Table of contents
List of Figures
List of Tables
List of Abbreviations
1.1 Purpose of this Paper
1.2 Course of Investigation
2 Concept of Overconfidence
2.1 Theoretical Concept
2.2 Managerial Overconfidence
3 Sources of Overconfidence
3.1 Principal Sources
3.2 Influencing Factors
3.2.4 Cultural Background
3.2.5 Task Familiarity and Performance
3.2.7 Importance of a Topic
3.2.8 Mental Condition
3.2.9 Situational Factors
4 Implications of Overconfidence
4.1 Better-Than-Average Effect
4.2 Implications for Finance
4.2.1 Implications for Security Trading
4.2.2 Implications for M&A
4.2.3 Implications for Corporate Investment
4.3 Implications for Entrepreneurship
4.4 Implications for Management
4.5 Implications for Economics
5 Motivation for my Research
5.1 Empirical Evidence
5.2 Experimental Evidence
6 Dual Reasoning and Overconfidence
6.1 Concept of Dual Reasoning
6.2 Differences in Reasoning
6.2.1 Fixed Factors
6.3 Reasoning Systems and Overconfidence
6.4 Differences in Training and Overconfidence
7 Experimental Approach
7.2 Preliminary Considerations
7.3 Experimental Setup
7.4 Additional Testing
7.5 Interpretation of Results
List of Figures
Figure 1: Academic Education of German DAX-Company CEOs
Figure 2: Distribution of Confidence Level
Figure 3: Mean Confidence Levels by Field of Studies
List of Tables
Table 1: Average Overconfidence of German CEOs According to Press Portrayal
Table 2: Output from the Regression
Table 3: Correlations of Independent Variables
Table 4: Labels Attached to Reasoning Systems
List of Abbreviations
Abbildung in dieser Leseprobe nicht enthalten
Chapter 1 Introduction
Many models in business and economics are based on the assumption that agents are “rational”. Barberis and Thaler (2003) propose two characteristics of what a “rational” agent is. The first one is that the agents update their beliefs according to Bayes’ law when they receive new information. The other characteristic is that the agent’s choices are consistent with Savage’s notion of Subjective Expected Utility. Countless experiments and observations in the field show, however, that these seemingly obvious characteristics are far from natural for humans.
Daniel Kahneman received the Nobel Memorial Prize in Economic Sciences in 2002 "for having integrated insights from psychological research into economic science, especially concerning human judgment and decision-making under uncertainty" (Nobel Foundation, 2008) He and many other prominent researchers have studied what assumptions of standard economic and financial models tend to be violated by actual human behavior, how those models could be adapted and, most importantly, why humans behave irrationally in the first place.
Looking for explanations of individual economic behavior in psychology is nothing new and can be traced back as far as to Adam Smith’s The Theory of Moral Sentiments (1759). Only during the last 50 years, however, have researchers like Kahneman, Tversky or Thaler begun to differentiate between individual cognitive biases and to study their impact on human behavior. Psychology has thereby gained a much higher relative value in economics and business and the idea of bound rationality has become a grave challenge for existing economic models.
1.1 Purpose of this Thesis
De Bondt and Thaler (1995) claim that “perhaps the most robust finding in the psychology of judgment is that people are overconfident”. In literature overconfidence has been blamed for economic bubbles and crises (Scheinkman and Xiong, 2003) as well as for international conflicts and wars (Johnson, 2004). Consequently, much research was conducted on the bias termed “overconfidence”, its roots and effects. These studies showed that overconfidence plays an important role in various fields like economics, finance, management or entrepreneurship. Understandably factors influencing an individual’s level of overconfidence have been of particular interest for researchers. In this context research has identified many factors, some more and some less amenable to influence. The question is therefore whether the factors found so far are conclusive or whether other factors of influence also exist.
The purpose of this thesis is to investigate this very question and to argue in favor of an individual’s major field of study as a factor of influence not studied yet. I propose that the field in which an individual has been educated in the sense of whether it is rather quantitatively or rather qualitatively oriented influences a person’s mode of thought and there over influences her proneness to overconfidence. While education has already been shown to impact one’s level of overconfidence previous research focused on the length and profoundness of education. The thesis at hand, however, examines the connection between overconfidence and the field in which a person has been educated. The issues covered are therefore how education and mind set are related, why a differentiation between “quantitative” and “qualitative” education makes sense in this context, and how different mind-sets influence an individual’s proneness to overconfidence.
1.2 Course of Investigation
This thesis is divided into two main parts: a literature review of the concept of overconfidence and an argumentative part where I propose quantitative education as a factor influencing an individual’s level of overconfidence that has not been investigated so far. First I shall describe the concept of overconfidence by presenting essential definitions and introducing managerial overconfidence as a special form of overconfidence. In addition to this I will present research regarding sources of overconfidence as well as consequences of overconfident behavior in the areas of finance, management, entrepreneurship, and economics.
After the theoretical foundation has been laid, I will present the results of a previous experiment as well as results from an empirical data set of German top managers. In both cases differences in individual levels of overconfidence can be observed, that cannot be explained by factors traditionally assumed to influence overconfidence. Since the only systematic connectedness seems to be that individuals from the fields of mathematics, natural sciences, and engineering exhibit lower levels of overconfidence I propose that these differences could stem from dissimilarly strong quantitatively oriented education of the subjects.
As a possible explanation for why such a relationship might exist, I suggest that focus of education may have an influence on individual levels of overconfidence through distinct ways of reasoning that are acquired and practiced during higher education. In order to corroborate this hypothesis I present the concept of dual processes of reasoning according to which humans use two distinct cognitive systems. I support my initial hypothesis with findings form psychological research showing that people make distinct use of these reasoning systems depending on how they have been educated. I further argue that the use of one of these systems fosters overconfidence while the other one inhibits it. Therefore, an individual’s disposition towards overconfidence could be influenced by the predominant use of one of these systems. Proceeding from this I will give an outline of an experimental design in which my hypothesis could be tested. After a brief summary of my arguments I end with concluding remarks.
2 Concept of Overconfidence
The term “overconfidence” might seem difficult to define. In economic literature it is often used as an umbrella term for a variety of effects and phenomena. At its hearts seems to be the notion that people tend to be optimistic in situations of uncertainty. The concept as a whole, however, remains vague. In the following chapter I will discuss the theoretical framework of the concept of overconfidence and clarify the different aspects that are subsumed under this term. Alongside with the presentation of conceptual work on overconfidence and its appearance I shall describe managerial overconfidence as a special form of overconfidence. Additionally I will present criticism that has been expressed regarding the bias itself as well as current methods used to demonstrate it.
2.1 Theoretical Concept
Baberis and Thaler (2002) differentiate between two manifestations of overconfidence: too narrow confidence intervals and bad assessment of probabilities. This differentiation is based on results from two different experiments. The first one is a confidence interval estimation game in which subjects are asked to report an interval for which they are X% sure that a certain variable, such as the number of murders in the USA in a certain year, lies in this interval. Alpert and Raiffa (1982) found that in such a setting the true number fell in their subjects’ 98% confidence intervals only 60% of the time. The second type of experiment uses a design in which subjects have to assess the probability of a certain random event. Fischhoff, Slovic and Lichtenstein (1977) found that events which subjects deemed certain occurred only with an 80% probability while events which the subjects considered impossible still occurred with approximately 20 % probability.
These two aspects are revisited by Ben-David, Graham and Harvey (2010). In an empirical analysis of stock market forecasts made by U.S. financial executives, the authors find that managers use too narrow confidence intervals while underestimating the element of chance. When asked to estimate future stock market returns, only 39% of the true returns fell into the executives’ forecasted 80% confidence intervals. Calling this finding a “miscalibration”, the authors equated the combination of the two aspects addressed by Baberis and Thaler with overconfidence.
The concept of miscalibration is also taken up by Englmaier (2007) in his review of overconfidence literature. While Ben-David, Graham and Harvey use the term as synonym with overconfidence, Englmaier refers to it as just one possible manifestation of overconfidence. Englmaier proposes that, apart from miscalibration, there are three other main outward forms of overconfidence, namely self-serving bias, illusion of control, and overoptimism.
Self-serving bias refers to people’s tendency to attribute success to their own abilities while blaming failure on external circumstances (Miller and Ross, 1975). The authors argue that people who expect a certain outcome tend to overestimate their influence on the outcome once it occurs. Fischhoff (1982) takes up this notion of people’s self-serving tendency and suggests that, after the occurrence of an uncertain outcome individuals forget their original forecasts and adjust their memory to fit the actual outcome. Furthermore, according to Fischhoff, people retrospectively exaggerate how certain they were about the outcome in the beginning.
Illusion of control, the second manifestation of overconfidence according to Englmaier, is the notion that individuals are often mistaken about the extent to which they can control future events while underestimating chance (Langer, 1975). In Langer’s original series of experiments, subjects perceived their likelihood of winning in a lottery to be higher when they could pick the numbers themselves rather than being given numbers by an external source. Further experimental evidence is presented by Allan and Jenkins (1980) from experiments in which subjects could press or not press a button resulting in a light flash or not. The authors conducted the experiment with two groups of subjects. One group could technically determine by a certain probability the reaction of the light by their decision to press or not press the button, while for the other group the light’s reaction was entirely random. Even though subjects in the second group had no influence on the light they still reported to have the impression that they were the ones controlling the light.
Overoptimism, which is the fourth outward appearance of overconfidence according to Englmaier, refers to people’s tendency to be overly optimistic about the outcome of their actions. This often goes along with underestimation of the likelihood of unfavorable outcomes. One of the first studies to approach overoptimism is Marks (1951). In a simple card drawing game with nine to twelve-year old children, Marks found that subjects’ reported expectations of the outcomes of the (random) card drawings were influenced by the desirability of the outcome. Irwin (1953) confirmed this finding for adults by showing that even grown-up subjects believed the likelihood of a random card drawing to be higher when a certain card was favorable and lower when it was unfavorable. Weinstein (1980) also showed that people believe that good things will happen to them more often than to others.
If self-serving bias is formulated as people’s tendency to overestimate their abilities, then miscalibration could be one possible manifestation of such an overestimation. The combination of these two forms of overconfidence makes Englmaier’s four forms of overconfidence identical to the three positive illusions leading to an unrealistically positive attitude towards oneself described by Taylor and Brown (1988) as well as Kruger et al. (2009).
2.2 Managerial Overconfidence
The term managerial overconfidence is generally used without explicit definition in research when referring to overconfident behavior of individuals in managing positions and its consequences. Nicolosi (2006) explicitly questions whether corporate executives can be expected to fall prey to the same cognitive bias as experimental subjects and individual investors. Using financial data from US companies, the author investigates this question and finds that even if individual executives exhibit behavior that is associated with overconfidence, companies and markets account for this and the resulting effects are insignificant.
Contrary to Nicolosi, most studies on the matter of overconfidence do not make a distinction between “ordinary” experimental subjects and corporate executives. Moreover, there are numerous studies showing that managers do exhibit overconfidence (e.g. Malmendier and Tate, 2005a; Malmendier and Tate, 2005b; Ben-David et al., 2006). Consequently, the question rises as to why overconfidence of corporate executives has been granted so much attention in financial literature.
Statistical regressions with data from corporate policies and individual CEO characteristic coefficients have shown that individual characteristics of top-level executives are significantly related to corporate decisions (Bertrand and Shoar, 2003). Barber and Odean (2001) sum up that overconfidence is greatest for tasks that are difficult, when making forecasts where the outcome has a low predictability, and during undertakings that lack fast and clear feedback.
All these characteristics are typically met in managerial decision environments. This means that corporate managers are, on the one hand, the most prone to overconfidence and, on the other hand, their decisions can harm the most if biased. This “double danger” makes managerial overconfidence and its consequences particularly interesting for research.
Countless studies show experimental findings, which researchers have interpreted as evidence for the cognitive bias of overconfidence. Overconfidence seems to have been an accepted fact, and research focused on the reasons for and the consequences of this bias when some researchers started calling it into question in the 1990s. Instead of the concept of overconfidence as a cognitive bias, researchers proposed rational explanations for the behavior formerly interpreted as overconfidence.
In economic literature, one common way of demonstrating overconfidence is asking subjects to answer general knowledge questions and then ask them to indicate their level of certainty of their answer’s correctness. Erev et al. (1994) criticize this method and argue that the discrepancy between accuracy and level of confidence when answering a question may also result from a regression effect. They describe the probability of a correct answer to a test question as a function of knowledge and the element of chance, since a question, particularly a multiple-choice question, can also be answered correctly by guesswork. When looking at accuracy as a function of the level of certainty because of the element of chance the authors show the existence of a regression effect. The predicted accuracy lies closer to the average which means that accuracy is relatively lower for high levels of certainty and relatively higher for low levels of certainty. This argumentation is consistent with the hard-easy effect.1 Erev and colleges’ point is revisited by Budescu et al. (1997) as well as Klayman et al. (1999). Both studies estimate the unsystematic error in their experiments in order to distinguish between unsystematic error and systematic bias. The authors find that systematic overconfidence exists, but, due to unsystematic errors, it is smaller than previously assumed.
More criticism is raised by Gigerenzer et al. (1991) and Juslin (1994) who challenge the fairness of the questions used in overconfidence-experiments. These authors claim that when choosing two-choice questions (e.g. Which city has more inhabitants? Dortmund or Dresden?), experimenters tend to take misleading ones which are answered wrong rather than right if the subject does not know the answer with certainty. According to Gigerenzer et al. the subject constructs a probabilistic mental model when being confronted with such a question. If they do not know the correct answer with certainty people consider general cues like “has an airport”, “famous sights” or “large companies’ headquarters” in order to assess the cities’ sizes. The criticism is that researchers often choose pairs for which the subject would answer wrongly using such cues and therefore mislead subjects. In the aforementioned example a subject would probably think Dresden is larger than Dortmund because it is much more famous in terms of architecture and culture and has an international airport. Dortmund would, however, be the correct answer as it has around 581,000 inhabitants while Dresden has only 530,000. In an experiment Gigerenzer and his colleges showed that overconfidence disappears if test-questions are chosen randomly and representatively. This argument was revisited by Brenner et al. (1996) and again by Klayman et al. (1999). Neither study could replicate the results of Gigerenzer et al.
A very basic point is furthermore made by Svenson (1981), who addresses one of the classical ways to demonstrate overconfidence, namely to ask subjects if they are better than the average in a certain area (e.g. car driving). In most studies that use this method, significantly more than 50% of the subjects place themselves as better than average which is logically impossible and commonly interpreted as overconfidence.2 Svenson points out that different people may have different definitions of being skillful in the areas studied (e.g. car driving skills could be measured by the number of traffic warrants received or the number of accidents had or prudence perceived by other drivers and so on) and, therefore, theoretically all the subjects assessing themselves as above average could be in fact merely mean above average as according to their personal definition.
Benoît and Dubra (2011) take up this method of demonstrating overconfidence and challenge the notion that people’s tendency to regard themselves as better than average automatically indicates overconfidence, as it is mostly found in economic literature. The authors argue that (wrongly) placing oneself above the average based on performance in a certain task can be perfectly Bayes-rational if one does not have any contradicting information available. Benoît and Dubra use the example of car drivers of low, middle, and high skill. Given that the probability of a low-skilled driver to have an accident is the highest and the probability of a high-skilled driver is the lowest a driver who has never before had an accident would rationally place herself as high-skilled. This, however, might be wrong and the real reason she has never had an accident was pure luck. Nevertheless the authors do not challenge the notion of overconfidence itself but rather question the validity of certain experimental evidence.
The criticism presented in this chapter is appropriate and benefits research on overconfidence. Challenging certain methods of demonstrating overconfidence allows for a more valid analysis and interpretation of experimental and empirical data. It stresses the importance of clear differentiation between rational, albeit seemingly odd behavior and the real irrational bias of overconfidence. In the following chapter I will examine the concept of overconfidence more closely by presenting research on the sources of this bias as well as factors influencing its appearance.
3 Sources of Overconfidence
Some authors propose explanations for such “irrational” behavior as overconfidence deliberately excluding systematic biases from human judgment. For example Stanovic and West (2000) see performance errors, human computational limitations, wrong norms being applied by experimenters and differences in the comprehension of a task by subjects as the four main reasons. In this thesis, however, I follow the more prominent school of thought which regards overconfidence as a cognitive bias keeping while in mind the critical points discussed in the previous chapter. In this chapter, I will first explore the reasons for overconfident behavior examined so far and in a second step review studies on factors influencing the extent to which people show overconfidence.
3.1 Principal Sources
One explanation for the origins of overconfidence is presented by Griffin and Tversky (1992). According to them, the reason why people are overconfident is their biased mode of processing information. The authors divide information into two elements, strength and weight. By strength they refer to the potential importance of information, such as the magnitude of the consequences of an event. By weight they refer to the credibility or reliability of the information. Griffin and Tversky claim that overconfidence is the consequence of a focus on strength of information while its weight is being adjusted. If, on the contrary, a person emphasizes weight while subordinating the strength of information, the result is underconfidence.
A slightly different approach is taken by Koriat et al. (1980) and Klayman and Ha (1989). They also blame people’s assessment of information in a decision-making process for overconfidence but suggest that the main reason is people’s selective perception of evidence. In their eyes people tend to quickly draw conclusions and then focus on evidence supporting it while irrationally neglecting any information to the contrary. Similarly McKenzie (1997) argues that overconfidence occurs because people focus on evidence for and against a quickly drawn first conclusion while ignoring any alternative conclusions with information supporting them.
Both approaches are based on the idea that once a decision is made people tend to hold on to it and are unlikely to change it even if there is evidence against their decision. This argumentation is supported by cognitive dissonance theory. Knox and Inkster (1968) conducted an experiment in which they asked people before and right after they placed their bets in a horse race about how certain they were that their horse would win. The results showed that right after the decision was made the better’s confidence level rose, which the authors interpreted as post-decisional dissonance reduction. Similarly, Frenkel and Doob (1976) found that voters’ confidence that their preferred candidate would win rose right after the casting of their vote. In both studies subjects reduced emotional discomfort caused by imagined failure through mentally adjusting the probability of a loss. Both studies dealt with decisions that are ultimate and cannot be changed. Koriat et al. (1980), Klayman and Ha (1989) as well as McKenzie (1997) confirmed these findings for interlocutory decisions in a judgmental process and showed that post-decisional dissonance reduction could still well occur on this micro-decision level.
Another psychological factor that evokes overconfidence is an individual’s self-image (Blanton et al., 2001). According to Blanton et al. every individual has a desire to see herself as competent and knowledgeable. This self-image hinders self-doubt and leads to an irrational confidence in one’s own decisions. Fiske and Taylor (1991) add that confident behavior is often perceived as an indication of competence. Therefore, even if an individual has an accurate self-assessment, she might want to cast herself in a better light, which leads to underassessment of risks and increased confidence.
Russo and Schoemaker (1992) present three more generalized reasons for overconfidence: cognitive bias, physiological causes and motivational factors. Furthermore, they subdivide the element of cognitive bias into availability bias, anchoring bias, confirmation bias, and hindsight.
Availability bias refers to a concept introduced by Tversky and Kahneman (1973). They claim that people assess the likelihood of an event depending on how easily they can remember similar outcomes. An example for this judgmental heuristic would be that many people highly overestimate the number of shark attacks because the picture of a shark attacking a human comes to mind very easily due to horror movies and excessive news coverage of singular events. Judgments based on this heuristic are often misleading and can promote overconfidence as probabilities are assessed wrongly.
The second heuristic that misleads people and thus promotes overconfidence is anchoring bias. This concept introduced by Tversky and Kahneman (1974) refers to the tendency of people to rely irrationally heavily on single pieces of information during a decision-making process. Russo and Schoemaker presented questions to subjects, in which a number (e.g. the length of the river Nile) had to be guessed. One group had to make a best guess first and then come up with a 90% confidence interval around this number. This resulted in the subjects taking their first guess as an anchor point and then choosing too narrow confidence intervals. The other group of subjects was asked to immediately provide a confidence range without making a first guess. As a consequence they choose more accurate confidence intervals and thus showing less overconfidence.
The concept of confirmation bias, which Russo and Schoemaker see as a third bias leading to overconfidence, refers to people’s tendency to seek confirming evidence once a first prediction has been made while ignoring any disconfirming evidence. Russo and Schoemaker integrate the ideas of Koriat et al. (1980) and Klayman and Ha (1989) described above. As in previous studies, Russo and Schoemaker raise awareness for the consequences of wrong assessment of arguments used in a decision-making process that result from confirmation bias.
The fourth and last cognitive bias Russo and Schoemaker propose to be responsible for overconfidence is hindsight, which is the illusion of predictability of an outcome in retrospect. First investigated by Fischoff and Beyth (1975) hindsight bias is accountable for people’s impression that many uncertain events are much more predictable than they really are. Russo and Schoemaker consider this false assessment of chance an important reason for wrong assessment of own forecasting accuracy and thus an unduly confident behavior.
In addition to the mental causes presented above Russo and Schoemaker point at biochemical processes as a possible physiological explanation for overconfidence. The authors claim that hormones, such as adrenalin and endorphins, produced in the human body as a response to strong emotional reactions, lead to exaggerated self-confidence. Russo and Schoemaker draw a parallel to the effects of alcohol: just like alcohol inhibits a person’s response time and thus one should not drive under the influence of alcohol, strong emotions facilitate overconfidence and therefore one should not make decisions while in an emotional agitation.
As a third factor, the authors propose the vital motivational effect of confident behavior. Russo and Schoemaker point at the psychological finding that optimism has motivational value. According to them, an individual’s performance in a certain task highly depends upon her motivation for success. A confident, and thus easily overconfident, attitude is indispensable for motivation. The authors quote German poet Johann Wolfgang von Goethe, who wrote, “for a man to achieve all that is demanded of him he must regard himself as greater than he is.” Furthermore, the authors argue that people often equate confidence with competence. Hence, people cannot afford to appear unconfident given they do not want to be considered incompetent.
3.2 Influencing Factors
The concepts presented above provide possible reasons for overconfidence. However, the occurrence of overconfident behavior also depends upon several factors impacting the probability that an individual will fall prey to overconfidence as well as the extent to which she will differ from a rational view. In the following I will more closely examine such factors and current literature on how fostering or inhibiting those factors seem to be.
In a study of university students who had to predict their own performance prior to a test as well as their expected performance after the test Grimes (2002) found that although the students were relatively homogenous with respect to age (average age of 20.5 years; standard deviation of less than 2.2 years), age was a statistically highly significant factor for the level of overconfidence. While younger students showed a high level of overconfidence, older students were much more accurate in their self- assessment. Similar results are presented by Bertrand and Schoar (2003). Studying data of large U.S. firms and their top executives the authors find that older managers tend to be more conservative while younger managers take more risks and exhibit greater confidence.
However, when studying age as a factor influencing overconfidence, it is difficult to separate age from experience. Malmendier and Nagel (2010) studied individuals who experienced economic depressions and booms in their life to see whether macroeconomic experience influences risk-taking behavior. They found that people who had experienced a great depression were less willing to take financial risks, while individuals who mainly experienced high stock returns were less averse to risk. If the degree of risk aversion which is often linked to the extent of overconfidence is influenced by events like the great depression in the 1930s or the “Wirtschaftswunder” (economic boom) in Germany in the 1950s, this is a serious factor of distortion when correlation between age and overconfidence is studied.
When assessing education as a factor influencing the level of overconfidence, two contradictory effects might come to mind: On one hand the better a person is educated the more competent she becomes and the higher her accuracy can be expected to be. On the other hand, education, especially for young people just out of school, could give an individual an illusion of safety and lead to overestimation of actual abilities. While both possibilities are realistic, most psychological literature supports the second one.
One example of education hampering overconfidence is discussed by Lundeberg et al. (1994). In an experimental bidding game with male and female undergraduate and graduate students, the authors find that undergraduate students, especially male ones, exhibit more overconfidence than graduate students in the study.
Contrary to that, Graham et al. (2009) find that their constructed measure of investors’ self- perceived competence and thus their overconfidence is highly positively correlated with the investors’ undergraduate and graduate education. This is supported by Mayhew and Simpson (2002), who argue that, in the context of driving safety, special drivers’ education in emergency maneuvers and collision avoidance techniques fosters the drivers’ overconfidence. Thus, paradoxically, special training increases the risk of an accident rather than reducing it.
The hypothesis that education fosters overconfidence rather than decreasing it is also supported by empirical evidence from professionals in the field of finance. In a study of U.S. CFOs’ forecasts about stock market returns as well as the corresponding confidence intervals given by the CFOs, Ben-David et al. (2006) find that the better the CFOs were educated, the more they exhibited overconfidence as measured by miscalibration of their forecasts.
Stereotypically men are often considered to show more overconfidence than women. There is a lot of empirical and experimental evidence suggesting that this is indeed the case. According to Nowell and Alston (2007), male students exhibit greater overconfidence than their female fellow students when asked to predict their grade at completion of a university course. Similarly, Soll and Klayman (2004) show in a classical year guessing experiment with confidence intervals, that male subjects overestimate their accuracy more than female subjects.
Similar evidence also exists for business-related decisions. In a Gallup poll for PaineWebber with approximately 15,000 between 1998 and 2000 individual investors were asked to predict the total market return as well as the return of their own portfolio for the next twelve months. According to Barber and Odean (2001) as well as Graham et al. (2005), this data shows that both, men and women, on average predicted to outperform the market, but men did so significantly more.3 Combining this data with the idea that overconfidence leads to excessive trading and ultimately to underperformance, Barber and Odean (2001) analyzed trading data of male and female individual investors.4 Their results show that male investors trade more than their female counterparts and perform less well, which again confirms the assumption that men show more overconfidence than women when it comes to financial decisions.
However, gender differences in overconfidence seem to depend on the task studied. Lundeberg et al. (1994) argue that gender differences can be found mainly in areas perceived to be masculine domains. In accordance with this theory, Lichtenstein and Fischhoff (1981) did not find any gender differences for the levels of overconfidence for general knowledge questions. Dunning et al. (2003) found that female subjects who had to evaluate their performance on typically male-oriented tasks rated themselves worse than even those male subjects who exhibited underconfidence.
3.2.4 Cultural Background
The consensus among researchers seems to be that cultural background does influence the level of overconfidence although it is disputed how exactly it does so and what the reasons for this influence could be. In a groundbreaking study Wright et al. (1978) asked British and Hong-Kong subjects to answer multiple-choice questions while stating how certain they were about their answers. The researchers obtained results that served as the basis for much further research. Both groups of subjects were overconfident about the correctness of their answers, but the Asian subjects were significantly more overconfident than their Western counterparts.
Yates et al. (1996) as well as Yates et al. (1997) offer two different explanations for why Asians might be more overconfident than individuals from Western cultures. In their first paper the authors offer the differences in education systems as one explanation. According to them in Western cultures students are generally encouraged to think critically whereas in Asian cultures tradition plays a large role and viewpoints of others as well as oneself are challenged much less. This hypothesis is supported by evidence form an overconfidence comparison study between Chinese students from Fujian and Chinese students from Singapore. Li et al. (2006) conducted an experiment to investigate whether Chinese students would also exhibit more overconfidence in comparison to students from Singapore, even though the Singaporean students were descendants from Fujian-Chinese and therefore shared a common culture and cultural heritage. The authors could thereby more or less eliminate other cultural factors and isolate the effects of different educational systems. The results confirmed the authors’ hypothesis that Singaporean students educated based on Western standards were less overconfident than the Chinese students.
In Yates et al. (1997) the difference in expression of convincement is added as another possible explanation for different levels of overconfidence. Since in Asian cultures the saving of face is crucial, individuals from Asian cultures often express certainty about something they do not know because admitting to not knowing something would result in a loss of face.
A different approach is taken by Weber and Hsee (1998). These researchers studied transaction prices of financial options and showed through experiments that Chinese subjects underestimated risks in this field more strongly than American subjects. The authors offer the explanation that because of Chinese collectivism and their culture of solidarity, risks are perceived as lower than they would be in American culture where risks are generally borne individually. This hypothesis is tested in Hsee and Weber (1999). The authors studied riskaversion of Chinese and American subjects in several different areas. The results showed that a significant difference in risk perception can be found only when it comes to financial risks, but not in academic or medical situations. With financial losses, Chinese society might step in to help the individual. In academia and medicine, however, risks are normally borne individually in both cultures. This confirms the authors’ hypothesis that the nature of risk bearing is the reason for the observed cultural differences.
3.2.5 Task Familiarity and Performance
It is obvious that competence and experience in a certain task increase accuracy. The question, however, of how they affect confidence has been studied with very different results. Langer (1975) studied the impact of familiarity with a setting on overconfidence. He gave some subjects lottery tickets which were familiar to them, while another subject group received unfamiliar ones. The result was that even perceived familiarity with a setting (winning still depended on pure chance as all tickets had the same likelihood of winning) increased the illusion of control over the outcome and thus overconfidence of subjects. These findings suggest that experience and familiarity with a task generally increase overconfidence as individuals feel saver and perhaps more daring in familiar environments.
The notion that task familiarity fosters overconfidence while unknown and difficult tasks decrease it is mentioned by Lichtenstein, Fischhoff and Phillips (1980) as the hard-easy- effect, meaning that with hard tasks, typically, underconfident behavior can be observed and for easy tasks mainly overconfidence occurs.
This is also supported by Hoelzl and Rustichini (2005), who had their subjects take a vocabulary test with easy and difficult words. Afterwards, subjects could vote whether the pay-off for the experiment should depend upon the relative performance of each subject (the best 50% would receive a payment of about 10 USD) or based on a lottery in which a randomly determined 50% of the subjects would receive the payoff of 10 USD. The vast majority of subjects solving the easy test voted for the payoff to depend upon relative performance, while most subjects solving the difficult test preferred the lottery. The authors interpret this behavior as the subjects confusing “being good” with “being better”. Subjects solving the easy test anticipated a good absolute performance and transferred their good feeling to relative performance. Thus familiarity with the vocabulary tested fostered overconfidence while dealing with unfamiliar words and thus an unfamiliar task had the opposite effect.
Contrary to these results, Oskamp (1965) finds in a study of clinical judgment that while both, professional psychologists and psychology students, exhibit overconfidence in their judgment professionals with years of experience do so less than inexperienced students. Ben-David et al. (2006) find in their CFO stock market prediction study that, on top of education as a factor that increases miscalibration, overconfidence is negatively correlated with professional experience.
A possible explanation for these contradictory findings is given by Gervais and Odean (2001). They propose that the effect of experience and familiarity on overconfidence cannot be determined generally but depends upon the degree of experience. Early in an individual’s career she has acquired little experience while having the illusion of competence and experience. At this stage, overconfidence is increased by experience. Later, in a professional career, the person might have gained more insight and learned to account for overconfidence, so that experience then decreases overconfidence.
The studies presented above generally show that experience does have an effect upon overconfidence, while questioning whether this effect is positive or negative. Nevertheless some renowned studies found that experience has no effect on overconfidence at all. Both, Lichtenstein and Fischhoff (1977) and Allen and Evans (2005) could not show a statistically significant correlation between experience and overconfidence using data from an experimental bidding game.
One reason why feedback may reduce overconfidence is that with increasing quantity and quality of available information, accuracy of assessment also increases. Park and Santos-Pinto (2010) compared overconfidence of poker and chess players, whom they asked to predict their relative performance before a tournament. Although both groups were found to overestimate their performance, on average the poker players’ predictions were close to random guesses while the chess players were much more accurate and less overconfident. The authors explain these results with the different roles that luck and skills play in the two games.
1 See chapter 3.2.5 for an explanation of the hard-easy effect.
2 See chapter 4.1 for a discussion of the better-than-average effect.
3 Men predicted an outperformance of the market by 2.8 percentage points while women predicted an outperformance by 2.2 percentage points. (t = 3.3).
4 See chapter 4.2.1 for further discussion.