
etw
een
the
act
ual
port
folio
retu
rnan
dth
eex
pec
ted
port
folio
retu
rn.
Th
eex
pec
ted
port
folio
retu
rnd
efin
edas
the
fact
or
load
ings
esti
mate
dover
the
pre
vio
us
60
month
sti
mes
thei
rre
spec
tive
month
61
fact
or
retu
rns.
We
can
rep
lica
teth
isst
epev
ery
month
,an
dth
enaver
age
the
tim
ese
ries
of
thes
eab
norm
al
retu
rnan
dfa
ctor
load
ing
esti
mate
s.N
on
-OC
CE
Os
refe
rsto
the
firm
sw
ith
non
-over
con
fid
ent
CE
Os,
an
dO
CC
EO
sre
fers
toth
efi
rms
wit
hover
con
fid
ent
CE
Os.
Inp
are
nth
eses
are
New
ey-W
est
(1987)
ad
just
edfo
rse
rial
corr
elati
on
an
dh
eter
osc
edast
icit
yt-
stati
stic
s.
Lon
gh
old
erF
am
a-F
ren
chT
hre
e-F
act
or
Mod
elC
arh
art
Fou
r-F
act
or
Mod
el
αb
sh
αb
sh
m
Pan
elA
:F
ull
Sam
ple
Equ
al
Non
-OC
CE
Os
0.0
061***
1.0
972***
0.2
705***
-0.5
160***
0.0
067***
1.1
008***
0.2
566***
-0.5
273***
-0.0
797
Wei
ght
(3.2
5)
(19.8
2)
(2.9
2)
(-4.5
5)
(3.3
5)
(20.4
8)
(2.7
4)
(-4.4
4)
(-1.0
6)
OC
CE
Os
0.0
004
1.1
951***
0.1
454
-0.2
672*
0.0
023
1.2
067***
0.0
996
-0.3
046**
-0.2
628***
(0.1
5)
(17.0
9)
(1.3
1)
(-1.8
3)
(0.8
4)
(18.2
4)
(0.9
4)
(-2.0
8)
(-2.9
3)
Diff
eren
ce0.0
057**
-0.0
979
0.1
251
-0.2
488*
0.0
044
-0.1
060
0.1
570
-0.2
227*
0.1
831*
(2.1
3)
(-1.5
3)
(1.2
2)
(-1.9
0)
(1.6
0)
(-1.6
5)
(1.5
4)
(-1.7
0)
(1.9
4)
Valu
eN
on
-OC
CE
Os
0.0
065***
0.9
987***
-0.1
447*
-0.6
719***
0.0
061***
0.9
965***
-0.1
358*
-0.6
646***
0.0
509
Wei
ght
(3.8
5)
(19.3
7)
(-1.9
4)
(-6.7
2)
(3.4
2)
(19.0
6)
(-1.7
3)
(-6.5
2)
(0.8
4)
OC
CE
Os
0.0
001
1.0
305***
-0.3
117***
-0.4
790***
0.0
005
1.0
325***
-0.3
196***
-0.4
854***
-0.0
453
(0.0
7)
(17.1
6)
(-3.1
5)
(-4.8
1)
(0.2
2)
(16.9
4)
(-3.1
5)
(-4.9
5)
(-0.5
9)
Diff
eren
ce0.0
063***
-0.0
318
0.1
670
-0.1
929
0.0
056**
-0.0
360
0.1
838
-0.1
792
0.0
962
(2.6
7)
(-0.4
3)
(1.4
4)
(-1.4
6)
(2.3
5)
(-0.4
8)
(1.5
2)
(-1.3
5)
(1.0
7)
28
Tab
le4
Con
tinued
Lon
gh
old
erF
am
a-F
ren
chT
hre
e-F
act
or
Mod
elC
arh
art
Fou
r-F
act
or
Mod
el
αb
sh
αb
sh
m
Pan
elB
:R
ollin
gR
egre
ssio
nM
eth
od
Equ
al
Non
-OC
CE
Os
0.0
048*
0.0
079***
1.0
845***
-0.5
768***
0.0
066**
0.0
090***
1.0
959***
-0.6
198***
-0.1
591***
Wei
ght
(1.8
5)
(75.2
2)
(193.9
9)
(-21.1
3)
(2.5
0)
(64.0
0)
(195.4
9)
(-20.9
6)
(-16.2
7)
OC
CE
Os
-0.0
032
0.0
003
1.1
048***
-0.3
152***
-0.0
016
0.0
019***
1.1
193***
-0.3
509***
-0.2
079***
(-1.0
7)
(1.0
8)
(119.0
1)
(-9.5
3)
(-0.5
4)
(6.6
2)
(124.8
2)
(-10.1
8)
(-19.6
0)
Diff
eren
ce0.0
080**
0.0
075***
-0.0
203*
-0.2
616***
0.0
082**
0.0
071***
-0.0
234**
-0.2
689***
0.0
487***
(2.0
3)
(23.2
4)
(-1.8
8)
(-6.1
0)
(2.0
7)
(22.3
6)
(-2.2
1)
(-5.9
2)
(3.3
8)
Valu
eN
on
-OC
CE
Os
0.0
044*
0.0
068***
1.0
124***
-0.6
948***
0.0
042*
0.0
067***
1.0
118***
-0.6
898***
0.0
108**
Wei
ght
(1.9
1)
(50.9
4)
(299.9
3)
(-30.6
7)
(1.7
9)
(45.4
4)
(294.0
2)
(-30.2
9)
(2.4
4)
OC
CE
Os
-0.0
041*
-0.0
002
0.9
559***
-0.5
910***
-0.0
043*
-0.0
000
0.9
566***
-0.5
806***
-0.0
016
(-1.8
5)
(-0.5
6)
(223.0
3)
(-34.7
4)
(-1.9
3)
(-0.0
9)
(216.3
0)
(-33.7
4)
(-0.2
0)
Diff
eren
ce0.0
085***
0.0
070***
0.0
565***
-0.1
037***
0.0
085***
0.0
068***
0.0
552***
-0.1
092***
0.0
124
(2.6
6)
(18.1
7)
(10.3
5)
(-3.6
6)
(2.6
3)
(18.1
9)
(9.8
5)
(-3.8
3)
(1.3
4)
Th
esy
mb
ols
***,
**,
*in
dic
ate
stati
stic
al
sign
ifica
nce
at
the
1,
5,
10
per
cent
level
s,re
spec
tivel
y.
29
Tab
le5:
Subsa
mple
Tes
tsof
Lon
g-T
erm
Abnor
mal
Ret
urn
sfo
rC
alen
dar
-Tim
eP
ortf
olio
Th
ista
ble
pro
vid
esab
nor
mal
stock
retu
rns
for
the
sam
ple
incr
easi
ng
thei
rR
&D
from
1980
to1994.
The
sam
ple
are
div
ided
into
hig
h-t
ech
and
low
-tec
hfi
rms
usi
ng
the
defi
nit
ion
inB
row
n,
Fazz
ari
,an
dP
eter
son
(2009).
Th
eα
refe
rsto
the
ab
norm
al
retu
rnm
easu
re.
See
Tab
le3
for
ad
etai
led
des
crip
tion
of
the
test
pro
ced
ure
s.In
pare
nth
eses
are
New
ey-W
est
(1987)
ad
just
edfo
rse
rial
corr
elat
ion
and
het
eros
ced
asti
cityt-
stat
isti
cs.
Lon
gh
old
erF
am
a-F
ren
chT
hre
e-F
act
or
Mod
elC
arh
art
Fou
r-F
act
or
Mod
el
αb
sh
αb
sh
m
Pan
elA
:H
igh
-Tec
hS
am
ple
Equ
al
Non
-OC
CE
Os
0.0
096***
1.0
970***
0.3
442**
-0.6
777***
0.0
103***
1.1
012***
0.3
318**
-0.6
948***
-0.1
074
Wei
ght
(3.8
8)
(14.0
7)
(2.1
5)
(-4.2
7)
(3.9
7)
(14.8
2)
(2.0
8)
(-4.1
2)
(-1.0
7)
OC
CE
Os
0.0
038
1.1
970***
0.2
450
-0.5
164***
0.0
064*
1.2
114***
0.2
025
-0.5
751***
-0.3
697***
(1.0
1)
(13.0
7)
(1.5
4)
(-2.6
8)
(1.7
2)
(14.7
3)
(1.4
1)
(-2.8
9)
(-3.5
1)
Diff
eren
ce0.0
058
-0.0
999
0.0
991
-0.1
613
0.0
039
-0.1
102
0.1
293
-0.1
197
0.2
623**
(1.5
2)
(-1.2
5)
(0.5
8)
(-0.8
5)
(1.0
0)
(-1.3
6)
(0.7
5)
(-0.6
2)
(2.3
5)
Valu
eN
on
-OC
CE
Os
0.0
100***
1.0
075***
-0.2
625**
-0.9
416***
0.0
102***
1.0
088***
-0.2
665**
-0.9
471***
-0.0
350
Wei
ght
(4.7
5)
(16.8
9)
(-2.3
7)
(-7.9
3)
(4.5
9)
(17.1
8)
(-2.4
3)
(-7.8
1)
(-0.4
7)
OC
CE
Os
0.0
036
1.0
183***
-0.3
451**
-0.7
283***
0.0
045*
1.0
232***
-0.3
594***
-0.7
481***
-0.1
243
(1.4
6)
(14.2
4)
(-2.5
9)
(-5.4
7)
(1.7
4)
(14.4
9)
(-2.6
6)
(-5.8
2)
(-1.5
6)
Diff
eren
ce0.0
064**
-0.0
109
0.0
826
-0.2
133
0.0
057**
-0.0
143
0.0
928
-0.1
991
0.0
893
(2.5
0)
(-0.1
5)
(0.5
2)
(-1.4
5)
(2.1
1)
(-0.1
9)
(0.5
8)
(-1.3
6)
(0.9
7)
Pan
elB
:L
ow
-Tec
hS
am
ple
Equ
al
Non
-OC
CE
Os
-0.0
028
1.1
981***
-0.1
156
-0.0
528
0.0
011
1.1
509***
-0.2
675*
-0.1
636
-0.4
591***
Wei
ght
(-0.9
0)
(10.9
6)
(-0.7
2)
(-0.2
3)
(0.3
7)
(11.0
8)
(-1.8
4)
(-0.8
0)
(-2.9
6)
OC
CE
Os
-0.0
035
1.1
307***
0.2
145
0.2
953
-0.0
029
1.1
224***
0.1
878
0.2
759
-0.0
808
(-0.9
3)
(9.1
5)
(1.0
5)
(1.2
9)
(-0.7
5)
(8.8
8)
(0.9
8)
(1.2
6)
(-0.4
9)
Diff
eren
ce0.0
008
0.0
674
-0.3
301
-0.3
481
0.0
039
0.0
285
-0.4
553**
-0.4
394
-0.3
784*
(0.1
7)
(0.4
0)
(-1.4
7)
(-1.2
2)
(0.9
1)
(0.1
9)
(-2.1
8)
(-1.6
)(-
1.8
9)
Valu
eN
on
-OC
CE
Os
-0.0
027
1.1
236***
-0.5
588***
0.1
164
-0.0
032
1.1
295***
-0.5
396***
0.1
304
0.0
581
Wei
ght
(-0.6
6)
(7.6
7)
(-3.0
8)
(0.4
4)
(-0.6
8)
(7.9
9)
(-2.9
2)
(0.4
8)
(0.2
1)
OC
CE
Os
-0.0
026
0.9
466***
-0.1
200
-0.0
094
-0.0
029
0.9
495***
-0.1
108
-0.0
027
0.0
278
(-0.6
6)
(7.2
8)
(-0.6
9)
(-0.0
4)
(-0.6
7)
(7.0
5)
(-0.6
5)
(-0.0
1)
(0.1
6)
Diff
eren
ce-0
.0001
0.1
769
-0.4
388*
0.1
257
-0.0
003
0.1
801
-0.4
287*
0.1
330
0.0
303
(-0.0
1)
(0.8
5)
(-1.9
4)
(0.3
5)
(-0.0
5)
(0.8
5)
(-1.8
9)
(0.3
6)
(0.0
9)
30
Table 6: Long-Term Abnormal Operating Performance
We compute sample firm’s changes in operating performance for five years following the year in which theyunexpected increase R&D by an economically significant amount. The definition of OPM measure aredescribed in Appendix. We measure abnormal operating performance minus its matched firm’s operatingperformance. We choose matched firms that do not have the same corporate event as the sample in theevent year. We select the a group of control firms, in the same two-digit SIC code as the sample firm, thatdo not unexpectedly significant increase its R&D during the sample firm’s R&D increase year. From thosescreened firms, we choose a firm as the matched firm that has the closest OPM with sample firm’s OPM priorthe sample firm’s R&D increase year. We also create another group of matched firms based on the samplefirm’s characteristics such as size, book-to-market ratio, and momentum. In the beginning of a sample firm’sR&D increase year, we choose a matched firm that the market equity value is within 30% of the marketequity value of the sample firm and then we choose the jointly lowest absolute value of the difference in thecharacteristics. Non-OC CEOs refers to the firms with non-overconfident CEOs, and OC CEOs refers to thefirms with overconfident CEOs. We use the Kruskal-Wallis test to test the median difference.
Year LongholderIndustry and Performance Match Characteristic Match
N Mean Median N Mean Median
Panel A: Changes in OPM1
Non-OC CEOs 96 0.0376*** 0.0119** 96 0.0013 0.0043-1 to +1 OC CEOs 53 −0.0004 −0.0023 53 −0.0210* −0.0093*
Difference 0.0380** 0.0143** 0.0223 0.0136*
Non-OC CEOs 95 0.0719*** 0.0279*** 95 0.0495 0.0047-1 to +2 OC CEOs 52 −0.0006 −0.0092 52 −0.0233** −0.0123*
Difference 0.0725*** 0.0370*** 0.0728** 0.0170*
Non-OC CEOs 93 0.0719*** 0.0401*** 95 0.0379** −0.0032-1 to +3 OC CEOs 52 0.0168 0.0161 52 −0.0120 −0.0013
Difference 0.0552** 0.0240* 0.0499** −0.0020
Non-OC CEOs 92 0.0885*** 0.0538*** 94 0.0555*** 0.0184**-1 to +4 OC CEOs 52 0.0230 0.0154 52 −0.0127 −0.0069
Difference 0.0656** 0.0384** 0.0682*** 0.0253**
Non-OC CEOs 91 0.0962*** 0.0392*** 93 0.0993*** 0.0249***-1 to +5 OC CEOs 52 0.0211 0.0248 52 −0.0082 0.0117
Difference 0.0751** 0.0144* 0.1075*** 0.0132**
Panel B: Changes in OPM2
Non-OC CEOs 96 0.0347** 0.0105** 96 0.0036 0.0035-1 to +1 OC CEOs 53 −0.0043 0.0096 53 −0.0232 −0.0099
Difference 0.0390 0.0009 0.0268 0.0134*
Non-OC CEOs 95 0.0542*** 0.0387*** 95 0.0435 0.0126-1 to +2 OC CEOs 52 0.0140 0.0334* 52 −0.0271* −0.0098
Difference 0.0402* 0.0053 0.0706* 0.0224
Non-OC CEOs 95 0.1694* 0.0381*** 95 0.0371* 0.0100-1 to +3 OC CEOs 52 0.0487* 0.0340** 52 −0.0108 0.0040
Difference 0.1207 0.0041 0.0478* 0.0060
Non-OC CEOs 94 0.1505*** 0.0835*** 94 0.0660*** 0.0323***-1 to +4 OC CEOs 52 0.0375* 0.0293* 52 −0.0124 −0.0076
Difference 0.1130** 0.0541** 0.0784*** 0.0399**
Non-OC CEOs 90 0.0937*** 0.0981*** 93 0.1065*** 0.0445***-1 to +5 OC CEOs 52 0.0394* 0.0432** 52 −0.0065 0.0152
Difference 0.0543* 0.0549 0.1129*** 0.0293**
The symbols ***, **, * indicate statistical significance at the 1, 5, 10 percent levels, respectively.
31
Table 7: Subsample Tests for Long-Term Abnormal Operating Performance
This table reports the abnormal operating performance for the sample increasing their R&D from1980 to 1994. The sample are divided into high-tech and low-tech firms using the definition inBrown, Fazzari, and Peterson (2009). See Table 5 for a detailed description of the test procedures.
Year Longholder
Industry and Performance Match Characteristic Match
OPM1 OPM2 OPM1 OPM2
N Mean N Mean N Mean N Mean
Panel A: High-Tech Sample
Non-OC CEOs 76 0.0323** 76 0.0331* 76 0.0077 76 0.0111-1 to +1 OC CEOs 38 0.0051 38 0.0057 38 −0.0250 38 −0.0285
Difference 0.0272 0.0273 0.0327 0.0396
Non-OC CEOs 75 0.0732*** 75 0.0551*** 75 0.0686* 75 0.0584-1 to +2 OC CEOs 37 0.0083 37 0.0231 37 −0.0284* 37 −0.0339*
Difference 0.0649** 0.0320 0.0970** 0.0922**
Non-OC CEOs 73 0.0828*** 75 0.2097* 75 0.0577*** 75 0.0561**-1 to +3 OC CEOs 37 0.0290 37 0.0372 37 −0.0179 37 −0.0167
Difference 0.0538* 0.1725 0.0755*** 0.0728**
Non-OC CEOs 72 0.0858*** 74 0.1680*** 74 0.0778*** 74 0.0895***-1 to +4 OC CEOs 37 0.0287 37 0.0529* 37 −0.0206 37 −0.0199
Difference 0.0572* 0.1152* 0.0984*** 0.1094***
Non-OC CEOs 71 0.1053*** 70 0.1004*** 73 0.1259*** 73 0.1340***-1 to +5 OC CEOs 37 0.0266 37 0.0548* 37 −0.0191 37 −0.0161
Difference 0.0787** 0.0456 0.1450*** 0.1501***
Panel B: Low-Tech Sample
Non-OC CEOs 20 0.0579* 20 0.0406 20 −0.0230 20 −0.0249-1 to +1 OC CEOs 15 −0.0143 15 −0.0298 15 −0.0110 15 −0.0099
Difference 0.0722** 0.0704** −0.0119 −0.0150
Non-OC CEOs 20 0.0670* 20 0.0506* 20 −0.0220 20 −0.0125-1 to +2 OC CEOs 15 −0.0227* 15 −0.0085 15 −0.0107 15 −0.0105
Difference 0.0897** 0.0591* −0.0114 −0.0020
Non-OC CEOs 20 0.0322 20 0.0183 20 −0.0363* 20 −0.0345*-1 to +3 OC CEOs 15 −0.0134 15 0.0772 15 0.0023 15 0.0037
Difference 0.0456* −0.0589 −0.0386 −0.0383
Non-OC CEOs 20 0.0983 20 0.0855* 20 −0.0270* 20 −0.0207-1 to +4 OC CEOs 15 0.0090 15 −0.0004 15 0.0067 15 0.0062
Difference 0.0893 0.0859* −0.0337 −0.0270
Non-OC CEOs 20 0.0636 20 0.0703** 20 0.0022 20 0.0060-1 to +5 OC CEOs 15 0.0074 15 0.0015 15 0.0188 15 0.0173
Difference 0.0562 0.0688* −0.0166 −0.0113
The symbols ***, **, * indicate statistical significance at the 1, 5, 10 percent levels, respectively.
32
Table 8: Cross-Sectional Regression of Long-Term Abnormal Operating Performance
In this table, we estimate the equation as follows:
OPMit = θ0 + θ1OCit + xit−1β + εit
where OPMit refers to the five-year abnormal operating performance following R&D significantly increases for firm i inyear t. OCit refers to the CEO overconfidence indicator, Longholder, for firm i in year t, and xit−1 is a vector of controlvariables of firm i in year t− 1. The definitions of control variables, x, are described in Appendix. In addition, we measurethe following equation by adding the high-tech indicator as the explanatory variable and its interaction term between CEOoverconfidence indicator:
OPMit = θ0 + θ1OCit + θ2HTit + θ3OCit ×HTit + xit−1β + εit
where OPMit refers to the five-year abnormal operating performance following R&D significantly increases for firm i inyear t. OCit refers to the CEO overconfidence indicator, Longholder, for firm i in year t. HTit refers to the high-techindustries indicator, and xit−1 is a vector of control variables of firm i in year t − 1. Standard errors are clustered atfirm level, heteroskedasticity consistent, and reported in the parentheses. The bottom presents the results of Wald test forθ1 + θ3 = 0, and P -values are reported in the parentheses.
Industry and Performance Match Characteristics Match
OPM1 OPM2 OPM1 OPM2
(1) (2) (3) (4) (5) (6) (7) (8)
Intercept −0.062 −0.147 −0.052 −0.084 −0.090 −0.340 −0.0661 −0.374(−0.31) (−0.68) (−0.31) (−0.49) (−0.50) (−1.11) (−0.33) (−1.30)
Longholder −0.070 −0.027 −0.057 −0.060 −0.079 0.092 −0.082 0.081(−2.07)** (−0.47) (−1.71)* (−1.12) (−1.58) (1.07) (−1.71)* (0.87)
High Tech 0.090 0.043 0.274 0.293(1.20) (0.66) (1.91)* (2.04)**
Longholder×High-Tech −0.060 0.003 −0.244 −0.225(−0.81) (0.04) (−1.76)* (−1.51)
Tobin q 0.009 0.010 0.029 0.030 0.084 0.071 0.069 0.086(0.42) (0.44) (1.12) (1.14) (2.40)** (1.95)* (1.80)* (2.66)**
R&D/Asset −0.153 −0.408 0.004 −0.160 −1.002 −1.467 −0.774 −1.788(−0.34) (−0.76) (0.01) (−0.30) (−1.41) (−2.58)** (−1.12) (−3.09)***
ROA 0.235 0.203 −0.198 −0.263 −0.992 −0.685 −0.672 −1.056(0.60) (0.51) (−0.47) (−0.59) (−1.90)* (−1.45) (−1.38) (−2.01)*
Sales Growth −0.110 −0.106 0.000 0.009 0.034 0.052 0.050 0.042(−1.52) (−1.52) (0.00) (0.07) (0.16) (0.27) (0.25) (0.20)
Cash 0.087 0.127 0.173 0.187 0.178 0.388 0.264 0.308(0.47) (0.68) (1.00) (1.05) (0.63) (1.31) (0.91) (1.07)
PPE/Emp 0.000 0.001 0.000 0.000 −0.001 0.000 −0.001 0.000(0.27) (0.66) (−0.15) (0.07) (−1.12) (0.03) (−0.79) (−0.14)
Stock Ownership 0.157 0.093 −0.025 −0.042 0.181 −0.165 0.063 −0.042(0.51) (0.30) (−0.08) (−0.12) (0.45) (−0.41) (0.17) (−0.10)
Vested Options 2.435 1.976 2.460 2.211 7.420 5.626 6.901 6.046(0.93) (0.70) (0.65) (0.57) (1.37) (1.13) (1.32) (1.18)
(Vested Options)2 −36.208 −29.684 −65.129 −62.168 −114.973 −83.176 −102.889 −94.651(−0.86) (−0.66) (−1.08) (−1.01) (−1.67) (−1.27) (−1.51) (−1.43)
Year Effects Yes Yes Yes Yes Yes Yes Yes YesSIC 2-digit Effects Yes Yes Yes Yes Yes Yes Yes Yes
Obs 128 128 127 127 130 130 130 130
Adjusted R2 0.3331 0.3386 0.1532 0.1386 0.1846 0.2157 0.1758 0.2225
Wald Tests: θ1 + θ3 = 0 −0.087 −0.057 −0.152 −0.144P -value (0.04)** (0.14) (0.06)* (0.09)*
The symbols ***, **, * indicate statistical significance at the 1, 5, 10 percent levels, respectively.
33