University of Amsterdam
Amsterdam School of Economics MSc Economics
Master’s Thesis Behavioural Economics & Game Theory (15 ECTS)
Are lucky investments from the past affecting individual investors’
future decisions?
Author: J.J. van Delft Student number: 11356758 Thesis supervisor: Ms. G. Romagnoli Second assessor: Dr. J.J. van der Weele Finish date: August 11, 2017 NuStatement of Originality
This document is written by Student Joeri Jan van Delft who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
This paper examines experimentally whether individual investors are subjected to the self-attribution bias. It is found that lucky participants were more likely to act against the advice of financial institutions compared to unlucky participants, indicating that the self-attribution bias, which is caused by people’s tendency to address their successes to ability and failures to bad luck, is present among this subject pool. This bias reduced participants’ investment returns from 1.9% to 1.1%. Actual individual investors can benefit from this study by being aware of this bias and overcome it, for them not to trade excessively. Financial institutions can also benefit from this finding by making individual investors aware of this bias, which in turn can improve their reputation. Further research is needed to support the robustness of the results and to find the actual loss in returns that the self-attribution bias causes to individual investors in financial markets.
JEL Classification: G02, G11
List of Tables
Table 1: Overview of experimental design ... 18 Table 2: Independent variables ... 21 Table 3: Overview of data used in part 2 of the experiment ... 23 Table 4: Overview of participants’ performance in part 1 ... 25 Table 5: Overview of the participants’ tendency to AAA ... 26 Table 6: Buy versus sell recommendations. ... 27 Table 7: Multiple Regression Analyses of AAA ... 28 Table 8: Mean return in part 2 ... 29 Table 9: Multiple Regression Analysis of Returns ... 30 Table 10: Pearson Correlation Matrix and collinearity statistics ... 30Table of Contents
1 Introduction ... 6
2 Literature review ... 9
2.1 Self-attribution bias ...9
2.2 Self-attribution bias in experiments ...10
2.3 Stock analyses ...11
2.4 Stock recommendations ...12
3 Methodology ... 15
3.1 Experimental design ...15
3.2 Mean comparison analyses ...17
3.3 Standard Multiple Regression Analysis ...19
4 Data ... 22
5 Results ... 24
5.1 Demographics of subject pool ...24
5.2 Results of part 1 of the experiment ...24
5.3 Results of part 2 of the experiment ...25
5.4 Multiple regression analysis ...27
5.5 Returns ...28
5.6 Limitations ...29
6 Discussion ... 32
6.1 Mean comparisons within groups ...32
6.2 Buy versus sell advices ...33
6.3 Multiple Regression Analyses ...34
6.4 Earnings ...35
6.5 Potential improvements of the experiment ...35
1 Introduction
In financial markets overconfidence emerges in two forms: overoptimism and miscalibration. As a result, future cash flows are overestimated and its volatilities are underestimated (Malmendier & Tate, 2005; Ben-David, Graham & Harvey, 2013). Overconfidence causes excessive trading, leading to lower net returns (Barber & Odean, 2000). Various factors affect the amount of overconfidence that individuals have. One potential source of overconfidence is the self-attribution bias which is caused by people addressing their successes to skills and failures to bad luck (Miller & Ross, 1975; Zuckerman, 1979). In that case, people can become overconfident after situations where being lucky instead of being skilled caused them to perform well.
Previous research shows that it is not uncommon for financial executives and analysts to be subjected to this bias (Libby & Rennekamp, 2012; Hilary & Hsu, 2011; Billett & Qian, 2008; Hilary & Menzly, 2006). This is not surprising considering that financial markets are hard to predict and people expect successes as a consequence of their actions (Miller & Ross, 1975). Given these points, it is interesting to examine whether the self-attribution bias is also causing individual investors to be overconfident in financial markets. In this paper, it is examined whether participants considered their past performance, based on random guesses, in future decision making. To my knowledge, no research has been executed before concerning the self-attribution bias among individual investors. I aim to contribute to the literature by filling in this gap.
To obtain the desired results, an experiment consisting of three parts is conducted. In a nutshell, in part 1 participants are shown historical stock charts of ten months of different unspecified companies and are asked to estimate whether this stock will decrease or increase in the next two months. After four rounds, the participants receive feedback about their performance. In the next part, similar historical stock charts are shown to the participants including an advice by an unspecified financial institution. Subsequently, they are asked to choose between the current stock price or the stock price in two months from that moment. To obtain some background knowledge about the participant, a small questionnaire is conducted in the third part. Hypothesized is that the self-attribution bias causes participants who were lucky in the first part are more likely to act against the advice of the financial institution compared to the unlucky participants.
Fama (1970) argued that markets are efficient, meaning that historical stock prices are completely uninformative. Accordingly, different studies suggest that technical analyses (based on
historical stock prices) are not profitable for investors (Neftci, 1991; Marshall, Cahan & Cahan, 2008). Besides, stock recommendations by financial institutions are valuable to individual investors if one trades to it in time (Womack, 1996; Barber, Lehavy, McNichols & Trueman, 2001; Boni & Womack, 2006; Jegadeesh & Kim, 2006). Considering these findings, acting against the advice of financial institutions (from now on, AAA) reduces the investment returns of investors in general. Therefore, the self-attribution bias has detrimental effects on the returns of individual investors who are subjected to this bias. The results of this study confirm the negative effects of this bias as it reduces participants’ returns from 1.9% to 1.1%.
In this study, the mean comparison and multiple regression analyses showed that participants who were lucky in the first part of the experiment, acted more against the advices of financial institutions. The regression analysis reported an increase of 3.9% to 4.5% AAA for every correct guess in part 1. Although it is hard to interpret these numbers in real-life situations, the main takeaway from this study is that individual investors’ future decisions are influenced by the self-attribution bias. This means that lucky investments in the past causes overconfidence in the future, resulting in more detrimental investment decisions. The difference in the number of times lucky and unlucky participants AAA was mainly found for buy recommendations and to a lesser extent for sell recommendations. Additionally, this study found that gender and risk-attitude did not affect the tendency to AAA.
The results of this study have potentially relevant implications for individual investors and financial institutions. As for the individual investors, it can be of great importance to be aware of their potential subjection to the self-attribution bias in financial markets. Excessive trading and ignoring stock recommendations are two examples of harmful consequences of this bias. Additionally, awareness of the existence of this bias is also an opportunity for financial institutions to inform their customers, the individual investors, about this bias. They should not inform their customers only about the self-attribution bias, but also about the fact that people generally believe that they are not subjected to this bias (Libby & Rennekamp, 2012). For individual investors, it is important to be aware of this bias not to trade excessively because it leads to lower net returns. Although providing this kind of information reduces the number of trades per investor leading to lower transaction fees for financial institutions, it can also be important to improve their reputation. Logically, informing customers about this bias can help to achieve this. In the end, a better reputation can lead to more future customers.
In the next section, the existing literature concerning the self-attribution bias, stock analyses and stock recommendations will be reviewed. After that, the methodology will be discussed in Section 3 where the experimental design and the various performed analyses will be explained. In Section 4, the data concerning the historical stock charts and the financial institutions’ recommendations to set up this experiment will be briefly discussed. Next, Section 5 will provide all results and potential limitations, followed by a discussion of the results and the experiment in Section 6. In Section 7, a conclusion will be drawn up.
2 Literature review
2.1 Self-attribution bias
There is a tendency for people to believe that their successes are caused by skills while their failures are due to bad luck (Miller & Ross, 1975; Zuckerman, 1979). This phenomenon, which is called the self-attribution bias, is a source of overconfidence. In general, people feel more responsible if an outcome is expected and they usually expect successes in consequence of their behaviour. Consequently, it is easier to relate successes to behaviour than it is for failures. Miller and Ross (1975) argue that the self-attribution bias is a bias in information-processing and they attribute this to people’s perception of the relationship between their behaviour and its outcome. However, in a later study by Zuckerman (1979), it is argued that the motivational process is more likely to cause the self-attribution bias. The fact that people need to maintain self-esteem cause them to relate success to internal factors and they attribute their failures to external factors for self-protection.
In financial markets, overconfidence is present in two major forms; participants being overprecise as well as feeling better than average (Malmendier & Tate, 2005; Ben-David, Graham & Harvey, 2013). Investors’ overprecision leads to miscalibration leading to an underestimation of the volatilities of returns and stock price movements. That feeling better than average is related to overoptimism can be explained by the buy and sell mechanism of stocks. When an investor buys a stock for a certain price, there must be another investor who wants to sell these stocks for that price. Thus, buying a stock implies that the buyer knows better than the seller or the buyer is more optimistic about the continuation of the stock price. Many participants in financial markets, like corporate managers, analysts and traders, can show different forms of overconfidence. In this part, overconfidence of agents and investors will be briefly discussed.
Ben-David, Graham and Harvey (2007) showed that most financial executives are miscalibrated regarding stock market predictions. As an illustration, an experiment by Libby and Rennekamp (2012) showed that financial executives are generally subjected to the self-attribution bias as they are more likely to attribute their successes to internal than to external factors. Another research showed that financial executives tend to be more precise, though less accurate, in their subsequent forecasts after predicting previous forecasts more accurately compared to others.
Moreover, their subsequent forecasts show an increase in deviation from most analysts’ predictions (Hilary & Hsu, 2011). Similarly, financial executives who successfully executed their first few M&A deals are more likely to acquire again, although their successes could not be attributed to their skills (Billett & Qian, 2008). As a matter of fact, many financial executives know about the self-attribution bias but think they are not subjected to it (Libby & Rennekamp, 2012).
Like financial executives, miscalibration due to the self-attribution bias is also present among market analysts. The predictions of analysts who performed excellent in previous periods, appeared to be less accurate in subsequent periods compared to the median analysts. Moreover, the predictions deviated more often from the general opinion among other analysts (Hilary & Menzly, 2006).
Odean (1998) argues that people are generally overconfident, and that investors are no exception considering the different biases investors are subjected to. One example of a consequence of overconfidence is that it increases trading volumes among individual investors. In a later study, Odean (1999) calls the trading volumes of these individual investors even excessive. Excessive, because it deteriorates earnings as it is shown that individual investors’ net returns decrease with an increase in the number of trades in securities (Barber & Odean, 2000). Additionally, Gervais and Odean (2001) showed that traders who are new to the market are more likely to attribute their successes to skills resulting in overconfidence. After gaining more experience, traders are better able to link their performance to ability.
In summary, as a result of the self-attribution bias, overconfidence causes managers to be overly active in M&A deal. In addition, it causes managers and analysts to be overprecise but less accurate in their forecasts. Furthermore, it causes traders to trade too often, especially when they have little experience, leading to a deterioration of their net returns. Nevertheless, to my knowledge, no research has been done on the self-attribution bias among individual investors. By filling in this gap, this study aims to contribute to the literature.
2.2 Self-attribution bias in experiments
An experimental study by Libby and Rennekamp (2012) showed that experienced financial and accounting managers’ decisions, whether to provide earnings forecasts, can depend on past performance. When managers believe that they can predict earnings with a certain accuracy, the
benefits will exceed the costs and therefore they will issue the earnings forecast. Their experiment consisted of two parts with different levels of trivia questions. The participants were asked to rate their performance in the first part on internal versus external factors. It appeared that managers placed greater weight on internal factors (skill and effort) compared to external factors (difficulty and luck) when performance was positive and vice versa. Positive past performance triggered managers to be overconfident resulting in the belief that future performance will improve. In the experiment this resulted in managers committing to improve performance in the second round. The commitment would increase(decrease) their returns if it turned out to be correct(incorrect). To put the experiment into perspective, the authors compared their experiment to real-life business in which the difficulty of the task and the ability to perform the task is rather uncertain and compensation increases with performance. Regardless of the driving force of positive past performance, managers were more likely to provide forecasts when their firm is performing well. In accordance to the field data described in the previous subsection, the results from this experiment confirm that experienced financial and accounting managers are subjected to the self-attribution bias. Since the experimental and field data results both suggest that managers are subject to this bias, it is more likely that the results from this study can also be reflected on actual individual investors.
2.3 Stock analyses
Fundamental and technical analyses are two ways to predict future price changes and accordingly, to determine whether to buy or sell a stock. Companies’ financial statements are the essential part for a fundamental analysis. In contrast, the historical stock chart is the part on which a technical analysis is based. Logically, the expectation that a stock price will increase(decrease) in the future encourage investors to buy(sell) this stock.
For fundamental analysts, the expectation for a stock price to increase depends on the current market value and the estimated value by their analyses. Different methods like the Dividend Discount Model, the Discounted Cash Flow Model and the Adjusted Present Value model are used to value companies. Next to that, analysts also include the competitors and industry into their valuation. By way of contrast, technical analyses are less comprehensive as the analysts only consider the historical stock prices of that company. Following the theory of Fama (1970)
stating that capital markets are efficient in the sense that current stock prices completely reflect all information currently available, technical analyses are considered to be completely worthless. It is true that there exists autocorrelation on daily lagged returns, as is shown by i.a. Akgiray (1989). However, daily lagged returns are not relevant in this study because the aim is to find the self-attribution bias among individual medium- and long-term investors. According to Sewell (2011), who in his survey covered various studies, there is no significant autocorrelation on stocks’ daily, weekly and monthly returns. That financial markets are efficient with the implication that technical analysis is useless, corresponds to the finding of Neftci (1991) concluding that even well designed prediction rules are of no use in forecasting. This finding is confirmed in a later study by Marshall, Cahan and Cahan (2008) who, after taking the data snooping bias into account, did not find any profitable rules when testing almost eight thousand different technical trading rules.
In a nutshell, despite the disagreement on the existence of daily autocorrelation of returns, there is no reason to believe that daily autocorrelation can predict the future stock price two months from now. Therefore, to predict the future stock price of a company in two months, technical analyses are considered useless while fundamental analysis is at least to some extent accurate. The assumption that financial markets are efficient is crucial in this study.
2.4 Stock recommendations
The value of stock recommendations for individual investors is rather uncertain. Within these stock recommendations there are remarkable features that should be discussed before drawing any conclusions. First, there is a difference between recommendations from institutional investment firms and brokerage firms or from banks. While the information provided by institutional investment firms is usually private to the buyer of the analysis only, brokerage firms and banks publish their recommendations publicly. In this study, the focus will be on recommendations from banks and brokerage firms, which are called the sell-side analysts.
One crucial issue with sell-side analysts’ recommendations is the existing conflicts of interest. In a firm, the objectives of the corporate finance and the brokerage division differ, leading to potential unreliable recommendations. While the goal of a brokerage division is to maintain or even improve the reliability of their recommendations and to trade many securities to obtain transaction fees, the corporate finance division concerns more about the accessibility to gather new
information about the analysed company and the likelihood to do business with this company in the future. Another problematic issue is the incentive for banks to influence stock price changes to increase investment banking profits (Michaely & Womack, 2005). Because of the interests of the bank, brokerage divisions tend to be overoptimistic about the analysed firm. In fact, the buy/sell ratio of recommendations was to-1 twenty to 25 years ago (Rajan & Servaes, 1997). This 10-to-1 ratio becomes even larger when there is an investment banking relationship between the recommending and the recommended firms (Michaely & Womack, 1999). From this finding, one might also argue that a sell recommendation is more reliable than a buy recommendation, because of the additional incentives that can go along with a buy recommendation. This statement is supported by Francis and Soffer (1997) who argue that, because of the incentives for financial institutions, sell recommendations are issued with more care leading to less valuation errors.
Considering these biases one might think that analysts’ recommendations are useless. In this paragraph, the real value of recommendations will be discussed. To begin with, Womack (1996) showed that there is generally a positive(negative) price drift following positive(negative) recommendation in the three subsequent months. Corresponding results are provided by Barber et al. (2001) who found that stocks with the highest recommendations outperform the ones with the lowest by about 9% per year. Additionally, results by Boni and Womack (2006) showed that it is possible to outperform the market by creating a portfolio of long positions in positively recommended stocks and short positions in negatively recommended stocks within one industry. Likewise, the study by Jegadeesh and Kim (2006) showed that such a portfolio outperforms the market, especially in the US and Japan. Furthermore, analysts’ recommendations appear to correct most of the time because there is a bias in the stocks chosen to recommend. By recommending stocks with a positive momentum, high volume or high growth, it seems that these stocks usually outperform others, resulting in the positive value for investors (Jegadeesh, Kim, Krische and Lee 2004).
Although there is a big controversy surrounding the incentives for financial institutions in their recommendations, it appears that these recommendations contain valuable information regarding the prediction of future stock prices. Be that as it may, one should be aware of the financial institutions’ tendency to exaggerate forecasts (Dimson & Marsh, 1984) and of the duration of validity of the recommendation. This duration differs per sell recommendation, buy recommendation or target price, but after some time (6 weeks to 6 months) the drift ends (Womack,
1996; Logue & Tuttle, 1973). As a final point, to earn abnormal returns, investors must ideally react within two weeks after the recommendation is published. Increasing the duration between the date of publication and date of trading, as well as less frequent updating of the portfolio, reduce investors’ profits (Barber et al., 2001).
3
Methodology
This section first explains the design of the experiment. After that, two ways of examining the effect of the self-attribution bias on future decision making will be discussed. First, the difference in the tendency to follow the advice of the financial institution based on all variables are examined separately, where the variables are divided into at least two different groups. Second, standard multiple regression analyses are performed to examine the effect of the self-attribution bias, including the role of various control variables.
3.1 Experimental design
To examine whether individual investors are subject to the self-attribution bias, an experiment is conducted. Many participants were needed and little funding was available to conduct this research. Therefore, it was useful to conduct this experiment online. By this way, I could gather participants from all over the country at any moment in time. oTree is an open-source programming application that is used to create this online experiment (Chen, Schonger & Wickens, 2016). In total, the experiment consisted of three different parts. All three parts will be extensively described in the next three parts.
To be able to demonstrate that participants are subject to the self-attribution bias, there is a need to separate the whole subject pool into two groups; the well and the poorly performing group. Part 1 of the experiment takes care of this. The task in this part is to estimate whether a stock price will increase or decrease when the only information that is provided is an historical stock chart. Keep in mind that the literature shows that one cannot be skilled in this task and performance is solely based on luck. At the end of this part, after four rounds, feedback is provided to give the participants an idea of how “skilled” they are in this task.
Appendix A shows the instructions for part 1 of the experiment. The instructions were visible during the whole first part. In the instructions, it becomes clear that the participants can expect four rounds where a historical stock chart of ten subsequent months of a publicly listed company is presented. Subsequently, participants are asked to estimate whether the stock price has increased or decreased over the two subsequent months (see Appendix B for part 1 Round 1). Participants only receive this chart, without any information about the period or the company. To
avoid that participants could recognize which company’s stock price is presented, the current stock price was normalized to ten. After every round the participants immediately receive feedback about the continuation of the stock price (see Appendix C for Round 1). Part 1 serves two purposes. On the one hand, participants will become acquainted with the task. On the other hand, it can make participants believe that they are skilled in doing the task. The latter is needed in part 2, where the task is similar. To emphasize the participants’ performance in part 1, a complete overview is showed to them after the last round of part 1 (see Appendix D).
Besides, a personal characteristic that can affect the tendency to AAA is financial knowledge. A Likert scale is used to learn the participants’ financial knowledge. At the end of this part, the participants are asked to rate, on a scale from one to seven, how important luck and skills are in this game. Regardless of the results, it is expected that the more financial knowledge a participant has, the more he or she addresses his or her performance to luck.
Next, to see whether participants are subjected to the self-attribution bias the second part of the experiment comes in. In addition to the historical chart of part 1, part 2 also provides an actual recommendation by a financial institution whose name is unannounced to the participants. Instead, participants are told that the additional information is provided by a well-known financial institution. In contrast to the previous part, participants do not receive any feedback about their performance to keep the believes about their ability constant during this entire part. Again, participants are asked to estimate whether the stock price has increased or decreased over the two subsequent months. Another difference to part 1 is that participants are financially incentivized to make decisions according to their preferences. In all rounds of part 2, current stock prices will be normalized to 10 points. Participants are asked to choose between two options (see Appendix E). They can either choose the current stock price of 10 points, or to have the stock price in points in two months from now. Suppose that the future stock price is the chosen option and the stock price has increased(decreased) with e.g. 20%, the participant earns 12(8) points. Except for the different stocks and their corresponding recommendations, all rounds in part 2 are equal. All eight earnings add up to the participant’s total earnings. Selecting the best performing participant for payment can induce excessive risk taking. Therefore, at the end of the experiment, one participant was randomly selected for payment. The information is communicated in the instructions of part 2 which can be found in Appendix F. Again, those instructions were visible during the entire part.
After this part, the number of AAA can be checked. Considering that historical stock prices are worthless in these games, the best thing participants can do is to follow the advice. The unrealistic belief that the information provided to the participants is of higher quality compared to the information the financial institution had at the time of the recommendation can induce the participants to AAA. Another reason for participants to AAA is the belief they are more skilled in predicting future stock prices based on the uninformative historical stock prices.
Altogether, it is expected that participants who had more correct guesses (from here on, CG) in part 1, are more likely to AAA. In this study, the unlucky participants (0, 1 or 2 CG) are compared to the lucky participants (3 or 4 CG). Additionally, OLS-regressions will show whether there is a continuous relationship between CG and AAA.
Furthermore, as for individual investors, other personal characteristics like gender and risk-attitude can show differences in the subjection to the self-attribution bias. A questionnaire is conducted in part 3 to obtain these personal characteristics. Moreover, the risk-attitude of participants might influence the likelihood of investing in these stocks at all. The Holt & Laury (2002) risk-attitude test is adopted in this research to assess the participants’ risk-attitude. A concise overview of the different parts can be found in Table 1.
3.2 Mean comparison analyses
To find out to what extent participants are subjected to the self-attribution bias, the effect of the percentage of correct guesses in the first part on the tendency to follow the advice of the financial institution in the second part is examined. To be more specific, the number of times that participants AAA is the measurement that is considered to assess the subjection to the self-attribution bias. In the first part of the analysis, the group of participants is divided into two groups consisting of the lucky and the unlucky group. The division took place after the experiment is concluded, based on the performance in the first part of the experiment. Considering the existing literature, suggesting that markets are almost completely efficient and that historical stock charts are almost completely uninformative; these groups will be called the “lucky” and the “unlucky” group. The lucky group consists of participants with three or four CG and the remaining participants together form the unlucky group. Next, the average number of AAA of lucky participants will be compared to the average number of AAA of unlucky participants. A higher
Table 1: Overview of experimental design
This table provides a brief overview of the content of all three parts of the experiment. Part Description
1 Four rounds where participants are asked to estimate a stock price increase or decrease. Only information provided is the historical stock chart. Subsequently, provision of feedback and financial knowledge test.
2 Eight rounds where participants are asked to either buy or not buy the stock. Information provided in this round consists of stock chart and analysts’ recommendation.
3 Questionnaire; Gender and Holt & Laury (2002) test
mean of AAA of the lucky group compared the unlucky group, indicates that the self-attribution bias is present among the participants.
Furthermore, additional factors are tested the same way as described before. Factors like gender, financial knowledge and risk-attitude are possibly affecting the tendency to follow the advice of the financial institutions on their own. Logically, the decisions of females are compared to the decisions of males. However, as for financial knowledge and risk-attitude, the tests are slightly different because there are more than two groups present.
Regarding financial knowledge, the participants are divided into two approximately equally sized groups. Subsequently, a comparison will be made between the group with the highest and the lowest financial knowledge. As mentioned before, financial knowledge is simply measured by asking the participants to rate the on what the performance in the first part of the experiment is primarily based, skill or luck. Participants with a financial background are supposed to know that historical stock charts are almost completely uninformative and are therefore expected to address performance mostly to luck instead of skills, by this way one can distinguish the experts from the laymen.
Lastly, the subject pool is divided into three approximately equally sized groups, based on attitude for which the Holt & Laury (2002) test is used. The first group contains the risk-seeking participants, the second group consists of risk-neutral and slightly risk-averse participants and the more risk-averse participants together form the third group. Subsequently, the risk-seeking and the risk-averse group are compared to examine the effect of risk-attitude on the tendency to follow the advice of financial institutions.
3.3 Standard Multiple Regression Analysis
Instead of just dividing participants into different groups based on certain characteristics and subsequently examining the differences in their tendency to follow the recommendations, Standard Multiple Regression Analysis (MRA) is another method that can combine all factors to analyse the self-attribution bias among the participants. In this MRA, the number of AAA of each participant is the dependent variable. It is a measure that counts how often participants believe that they are better able to estimate the future stock price compared to a financial institution. While the aim is to find the causal effect of the number of CG on the tendency to AAA, augmenting the regression with more independent regressors improves the estimates of the causal effects. In this section, all independent variables are separately discussed.
To begin with, the essential variable of this study is the number of CG in part 1 in the experiment and its effect on the tendency to follow the advice of financial institutions. As described before, people usually address their successes to skills and their failures to bad luck. In this experiment, neither successes nor failures can be addressed to skills, but it is expected that the participants are not aware of this. Because participants relate their performance in part 1 to skills, it is expected that the higher the number of CG, the higher the number of AAA. Note that in this analysis, the variable is not binary but instead a continuous variable based on the number of CG.
Next, gender can have a significant effect on the subjection to the self-attribution bias. Barber and Odean (2001) showed that men are more overconfident in financial markets than women, resulting in excessive trading and less net returns. As for this experiment, it is hypothesized that men are more likely than women to AAA, especially when they perform well in the first round.
Additionally, there is financial knowledge, which is also a continuous variable, determined by a self-reported skill-luck ratio in guessing future performance of a stock. Participants with a financial background are supposed to know that this experiment is primarily based on luck, because markets are largely efficient. They are also supposed to know that recommendations are at least, to some extent, of value. Therefore, it is expected that participants with less financial knowledge tend to act more against the advice of financial institutions.
Lastly, the risk-attitude is a continuous variable based on the number of safe choices based on the risk-attitude test. Its effect on the tendency to AAA is harder to predict. On the one hand,
participants who are extremely risk averse choose the certain outcome in every round, regardless of the historical stock chart or the recommendation. On the other hand, participants who are (slightly) risk averse can also choose to frequently follow the recommendations to go for the highest expected return in every round. Besides, if participants gamble eight times in a row with favourable odds in stock markets, the risk taken is not that big. In contrast, risk-loving participants are probably paying less attention to the recommendations and more seeking for sensation. The premise that risky behaviour in finance is positively related to sensation seeking is showed by Horvath and Zuckerman (1993). The previously discussed independent variables are summarized in Table 2 for clarification purposes, including the corresponding measurements, hypothesized effect on the dependent variable and the explanation for the hypothesized effect.
To estimate the number of times that participants AAA, the following OLS-regression is performed:
(1) %%%& = ( + (*+,-&+ (./012/3&./012/+ (
.456&+ (789%&+ :;
In this estimation model, the dependent variable: %%%&, is the number of times that participant (p)
acts against the advice of the financial institution. Next, the independent variables that are included are; ,-&, which is the number of correct guesses given in part 1 of the experiment by participant
(p), 3&./012/, which is a dummy variable for gender, 56&, which is a measure of financial
knowledge and 9%&, which is the risk-attitude measure.
In the end, three different OLS-regressions are performed. In the first model, the CG variable is the only variable estimating the AAA coefficient. In the second one, all mentioned variables are included. Subsequently, it is possible to compare this regression to the first one to observe how the additional regressors improve the estimates of the causal effects. In the last model, the FK variable is excluded from the model because of its high correlation with the CG variable. This issue will be discussed in Section 6.4.
Table 2: Independent variables
This table lists all independent variables with its corresponding measurement in column 1 and 2 respectively. Next, column 3 and 4 shows the expected direction and explanation for the expected direction. The expected direction indicates the direction of the variable’s correlation with the independent variable (tendency to AAA).
Independent
variable Measurement
Expected
direction Explanation
,-& Number of correct guesses
part 1 Positive
The self-attribution bias can make the participant feel that the participant is really skilled in this game, leading to more ignorance of the financial recommendations.
3&./012/ 1 for female, 0 for male Negative
Men are expected to be more overconfident in financial markets. This means that women are probably less likely to act against the advice of recommendations.
56&
Assessment of performance based on a scale from 1 (skill) or 7 (luck)
Negative Financial experts know that markets are efficient and thus skills are of little influence on this experiment. Therefore, high FK results in less AAA because it is beneficial to follow the advice of financial institutions.
9%& Number of safe choices in
Holt & Laury (2002) test Negative
Risk-loving particpants are expected to pay little attention to
recommendations, while risk-averse participants are more likely to follow the advice to for for the higher odds.
4
Data
In this section, the data that is used in the experiment will be briefly discussed. In the experiment, actual historical stock prices are used, as well as actual stock recommendations. Data on stock prices of twelve different companies for part 1 and part 2 of the experiment were collected from Thomson Reuters’ Datastream. Whereas the eight different stock recommendations for part 2 of the experiment were collected from Microsoft’s Yahoo!Finance.
Because part 1 of the experiment was just a guessing game, the stocks and the time range could have been randomly selected. Eventually, historical stock prices of the following companies were converted into the graphs that were needed in this part; ArcelorMittal, Koninklijke Philips Electronics N.V., The Walt Disney Company and Chipotle Mexican Grill, Inc.
In contrast to part 1, the stocks and time periods in part 2 were selected with more care. There were a few requirements that the combination of stocks had to meet to make this game appropriate to answer the research question. First, following the advice of the financial institutions must be more beneficial for the participants compared to random guessing in order to make the self-attribution bias deteriorating your returns. Next to that, both buy and sell recommendations were included in the experiment since the literature suggests that sell recommendations are more credible than buy recommendations. By this way, the different recommendations can also be compared. A combination of the eight companies, presented in Table 3, met these requirements. Table 3 provides an overview of the companies used in the experiment, specifications about the recommendation, the final stock price and the corresponding net return of the two months after the recommendation was published.
From Table 3 Columns 4 and 7 it can be derived that five of the eight recommendations were correct and that following every recommendation generates a return of 4.03% per round. Compared to buying every stock at the date of recommendation, generating a return of 2.55% per round, following all recommendations is the better strategy. Logically, not buying any stock generates a return of 0.00%.
Table 3: Overview of data used in part 2 of the experiment
The recommendation of each company listed in column 1 is issued on the date shown in column 2 by the analyst listed in column 3. The content of the advice is listed in column 4. Column 5 and 6 report the stock price of the recommended company on the date the recommendation is issued and the price two months after the recommendation is published. Column 7 reports the return in the two months after the recommendation is published.
Company Date of
recommendation Analyst Advice
Price at date of recommendation Price two months later Return last two months Activision Blizzard,
Inc. 17/01/2014 CRT Capital Buy 16.96 21.43 26.36%
Advanced Micro
Devices, Inc. 11/07/2011 JMP Securities Sell 6.76 6.52 (3.55%)
Banco Bilbao Vizcaya
Argentaria, S.A. 10/11/2016 Citigroup Sell 6.69 6.73 0.06%
FedEx Corporation 10/06/2015 Stifel Buy 182.32 169.40 (7.09%)
Intel Corporation 16/11/2012 Standpoint research Buy 20.19 22.11 9.51%
Johnson & Johnson 20/03/2009 Swiss Bank Buy 51.67 55.87 8.13%
Morgan Stanley 18/07/2007 Punk, Ziegel & Co Sell 70.85 64.92 (8.38%)
5
Results
In this section, a complete overview will be presented of the experimental results and subsequently the results from the analyses. To begin with, the demographics of the subject pool will be reported followed by an overview of the participants’ performance in part 1 of the experiment. After that, the participants’ tendency to AAA is presented and the corresponding within group comparisons are made. Subsequently, the results of the different MRAs are shown. To conclude this section, some limitations of this study are discussed.
5.1 Demographics of subject pool
In total, 74 subjects participated in this experiment; 27 females and 47 males. The average age of the subject pool is 27.1 years with a standard deviation of 10.3 years. Furthermore, the average financial knowledge is 4.7 on a scale from one to seven. Out of the 74 participants, ten participants showed inconsistent behaviour or preferred the certain 2 euros over the certain 3.85 euros in the risk-attitude test. Therefore, the results from their risk-attitude test are not taken into consideration in the analyses since I assume that these decisions are caused by mistakes. The average number of safe choices of the remaining 64 participants was 4.6 out of 10, with a standard deviation of 1.45.
5.2 Results of part 1 of the experiment
To begin with, as specified before, all variables are divided into two or more groups to perform a mean comparison test within these groups. First, a general overview will be provided on how these groups performed in the first part of the experiment. As mentioned before, the subject pool is divided into females and males, participants with high and with low financial knowledge, and risk-loving, risk-neutral and risk-averse participants.
Table 4 Panel A reports the performance of all participants of part 1 in the experiment. On average, 58.5% of all guesses were correct. Next, the performance of participants is also categorised on gender in Table 4 Panel B, risk-attitude in Table 4 Panel C and financial knowledge in Table 4 Panel D. The results show that there are no gender differences in the percentage of CG
Table 4: Overview of participants’ performance in part 1
This table reports the performance of all participants in part 1 of the experiment. Column 2 reports the number of participants, column 3 reports the average number of CG in percentages and the corresponding SD is shown in column 4. Participants are also categorised based on gender (Panel B), risk-attitude (Panel C) and financial knowledge (Panel D). Panel D shows the only difference in performance that is significant at the 5% level, two-sided test. Note that the number of participants in Panel C do not add up to 74. Ten participants gave inconsistent or irrational answers in the risk-attitude test and their risk-attitude measure is therefore disregarded.
Group Number of participants Average number of CG in % Standard deviation in % Panel A: All participants
All participants 74 58.5 22.0 Panel B: Gender Female 27 53.7 24.7 Male 47 61.2 20.1 Panel C: Risk-attitude Risk-averse 20 53.8 18.6 Risk-neutral 28 55.4 23.9 Risk-loving 16 65.6 22.1
Panel D: Financial knowledge
Low 36 69.4 15.9
High 38 48.0 22.0
(women: 53.7% vs. men: 61.2%, p=0.161, t-test). Additionally, there are also no performance differences between risk-averse and risk-loving participants (risk-averse: 53.8% vs. risk-loving 65.6%, p=0.088, t-test). Correspondingly, a one-way analysis of variance (ANOVA) shows that risk-attitude did not affect the percentage of CG (p-value of F-test: 0.227). In contrast, there is a significant difference in the percentage of CG based on financial knowledge (low: 69.4% vs high: 48.0%, p<0.0001, t-test). In the next section, the origin of this highly significant difference will be discussed.
5.3 Results of part 2 of the experiment
This section compares the tendency to AAA within groups, resulted from part 2 of the experiment. Table 5 provides an overview with the different groups and their corresponding number of participants, the percentage of participants AAA and additionally the standard deviation in percentages.
The first thing to note from Table 5 in Panel A is that participants acted on average 41.2% against the advice of financial institutions. Next, Table 5 Panel B shows that there is a significant difference of 8.8% in the tendency to AAA between those groups (lucky: 46.0% vs unlucky: 37.2%, p=0.017, t-test). Furthermore, Table 5 Panel C reports the participants’ tendency to AAA
Table 5: Overview of the participants’ tendency to AAA
This table reports the participants’ tendency to AAA in the experiment. Colum 2 reports the number of participants, column 3 reports the average number of AAA in percentages and the corresponding SD is shown in column 4. Participants are also categorised based on luck in part 1 (Panel B) gender (Panel C), risk-attitude (Panel D) and financial knowledge (Panel E). Panel B and E show the only differences in participants’ tendency to AAA that are significant at the 5% level, two-sided test. Note that the number of participants in Panel C do not add up to 74. Ten participants gave inconsistent or irrational answers in the risk-attitude test and their risk-attitude measure is therefore disregarded.
Group Number of participants Average number AAA in % Standard deviation in % Panel A: All participants
All participants 74 41.2 15.9
Panel B: Luck in part 1
Lucky 34 46.0 14.0 Unlucky 40 37.2 16.4 Panel C: Gender Female 27 36.6 17.3 Male 47 43.9 14.5 Panel D: Risk-attitude Risk-averse 20 36.3 15.7 Risk-neutral 28 42.4 14.2 Risk-loving 16 46.1 14.9
Panel E: Financial knowledge
Low 36 46.9 14.5
High 38 35.9 15.4
based on gender. The results show that there is no difference between women and men (female 36.6% vs male: 43.9%, p=0.057, t-test). Similarly, Table 5 Panel D shows no differences in the tendency to AAA between loving and averse participants (averse: 36.3% vs. risk-loving 46.1%, p=0.064, t-test). This is confirmed by the performed ANOVA showing that the tendency to AAA of different groups based on risk-attitude is not different from each other (p-value of F-test: 0.135). Lastly, in Table 5 Panel E reports a significant difference in the tendency to AAA based on financial knowledge (low: 46.9% vs. high 35.9%, p=0.0023, t-test).
Besides, comparing the decisions by participants after buy or sell recommendations, Table 6 shows that the difference in the tendency to AAA based on buy or sell recommendations is significant (buy: 45.9% vs. sell: 33.3%, p=0.0020, t-test). Moreover, the higher tendency to act against buy recommendations is found in both the lucky and the unlucky group. For the lucky participants, the different reaction to buy and sell recommendations of 13.9% is significant (p=0.0121, t-test), and the difference of 11.5% for the unlucky participants is significant as well (p=0.0486, t-test). Additionally, as for the tendency to act against buy recommendations, the difference between the lucky and unlucky group of 9.7% is significant (p=0.0486, t-test). However, the difference between the groups of 7.3% for sell recommendations is not significant (p=0.2721, t-test)
Table 6: Buy versus sell recommendations.
This table compares participants’ tendency to act against buy advices to their tendency to act against sell advices. Additionally, participants are divided in lucky and unlucky groups to obtain a better view on their reactions to different advices. In all three groups, the difference in reaction to buy and sell recommendations is statistically significant, based on a 5% significant level, two-sided test. Additionally, the difference in reaction to buy recommendations between lucky and unlucky participants is also significant.
Participants’ tendency to AAA
All participants (N=74) Lucky participants (n=34) Unlucky participants (n=40)
Buy 45.9% (20.1%) 51.2% (18.6%) 41.5% (20.5%)
Sell 33.3% (28.1%) 37.3% (25.6%) 30.0% (30.0%)
5.4 Multiple regression analysis
From the previous subsection, it becomes clear that differences exist between lucky and unlucky participants and participants with more and less financial knowledge in their tendency to AAA, when looking at these characteristics separately. In this section, participants are not divided into groups, but instead, variables like the number of CG, risk-attitude and financial knowledge become continuous measures. Moreover, this section combines these variables into one OLS-regression to predict the tendency to AAA.
In Table 7, the coefficient estimates of three OLS-regressions are shown. The first regression, where the number of CG in part 1 is the only independent variable, shows that an increase of one correct guess in part 1, predicts an increase of 0.36 more AAA in the next eight decisions. This increase of 4.6% is statistically significant. However, notice that the Adjusted R2 is just 0.0507, meaning that only 5.07% of the variability of the dependent variable is explained by this independent variable. The second model shows that when all regressors are included in the model, they all appear to be insignificant estimators of the dependent variable. In Section 6.4 is becomes clear that the FE variable should be disregarded from this model. Therefore, the third model is more important. In this model, the gender and risk-attitude variables are the additional regressors to improve the estimate of the causal effects. According to the estimates of Table 7, the CG variable does affect the tendency to AAA. However, including the gender and risk-attitude variables decreases the effect of the performance in part 1 on the tendency to AAA. Moreover, the standard error of the CG regressor increases, resulting in the CG to be less significant compared to the model where the CG variable was the only one. Corresponding to the mean comparison analyses, it appears that both gender and risk-attitude have no effect on the tendency to AAA. Nevertheless, including these two insignificant estimators increases the Adjusted R2 by more than 50%.
Table 7: Multiple Regression Analyses of AAA
This table reports the results of three OLS-regression models. The dependent variable (AAA=) is the number of times participant (p) acted against
the advice of the financial institution in part 2 of the experiment. To estimate this, different combinations of independent regressors are included in the MRA, like the number of questions answered correctly in part 1 of the experiment (CG=), the gender effect (D=ABCDEB), an ability effect (FK=)
consisting of a self-reported scale assessment between skill and luck and the risk-attitude effect (RA=) consisting of the participants’ number of safe
choices in the Holt & Laury (2002) risk-attitude test. A dash (-) is shown for the parameter not estimated in the model. The parameter estimates are presented in the table with corresponding robust standard errors directly beneath the parameter in parentheses. In the last two rows, the RIand the
Adj. RI for each regression are presented.
* Indicate statistical significance at the 10% level (two-sided test) ** Indicate statistical significance at the 5% level (two-sided test) *** Indicate statistical significance at the 1% level (two-sided test)
Model (1): %%%&= ( + (*+,-&+ :;
Model (2): %%%&= ( + (*+,-&+ (./012/3&./012/+ (789%&+ (.M56&+ :;
Model (3): %%%&= ( + (*+,-&+ (./012/3&./012/+ (789%&+ :;
Regressor (1) (2) (3) (*+ 0.364** (0.161) (0.201) 0.178 (0.171) 0.306* (./012/ - (0.279) -0.225 (0.284) -0.284 (78 - (0.087) -0.110 (0.085) -0.115 (.4 - (0.108) -0.119 - ( 2.447*** (0.427) 4.011*** (1.002) 3.171*** (0.670) R2 0.0637 0.1340 0.1191 Adj. R2 0.0507 0.0807 0.0791
5.5 Returns
In part 2 of the experiment, participants could earn points based on their decisions. According to the previous subsections, the results show that participants are subject to the self-attribution bias. In this section, the effects of this bias on participants’ returns will be examined. First the returns of lucky participants will be compared to the returns of the unlucky participants. Next, a regression analysis will show the relation between participants’ number of AAA and their returns for both lucky and unlucky participants.
Table 8 reports the mean return of participants in the second part of the experiment. The difference in return of 0.82 percentage points is quite substantial as the mean return of unlucky participants is about 75% higher than the mean return of lucky participants (p=0.0398, t-test).
Additionally, to estimate the return of participants, the following OLS-regression is performed:
Table 8: Mean return in part 2
This table reports the mean returns and corresponding standard deviations in percentages of participants categorised on their performance of part 1 of the experiment.
Group Number of participants Return in % Standard deviation in %
Unlucky 40 1.92 1.47
Lucky 34 1.10 1.90
In this estimation model, the dependent variable: (9OP(QR&) is the total return in percentages of
participant (p) of all eight rounds in part 2 of the experiment. Next, the independent variables that are included are: the number of AAA in part 2 of the experiment (%%%&), the luck effect (3&STUVW),
and the interaction effect (%%%&∗ 3&STUVW). Table 9 reports a significant negative effect of the
number of AAA on returns. An increase of one AAA, decreases returns with 0.4 percentage points. The negative effect of AAA on returns is even larger for lucky participants. For them, an increase of one AAA results in a decrease in returns of 0.9 percentage points. The Adjusted R2 of 0.26
suggests that the number of AAA is an important estimator of participants’ returns in this experiment.
In this experiment, the unlucky participants earned not only half of what they could have earned if they did not rely on their own judgement of historical stock charts. By performing well in the first part of the experiment, the self-attribution effect made the reliance on participants’ own judgement even worse. Instead of earning 4.03% per round, they achieved an average return of only 1.10%, that is a reduction of 72.7%.
5.6 Limitations
In this section, different threats to internal and external validity will be discussed. To begin with, it is assumed that in this study reverse causation is not present for the simple reason that ,-& is
determined before %%%& and the gender and risk-attitude variables are personal characteristics that
are also not affected by %%%&.
Next, one potential issue with these independent variables is a high correlation between two of them. Table 10 Panel A depicts all correlations between the independent variables included in equation (1) with its corresponding significance. There is one correlation, between number of CG in part 1 and the self-assessed skill/luck ratio, that is remarkably high and significant. This
Table 9: Multiple Regression Analysis of Returns
This table reports the results of the OLS-regression model where the dependent variable (Return=) is the total return in percentages of participant
(p) in all eight rounds in part 2 of the experiment. To estimate this, the following regressors are included in the model: the number of AAA in part 2 of the experiment (AAA=), the luck effect (D=`abcd), and the interaction effect (%%%&∗ 3&STUVW). The parameter estimates are presented in the table
with corresponding robust standard errors directly beneath the parameter in parentheses. In the last two rows, the RIand the Adj. RI for each
regression are presented.
* Indicate statistical significance at the 10% level (two-sided test) ** Indicate statistical significance at the 5% level (two-sided test) *** Indicate statistical significance at the 1% level (two-sided test)
9OP(QR&= ( + (888%%%&+ (STUVW3&STUVW+ (XY;%%%&3&STUVW+ :;
Regressor Equation (2) (888 -0.402** (0.167) (STUVW (1.013) 1.300 (eY;/f1U;XgY -0.500* (0.280) ( 2.447*** (0.427) R2 0.2634 Adj. R2 0.2318
Table 10: Pearson Correlation Matrix and collinearity statistics
This table displays the estimated correlation coefficients between the different independent variables in Panel A. The independent variables’ variance inflation factors are displayed in Panel B. Corresponding p-values are presented beneath the coefficients in parentheses.
* Indicate statistical significance at the 5% level (two-sided test)
Panel A: Pearson Correlation Matrix
,-& 3&./012/ 56& 9%&
,-& 1.0000 3&./012/ -0.2287* (0.0500) 1.0000 56& 0.5979* (0.000) -0.1618 (0.1684) 1.0000 9%& -0.2694* (0.0241) 0.0020 (0.9872) -0.2070 (0.0856) 1.0000
Panel B: Collinearity statistics
Variance inflation factor 1.71 1.65 1.08 1.05
skill/luck question is implemented to distinguish the financial experts from the noobs. In accordance to the theories of Miller and Ross (1975) and Zuckerman (1979) it is not surprising that the participants who performed well in the first part were more likely to address their performance to skills instead of luck, and vice versa, resulting in the high correlation between these variables. However, according to Gunst and Mason (1980), a correlation below 0.6 is not causing multicollinearity and is therefore not alarming.
Another measure that can be used to assess the presence of multicollinearity in the MRA is the variance inflation factor (VIF). The VIFs of all independent variables are depicted in Table 8 Panel B. Here one can observe that the highest VIF is 1.71 meaning the variance increased by 71% because of multicollinearity. This is still below the threshold of a VIF of five1, meaning that
this measure also suggests that multicollinearity in this MRA is not problematic.
Despite these reassuring statements, it can be concluded from the MRA results (Table 7) that there is, in fact, a problem caused by collinearity. Because of the inclusion of the FK variable, which is highly correlated to the CG variable, the latter becomes an insignificant estimator of the AAA. Therefore, one should not pay too much attention to the FE variable in this study. Nevertheless, this does imply that financial knowledge is not important at all.
This brings us to the next topic, the omitted variable bias. Although the way to measure financial knowledge in this study did not work out, it is still expected that it can significantly influence the tendency to AAA. Supposing that financial experts are aware of the efficient financial market hypothesis, it is expected that financial experts are less inclined to AAA compared to people with less financial knowledge. Therefore, a better measure of financial knowledge can improve the results on this topic significantly.
One issue regarding the possibility to generalize the results of this study is the stakes in the experiment. First, the stakes were considerably low, with an expected payoff per participant of about one euro and no initial investment. Second, the stakes did not depend on the participants’ income and capital. It is quite likely that people act differently when deciding about their own money instead of points in an experiment. Besides, it is unlikely that people invest in a stock based on just a historical stock chart.
In summary, the results from the OLS-regressions are not appropriate to forecast individual investors’ tendency to AAA in real-life decisions. As for the external validity, the results show that individual investors relate past performance to skills and their future decisions are affected by this past performance.
1
6 Discussion
In this section, all results will be discussed. First, the results of the different groups will be analysed. Next, the comparison of buy and sell advices are briefly discussed. After that, the different MRAs are reviewed. To conclude, potential improvements of this experiment will be discussed to obtain a better view on the self-attribution bias of individual investors in the future.
6.1 Mean comparisons within groups
In the within group comparison, a statistically significant difference of 8.8% in the tendency to AAA was observed between lucky and unlucky participants. As expected, it is likely that participants who performed well in the first part of the experiment, believed they were skilled although it was just about being lucky. Subsequently, those lucky participants decided to follow the advice of financial institutions less often compared to the unlucky participants, while the advices were valuable to the participants. The participants’ future decisions are based on random past performance. That can be attributed to the self-attribution bias, as hypothesized.
Next to that, differences in gender did not affect the tendency to AAA of financial institutions. In contrast to what the literature suggests, men in this subject group do not show significantly more overconfidence than women in the form of AAA. Although the difference of 0.58 cannot be generalized because it was insignificant, it still showed that men tended to AAA more often than women, in this experiment specifically.
In addition, risk-attitude did also not affect the tendency to AAA. This is probably caused by the diverse effects that risk-attitude can have on investing in stocks. On the one hand, highly risk-averse participants choose the certain values so they do not have to gamble, resulting in acts against all five buy advices. On the other hand, risk-averse participants can choose to always follow the advices to go for the highest expected return in every round, while risk-loving participants can be looking for sensation regardless of the recommendation.
Lastly, the comparison between high and low financial knowledge showed a highly statistically significant difference of 11.0% regarding the tendency to AAA. However, I fear that this is mostly caused by an improper design. From the Results section, it is made clear that the
correlation between CG and FE is almost 0.6. Additionally, the question to learn financial knowledge was asked after the results of part 1 were shown to the participant. Also, as is mentioned before, people address their successes to skills and failures to bad luck. Consequently, participants who performed well in part 1 were more likely to address this to skills and participants who performed worse were more likely to address this to bad luck. Altogether, this caused the relatively high correlation between the two variables. The large difference in AAA between high and low FK groups can be addressed to the facts that the lucky participants were more likely to AAA and that the number of CG is highly correlated with the FK variable. Because of this improper design, the results of the effect of financial knowledge on the likelihood to AAA should be taken with a pinch of salt.
6.2 Buy versus sell advices
Conflicts of interest within financial institutions can lead to differences in credibility between buy and sell advices. Buy advices can have a diversity of meanings. The most obvious one is financial institution expecting the stock price to increase within a limited amount of time. If this turns out to be correct, investors are satisfied with the advice which in turn improves the reputation of the financial institution. If the stock is expected to decrease in the near feature, the same can be argued for a sell advice. However, there are more reasons for financial institutions to issue a buy advice. When financial institutions hold a recommending stock in their portfolio, a buy advice signals that a company is doing well. Subsequently, investors buying the stock could then increase the stock price from which it can profit. Another reason to publish a buy advice, or not to publish a sell advice, on a stock creates goodwill for the financial institution from the recommending company. This can increase the accessibility to gather information of the company and can increase the chance to do business with this company in the future. On average, participants acted 12.7 percent points more against buy advice than against sell advices, this shows that participants behaved according to this phenomenon.
6.3 Multiple Regression Analyses
In accordance with the mean comparisons, the first model shows that the number of CG significantly affects participants’ tendency to AAA. However, when the gender, risk-attitude and financial knowledge variables are included in the regression, none of the variables appear to be significantly influencing the AAA anymore. This is in stark contrast to what is found in the group comparisons where CG was statistically significant and FK was even highly statistically significant affecting the AAA. Before, it was stated that the collinearity between CG and FK is not problematic, since the correlation is below 0.7 and the VIF is far below the threshold of 5. Nevertheless, the collinearity between these variables is likely to be the main factor causing both variables to be insignificant in the MRA where both are included.
By excluding the FK variable from the regression, similar results appear in the MRA compared to the mean comparison analysis. In both analyses the CG appear to be a significant estimator on participants’ tendency to AAA while the gender and risk-attitude measures are insignificant estimators. Since the results from the MRA are similar to the mean comparison analyses, the same explanation can be given for these variables not being a significant estimator of the tendency to AAA. Although the most important variable for this study, CG, appears to be significant in all cases where the FK variable is excluded from the MRA, it only explains 5% to 8% of the variability of the dependent variable. Logically, there are many other factors that play a role in deciding whether to choose the current or future stock value in an experiment where only a recommendation is provided.
Disregarding the FK variable, the OLS-coefficient of the CG is between 0.31 and 0.36 in the first three models. This means that an increase of one correct guess in the first round, increases the number of AAA with 3.9% to 4.5%. It is hard to put these numbers in real-life examples, since it was just an experiment with forced decisions individual investors do not usually make in real-life. From these results, it can be concluded that past performance does affect future decisions. More specifically, good past investment decisions due to luck, causes investors to overestimate their own ability. As a result, these lucky investors are less inclined to follow the advice of financial institution, which is, in general, deteriorating returns.
