Printer Friendly

CEO overconfidence and long-term performance following R&D increases.

We examine the relation between Chief Executive Officer (CEO) overconfidence and significant increases in research and development (R&D) expenditures. Although prior studies reveal a significantly positive market reaction to increases in R&D expenditures in both the long and short run, we find that long-run stock performance is positive only for firms whose CEOs are not overconfident. Our findings, which may be attributable to overinvestment and the overestimation of future cash flows, imply that R&D resulting from overconfident behavior does not provide any value to firms.

**********

Many prior studies in finance report that overconfident Chief Executive Officers (CEOs) affect the investment decisions of their firms. The theory regarding CEO overconfidence essentially stems from a prominent stylized fact, namely, the "better-than-average" effect in the psychology literature (Larwood and Whittaker, 1977; Svenson, 1981; Alicke, 1985). Psychology studies suggest that people generally overestimate their wisdom and skills relative to an average benchmark. As a result, they are more likely to attribute positive outcomes to their own actions and negative outcomes to bad luck or external factors.

Several prior studies attribute overconfidence to three main factors: 1) the illusion of control, 2) a high degree of commitment to specific results, and 3) abstract reference points (Weinstein, 1980; Alicke et al., 1995). These three factors trigger overconfidence and are closely related to the role of the CEO and the characteristics of corporate investment decisions. Specifically, CEOs, as the highest ranking officer in their companies, believe that strategic outcomes are under their control. In addition, they are typically highly committed to better performance of the firm and thus are driven to produce particular outcomes. Moreover, corporate investment decisions are complicated, and their success can be affected by various factors. With such abstract reference points, CEOs may overestimate their ability to select profitable investment projects.

Focusing on the overestimation of future cash flows, Malmendier and Tate (2005) propose a simple model to demonstrate empirically that managerial overconfidence may cause corporate investment distortions. They find that the sensitivity of investment to cash flows is strongest when discernible overconfidence is present. Malmendier and Tate (2008) suggest that overconfidence among CEOs may help to explain merger and acquisition (M&A) decisions. Since overconfident CEOs tend to overestimate their ability to generate higher returns, they pursue value-destroying M&As and pay excessive amounts for target companies.

Malmendier and Tate (2008) find that overconfident CEOs are, in general, more likely to engage in M&As. Specifically, they find that the probability of a CEO making an acquisition is 65% higher if the CEO is regarded as overconfident. The probability of a merger is greater if the merger involves diversification and does not require external financing. When compared to merger announcements by non-overconfident CEOs, the market reaction to announcements by overconfident CEOs is significantly negative.

Ben-David, Graham, and Harvey (2013) provide an alternate theoretical framework for the analysis of the relation between managerial overconfidence and corporate policies. They suggest that overconfident managers either tend to underestimate the volatility of their firms' future cash flows or overweight private signals relative to public information. They posit and find that based on an empirical examination of survey data from hundreds of Chief Financial Officers (CFOs), overconfident managers tend to invest more than less confident managers.

Goel and Thakor (2008) propose a theoretical model on the relations among overconfidence, CEO selection, and corporate governance. The study focuses on three types of CEOs: 1) excessively overconfident, 2) moderately overconfident, and 3) excessively diffident CEOs. Their model provides an explanation of the paradox that overconfident managers are more likely to be promoted to CEO, but later forced to leave the office. In addition, they predict that when compared to moderately confident CEOs, excessively overconfident and excessively diffident CEOs face a greater likelihood of forced turnover. Goel and Thakor (2008) further demonstrate that excessively overconfident and excessively diffident CEOs ultimately reduce the value of the firm as a result of overinvestment and underinvestment, respectively. In a subsequent empirical examination, Campbell et al. (2011) find strong support for Goel and Thakor's (2008) model.

Other empirical studies focusing on relevant research include the following. Malmendier, Tate, and Yan (2011) examine the relation between overconfidence and major financial decisions. In addition, Billett and Qian (2008) study the linkage between overconfidence and M&A frequencies, and Liu and Taffler (2008) focus on the relation between overconfidence and M&A decision making. Finally, Hribar and Yang (2010) find that overconfidence increases the issuance of overly optimistic management earnings forecasts, thereby leading to greater earnings management.

Although, as noted, significant prior research investigates issues related to CEO overconfidence and associated financial decisions, few studies examine CEO overconfidence and research & development (R&D) expenditures. To the best of our knowledge, only two studies address a similar topic. Galasso and Simcoe (2011) and Hirshleifer, Low, and Teoh (2012) both examine the association between CEO overconfidence and CEO behavior with regard to innovation and find that overconfident CEOs are more likely to move their firms toward innovation. These studies also examine the relation between CEO overconfidence and patent citations and provide evidence that overconfident CEOs obtain greater numbers of patents and patent citations.

Our study contributes to the extant literature by combining CEO overconfidence and R&D investment decisions, but differs from the analyses of Galasso and Simcoe (2011) and Hirshleifer et al. (2012) in two important ways. First, we investigate the relation between CEO overconfidence and the effects of significant increases in R&D expenditures. Both Galasso and Simcoe (2011) and Hirshleifer et al. (2012) examine the general relation between overconfident CEOs and their innovation behavior. Because a firm is likely to increase its R&D expenditures as it grows, regular R&D increases may not have an economical impact on the firm. Therefore, we specifically focus only on firms that increase their R&D by an economically significant amount. Additionally, we examine long-run abnormal stock returns and operating performance following significant increases in R&D expenses from the perspective of CEO overconfidence. Neither Galasso and Simcoe (2011) and Hirshleifer et al. (2012) examine the long-run performance after R&D increases. Since R&D benefits are usually realized over a longer horizon, long-run abnormal performance is an important issue to explore.

To motivate our study, we argue that the R&D decisions of overconfident CEOs are an interesting research issue for at least three reasons. First, increases in R&D represent a managerial decision taken by CEOs. Eberhart, Maxwell, and Siddique (2004) argue that in the long run, significant increases in R&D are beneficial to the operating performance of firms. Prior studies argue that managerial overconfidence often results in overinvestment. Such a perspective raises the question as to whether overconfident CEOs make profitable R&D investment decisions (Malmendier and Tate, 2005, 2008; Goel and Thakor, 2008; Ben-David et al., 2013). We conjecture that overconfident CEOs make less beneficial R&D decisions than those who are not overconfident. Additionally, many prior studies find that market reaction to increases in R&D expenditures is significantly positive (Chan, Martin, and Kensinger, 1990; Zantout and Tsetsekos, 1994; Szewczyk, Tsetsekos, and Zantout, 1996; Eberhart et al., 2004). However, since overconfident CEOs tend to overinvest, the quality of their R&D projects may be lower. Thus, we posit that firms with overconfident CEOs may not experience long-run positive drift of stock returns following R&D increases due to overinvestment.

Moreover, although R&D expenditures are just one type of investment decision made within a firm, they have unique features that differ significantly from other types of long-term investments. Unlike investments, such as capital expenditures and mergers, R&D expenditures convey not only tangible information, but also reflect intangible information regarding the future cash flows of the firm. R&D investment often results in new technology or a new product, but it can be risky and profits (or losses) may take years to be fully realized. Einhorn (1980) points out that overconfidence tends to be more severe with ambiguous and deferred feedback. Since it takes an extended period of time to determine the success of an R&D project (i.e., the information is less tangible), CEO overconfidence may play an even more important role relative to tangible investments.

We use various methodologies to detect the long-run abnormal stock returns and operating performance of our sample firms. We find consistent evidence that non-overconfident CEOs exhibit significantly positive long-run abnormal stock returns following increases in R&D. We do not find similar evidence for overconfident CEOs. These findings suggest that overconfident CEOs tend to overestimate the future prospects of their investment projects. Our results differ from the prior studies that conclude that investors, in general, earn positive abnormal stock returns following increases in R&D expenditures (Chan et al., 1990; Eberhart et al., 2004). In addition, we find that R&D investment is beneficial, measured by operating performance, only when the firm's CEO does not exhibit overconfidence. The results on both the stock and operating performance are robust for the high-tech samples, but weaker for the low-tech samples.

The remainder of this paper is organized as follows. Section I presents a description of the data, while Section II offers an introduction and explanation of the methodology adopted for our study. Section III provides an analysis and discussion of the empirical results. Finally, in Section IV, we present our conclusions.

I. Data

A. Sample Construction

The construction of our data set on CEO overconfidence is based on Yermack (1995) and Hall and Liebman (1998) and is identical to Malmendier and Tate's (2008) data set. The sample consists of 477 large publicly traded US firms from 1980 to 1994. To be eligible for inclusion in the sample, a firm must be listed among Forbes magazine's largest firms in the United States at least four times from 1980 to 1994. The data set is constructed on the basis of complete information regarding the stock and option holdings of CEOs, thereby providing a comprehensive and detailed picture of the portfolio of each CEO during his/her tenure.

The stock price data for our sample are obtained from the Center for Research in Security Prices (CRSP). We include only ordinary common equities with CRSP share type codes of 10 or 11 for our estimation of abnormal stock returns. The accounting data used for the computation of the abnormal operating performance of the sample firms are retrieved from the Compustat database.

We follow Eberhart et al. (2004) to impose three major requirements for our measure of unexpected increases in R&D expenditures. First, we define an unexpected R&D increase as an increase in the ratio of R&D to the assets of a firm. Next, we concentrate on those firms with an R&D intensity level (defined as the ratios of R&D to assets and R&D to sales) in excess of 5%. Finally, we focus only on an economically significant increase in R&D; thus, the firm must have increased both its dollar R&D and its ratio of R&D to assets by at least 5%. To allow the dissemination of the accounting data information throughout the market, we follow Eberhart et al. (2004) and begin measuring the returns from the fourth month after the fiscal year-end in which the firm increased its R&D. The definitions of all variables are provided in the Appendix. (1)

B. Measuring Overconfidence

Our CEO overconfidence measures are the same as those adopted by Malmendier and Tate (2005, 2008), who used panel data on the personal portfolios of CEOs to identify whether they were overconfident based on the exercise of executive options. CEOs obtain the right to purchase shares in their company from executive options with the exercise price of the options usually set as the stock price on the grant date. The duration of most executive options is 10 years, and the vesting period during which the options are nonexercisable is normally four years.

Merton (1973) argues that investors should not exercise their options early, essentially for European call options, as the time value is nonnegative for nondividend paying stocks. However, this logic may not be applicable to executive options as they are nontradable and CEOs cannot short sell their holdings to hedge the idiosyncratic risk of their firms' stocks. Furthermore, CEOs' wealth is directly exposed to firm risk because they invest their human capital in the firm and a large proportion of their compensation is equity-based. As a result, risk-averse CEOs should exercise their option grants early at times when the stock price is sufficiently high. Lambert, Larcker, and Verrecchia (1991) propose a theoretical framework to demonstrate that risk-averse CEOs should exercise in-the-money executive stock options prior to expiration to avoid their exposure to firm-specific risk. In addition, Hall and Murphy (2002) discuss the optimal exercise and valuation of executive stock options. They suggest that the optimal schedule for exercise depends upon CEO wealth, their risk aversion level, and diversification. In our sample, CEOs persistently refuse to exercise deep in-the-money vested options until the year of expiration. Our interpretation of this delay in exercise behavior is overconfidence (i.e., overestimation of the future returns of the firm). Malmendier and Tate (2008) find that alternative reasons for such behavior, including positive inside information, signaling, board pressure, risk tolerance, and taxes, fail to adequately explain the delay in exercise behavior among overconfident CEOs. They provide strong empirical evidence that overconfident CEOs believe that they are capable of making profitable M&A decisions.

Malmendier and Tate (2008) define their CEO overconfidence indicator variable, longholder, as CEOs, who at least once during their tenure, keep an option until the year of expiration even though that option was at least 40% in-the-money entering its final year. The exercise threshold of 40% is calibrated from Hall and Murphy's (2002) model, which assumes that CEOs have a constant relative risk aversion of three and that 67% of their wealth is in the firm's stock. (2) However, the longholder may result in a forward-looking bias. (3) In other words, consider a CEO with tenure from 1985 to 1995. He/she is classified as an overconfident CEO from the year 1985 even if it was the year 1990 when he/she for the first time did not exercise options that were at least 40% in the money with one year to maturity. Thus, to mitigate this bias, we adopt another Malmendier and Tate (2008) overconfidence indicator, postlongholder. Postlongholder categorizes CEOs as overconfident only after the year that they, for the first time, keep an option that ever exceeded 40% in-the-money until expiration.

II. Methodology

There are several studies focusing on the estimation of long-run abnormal stock returns. Barber and Lyon (1997) demonstrate that the arithmetical summation of returns (i.e., cumulative returns) does not reflect the actual returns of investors as precisely as buy-and-hold abnormal returns (BHARs). Fama (1998) argues that event-time returns exhibit a cross-sectional dependence problem that leads to a downward bias in the standard error, while Lyon, Barber, and Tsai (1999) suggest that the adoption of the calendar-time return method can mitigate the cross-sectional correlation problem. Some debate also surrounds the use of value-weighted versus equal-weighted calendar-time returns. Loughran and Ritter (2000) argue that the equal-weighted scheme is more appropriate, since mispricing is more likely to occur among smaller firms. Alternatively, Fama (1998) argues that a value-weighted portfolio is more appropriate, essentially because it considers the total wealth effects experienced by investors.

Although considerable debate continues to surround the empirical estimation of long-run abnormal stock returns, the two widely adopted empirical methods are: 1) the BHARs of the firm relative to a benchmark firm, and 2) calendar-time abnormal returns (CTARs) in which a factor model is used to estimate the risk-adjusted returns.

A. Buy-and-Hold Abnormal Returns

We compute the one-, three-, and five-year BHARs for all of our sample firms with a threemonth lag following the end of the fiscal year with a significant increase in R&D. [BHAR.sub.i,T1,T2] for the ith sample firm, from month [T.sub.1] to [T.sub.2] is expressed as follows:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (1)

where [R.sub.i,t] is the return of the sample firm in event month t, and [R.sub.b,t] is the return of the benchmark over the same period. The test statistics used are as follows:

t = [square root of n] x [bar.BHAR]/[sigma](BHAR), (2)

where [bar.BHAR] is the average across firms, [sigma](BHAR) is the cross-sectional standard deviation of the BHARs of all sample firms, and n is the number of firms.

Barber and Lyon (1997) note that the use of control firms effectively eliminates both new listings and rebalancing biases and also addresses the problem of skewness. Thus, we apply their matching firm approach to find our benchmark. To fit the research design for this study, we set several requirements for our benchmarks. First, we restrict all benchmark firms to the same CEO type as the sample firms in the year of significant R&D increases. Additionally, we require the matching firms to have the same two-digit standard industrial classification (SIC) codes as the sample firms. Finally, we follow most related empirical studies by identifying all firms with a market value of equity within 30% for the market value of equity of the sample firm at the beginning of the year in which the firm experiences a significant increase in R&D. From this set of firms, we then select the matching firm with the closest book-to-market (BM) equity ratio to that of the sample firm.

Momentum controls for the potential bias that a CEO is more likely to be classified in the overconfident group if the stock price has increased sufficiently in the past for the CEO's options to turn in-the-money. Therefore, we also choose matching firms by controlling firm size, BM, and momentum, which is defined as the returns for the previous 12-month period. Similar to the previous matching criteria, after controlling CEO type and industry, we identity all firms with a market value of equity and BM equity ratio within 30% of the two variables of the sample firm at the beginning of the year in which R&D increases. From this set of firms, we then select the matching firm that has the closest momentum to the sample firm. (4)

We cannot always find a matching firm for our samples. However, because the percentage of unmatched observations is small, we delete those observations from our analysis. For both matching criteria, if the first matching firm is delisted or has missing data, we replace it with the second matching firm based on the matching criteria. Finally, the matching pools are different for calculating BHARs of different horizons. Specifically, for the one-year BHARs, we exclude firms that have had significant R&D increases within the past one year from the potential pool of matching firms, and for the five-year BHARs, we exclude firms that have had significant R&D increases within the past five years from the pool of matching candidates.

B. Calendar-Time Abnormal Returns

Fama (1998) points out that the problem of cross-sectional dependence of the returns is particularly severe when the sample event dates are close together. As such, we use a calendar-time approach to help mitigate this bias. (5) We use the following equation, which is based on the Fama-French (1993) three-factor model, to estimate the average monthly abnormal returns, [[alpha].sub.i]:

[R.sub.pt] - [R.sub.ft] = [[alpha].sub.p] + [[beta].sub.p]([R.sub.mt] - [R.sub.ft]) + [s.sub.p][SMB.sub.t] + [h.sub.p][HML.sub.t] + [[epsilon].sub.pt], (3)

where [R.sub.pt] is the monthly return on the equal- or value-weighted portfolio in calendar month t(a sample stock is included if month t is within the 60-month period following its R&D increase), [R.sub.ft] is the return of a one-month Treasury bill, [R.sub.mt] is the CRSP value-weighted market index return, [SMB.sub.t] is the difference in the returns on the value-weighted portfolios of small and large stocks, and [HML.sub.t] is the difference in the returns on the value-weighted portfolios of high and low BM stocks. (6) We also follow Carhart's (1997) model to measure the monthly abnormal return following an increase in R&D by estimating the intercept from the following equation:

[R.sub.pt] - [R.sub.ft] = [[alpha].sub.p] + [[beta].sub.p]([R.sub.mt] - [R.sub.ft]) + [s.sub.p][SMB.sub.t] + [h.sub.p][HML.sub.t] + [m.sub.p][UMD.sub.t] + [[epsilon].sub.pt], (4)

where [UMD.sub.t] is the difference in the returns on a value-weighted portfolio of high- and low-momentum stocks. (7)

Berk, Green, and Naik (2004) argue that the systematic risk of a firm may change as a result of R&D investments. To account for the time-varying risk, we also estimate each factor loading in Equations (3) and (4) based on a rolling regression approach. Specifically, we use the first 60 months to estimate the factor loadings in Equations (3) and (4), and then estimate the abnormal return in the 61st month as the difference between the actual and expected portfolio returns. The expected portfolio return is computed as the estimated factor loadings multiplied by 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.

C. Operating Performance Measures

We compute the operating performance of the sample firms from the first year to the fifth year following the year in which the firm experiences a significant increase in R&D expenditures. Following Eberhart et al. (2004), we use earnings before interest and tax (EBIT) divided by sales (OPM1) and the sum of EBIT and after-tax R&D divided by sales (OPM2) as our operating performance measures (OPMs). We define abnormal operating performance as the raw operating performance of a sample firm minus the operating performance of its control firm.

We require the matching firms to have the same CEO type as our sample firms. We then use two specific criteria for our selection of the matching firms. First, following prior studies (Barber and Lyon, 1996), we set the criteria based on the industry of the firms and their pre-event performance. As such, we select a group of control firms that have the same two-digit SIC code as the sample firm and that do not experience any significant increases in R&D during the same year as the sample firms. For each of our sample firms, we then select a matched control firm from these screened firms. Specifically, we select the firm with the closest OPMs to those of the sample firm during the year prior to the year of the R&D increase.

Additionally, we form another group of control firms based on the characteristics of the sample firm: namely, firm size, BM, and momentum. At the beginning of the year in which a sample firm experiences an economically significant increase in R&D, we select a matching firm for which the market value of equity is within 30% of that for the sample firm, and then select the joint lowest absolute value of the difference in size, BM equity ratio, and momentum.

We cannot always find a matching firm for our samples. However, since the percentage of unmatched observations is small, we delete those observations from our analysis. For both matching criteria, if the first matching firm is delisted or has missing data, we replace it with the second matching firm based on the matching criteria. We further note that we exclude firms that have had significant R&D increases within the past five years from the pool of matching candidates. Finally, we compare the relative operating performance improvements between the sample firm and the control firm. (8)

III. Empirical Results

A. Descriptive Statistics

The descriptive statistics for the sample firms are reported in Panel A of Table I where all of the dollar variables are adjusted to 1994 dollars. There are 121 firm-year observations for non-overconfident CEOs and 18 firm-year observations for overconfident CEOs. For the year in which the sample firms experience an increase in R&D expenditures, those firms with overconfident (non-overconfident) CEOs had mean annual sales of US $4,759 million (US $5,280 million). For those firms with overconfident (non-overconfident) CEOs, the mean book value of total assets was US $4,246 million (US $4,492 million), and the mean market value of common stocks was US $7,581 million (US $4,960 million). We compute the ratio of BM equity following Daniel and Titman's (1997) definition of book equity. The mean and median BM equity ratios for firms with overconfident CEOs are not significantly different from those for firms with non-overconfident CEOs.

Panel A of Table I indicates that almost all of the variables on firm characteristics are statistically indifferent between firms with overconfident CEOs and firms with non-overconfident CEOs for both mean and median difference tests. The mean and median difference tests on the R&D intensity measures between overconfident and non-overconfident CEOs are also insignificant in most cases. In sum, based on the mean and median difference tests, the results in Table I suggest that for firms with significant increases in R&D, firm characteristics are not statistically significantly different between firms with overconfident CEOs and firms with non-overconfident CEOs.

In order to reveal the effectiveness of matching, we also report the descriptive statistics for the matching firm in Panel B of Table 1. The matching criteria are based on size/BM/CEO type/industry. Given data limitations, matching is not expected to be perfect. However, we note, from the comparison of Panels A and B, that the means and medians of market capitalization and BM equity are similar between sample and matching firms. This finding suggests that our matching effectiveness is reasonably sound.

B. Buy-and-Hold Abnormal Returns

Table II reports the one-, three-, and five-year BHARs following significant increases in R&D based on two sets of matching criteria. (9) Panel A matches firms based on size/BM/CEO type/industry, while Panel B includes the additional criteria, momentum, to the existing matching approach. We use postlongholder as the measure of CEO overconfidence to mitigate the concern regarding forward-looking bias. The median test results of the BHARs in Table II are similar to those for the mean test. As such, our continuing discussion concentrates on the results of the mean test.

Firms with non-overconfident CEOs generally exhibit positive abnormal stock returns following an increase in R&D. Panel A (Panel B) of Table II indicates that non-overconfident CEOs earn about 4% (5%) for a one-year investment, 30% (33%) for a three-year investment, and 56% (67%) for a five-year investment. For the longer holding periods of three and five years, firms with non-overconfident CEOs exhibit significantly positive abnormal stock returns when compared to their benchmarks.

Conversely, the one- and five-year BHARs for those firms with overconfident CEOs do not exhibit any significant difference from zero, with the three-year BHARs for overconfident CEOs exhibiting significantly negative returns. Furthermore, for the three-year and five-year holding periods, the difference in BHARs between firms with non-overconfident and overconfident CEOs is positively significant. In other words, the average BHARs for firms with non-overconfident CEOs are significantly greater than those of firms with overconfident CEOs.

Although Eberhart et al. (2004) suggest that because R&D expenditures usually convey intangible information, investors tend to underreact to the R&D-investing firm's stock price. Our findings suggest that this argument only holds for non-overconfident CEOs. Since CEOs who are overconfident consider themselves to be better-than-average managers, they may overestimate their future cash flows after investing in an R&D project. In addition, for firms with an overconfident CEO, the increase in R&D expenditures itself may be another form of overinvestment.

Next, we conduct a multivariate analysis of the relation between CEO overconfidence and long-term stock performance. We use the five-year BHARs in Table II as dependent variables in cross-sectional regressions. Table III provides the results. Models (1) and (3) indicate that after controlling for other factors, the BHARs following an increase in R&D are significantly lower for those firms with overconfident CEOs than for those firms with non-overconfident CEOs.

Furthermore, many prior studies recognize that R&D expenditures may have very different effects on high-tech and low-tech industries essentially because the R&D investment attributes of high-tech industries are more complicated than those of low-tech industries (Chan et al., 1990; Eberhart et al., 2004; Brown, Fazzari, and Petersen, 2009). Therefore, we include high-tech industry dummies in Models (2) and (4) of Table III, along with the interaction term between CEO overconfidence and high-tech industry. (10) At a first glance, in contrast to the results for [[theta].sub.1] in Models (1) and (3), we find that the coefficients on [[theta].sub.1] are no longer significantly negative. (11) However, the coefficients of [[theta].sub.3] in Models (2) and (4) are significantly negative. These results suggest that after controlling other factors, the difference in BHARs between overconfident and non-overconfident CEOs is more negative for high-tech industries. In addition, to examine the total effect of CEO overconfidence within high-tech industries, we test the sum of [[theta].sub.1] and [[theta].sub.3]. The penultimate row in Table III reports the results of the Wald test for [[theta].sub.1] + [[theta].sub.3] = 0. As the results indicate, the Wald test is significantly negative for both Models (2) and (4).

In sum, we find that the long-term drift in stock performance following an increase in a firm's R&D expenditures is significantly lower for overconfident CEOs than for non-overconfident CEOs. These results suggest that investor underreaction to the intangible information content of a significant increase in R&D is only applicable to firms with non-overconfident CEOs. One possible explanation is that overconfident CEOs may invest in nonprofitable projects resulting from their overestimation of future cash flows or an underestimation of the risk from R&D investment. In addition, we find that this phenomenon is more pronounced for high-tech firms. (12)

C. Calendar-Time Abnormal Returns

We estimate the risk-adjusted abnormal returns under the calendar-time approach using the Fama-French's (1993) three-factor and Carhart's (1997) four-factor models. The results for the long-term abnormal stock returns following an increase in R&D under the calendar-time approach are reported in Table IV. Panel A presents the ordinary least squares results, while Panel B provides the rolling regression results. (13)

As Panel A of Table IV indicates, the monthly average abnormal returns of non-overconfident CEOs are significantly positive under the equal- and value-weighted schemes in both the Fama-French's (1993) three-factor model (0.74% and 0.63%, respectively) and the Carhart's (1997) four-factor model (0.90% and 0.56%, respectively). However, the abnormal returns of overconfident CEOs are not significantly different from zero in either model for either the equal- and value-weighted schemes. When adopting the Fama-French's (1993) three-factor model, the differences between the abnormal returns of firms with non-overconfident CEOs and firms with overconfident CEOs are significantly positive for both the equal- and value-weighted calendar-time portfolios. Furthermore, when we carry out our estimations using the Carhart's (1997) four-factor model, the difference between the abnormal returns of non-overconfident and overconfident CEOs is also significantly positive for both portfolio-weighted schemes.

Turning to the rolling regression results reported in Panel B of Table IV, in both the Fama-French's (1993) three-factor and Carhart's (1997) four-factor models, the difference between the value-weighted abnormal returns of firms with non-overconfident CEOs and firms with overconfident CEOs is significantly positive. We note that the stronger results from a static model in Panel A may be due to the imperfect measurement of changing risk characteristics. Furthermore, for firms with non-overconfident CEOs, the abnormal stock returns of the calendar-time portfolio are significantly positive in both the Fama-French's (1993) three-factor and Carhart's (1997) four-factor models. These results suggest that our findings remain robust when taking into consideration time-varying portfolio risk.

The subsample test results concerning the abnormal stock returns for the calendar-time portfolio are presented in Table V, where the sample firms are divided into high-tech and low-tech firms following the criteria proposed in Brown et al. (2009). For non-overconfident CEOs, we find significantly positive abnormal returns for the high-tech sample across all categories. In contrast, for overconfident CEOs, the abnormal returns computed under the Fama-French's (1993) and Carhart's (1997) models are insignificantly different from zero for both the value- and equal-weighted schemes. For both the Fama-French's (1993) and Carhart's (1997) models, the differences in the abnormal returns between firms with non-overconfident CEOs and firms with overconfident CEOs are significantly positive under both the equal- and value-weighted scheme.

For low-tech firms, the value- and equal-weighted abnormal returns are insignificantly different from zero for the firms with non-overconfident CEOs and the firms with overconfident CEOs under the Fama-French's (1993) three-factor and Carhart's (1997) four-factor models. Additionally, for low-tech firms, the difference between the abnormal returns of firms with non-overconfident CEOs and firms with overconfident CEOs is also insignificantly different from zero. Thus, the results for the CTARs are consistent with those for the BHARs. (14)

D. Operating Performance

Table VI provides the empirical results regarding the abnormal operating performance for the sample firms. Since the results of the median and mean tests of abnormal operating performance are similar, we focus on the mean results. (15) Panel A reports the changes in the abnormal operating performance for OPM1 following an increase in R&D. The results based on the CEO type, industry, and performance-matching criteria indicate that in most cases following an increase in R&D, firms with non-overconfident CEOs improve their operating performance relative to their matching firm. This finding is consistent with prior literature in which R&D investment is associated in the long run with superior operating performance.

Conversely, following an increase in R&D, firms with overconfident CEOs do not perform any better than their benchmarks, a result that is consistent with the belief that overconfident CEOs overinvest. Consequently, their abnormal operating performance is generally negative. These results suggest that R&D investment is more beneficial to a firm when the CEO does not exhibit overconfidence. We compare the relative changes in abnormal operating performance between firms with overconfident CEOs and firms with non-overconfident CEOs using CEO type, industry, and performance-matching criteria and find that the difference is significantly positive after the third year following an increase in R&D.

Similarly, when using the CEO type and firm characteristics as the matching criteria, the abnormal operating performance for non-overconfident CEOs is positive although weaker after the fourth year following an increase in R&D. The abnormal operating performance for overconfident CEOs is still significantly negative after the second year following R&D increases. Consistent with CEO type, industry, and performance-matching criteria, results from the CEO type and characteristics-matching criteria exhibit a significantly positive difference between the abnormal operating performance of firms with overconfident CEOs and firms with non-overconfident CEOs after the third year following a significant increase in R&D.

Changes in the abnormal operating performance of OPM2 following an increase in R&D are reported in Panel B of Table VI. In general, the results based on CEO type, industry, and performance-matching criteria indicate that following an increase in R&D, firms with nonoverconfident CEOs experience significant improvement in their profit margins, whereas the abnormal operating performance of firms with overconfident CEOs remains significantly negative. Overall, the difference between the abnormal operating performance of firms with nonoverconfident CEOs and firms with overconfident CEOs is significantly positive.

When CEO type and firm characteristics are the matching criteria, the results are qualitatively similar to those obtained using industry and performance-matching firms. The difference between the abnormal operating performance of firms with non-overconfident CEOs and firms with overconfident CEOs remains significantly positive after the fourth year following an increase in R&D. Although Eberhart et al. (2004) argue that R&D investment is generally beneficial, the results in Table VI suggest that an increase in R&D is related to better operating performance only for those firms whose CEOs do not exhibit overconfidence. For firms with overconfident CEOs, such R&D investment projects may prove to be inefficient.

We conclude our analysis with a multivariate examination of the relation between the relative change of the five-year abnormal operating performance and CEO overconfidence. Table VII reports the results of the cross-sectional regression on the five-year abnormal operating performance of Table VI. The coefficient in Models (1), (3), (5), and (7) overall suggest that following a significant increase in R&D, the abnormal operating performance for firms with overconfident CEOs is likely to be lower than that for firms with non-overconfident CEOs. This result is consistent with our empirical finding reported previously. Models (2), (4), (6), and (8) include a high-tech dummy and its interaction term with the CEO overconfidence measure as the explanatory variable. The coefficients of [[theta].sub.3] are significantly negative for all models. Furthermore, for Models (2), (4), (6), and (8), the results of the Wald test in the penultimate row of Table VII are negatively significant. The results of the significant [[theta].sub.3] and the Wald tests suggest that the difference in the abnormal operating performance between firms with non-overconfident CEOs and firms with overconfident CEOs following an increase in R&D is more negative in high-tech industries.

To summarize, we do not find a positive long-term drift in the operating performance following a significant increase in R&D expenditures in firms with an overconfident CEO. These results indicate that a R&D investment increase is associated with better long-run operating performance only in firms with non-overconfident CEOs. One possible explanation for this finding may be that overconfident CEOs may overinvest and overestimate future cash flows from R&D investments. As expected, this phenomenon is more pronounced in high-tech industries. (16)

IV. Conclusion

We investigate the relation between significant increases in R&D and CEO overconfidence. Numerous recent studies have investigated the ways in which CEO overconfidence may affect the corporate decisions of a firm. For example, Malmendier and Tate (2005, 2008) suggest that overconfident CEOs usually have a negative effect on their firms' investment decisions due to their overestimation of future cash flows and overinvestment. However, their studies, which employ data from capital expenditures and M&A events, focus only on the tangible investments of a firm. Alternatively, R&D expenditures not only convey tangible information, but also reflect intangible information on the future cash flows of the firm. R&D investments often result in new technology or a new product, but can also be risky and the results may take years to realize. Einhom (1980) notes that overconfidence tends to be more severe if the feedback on decisions is slow or inconclusive than if the feedback is clear and rapid. Since R&D projects require an extended time period to realize success (or failure) and the information associated with R&D investments is less tangible, CEO overconfidence can play an even more important role in R&D investments relative to tangible investments. To the best of our knowledge, our investigation is the first study to link CEO overconfidence to R&D increases and to provide empirical evidence that CEO overconfidence clearly affects firms' R&D investment decisions.

We examine the long-run abnormal stock returns and operating performance of firms using various methodologies including BHARs and CTARs under different matching firms and asset pricing models. Our results indicate that following a significant increase in R&D, investors in firms whose CEOs do not exhibit overconfidence earn abnormal returns on their stocks. Furthermore, long-run abnormal stock returns are significantly greater for firms with nonoverconfident CEOs than for firms with overconfident CEOs.

We also find consistently strong evidence that increasing R&D is a less beneficial investment decision for those firms with overconfident CEOs. In other words, the long-run abnormal operating performance of firms with non-overconfident CEOs is significantly better than that for firms with overconfident CEOs. Our cross-sectional regression analyses also demonstrate that overconfidence is associated with significantly lower abnormal stock returns and operating performance following a significant increase in R&D. These findings are stronger for high-tech industries than for low-tech industries, likely because R&D expenditures are more essential to the former.

Our results contribute to the link between behavioral corporate finance and R&D investment decisions and fill the gaps in the extant literature by undertaking an in-depth examination of the impact of increases in R&D using an option-based overconfidence measure. Overall, our empirical findings suggest that overconfident CEOs harm the performance of their firms in relation to significant R&D increases. This may be due to the overestimation of future cash flows and overinvestment. Therefore, we find that the positive long-term drift following an increase in R&D reported in general by Eberhart et al. (2004) is only found in those firms with nonoverconfident CEOs.
Appendix: Variable Definitions

Variables                               Definitions

Panel A. CEO Overconfidence Measure and Characteristics **

Longholder            Dummy variable: Equal to one if the CEOs, at
                      least once during their tenure, hold an option
                      grant until the last year before expiration even
                      though the option grant was at least 40%
                      in-the-money entering its last year, and zero
                      otherwise.

Postlongholder        Dummy variable: Equal to one only after the year
                      that the CEOs for the first time keep an option
                      grant that was at least 40% in-the-money
                      entering its last year, and zero otherwise.

CEO stock ownership   Percentage of common stock owned by CEOs and
                      their immediate family.

CEO vested options    CEO's holdings of options that are exercisable
                      within six months divided by the common shares
                      outstanding, and multiplied by 10.

Panel B. Firm Characteristics

Market value of       Fiscal year end price (Compustat Item 25)
common equity         multiplied by outstanding shares (Compustat Item
                      199).

Preferred stock       Redemption value (Compustat Item 56) if
                      available; otherwise, use the liquidating value
                      (Compustat Item 10) if it is available and, if
                      not available, use the carrying value (Compustat
                      Item 130).

Book value of common  Stockholder's equity (Compustat Item 216) plus
equity                any deferred tax (Compustat Item 74) and any
                      investment tax credit (Compustat Item 208),
                      minus any preferred stock.

Market value of       Book value of assets (Compustat Item 6) minus
assets                the book value of common equity plus the market
                      value of common equity.

Tobin q               Market value of assets to the book value of
                      assets.

BM Equity             Book value of common equity to the market value
                      of common equity.

Return on assets      Ratio of operating income before depreciation
                      (Compustat Item 13) to the book value of assets.

BM Equity             Book value of common equity to the market value
                      of common equity.

Return on assets      Panel B. Firm Characteristics Ratio of operating
                      income before depreciation (Compustat Item 13)
                      to the book value of assets.

Sales growth          Sum of sales (Compustat Item 12) minus prior
                      year sales and divided by prior year sales.

Cash                  Ratio of cash and short-term investments
                      (Compustat Item 1) to the book value of assets.

PPE/Emp               Ratio of property, plant, and equipment
                      (Compustat Item 7) to the employees (Compustat
                      Item 29).

Panel C. R&D Intensity Measure

R&D/Assets            R&D (Compustat Item 46) divided by the book
                      value of assets.

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 (one minus taxes
                      (Compustat Item 16) over taxable income
                      (Compustat Item 170) multiplied by R&D) divided
                      by sales.

** We thank Ulrike Malmendier for providing data in Panel A.


References

Alicke, M., 1985, "Global Self-Evaluation as Determined by the Desirability and Controllability of Trait Adjectives," Journal of Personality and Social Psychology 49, 1621-1630.

Alicke, M., M. Klotz, D. Breitenbecher, T. Yurak, and D. Vredenburg, 1995, "Personal Contact, Individuation, and the Better-than-Average Effect," Journal of Personality and Social Psychology 68, 804-825.

Andre, P., M. Kooli, and J.-F. LHer, 2004, "The Long-Run Performance of Mergers and Acquisitions: Evidence from the Canadian Stock Market," Financial Management 33, 27-43.

Barber, B.M. and J.D. Lyon, 1996, "Detecting Abnormal Operating Performance: The Empirical Power and Specification of Test Statistics," Journal of Financial Economics 41, 359-399.

Barber, B.M. and J.D. Lyon, 1997, "Detecting Long-Run Abnormal Stock Returns: The Empirical Power and Specification of Test Statistics," Journal of Financial Economics 43, 341-372.

Ben-David, I., J. Graham, and C. Harvey, 2013, "Managerial Miscalibration," Quarterly Journal of Economics, forthcoming.

Berk, J., R. Green, and V Naik, 2004, "Valuation and Return Dynamics of New Ventures," Review of Financial Studies 17, 1-35.

Billett, M. and Y. Qian, 2008, "Are Overconfident CEOs Born or Made? Evidence of Self-Attribution Bias from Frequent Acquirers," Management Science 54, 1037-1051.

Braga-Alves, M. and K. Shastri, 2011, "Corporate Governance, Valuation, and Performance: Evidence from a Voluntary Market Reform in Brazil," Financial Management 40, 139-157.

Brown, J., S. Fazzari, and B. Petersen, 2009, "Financing Innovation and Growth: Cash Flow, External Equity, and the 1990s R&D Boom "Journal of Finance 64w, 151-185.

Campbell, T.C., M. Gallmeyer, S. Johnson, J. Rutherford, and B. Stanley, 2011, "CEO Optimism and Forced Turnover f Journal of Financial Economics 101, 695-712.

Carhart, M., 1997, "On Persistence in Mutual Fund Performance," Journal of Finance 52, 57-82.

Carter, R., F. Dark, I. Floros, and T. Sapp, 2011, "Characterizing the Risk of IPO Long-Run Returns: The Impact of Momentum, Liquidity, Skewness, and Investment," Financial Management 40, 1067-1086.

Chan, K., J. Cooney, Jr., J. Kim, and A. Singh, 2008, "The IPO Derby: Are There Consistent Losers and Winners on This Track?" Financial Management 37, 45-79.

Chan, S.H., J. Martin, and J. Kensinger, 1990, "Corporate Research and Development Expenditures and Share Value," Journal of Financial Economics 26, 255-276.

Daniel, K. and S. Titman, 1997, "Evidence on the Characteristics of Cross Sectional Variation in Stock Returns," Journal of Finance 52, 1-33.

Eberhart, A., W. Maxwell, and A. Siddique, 2004, "An Examination of Long-Term Abnormal Stock Returns and Operating Performance Following R&D Increases," Journal of Finance 59, 623-650.

Einhorn, H., 1980, "Overconfidence and Judgment," New Directions for Methodology of Social and Behavioral Science 4, 1-16.

Fama, E., 1998, "Market Efficiency, Long-Term Returns, and Behavioral Finance," Journal of Financial Economics 49, 283-306.

Fama, E. and K. French, 1993, "Common Risk Factors in the Returns on Bonds and Stocks," Journal of Financial Economics 33, 3-56.

Galasso, A. and T. Simcoe, 2011, "CEO Overconfidence and Innovation," Management Science 57, 1469-1484.

Gao, H., 2010, "Market Misvaluation, Managerial Horizon, and Acquisitions," Financial Management 39, 833-850.

Goel, A. and A. Thakor, 2008, "Overconfidence, CEO Selection, and Corporate Governance," Journal of Finance 63, 2737-2784.

Hall, B. and J. Liebman, 1998, "Are CEOs Really Paid Like Bureaucrats?" Quarterly Journal of Economics 113, 653-691.

Hall, B. and K. Murphy, 2002, "Stock Options for Undiversified Executives," Journal of Accounting Economics 33, 3-42.

Hirshleifer, D., A. Low, and S. Teoh, 2012, "Are Overconfident CEOs Better Innovators?" Journal of Finance 67, 1457-1498.

Hribar, P. and H. Yang, 2010, "Does CEO Overconfidence Affect Management Forecasting and Subsequent Earnings Management?" University of Iowa Working Paper.

Ikenberry, D. and S. Ramnath, 2002, "Underreaction to Self-Selected News: The Case of Stock Splits," Review of Financial Studies 15, 489-526.

Lambert, R., D. Larcker, and R. Verrecchia, 1991, "Portfolio Considerations in Valuing Executive Compensation," Journal of Accounting Research 29, 129-149.

Larwood, L. and W. Whittaker, 1977, "Managerial Myopia: Self-Serving Biases in Organizational Planning," Journal of Applied Psychology 62, 94-198.

Liu, Y. and R. Taffler, 2008, "CEO Overconfidence in M&A Decision Making and Its Impact on Firm Performance," University of Edinburgh Working Paper.

Loughran, T. and J. Ritter, 2000, "Uniformly Least Powerful Tests of Market Efficiency," Journal of Financial Economics 55, 361-389.

Lyon, J.D., B.M. Barber, and C.-L. Tsai, 1999, "Improved Methods for Tests for Long-Run Abnormal Stock Returns," Journal of Finance 54, 165-201.

Malmendier, U. and G. Tate, 2005, "CEO Overconfidence and Corporate Investment," Journal of Finance 60, 2661-2700.

Malmendier, U. and G. Tate, 2008, "Who Makes Acquisitions? CEO Overconfidence and the Market's Reaction," Journal of Financial Economics 89, 20-43.

Malmendier, U., G. Tate, and J. Yan, 2011, "Overconfidence and Early-Life Experience: The Impact of Managerial Traits on Corporate Financial Policies," Journal of Finance 66, 1687-1733.

Merton, R., 1973, "Theory of Rational Option Pricing," Bell Journal of Economics and Management Science 4, 141-183.

Newey, W.K. and K.D. West, 1987, "A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix," Econometrica 55, 703-708.

Svenson, O., 1981, "Are We All Less Risky and More Skillful Than Our Fellow Drivers?" Acta Psychologica 47, 143-148.

Szewczyk, S., G. Tsetsekos, andZ. Zantout, 1996, "The Valuation of Corporate R&D Expenditures: Evidence from Investment Opportunities and Free Cash Flow," Financial Management 25, 105-110.

Weinstein, N., 1980, "Unrealistic Optimism about Future Life Events," Journal of Personality and Social Psychology 39, 806-820.

Yermack, D., 1995, "Do Corporations Award CEO Stock Options Effectively?" Journal of Financial Economics 39, 237-269.

Zantout, Z. and G. Tsetsekos, 1994, "The Wealth Effects of Announcements of R&D Expenditures Increases," Journal of Financial Research 17, 205-216.

(1) The market value of common equity, market capitalization, and size is identically defined. We use them interchangeably in this study.

(2) Malmendier and Tate (2008) show that the particular choice of parameter values does not affect their results. An alternative measure following Malmendier and Tate (2008), holder67, defines CEOs as overconfident if they fail to exercise the options within five years before expiration despite a 67% or more increase in the stock price from the grand date. Our empirical results are qualitatively similar for the use of these two overconfidence measures.

(3) We thank the anonymous reviewer for this comment.

(4) Our matching criteria based on size, BM, and momentum are also adopted by Gao (2010) and Carter et al. (2011).

(5) The approach of using factor models to measure abnormal returns is also adopted by Andre, Kooli, and LHer (2004), Chan et al. (2008), and Braga-Alves and Shastri (2011).

(6) We also compute the risk-adjusted returns by using 12- and 36-month periods. The empirical results are qualitatively similar to those based on a 60-month period.

(7) We thank Ken French for providing data on [R.sub.m], [R.sub.f], SMB, HML, and UMD. We obtain the data through the Wharton Research Data Services website.

(8) Our matching criteria are also adopted by Ikenberry and Ramnath (2002) and Eberhart et al. (2004).

(9) To avoid extreme values confounding our results, we compute the BHARs by trimming and winsorizing at the 5th and the 95th percentiles of our sample and find that the results are similar. Since some R&D increase firms appear multiple times in our samples, we also estimate the BHARs for nonoverlapping samples. The results remain qualitatively similar to those for the full sample.

(10) Accordingly, we follow the industry categorization method proposed by Brown et al. (2009) to separate our samples into high-tech and low-tech industries. The high-tech industries are classified as firms with three-digit SIC codes 283, 357, 366, 367, 383, 384, and 737. In Table III, there are 14 firm-year observations for high-tech firms with overconfident CEOs and 82 firm-year observations for high-tech firms with non-overconfident CEOs.

(11) We note that the coefficients on [[theta].sub.1] are positively significant for Models (1) and (3) suggesting that for low-tech firms, the firms with overconfident CEOs perform better than those with non-overconfident CEOs. However, the CTAR results for low-tech firms in Panel B of Table V do not support this finding.

(12) To ensure that the different performances of BHARs between firms with overconfident CEOs and those without following R&D increases are indeed due to the inefficiency of R&D investments rather than the inefficiency of other corporate policies of overconfident CEOs, we also examine the five-year BHARs for the sample of R&D decreases. Our findings demonstrate no significant difference in the long-run stock performance between firms with overconfident CEOs and those without.

(13) We also estimate the abnormal returns using purged factors, which eliminate the sample firms from the factor portfolio construction. The results are qualitatively similar.

(14) To ensure that the different performances of CTARs between firms with overconfident CEOs and those without following R&D increases are indeed due to the inefficiency of R&D investments rather than the inefficiency of other corporate policies of overconfident CEOs, we also examine the CTARs for the sample of R&D decreases. Our findings indicate no significant difference in abnormal stock returns between firms with overconfident CEOs and those without.

(15) To avoid extreme values confounding our results, we compute abnormal operating performance by trimming and winsorizing at the 5th and the 95th percentiles of our sample and find that the results are similar. Because some R&D increase firms appear multiple times in our samples, we also estimate the abnormal operating performance for nonoverlapping samples. The results remain qualitatively similar to those for the full sample.

(16) To ensure that the different of operating performance between firms with overconfident CEOs and those without following R&D increases is indeed due to the inefficiency of R&D investments rather than the inefficiency of other corporate policies of overconfident CEOs, we also examine the five-year operating performance for the sample of R&D decreases. Our findings show that there is no significant difference in long-run operating performance between firms with overconfident CEOs and those without.

We are grateful to Ulrike Malmendier for providing us with the CEO overconfidence data and William Christie, Marc Lipson (Editor), and especially an anonymous reviewer for constructive comments. The study has benefitted from comments from the conference participants at the 2010 NTU International Conference on Finance, the 6th International Conference on Asian Financial Markets, PKU/NTU Finance Conference, Macao International Symposium on Accounting and Finance, the 2012 EFA Meeting, the 2012 AsianFA Meeting, and the 2012 FMA Annual Meeting, as well as seminar participants at National Chengchi University, University of Reading, National Cheng Kung University, Chongqing University, and National Taiwan University.

Sheng-Syan Chen, Keng-Yu Ho, and Po-Hsin Ho *

* Sheng-Syan Chen is a Professor in the Department of Finance at National Taiwan University, Taipei, Taiwan. Keng-Yu Ho is an Associate Professor in the Department of Finance at National Taiwan University, Taipei, Taiwan. Po-Hsin Ho is an Assistant Professor in the Department of Finance, National United University, Miaoli, Taiwan.
Table I. Descriptive and Summary Statistics

This table presents the summary statistics of the sample and
matching firms. The sample period covers 1980-1994. In Panel A, the
sample firms, either with overconfident CEOs or with non-overconfident
CEOs, are those with economically significant
increases in R&D. The matching criteria in Panel B are based on
size-BM-CEO type-industry. The measure of overconfident CEOs,
postlongholder, follows Malmendier and Tate (2008). Postlongholder
is a binary variable that takes a value of one only after the year
that the CEO for the first time keeps an option grant that was at
least 40% in-the-money entering its last year, and zero otherwise.
The sales, total assets, market capitalization, and BM equity
variables, which are measured at the beginning of the year of the
increase in R&D by a sample firm, are adjusted by the consumer price
index to reflect 1994 dollars. See the Appendix for variable
definitions. The R&D intensity ratio is also measured at the
beginning of the year of the increase in R&D. Mests (Kruskal-Wallis
tests) are employed to test for the difference between the means
(medians) for the firms with overconfident CEOs and firms with
non-overconfident CEOs, while the p-values are reported in the last two
columns. The choice of using equal variance or nonequal variance in
the Mests is based on the variance test.

                           Non-overconfident CEOs
                            (Firm-Year, n = 121)

                      Mean       Median     Std. Dev.

Panel A. Sample Firms

Sales ($MM)           5,279       1,704      10,831
Total assets          4,492       1,476      8,217
  ($MM)
Market                4,959       2,732      6,894
  capitalization
  ($MM)
BM equity             0.42        0.39       0.23
R&D Intensity Measure (%)
  R&D/Assets          9.36        8.60       4.16
  R&D/Sales           9.60        7.85       12.03

Panel B. Matching Firms

Sales ($MM)         6,327.40    3,005.40    13,469.25
Total assets        5,760.37    2,409.93    10,636.74
  ($MM)
Market              4,952.44    2,511.18    6,242.55
  capitalization
  ($MM)
BM equity             0.53        0.44        0.29
R&D Intensity Measure (%)
  R&D/Assets          3.25        2.55        3.68
  R&D/Sales           2.88        1.90        3.53

                             Overconfident CEOs
                             (Firm-Year, n = 18)

                      Mean       Median     Std. Dev.

Panel A. Sample Firms

Sales ($MM)           4,758       3,692       3,972
Total assets          4,245       3,475       3,607
  ($MM)
Market                7,580       2,658       7,812
  capitalization
  ($MM)
BM equity             0.44        0.46        0.17
R&D Intensity Measure (%)
  R&D/Assets          8.96        7.48        3.34
  R&D/Sales           7.72        7.04        2.55

Panel B. Matching Firms

Sales ($MM)         8,610.18    6,907.35    7,927.53
Total assets        6,781.46    2,136.30    8,154.14
  ($MM)
Market              7,255.22    2,847.12    7,277.52
  capitalization
  ($MM)
BM equity             0.44        0.42        0.17
R&D Intensity Measure (%)
  R&D/Assets          2.94        3.32        2.70
  R&D/Sales           2.29        1.93        2.18

                          Difference

                      Mean       Median

Panel A. Sample Firms

Sales ($MM)         0.70        0.07 *
Total assets        0.83        0.13
  ($MM)
Market              0.14        0.18
  capitalization
  ($MM)
BM equity           0.68        0.78
R&D Intensity Measure (%)
  R&D/Assets        0.70        0.69
  R&D/Sales         0.01 **     0.35

Panel B. Matching Firms

Sales ($MM)         0.31        0.04 **
Total assets        0.70        0.30
  ($MM)
Market              0.10        0.17
  capitalization
  ($MM)
BM equity           0.07 *      0.31
R&D Intensity Measure (%)
  R&D/Assets        0.73        0.71
  R&D/Sales         0.34        0.95

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table II. Buy-and-Hold Abnormal Returns (BHARs) Following R&D
Increases

This table presents the one-, three-, and five-year BHARs following
increases in R&D among the sample firms. For the ith sample firm
from month [T.sub.1] to [T.sub.2], [BHAR.sub.i,T1,T2] is expressed
as:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where [R.sub.i,t] is the return of the sample firm in event month t,
and [R.sub.b,t] is the return of the benchmark over the same period.
The test statistics is

t = [square root of n] x [bar.BHAR]/[sigma](BHAR),

where [bar.B H A R] refers to the average across firms' BHARs,
[sigma] {BHAR) is the cross-sectional standard deviation of BHARs
for all sample firms, and n is the number of firms. For each event
firm, we compute corresponding one-, three-and five-year BHARs from
the year of the increase in R&D with a three-month lag, using a
benchmark control firm approach. We use postlongholder to measure
CEO overconfidence. In Panel A, we restrict all benchmark firms to
the same CEO type as sample firms in the year of significant R&D
increases. We also require that the matching firms having the same
two-digit SIC codes as sample firms. We identify all of the firms
with a market value of equity within 30% of the market value of
equity of the sample firm in the beginning year of the increase in
R&D. From this set of firms, we then select the firm that has the
closest BM equity ratio to the sample firm and compute the mean and
median BHARs of those firms with overconfident and non-
overconfident CEOs. Panel B adds momentum (defined as the previous
12-month returns) as an additional criterion for matched control
firms. Non-OC CEOs represent the non-overconfident CEO samples, and
OC CEOs represent the overconfident CEO samples. We use t-test
(Kruskal-Wallis test) to test for the mean (median) difference.

                     1-Year BHAR

                N        Mean          Median

Panel A. Size/BM/CEO Type/Industry Match

Non-OC CEOs    121    0.0440        -0.0047
OC CEOs        18    -0.0850        -0.0472
Difference            0.1291         0.0425

Panel B. Size/BM/Mom/CEO Type/Industry Match

Non-OC CEOs    120    0.0542        -0.0004
OC CEOs        18    -0.1189        -0.0472
Difference            0.1731         0.0468

                     3-Year BHAR

                N        Mean          Median

Panel A. Size/BM/CEO Type/Industry Match

Non-OC CEOs    121    0.2969 **      0.0888 *
OC CEOs        18    -0.4906 **     -0.4125 *
Difference            0.7875 **      0.5013 **

Panel B. Size/BM/Mom/CEO Type/Industry Match

Non-OC CEOs    120    0.3311 ***     0.1348 **
OC CEOs        18    -0.4817 *      -0.3019 *
Difference            0.8128 ***     0.4368 **

                     5-Year BHAR

                N        Mean          Median

Panel A. Size/BM/CEO Type/Industry Match

Non-OC CEOs    121    0.5566 **      0.0902
OC CEOs        18    -0.4387        -0.7017
Difference            0.9954 **      0.7919 **

Panel B. Size/BM/Mom/CEO Type/Industry Match

Non-OC CEOs    120    0.6693 ***     0.3253 **
OC CEOs        18    -0.6147        -0.6099
Difference            1.2840 **      0.9352 **

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table III. Cross-Sectional Regression of BHARs Following R&D
Increases

We estimate the following regression:

[BHAR.sub.it] = [[theta].sub.0] + [[theta].sub.1] +
[beta][x.sub.it-1] + [[epsilon].sub.it],

where [BHAR.sub.it] refers to the five-year BHARs calculated from
Table II for firm i in year t. [OC.sub.it] refers to the CEO
overconfidence indicator, postlongholder, for firm i in year t, and
[x.sub.it-1] is a vector of control variables from firm i in year t
- 1. The definitions of control variables (x) are described in the
Appendix. We also measure the following equation by adding a
high-tech indicator as the explanatory variable alone with its
interaction term with the CEO overconfidence indicator:

[BHAR.sub.it] = [[theta].sub.0] + [[theta].sub.1] O[C.sub.it] +
[[theta].sub.2]H[T.sub.it] + [[theta].sub.3] O[C.sub.it] x
H[T.sub.it] + [beta][x.sub.it-1] + [[epsilon].sub.it],

where [BHAR.sub.it] refers to the five-year BHARs for firm i in year
t; O[C.sub.it] refers to the CEO overconfidence indicator,
postlongholder, for firm i in year t; H[T.sub.it] refers to the
high-tech industries indicator; and [x.sub.it-1] is a vector of
control variables of firm i in year t-1. Firms are classified into
high-tech and low-tech firms using the definition proposed in Brown
et al. (2009). The year effects and the SIC 2-digit effects are
controlled in addition to the matching done on year and industry.
The t-statistics reported in the parentheses are based on clustered
and heteroskedasticity consistent standard errors. The penultimate
row presents the results of the Wald tests for [[theta].sub.1] +
[[theta].sub.3] = 0, and p-values are reported in the parentheses.

                                  Size/BM/CEO Type/
                                    Industry Match

                                 (1)            (2)

Intercept                     -0.275         -2.461
                             (-0.12)        (-1.00)
Postlongholder                -0.536 **       3.888 ***
                             (-2.70)         (3.83)

High-Tech                                     2.107 ***
                                             (3.03)
Postlongholder x High-Tech                   -4.613 ***
                                            (-4.30)
Size                          -2.336         -4.524
                             (-0.56)        (-1.44)
BM                             1.652          1.790
                              (0.79)         (0.92)
Tobin q                       -0.094         -0.102
                             (-0.81)        (-0.91)
R&D/Sales                      0.537         -2.248
                             (-0.29)        (-1.43)
ROA                            6.307          6.727
                              (1.36)         (1.47)
Sales growth                  -0.740         -0.801
                             (-0.59)        (-0.67)
Cash                           2.595          3.179 *
                              (1.17)         (2.00)
PPE/Emp                       -0.008         -0.002
                             (-1.44)        (-0.41)
Stock ownership               18.805 ***     18.027 ***
                              (6.10)         (6.25)
Vested options                -4.128         -6.969
                             (-0.23)        (-0.41)
[(Vested options).sup.2]       3.694          5.830
                              (0.21)         (0.34)
Year effects                 Yes            Yes
SIC 2-digit effects          Yes            Yes
Obs                          125            125
Adjusted [R.sup.2]             0.251          0.300
Wald tests:                                  -0.725 ***
  [[theta].sub.1] +
  [[theta].sub.3] = 0
p-value                                      (0.002)

                                Size/BM/Mom/CEO Type/
                                   Industry Match

                                 (3)            (4)

Intercept                      0.592         -1.374
                              (0.31)        (-0.67)
Postlongholder                -0.844 **       3.256 ***
                             (-2.78)         (3.24)
High-Tech                                     1.897 **
                                             (2.85)
Postlongholder x High-Tech                   -4.270 ***
                                            (-4.64)
Size                          -3.469         -5.438
                             (-0.67)        (-1.23)
BM                             1.193          1.317
                              (0.61)         (0.72)
Tobin q                        0.070          0.063
                              (0.40)         (0.37)
R&D/Sales                     -1.056         -2.596
                             (-0.54)        (-1.33)
ROA                            4.059          4.437
                              (0.94)         (1.06)
Sales growth                  -0.936         -0.991
                             (-0.80)        (-0.87)
Cash                           1.567          2.092
                              (0.55)         (0.82)
PPE/Emp                       -0.009         -0.004
                             (-1.47)        (-0.65)
Stock ownership               14.988 ***     14.288 ***
                              (3.20)         (3.13)
Vested options                -4.602         -7.158
                              (0.31)        (-0.49)
[(Vested options).sup.2]       5.752          7.674
                              (0.38)         (0.53)
Year effects                 Yes            Yes
SIC 2-digit effects          Yes            Yes
Obs                          125            125
Adjusted [R.sup.2]             0.172          0.213
Wald tests:                                  -1.014 **
  [[theta].sub.1] +
  [[theta].sub.3] = 0
p-value                                      (0.020)

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table IV. Long-Term Abnormal Returns for Calendar-Time Portfolio

This table presents the abnormal stock returns for the sample firms
with increases in R&D from 1980 to 1994. Panel A reports the results
for the full sample, while Panel B provides the results of the
rolling regression method. We use the Fama-French (1993) three-
factor model to estimate the abnormal returns based upon the
following equation:

[R.sub.pt] -[R.sub.ft] = [[alpha].sub.p] + [[beta].sub.p]
([R.sub.mt] -[R.sub.ft]) + [s.sub.p][SMB.sub.t] +
[h.sub.p][HML.sub.t] + [[epsilon].sub.pt],

where [R.sub.pt] is the monthly return on a equal-or value-weighted
portfolio in calendar month t (where a sample stock is included if
month t is within the 60-month period following its R&D increase);
[R.sub.ft] is the one-month treasury bill return; [R.sub.mt] is the
CRSP value-weighted market index return; SMB, is the difference
between the returns on the value-weighted portfolios for small and
large stocks; and HML, is the difference between the returns on the
value-weighted portfolios for high and low BM stocks. We also use
the Carhart (1997) four-factor model to measure the monthly abnormal
returns following an increase in R&D by estimating the intercept
from the following equation:

[R.sub.pt] -[R.sub.ft] = [[alpha].sub.p] + [[beta].sub.p]([R.sub.mt]
-[R.sub.ft]) + [S.sub.p][SMB.sub.t] + [h.sub.p][HML.sub.t] +
[m.sub.p][UMD.sub.t] + [[epsilon].sub.pt],

where [UMD.sub.t] is the difference between the returns on a value-
weighted portfolio of high-and low-momentum stocks. In Panel B, we
use the first 60 months to estimate the factor loadings for
Equations (3) and (4), and then estimate the abnormal return in
month 61 as the difference between the actual and expected portfolio
returns. The expected portfolio return is defined by the factor
loadings estimated over the previous 60 months multiplied by their
respective month 61 factor returns. We replicate this step for every
month and then average the time series of these abnormal returns and
factor loading estimates. The measure of CEO overconfidence is
postlongholder, detailed in the Appendix. Non-OC CEOs represent the
non-overconfident CEO samples, while OC CEOs represent the
overconfident CEO samples. Newey-West (1987) t-statistics, reported
in parentheses, are adjusted for serial correlation and
heteroskedasticity.

                  Fama-French Three-Factor
                            Model

                   [alpha]         [beta]

Panel A. Full Sample

Equal weight
  Non-OC CEOs     0.0074 ***       1.1446 ***
                 (2.75)          (17.65)
  OC CEOs         0.0010           1.0503 ***
                 (0.31)          (14.51)
  Difference      0.0064 **        0.0943
                 (2.08)           (1.35)

Value weight
  Non-OC CEOs     0.0063 ***       1.0387 ***
                 (3.52)          (22.22)
  OC CEOs        -0.0009           0.8887 ***
                (-0.39)          (12.39)
  Difference      0.0072 ***       0.1501 **
                 (2.93)           (2.11)

Panel B. Rolling Regression Method

Equal weight
  Non-OC CEOs     0.0081           1.1236 ***
                 (1.48)         (132.66)
  OC CEOs        -0.0006           1.0233 ***
                (-0.07)         (148.33)
  Difference      0.0087           0.1003 ***
                 (0.92)           (9.19)
Value weight
  Non-OC CEOs     0.0073 **        1.0234 ***
                 (2.20)         (242.07)
  OC CEOs        -0.0037           0.8739 ***
                (-0.75)         (135.30)
  Difference      0.0111 ***       0.1495 ***
                 (1.85)          (19.36)

                   Fama-French Three-Factor
                           Model

                      s               h

Panel A. Full Sample

Equal weight
  Non-OC CEOs      0.3230 **      -0.5032 ***
                  (2.31)         (-3.26)
  OC CEOs          0.1610         -0.1217
                  (1.41)         (-0.59)
  Difference       0.1621         -0.3815 **
                  (1.31)         (-2.07)

Value weight
  Non-OC CEOs     -0.2966 ***     -0.6065 ***
                 (-3.65)         (-5.13)
  OC CEOs         -0.3340 ***     -0.6802 ***
                 (-3.03)         (-4.69)
  Difference       0.0374          0.0737
                  (0.30)          (0.46)

Panel B. Rolling Regression Method

Equal weight
  Non-OC CEOs      0.2817 ***     -0.6336 ***
                 (11.32)        (-14.41)
  OC CEOs          0.1395 ***     -0.2398 ***
                 (17.30)         (-5.45)
  Difference       0.1421 ***     -0.3938 ***
                  (5.43)         (-6.33)
Value weight
  Non-OC CEOs     -0.3462 ***     -0.7158 ***
                (-24.66)        (-19.85)
  OC CEOs         -0.3107 ***     -0.6924 ***
                (-47.42)        (-48.83)
  Difference      -0.0355 ***     -0.0234
                 (-2.29)         (-0.60)

                            Carhart Four-Factor Model

                   [alpha]         [beta]             s

Panel A. Full Sample

Equal weight
  Non-OC CEOs     0.0090 ***       1.1668 ***      0.2728 *
                 (3.39)          (19.54)          (1.92)
  OC CEOs         0.0023           1.0676 ***      0.1218
                 (0.69)          (14.10)          (1.01)
  Difference      0.0067 **        0.0992          0.1510
                 (2.24)           (1.46)          (1.12)

Value weight
  Non-OC CEOs     0.0056 ***       1.0292 ***     -0.2750 ***
                 (2.78)          (21.22)         (-3.28)
  OC CEOs        -0.0018           0.8763 ***     -0.3060 ***
                (-0.74)          (13.50)         (-2.62)
  Difference      0.0074 ***       0.1529 **       0.0310
                 (2.87)           (2.23)          (0.23)

Panel B. Rolling Regression Method

Equal weight
  Non-OC CEOs     0.0124 **        1.1317 ***      0.1850 ***
                 (2.18)         (165.32)          (7.03)
  OC CEOs         0.0027           1.0275 ***      0.0708 ***
                 (0.35)         (169.49)          (7.17)
  Difference      0.0097           0.1043 ***      0.1141 ***
                 (1.01)          (11.40)          (4.06)
Value weight
  Non-OC CEOs     0.0065 *         1.0219 ***     -0.3231 ***
                 (1.96)         (230.17)        (-22.13)
  OC CEOs        -0.0048           0.8687 ***     -0.2687 ***
                (-0.96)         (138.75)        (-27.03)
  Difference      0.0114 *         0.1532 ***     -0.0544 ***
                 (1.88)          (19.96)         (-3.08)

                  Carhart Four-Factor Model

                      h               m

Panel A. Full Sample

Equal weight
  Non-OC CEOs     -0.5394 ***     -0.1926
                 (-3.05)         (-1.66)
  OC CEOs         -0.1499         -0.1501
                 (-0.71)         (-1.44)
  Difference      -0.3895 **      -0.0425
                 (-2.01)         (-0.39)

Value weight
  Non-OC CEOs     -0.5909 ***      0.0829
                 (-5.30)          (1.09)
  OC CEOs         -0.6600 ***      0.1072
                 (-4.52)          (1.26)
  Difference       0.0691         -0.0243
                  (0.41)         (-0.26)

Panel B. Rolling Regression Method

Equal weight
  Non-OC CEOs     -0.7639 ***     -0.2729 ***
                (-15.05)        (-31.39)
  OC CEOs         -0.3223 ***     -0.1872 ***
                 (-6.62)        (-18.89)
  Difference      -0.4416 ***     -0.0857 ***
                 (-6.28)         (-6.50)
Value weight
  Non-OC CEOs     -0.6828 ***      0.0675 ***
                (-18.87)         (12.34)
  OC CEOs         -0.6227 ***      0.1265 ***
                (-49.71)          (8.87)
  Difference      -0.0601         -0.0590 ***
                 (-1.57)         (-3.87)

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table V. Subsample Tests of Long-Term Abnormal Returns for
Calendar-Time Portfolio

This table provides details of the abnormal stock returns for the
sample of firms increasing their R&D from 1980 to 1994. The full
sample is divided into high-and low-tech firms using the definition
proposed by Brown et al. (2009), where a refers to the abnormal
return measure. A detailed description of the test procedures is
provided in Table IV. Newey-West (1987) t-statistics, reported in
the parentheses, are adjusted for serial correlation and
heteroskedasticity.

                     Fama-French Three-
                        Factor Model

                  [alpha]         [beta]

Panel A. High-Tech Sample

Equal weight
  Non-OC CEOs     0.0101 ***     1.1071 ***
                 (3.71)        (13.36)
  OC CEOs         0.0024         1.0216 ***
                 (0.64)        (11.11)
  Difference      0.0077 **      0.0855
                 (2.19)         (0.96)
Value weight
  Non-OC CEOs     0.0090 ***     0.9980 ***
                 (4.74)        (17.11)
  OC CEOs         0.0007         0.8525 ***
                 (0.28)        (12.76)
  Difference      0.0084 ***     0.1454 *
                 (3.58)         (1.78)

Panel B. Low-Tech Sample

Equal weight
  Non-OC CEOs    -0.0025         1.2057 ***
                (-0.98)        (14.69)
  OC CEOs        -0.0008         0.8861 ***
                (-0.18)         (9.48)
  Difference     -0.0017         0.3196 ***
                (-0.37)         (3.24)
Value weight
  Non-OC CEOs    -0.0032         1.0812 ***
                (-1.09)        (10.75)
  OC CEOs        -0.0012         0.8793 ***
                (-0.29)         (8.53)
  Difference     -0.0020         0.2019 *
                (-0.39)         (1.92)

                     Fama-French Three-
                        Factor Model

                     s               h

Panel A. High-Tech Sample

Equal weight
  Non-OC CEOs     0.3718 ***      -0.6735 ***
                 (2.11)          (-3.41)
  OC CEOs         0.1288          -0.3986
                 (0.82)          (-1.64)
  Difference      0.2430          -0.2749
                 (1.46)           (1.18)
Value weight
  Non-OC CEOs    -0.2977 **       -0.9598 ***
                (-2.55)          (-7.54)
  OC CEOs        -0.3469 ***      -0.8551 ***
                (-2.88)          (-6.22)
  Difference      0.0492          -0.1047
                 (0.32)          (-0.73)

Panel B. Low-Tech Sample

Equal weight
  Non-OC CEOs     0.0817           0.0488
                 (0.74)           (0.38)
  OC CEOs         0.2427 *         0.0887
                 (1.67)           (0.41)
  Difference     -0.1610          -0.0399
                (-0.96)          (-0.18)
Value weight
  Non-OC CEOs    -0.2054           0.1268
                (-1.61)           (0.83)
  OC CEOs         0.1665           0.1050
                 (1.08)           (0.51)
  Difference     -0.3718 **        0.0217
                (-2.22)           (0.10)

                         Carhart Four-Factor Model

                  [alpha]         [beta]           s

Panel A. High-Tech Sample

Equal weight
  Non-OC CEOs     0.0117 ***     1.1374 ***     0.3219 *
                 (4.33)        (15.06)         (1.79)
  OC CEOs         0.0047         1.0636 ***     0.0597
                 (1.28)        (11.56)         (0.40)
  Difference      0.0070 **      0.0738         0.2622
                 (2.10)         (0.82)         (1.50)
Value weight
  Non-OC CEOs     0.0095 ***     1.0064 ***    -0.3116 ***
                (-4.64)        (16.51)        (-2.64)
  OC CEOs         0.0008         0.8537 ***    -0.3488 ***
                 (0.30)        (12.72)        (-2.76)
  Difference      0.0087 ***     0.1527 *       0.0372
                 (3.52)         (1.97)         (0.24)

Panel B. Low-Tech Sample

Equal weight
  Non-OC CEOs    -0.0017         1.2199 ***     0.0524
                (-0.64)        (13.88)         (0.42)
  OC CEOs        -0.0020         0.8647 ***     0.2869 *
                (-0.52)         (9.49)         (1.96)
  Difference      0.0003         0.3552 ***    -0.2345
                 (0.07)         (3.37)        (-1.41)
Value weight
  Non-OC CEOs    -0.0045         1.0581 ***    -0.1576
                (-1.50)        (11.68)        (-1.27)
  OC CEOs        -0.0027         0.8540 ***     0.2186
                (-0.67)         (8.70)         (1.41)
  Difference     -0.0019         0.2040 *      -0.3762 **
                (-0.38)         (1.83)        (-2.24)

                 Carhart Four-Factor Model

                     h              m

Panel A. High-Tech Sample

Equal weight
  Non-OC CEOs    -0.7040 ***    -0.2292 *
                (-3.26)        (-1.91)
  OC CEOs        -0.4409 *      -0.3177 ***
                (-1.78)        (-2.86)
  Difference     -0.2632         0.0884
                (-1.11)         (0.72)
Value weight
  Non-OC CEOs    -0.9683 ***    -0.0637
                (-7.43)        (-0.88)
  OC CEOs        -0.8562 ***    -0.0087
                (-6.21)        (-0.13)
  Difference     -0.1120        -0.0549
                (-0.75)        (-0.63)

Panel B. Low-Tech Sample

Equal weight
  Non-OC CEOs     0.0354        -0.1184
                 (0.27)        (-0.99)
  OC CEOs         0.1088         0.1789
                 (0.54)         (1.19)
  Difference     -0.0734        -0.2973 *
                (-0.35)        (-1.98)
Value weight
  Non-OC CEOs     0.1486         0.1933
                 (1.05)         (1.32)
  OC CEOs         0.1288         0.2109
                 (0.66)         (1.41)
  Difference      0.0197        -0.0175
                 (0.09)        (-0.11)

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table VI. Long-Term Abnormal Operating Performance Following R&D
Increases

In this table, we compute the changes in abnormal operating
performance for the sample firms for the five years after the year
in which there was a significant increase in R&D. The abnormal
operating performance is measured as the raw operating performance
of the sample firm minus the operating performance of its matching
firm. See the Appendix for definitions of the operating performance
measures (OPMs). We use postlongholder to measure CEO
overconfidence. We also require matching firms that have the same
CEO type as our sample firms. Matching firms are selected that do
not have the same corporate event as the sample firm in the event
year. We select a group of control firms, which have had no
significant increase in R&D during the same period as the sample
firms, from the same two-digit SIC code as the sample firm. From
these screened firms, we then select a firm, as the matching firm,
with the closest OPM to that of the sample firm in the year prior to
the year of the increase in R&D for the sample firm. We also create
an additional group of matching firms based on firm characteristics
including firm size, BM equity ratio, and momentum. At the beginning
of the year of the sample firm's increase in R&D, we select a
matching firm, the market equity value of which is within 30% of the
market equity value of the sample firm, and then choose the joint
lowest absolute value of the difference in the characteristics. Non-
OC CEOs represent the non-overconfident CEO samples, while OC CEOs
represent the overconfident CEO samples, t-tests (Kruskal-Wallis
tests) are used to test the mean (median) difference.

                        CEO Type/
                   Industry/Performance Match

Year            N     Mean          Median

Panel A. Changes in OPM1

-1 to +1
  Non-OC CEOs   123    0.0018        0.0039
  OC CEOs       22    -0.0186       -0.0135
  Difference           0.0204        0.0174
-1 to +2
  Non-OC CEOs   121    0.0016        0.0037
  OC CEOs       22    -0.0252       -0.0227
  Difference           0.0267        0.0264
-1 to +3
  Non-OC CEOs   121    0.0112        0.0083
  OC CEOs       22    -0.0424 **    -0.0292 *
  Difference           0.0537 **     0.0374 **
-1 to +4
  Non-OC CEOs   119    0.0161        0.0146 *
  OC CEOs       23    -0.0550 ***   -0.0323 **
  Difference           0.0711 ***    0.0469 ***
-1 to +5
  Non-OC CEOs   118    0.0133        0.0166 *
  OC CEOs       23    -0.0574 **    -0.0452 **
  Difference           0.0706 **     0.0618 ***

Panel B. Changes in OPM2

-1 to +1
  Non-OC CEOs   123    0.0002        0.0099
  OC CEOs       22    -0.0219       -0.0157
  Difference           0.0220        0.0257
-1 to +2
  Non-OC CEOs   121    0.0008        0.0190
  OC CEOs       22    -0.0264       -0.0075
  Difference           0.0272        0.0265
-1 to +3
  Non-OC CEOs   121    0.0187        0.0169 **
  OC CEOs       22    -0.0460 *     -0.0247
  Difference           0.0647 *      0.0416 **
-1 to +4
  Non-OC CEOs   119    0.0301 **     0.0259 *
  OC CEOs       23    -0.0715 ***   -0.0525 **
  Difference           0.1016 ***    0.0784 ***
-1 to +5
  Non-OC CEOs   118    0.0279 **     0.0351 ***
  OC CEOs       23    -0.0680 **    -0.0505 **
  Difference           0.0959 ***    0.0855 ***

                         CEO Type/
                    Characteristics Match

Year            N     Mean          Median

Panel A. Changes in OPM1

-1 to +1
  Non-OC CEOs   123   -0.0202 **    -0.0121 ***
  OC CEOs       26    -0.0216 *     -0.0052
  Difference           0.0014       -0.0069
-1 to +2
  Non-OC CEOs   121   -0.0108       -0.0058
  OC CEOs       26    -0.0295 *     -0.0055
  Difference           0.0186       -0.0002
-1 to +3
  Non-OC CEOs   121   -0.0016       -0.0063
  OC CEOs       26    -0.0400 **    -0.0161
  Difference           0.0384'       0.0098
-1 to +4
  Non-OC CEOs   119    0.0054        0.0073
  OC CEOs       27    -0.0515 **    -0.0075
  Difference           0.0569 **     0.0148
-1 to +5
  Non-OC CEOs   118    0.0090        0.0214
  OC CEOs       27    -0.0592 **    -0.0079
  Difference           0.0682 **     0.0293 **

Panel B. Changes in OPM2

-1 to +1
  Non-OC CEOs   123   -0.0200       -0.0055
  OC CEOs       26    -0.0211        0.0025
  Difference           0.0011       -0.0080
-1 to +2
  Non-OC CEOs   121   -0.0182        0.0020
  OC CEOs       26    -0.0283        0.0047
  Difference           0.0101       -0.0026
-1 to +3
  Non-OC CEOs   121   -0.0045        0.0031
  OC CEOs       26    -0.0384 *      0.0012
  Difference           0.0340        0.0019
-1 to +4
  Non-OC CEOs   119    0.0133        0.0136
  OC CEOs       27    -0.0580 **    -0.0005
  Difference           0.0713 **     0.0141 **
-1 to+5
  Non-OC CEOs   118    0.0123        0.0213
  OC CEOs       27    -0.0595 *      0.0020
  Difference           0.0718 **     0.0193 *

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table VII. Cross-Sectional Regression of Long-Term Abnormal Operating
Performance Following R&D Increases

We estimate the following two equations. The second equation
includes a high-tech indicator as the explanatory variable, along
with its interaction term with the CEO overconfidence indicator
(postlongholder):

[OPM.sub.it] = [[theta].sub.0] + [[theta].sub.1]O[C.sub.it] +
[beta][x.sub.it-1] + [[epsilon].sub.it],

[OPM.sub.it] = [[theta].sub.0] + [[theta].sub.1]O[C.sub.it] +
[[theta].sub.2][HT.sub.it] + [[theta].sub.3][C.sub.it] x [HT.sub.it]
+ [beta][x.sub.it-1] + [[epsilon].sub.it],

where [OC.sub.it] refers to the CEO overconfidence indicator for
firm i in year t; [OPM.sub.it] refers to the five-year abnormal
operating performance reported in Table VI for firm i in year t; and
[x.sub.it-1] is a vector of control variables for firm i in year t-1.
The definitions of the control variables (x) are provided in the
Appendix. Firms are classified into high-tech and low-tech firms
using the definition proposed in Brown et al. (2009). The year
effects and the SIC 2-digit effects are controlled in addition to
the matching done on year and industry. The t-statistics reported in
the parentheses are based on clustered and heteroskedasticity
consistent standard errors. The penultimate row presents the results
of the Wald tests for [[theta].sub.1] + [[theta].sub.3] = 0, and p-
values are reported in the parentheses.

                                      CEO Type/
                                 Industry/Performance
                                       Match

                                         OPM1

                                 (1)            (2)

Intercept                      0.094          0.060
                              (1.34)         (0.74)
Overconfident                 -0.046 *        0.151 ***
                             (-1.91)         (3.12)
High-Tech                                     0.039
                                             (1.69)
Postlongholder x High-Tech                   -0.212 ***
                                            (-4.16)
Tobin q                       -0.001         -0.002
                             (-0.12)        (-0.20)
R&D/Sales                     -0.095         -0.122
                             (-0.77)        (-0.83)
ROA                           -0.087         -0.071
                             (-0.34)        (-0.27)
Sales growth                  -0.011         -0.011
                             (-0.59)        (-0.62)
Cash                           0.129 *        0.143 **
                              (1.74)         (2.12)
PPE/Emp                       -0.000         -0.000
                             (-1.37)        (-0.86)
Stock ownership                0.525 ***      0.498 ***
                              (8.26)         (8.44)
Vested options                -0.628         -0.650 *
                             (-1.69)        (-1.91)
[(Vested options).sup.2]       0.479          0.485
                              (1.26)         (1.38)
Year effects                 Yes            Yes
SIC 2-digit effects          Yes            Yes
Obs                          126            126
Adjusted [R.sup.2]             0.306          0.324
Wald tests:
  [[theta].sub.1] +                          -0.061 ***
  [[theta].sub.3] = 0
p-value                                      (0.002)

                                      CEO Type/
                                 Industry/Performance
                                       Match

                                         OPM2

                                 (3)            (4)

Intercept                      0.153 **       0.065
                              (2.85)         (1.16)
Overconfident                 -0.067 **       0.051
                             (-2.24)         (0.80)
High-Tech                                     0.104 ***
                                             (3.85)
Postlongholder x High-Tech                   -0.137 **
                                            (-2.35)
Tobin q                        0.021 ***      0.020 ***
                              (3.58)         (3.54)
R&D/Sales                     -0.655 ***     -0.735 ***
                             (-5.49)        (-7.58)
ROA                           -0.770 ***     -0.781 ***
                             (-3.20)        (-3.19)
Sales growth                  -0.001         -0.002
                             (-0.02)        (-0.07)
Cash                           0.300 **       0.328 **
                              (2.21)         (2.76)
PPE/Emp                        0.000          0.000
                              (0.12)         (0.90)
Stock ownership                0.504 ***      0.468 ***
                              (5.41)         (5.04)
Vested options                -1.145 ***     -1.288 ***
                             (-3.48)        (-4.90)
[(Vested options).sup.2]      -1.162 ***      1.272 ***
                              (3.69)         (5.04)
Year effects                 Yes            Yes
SIC 2-digit effects          Yes            Yes
Obs                          126            126
Adjusted [R.sup.2]             0.379          0.411
Wald tests:
  [[theta].sub.1] +                          -0.086 ***
  [[theta].sub.3] = 0
p-value                                      (0.001)

                                      CEO Type/
                                Characteristics Match

                                         OPM1

                                 (5)            (6)

Intercept                      0.081          0.018
                              (1.35)         (0.29)
Overconfident                 -0.062 *        0.039
                             (-1.97)         (1.10)
High-Tech                                     0.076 **
                                             (2.38)
Postlongholder x High-Tech                   -0.118 ***
                                            (-4.27)
Tobin q                       -0.010         -0.010 *
                             (-1.72)        (-1.95)
R&D/Sales                      0.143          0.089
                             (-0.85)         (0.44)
ROA                           -0.046         -0.044
                             (-0.13)        (-0.12)
Sales growth                  -0.006         -0.008
                             (-0.23)        (-0.36)
Cash                           0.176          0.198 **
                              (1.62)         (2.12)
PPE/Emp                       -0.000         -0.000
                             (-0.68)        (-0.23)
Stock ownership                0.254 ***      0.221 ***
                              (2.11)         (1.82)
Vested options                -0.104         -0.219
                             (-0.25)        (-0.56)
[(Vested options).sup.2]      -0.151         -0.059
                             (-0.38)        (-0.15)
Year effects                 Yes            Yes
SIC 2-digit effects          Yes            Yes
Obs                          130            130
Adjusted [R.sup.2]             0.140          0.159
Wald tests:
  [[theta].sub.1] +                          -0.079 ***
  [[theta].sub.3] = 0
p-value                                      (0.003)

                                      CEO Type/
                                Characteristics Match

                                         OPM2

                                 (7)            (8)

Intercept                      0.174 ***      0.053
                              (4.06)         (1.15)
Overconfident                 -0.048          0.052
                             (-1.67)         (1.35)
High-Tech                                     0.144 ***
                                             (5.01)
Postlongholder x High-Tech                   -0.125 ***
                                            (-4.44)
Tobin q                        0.020 ***      0.019 ***
                              (3.59)         (3.29)
R&D/Sales                     -0.720 ***     -0.828 ***
                             (-4.54)        (-6.67)
ROA                           -0.910 ***     -0.926 ***
                             (-4.33)        (-4.52)
Sales growth                   0.037          0.033
                              (1.51)         (1.60)
Cash                           0.375 **       0.413 **
                              (2.14)         (2.79)
PPE/Emp                        0.000          0.000 *
                              (0.70)         (1.88)
Stock ownership                0.263 ***      0.213
                              (1.89)         (1.66)
Vested options                -0.438         -0.662
                             (-0.92)        (-1.72)
[(Vested options).sup.2]       0.310          0.492
                              (0.67)         (1.32)
Year effects                 Yes            Yes
SIC 2-digit effects          Yes            Yes
Obs                          130            130
Adjusted [R.sup.2]             0.263          0.323
Wald tests:
  [[theta].sub.1] +                          -0.073 ***
  [[theta].sub.3] = 0
p-value                                      (0.001)

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.
COPYRIGHT 2014 Financial Management Association
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2014 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:chief executive officer
Author:Chen, Sheng-Syan; Ho, Keng-Yu; Ho, Po-Hsin
Publication:Financial Management
Article Type:Report
Geographic Code:1USA
Date:Jun 22, 2014
Words:14309
Previous Article:Determinants and effects of corporate lobbying.
Next Article:The effect of executive stock options on corporate innovative activities.
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters