How much does expertise reduce behavioral biases? The case of anchoring effects in stock return estimates.
The question of how individuals form stock market return expectations has received increasing attention, and survey-based studies directly asking for these expectations have appeared in recent years (e.g., Welch, 2000; Bange, 2000; Fraser, 2004; Graham and Harvey, 2005; Glaser and Weber, 2005). However, we know very little about how financial advisers form their return expectations. Advisers and other financial market professionals play an influential role in matching financial products and individual investors, and their return expectations are likely to have a significant impact on their clients' asset allocation decisions.
We are particularly interested in finding out if professionals' expectations are affected by behavioral biases. Numerical estimates are often influenced by an initial starting value, a process known as "anchoring and adjustment." (1) Anchoring is one of the three original decisionmaking heuristics, and since the classic study by Tversky and Kahneman (1974), anchoring effects have been documented in studies on, for example, real estate valuation (Northcraft and Neale, 1987), effort or task motivation (Switzer and Sniezek, 1991), and consumers' purchasing decisions (Wansink, Kent, and Hoch, 1998). Anchoring may also be a powerful force in financial markets (see Shiller, 1999, for discussion). Some studies have shown that experts, too, can suffer from anchoring bias (Northcraft and Neale, 1987; Englich, Mussweiler, and Strack, 2006). (2) On the other hand, a vast majority of studies on anchoring, as well as other experimental studies in decision making, have used student subjects. A strong presupposition in the economics literature is that more sophisticated subjects behave fundamentally differently, as they learn from experience to avoid biases and their behavior is also influenced by higher incentives. A series of field experiments utilizing the market for sports memorabilia reported in List (2003, 2004a, 2004b, 2006) support the notion that experience indeed attenuates behavioral biases in general. Hence, anchoring effects may not be so important in practical decision-making contexts.
Previous studies comparing the decision making of financial market professionals and students find that whether or not professionals are less biased depends on the context. Haigh and List (2005) find that the floor traders at the Chicago Board of Trade (CBOT) demonstrate a greater degree of myopic loss aversion than students. Alevy, Haigh, and List (2007) find that students more closely follow Bayes' rule, whereas CBOT professionals are better at assessing the quality of public information, and thus earn higher profits.
In this paper, we carry out three controlled experiments: we asked 300 Scandinavian financial advisers and other professionals about their stock return expectations, while varying the information that is first provided to them. Our paper differs from prior manipulation experiments in finance. Jordan and Kaas (2002) investigate the effect of viewing different types of mutual fund ads on expectations. Bloomfield and Michaely (2004) study the effect of variation in beta and book-to-market to perceptions on expected returns, risk, and mispricing. Glaser, Langer, Reynders, and Weber (2007) study the impact of asking for subjects' expectations in terms of returns versus prices. We, on the other hand, focus on the impact of historical stock market facts presented to the subjects--or the subjects' recollection of those facts--n their future return estimates. For example, in one version we tell the subjects what the historical stock market return has been, and in another version we ask the subjects to provide an estimate of the historical return themselves.
With the notable exception of Bloomfield and Michaely (2004), this is, to our knowledge, the only paper utilizing financial market professionals to study the impact of different background information or presentation format on estimates; other studies have used students. We gather our data on location, using field seminars and professional education courses. We achieve a participation rate of 100% in many events, and the overall participation rate is about 80%. The participants come to the location without knowing that they are about to take part in an experiment. This approach combines a high participation rate, minimal selection bias, and a controlled environment. Given these features, this study can be classified as a "framed field experiment." (3)
Our results show that the skepticism toward experimental evidence with unsophisticated subjects is warranted. We find that expertise indeed significantly attenuates behavioral bias. In a test of the classic anchoring effect applied to stock market return estimates (Experiment 2), students show an effect that is several times higher than with professionals. (4) Nevertheless, we find interesting effects also in the experts' estimates. The more specific results of our three experiments are summarized as follows:
In Experiment 1, professionals filling out questionnaires showing the 4.5% (5) average real return in European countries during the last century expect, on average, 4.6% from European stocks during the next 20 years. In the complementary mode, we ask the subjects to estimate the historical returns themselves. Subjects asked to give self-estimated historical returns expect 3.4 percentage points higher future real returns than those who were shown the 4.5% figure. We have verified that the result is not due to money illusion. The 4.5% is, of course, not just some random number, but one that is also relevant for future forecasts. Still, the clustering of the future estimates around 4.5% is surprisingly strong, given that there could be many reasons to expect the next 20 years to be different from a 100-year history. Furthermore, when we ask about the extent to which the knowledge of history affected the future estimates, we find that those who say it had only a minor effect or no effect at all anchor to the 4.5% just as strongly as those who say history had a major effect. The subjects also anchor tightly to their own historical estimates.
Subjects significantly overestimate historical returns. Goetzmann and Peles (1997) find that individual investors tend to overestimate the past returns of the mutual funds they hold, often by several percentage points. Our results show that this optimism applies to general market returns as well, and this is also demonstrated by the professionals. (6)
In Experiment 2, we provide anchors of 20% or 2%, in pairs of questionnaires, ask for a comparison between the anchor and the future estimate, and then separately ask for the future estimate. This procedure replicates the design of the classic anchoring studies. Professionals show a statistically significant anchoring effect, giving 1.8 to 2.4 percentage points higher return estimates on average, when exposed to the higher anchor. The magnitude of this difference is economically significant, considering recent equity premium forecasts as low as 3%.
Neither a higher level of expertise nor longer work experience among the professionals diminishes the effect in Experiment 2. However, among students the effect decreases along two measures of financial sophistication: being a finance major student and having personal investment experience. The most sophisticated student group (personal investment experience and finance as a major subject) shows an anchoring effect of approximately one-third of the least sophisticated student group (no personal experience and a major other than finance).
In Experiment 3, we investigate the impact of a purely qualitative manipulation, that is, without mentioning any numbers. This constitutes a test of what psychologists refer to as the "basic anchoring effect." The results do not show any statistically significant effects, however.
The main results of the experiments are summarized in Figure 1. The systematic decline of the effect by increasing sophistication, as observed in Experiment 2, is demonstrated in Figure 2. In the remainder of the paper, we start by describing the experimental setting and the data in Section I, and present the results in Section II. We discuss some of the implications of our findings in the concluding Section III.
I. Experimental Setting
We conduct three experiments to study the impact of varying background information and anchoring effects among financial market professionals. Each of the three experiments employs a pair of questionnaires with different modes. In all experimental modes, the subjects are asked for their estimate of the future 20-year real return on the stock market. The modes differ regarding the information presented or asked for, before asking the question on future returns. The participants, sessions, and questionnaires related to each experiment are described in this section.
Data are collected during September 2004 to March 2006 and involve 13 data-gathering sessions, 12 in Finland and one in Sweden. The subjects comprise financial market professionals in 11 sessions and students in two sessions. We conduct three experiments using a between-subjects design. We pool data from different sessions within an experiment, but keep the data on the students separate from those on the professionals.
The participants include field experts such as financial advisers, institutional investors, asset managers, and analysts. The most frequent job title in the professional sample is financial adviser (19%). (7) Other typical backgrounds include analyst, investments expert, broker, wealth manager, stock specialist, and administrative staff, but director and manager levels are well represented. This subject pool is henceforth collectively referred to as financial market professionals, or briefly as professionals.
In addition to the professional subject sample, a control sample consisting of undergraduate finance students is elicited. The survey is carried out in an investments course that these students attend at the Helsinki School of Economics (HSE). The course is mandatory for students majoring in finance, and it is typically their second course in finance. All students attend an elementary finance course and have thus been exposed to the basics of stock markets, including stock returns. The students are at the beginning of their specialization in university business studies, and have limited work experience in financial markets.
Descriptive statistics of survey participants are shown in Appendix A. We attempt to control for the influence of heterogeneous subject characteristics within an experiment. The survey design and the data gathering procedure minimize the heterogeneity in the pairs of samples. Appendix A records the number of participants in each of the experiments and confirms that the parallel samples have near resemblance. The characteristics of the samples in the different experimental conditions are quite similar in the distribution of gender and the years of professional market following.
The total number of participants is 513, of whom 300 are financial market professionals, and 213 are students. Two answer sheets were returned empty and five answer sheets were excluded from the analysis due to obvious misunderstanding of the questions.
The sessions are held in connection with events in which we can reach entire groups of financial market professionals. These events are narrowly targeted field seminars on financial markets, or professional education sessions. In some events, the survey participants are from a single financial services firm, and occasionally they represent different organizational backgrounds but still hold approximately parallel functional posts in the industry. At the beginning of the event, after all the participants have arrived and are seated in an auditorium or a classroom, they are asked to participate in a voluntary study on stock market return expectations. The participants are neither informed of the experiment's manipulative design, the research objectives, nor warned against estimation biases.
The participants come to the event without knowing that an experiment will be conducted. The overall participation rate is about 80%. However, in most sessions we achieved a participation rate of 100%. The overall rate is pushed down by one session with about 50% participation involving only Experiment 3 (the experiments are described in the later sections). In that session, we did not manage to collect the answer sheets on location but had them delivered to us later. We suspect that the lower rate in that session is mainly due to questionnaires being misplaced or forgotten, and not so much to nonparticipation bias.
At the beginning of the experiment, the participants are given detailed instructions on filling out the questionnaires. The participants are first generally advised that their answers can be based on intuitive estimates, as no information sources, including discussion with other participants, are allowed. Prior to filling out the questionnaires, the participants are given detailed instructions on what is being asked, and the relevant terms are defined. Specifically, it is explained that geometric real total return of stocks means that the invested capital is compounded and that this corresponds to how investment returns are usually measured, real return is what is left of nominal returns after inflation has been deducted, and total return means capital gains plus reinvested dividends, and taxes and other costs are ignored. The instructions also state that the historical estimates refer to the countries indicated in the question, and the future estimates concern the European stock market. (8) In addition, the participants are told that by average development the experimenter means an equally weighted average of all countries. One of the authors was present in each session to answer any questions regarding the definition of the terms used.
One experimental session lasts 15 to 20 minutes, of which 5 minutes are spent describing the study and explaining the instructions. As mentioned, the questionnaires ask about stock market return expectations, and the explicit features of the questions are discussed in the next section describing the experimental models. In addition to asking for the stock market estimates, we also request estimates related to volatilities and fixed income instruments, but these data are not utilized in this study.
We make every attempt to distribute the questionnaire sheets randomly to the participants in the events. In each event, the respondents receive questionnaires that relate to one experiment, and the participants are divided into two approximately equal groups: those who receive a questionnaire in one mode, and those getting the corresponding questionnaire in the other mode. In one event with 160 participants, questionnaires related to two experiments were distributed, that is, four questionnaire versions.
We rely entirely on the intrinsic motivation of the subjects to perform well, and offer no compensation or rewards for accuracy. However, a standard assumption in economics is that people react to monetary incentives, and that incentives improve performance. Psychologists, on the other hand, tend to assume that human cognitive processes cannot be easily (if at all) improved by imposing monetary incentives, and that even the opposite can happen. Intrinsic motivation could be compromised by the presence of extrinsic motivation. According to this argument, monetary incentives can crowd out other incentives from which a subject derives utility (for a formal model, see Benabou and Tirole, 2003). Camerer and Hogarth (1999) review 74 experimental studies that use a varying degree of monetary incentives. They find that the presence of a financial incentive improves performance in tasks where one needs to pay close attention (such as memory- and recall-related tasks) and in dull tasks where intrinsic motivation may be low (such as coding words). In other tasks, incentives can actually hinder performance, for example, by causing anxiety. In most cases, however, the presence or absence of a financial incentive does not affect mean performance.
In our experiments, we could not provide the subjects with meaningful monetary incentives based on accuracy, since we ask for 20-year return estimates. However, some characteristics of the experiments suggest that the results would not markedly change, even in the presence of monetary incentives. First, the subjects enter the events without knowing that they will participate in an experiment, so they are not seeking to pocket extra money. Second, people with low intrinsic motivation can elect not to participate, and there are some who make that choice. Third, conditional on participating, we expect the subjects to be motivated to perform well in a task that is central in their work. Fourth, there is no incentive to lie. Disclosing a truthful estimate of 20-year returns can neither be used against the person providing the estimate nor is there any benefit in hiding one's information.
D. Experiment 1: Disclosed versus Self-Estimated Historical Return
In the first experiment, half of the subjects are told that the mean annualized real stock return in European countries in 1900-2000 was 4.5%, which is based on Dimson, Marsh, and Staunton (2002). This mode is denoted as "disclosed." Historical return figures have the potential to be powerful anchors, because they can provide relevant information of future returns. For example, corporations frequently use historical returns as estimates for their cost of equity capital (Graham and Harvey, 2001; Brounen, de Jong, and Koedjik, 2004). Welch (2000) surveys over 200 academic economists for their estimates of the future 30-year equity premium. His questionnaire refers to the widely used Ibbotson Associates return figures. At the time of the survey, the arithmetic average return on the S&P 500 index over the US Treasury bill return since 1926 was 8.2%, and this was shown to the participants. He observes that about one-fifth of the respondents picked an equity premium estimate equal to the particular Ibbotson figure quoted in the survey.
No historical figures are disclosed to the other half of the subjects; instead, they are asked to provide this information themselves ("estimated" mode). The subjects know that their recollection of history cannot be as informative as the correct figure. On the other hand, prior studies have found self-generated anchors to be more robust than externally imposed anchors (Davies, 1997; Mussweiler and Strack, 1999).
If the subjects are conscious of the impact of their knowledge of stock market history on their estimates of future returns, we would expect to see a smaller difference in estimates between the anchor modes for the group of respondents stating a low relevance of history.
Tversky and Kahneman (1986) and Wilson, Houston, Eitling, and Brekke (1996) suggest that the anchoring effect is an unconscious, or automatic, cognitive process. The subjects may feel that their estimates are not influenced by the anchor when they actually are, and vice versa (Wilson et al., 1996). To measure this effect, we ask the subjects to evaluate how much their prior knowledge of historical development affected their estimates of the future. Evaluations are given on a scale of 1 to 5, 1 denoting that the future estimate is completely independent of history and 5 denoting that the future estimate is directly derived from history. We call this the relevance-of-history score. It is similar to Brewer and Chapman's (2002) influence-to-self variable, which sorts respondents into those who feel that their estimates were affected by the anchor and those who do not.
E. Experiment 2: Forced Comparison to a High versus Low Anchor
In Experiment 2, half of the subjects are told that Sweden experienced a particularly good 20-year period in the stock market during 1980-2000, and that the real return for that time period was over 20% per year ("high" mode). The other half were told that the recent 20-year period (1984-2004) in the Japanese stock market was particularly weak: stock prices appreciated strongly until the end of the 1980s, but thereafter the development was very poor, and that the real return for the entire 20-year period was less than 2% per year ("low" mode). The subjects in both modes are then asked whether they think the average development in the European Union (EU) countries' stock markets during the next 20 years will be higher or lower than the anchor value of 20% or 2% that they were provided with. After that, the subjects are asked to state their estimate for the future returns. This experiment most closely resembles the classic anchoring study design in Yversky and Kahneman (1974).
F. Experiment 3: Selecting Countries with Good versus Bad Stock Market Episodes
In this experiment, half of the subjects are asked to identify three countries with the worst 20-year stock market periods somewhere during 1900-2000 ("bad" mode), and the other half are asked for three countries with the best 20-year periods ("good" mode). Both modes present a list of the following countries: Australia, Belgium, Canada, Spain, South Africa, the Netherlands, Ireland, United Kingdom, Italy, Japan, France, Sweden, Germany, Switzerland, Denmark, and the United States. The subjects are then asked to provide a numerical estimate (annualized percentage return) for the single best or worst 20-year history.
The action of choosing countries with prolonged good or bad stock market episodes may increase the accessibility of instances gathered from memory that are consistent with the presented frame. This can give rise to the basic anchoring effect (Wilson et al., 1996), in other words, anchoring in the absence of a forced comparison of the target and the anchor.
G. Synthesis of the Different Modes
The experiments differ in the following respects: Experiment 2 has an externally given anchor, while Experiment 3 utilizes self-provided anchors. In Experiment 1, the nature of the anchor depends on the mode: it is external in the disclosed mode, and self-provided in the estimated mode. Explicit comparison to the anchor is only asked for in Experiment 2. Experiments 2 and 3 contain semantic priming, which means that stock market returns are referred to either in a negative or a positive light in the questions. The disclosed mode of Experiment 1 provides a highly informative anchor, as the historical return figure is disclosed. The estimated mode of Experiment 1, on the other hand, provides no information. Experiment 3 also contains no information, while the Swedish and Japanese priming in Experiment 2 provides some information. The features of the experiments are summarized in Table I. Appendix B presents the English translations of the questions, which were originally in Finnish and Swedish.
Some basic features are common to all the modes. As mentioned, in all modes the final question asks for a subjective estimate of stock market real returns over the following 20 years. In addition to the common estimation task, we ask for some personal information. The questionnaires request the professional subjects to indicate their sex, position in the firm, and years of professional experience related to the stock market. Students indicate their sex, major subject, and whether they have made any stock market investments of their own.
In this section, we present the results of the three experiments and then discuss the role of experience. We also discuss the robustness of the results.
A. Experiment 1: Disclosed versus Self-Estimated Historical Return
This experiment utilizes a pair of questionnaires that compare the effects of an anchor value that is externally provided, informative, and plausible, vis-a-vis a self-generated anchor. In the disclosed mode, the realized past annual real returns on stocks is disclosed to the respondents, whereas in the estimated mode subjective estimates of the past returns are elicited from the respondents themselves. In both modes, the subjects are asked for estimates of future returns.
Table II shows that the future real return estimate is nearly twice as high in the estimated mode (mean = 8.05%) compared to the disclosed mode (mean = 4.62%). The difference of 3.43% between the modes is highly statistically significant (t = 6.50, p < 0.01). The result in the disclosed mode is very close to the actual historical figure of 4.5% disclosed in the questionnaire. Comparing the estimates within each subject shows that the future estimates are often close to the historical figures, irrespective of whether they are self-estimated (estimated mode) or provided (disclosed mode). The future estimate is within [+ or -] 0.5 percentage points of the historical estimate in 47.6% of the answers in the estimated mode, and in 53.1% in the disclosed mode. Overall, the subjects thus anchor strongly to what they perceive to be the historical norm for returns.
This result is in line with the finding of Welch (2000), that equity premium estimates of economists are anchored to the Ibbotson estimates of historical returns. However, utilizing our relevance-of-history score allows further analysis of this issue. A majority of the subjects place a high weight on their knowledge of historical returns when assessing future returns. A score of 4 or 5 is reported in 51.1% of the answers in the estimated mode, and 60.6% in the disclosed mode. To analyze the extent to which forming the future expectation is a conscious effort, the sample is divided, placing scores 4 and 5 (high weight on history) in one group, and scores of 1 and 2 (low weight on history) in the other, and leaving out scores of 3. (9) Results of this test are reported in Panel B of Table II. The group stating low weight on history demonstrates an even larger difference (3.78%, t = 4.97, p < 0.01) between the modes. In other words, those who say that history does not help much in forecasting the future give a much lower estimate of the future when they are told the historical return figure. This result suggests that the subjects are not conscious of the anchor's influence. The larger difference in this low-relevance sample is driven by both a higher mean estimate in the estimated history anchor mode and a lower mean estimate in the disclosed history anchor mode.
The analysis of the relevance-of-history score suggests that the results of Experiment 1 cannot be explained simply by professionals believing that future returns equal past returns, while unaware of the true past returns. Furthermore, if someone believes that a return calculated over 100 years of history is highly relevant for the future, and works in the profession, it seems they should be aware of that return. We find this is not the case. We also show later that the same applies when we limit the sample to those who have more than two years of professional experience with the stock market. Overall, the findings in this experiment support the hypothesis of Wilson et al. (1996) on unintentional anchoring.
It is rational for higher level experts to place more weight on private forward-looking information and correspondingly less weight on history, compared to less qualified experts. This would imply a higher relevance-of-history score for the less qualified experts. To investigate this hypothesis, we classify as higher level experts those with job positions requiring higher analytical skills (strategists, specialists, and analysts), solid knowledge of financial markets (brokers, dealers, bank directors, and portfolio managers), and other high-level professional specialization (lawyers and controllers). These job positions represent 35% of the subjects. The rest of the positions comprise financial advisers and support function titles (e.g., portfolio assistant), and other positions of lower analytical requirements (sales manager and product manager). Some arbitrary cut-offs had to be made in the categorization due to unclear definition of the job position, but these were, as a rule, considered as less qualified experts. In this experiment, 15 participants did not indicate their job position. We find that 55.6% of higher-level experts report a high relevance-of-history score (4 or 5), and 57.5% of the less qualified experts do so. These ratios are, however, statistically indistinguishable, and thus we cannot conclude that higher level experts place less weight on history.
B. Experiment 2: Forced Comparison to a High versus Low Anchor
Anchor manipulation in this experiment involves the combination of numerical facts and semantic priming of historical stock market development. One questionnaire mode (high mode) describes the development of the Swedish stock market during the period 1980-2000, when the annual real return was over 20%. This case is referred to as exceptionally good in terms of general stock market development. On the other hand, the other questionnaire mode (low mode) presents the dismal case of the Japanese stock market, which experienced a real return of less than 2% annually during 1984-2004. This case is referred to as exceptionally weak in terms of general stock market development. The future return estimates are again requested as an average of EU countries.
Table III gives the results of Experiment 2. In the high mode, the mean return estimate for stocks is 9.19%, while in the low mode it is 7.38%. This translates into a difference of 1.81% between the high and low modes (t = 2.38, p < 0.05). The estimates made by financial market professionals are thus significantly biased by the questionnaire mode in the direction predicted by the anchoring hypothesis. Students exhibit an even stronger effect: the mean return estimate for students is 14.14% in the high mode and 6.74% in the low mode. The difference is a striking 7.40% (t = 9.63, p < 0.01).
This experiment involves a forced comparison to anchor values, as respondents identify their belief as to whether the future return will be higher (lower) than a threshold of 20% in the high mode (2% in the low mode). It appears that our professional subjects consider the return threshold of 20% and 2% to be extreme, as none of them estimate future returns outside that range. Two student subjects give estimates slightly under 2%.
The Swedish and Japanese cases presented to the subjects clearly offer much less information about the general level of historical returns than the overall average used in Experiment 1. They nevertheless contain some information. An alternative way to think about these results could be to treat them as a pure information effect: subjects learn a new piece of information and update their beliefs. If one makes this interpretation, it follows that the subjects must have very diffuse priors in order for the new piece of information to have an effect of this magnitude, and they also fail to adjust sufficiently toward the mean. No matter what the exact thought process is, the consequence is the same: the anchor value has a large impact on the estimates.
Anchoring studies in psychology have also used purely random anchors. For example, in the context of stock returns, anchoring to a random number could be tested as follows: ask subjects to write down the last digit of their phone number, and then ask whether future returns will be higher or lower than this figure. However, we did not perform such a test because random anchors are rarely encountered in practical decision-making settings. A likelier scenario in the work of financial market professionals is the presence of some weakly informative anchor, such as the previous year's return from their employer's flagship equity fund.
C. Experiment 3: Selecting Countries with Good versus Bad Stock Market Episodes
This experiment evaluates the basic anchoring effect hypothesis of Wilson et al. (1996). We expect a smaller anchoring effect in this experiment, as semantic priming is likely to be a weaker activator of anchoring and adjustment than numerical anchors.
In this experiment, historical stock returns are framed without explicit numerical information. In the good (bad) mode, the questionnaire refers to exceptionally good (bad) stock market development in the past, and the subjects are asked to identify such cases from a list of countries, and to provide estimates of how much the returns were. The total number of participants in this experiment is 186, of whom 111 are financial market professionals and 75 are students.
Table IV shows the results of Experiment 3. The differences in the mean stock return estimates between the two modes are in the direction predicted by the basic anchoring effect for both professional and student subjects. Estimates in the good mode are 0.55 percentage points higher for professionals and 0.78 percentage points higher for students. However, these differences are not statistically significant.
The analysis of the professional and student samples shows that expertise clearly mitigates anchoring effects. Our data enable analyzing the effect of the degree of expertise further within the two subject pools. In the questionnaire, professional subjects state the number of years that they have followed stock markets professionally. We categorize as experienced professionals those with more than two years of stock market-related work experience. In the majority of the samples, setting the limit to two years implies an exclusion of a nontrivial number of participants, yet without losing too many observations.
The results of the interaction of expertise and anchoring are presented in Table V. Excluding the less experienced professionals slightly reduces the gap between estimated and disclosed modes in Experiment 1 (from 3.43% to 2.92%). However, this exclusion increases the gap between the low and high modes in Experiment 2 (from 1.81% to 2.45%). In unreported results, we also separately investigate higher-level experts in Experiment 2 and find the difference between the conditions for these subjects to be 2.13%. (10)
In the student sample, the importance of field experience is analyzed by comparing students who have made stock market investments themselves to those who have not. Students who engage in stock market investments may be more informed of market developments than students who have no personal incentive to monitor the markets closely. The subject pool of students is almost evenly distributed in this respect in all the subsamples except for the bad mode in Experiment 3. As in the base case in Experiment 3 with the whole student sample, we find no statistically significant anchoring effect in stock return estimates of investor and noninvestor students. In Experiment 2, there is a very large anchoring effect for students with no investment experience: the sample difference between the mean stock return estimates in the high and low modes is 9.57% (t = 9.50, p < 0.01). Students with investment experience also show a strong anchoring effect (5.01%, t = 4.56, p < 0.01), but the magnitude is about half of that of noninvestors.
Figure 2 presents a further analysis of the role of experience in Experiment 2. In addition to the subject categories already discussed, we partition the student sample based on their major subject (roughly half of the subjects are finance majors, the rest major in subjects like accounting, marketing, international business, and economics). It is not clear ex ante whether finance majors, for example, would be more sophisticated in this experiment than students of any major who have personal investment experience. However, we do expect finance students with experience to be the most sophisticated student category, and, correspondingly, students with other majors and no personal experience to be the least sophisticated. We therefore also include these two groups based on the interaction of major subject and experience. The results in Figure 2 corroborate these expectations. While the anchoring effect of the least sophisticated student category is 4.3 times higher than with professionals having two or more years of experience, it is "only" 50% higher in the most sophisticated student category.
E. Robustness Checks
One potential explanation for the relatively high numbers in the estimated mode of Experiment 1 is that subjects confuse real returns with nominal return, and the disclosed mode provides them with information of the general level of real returns. This would be a form of money illusion. To test for this possibility, we run a new questionnaire version that requests the estimates in nominal terms rather than in real terms, and asks for inflation estimates. We collect responses from 22 professionals, all in the estimated mode. We calculate an implied real return estimate of future stock return by subtracting expected inflation from the expected nominal stock return. This gives 7.65%, which is only 0.40 percentage points lower than in Table II when asking directly for real returns in the estimated mode, and the difference is not statistically significant. The implied real historical return is 9.95%, which is actually higher than what we obtain when asking directly for real returns. Based on this result, we conclude that confusion between real and nominal returns is unlikely to drive our results.
To test for the impact of possible extreme estimates, we perform the following outlier analysis. First, we take logarithms of the estimates to produce an approximately normally distributed variable. In each mode, we exclude return estimates located further than 2.58 standard deviations from the mean corresponding to a 99% interval for the normal distribution. There are only a couple of such cases, and accordingly the results remain virtually unchanged. By employing a stricter condition and excluding in each sample one, two, or three highest and lowest values, the sample differences do not disappear and t-statistics become even more significant.
To better account for the differences in the professional subjects' characteristics, we also run a regression where the dependent variable is the future return estimate, and the explanatory variables are a dummy for the questionnaire mode, a gender dummy, a dummy indicating higher-level expertise (see Section II.A for a description of this variable), (log) years of market following, and its squared term. The results are qualitatively the same as those obtained with the univariate tests reported above, that is, the differences between the modes are significant in Experiments 1 and 2, but not in Experiment 3.
In three controlled experiments with 300 financial market professionals, we find significant anchoring effects in their long-term future stock return estimates. The main results of the experiments are summarized in Figure 1. Anchoring is most robust with self-estimated or explicitly given numerical stock market return anchors.
Experimental evidence often raises the question of whether the results hold outside the experimental setting. The main cause for this concern for generalizability is the use of inexperienced subjects. Another concern is the relevance of the actual questions in a practical decision-making context. For example, Tversky and Kahneman's (1974) question of the percentage of African countries in the UN is a completely artificial decision-making situation for most subjects. These concerns should weigh less in our study, given that we have professional subjects, and we ask for a number that is critically important in their work. Hence, the study can be described as a framed field experiment.
We find that expertise indeed significantly attenuates behavioral biases. A test of the classic anchoring effect applied to stock market return estimates shows that the effect obtained with students is several times higher than the effect obtained with professionals. When we condition on the degree of expertise within the professionals, we do not find further mitigation of the anchoring effect. This conclusion is drawn based on the analysis of the subjects with longer work experience or a position involving higher level expertise. This is consistent with the idea that there are limits to debiasing.
The results suggest that financial market professionals may not hold steady return expectations. Perhaps they behave differently in actual conversations with their clients. They have various information sources and analysis tools at their disposal, so perhaps the use of these resources leads to a different answer. However, we believe that the long-term stock market return expectation is such a key figure in the financial industry that the professionals may well have an answer at hand, rather than starting to search for information with every new client. We therefore believe that the subjects' behavior in our experiments is a good proxy for their behavior on the job. Since the advisers are affected by the experimental manipulations, it is plausible that situational factors in a given client conversation also influence the estimates they provide at work. For example, recent returns may serve as an anchor, causing unconscious extrapolation into future return estimates based on only a few data points. This could lead to inconsistent advice.
We did not provide monetary incentives for the subjects in the experiments. Instead, we relied on their intrinsic motivation and asked them to do their best. Although the presence of a financial incentive can lead to better performance in estimation tasks, prior literature shows that it often does not make a difference. As discussed before, some features specific to our experiments also suggest that the lack of monetary incentives probably has only a minor effect on the results. If anything, offering incentives might lead to an even greater effect of expertise. Assuming that all subjects increase their cognitive effort, the experts could improve their performance more if they had more cognitive resources to draw on. This issue could be addressed in future studies.
As financial products increase in number and complexity, investors have a greater need for information and advice in order to make appropriate decisions. When developing investment strategies, educating clients, and implementing investment plans, it is necessary to consider a number of economic factors, of which perhaps the most crucial is the expectation for future equity returns. Our results highlight the importance of providing financial market experts with neutral and informative data on stock market returns. Advisers should also be warned of possible unconscious anchoring effects.
Appendix A. Descriptive Statistics of Experiment Participants
This table reports the descriptive statistics of the survey data with which we test anchoring effects in long-term stock market return estimates. For each of the three experiments featured in this survey, the table reports the number of respondents separately for each sample characterized by the questionnaire modes and the distribution of male and female respondents for both financial market professionals and student subjects. The table also reports sample median, minimum, and maximum number of years the professional respondents report they have followed the financial market for a profession. The total sample includes 22 professional subjects in addition to what is reported in this table. These 22 subjects are used in an additional test asking for nominal instead of real returns, as described in Section II.E. Panel A. Experiment 1 Condition Total Estimated Disclosed Professionals total, no. 47 36 83 Male 29 16 45 Female 16 19 35 Gender not reported 2 1 3 Market Following, Years Median 5 10 Min 0 0 Max 30 30 Panel B. Experiment 2 Condition Total High (20%) Low (2%) Professionals total, no. 30 43 73 Male 13 20 33 Female 14 23 37 Gender not reported 3 0 3 Market following, years Median 6 4.5 Min 0 0 Max 20 25 Students total, no. 61 78 139 Male 40 62 102 Female 21 16 37 Panel C. Experiment 3 Condition Total Good Bad Professionals total, no. 62 60 122 Male 40 38 78 Female 17 20 37 Gender not reported 5 2 7 Market following, years Median 6 8 Min 0 0 Max 31 25 Students total, no. 40 34 74 Male 28 24 52 Female 12 10 22
Appendix B. English Translations of the Questions
Experiment 1--Estimated Mode
* Estimate the historical stock market development during the period of 1900-2000 in Europe; indicate an equally weighted average of those European countries in which a stock exchange has been in operation for the whole period; the countries are Belgium, Spain, The Netherlands, Ireland, the United Kingdom, Italy, France, Sweden, Germany, Switzerland, Denmark. Geometric real return per annum, %?
* Estimate the average stock market development in the EU countries during the next 20 years. Geometric real return per annum, %?
* Estimate how much your prior knowledge on historical development affected your estimate of future development of stock markets. Scale 1-5, 1 = Not at all, my estimate is totally independent of the past, 5 = Very much, I believe that the historical result is directly the best estimate for the future.
Experiment 1--Disclosed Mode
* Same questions and wording as in Experiment 1--estimated mode, but instead of asking the subjects to estimate the historical returns, the questionnaire discloses the historical stock market development of 4.5%, defined exactly as in the estimated mode.
Experiment 1--Estimated Mode, Nominal Returns
* Same questions and wording as in Experiment 1--estimated mode, except that real returns are replaced by nominal returns, and we also ask for inflation estimates.
Experiment 2--High Mode
* Sweden experienced a particularly good 20-year period in the stock market during 1980-2000. The real return for that time period was over 20% per year. How do you think the average development among the EU countries is likely to compare to that number during the next 20 years--will the return be higher or lower than 20% per year?
* Estimate the average stock market development in the EU countries during the next 20 years. Geometric real return per annum, %?
Experiment 2--Low Mode
* The latest 20-year period (1984-2004) in the Japanese stock market was particularly weak. Stock prices appreciated strongly until the end of the 1980s, but thereafter the development was very poor. The real return for the entire 20-year period was less than 2% per year. How do you think the average development among the EU countries is likely to compare to that number during the next 20 years--will the return be higher or lower than 2% per year?
* Estimate the average stock market development in the EU countries during the next 20 years. Geometric real return per annum, %?
Experiment 3--Good Mode
* Which countries, according to your estimate, experienced a particularly good consecutive 20-year period in the stock market real return at some point during 1900-2002? Choose three countries (Australia, Belgium, Canada, Spain, South Africa, the Netherlands, Ireland, the United Kingdom, Italy, Japan, France, Sweden, Germany, Switzerland, Denmark, the United States).
* How much do you think stocks returned in real terms in the best 20-year period, considering all countries and all 20-year periods, per year?
* Estimate the average stock market development in the EU countries during the next 20 years. Geometric real return per annum, %?
Experiment 3--Bad Mode
* Same questions and wording as in Experiment 3--good mode, except for adjective "good" replaced with "bad."
We thank Matti Keloharju, Samuli Knupfer, Jukka Perttunen, Matti Suominen, and the seminar participants at the Finnish Graduate School of Finance Summer Workshop 2006 for comments, and Marja-Leena Sarvikivi for help in translating questionnaires into Swedish. We acknowledge financial support from the Academy of Finland. Kaustia further acknowledges financial support from the Finnish Foundation for Advancement of Securities Markets.
Alevy, J.E., M.S. Haigh, and J.A. List, 2007, "Information Cascades: Evidence from a Field Experiment with Financial Market Professionals," Journal of Finance 62, 151-180.
Alevy, J.E., C.E. Landry, and J.A. List, 2007, "Anchors Away: A Field Experiment on Anchoring of Consumer Valuations," University of Nevada, East Carolina State University, and University of Chicago Working Paper.
Bange, M.M., 2000, "Do the Portfolios of Small Investors Reflect Positive Feedback Trading?" Journal of Financial and Quantitative Analysis 35, 239-255.
Benabou, R. and J. Tirole, 2003, "Intrinsic and Extrinsic Motivation," Review of Economic Studies 70, 489-520.
Bloomfield, R. and R. Michaely, 2004, "Risk or Mispricing? From the Mouths of Professionals," Financial Management 33, 61-81.
Brewer, N.T. and G.B. Chapman, 2002, "The Fragile Basic Anchoring Effect," Journal of Behavioral Decision Making 15, 65-77.
Brounen, D., A. de Jong, and K.G. Koedjik, 2004, "Corporate Finance in Europe: Confronting Theory with Practice," Financial Management 33, 71-101.
Camerer, C.F. and R.M. Hogarth, 1999, "The Effects of Financial Incentives in Experiments: A Review and Capital-Labor Production Framework," Journal of Risk and Uncertainty 19, 7-42.
Davies, M.F., 1997, "Belief Persistence after Evidential Discrediting: The Impact of Generated versus Provided Explanations on the Likelihood of Discredited Outcomes," Journal of Experimental Social Psychology 33, 561-578.
Dimson, E., P. Marsh, and M. Staunton, 2002, Triumph of the Optimists: 101 Years of Global Investment Returns, Princeton, NJ, Princeton University Press.
Englich, B., T. Mussweiler, and F. Strack, 2006, "Playing Dice with Criminal Sentences: The Influence of Irrelevant Anchors on Experts' Judicial Decision Making," Personality and Social Psychology Bulletin 32, 188-200.
Fraser, P., 2004, "How Do US and Japanese Investors Process Information, and How Do They Form Their Expectations of the Future? Evidence from Quantitative Survey Based Data," Journal of Asset Management 5, 77-90.
Glaser, M., T. Langer, J. Reynders, and M. Weber, 2007, "Framing Effects in Stock Market Forecasts: The Difference between Asking for Prices and Asking for Returns," Review of Finance 11, 325-357.
Glaser, M. and M. Weber, 2005, "September 11 and Stock Return Expectations of Individual Investors," Review of Finance 9, 243-279.
Goetzmann, W.N. and N. Peles, 1997, "Cognitive Dissonance and Mutual Fund Investors," Journal of Financial Research 20, 145-158.
Graham, J.R. and C.R. Harvey, 2001, "The Theory and Practice of Corporate Finance: Evidence from the Field," Journal of Financial Economics 61, 187-243.
Graham, J.R. and C.R. Harvey, 2005, "The Long-Run Equity Risk Premium," Finance Research Letters 2, 185-194.
Haigh, M.S. and J.A. List, 2005, "Do Professional Traders Exhibit Myopic Loss Aversion? An Experimental Analysis," Journal of Finance 60, 523-534.
Harrison, G.W. and J.A. List, 2004, "Field Experiments," Journal of Economic Literature 42, 1009-1055.
Jordan, J. and K.P. Kaas, 2002, "Advertising in the Mutual Fund Business: The Role of Judgmental Heuristics in Private Investors' Evaluation of Risk and Return," Journal of Financial Services Marketing 2, 129-140.
List, J.A., 2003, "Does Market Experience Eliminate Market Anomalies?" Quarterly Journal of Economics 118, 41-71.
List, J.A., 2004a, "Neoclassical Theory versus Prospect Theory: Evidence from the Marketplace," Econometrica 72, 615-625.
List, J.A., 2004b, "Substitutability, Experience, and the Value Disparity: Evidence from the Marketplace," Journal of Environmental Economics and Management 47, 486-509.
List, J.A., 2006, "Using Hicksian Surplus Measures to Examine Consistency of Individual Preferences: Evidence from a Field Experiment," Scandinavian Journal of Economics 108, 115-134.
Mussweiler, T. and F. Strack, 1999, "Hypothesis-Consistent Testing and Semantic Priming in the Anchoring Paradigm: A Selective Accessibility Model," Journal of Experimental Social Psychology 35, 136-164.
Northcraft, G.B. and M.A. Neale, 1987, "Experts, Amateurs, and Real Estate: An Anchoring-and-Adjustment Perspective on Property Pricing Decisions," Organizational Behavior and Human Decision Processes 39, 228-241.
Nutt, S.R., J.C. Easterwood, and C.M. Easterwood, 1999, "New Evidence on Serial Correlation in Analyst Forecast Errors," Financial Management 28, 106-117.
Shiller, R.J., 1999, "Human Behavior and the Efficiency of the Financial System," in J.B. Taylor and M. Woodford, Eds., Handbook of Macroeconomics, Vol. 1C, Amsterdam, Elsevier.
Switzer, F.S. and J.A. Sniezek, 1991, "Judgment Processes in Motivation: Anchoring and Adjustment Effects on Judgment and Behavior," Organizational Behavior and Human Decision Processes 49, 208-229.
Tversky, A. and D. Kahneman, 1974, "Judgment under Uncertainty: Heuristics and Biases," Science 185, 1124-1130.
Tversky, A. and D. Kahneman, 1986, "Rational Choice and the Framing of Decisions," Journal of Business 59, 251-278.
Wansink, B., R.J. Kent, and S.J. Hoch, 1998, "An Anchoring and Adjustment Model of Purchase Quantity Decisions," Journal of Marketing Research 35, 71-81.
Welch, I., 2000, "Views of Financial Economists on the Equity Premium and on Professional Controversies," Journal of Business 73, 501-537.
Wilson, T.D., C.E. Houston, K.M. Eitling, and N. Brekke, 1996, "A New Look at Anchoring Effects: Basic Anchoring and Its Antecedents," Journal of Experimental Psychology: General 125, 387-402.
(1) To demonstrate the anchoring effect, Tversky and Kahneman (1974) use the following test: subjects are asked to make a comparative judgment that involves the estimation of a quantity that people typically do not know exactly (e.g., "Is the percentage of African countries in the United Nations higher or lower than 20?"). After that, they are asked to estimate the figure ("What is the percentage of African countries in the United Nations?"). The estimates that people give are strongly influenced by the hurdle rates they are presented with in the first, comparative part, of the question.
(2) See also the recent paper by Alevy, Landry, and List (2007). They conduct a field experiment of anchoring at a sportscard tradeshow, and find an effect only for subjects with less than one year of market experience.
(3) In the taxonomy of Harrison and List (2004), a framed field experiment is one that uses a subject pool of field experts rather than the standard experimental subject pool of students, and employs an experimental frame that can naturally occur in the field context instead of an abstract decision-making situation. Experience of the subjects is generated in the field and is tied to the task.
(4) We use data gathered from questionnaires using 213 business students at the Helsinki School of Economics (HSE) who already possess basic financial literacy. Prior to the experiment, they have all completed an elementary finance course covering the basics of stock markets, and they have also been exposed to the general level of historical stock returns.
(5) This number is based on Dimson, Marsh, and Staunton (2002).
(6) Several papers have documented forward-looking optimism by security analysts (see, e.g., Nutt, Easterwood, and Easterwood, 1999).
(7) of the total number of 300 professional experiment subjects, 45 did not disclose their organizational position.
(8) The first data samples involving 55 participants were gathered in Finland, and in these sessions the subjects were requested to provide estimates for the Eurozone. We then became aware of the possibility to test Swedish subjects. Both Finland and Sweden are EU countries, but while Finland is part of the Eurozone, Sweden is not. We wanted to include the Swedish subjects' home market, and thus extended the set of countries to all EU countries. We applied this definition to all further sessions in Finland as well. This is an apparent inconsistency between the first sessions and the rest of the sessions. However, the mean return estimates for EU countries and Euro countries are statistically indistinguishable, and we do not believe this to be a source of bias.
(9) Alternatively, the division can be performed by including scores of 3 in the group with a low weight on history. The median score is 4, so this produces a more even split. The results are similar under both definitions.
(10) The definition of a higher level expert is explained in connection with the results of Experiment 1.
Markku Kaustia, Eeva Alho, and Vesa Puttonen *
* Markku Kaustia is an Assistant Professor, Eeva Alho is a Ph.D. Student, and Vesa Puttonen is a Professor in the Department of Finance and Accounting, Helsinki School of Economics, Helsinki, Finland.
Table I. Summary of the Experiments Experiment 1 Experiment 2 Anchor level Self-generated Semantic priming and versus externally a forced comparison to imposed numerical a numerical value anchor Modes Estimated and High and low referred to as disclosed Estimated High Decision frame Estimate the annual The Swedish stock market (i.e., the stock market real provided exceptionally presentation form return in European good real return in of return history) countries in the 1980-2000, over 20% 20th century annually. Do you expect the future return in Europe to be over or under 20%? Disclosed Low Historical real The Japanese stock return for stocks market provided given as 4.5% exceptionally poor real return in 1984-2004, less than 2% annually. Do you expect the future return in Europe to be over or under 2%? Additional Dependence of esti- Forced comparison to mates on historical 20% (2%) knowledge directly asked from subjects (relevance-of- history score) Participants 83 professionals 73 professionals and 139 students Experiment 3 Anchor level Semantic priming Modes Good and bad referred to as Good Decision frame Select from a list of countries those (i.e., the three with the best episodes of 20-yeas presentation form periods in stock market real returns of return history) Bad Select from a list of countries those three with the worst episodes of 20-year periods in stock market real returns Additional Subjects asked to estimate the return of the best (worst) episode Participants 122 professionals and 74 students Table II. Experiment 1 with Self-Generated versus Externally Imposed Informative Numerical Anchor This table reports the mean estimates for the future 20-year stock market real return in annual terms made by financial market professionals in Experiment l. In the estimated condition, the subjects are first asked to provide an estimate for historical returns themselves (1900-2000, an average of 11 European countries). In the disclosed mode, the correct historical return figure of 4.5% is disclosed in the questionnaire. Diff. indicates the sample difference, estimated less disclosed. The standard two-sample t-test on difference is reported in parentheses. Panel A presents the results for the full sample, while Panel B divides the subject pool in two based on the self-assessed relevance of prior knowledge of historical stock market development for the estimate formation (relevance-of- history score). The score takes values from 1 to 5, with 5 denoting the highest weight on history and 1 the lowest. Low weight on history refers to scores 1-2 and high weight on history to scores 4-5. Condition Dif. Estimated Disclosed Panel A. Full Professional Sample 8.05 4.62 3.43 *** (N = 43) (N = 33) (6.50) Panel B. Sample Partitioned Based on Relevance-of-History Score Low weight on history 8.59 4.81 3.78 *** (N = 11) (N = 7) (4.97) High weight on history 7.61 4.76 2.85 *** (N = 22) (N = 20) (4.05) *** Significant at the 0.01 level. Table III. Experiment 2 with Forced Comparison to a Numerical Anchor This table reports the mean estimates for the future 20-year stock market real return in annual terms made by financial market professionals and students in Experiment 2. In the high condition, subjects are told about the exceptionally good real returns during the recent 20-year period in Sweden, quoting 20% real returns. In the low condition, subjects are told about the exceptionally bad real returns during the recent 20-year period in Japan, quoting 2% real returns. The respondents then make a subjective estimate of whether the future return in Europe will exceed or underperform the 20% or 2% threshold. Diff. indicates the sample difference, high less low. The standard two-sample t-test on difference is reported in parentheses. Condition Diff. High (20%) Low (2%) Professionals 9.19 7.38 1.81 ** (N = 29) (N = 43) (2.38) Students 14.14 6.74 7.40 *** (N = 60) (N = 75) (9.68) *** Significant at the 0.01 level. ** Significant at the 0.05 level. Table IV. Experiment 3 with Semantic Priming This table reports the mean estimates for the future 20-year stock market real return in annual terms made by financial market professionals and students in Experiment 3. In the good condition, the subjects select what they think are the three countries with the best 20-year stock market episodes from a given set of countries (Australia, Belgium, Canada, Spain, South Africa, the Netherlands, Ireland, the United Kingdom, Italy, Japan, France, Sweden, Germany, Switzerland, Denmark, and the United States). In the bad condition, the subjects are asked to identify the three countries with the worst episodes. Diff, indicates the sample difference, good less bad. The standard two-sample t-test on difference is reported in parentheses. Condition Diff. Good Bad Professionals 8.11 7.56 0.55 (N = 55) (N = 56) (0.79) Students 9.62 8.84 0.78 (N = 39) (N = 34) (0.82) Table V. Further Analysis of the Impact of Experience This table reports the mean estimates for the future 20-year stock market real return in annual terms made by financial market professionals and students in all experiments. In the professional subject sample, respondents with less than two years of market following have been excluded. The student subject pool is divided into two groups based on having personal experience of stock market investment. Dif indicates the sample difference between the parallel experiment modes, and the standard two-sample t-test on difference is reported in parentheses. Panel A. Experiment 1 Condition Diff. Estimated Disclosed Professionals (> 2 years) 7.88 4.96 2.92 *** (N = 30) (N = 24) (6.63) Panel B. Experiment 2 Condition Diff. High (20%) Low (2%) Professionals (> 2 years) 9.60 7.15 2.45 *** (N = 25) (N = 31) (2.88) Students Investing experience 13.13 8.12 5.01 *** (N = 28) (N = 36) (4.56) No investing experience 15.03 5.46 9.57 *** (N = 32) (N = 39) (9.50) Panel C. Experiment 3 Condition Diff. Good Bad Professionals (> 2 years) 7.84 7.40 0.44 (N = 36) (N = 41) (0.50) Students Investing experience 9.61 8.97 0.64 (N = 20) (N = 9) (0.44) No investing experience 9.63 8.79 0.84 (N = 19) (N = 25) (0.61) *** Significant at the 0.01 level. Figure 1. Mean Expected 20-Year Real Stock Return Estimates across the Experiments by Subject Category In the estimated condition of Experiment 1, subjects give first an estimate of historical stock returns, while the historical return of 4.50% is disclosed to subjects in the disclosed condition. In the high condition of Experiment 2, subjects are told about exceptionally good returns during a recent 20-year period in Sweden, quoting 20% real returns. In the low condition, they are told about exceptionally bad returns during a recent 20-year period in Japan, quoting 2% real returns. In conditions good and bad of Experiment 3, subjects are asked to select three countries with exceptionally good or bad stock market episodes during the last century. Financial Market Professionals Experiment 1 Estim 8.1 Disc 4.6 Experiment 2 High 9.2 Low 7.4 Experiment 3 Good 8.1 Bad 7.6 Students Experiment 2 High 14.1 Low 6.7 Experiment 3 Good 9.6 Bad 8.8 Note: Table made from bar graph. Figure 2. Anchoring Effect in Experiment 2 by Subject Type In the high condition, subjects are told about the exceptionally good real returns during the recent 20-year period in Sweden, quoting 20% real returns. In the low condition, subjects are told about the exceptionally bad real returns during the recent 20-year period in Japan, quoting 2% real returns. The respondents then make a subjective estimate of whether the future return in Europe will exceed or underperform the 20% or 2% threshold. Difference in Estimated Future Returns Between Conditions (%) Student, Other Major, No 10.6 Experience Student, No Experience 9.6 Student, Other Major 8.4 Student, Finance Major 6.1 Student, with Experience 5 Student, Finance Major, with 3.7 Experience Professional > 2 Years 2.5 Note: Table made from bar graph.
|Printer friendly Cite/link Email Feedback|
|Author:||Kaustia, Markku; Alho, Eeva; Puttonen, Vesa|
|Date:||Sep 22, 2008|
|Previous Article:||Tax-adjusted discount rates with investor taxes and risky debt.|
|Next Article:||Corporate debt issuance and the historical level of interest rates.|