Master Thesis – Organization Economics
Does There Exist a Relationship Between the Level of Overconfidence
of the CEO and the Firm Value? Investigation of Chinese Innovative
Industries
Supervisor: Silvia Dominguez Martinez
Student: Wu Bi
Student Number: 10828648
Date: 2015-07-25
1
Abstract
This paper investigates the relationship of the CEO’s confidence level and firm value in Chinese innovation industry during the period 2010 to 2014. We employee alternative proxies for CEO’s confidence based on relative compensation and forecast bias. Our result shows that CEO’s confidence level positively affects R&D expenditure of the firm. We also find a quadratic relationship between CEO’s confidence level and firm value under compensation-based measure. But we find a negative linear relationship under forecast-based measure. Furthermore, we find that overconfident CEO affect firm value by R&D expenditure only under forecast-based measure. Since the result inconsistence between two measure, we should interpret the result with cautious.
Keywords
2
Catalogue
1. Introduction ... 1
2. Related literature and hypothesis ... 6
2.1 Overconfidence ... 6
2.2 Innovation investment ... 8
2.3 CEO's overconfidence and innovation investment ... 9
2.4 Innovation investment and firm value ... 11
2.5 CEO's overconfidence and firm value ... 13
3. Methodology ... 15 3.1 The data ... 15 3.2 Empirical model ... 16 4. Results ... 19 4.1 Descriptive statistics ... 19 4.2 Correlation Analysis ... 22 4.3 Linear regression ... 25 5. Discussion ... 31 6. Conclusion ... 35 7. Appendix ... 37 8. References ... 43
1
1. Introduction
According to 2014 Global R&D Report from the R&D magazine global, China’s R&D investment is about 61% that of the U.S. in 2014 and is expected to exceed that of U.S. by 2022. In last 20 years, China’s R&D spending has increased in a rate from 12% to 20% annually. The strong growth of China’s R&D spending not only has a link with economic growth but also to be very likely influenced by Chinese leadership: eight of nine members of China’s Standing Committee of the Political Bureau have engineering degrees. According to the investigation of Chinese National Bureau of Statistics in 2013, China’s R&D spending reaches 1184.66 billions RMB in that year. This corresponds to an increase of 15% as compared to 2012. There are 7 industries have more than 50 billions RMB R&D investment and account for 61% of total R&D investment in China.
CEOs of Chinese companies tend to be more confident about future when compared to US, Japan and Germany. The PwC 17th Annual Global CEO Survey in 2014 shows that 47 percent of Chinese CEOs are “very confident” about their companies’ prospects compares to 39 percent of global average. This advantage in confidence exists from 2010. The survey also shows that innovation is an important theme among Chinese CEOs. 52 percent of CEO in China seeks the growth in product and service innovation compare to 35 percent of global average. When Chinese CEOs are asked the trend to transform business in five years forwards, more than 80 percent of them mentioned technical advances. Considering for example, Yang Yuanqing, the CEO of Lenovo Group, which overtook HP as the largest PC company in the world in 2013. According an article in The Wall Street Journal1, Yang is “confident about the long-term outlook for its business in China despite the slowdown in the country’s economic growth.” Despite the slowdown in growth, Yang invested aggressively in the mobile device business and acquired Motorola in 2014 from Google. Again, Yang showed his confident and claims that “I am confident we will be successful with this process,
2 and that our companies will not only maintain our current momentum in the market, but also build a strong foundation for the future.”2
This thesis will examine the effect of overconfidence of CEOs in Chinese listed companies. The following research questions will be investigated: First, does there exist a relationship between the level of overconfident of the CEO and his investment in innovation in Chinese innovation industry? Second, how does the level of CEO’s overconfidence affect firm value?
Behavioral economists have challenged the traditional rational hypothesis and discussed the behavior of boundedly rational decision-makers. People cannot behave fully rational since they have psychological biases, such as framing effect, self-serving bias and status-quo bias. These psychological biases have different effects on human’s perception and decision-making. In recent studies, researchers have investigated the effect of manager’s psychological biases on corporate decision-making. Graham, Harvey and Puri (2013) apply a psychometric test to US CEOs to determine how their personal traits, such as optimism and risk tolerance, are related to their financial decisions and their compensation. In their study, they find that CEO who is risk loving is more likely to make acquisitions. Moreover, optimistic CEOs tend to have more debt, especially short-term debt.
The focus of this paper will be on overconfidence. Mahajan (1992) defines overconfidence as “an overestimation of the probabilities for a set of events” (p.3). Kruger and Dunning(1999) relate overconfidence to how individuals rate their own abilities. They show that individuals tend to perceive themselves as “better than average”. Previous studies have revealed that overconfidence is a prevalent and persistent human character trait. The effect of overconfidence is also found to be a prevalent trait on CEOs. Cooper, Woo and Dunkelberg (1988) document that more than half of the researched entrepreneurs think their business has higher chance of success, given the same kind of business. These psychological biases might
2 See “Google sells Motorola Mobility to Lenovo for US$2.9bn” The Siliconrepublic January, 30,
3 have a large effect on firm value. Malmendier and Tate (2008) argue that overconfident CEOs underestimate the risks and undertake more acquisitions than CEOs in average, which destroys firm value.
This paper differs from most prior research, because it mainly focuses on innovative investments. Innovation adoptions are found in the study of Bantel and Jackson (1989) to be influenced by top managers’ personal characteristics. CEO, as corporate organizer, can influence workers’ morale and job satisfaction, improve working climate, which motivate and reward creativity in work (Elenkov, Judge, and Wright 2005). The study of Martin and Terblanche (2003) find that firm environment such as strategy, organization structure, support mechanism and behavior influence creativity and innovation.
The sample of this paper is constrained to the innovation industry. Innovation industry has high intensity of R&D expenditure. The reason to study this industry is that Hirshleifer, Low and Teoh (2012) investigate U.S. company and find the effect of overconfident managers on innovation is strong only in innovation industry since innovation is important among those companies. This paper studies the Chinese innovation industry. This industry has grown extremely fast in the recent years as discussed earlier.
This paper implements an empirical study in Chinese listed innovative companies to investigate how Chinese CEO's overconfidence affects innovative investment and firm value. This topic is worth to investigate since firstly it is important to know how overconfident CEOs affect firm investment decision-making, especially in Chinese innovative industry and secondly whether those overconfident CEOs are beneficial to the firm value. China is now experiencing a complicated and dynamic transitional market, which is a rich context to influence innovation (Zhou and Li, 2012). Zhang and Li (2010) also point out that the competition in China is increasing since China plays an important role on global economics. This forces the companies in China to develop radical innovations. The combination of these elements makes investigating the Chinese market interesting and relevant.
4 There are two very related studies. One is Galasso and Simcoe (2010), who investigate 627 CEO in U.S. firm during 1980 to 1994. Another is the research of Hirshleifer, Low and Teoh (2012), which investigate 2,577 CEOs between 1993 and 2003. Both of these two studies build on the overconfidence measure developed by Malmendier and Tate (2005a,b; 2008) and study the overconfidence effect on patent and citation count. They find a positive relationship between overconfidence and innovation, but Hirshleifer, Low and Teoh (2012) find the result only in innovation industry. Our study differs firstly because we focus on Chinese companies, a totally different cultural and economical environment from the U.S., secondly we apply different ways to measure CEO’s overconfidence and lastly we investigate a more recent sample.
The second research question is related to Goel and Thakor (2008) and Gervais, Heaton and Odean (2011). These two studies build two different theoretical models but reach a similar conclusion: CEO's overconfidence benefits firm value when it is moderate, but when it goes to extreme, it is detrimental to firm value. This thesis will test the theoretical predictions for innovative industry in China.
Previous studies have used the method developed by Malmendier and Tate (2005a,b; 2008) to measure overconfidence. Malmendier and Tate use two proxies based on option exercise behavior and press coverage. The option exercise measure defines the CEOs as overconfidence that fail to exercise 67% in-the-money option with 5 years remaining duration. The press coverage measure uses Press based measure, the author use the Press portrays in The New York Times, Business Week, Financial Times, The Economist and The Wall Street Journal and record the numbers of articles that use words such as “confident” “optimistic” to describe the CEO. The study of Galasso and Simcoe (2011) and the study of Hirshleifer et al (2012) used these two proxies to study the relationship between overconfidence and firm decision.
These measures are not available for the data used in this paper because the option incentive is not prevalent in China. Thus there is not sufficient data for investigation. Moreover, the
5 data of press coverage is hard to obtain since there is not valid source. Therefore we use alternative proxies for CEO's overconfidence. The first proxy we use is CEO’s self-importance developed by Hayward and Hambrick (1997), which is measured as CEO’s relative compensation. The related compensation is calculated by the ratio of CEO’s compensation and the compensation of second highest officer. The second proxy is based on the study of Hriber and Yang (2006). They relate overconfidence to a tendency to issue unduly optimistic earnings forecasts. Therefore, we calculate the degree of CEO’s overconfidence as the difference of earnings forecasts and real earnings.
Our study reveals a positive significant relationship between CEO’s overconfidence and R&D expenditure. The result also indicates that overconfident CEO influence firm value by invest more heavily in innovation. However, the relative compensation based measure indicates a quadratic relationship between overconfidence and firm value while the forecast base measure indicates a linear relationship of it. Since we do not get same result by our two measures, we should interpret our results with caution.
There are some weaknesses of this study. Firstly, because of data unavailability, we do not built a more widely used proxy as Malmendier and Tate (2005a,b; 2008) used, but use a part of the proxy (self-importance) from the study of Hayward and Hambrick (1997) to show overconfidence. This might biases our results. Secondly, there only exists small positive but insignificant correlation between the two measures. This shows that the measure of these two proxies overlaps little. Third, we only include three control variables in the test. Failing to find more control variables might result in omitted variable bias. In the discussion of this paper, more attention will be paid to these weaknesses and how they might affect our findings.
The contribution of this study is that we investigate the Chinese innovative industry, which is experiencing a highly complicated and dynamic transitional period. Although we find contradicting results on the relationships between overconfidence and firm value, we still
6 have some other interesting results: First, we find a robust positive effect CEO’s overconfidence on innovation investment in both measures. This suggests overconfident CEO in innovative companies invest more heavily in innovation. Second, we find that under the forecast based measure R&D expenditure seems to influence firm value. This result is robust under forecast-based measure.
The paper proceeds as follows. Part 2 we review some literature related to our topic. Part 3 we describe the data used and the empirical model. In part 4 the result will be discussed. We conclude our paper in the last part.
2. Related literature and hypothesis
2.1 OverconfidenceOverconfidence builds on the prominent stylised “better than average” effect from social psychology research. Early researchers like Larwood and Whittaker (1977) defined this effect as a tendency to overstate one’s ability on average. Camerer and Lovallo(1999) show overconfident individuals perceive their ability of making profit is higher as compared to others. Hirshleifer, Low and Teoh describe overconfidence as “the tendency of individuals to think that they are better than they really are in terms of characteristics such as ability, judgment, or prospects for successful life outcomes”(Hirshleifer, Low and Teoh, 2011, p.1458). The study of Hayward and Hambrick (1997) use the term “hubris” to relate to overconfidence. They take the definition of “hubris” from the dictionary: “exaggerated pride or self-confidence, often resulting in retribution”(P.106). In their study, they perceive overconfidence as that it affect individuals when assessing their ability to extract benefit. Overconfidence affects individuals endogenously. The study of Malmendier and Tate (2005a,b; 2008) classify individuals as overconfident when they persistently expose themselves to risk. This classification indicates that overconfident individuals have different risk sensitivity. Similar to their studies, we consider overconfident individuals as individuals who are optimistic about prospects. In line with the study of Camerer and Lovallo (1999) and
7 Malmendier and Tate (2005a,b; 2008) use the term “confidence” to describe the “better than average” effect and focus on managers’ psychological biases.
Prior literature indicates that overconfidence, as a personal characteristic, is prevalent and prominent, but differs along many dimensions. Svenson(1981) first show that large majority of people rate their driving ability as “better than average” in a survey. Camerer and Lovallo (1999) find that people are more likely to perceive themselves as “better than average” when they think they are good at trivia questions about sports or current events in an experimental setting. This result indicates “better than average” effect is skilled dependent. Moore and Kim (2003) show that these effects become stronger when the relationship between action and result is complex and the outcome is abstract. Furthermore, gender effects play an important role in overconfidence. Barber and Odean (2001) in their study investigate the common stock investment behavior among men and women. They find that men trade 45% more excessively than women but suffer more losses. Ben-David, Graham and Harvey (2007) investigate managerial overconfidence and optimism in survey among Chief Financial Officers (CFOs) and find both characteristics persist over time while overconfidence exhibits stronger effects. This suggests that the “better than average” effect remains relative stable and persistent over time for an individual.
Some researchers have found that this “better than average” effect have higher incident rate in entrepreneurs than non-entrepreneurs. Overconfidence is observed by researchers in many occupations such as lawyers (Wagenneer and Keren 1986), physicians and nurses (Baumenn, Deber and Thompson 1991) and engineers (Kidd 1970). These jobs involve difficult tasks and are highly skill dependent. Entrepreneurs are also found to be overconfident when they make decisions. The result of the survey conducted by Cooper, Woo and Dunkelberg (1988) shows that given the same kinds of business, 68% of the entrepreneurs thought their businesses have higher odds to success than others, while only 5% thought their chances are poorer. Odean (1998) develops a model that shows overconfident investors overestimate the precision of the knowledge about financial security. In the model, overconfident investors hold unrealistic belief about their judgments towards financial return as well as the precision
8 of their estimation. Experimental evidence by Camerer and Lovallo (1999) indicates that an individual’s overestimate towards his or her ability relative to his or her counterparts’ leads to excessive entry to the entrepreneurship game, which suggest that overconfident individuals self-select into the market.
Malmendier and Tate (2005a,b) in their study of CEO’s overconfidence and corporate investment provide reasons why high-rank executives are more likely to be overconfident. Malmendier and Tate suggest overconfident CEOs tend to use more internal fund instead of external since they overestimate the return generated from internal funds, they believe that their internal financial resources are enough for their investment decisions. On the other hand, overconfident CEOs might underestimate the risk of their decisions. Bernardo and Welch (2001) suggest that overconfident entrepreneurs tend to believe they have more information and perceive their information more valuable than it actually is. Previous studies show that CEO’s overconfidence is correlated to their job characteristics. For example, Langer (1975) shows that an individual is more optimistic about the outcome when he believes he has control towards it. Furthermore, CEOs, in most cases, make the final decision for the firm and they are believed to control the firm. Such a position may make them to believe they can control the outcome (March and Shapira, 1987). Moreover, Weinstein (1980) suggest that highly committed individuals are prone to be more optimistic about the outcome. We expect that CEOs are highly committed since their value and compensation is largely dependent on their performance. In a nutshell, there is strong evidence support that CEO’s psychological bias is relatively strong and can influence firm decisions.
2.2 Innovation investment
Innovation is a process of translating the idea into new product, service and process (Bissant and Tidd, 2007). Bolwijn and Kumpe (1990) suggest that innovation is not limited to new technologies, new products and services, but also include expanding new markets, setting up new business, formulating new missions. The description by Crossan and Apaydin is more detailed: “innovation is a production or adoption, assimilation and exploitation of a value-added novelty in economic and social spheres, renewal and enlargement of products, services
9 and markets, development of new methods of production and the establishment of new management systems. It is both a process and an outcome.”(Crossan and Apaydin, 2010, p1155)
Innovation investment is reflected by Research and Development (R&D) in companies. Innovative investment not only allows companies to develop new technologies and provide new products and services, but also strengthens the firm’s competitiveness. Cohen and Levinthal (1989) suggest that innovation investment also develops firm’s ability to obtain knowledge from the outside environment. Furthermore, Bolwijn and Kumpe (1990) indicate that innovation can reduce cost, improve quality and increase flexibility of the firm. They also claim that the ultimate goal of innovation is to provide best product respect to price, quality and performance. Moreover, Page (2005) claims that the challenge of innovation involving difficulties, uncertainty and conflicts. Innovative investment is full of risk and uncertainty since the process is highly complicated and the outcome is mostly ambiguous.
Although both innovative investment and general investment can enhance the competitiveness of the firm, there still exists remarkable difference between them. Innovation investment is usually characterized by its highly knowledge dependent process and is susceptive to managers’ knowledge. Bantel and Jackson (1989) find that firms with more educated managers who have different background yield more innovation outcomes. Bolwijn and Kumpe (1990) claim that the success of innovative company depends on the knowledge and expertise of its employees, which means human capital is important in innovative company. Barker and Mueller (2002) find that CEO characteristics such as age, career experience, tenure, and education background can largely explain the sample variance in corporate R&D investment. There is strong evidence that innovation investment is more concerned about knowledge capital and human capital in a firm.
2.3 CEO's overconfidence and innovation investment
Overconfidence affects CEOs’ investment decisions from a psychological way. Overconfidence affects how CEOs interpret situations by distorting their assessment on their
10 ability and resource and therefore influences their judgment on potential risk of their decision-making (Malmendier and Tate 2005a,b; 2008; Hirshleifer and Luo, 2001).
There are two studies that investigate the relationship between overconfidence and innovation. One is the study of Galasso and Simcoe (2010). They predict that the propensity to innovate of overconfident CEOs increases when the return of innovation is high. They also predict that the increased innovation propensity might offset the negative impact by failure merger and acquisition (M&A) and suboptimal investment decision. They perform an empirical study using a sample of 627 CEOs during the period 1980 to 1994, based on the measure developed by Malmendier and Tate (2005a,b; 2008). The author finds that overconfidence of the CEO is correlated with a 25% to 35% increase in innovation outcome. This supports the prediction.
Close to this study, Hirshleifer, Low, and Teoh (2012) also built on the measure of Malmendier and Tate (2005a,b; 2008). They consider the period 1993 to 2003 in their study. Their findings are that firms with overconfident CEOs invest more heavily on innovation and produce more innovation outcomes. Innovation outcomes are measured by patent and citation count. These empirical results show that innovation investment increase with managerial overconfidence.
There are three reasons that support that CEOs are more likely to be overconfident when they are making innovation investment decision. Firstly, prior research shows that the effect of overconfidence is stronger when the situation is skill oriented and highly abstract, the outcome predictability is low and the feedback is slow and ambiguous (Moore and Kim, 2003; Griffin and Tversky, 1992; Camerer and Lovallo 1999). Moore and Kim (2003) find this effect become strongest when the direct comparisons between individuals are hard and the link between action and result is complex and outcome is abstract. Since the innovation outcome is usually intangible (like patents and knowledge), the link between investment and innovation outcome is unclear and the comparison among innovation outcome is hard, we therefore expect that the relationship between overconfidence and innovation decisions is stronger in the innovation industry.
11 Secondly, Camerer and Lovallo (1999) test the personal overconfidence and economic decisions simultaneously in an experimental setting. This laboratory study find that overconfident participants keep excessively entering the market even though the market is overcrowded, since they hold the belief that they are capable to prevail in the game. They also find that the “better than average” effect is especially strong among highly skilled individuals, since they find that there are more excessive entries when participants know their payoffs are dependent on skill. We assume CEOs in innovation industry companies have certain education background and professional knowledge or skill in this field. This might leads to CEO overestimate the success of investment when making decision. We therefore expect that CEOs in innovation industry companies invest more in innovation.
Lastly, early research shows that feedback can be ambiguous, misleading and hard to process and reinforce psychological bias when making decisions, even for an expert (Einhorn, 1980; Northcraft and Neale, 1986). Einhorn (1980) also points out that the effect of overconfidence is stronger when the feedback is deferred. Since the outcome of innovation investment projects take a long time to resolve and therefore overconfidence tends to be more severe (Hirshleifer, Low, and Teoh 2010), we expect that the effect of overconfidence is stronger when making innovation investment. There is strong evidence to believe that overconfident CEOs tend to invest more heavily in innovation. We therefore hypothesis that:
Hypothesis 1: Overconfident CEOs will invest more in innovation.
2.4 Innovation investment and firm value
Innovation helps the organization to improve efficiency and reduce costs. Overconfident CEOs tend to explore new environments instead of imitate the competitors, this behavior help to bring the organization valuable information (Bernardo & Welch, 2001). The research by Hirshleifer, Low and Teoh (2012) show that overconfident CEOs in the innovative industry bring more corporate patent rights compared to normal confident CEOs. They help improving
12 the organization innovation efficiency and thus bring higher innovative success. The research of Korea companies shows that the R&D investment is positively related to the stock price. Furthermore, investors think R&D investment can bring positive net present value (Hana & Manry, 2004). However, some studies show the opposite results. A larger innovation investment does not necessarily yield a higher firm value since not all innovation investment translates into firm value. Prior research shows that R&D investment in computer industries is negatively related to the total shareholder return. The result can be explained by CEO opportunism, which cause the failure for innovation investment to increase firm value (Mank & Nystrom, 2001). Hall, Jaffe and Trajtenberg (2005) study the relationship between market value and patent citations using Tobin’s Q and the ratio of R&D expenditure and asset during the year 1963 to 1995. They find a strong positive relationship between patent citation and firm value. The result shows that, on average, patent citation increase firm value by 3%. Although the authors show a positive relationship between innovation outcome and firm value, they also suggest that patents might be detrimental to firm value. For example, if patents spread too many different fields, the firm might find it hard to benefit, leading to inefficient innovation.
Previous research does not provide clear direction of the effect of innovation investment on firm value. Mairesse and Sasseneou (1991) suggest that it is difficult to establish a reliable statistical relationship between innovation investment and productivity, since innovation investment affects the firm in various aspects and it may vary from one firm to another. To avoid endogeneity problem, Hirshleifer, Low and Teoh (2012) do not run a direct regression between firm value and innovation investment or managerial overconfidence, instead they use an instrument for exogenous growth opportunity and find a positive relationship.
The study of Loof and Heshimati (2002) investigate relationship between knowledge capital and firm performance heterogeneity among Swedish manufacturing. The empirical result of their study shows that knowledge capital, defined as the ratio of innovation sales and total sales, increases with innovation investment. Furthermore, the study also reveals that knowledge capital is a significant factor to contribute to firm performance heterogeneity
13 among Swedish firms. They find firm productivity increases largely with knowledge capital when controlling for sector, size, capital intensity, human capital, recent establishment, merge and closure. Their findings indicate that innovation investment might positively affect firm value through innovation investment. We therefore hypothesis that:
Hypothesis 2: There exists a positive relationship between innovation investment and firm value.
2.5 CEO's overconfidence and firm value
The effect of CEO's overconfidence on firm value has been investigated in previous empirical studies. Hirshleifer, Low and Teoh (2012) find that overconfident CEOs are better at transferring external growth opportunities (instrument variable) into firm value, which suggest that managerial overconfidence benefits the firm. The result of Galasso and Simcoe (2011) show that overconfident CEO is correlated with a 25% to 35% increase in citation-weighted patent count, which means overconfident CEOs gain more patent and citation in every dollar of investment. Since they use an indirect measurement of innovation outcome and they do not have information about opportunity cost, they cannot explicitly conclude that effect of overconfidence on firm value. Moreover, the study of Hall, Jaffe and Trajtenberg (2005) indicate a positive relationship between patent citations and firm value, which means the higher innovation outcome, the higher firm value. We might induce that overconfident on average benefits firm value by obtaining more innovation. Hence, overconfident CEOs might influence firm value indirectly by investing more in innovation.
Overconfidence could also have a passive impact towards firm value when it goes to an extreme. Roll (1986) is the first one who studies the firm outcome with overconfident decision makers in bidding firm, he hypothesized that corporate takeovers could be explained by decision makers’ hubris and linked the winner's curse to the takeover contest. He finds that overconfidence hurts firm outcome. In later studies, researchers find this psychological bias might worsen outcome. For example, Malmendier and Tate (2005a) suggest
14 overconfident CEOs rely more heavily on internal fund instead of external fund since they overestimate the return generated from internal fund. Hayward and Hambrick (1997) indicate such misinterpretation might leads to paying higher premiums for acquisitions. Malmendier and Tate (2008) find that overconfident CEOs are more likely to undertake more acquisitions that on average do not create value, which destroys firm value. The results suggest overconfident would help to increase innovative growth.
Theoretical studies such as Goel and Thakor (2008) and Gervais, Heaton and Odean (2011) find a non-linear relationship between overconfident manager and firm value and conclude that extreme overconfident turns out to be detrimental. Goel and Thakor (2008) develop a two period selection model with one CEO, several risk averse subordinate managers and risk neutral shareholders. In their theoretical model, CEO decides the strategy of the firm, which influences the return of the investment projects. Those projects are selected by and implemented by managers. The payoff of managers will be decided by managers’ decision and ability. The shareholders will decide to replace incumbent CEO with subordinate manager. The author find that moderate overconfident CEOs overestimate the precision of their information and overreact to it, and they benefit firm value by mitigating underinvestment problem. Their study also finds that overconfident managers are more likely to be promoted to CEO position in the model, which also suggests that overconfident decision makers benefit the firm. However, although low risk adverse CEO always benefit the firm, their study subsequently shows that extreme overconfident CEO hurt firm value.
Another theoretical study by Gervais, Heaton and Odean (2011) develop a capital budgeting model and find similar results to Goel and Thakor. The model by Gervais, Heaton and Odean consist a risk neutral firm that hire a risk averse manager who make investment decision after receiving some noisy signals about the quality of the projects. If the manager is overconfident, then he overestimates the precision of his judgement from the signal, and unduly implements the project when his information is positive and vice versa. They show that economic surplus increase with overconfidence when overconfidence is mild. They latter introduce the competitive labor market. In the competitive labor market, firm competes to hire managers
15 who are good at transfer economic surplus for the firm. They find that overconfident managers guarantee to implement valuable risky project, which benefit the firm. However, when overconfidence goes to an extreme, managers will be always worse off and this is detrimental to the firm. Both theoretical studies suggest that moderate overconfident CEOs are good for firm value while extremely overconfident CEO hurt firm value. In our study, we expect an inverse U-shape as Goel and Thakor (2008) and Gervais, Heaton and Odean (2011) indicate. We therefore hypothesize:
Hypothesis 3: Moderate confident CEOs will yield a higher firm value while extreme overconfident CEOs detrimental to firm value.
3. Methodology
3.1 The dataThis study focuses on the listed companies in Chinese innovative industry. Sample year is from 2010 to 2014. This period excludes the financial crisis in 2008, which might affect our result. Some data are missing in the database due to unclear reasons. We delete the firm-year from our sample if there is any missing data in our independent, dependent or control variables. CEOs, in our sample, are required to keep unchanged for firm-year to ensure our observation is the same. Therefore, our panel data is unbalanced. We include 1038 firm-year observations and 406 listed companies in the relative compensation based sample. For forecast based sample, we include 421 firm-year observations and 202 companies. We then classify the innovative industry base on the most recent category released by National Bureau of Statistics of China (NBSC) in 2013. This category is based on the criteria that the intensity of R&D expenditure is relatively high. This category includes 6 industries of manufacturing firms and 9 industries of service firms. Pharmaceutical manufacturing, aviation, spacecraft and equipment manufacturing, electronics and communications equipment manufacturing, computer and office equipment manufacturing, medical equipment and instrument manufacturing, chemical manufacturing information are the six manufacturing industries in
16 the list. Information services, E-commerce services, inspection services, high-tech professional and technical services, research and design services, technological transformation services, intellectual property and related legal services, environmental monitoring and management services are the nine industries in the list.
We use several databases to construct the sample. CZMAR provide the data of top executive and chairman's compensation. We use it to construct our first measure of CEO’s overconfidence. CZMAR also provide the data of firm value, we use Tobin’s Q. RESSET provide the data of our second overconfidence measure, which is composed by CEO’s prediction towards the net profit of the year and the real net profit of from the annual report. We also get the data of innovation investment from RESSET, in which we obtain the R&D expense from the annual report. Other variables such as the size of the firm, the growth of the firm, the current asset and the total asset are collected from CZMAR. Current asset shows the asset constantly flow in and out of the firm such as cash and inventory. Any asset expected to last or be used for less than one year is considered as current asset. Total asset refers to the final amount of asset present in the balance sheet.
3.2 Empirical model 3.2.1 Regression functions
To test the relationship between CEO’s confidence level and R&D expenditure, we apply the regression functions shows as follows:
Where RD is R&D expenditure, CON is CEO confidence, TAGR is total asset growth rate, ALR is asset liability ratio, CA is current asset and TA is total asset. YEAR* and IND* is the fix year and industry effect.
17 To test the relationship between firm value and CEO confidence level, we construct a quadratic function according to the theoretical studies by Goel and Thakor (2008) and Gervais, Heaton and Odean (2011). The quadratic regression function shows below:
Where Tobin’s Q shows firm value, CON is CEO confidence, SQU(CON) is the quadratic term of CEO’s overconfidence. TAGR is total asset growth rate, ALR is asset liability ratio, CA is current asset and TA is total asset. YEAR* and IND* is the fix year and industry effect.
3.2.2 First Measuring overconfidence
Following the study of Hayward and Hambrick (1997), we use CEO’s self-importance to show managerial overconfidence. Hayward and Hambrick (1997) in their study construct a model of CEO hubris, which is composed as recent performance, media praise and CEO’s self-importance. Since the efficiency of stock market in China is still low (Lim, Brooks and Kim, 2008; Wang et al, 2010), we do not use the “abnormal return on stock” to measure recent performance as the author does. Moreover, the media praise is not available due to limited source. Therefore we only consider self-importance. Hayward and Hambrick (1997) test the effect of overconfidence first by the three separated hubris variables. They find the results are significant and all these three variables are positively correlated to acquisition premiums. They subsequently test the hubris factor and find a similar result as the result of three separated hubris variables. Concerning the availability of data in China, we choose CEO’s self-importance as a proxy for overconfidence. As Hayward and Hambrick (1997), we use CEO relative compensation to measure CEO self-importance. Relative compensation is calculated by CEO cash compensation divided by the compensation of second highest officer. CZMAR database is the source of this proxy.
18 3.2.3 Second measuring overconfidence
Heaton (2002) predicts that overconfident managers hold a biased cash flow forecast. Hriber and Yang (2006) found that overconfident CEOs have a tendency to issue unduly optimistic earnings forecasts. Our second measure of overconfidence employed is a forecast based measure similar to the measure used by Ben-David, Graham and Harvey (2007). They measure the confidence level by asking executives to forecast future net profit. In the survey, Ben-David, Graham and Harvey ask executives about their prediction towards stock index. Their survey questions are arranged as “Over the next year, I expect the S&P 500 will be….” In our study, we use data taken from the forecast report issued in every three quarterly report, which is approved by CEOs. We use the difference of net profit forecast and the real net profit of the sample year to measure CEO’s overconfidence. The degree of CEO’s overconfidence is larger when the difference of CEO net profit estimation and real net profit is larger.
3.2.4 Other variables
We collect the innovation investment data from the balance sheet in annual report. For the firm market value, we use Tobin’s Q as a proxy, which is the ratio between a physical asset's market value and its replacement value. The study of Malmendier and Tate (2005a,b) show that CEOs are highly sensitive to cash flow when making investment decisions. We therefore control cash flow by including current asset and we expect a positive relationship between overconfidence and cash flow. Some factors relative to firms’ market value and innovation investment also need to be controlled, such as the size of the firm, the growth of the firm and the total asset. The change of these factors might influence the R&D expenditure and firm value (Hirshleifer, Low, and Teoh, 2012). We expect that all these factors are positively correlated to the R&D expenditure and firm value. To control the industry and time effect, we control the year and industry. The industry is based on the criteria of “National Industry Classification”(GB/T 4754-2011) released by National Bureau of Statistics of China (NBSC).
19
4. Results
4.1 Descriptive statistics
Table 1
Summary Statistics
The table provides the mean, median, standard deviation, maximum and minimum value of the variables used in the study. The sample is collected from Chinese database CZMAR and RESSET. The sample includes the Chinese listed company in innovation industries defined by National Bureau of Statistics of China (NBSC) in 2013. The sample period is from the year 2010 to 2014. This period consists five years and exclude financial crisis in 2008. Panel A measures the CEO confidence using the relative compensation of CEOs. Panel B measures the CEO’s overconfidence base on the CEO forecast towards the net profit of the firm. The relative compensation measure is calculated by the ratio of CEOs’ salary and the salary of second highest officer. The forecast-based measure defines CEO confidence using the difference of CEO’s net profit forecast and firm’s true net profit.
Panel A: Compensation-Based Measure of Confidence
Variable Compensation-Based Measure (N=1038)
Mean Median Std. Dev. Min Max
Dependent variables Tobin’s Q 2.24 1.82 1.36 0.84 15.11 R&D (million) 47.03 13.73 170.22 0.02 3483.51 Independent variable Relative compensation 1.30 1.17 0.50 1 10 Control variables
Total asset growth rate (%) 0.22 0.12 0.50 -0.40 11.44
Asset liability ratio 0.37 0.36 0.21 0.01 1.16
Current asset (billion) 4.71 1.36 13.34 0.04 133.54
Total asset (billion) 7.29 2.40 18.45 0.14 179.16
20 sample in which overconfident CEOs are measured based on the relative compensations of CEOs. Panel B shows the descriptive statistics of the sample measured by forecast-based confidence proxy. There are 1038 observations in Panel A and 421 observations in Panel B.
In panel A, we observe that the mean of Tobin’s q is 2.24 and the mean of the R&D is 47.03 million. However, both Tobin’s Q and R&D expenditure are found to have some observations with large amount. The difference of R&D investments is specially large since the maximum R&D investment is 3483.51 million while the lowest R&D investment is only 0.02 million. Moreover, the mean of R&D expenditure is almost four times higher than its median. This indicates that there exists some large value affect the mean of R&D expenditure.
With respect to relative compensation-based measure of confidence, the mean of the relative compensation is above 1. This means all CEOs in the sample obtain the highest salaries within the organizations. The average is relative low (1.30), so we can find that the distances between the salary of CEOs and the salary of second highest officer are not high on average. However, in some cases, the salary distance between CEO and the second highest salary officer is very large. The maximum value of relative compensation is 10, which means the salary of the CEOs is 10 times higher than the second highest salary employee. This could be an outlier in the sample. We will discuss it in the discussion part.
Moreover, the means of the control variables are 0.22% (total asset growth rate); 0.37 (asset liability ratio), 4.71 billion (current asset), 7.29 billion (total asset), respectively. Consistent with R&D expenditure, total asset growth rate, current asset and total asset are found to have some large values. In addition, relative stable asset liability ratio shows that innovation firms do not have very high debt.
21
Table 1-Continued
Panel B: Forecast-Based Measure of Confidence
Variable
Forecast-Based Measure (N=421)
Mean Median Std. Dev. Min Max
Dependent variables Tobin’s Q 2.14 1.74 1.26 0.95 8.87 R&D (million) 34.32 9.54 156.75 0.02 2932.15 Independent variable Forecast bias -10.99 -4.09 72.09 -984.69 371.42 Control variables
Total asset growth rate (%) 0.24 0.11 0.52 -0.40 4.57
Asset liability ratio 0.30 0.24 0.22 0.01 1.16
Current asset (billion) 1.80 0.93 5.15 0.03 76.41
Total asset (billion) 3.07 1.40 7.73 0.13 100.06
Panel B shows the descriptive statistics whose independent variable is based on the forecast measurement. Tobin’s Q and R&D expenditure shows a similar distribution as in the sample under relative compensation measure. However, compares to the data under relative compensation measure, both Tobin’s Q and R&D expenditure have lower values. Tobin’s Q and R&D expenditure in Panel B have both lower values of mean and maximum compares to the values in Panel A. This might indicates that the companies in the sample under forecast-based measure are the companies with smaller size and lower R&D expenditure.
The mean of forecast-based measure overconfidence is -10.99. This shows that in most cases, CEOs underestimate their companies’ net profit. The mean of forecast-based measure is lower than its median, which means that some CEOs tend to largely underestimate the net profit of the firm.
22 The means of control variables are 0.24 (total asset growth rate), 0.30 (asset liability ratio), 1.80 (current asset) and 3.07 (total asset), respectively. The companies scales differ greatly, for example, the minimum value of total asset is 0.13 billion while the maximum value of total asset reaches 100.06 billion. Consistent with the compensation-based measure, the control variables show a similar distribution but with lower values. The mean of current asset and total asset are 1.8 and 3.07 respectively, which are especially lower than those in the sample under relative compensation measure.
4.2 Correlation Analysis
Table 2
Correlation Analysis
The table provides the correlation of all the variables used in the study. Panel A measures the CEO confidence using the relative compensation of CEOs. Panel B measures the CEO’s overconfidence base on the CEO forecast towards the net profit of the firm. The numbers in the brackets under the correlation coefficient value is the p value of the correlation. *, ** and *** measure significance at the 10%, 5% and 1% level, respectively.
Panel A: Compensation-Based Measure R&D Tobin’s Q Confide
nce (Compe nsation) Total asset growth rate Asset liability ratio Current asset Total asset R&D 1.000 Tobin’s Q -0.071** (0.022) 1.000 Confidence (Compensation) 0.075** (0.016) 0.034 (0.269) 1.000 Total asset growth rate -0.012 (0.707) 0.010 (0.748) 0.007 (0.823) 1.000 Asset liability ratio 0.190*** (0.000) -0.157*** (0.000) 0.007 (0.828) 0.059* (0.058) 1.000 Current asset 0.396*** (0.000) -0.185*** (0.000) 0.034 (0.278) 0.034 (0.272) 0.318*** (0.000) 1.000 Total asset 0.389*** (0.000) -0.200*** (0.000) 0.036 (0.253) 0.029 (0.352) 0.346*** (0.000) 0.989*** (0.000) 1.000
23 To avoid multi-collinearity problem, it is important to check the correlation between each variables. Table 2 shows the correlation analysis of all the variables in the study. The analysis in Panel A is based on the relative compensation-based measure. In Panel B, the correlation is based on forecast-based measure.
From Panel A, we find that the Tobin’s Q is negatively correlated to the R&D, with the coefficient equals to -0.071 and the significant level is 0.05. Moreover, the CEO’s overconfidence is positively correlated to the R&D, with the coefficient equals to 0.075 and the significant level reaches the 0.05. Furthermore, the asset liability ratio (coefficient=0.190), current asset (coefficient=0.396) and total asset (coefficient=0.389) all positively correlated to the R&D, the significant level is 0.01. However, the asset liability ratio (coefficient=-0.157), current asset (coefficient=-0.185) and total asset (coefficient=-0.200) are negatively correlated to Tobin’s Q, the significant level also reaches 0.01. Besides, the asset liability ratio, current asset and total asset are all positive correlated to each other, the significant value reaches 0.01. Furthermore, total asset growth rate is positive related to asset liability ratio, the coefficient equals to 0.059, but the significant level only reaches 0.1. However, we find a strong correlation between total asset and current asset, with a significant coefficient every close to 1 (coefficient=0.989). We can conclude that current asset and total asset are very similar, therefore we only include current asset in our regression under compensation-based measure.
Panel B shows the results of correlation analysis in which the measure of CEO’s overconfidence based on the forecast measure. In Panel B, we find that CEO’s overconfidence is negatively correlated to the R&D, the coefficient equals -0.2963 and the p value is lower than 0.01. Accordingly, the asset liability is positive correlate to the R&D investment, in which the coefficient equals to 0.162 and the p value is lower than 0.01. Besides, the current asset and total asset are negative correlated to Tobin’s Q, the coefficients are -0.132 and -0.151 respective and the significant level all reaches 0.01. Furthermore, the forecast bias is negative correlated to the current asset (-0.625) and total asset (-0.645) and
24 the significant level is 0.01. The current asset is positive correlated to asset liability ratio. The coefficient equals to 0.280 and the significant level is 0.01.
Table 2-Countinued Panel B: Forecast-Based Measure R&D Tobin’s Q Confidence
(Forecast) Total asset growth rate Asset liability ratio Current asset Total asset R&D 1.000 Tobin’s Q -0.062 (0.202) 1.000 Confidence (Forecast) -0.2963*** (0.000) 0.017 (0.722) 1.000 Total asset growth rate -0.043 (0.398) 0.027 (0.590) -0.027 (0.600) 1.000 Asset Liability ratio 0.162*** (0.001) 0.022 (0.662) 0.073 (0.137) -0.066 (0.190) 1.000 Current asset 0.816*** (0.000) -0.132*** (0.007) -0.625*** (0.000) 0.038 (0.454) 0.280*** (0.000) 1.000 Total asset 0.754*** (0.000) -0.151*** (0.002) -0.645*** (0.000) -0.036 (0.473) 0.339*** (0.000) 0.977*** (0.000) 1.000
Consistent with the correlation analysis of relative compensation measure, the current asset and the total asset shows a highly correlation with a coefficient equals to 0.977. Meanwhile, the current asset also positively correlate to the R&D, with the coefficient equals to 0.816. The total asset also positively correlate to R&D, with a coefficient equals to 0.754. As a result, we only exclude total asset from our control variable in the later regression.
Table 3 shows the result of correlation analysis between compensation-based measure and forecast-based measure. The table shows that the forecast-based measure does not highly correlated to compensation-based measure. Therefore, we can conclude that the internal consistency of these two measures is weak. The results should be interpreted with caution.
25 Table 3
Correlation Analysis
The table provides the correlation of compensation-based measure and forecast-based measure of confidence. This correlation test uses the observations that both measures have. The numbers in the brackets under the correlation coefficient value is the p value of the correlation. *, ** and *** measure significance at the 10%, 5% and 1% level, respectively.
Correlation Analysis
Compensation-Based Measure Forecast-Based Measure Compensation-Based Measure 1.000
Forecast-Based Measure 0.012 (0.823)
1.000
4.3 Linear regression
Table 4 shows the results of linear regression analysis. Regression in Panel A is by compensation-based measure. The result shows that CEO’s overconfidence is positive correlated to R&D investments with the coefficient equals to 21.183 and 0.05 significance level. After adding the control year and industry effect, the relationship between CEO’s overconfidence and R&D investment is also significant. This result indicates a significant positive relationship between overconfidence and R&D expenditure. Panel B are the regressions by forecast-based measure. The result shows again CEO’s overconfidence is positively related to the R&D investment, with the coefficient equals to 0.808. When controlling the industry and year effect, the coefficient increased to 0.857. Base on the regression of two different measures, we can conclude that the CEO’s overconfidence can lead to higher R&D investment. When considering the fixed effect of industries and years, the relationship between CEO’s overconfidence and R&D investment is still significant. This shows our empirical results are robust. Therefore, we accept our first hypothesis.
26 Table 4
Overconfident and R&D Expenditure
The table provides the regression result of R&D expenditure and CEO’s confidence. The sample includes the Chinese listed innovation industry companies during the year 2010 to 2014. Confidence
(Compensation) is a proxy of CEO’s overconfidence measured by CEO’s relative compensation
compares to the compensation of second highest officer. Confidence (Forecast) is another proxy of CEO’s overconfidence measured by the difference of CEO prediction of net profit and the real net profit. The numbers in the brackets under the coefficient value is the t value. *, ** and *** measure significance at the 10%, 5% and 1% level, respectively.
Panel A: Compensation-Based Measure Dependent variable = R&D Expenditure (million)
(1) (2) (3) Confidence (Compensation) 25.581*** (2.41) 21.183*** (2.18) 17.616** (1.82)
Total asset growth rate -9.866
(-1.02)
-16.782* (-1.75)
Asset liability ratio 58.472***
(2.44)
92.936*** (3.72)
Current asset (billion) 4.744***
(12.39)
5.021*** (12.89)
Industry effect No No Yes
Year effect No No Yes
Constant 13.697 -22.358 -84.590
Adjusted R2 0.0046 0.163 0.197
Panel B: Forecast-Based Measure
Dependent variable = R&D Expenditure (million)
(1) (2) (3) Confidence (Forecast) -0.644*** (-6.35) 0.807*** (11.39) 0.857*** (11.78)
Total asset growth rate 0.056
(0.01)
-1.829 (-0.22)
Asset liability ratio -88.542***
(-4.50)
-83.816*** (-3.50)
Current asset (billion) 32.910***
(31.94)
33.608*** (31.58)
Industry effect No No Yes
Year effect No No Yes
Constant 27.241 11.739 -2.829
27 Table 5
Overconfident and Firm value
The table provides the quadratic regression result of firm value and CEO’s confidence. The sample includes the Chinese listed innovation industry companies during the year 2010 to 2014. Confidence
(Compensation) is a proxy of CEO’s overconfidence measured by CEO’s relative compensation
compares to the compensation of second highest officer. Confidence (Forecast) is another proxy of CEO’s overconfidence measured by the difference of CEO prediction of net profit and the real net profit. Square Confidence (Compensation) is the quadratic term of Confidence (Compensation).
Square Confidence (Forecast) is the quadratic term of Confidence (Forecast). The numbers in the
brackets under the coefficient value is the t value. *, ** and *** measure significance at the 10%, 5% and 1% level, respectively.
Dependent variable = Tobin’s Q
(1) (2) (3) (4) Confidence (Compensation) 0.457*** (2.60) 0.463*** (2.81) Square confidence (Compensation) -0.055** (-2.23) -0.057** (-2.50) Confidence (Forecast) -0.309** (-2.06) -0.148 (-1.01) Square confidence (Forecast) -0.021 (-1.07) -0.006 (-0.34)
Total asset growth rate 0.056 (0.68) -0.026 (-0.35) 0.040 (0.32) 0.020 (0.17) Asset liability ratio -0.659**
(-3.23) -0.751*** (-3.75) 0.370 (1.19) 0.358 (1.01) Current asset (billion) -0.015**
(-4.83) 0.014*** (-4.68) -0.051** (-3.02) -0.046** (-2.88)
Industry effect No Yes No Yes
Year effect No Yes No Yes
Constant 2.049 2.591 2.130 2.904
Adjusted R2 0.046 0.200 0.020 0.157
28 Tobin’s Q. In our hypothesis, we predict a quadratic relationship between CEO’s overconfidence and firm value. We therefore test our hypothesis in table 5. According to the table, when base on the compensation-based measurement, both the confidence and the squared value of confidence are significant. This indicates a quadratic relationship between overconfidence and firm value. The coefficient of CEO’s square overconfidence is negative (-0.055) and the significance level 0.05. After controlling for industry and year effect, the relationship between CEO’s square overconfidence and R&D investment remains significant and negative, the p value remains in 0.05 levels and the coefficient equals -0.058. The table also shows a positive coefficient for confidence (coefficient=0.457, p<0.01). After controlling for industry and year effect, the coefficient of overconfidence remains positive and significant (coefficient=0.463, p<0.01). Thus, we can conclude that the firm’s value (reflected by Tobin’s Q) increases with the CEO’s confidence. However, for large values of confidence, the Tobin’s Q decreases with the CEO’s confidence. There seems to exist an inverse U relationship between confidence and firm value.
For the forecast-based measurement, however, the squared value of the confidence measure is not significant. After control the total asset growth rate, current asset, and asset liability ratio, the coefficient remains insignificant. Meanwhile, we can find that CEO’s overconfidence is significantly negatively correlated to Tobin’s Q. However, after adding the time effect and fix effect, the relationship becomes insignificant (coefficient=-0.148, p>0.1). Therefore, we can conclude that for the forecast-based measure no quadratic relationship between confidence and firm value is found. Although by the relative-compensation measurement, the quadratic relationship is significant, we still cannot draw any conclusion about hypothesis 3. The final step is to test how controlling for R&D expenditures influences the relationship between confidence and firm value. For the relative-compensation measure the results are presented in Table 6. Since no quadratic relationship was found for the forecast based measure, we then consider fort his measure a linear relationship between CEO confidence and Tobin’s Q. The results for the forecast based measure are presented in Table 7.
29 Table 6
R&D Expenditure Mediating Effect in Confidence (Compensation)
The table provides the quadratic regression result of firm value and CEO’s confidence (Compensation). The sample includes the Chinese listed innovation industry companies during the year 2010 to 2014. Confidence (Compensation) is a proxy of CEO’s overconfidence measured by CEO’s relative compensation compares to the compensation of second highest officer. Square
Confidence (Compensation) is the quadratic term of Confidence (Compensation). The numbers in the
brackets under the coefficient value is the t value. *, ** and *** measure significance at the 10%, 5% and 1% level, respectively.
Dependent variable = Tobin’s Q
(1) (2) Confidence (Compensation) 0.455*** (2.57) 0.466*** (2.82) Square confidence (Compensation) -0.054** (-2.21) -0.058*** (-2.51) R&D (million) 0.000 (0.14) -0.000 (-0.26) Total asset growth rate 0.056
(0.69)
-0.027 (-0.36) Asset liability ratio -0.661**
(-3.23) -0.746*** (-4.25) Current asset -0.015** (-4.55) -0.014*** (-4.25)
Industry effect No Yes
Year effect No Yes
Constant 2.052 6.855
Adjusted R2 0.046 0.190
Table 6 tests the mediating effect of R&D expenditure in the relationship of firm value and CEO’s confidence base on relative compensation measure. After adding R&D expenditure into our model (1) and (2), we can find there are still significant relationship between CEO’s overconfidence (compensation-based) and Tobin’s Q. Consistent with the result in table 5, the coefficient of square confidence (compensation) is negative and significant both without controlling industry and year effect (coefficient=-0.054, p<0.05) and with controlling
30 industry and year effect (coefficient=-0.059, p<0.01). Moreover, the coefficient of CEO’s confidence is also significantly positive related to Tobin’s Q (coefficient=0.456, p<0.01) even after controlling the industry and year effect (coefficient=0.473, p<0.01). However, since the coefficient of R&D expenditure is 0 and not significant, we cannot conclude that the R&D is the mediator of the relationship between CEO’s overconfidence (compensation-based measure) and Tobin’s Q.
Table 7
R&D Expenditure Mediating Effect in Confidence (Forecast)
The table provides the linear regression result of firm value and CEO’s confidence (Forecast). The sample includes the Chinese listed innovation industry companies during the year 2010 to 2014.
Confidence (Forecast) is a proxy of CEO’s overconfidence measured by CEO’s relative
compensation compares to the compensation of second highest officer. The numbers in the brackets under the coefficient value is the t value. *, ** and *** measure significance at the 10%, 5% and 1% level, respectively.
Dependent variable = Tobin’s Q
(1) (2) (3) (4) Confidence (Forecast bias) -0.202* (-1.81) -0.113* (-1.68) -0.400** (-3.12) -0.239* (-1.91) R&D (million) 0.002** (3.07) 0.001* (1.91) Total asset growth rate 0.048
(0.39) 0.022 (0.18) 0.048 (0.39) 0.024 (0.21) Asset liability ratio 0.356
(1.14) 0.346 (0.98) 0.572* (1.81) 0.468 (1.30) Current asset (billion) -0.550***
(-3.39) -0.047*** (-3.03) -0.135*** (-4.42) -0.098*** (-3.21)
Industry effect No Yes No Yes
Year effect No Yes No Yes
Constant 2.100 3.076 2.110 3.080
Adjusted R2 0.020 0.159 0.041 0.165
31 forecast base measure and firm value. In the first column, we apply a linear regression on Tobin’s Q and CEO’s overconfidence based on forecast measure. Due to the insignificance of the quadratic relationship between CEO’s overconfidence and Tobin’s Q based on the forecast-based measurement, we exclude the variable of square of CEO’s overconfidence. The results turn out to be significant. Specifically, through table 10, we can figure out that there is a negative relationship between CEO’s overconfidence and Tobin’s Q (coefficient=-0.202, p<0.1), after control the fixed variables include industries and year, the relationship remain significant but the p value has increased (coefficient=-0.113, p<0.1), which means our empirical results are robust by the fixed effect analysis. After we add R&D investment the relationship between CEO’s overconfidence and Tobin’s Q remains significant and becomes stronger. After control the industries and years, the significant level reduce to 0.05 and the coefficient equals to -0.286. However, the coefficient of R&D expenditure is significant (coefficient=0.002, p<0.05), even after controlling the industry and year effect. Therefore, we find that introducing R&D expenditure under forecast-based measure influences the relationship between confidence and firm value. However, we cannot conclude that R&D expenditure is mediating the relationship. The reason is that R&D expenditures mediates the relationship if by introducing R&D expenditures the relationship between overconfidence and firm value becomes weaker than when R&D expenditures where not included. This is not what is reflected in Table 7. After introducing R&D expenditures the coefficient of confidence becomes larger in absolute terms. Without industry and year affect the absolute value of the coefficient increases from 0.202 to 0.400, and with industry and year effects, the absolute value of the coefficient increases from 0.113 to 0.239.
5. Discussion
This paper investigates the relationship between CEO’s overconfidence and firm value in Chinese innovation listed company during the year 2010 to 2014. Although some results are found to be significant and consistent with our hypothesis, there are some problems that might affect the results of this study. These problems might explain the inconsistence of the
32 results between two measures as well as the result between our hypotheses. This section describes a variety of problems that might affect the result of the study.
The first problem is the problem of proxies. The proxies used in this study cannot perfectly measure CEO’s confidence. Since data unavailability in China, we do not apply the most widely used proxies developed by Malmendier and Tate (2005a,b; 2008). The proxies developed by Malmendier and Tate have been used in several studies Failing to use their proxies might leads to lower accuracy of measure. Furthermore, the low correlation between compensation-based measure and forecast-based measure of confidence indicates that these two proxies are not measuring the same thing.
We measure the CEO’s confidence by their relative compensation. Relative compensation can affect CEO’s self-importance (Hayward and Hambrick, 1997), which leads to CEO’s confidence. However, this measure might also capture some other factors, for example, CEO’s preference of equity. Since individuals are mostly inequity averse (Carlsson et al, 2005; Schwarze and Harpfer, 2002), they do not want to receive unequal earning when compares to others. CEOs might decrease the inequity of salary when they are inequity averse. Therefore, relative compensation might in certain extent shows the level of CEO’s inequity aversion.
If we assume Chinese CEOs are mostly inequity averse as normal people, then most of CEOs would not like to receive a salary much higher than the salary of second highest officer. Even an overconfident CEO have high self-importance, he or she might receive a salary not much higher than the salary of second highest officer. Therefore, inequity aversion might lead to lower relative compensation in our sample.
Previous summary statistic (Table 1, Panel A) shows that the variable the maximum value of Confidence (Compensation) is much higher than its mean (mean=1.29; max=10). When taking away the maximum value (Confidence=10), we find the new maximum value is 4.912. This indicates an outlier exist in our tests (Outlier: confidence=10). To test whether the
