Master Thesis Msc. Finance
Willem T. Delen Rijksuniversiteit Groningen
Faculty of Economics MSc Finance
Supervisor: Prof. Dr. R v. d. Meer
Overconfidence: Charting this Bias in a Taxonomy.
Some Evidence of Excessive Optimism in the Dutch Stock Market 1986-2015.
Key words: AEX, Behavioural Finance, Excessive Optimism, Overconfidence, Taxonomy,
Table of contents Abstract 3 1. Introduction 4 o 1.1 Research question 5 o 1.2 Overconfidence dimensions 6 2. Literature review 9 o 2.1 Overconfidence 9 o 2.2 Taxonomy 13 o 2.3 Miscalibration 16 o 2.4 Better-than-average effect 18 o 2.5 Illusion of control 19 o 2.6 Excessive optimism 19 o 2.7 Excessive trading 21
o 2.8 Dutch market and financial bubbles 22
3. Methodology and data 24
o 3.1 Data 24
o 3.2 AEX-Index 25
o 3.3 Consumer confidence index 26
o 3.4 Economic climate 27 o 3.5 Unemployment rate 27 o 3.6 GDP-growth rate 28 o 3.7 Budget deficit 29 o 3.8 Inflation rate 29 4. Results 30
o 4.1 Economic climate measurement 30
o 4.2 Autocorrelation 32
o 4.3 Multicollinearity 33
o 4.4 Overconfidence measurement 33
5. Conclusion 37
6. Limitations and further research 39
Appendix A, figures 40
Appendix B, tables 44
Appendix C, literature summaries 51
References 52
Datasets 57
List of figures and tables
o Figure 2: Miscalibration according to confidence interval and actual results. 17
o Table 1: OLS of AR01 and CI. 18
o Table 2: OLS of exponential AR01 and CI. 18
o Table 3: AR of EC and relevant variables. 30
o Table 4: AR of EC and relevant variables with lags. 31
o Figure 3: ECt and relevant variables URt, BDt, IRt, and GDPt.. 33
o Figure 4: ECt and ÊCt. 34
o Table 5: AR of AEXt and OCt measurement. 35
o Figure 5: AEX-index. 40
o Figure 6: AEX-volumes. 40
o Figure 7: CCI-index. 41
o Figure 8: CCI-index seasonally adjusted. 41
o Figure 9: EC-index. 42
o Figure 10: EC-index seasonally adjusted. 42
o Figure 11: SP-index. 43
o Figure 12: SP-index seasonally adjusted. 43
o Table 6: Regression of EC and variables. 44
o Table 7: Regression of EC and variables. 44
o Table 8: OCt measurement on the AEX volumes. 45
o Table 9: Final model including AEX-index lag. 45
o Table 10: Regression of ECt and variables including ECt-1. 46
o Table 11: Regression of OCt with ECt-1 included on AEX. 46
o Table 12: Regression of ECt and variables without lags. 47
o Table 13: Regression of OCt without lags on AEX. 47
o Table 14: Regression of ECt and variables without lags including EC-1. 48
o Table 15: Regression of OCt with EC(-1) included on AEX. 48
o Table 16: Autocorrelation BDt-3 on URt. 49
o Table 17: Autocorrelation BDt-6, URt-3 on IRt. 49
o Table 18: Autocorrelation BDt-7,URt-4, IRt-1 on GDPt. 50
o Table 19: Autocorrelation BDt-12,URt-9, IRt-6, GDPt-5 on ECt. 50
o Table 20: Summary of articles related to miscalibration. 51
Abstract
Introduction
The purpose of this thesis is to raise awareness of the overconfidence bias and its relation to returns on the Dutch stock market. Overconfidence is an important aspect of behavioural finance as every individual is exposed to this behaviour to some extent and hence has a big influence on micro- as well as macro economy. The research will be done by investigating a large amount of available information on stock trading and economic factors in the Netherlands over the sample period 1986 - 2015.
The Dutch market is specifically interesting to investigate as it is subject to a small but well developed economy. Additionally, the stock market has actually been invented in the Netherlands and lists famous multinationals like Heineken, Royall Dutch Shell and Unilever on the Amsterdam Stock Exchange (AEX). Individuals in the Netherlands are generally considered to be conservative, with large investments in professional pension funds and participate in an individualistic culture (Donkers, Van Soest 1999). Therefore, as a global study on overconfidence found that individualistic cultures experience more overconfident behaviour, overconfidence should be prevalent in the Netherlands (Chui, Titman & Wei 2010).
overpriced, to react even more aggressive by selling excessively and lose even more faith in the economy. Studies suggest that this is also evident in the currency market where overconfident investors tend to overreact on future inflation, causing exchange rates to drop with unnecessary amounts (Burnside et al. 2011). In addition, (De Bondt 2013), also wrote about the economic crisis starting in 2008, known as the Great Recession, and the bubble that caused the economy to force into recession. He mentions the same arguments as before, which is that investors have a false perception of the economy and contributes excessive optimism as a major cause for this bubble to grow and, eventually, burst. On top of that, he states that investors throughout the decades have always demonstrated the same patterns of behaviour prior to economic distress and acknowledges the importance of confidence in the market. For these reasons, overconfidence is an interesting bias to test, as it is clearly existent, but yet a recurring problem among individuals.
1.1 Research question
The main challenge of testing overconfidence is to find a way to measure this bias. Therefore the research question for this thesis is:
Does overconfidence affect the Dutch stock market, and if so, how?
Typically, in finance, overconfidence is measured by either making an estimation of one or more of the behaviours such as:
1. Miscalibration
2. The better-than-average effect 3. Illusion of control
4. Excessive optimism
optimism is a pattern of psychological behaviour where individuals are overly optimistic about the future (Taylor, Brown 1988). A short summary of the dimensions is given next.
1.2 Overconfidence dimensions
Miscalibration
Miscalibration, used as the first indicator for overconfidence, has been studied broadly. It is stated, for example, that market returns with an 80% confidence interval are realized only 36% of the time. It is therefore said that miscalibrated executives follow more aggressive policies which result in more debt-financed investments (Ben-David, Graham & Harvey 2010). When individuals are asked about their confidence interval though, i.e. the assumption of correctness in percentage points, and they are wrong, it does not automatically mean that they used the wrong confidence interval. However, when a large sample is asked about their confidence interval, and a large amount of the outcomes appears to be outside this confidence interval, there is definitely something wrong with such assessments (Lichtenstein, Fischhoff & Phillips 1977). Calibration in this article is related to realism, whereas miscalibration can be referred to as unrealistic assumptions. While measures of overconfidence are related to trading volume, miscalibration focuses only on the confidence intervals (Glaser, Weber 2007).
Better-than-average-effect
The better-than-average effect has been analyzed in several different fields of study and is hence a well-established effect. It has been stated, for example, that 81% of new business owners expect their own company to have a 70% higher chance of succeeding, whereas only 39% of the respondents had similar expectations of a comparable company managed by someone else. These results are consistent with several other studies where respondents in, for example, a survey on college professors, 94% claimed to be above average on work-related performance (Odean 2014).
Illusion of control
demonstrate that the more personally investors are involved, the more they attribute successful decisions to their own contribution (Yarritu, Matute & Vadillo 2014). Previous studies indicate that it is not really possible to attach a number to this bias, but it has nevertheless been proven to be existent. With the illusion of control in mind it is important to learn from this bias that factors such as unpredictable events, e.g. a tsunami, are also affecting the outcome of stock indices, events of which individuals have no control whatsoever.
Excessive optimism
Excessive optimism can be explained as a bias where people overestimate the chance of good outcomes over bad ones (Clark, Friesen 2009). This means that investors mostly create positive scenarios while they tend to neglect the chance of natural disasters or economic turmoil (Chira, Adams & Thornton 2011). High past returns cause investors to increase their trading volume and overestimate economic conditions and thus become excessively optimistic about the future (Statman, Thorley & Vorkink 2006).
As it becomes clear in the literature that researchers struggle with finding a reliable measurement for overconfidence, the models which are frequently used are analysed. Additionally, an estimation of the economic climate in the Netherlands is constructed which is subtracted from the sub category of the Consumer Confidence Index (CCI) related to the economy. With this measurement, an estimation of the overconfidence bias is measured which is than analysed on the AEX-index.
Research question:
Does overconfidence affect the Dutch stock market, and if so, how?
Methodology:
(OC) = f (M , BA , IC , EO , Ɛ) Literature study on overconfidence.
Conclusion. Data collection: o AEX-Index o CCI o Unemployment Rate o GDP Growth Rate o Budget Deficit o Inflation Rate Results. (ÊCt) = (α + β1(URt) + β2(GDPt) + β3(BDt) + β4(IRt) + Ɛ) (OCt) = (ECt - ÊCt)
2. Literature review
Stock market investments appear to be one of the most profitable investments when managed properly (Jeremy J.. Siegel 2014). In his book, Stocks for the Long Run, originally published in 1994, Jeremy Siegel contributes several important findings to stock market investments. First to be mentioned, Siegel explains that average annual stock returns have been around 6.8%, after inflation, over a period of approximately 200 years. Compared to other investment vehicles, it is clear that the returns on stock holdings are considerably higher. Bonds, for example, had an average return of 1.7 percent during this period, whereas gold even had a negative average return of -0.4%. The most important finding of the book, however, is that stock investments have always had a positive return over any 20 year period.1 He describes that periods of recession are always outweighed by profitable periods, and hence acknowledges the importance of staying positive. Apart from these results, the book also states that the stock market crash from 1987 was caused by a financial bubble, whereas he does not provide strong evidence that the crash of 2008 was caused by one as well (Jeremy J.. Siegel 2014). These findings confirm the importance of investing in shares as a reliable source of income for the future. This is corroborated by the results of the AEX as mentioned above. One should bear in mind, though, that there is always a temporary turning point which can be explained by both overconfidence as well as economic leverage.
Unfortunately, economic factors influence the market whereas psychological factors impact individual behaviour and one should therefore be aware of such effects. In this section, overconfidence is defined. Specifically, it is described how this behaviour influences one’s decisions and potentially affects returns. Additionally, its appearance in literature and the evolution of this bias is analyzed by relating this to former research. The second part of the literature review covers the Dutch financial market and its characteristics.
2.1 Overconfidence
Overconfidence is the unjust bias of individuals when analyzing or predicting their own decisions. It consists of the overestimation of their own performance while neglecting factors such as chance. They typically assign successful decisions to their own performance while
1 Even when events like the Great Recession are included in a 20 year period, average returns have always been
failure is caused by other factors. In financial literature overconfidence is defined as how well individuals understand the limitations of their knowledge and their own abilities for making financial decisions (Chira, Adams & Thornton 2011). In psychology it is suggested that this behaviour is strongly related to the tendency to avoid discomforting feelings that include doubt, anxiety and fear as individuals demand certainty (Odean 2014). Psychologists have found several judgment biases, but it remains unclear which forms of behaviour affect economic decision making the most or if they affect economic behaviour at all (Glaser, Weber 2007). Therefore, this study will elaborate on the suggested psychological patterns of conduct related to overconfidence and how they potentially affect economic behaviour.
Overconfidence has always been part of the nature of mankind. Without overconfidence, people would never have expanded their borders or discovered new tools as it required the thought that things could improve. When studying literature related to overconfidence, people explain the phenomenon in all different subjects and cultures at all times (Estorick 1941). The first time overconfidence has been used as the main subject for a scientific article goes back to 1969, where Abelson wrote about overconfidence in American technology as the behaviour influencing scientific judgments (Abelson 1969). However, if we go back further, it is stated in the writing by (Pasha 1884), how naval war tactics can be influenced and distorted by this behaviour.2
In the quote from the article by (Pasha 1884) all the segments are included that makes overconfidence an interesting, but also problematic behaviour. The author explains that overconfidence neglects the possibility of chance and bad luck which should simply not be ignored, and besides regardless of how skilled an individual may be, it is never an assurance for success.
The existence of overconfidence has been proven in several studies and assigned to have a great impact on investor decisions and their results. One can suggest that a small amount of this emotion can result in great ideas turning out to be a big disappointment (Odean 2014). It
2 “Who, then, can predict what will be the result of the next naval engagement?That good sea-legs, skill at the
wheel, and presence of mind - qualities for which seamen of the Anglo-Saxon race are still so celebrated – will always tell in a struggle at sea, is not to be denied; but it cannot be kept too prominently in view that
overconfidence in such matters may lead to the most disastrous results. It should be borne in mind that the smallest rope accidentally fouling the screw, may deprive the ship of her manoeuvring power at a critical moment; or a shell penetrating the boiler from the unprotected deck above, may completely disable her.” (Pasha
is demonstrated that individual investors who hold common stock and trade very actively pay a significant penalty on their performance and show below-average results (Ivković, Sialm & Weisbenner 2008). Next to that, several interesting studies confirm the existence of overconfidence on a non-economic related level. For example, 80% of the respondents considered themselves above-average drivers which is lined to overconfidence (Svenson 1981). Other studies show similar results when individuals compare themselves to others on positive life events such as starting salaries, successful marriages or having gifted children (Weinstein 1980, Lin, Raghubir 2005). When on the other hand, people compare themselves on the likelihood of negative life events, such as being robbed or experiencing health problems, they attribute a below average chance that this will happen to them (Weinstein 1980, Harris, Hahn 2011).3
Behavioural explanations for the existence of overconfidence are described to be the result of investors overestimating the precision of their information as well as their inability to interpret this information in a correct way to make complex economic decisions (Huisman, van der Sar, Nico L & Zwinkels 2012). Studies suggest that the effort for an individual to analyze decisions objectively generate feelings of fear, which results in one’s brain trying to reduce these negative feelings by making one feel unrealistically capable and in control of one’s decisions (Odean 2014). Results of this investor behaviour are visible in an underdiversification within stock portfolios, as they overestimate the precision of their information for familiar stocks and hence invest in too many similar stocks and industries (Ivković, Sialm & Weisbenner 2008).
The above-mentioned effects have also been proved among CEO’s, who commonly demonstrate this behaviour as well.4 They typically have a preference for innovation and mostly experience underdiversification in their investment portfolios (Malmendier, Tate 2005, Galasso, Simcoe 2011). Additionally, it is said that overconfident CEO’s are 65% more likely to initiate acquisitions (Malmendier, Tate 2008).
3 Weinstein wrote several articles on excessive optimism in different subjects such as health, risk taking and
skills among jobs (Weinstein 1980, Weinstein 1987, Weinstein 1982, Weinstein, Klein 1996, Weinstein, Marcus & Moser 2005, Weinstein, Lachendro 1982, Weinstein 1984, Weinstein, Klein 1995, Weinstein 1989).
4 Galasso and Simcoe (2011), , for example, demonstrate that overconfident CEO’s are 48% more likely to
Overconfidence for this thesis is related to economic and investor decisions, as many studies demonstrate the negative impact of such behaviour on financial returns. It is stated, for example, that overconfident men reduce their net returns on average by 2.65% and overconfident women by 1.72% (Barber, Odean 2001, Deaves, Lüders & Luo 2008). Another interesting study, among twins, demonstrates that genes are a more important explanation for overconfidence than their environment (Cesarini et al. 2009).5 Several studies mention that the measure of overconfidence increases when tasks or decisions become harder (Moore, Healy 2008). An article that proves this provides an example where students were tested on overconfidence according to easy, medium and hard questions which resulted in an overconfidence measure between 30-40% for hard questions and an underconfidence measurement of approximately 2-3% for easy questions (Michailova 2010). Another study, however, contradict these results and state that overconfidence is more prevalent for easy tasks (Burson, Larrick & Klayman 2006). Participants in this study clearly demonstrated overconfidence for tasks such as the ability to use a computer mouse, while they actually showed strong underconfidence for difficult tasks such as juggling. Additionally, it appears that for some tasks, individuals generally demonstrated either over- or underconfidence, regardless of their level of expertise (Glaser, Langer & Weber 2013). It is made clear though that both over- and underconfidence have equally negative effects on returns (Kirchler, Maciejovsky 2002).
Besides these negative effects of overconfidence some studies actually attribute some positive results to this emotion (Dorn, Sengmueller 2009).6 For example, it may be of help to an individual to gain greater willpower when one has trouble to motivate oneself to achieve one’s goals (Bénabou, Tirole 2004). In addition, overconfidence regularly leads to a higher chance of success for risky activities (Compte, Postlewaite 2004). Without overconfidence Columbus would never had discovered America.
In finance there are several findings that are often summarized under the concept of overconfidence, being miscalibration, the better-than-average effect, illusion of control, and unrealistic optimism. The relation between these patterns of behaviour, however, has not been tested on a broad scale yet (Glaser, Weber 2007). In psychological research, though, only
5 Genes explain 16-34% variation in overconfidence while environmental differences explain 5-11% variation.
6 Besides explaining the advantages of entertaining yourself by participating on the stock market, this article
miscalibration is mostly used as a relevant explanation for overconfidence (Michailova 2010). The effect of these forms of conduct and their relation to financial decision making are therefore important to study and hence explained more thoroughly in this thesis.
2.2 Taxonomy
The Taxonomy used for this section, given below under figure 1, provides an overview of the articles related to overconfidence and the above-mentioned patterns of behaviour and are sorted according to the year of publication. The first thing to be noticed is that the notion of overconfidence appears in studies until 1969, but lacks any mentioning in the years thereafter, until 1996. However, the other four patterns of behaviour do appear in this period where excessive optimism is mentioned in all of these articles bar one. This is directly in line with the next finding, which is that 28 articles describe excessive optimism. Therefore, one can conclude that it is by far the most prevalent pattern of behaviour leading to overconfidence. Miscalibration, the better-than-average effect, and the illusion of control appear in 13, 16 and 11 articles respectively, which indicates that the latter is the least relevant measurement for overconfidence.
The first time an author used all four forms of conduct as a measurement for overconfidence in this Taxonomy was in 1998 (Kahneman, Riepe 1998). Thereafter this happened three more times, namely in 2004, 2005, and 2010. What immediately draws attention is that two from these articles were studied in Germany. This may indicate that European, and specifically, German studies focus on finding a reliable measurement for overconfidence. This is in contrast with the studies in the US, where most of the articles tend to focus on the existence, and the effects, of overconfidence. The Taxonomy demonstrates this, since a lot of articles from the US centre on overconfidence, without mentioning a single one of the four patterns of behaviour.
Figure 1: Taxonomy of the dimensions leading to overconfidence.
Article Market OC M BA IC EO
Progress in Naval Armament, 1884 US X
Morale in Contemporary England, 1941 UK X
Overconfidence in American Technology, 1969 US X
Unrealistic optimism about future life events. 1980 US X
Are we all less risky and more skilful than our fellow drivers? 1981 US X
X
(82%)
A progress report on the training of probability assessors, 1982
X
(50%=36%), (98%=60%)
Egocentrism as a source of unrealistic optimism, 1982 US X
Unrealistic optimism about susceptibility to health problems, 1982 US X
Why it won't happen to me: perceptions of risk factors and susceptibility. 1984 US X
Unrealistic optimism about susceptibility to health problems: Conclusions from a community-wide sample, 1987 US X
Illusion and Well-Being: A Social Psychological Perspective on Mental Health, 1988 X X X
Effects of personal experience on self-protective behavior. 1989 US X
Resistance of personal risk perceptions to debiasing interventions. 1995 US X
On the reality of cognitive illusions, 1996 US X X
Unrealistic optimism: Present and future, 1996 US X
Aspects of Investor Psychology, 1998 US X
X
(98%=80%) X X
X
(88%)
Do Investors Trade Too Much?, 2000 US X
Trading Is Hazardous to Your Wealth: The Common Stock Investment Performance of Individual Investors, 2000 US X
Boys Will be Boys: Gender, Overconfidence, and Common Stock Investment, 2001 US X
Simultaneous over and underconfidence evidence from experimental asset markets, 2002 US X
Trading on Illusions: Unrealistic Perceptions of Control and Trading Performance, 2003 UK X
An Empirical Evaluation of the Overconfidence Hypothesis, 2004 US X
Overconfidence and Trading Volume, 2004 GER X X X X X
CEO Overconfidence and Corporate Investment, 2005 US X X X X X
Judgemental overconfidence, self-monitoring, and trading performance in an experimental financial market, 2005 X X X X
Skilled or Unskilled, but Still Unaware of It: How Perceptions of Difficulty Drive Miscalibration in Relative Comparisons, 2005 US X X X
Smokers' unrealistic optimism about their risk, 2005 US X
Investor overconfidence and trading volume, 2006 US X
Overconfidence in wargames: experimental evidence on expectations, aggression, gender and testosterone, 2006 US X X X X
True Overconfidence in Interval Estimates: Evidence Based on a New Measure of Miscalibration, 2006 GER X X
Who Makes Acquisitions? CEO Overcon…dence and the Market’s Reactios, 2006 US X X
The Trouble with Overconfidence, 2007 US X X X
All That Glitters: The Effect of Attention and News on the Buying Behavior of Individual and Institutional Investors, 2008 US X
Behavioral Bias Within The Decision Making Process, 2008 US X X X X
Portfolio Concentration and the Performance of Individual Investors, 2008 US X
Trusting the Stock Market, 2008 NL, IT X
An experimental test of the impact of overconfidence and gender on trading activity, 2009 X X X
Heritability of Overconfidence, 2009 SW X
Overconfidence in Forecasts of Own Performance: An Experimental Study, 2009 US X X X X
Sensation Seeking, Overconfidence, and Trading Activity, 2009 FIN X X X
Trading as Entertainment? 2009 GER X
A New Measurement Method of Investor Overconfidence, 2010 NL X X
CEO Overconfidence and Innovation, 2010 US X X
Development of the overconfidence measurement instrument for the economic experience, 2010 GER X
X
(50%=30%),
(98%=60%) X X
X
(82%)
Individualism and Momentum around the World, 2010 Global X X X
Managerial Miscalibration, 2010 US X
X
(80%=36%) X
Investor Overconfidence and the Forward Premium Puzzle, 2011 US X
Understanding the Better Than Average Effect: Motives (Still) Matter, 2012 US X
X
(80%)
Overcoming Overconfidence, 2014 US X X
X
(81%)
Overconfidence and Diversification, 2014 US X
Illusion of Control, 2015 SP X X
2.3 Miscalibration
The variable miscalibration is often measured by analyzing individuals by means of the confidence interval of their personal predictions. Typically, they are asked about their 10% confidence interval related to predictions on such subjects as the economy or company results. The articles used for this paragraph are shown in the Taxonomy in which the results of miscalibration are summarized.7 The most remarkable result from these articles is that, when executives predicted their realized market returns with an 80% confidence interval, they were right only 36% of the time (Ben-David, Graham & Harvey 2010). Another research in which results were tested in a similar way, found that, with a 50% confidence interval, the true quantity was identified only 30% of the time. In addition, this was only 60% with 98% confidence intervals (Michailova 2010). An interesting and useful finding is that individuals tend to neglect an element of surprise estimated at 15-20%; so when people say they are 99% sure, it should be translated to a probability of 85% (Kahneman, Riepe 1998). The last study that specifically tested miscalibration showed a similar result, as subjects were asked for a 90% calibration and were right only 75% of the time (Keren 1991). Therefore, miscalibration as a variable of overconfidence indicates how reliable the predictions of an individual are.
Figure 2 below, relates to overconfidence and miscalibration and visualizes the results of the articles where this is tested specifically.8 The black line starting at point 0,0 and ending at
100,100 represents the border between overconfidence and underconfidence. The red line indicates the linear regression of the miscalibration statistics and demonstrates a decreasing upward slope. The X-axis represents the confidence intervals, whereas the Y-axis represents the percentage of actual correct responses. Therefore, miscalibration clearly visualizes overconfidence in figure 2, as all outcomes are found in the grey area. It means, that the actual amount of correct responses are always below the expected confidence intervals, as is shown in this figure. This may also explain why psychological studies define miscalibration as the only relevant definition for overconfidence, as the figure demonstrates the border between realism and either over- or underconfidence.9 From the red line one can conclude that, as investors become more confident about their predictions, their assumptions become less reliable as the gap between the percentage of predicted and right outcomes increases.
7 These are summarised in the Appendix in table 20.
8 The articles in the Taxonomy where miscalibration is given with percentages are used for this analysis.
9 It is stated in psychological studies that miscalibration is the only pattern of behaviour that can be used as a
Figure 2: Miscalibration according to confidence interval and actual results.
The table below shows the statistics of the OLS of miscalibration with AR01 indicating the actual results and CI the confidence intervals. It shows that an increase of the confidence interval by 1% increases the actual results by approximately 0.69% and thus makes higher confidence intervals relatively even less reliable.10 The R-values are between 0.7 and 0.63 which indicates that they explain a significant amount of the model. Besides, the T-value of CI indicates that this model is significant.
Dependent Variable: AR01 Method: Least Squares Date: 03/16/16 Time: 17:15 Sample: 1 6
Included observations: 6
Variable Coefficient Std. Error t-Statistic Prob.
C -4.561099 18.34871 -0.248579 0.8159
CI 0.694866 0.224132 3.100258 0.0362
R-squared 0.706133 Mean dependent var 50.33333
10 Actual results of 20% are on average subject to a confidence interval of 36%. A 100% increase of these actual
results towards 40% are on average subject to a confidence interval of 63%. Hence, (63%-36%)/36% results in a percentual change of 75%. These intervals are indicated in figure 2 with the white lines.
Underconfidence
Adjusted R-squared 0.632666 S.D. dependent var 19.44908
S.E. of regression 11.78771 Akaike info criterion 8.033194
Sum squared resid 555.8004 Schwarz criterion 7.963780
Log likelihood -22.09958 Hannan-Quinn criter. 7.755326
F-statistic 9.611600 Durbin-Watson stat 1.803940
Prob(F-statistic) 0.036212
Table 1: OLS of AR01 and CI.
When the same relation is checked exponentially in the statistics below, the model becomes less valid as the R-values are significantly lower and the Akaike AIC higher. This indicates that less can be explained in the exponential model which implies that the linear function is probably more valid. The T-values given for CI indicate that one can assume there is an exponential relation, but as stated before, the values for validity are weak and it is therefore highly unlikely that this model is useful.
Dependent Variable: AREXP Method: Least Squares Date: 03/16/16 Time: 17:20 Sample: 1 6
Included observations: 6
Variable Coefficient Std. Error t-Statistic Prob.
C -2749.651 2265.032 -1.213957 0.2915
CI 70.86479 27.66764 2.561288 0.0626
R-squared 0.621219 Mean dependent var 2848.667
Adjusted R-squared 0.526524 S.D. dependent var 2114.702
S.E. of regression 1455.118 Akaike info criterion 17.66476
Sum squared resid 8469473. Schwarz criterion 17.59535
Log likelihood -50.99429 Hannan-Quinn criter. 17.38690
F-statistic 6.560195 Durbin-Watson stat 1.602712
Prob(F-statistic) 0.062553
Table 2: OLS of exponential AR01 and CI.
2.4 Better-than-average effect
above average compared to others, independent of the field of investigation. As stated before, 81% of new-business owners assign their own company to have a 70% higher chance of succeeding, whereas only 39% of the respondents expected the same of a comparable company managed by someone else (Odean 2014). This indicates that individuals generally perceive their skills and abilities to be more developed than they actually are. This perception can thus be linked to overconfidence.
The better-than-average effect can also be analyzed by the Taxonomy. From the articles that indicate the better-than-average effect with a specific number, one can conclude that on average, approximately 83% of individuals are biased by this behaviour.11 One may therefore assume that 83% of the investors on the stock market overestimate their own contribution when compared to others. This is a clear example of overconfidence as this number should evidently be in the middle, at 50%, where half of the people perform above and half of the people perform below average.
2.5 Illusion of control
As mentioned before, the next variable related to overconfidence is the illusion of control. This variable describes the assumption that individuals have more control over outcomes when they, in fact, have little or no control. Individuals who are biased by this illusion underestimate the role of chance, which makes it a difficult variable to measure. However, it is demonstrated that individuals are either aware of this illusion, or not (Yarritu, Matute & Vadillo 2014). Another study suggested that the illusion of control actually increases with age, as students demonstrated the lowest vulnerability towards it (Chira, Adams & Thornton 2011). However it is evident that the illusion of control diminishes overall performance (Fenton-O Creevy et al. 2003). Therefore, this variable contributes to overconfidence in the sense that individuals misjudge their own contributions to certain outcomes.
2.6 Excessive optimism
The last variable, excessive optimism, is the most important one as a measurement for overconfidence in this study, since it can be closely linked to the consumer confidence index
(CCI). Excessive optimism is commonly measured by questioning individuals about their predictions of economic or individual financial situations. In a computer-based study, participants had to play a war game including several elements, such as policy and economics. Prior to the game, they were asked at which rank they thought they would finish. On average they did so at place 72.3 where 200 was the lowest. In this article the writers relate this to excessive optimism, as the participants were unskilled, but aware of factors, such as chance being part of the game (Johnson et al. 2006). For this thesis the overconfidence is used as an indicator of excessive optimism, as it consists of several elements that are related to testing this variable. A high level of the CCI, for example, may indicate that individuals are overly optimistic about the economy and, equally, stock returns, and hence respond too late to a declining market.
The CCI is measured by analyzing the answers of a monthly sample of approximately one thousand households which are asked about their perception of the economy and their personal financial situation. They are stated as follows:12
1. “According to your personal opinion, do you think that the economic situation in our country over the past twelve months improved, worsened or remained the same?” 2. “What is your personal opinion for the next twelve months? Will the economic
situation in the Netherlands improve, worsen or remain the same?”
3. “Related to furniture, a washing machine, a television set, and other durable goods. According to your personal opinion, is it for consumers a favourable or an unfavourable time to purchase similar valuable goods or neither favourable nor unfavourable?”
4. “Has the financial situation of your household over the past twelve months improved, worsened or remained unchanged?”
5. “What do you expect of the personal financial situation of your household? Will it improve, worsen or remain the same during the next twelve months?”
The only possible answers are that it is becoming better (optimist), that it is becoming worse (pessimist), or that it remains the same (neutral). Once obtained, the percentage point of
optimistic responses is deducted from the percentage points of pessimistic responses and the outcome is the indicator for the index.
For this thesis the effect of overconfidence on the Dutch stock market is tested. As variables for an indication of overconfidence, miscalibration, as well as the better-than-average effect, the illusion of control, and excessive optimism are discussed. Specifically, excessive optimism is used as a relevant measurement and as an indication of the overconfidence bias. This measurement of excessive optimism will be compared to the economic climate according to several factors given in the following section. When excessive optimism exceeds the expectations of the economic climate one can presume that there is overconfidence. Contrary, when the economic climate appears to be more favourable than the measurement of excessive optimism indicates one can presume that there is underconfidence.
Subsequently, the Dutch stock market index of the AEX is analyzed. These indices are then compared to the Dutch CCI. If these are significantly related to each other, a decrease of the AEX index should be followed by a fall in consumer confidence. In the next section these indices are explained more extensively and supplemented with other relevant data of the Dutch economy.
Due to its relevance and measurability, for this thesis excessive optimism is measured more specifically as an indication of overconfidence, while the other three are used as additional explanations. With these results, an estimation of the overconfidence bias is measured and applied on the Dutch stock market. In extension to excessive optimism, in the next section, excessive trading will be analyzed to discuss the relevance of optimism.
2.7 Excessive trading
1998), (Grinblatt, Keloharju 2009). These individuals tend to overestimate the precision of their knowledge and make typical bad decisions such as constructing underdiversified portfolios (Heller 2014). However, excessive trading is not necessarily caused by overconfidence. Fund managers, for example, increase their trading activity due to the inflow of new money, as a result of former successful performance (Eshraghi 2011). Therefore it is important to identify when excessive trading is related to overconfidence and when to other factors.
2.4 Dutch market and financial bubbles
The Netherlands boasts a highly developed financial environment with many experienced investors (Van Rooij, Lusardi & Alessie 2011). An array of banks, stockbrokers, financial, pension and mortgage advisors, insurance companies, pension funds and insurance agents and brokers cater for the financial needs of the Dutch households, and the economy at large (Alessie, Hochguertel & Van Soest 2000). Due to the fact that internet penetration rates are among the world’s highest, with 95.5 percent of all households on line, a lot of financial activities have become internet based. Between 2004 and 2014 the number of bank branches in the country dropped by 1,500, or 52 percent. More closures have been announced since.13
Founded in 1606, the Amsterdam Stock Exchange is considered the world’s oldest, still-functioning bourse. To regulate these antics the municipal authorities confined stock trading to two hours a day, and to a dedicated building, known as the Bourse (Schama 1988).
The development of the Amsterdam Bourse was closely linked to the activities of the Vereenigde Oostindische Compagnie, the Dutch East Indies Company, known by its acronym VOC. Established in 1602, the VOC is deemed the world’s first publicly held company whose shares were traded on the Bourse. The Bourse gained great notoriety in the late 1630s when it witnessed the first recorded financial bubble –and crash – known as the Tulip Mania of 1636/1637.
It concurred with a sharp rise in Amsterdam house prices and VOC stock prices, which doubled between 1630 and 1639. Tulips, unlike canal houses or shares in land reclamation ventures, did not demand much capital and became therefore the investment of choice for the
13 Data are mainly provided by the websites of DNB and the CBS. Relevant datasets can be found in the
less-wealthy, but rapidly expanding, Dutch middle classes who could afford a luxury as a bunch of flowers (Israel 1995). It was not so much the trade in the bulbs proper, but the trade of future rights to buy and sell bulbs that went soaring. Prices of the futures rose to such an extent that authorities in several Dutch cities eventually blew the whistle on this trade. It was in fact a derivatives bubble more than anything else. When this bubble burst in 1637 vast fortunes were lost. Being a Calvinist nation to the core, allegories to these speculative follies were written and painted as to teach the public of the risks of earning money without labour in exchange.
Nothing much was learnt from it, since asset bubbles continue to these days, which the South Sea Company Bubble in the late 1710s/1720s (Carswell 1961), the British Railway Mania of the 1840s and the Dot-com Bubble of 1997/2000 can attest. All are fine examples of self-promoting cycles founded solely on overoptimistic speculation (Kindleberger, Aliber 2011).
The AEX index reached an all-time high of 701.56 on 4 September 2000. The Dot-com bubble then burst and had the index fall to a low of 218.44 on March 2003. By 2007 the 500-mark was again crossed. A year later the Great Recession took its toll and the index closed at the end of 2008 at 245.91, a drop of nearly 52 percent in one year’s time. These fluctuations had a major impact on the assets of both households and institutions (Stango, Zinman 2009). The pension funds were especially hard hit since the fall of the stock markets coincided with a sharp drop in interest rates worldwide.
3. Methodology and data:
A typical way for measuring overconfidence is to combine the variables, i.e. miscalibration, as well as the better-than-average effect, the illusion of control, and excessive optimism. In psychological studies, only miscalibration is directly linked to overconfidence, but for financial studies the other variables are commonly used as well (Michailova 2010).14 Hence, the following formula is configured:
(OC) = f (M , BA , IC , EO , Ɛ) (1) In which: OC = Overconfidence M = Miscalibration BA = Better-than-average effect IC = Illusion of control EO = Excessive optimism Ɛ = Error term f : Rn → R.
Where f (x,y) = c+β1x+ β2y is a multivariate function, where c is a scalar constant, βi is an
undefined parameter (i є R) and x,y є R are undefined variables. The number of variables and corresponding parameters in this function are undefined and equal to n є R.
For the rest of this thesis these symbols indicate the variables as mentioned above.
3.1 Data
According to the Taxonomy, it becomes clear that the biggest challenge for defining OC is to find a reliable measurement. Therefore, as EO seems to be the most commonly used variable for OC, it will be used for this thesis as well. A good indicator for EO can be the CCI and will therefore be analysed as a proxy for OC. However, as there is nothing wrong with being
14 An assumption why miscalibration is defined as the only pattern of behaviour for overconfidence is mentioned
confident when the economy is in a favourable state, the difference between confidence and OC has to be defined. Therefore, the available data has to be defined first.
The period of observation for this thesis lies between April 1986 and November 2015, since monthly data of the CCI have only been made available from the start of this period. Data from the AEX (formerly EOE index) are available since the start of 1983 and are included in the Appendix. However due to the lack of monthly CCI data over the years between 1983-1986 AEX data published before 1983-1986 will not be used for this thesis. The CCI itself was established in May 1972. It used to be updated three times a year, until 1984. Afterwards it was published on a quarterly base, until April 1986 when it became a monthly quoted figure. The Dutch CCI is measured in the same way as in the US and the European Union. This period is relevant for this study, since there are sufficient data available and major global, political, and economic developments occurred. The most important data are related to the Dutch market as this thesis is mainly about the Netherlands.
3.2 AEX-index
index in October 2007 after its takeover. For this thesis the AEX index starting in April 1986 is being used, as it covers the same period for which the CCI numbers are available.
Besides the AEX-index, the trading volumes of the AEX are also includes in this thesis. The available period for this data, however, starts in 2003. Trading volumes consist of the total amount of traded stocks according to their values.
3.3 Consumer confidence index
The next set of data consists of figures on the Dutch consumer confidence, provided by the CBS, done through a monthly survey amongst approximately one thousand Dutch households. The survey consists of five questions related to the economy in general and to the financial situation of the households.15 As mentioned, the amount of positive outcomes are deducted from the negative outcomes. A positive number indicates that the Dutch consumers are on average more positive about the economy, whereas a negative number indicates that the consumers are on average more negative about the economy. The first observation used for this thesis lies in April 1986 and continues every 15th of the month until 15 November 2015.
The CCI is subdivided in several categories related to the economy and spending propensity. The regular CCI and its subdivisions are also seasonally adjusted. The first subdivision is related to the economic climate in the Netherlands. It is constructed in the same way as the standard CCI, by measuring the difference between positive and negative outcomes among the first two questions which are related to the economic climate in general. The second subdivision deals with the spending propensity by households in the Netherlands. Similar to the one on economic climate, the subdivision covering spending propensity is constructed by measuring the difference between positive and negative outcomes of the last three questions related to the spending propensity.15 This index indicates the financial situation of Dutch households and their intention of purchasing durable goods. Subsequently, all indices are seasonally adjusted. The reason for these seasonal adjustments are simple: individuals tend to be more positive during spring and summer. The CBS reckons that by seasonally adjusting the indices, the numbers can be compared more equally every month. Adjustments are made every January, commencing in April 1986. Additionally, several well known elements of the
economy influence the CCI such as the unemployment rate. Like the CCI, the unemployment rate is published by the CBS. The unemployment rate in the Netherlands indicates the percentage of unemployed members of the total labour force. One should consider these factors as well for making a reliable assumption of the economy.
3.4 Economic Climate
With the data above, it becomes clear that the questions relating to the economic climate give a more relevant comparison for the next analysis. This is as the CCI includes questions about spending propensity which should be excluded for analyses exclusively related to the economy.
Hence the economic climate (ÊC) as an indication of the economy for this thesis is measured by combining the Unemployment Rate (UR), GDP growth rate (GDP), the Budget Deficit (BD) of the government, and the Inflation Rate (IR). Combined, this results in the following formula:
(ÊCt) = (α + β1(URt) + β2(GDPt) + β3(BDt) + β4(IRt) + Ɛ) (2)
Where α is a scalar constant, βi is an undefined parameter (i є R).
Where the factors are explained next.
3.5 GDP
GPI consists of fixed investments, inventory investments and residential investments and is the sum of these measurements combined. Raw materials are excluded in this category. The purpose of this category is to measure the investments which have yet to be realised for the future.
GP includes the total expenditures of new goods and services by all the relevant government instances. Transfer payments are excluded for this category whereas spending on welfare projects are included. GP indicates an estimation of the financial activities of the government.
Finally, NE is the difference between foreign country’s spending on domestic goods (X) and domestic spending on foreign goods (M). When X > M, it indicates that there is a trade surplus, and when X < M, it indicates that there is a trade deficit. Subsequently, when X = M, it indicates that there is a trade balance.
The total measurement of the GDP indicates an estimation of the current situation of the economy. When GDP decreases for two quarters in a row, this generally indicates that the economy is in a recession (Smith 2016). The other way around, when GDP increases for two quarters in a row, this generally indicates that the economy is progressive. The problem with a progressive economy is that it generally does not hold as prices increase too quickly and companies use their reserves. For these reasons, one can say that a high GDP growth rate boosts confidence, whereas a low GDP growth rate decreases confidence.
3.6 Unemployment rate
3.7 Budget deficit
As an indicator for the economic climate according to the government, the budget deficit (BD) is used for this thesis and obtained from the Centraal Planbureau (CPB). The BD is measured by subtracting the expenditures of the government from the revenues of the
government. One can speak of a budget deficit when expenditures exceed revenues, and hence, when revenues exceed expenditures one can speak of a budget surplus. The BD is published once a year by the government and individuals will therefore consider this amount until the next will be published. Generally, a BD is financed by debt such as issuing treasury bonds. One year that specifically catches attention is the BD of 1995. With a BD of -8.59 this year is by far the highest deficit of the period, while the economy was in a favourable state.
Therefore, one should bear in mind that during this year, the government did a great pay-off structure of subsidies from housing corporations. Generally, a small BD results in a positive assumption of the economy, as individuals see that the government is in a healthy state. Hence, a large BD results in a negative assumption of the economy, as individuals see that the
government is making great losses.
3.8 Inflation rate
4. Results
In the following section the results of the relevant analyses are given. Starting with explaining how the model is optimised, onwards it will result in the OC measurement for this thesis.
4.1 Economic climate measurement
The first step for constructing the OC measurement, is to compare the ÊC to the EC of the CCI. Including all the variables named before gives the following Autoregression in relation to the EC.
Vector Autoregression Estimates Date: 03/22/16 Time: 16:30
Sample (adjusted): 1986M04 2015M12 Included observations: 357 after adjustments
Standard errors in ( ) & t-statistics in [ ] EC C -6.968091 (6.78159) [-1.02750] UR 1.291451 (1.07667) [ 1.19948] GDP 16.80386 (1.77415) [ 9.47147] BD 2.202563 (0.66406) [ 3.31680] IR -8.879792 (1.15507) [-7.68765] R-squared 0.388237 Adj. R-squared 0.381285 Sum sq. Resids 169037.6 S.E. equation 21.91393 F-statistic 55.84657 Log likelihood -1606.146 Akaike AIC 9.026029 Schwarz SC 9.080339 Mean dependent -13.44538 S.D. dependent 27.85961
In the regression it becomes clear that these variables explain approximately 40% of the EC. The T-statistics of the variables indicate that they can be considered valid and are therefore reliable indicators for the EC. Next, it is however important to optimise the model by including lags. This is as individuals are not immediately aware of the economic
circumstances and may for example notice the effects of their unemployment after several months. Therefore, after testing many possible outcomes, the following model appears to be the statistically most valid.16
Vector Autoregression Estimates Date: 03/21/16 Time: 23:57
Sample (adjusted): 1986M04 2015M12 Included observations: 357 after adjustments
Standard errors in ( ) & t-statistics in [ ] EC C -47.46974 (5.62416) [-8.44032] UR(-9) 10.08111 (0.91593) [ 11.0064] GDP(-5) 13.22257 (1.41353) [ 9.35431] BD(-12) 5.886755 (0.55988) [ 10.5143] IR(-6) -8.424422 (0.97301) [-8.65810] R-squared 0.585022 Adj. R-squared 0.580306 Sum sq. resids 114663.5 S.E. equation 18.04851 F-statistic 124.0594 Log likelihood -1536.867 Akaike AIC 8.637909 Schwarz SC 8.692219 Mean dependent -13.44538 S.D. dependent 27.85961
Table 4: AR of EC and relevant variables with lags.
In this model, the T-values indicate that these variables are significant at a 99% level, with approximately 300 degrees of freedom. Next, the model explains 58.5% of the EC and the Akaike AIC decreased with approximately 0.4 points compared to the first model, indicating that this model is more reliable.
4.2 Autocorrelation
The first thing that comes forward is that the UR has a positive effect on consumer confidence in the EC. This is contradictory to the findings in the literature, which indicates that there might be other factors influencing the model. An obvious explanation can be that the variables are influenced by autocorrelation. High UR’s, for example, will influence government
expenditures, and hence BD, as social expenditures by the government increase and tax incomes decrease (Abrams 1999). To test for autocorrelation, the independent variables have to be regressed with each other. Starting with a regression where URt is the dependant
variable and BDt-3 the independent variable indicates that autocorrelation for these variables is
significant.17 This is as the statistics suggest that a one point increase of the BD influences the UR with -0.48179. Combined with a T-value of approximately -18.5 and R-values of 0.465 the significance is confirmed.18 Using the coefficients of this regression provides a new
variable for the UR which is then subtracted from the original URt to define URautocor,t for the
next regressions. Continuing with the IR, autocorrelation with the two variables above seems to be significant as well. With a regression where IRt is the dependant variable, and UR autocor,t-3 and BD-6 are used as independent variables provides an even lower Akaike AIC than
before.19 Combined with T-values of -2.59 and 5.6 respectively indicate that autocorrelation is
highly significant. With the coefficients, it indicates that a one point increase of the UR influences the IR with -0.117 and the BD influences the IR with 0.13.20 Again, the coefficients from this regression are used to create the variable IRautocor,t for the next
regression. This last regression, where GDPt is the dependant variable does not show
significant autocorrelation like the ones before. The T-values of the IR and the UR indicate, especially for the UR, that autocorrelation is significant but not for the BD. Besides, with an
17 A three month lag over the BD is applied as the original model has a 9-month lag for the UR and a 12-month
lag for the BD. Therefore, 12 – 9 = 3 to keep the lags relevant.
18 Table 16 in the Appendix provides these results.
19 As mentioned in the footnote above, the IR has a lag of 6 months, indicating that the UR uses a 3-month lag
and the BD a 6-month lag.
even lower value of the Akaike AIC this assumption seems highly significant.21 Unfortunately,
when these new variables adjusted for autocorrelation are applied to the EC, the problem with a positive UR on consumer confidence are not resolved.22 This indicates that there are still
external factors influencing the outcome.
4.3 Multicollinearity
From these results one can assume that there is multicollinearity in the model.
Muticollinearity indicates that one or more of the variables are highly correlated within the model which is obviously the case. In the figure below the variables are visualised where one can clearly see that they are correlated.
-10 -8 -6 -4 -2 0 2 4 6 8 10 1 5 -1 -1 9 8 3 1 5 -1 2 -1 9 8 3 1 5 -1 1 -1 9 8 4 1 5 -1 0 -1 9 8 5 1 5 -9 -1 9 8 6 1 5 -8 -1 9 8 7 1 5 -7 -1 9 8 8 1 5 -6 -1 9 8 9 1 5 -5 -1 9 9 0 1 5 -4 -1 9 9 1 1 5 -3 -1 9 9 2 1 5 -2 -1 9 9 3 1 5 -1 -1 9 9 4 1 5 -1 2 -1 9 9 4 1 5 -1 1 -1 9 9 5 1 5 -1 0 -1 9 9 6 1 5 -9 -1 9 9 7 1 5 -8 -1 9 9 8 1 5 -7 -1 9 9 9 1 5 -6 -2 0 0 0 1 5 -5 -2 0 0 1 1 5 -4 -2 0 0 2 1 5 -3 -2 0 0 3 1 5 -2 -2 0 0 4 1 5 -1 -2 0 0 5 1 5 -1 2 -2 0 0 5 1 5 -1 1 -2 0 0 6 1 5 -1 0 -2 0 0 7 1 5 -9 -2 0 0 8 1 5 -8 -2 0 0 9 1 5 -7 -2 0 1 0 1 5 -6 -2 0 1 1 1 5 -5 -2 0 1 2 1 5 -4 -2 0 1 3 1 5 -3 -2 0 1 4 1 5 -2 -2 0 1 5 % E c on om ic v a ri a bl e s -50 -40 -30 -20 -10 0 10 20 30 40 50 C C I Ind e x
Unemployment Rate Monthly Budget Deficit Percentage GDP Inflation
GDP Growth Rate EC
Figure 3: ECt and relevant variables URt, BDt, IRt, and GDPt..
Obviously, the IRt is high in times of low URt and vice versa. This is also the case for the BDt
where an assumption has been made before. Multicollinearity, as is the case in this model, does not affect the reliability of the model, but only influences the individual predictors of the variables. This problem can therefore possibly be adjusted by including more variables or by applying less correlated variables.
21 Table 18 in the Appendix provides these results.
4.4 Overconfidence measurement
By applying the coefficients of table 4, these numbers can be used as beta’s for the variables of the ÊC. The model is therefore as follows:
(ÊCt) = f (-47.46974 + 10.08111*(URt-9) + 13.22257*(GDPt-5) +5.886755*(BDt-12) –
8.424422*(IRt-5) + Ɛ) (2)
Once the ÊCt, an estimation of ECt, is constructed, the estimation is subtracted from the
regular ECt. This is as the ÊCt gives an indication of the current situation of the economy
whereas the ECt indicates part of the confidence in the economy. It is however not realistic to
assume that individuals respond directly to economic movements, as they will become aware of the relevant factors after they are published. Therefore, the ÊCt is adjusted with several lags
in order to make a reliable model. With this data, if the ECt is higher than the ÊCt, it indicates
that there is overconfidence, and hence, if the ECt is lower than the ÊCt, it indicates that there
is underconfidence. Therefore, the following formula is constructed:
(OCt) = (ECt - ÊCt) (3)
In an ideal situation this number should always be 0, as this would indicate that the
confidence in the economy is realistic. This is however not the case, which implies that there is always some measurement of overconfidence or underconfidence. The results of both the EC’s are visualised in the following graph.
Once the new variable, OCt is measured, the next analysis can be done by applying it to the
AEX-index. With this measurement, one can speak of OC with a positive number and underconfidence with a negative number. This is, as when the ECt is higher than the ÊCt,
individuals have a too optimistic perception of the economy and vice versa.
With this data a regression is made to check its relation on the AEX-index. The results of this analysis can be seen in the table below.
Vector Autoregression Estimates Date: 03/22/16 Time: 11:43
Sample (adjusted): 1986M04 2015M11 Included observations: 356 after adjustments
Standard errors in ( ) & t-statistics in [ ] AEX Index C 321.4147 (7.99436) [ 40.2052] OC 2.832080 (0.44554) [ 6.35653] R-squared 0.102446 Adj. R-squared 0.099911 Sum sq. resids 8054151. S.E. equation 150.8371 F-statistic 40.40545 Log likelihood -2289.907 Akaike AIC 12.87588 Schwarz SC 12.89765 Mean dependent 321.3601 S.D. dependent 158.9884
Table 5: AR of AEXt and OCt measurement.
With these statistics one can see, with the R-values, that approximately 10% of the OCt
measurement, explains the movements of the AEXt. Next, a one point increase of the OCt
measurement indicates that the AEX responds with approximately 2.8 points. Applying the same steps for alternative models, like excluding the lags or including a lag on the left side of the regression, ECt-1, all result in statistically less valid regressions.23 This indicates that the
model as mentioned, seems to be the most reliable indicator for OCt. Alternatively, when the
analysis is done with the model as given above, on the AEX-volumes, the OCt measurement
still seems to be relevant according to the T-values and Akiake AIC. With R-values of 0.019,
however, the model explains a respectively small amount of the AEX-volumes indicating that it is more useful for the AEX-index.24 Besides, when a one month lag of the AEX-index is
applied to the original model, it results in R-values of 0.98. This indicates that when the
model includes a lag on the left hand side, a major part of the AEX movements is explained.25
24 Results of this regression are given in table 8 in the Appendix.
5. Conclusion
The purpose of this thesis is to raise awareness of the OC bias and its relation to returns on the Dutch stock market. By analyzing prior literature, it has been made clear in this thesis that OC is a behaviour that can be explained by several other dimensions. Studies prove that confidence always exceeds accuracy; which is what M specifies (Kahneman, Riepe 1998). In the same article it is stated that OC is either caused by EO or M, or a combination of both biases (Kahneman, Riepe 1998).
In order to analyse OC, EO has been tested extensively for this thesis. As a measurement for EO, an estimation of the economic climate (ÊC) in the Netherlands is made and subtracted from the EC of the CCI. For the ÊC, the variables UR, GDP, BD and IR are used as they are stated to be relevant indicators of the economy. With the measurement, ECt – ÊCt, a new
variable OCt is made which is regressed on the AEX-index. From this regression one can see
that a 1 point increase of the OCt measurement results in an increase of approximately 2.83
points on the AEX-index. This explains that OC, as measured accordingly, does indeed influence the AEX-index. With the R-values it also becomes clear that this measurement explains 10% of the movements on the AEX and one should therefore be aware of the OC bias. The lags, as given in the model, can be explained by the assumption that individuals do not respond directly to economic movements. The one-year lag of the BD, for example, can simply be explained by the fact that these numbers are published at the end of September each year and will therefore be born in mind for the whole year onwards. For the other variables, it can also be assumed that individuals notice the effect of these movements after a certain period of time.
Extensively to this study the Taxonomy visualizes all relevant literature to provide an overview of the mentioned dimensions and OC. Starting with M, this dimension has been tested by combining several unique studies, which are summarized in table 20 in the Appendix. With the results of these studies, figure 2 has been provided, which visualizes M as an indicator for OC in a graph. One of the major results of this study on M is that, as individuals increase their confidence intervals, the difference between predicted and actual outcomes increases as well. This indicates that higher confidence intervals are relatively even less reliable. In numbers this means that when individuals become 1% more sure about their outcome, the amount of correct outcomes increases with only 0.75%.
The next dimension relevant for OC according to the Taxonomy is the BA. When combining studies on the BA, according to the results, an average of 83% of the individuals were found to be biased by this behaviour.26 This indicates that most individuals overestimate their own skills, as naturally only 50% of the individuals perform above average and, likewise, 50% of the individuals perform below average. This clearly proves the contribution of this behaviour to OC in general.
Additionally, the IC has also been proved to significantly contribute to OC. Individuals who are in a depressive mood are more susceptible to the IC (Thompson 1999). This relates the IC to the tendency of investors to be loss-averse, as losses will always result in negative feelings. Turned the other way around, this partly explains the relation between the IC and OC as it explains why investors feel the need to make gains and their unwillingness to accept losses as a result of their own failure. However, in this study it becomes clear that there is no reliable measurement for this bias so far.
In short, the OCt measurement provided in this thesis, may well be a reliable indicator for OC
on the Dutch stock market. Therefore, one can conclude that OC does indeed influence this market. However, to what extent it diminishes overall returns is not clear. Therefore, the other relevant OC dimensions have to be investigated more thoroughly.
