Investment Committee Decisions: Potential Benefits, Pitfalls, and Suggestions for Improvement

Transcription

1 5/6/2010 5:01 PM J. W. Payne Last changes 4/8/2010 Investment Committee Decisions: Potential Benefits, Pitfalls, and Suggestions for Improvement John W. Payne Fuqua School of Business Duke University To appear in Perspectives on Behavioral Finance, A. Wood (Ed.) 1

2 Abstract Investment committees (teams) are responsible for many billions of dollars in investments worldwide. And, as the world of financial decision making becomes more complex and dynamic, the potential value of committee based decisions will increase. This chapter asks, and answers, three questions regarding the quality of investment committee (team) decision making. 1) When and why do collections of individuals perform better than the average individual judge? 2) When and why do groups or teams perform worse than the average decision maker? In particular, when might groups (committees) amplify, not mitigate, individual decision biases? 3) How might the processes of investment committee decision making be improved? It is argued that the quality of investment committee decisions is likely to matter most where individual judgmental biases also matter most, and therefore, if committees are going to be used to manage investments then the processes of good committee decision making must also be managed, and managed continually. 2

3 Introduction Investment committees (teams) are responsible for many billions of dollars in investments worldwide. Included in the decisions made by investment committees are asset allocations, judgments about future market conditions, choices regarding specific investments, and the hiring (and firing) of money managers. Most committees or teams meet face-to-face and use a strength in numbers decision rule (e.g., majority or consensus type scheme) 1. Committee or team-based decision making is also common in other areas of financial decision making. For example, it has been estimated that almost 60% of the actively managed equity mutual funds are managed by teams, up substantially from just 30% or so in 1992 (Bliss, Potter, & Schwarz, 2008). There are a number of reasons why committees or teams might be used to make financial decisions. For instance, in the area of nonprofits, it has been suggested that having a number of important people on an investment committee helps fund raising because such people will be more likely to give, and to get others to give. There is related evidence that the more members of a group are involved the making of a decision, the greater the commitment to the implementation of that decision. In the area of mutual fund management, it has been suggested that teams provide a more stable management structure than an individual (Kovaleski, 2000). That is, funds managed by committee are less subject to the impact of a single manager leaving the company. Another reason that has been offered for group or committee decision making is the diffusion of responsibility for a poor decision. 1 A consensus decision rule is often a 2/3ths majority rule in practice. That is, once 2/3ths of a group decide on a judgment or choice that tends to become the consensus decision. 3

4 The primary reason for the popularity of committee decision making, however, is the view that committees or teams will make better financial decisions. With teams, committees, or groups, it is felt that there will be more knowledge or information to be shared. This information argument for committee decision making appears even more compelling as the world of financial decision making becomes more complex and dynamic. Consequently, it becomes less likely that any single individual will have sufficient information and skills for good decision making. Gary Becker, the 1992 Nobel economics laureate, acknowledged this problem when he stated that the main problem with the modern financial system based on the widespread use of derivatives and securitization is that while financial specialists understand how individual assets functions, even they have limited understanding of the aggregate risks created by the system ( Wall Street Journal, 10/7/08). The potential of having more information, and the sharing of that information, is perhaps the strongest argument for group decision making. Teams also can provide error checking of the reasoning being presented during decision making. That is, simply having more of the relevant information is not enough; information must also be processed in the right way. Groups are seen as helping to avoid poor reasoning. For example, individuals frequently neglect or do not properly use baserate or distributional information when making probability judgments. The hope is that a member of a committee might correct such an error in reasoning by another member of the committee. In addition, humans are subject to random slips and mistakes in judgments. If the judgments of individual members of a group include random error or noise, then a statistical combination those individual judgments will cancel out those 4

5 errors and lead to the wisdom of crowds (Surowiecki, 2004). Finally, committees provide a way to incorporate different values, such as risk attitudes or a concern with social responsibility, into a decision. All of the reasons above suggest that investment committees have the potential of being a good way to manage investments. And, there is some empirical support for the superiority of group decision making. For example, Binder and Morgan (2005) find that group decisions about monetary policy questions are, on average, better than individual decisions. On the other hand, some view investment committees much more negatively they don t meet often, they act slowly, and when they act, they tend to make the wrong decisions (Robert Jaeger, quoted in the October, 2004, issue of Foundation and Endowment Money Management). Manzoni, Strebel, and Barsoux (2010) have also recently argued that diversity in thinking on corporate boards can backfire and lead to poorer performance. Substantial empirical support for this more negative view of group decision making exists. For example, Barber and Odean (2000) report that investment clubs performed worse than individual investors. To make the empirical results more unclear, Bliss, Potter, and Schwarz (2008) recently reported no statistically or economically significant differences in performance between individually-managed and team-managed mutual funds (p. 115). Thus, the empirical evidence on the performance of team-based versus individual financial decision making is mixed; although, the overall pattern of results do suggest limits of group decision making. For a recent review of the quality of group decision making in general, see Kerr and Tindale (2004). The quality of investment committee decisions will matter most when the investment decisions involve more complex valuations and subjective forecasts; i.e., 5

6 where individual judgmental biases may also matter most. Lo and Mueller (2010) recently provided a characterization of levels of uncertainty in markets. Committee decisions may matter more when the levels of uncertainty are higher, i.e., levels 4 and 5, i.e., when there are limits on what we can deduce or can know about the underlying phenomena generating whatever data is available. Also the quality of investment committee decisions will matter most when there are significant constraints on arbitrage possibilities. Examples of such decisions include investing in newer companies, extreme growth stocks, hedge funds, and new types of financial instruments. Jeremy Siegel in the Wall Street Journal (9/16/08) suggests, for instance, that Groupthink prevailed when the risks and values of new types of subprime real-estate loans and credit instruments were assessed. This chapter asks, and answers, three questions regarding the quality of team or investment committee decision making. 1) When and why do collections of individuals perform better than the average individual judge? 2) When and why do groups perform worse than the average decision maker? In particular, when might groups amplify, not mitigate, decision biases? 3) How might the processes of investment committee decision making be improved? The rest of this chapter is organized as follows: First, a simple model of individual judgment is presented, followed by a classic model of group judgment. Next, an illustration of how groups can outperform the average individual judge is presented. Third, a number of examples are presented where groups amplify, not mitigate, decision biases. Reasons why poor group decision making happens are discussed in terms of cognitive, motivational, and social interaction factors. A key point is that poor group 6

7 decision making occurs, to a large extent, because the members of the group are human, not because the members of the group are bad or individually not smart. The chapter concludes with suggestions for improving committee or team financial decision making. The focus of those suggestions is on the process of group decision making, and the need for continued monitoring of group processes. Truth, noise, and bias Individual judgments. Individual judgment can be thought of in terms of three components. The first is truth. That is, the part of a judgment that reflects the true state of the world or coherent preferences. Judgments also frequently include an element of random error or noise. That is, unpredictable deviations from truth. The third component is bias or predictable deviations from truth. Over the past four decades a growing body of research has shown that human judgments and choices frequently exhibit predictable biases. Ariely (2008) refers to such biases as predictably irrational behaviors. An example of a bias that is of great relevance to financial decision making is the overconfidence that people often show in the precision or accuracy of their judgments. Another bias, one of the most important in human judgment, is the tendency to search for confirming rather than disconfirming evidence of an initial hypothesis. A third decision bias that is related to preferences is the highly contextdependent nature of choice behavior. For instance, sometimes A is preferred to B when C is the third option in a choice set but B is preferred to A when D is the third option, even through neither C or D is ever chosen. To the extent that a judgment only includes a truth component plus random error (or the bias component is relatively small) then a group judgment is typically better than 7

8 the average of the individual judgments. This is because the group judgment will cancel out the noise component - the larger the group the better. This is the fundamental principle behind the wisdom of crowds. Although, it turns out that going from a single judge to even a small group (n=3 or better) can improve judgment substantially 2. See Hogarth (1978) for more on why aggregating even a few people s estimates usually suffices to boast accuracy. This positive feature of group or combined decision making does not require that the collection of individuals defining a group even meet. One could take a collection of individual judgments and just combine those judgments statistically (a simple average works well). An example of the wisdom of even small groups of judges is given below. On the other hand, to the extent that a decision is subject to substantial amounts of individual biases in judgment or choice, making that decision in a group often does not mitigate the biases seen in decisions. Instead, as reviewed below, group decision making can actually amplify biases in judgment and choice. That is, N heads can be worse than one head. Group judgment. The classic model of group performance (Steiner, 1972) sees the actual performance of a group as a function of the potential productivity of the group members minus process loss components that cause a group to perform at a level below its potential. One might also include the possibility of process gain due to the positive or synergistic aspects of a group of individuals working together. Group potential is a function of the number of independent judges with task-relevant information and skills making up a group. Group potential if often defined in terms of a statistical combination 2 A survey of investment committees by Arnold Wood (Martingale Asset Management) found that investment committees ranged from 3 to 12 in size with the median being 7 members. 8

9 of individual judgments. The concept of independent judges is stressed above because simply adding more and more members to an investment committee is not going to help if the added members share common background (knowledge, beliefs, and values) with existing committee members. Generally, diversity in knowledge and thought processes is more important in group composition than the size of the group. An example of a process loss, to be discussed later, is when a member of a group does not fully share the information that he or she might have that is relevant to a decision. Error checking of the facts or the reasoning being presented by one person by another member of the group is an example of a potential process gain. Unfortunately, the ubiquitous finding across decades of research is that groups usually fall short of their potential productivity they exhibit process loss (Kerr & Tindale, 2004),. While process loss may be a common result with group decision making, it does not occur in all cases. There are problem solving tasks in which groups come close to their apparent potential. Further, process interventions in group decision can mitigate process losses. Finally, Kerr and Tindale (2004) report that process gains, where group performance is better than any individual or a statistical combination of individual member efforts, have proven elusive or modest at best. While process gains have proven elusive, as discussed below, there are a number of studies showing that group judgments are frequently better than the average individual judgment. The best individual judgment is frequently better still, but it is often hard, unfortunately, to identify the best judge before the decision is made. Example of Group Accuracy in Judgment- N heads can be better than one. 9

10 Figure 1 shows a picture of a container full of nickels. How much money, in total, do you think is in that container? Now imagine that three individuals were asked that question as individuals. How well do you think the average individual (one of those three judges picked at random) would do on that estimation task? How well in terms of accuracy might the group of all three individuals do if asked to reach a consensus estimate? How well might that consensus group judgment do in comparison to a simple averaging of the three individual estimates? Recently threee colleagues (Jack Soll, Al Mannes, and Lehman Benson) and I asked 177 undergraduates to estimate the amount of money in the container (shown in Figure 1), first as individuals and then as part of groups of three persons. The correct answer is $ A measure of how close an estimate is to the correct answer is provided by the absolute deviation from the true answer. At one extreme, a randomly selected individual from the crowd of 177 people was found to deviate (+ or -) from the truth by a value of $8.33. Most people underestimated the amount of money in the container. At the other extreme the combined (average) estimate of all 177 respondents deviated from the truth by $4.55 (average estimate of $17.05). That is, there is a 45% improvement in the estimate by moving to the wisdom of crowd provided by the average of all 177 respondents. Picking just three judges at random, averaging their estimates, would lead to a 28% improvement in the estimate as compared to the randomly selected individual. The mean deviation for the average of three randomly selected individuals (a small group) was $6.02. Interestingly, the actual performance of the interacting groups of size three was worse, with an average of $6.77 (a 19% improvement rather than a 28% improvement). In part, this decrease in performance seems to have been the result of the 10

11 behavior of groups where there were two people with similar estimates (referred to as the middle and near judge) and one person with a substantially different estimate (referred to as the outlier judge). Generally, the opinion of the third judge whose estimate was further away (the outlier) was overly discounted. That is, the process used for reaching a group decision effectively reduced the size of the group from 3 to closer to 2. Interestingly, the discounting of the outlier opinion increased the more the outlier s opinion deviated from the opinions of the other two members of the group. While this makes sense if deviation of opinion is a sign of less knowledge by the outlier, it also has the effect of reducing the likelihood that the individual opinions to be combined will bracket truth. That is, one judge will produce an estimate that is below truth while another judge will produce an estimate that is above truth. Similar results were found in a study were the task was to forecast changes in month sales on the basis of cues like the level of website traffic. While outliers were generally less accurate than the two judges that were more in agreement, including the estimate of the outlier was important in collections of opinions doing better than any individual judgment. The results described above are common in the group judgment literature. Groups generally outperform a randomly selected individual. However, interacting groups frequently do worse than a simple statistical average of the individual judgments. Groups both are better but also show evidence of process loss. Why a committee might perform poorly: Sources of process loss There are many sources of process loss in group or committee decision making. Some of the sources are more cognitive and include poor information sharing and the amplification of biases in judgment through the search for confirming information. Other 11

12 sources are more motivational in nature and include lower levels of effort when one is part of a group ( social loafing ), and the introduction of goals, such as the desire to conform to a group opinion, that can lower accuracy. Also, influence in a group may be a function of factors that are not related to knowledge or skills. For instance, people who talk a lot may have more influence on a group judgment than is actually warranted. Or, as suggested above, one may treat opinions that are most similar to your own as necessarily being more valid. Finally, there is an illusion of group effectiveness that often exists. That is, people believe groups do much better than individual judges even though the evidence is mixed, at best. Do committees mitigate cognitive biases? As noted above, research has identified many biases in individual judgments and choices. See Weber and Johnson (2009) for a recent review. These individual thinking biases clearly impact financial decision making. A number of studies have investigated whether or not group decision making mitigates or amplifies or has no effect on the existence of biases in consensus decisions. Below are a few illustrations of this research. Overconfidence in judgment. Overconfidence in the accuracy and precision of one s knowledge is sometimes viewed as possibly the greatest deterrent to rational investment (J. Clements, Wall Street Journal, 2/27/2001). To illustrate one type of question asked in overconfidence studies consider the following: Over the next year, what do you expect that the average S & P 500 return will be (+ or -) %? Obviously the return could be lower (worse). Please give a lower bound to your estimate such that there is a 1-in-10 chance that it will be less than %. On the up side, please give an upper bound to your estimate such that is a 1-in-10 chance that the return will be 12

13 greater than %. Another form of an overconfidence question takes the following form: Next year (12 months from now), the competitor (in your industry) with the largest market share will have: a) less than or equal to 30% of the market, or b) greater than 30% of the market? What is the probability (.5 to 1.0) that your answer a) or b) is correct? A frequently used measure of the quality of such subjective judgments is to assess the proportion of times events said to have a certain probability (e.g.,.8) of occurrence (or of being correct) do in fact occur. A well calibrated probability assessor would be one where the frequency of occurrence was, in fact, 80%. Similarly, a well-calibrated 80% confidence interval around an estimate, such as the percent return of the S & P 500, should contain the actual return about 80% of the time. In general, people are not well-calibrated. For example, in a study of 7000 forecasts of the S & P 500 returns by top corporate executives (Ben-David, Graham, and Harvey, 2007), only about 38% of the forecasts were within the 80% confidence interval. Further, it appears that the gap between confidence and accuracy grows wider as the amount of information available to the judge increases (Tsai, Klayman, & Hastie, 2008). That is, confidence goes up quickly with more and more information but accuracy increases at a much slower rate. Do groups mitigate the overconfidence effect? Several studies have investigated this question. One by Plous (1995) nicely illustrates the findings. Individuals and groups of those individuals were asked to assess 90% confidence intervals for ten questions. Assuming perfect calibration one would expect that for 9 out of the 10 questions the true answer would fall within the 90% confidence interval. This did not happen for individuals or groups. Instead, for only about 3.1 of the 10 items did the true answer lie 13

14 within the confidence intervals assessed by individuals. Interacting groups assessment were only slightly better (4.2 out of 10). Had the individual judgments been combined statistically, the performance would have been much better (7.4 out of 10) indicating process loss. Interestingly, the groups studied by Plous thought they would be much better, on average, than the individuals. That is, people thought that groups would be much better than they actually were. This illusion of group effectiveness partially explains why group decision making is so popular. Sunstein and Hastie (2008) point out that deliberation among group members tends to increase confidence in the group s judgment or decision, whether or not such deliberation actually increases judgmental accuracy. In part this is due to the fact that deliberation usually reduces the variance in members judgments over time as people with each other. And, consensus is often taken as a sign of correct thinking. Interestingly, this same deliberation process can increase the variance in judgments on the same problem between groups. Planning fallacy. A common judgmental bias that is related to overconfidence is the planning fallacy (Buehler, Griffin, & Ross, 1994). Planning is defined in terms of the estimate of the time needed to complete a project. One striking result is that when people are asked to give optimistic and pessimistic estimates of the time needed to complete a project, more than 50% of the projects took longer than the pessimistic estimate. Another result was that people who were 70% confident in their best estimates of the time needed to complete a project were only correct about 40% of the time. 14

15 One reason for the planning fallacy seems to be that thoughts are focused more on the path to likely success of a project than on potential impediments. In addition, the distribution of actual outcomes experienced in the past for similar projects was given little or no weight. This discounting of past knowledge is a form of base-rate information neglect, a common judgmental bias itself. The planning fallacy seems slightly stronger when groups make the estimates. For example, Buehler, Messervey, and Griffin (2005) found that individuals tended to underestimate the time needed to complete a project (estimate = days, actual = days). Groups were even more optimistic (estimate = days). Similarly, in another study the actual task completion time was 2.30 days, on average, and the individual estimates were 1.87 days and the group estimates were 1.07 days. To summarize, groups do not seem to mitigate common judgmental biases. There are even times when groups seem to amplify the biases. In the next section this amplification effect is illustrated with preference tasks. Sunk costs. One of the most common decision errors is the sunk cost phenomenon. This phenomenon is also known as escalation of commitment. Essentially the phenomenon refers to the observation that people frequently persist in a project more than might be justified on the basis of projections regarding future costs and benefits, when there are prior costs or investments in that project. It has been suggested one reason for the sunk cost effect is the reluctance of people to admit mistakes. In other words, one way to hide a little mistake is to bury it under a potentially bigger one. Whyte (1993) has studied the frequency of sunk cost related decisions made by individuals and by groups. In a control condition, a project was presented to both 15

16 individuals and groups without any reference to a prior investment (sunk cost). Based on the description of the project, 29% of the individuals and 26% of the groups were in favor of going ahead with the project. However, in the case of a different set of individuals, when some sunk cost information was presented, about 69% of the individuals decided to go ahead with the project, showing the sunk cost effect. For groups, the sunk cost effect was even stronger. In that case, 86% of the groups voted to go ahead with a project when sunk cost information was presented. Further, in groups where only a minority of the members of the group initially favored escalation, about 50% of the decisions ended up in favor of continuing the project. In contrast, groups with an initial minority in favor of abandonment of the project only had about 2% of the final group decisions in the direction of abandonment. This suggests that an incorrect form of economic reasoning was actually more persuasive in group discussion than the more correct form of economic reasoning. Context effects in preference. Context-independence refers to the assumption that the relative ranking of any two options should not vary with the addition or deletion of other options. This is sometimes called the principle of independence of irrelevant alternatives. It is often viewed as a fundamental property of rational choice behavior. Despite its logical appeal, decision makers do not always satisfy context-independence in their choices. For example, Simonson and Tversky (1992) report a study in which given a choice between $6 and a Cross pen, only about 36 percent of the people selected the Cross pen. However, when a clearly inferior pen (a decoy) was added to the choice set (a pen selected only about 2% of the time), the percentage of people selecting the Cross pen over the $6 rose to 46%. 16

17 Slaughter, Bagger, and Li (2006) explored whether or not this context effect would occur with group decision making. The task was to select an employee based on ratings of sales and service potential to consumers. (This task is clearly related to investment committee decisions regarding the hiring of money managers.) One candidate for the marketing position (named Johnson) had an average rating in terms of sales but a high rating in terms of service. Another candidate (named Smith) had a high rating in terms of sales but only and average rating in terms of service. For half of the individual decision makers and half the groups, the choice problem was to select between Johnson, Smith, and a third candidate (named O Brien - J) who was clearly inferior to Johnson. O Brien - J had a similar high rating on service as Johnson but only a low rating on the sales dimension. For the other half of the individual decision makers and groups, the choice problem was to select between Johnson, Smith, and a third candidate (named O Brien S) who was clearly inferior to Smith. O Brien S had a similar high rating on sales as Smith but only a low rating on the service dimension. The percentage of individuals choosing Johnson over Smith when the third option was dominated by Johnson (O Brien J) was 83% compared to 59% when the third option was dominated by Smith (O Brien S). This effect was slightly stronger with group decisions. Ninety percent of the groups selected Johnson when that was the dominating candidate compared to only 49% when Smith was the asymmetrically dominating candidate. That is, groups may exhibit more context-sensitive preference than individuals. Polarization of attitudes and risk-taking. One potential plus of group decision making is the ability to bring to bear on a decision a diversity of attitudes and values. An implicit assumption is that the diversity of attitudes and values will cause less extreme 17

18 positions to be adopted by a group. The evidence is clear, however, that group discussion often leads to the polarization of attitudes, not the mitigation or compromising of attitudes. An excellent example of this effect is provided by Schkade, Sunstein, and Hastie (2007). In that research groups of citizens from Boulder Colorado (a predominantly liberal city) and groups of citizens from Colorado Springs Colorado (a predominantly conservative city) were asked to deliberate and reach consensus about three policy issues global warming, affirmative action, and civil unions for same-sex couples. The major effect of deliberation as part of a group was to make group members more extreme in their views than before they started to talk. Another related, and important, finding was that both liberal and conservation groups became more homogeneous; deliberation reduced internal diversity, see the earlier discussion of how this can also lead to greater confidence in decisions. The Schkade et al. (2007) results also suggest that initial diversity in a group beliefs and values will not persist over time. In an investment context, this polarization effect suggests that if you have a committee made up of a majority of slightly risk-taking people, the committee decisions will often exhibit even more risk-taking. On the other hand, if you have a committee made up of a majority of slightly risk-averse people, the committee decisions will often exhibit even more risk-aversion. Clearly this result is important for investment committees when dealing with the issue of asset allocation, among other issues. Myopic loss aversion refers to suggestion that people are myopic (short-sighted) in evaluating outcomes over time and are more sensitive to losses than to gains. Benartzi and Thaler (1995) suggest the concept of myopic loss aversion as an explanation of the equity premium puzzle. Sutter (2007) has shown that groups as well as individuals are 18

19 prone to myopic loss aversion in that they are more risk averse when considering investments in shorter evaluation and commitment periods of time. He does suggest, however, that group decision making attenuates the degree of loss aversion. This attenuation is due to the fact that in his study three-person teams invested more in a risky asset than individuals regardless of time period. To summarize, N heads are sometimes better than one and sometimes worse. In particular, groups do not always mitigate judgmental biases and risk attitudes. Why might groups amplify decision biases and attitudes? More generally, what are some of the reasons for process loss in group decision making? We next turn to a brief discussion of a few of the reasons for poor group decisions. The reasons are divided into cognitive factors and motivational factors, although actual committee decisions will typically include a variety of both types of factors. What causes process losses? Poor information sharing. For groups to be most effective there needs to be both different information held by the different members of a group, and that the different information must also be shared among the group members. Different information held by different members of a group reflects the group s potential. Sharing the unique and different information held by different members of a group is an important part of the decision process. Consequently, over the past two decades there has been much research devoted to investigating the role of shared versus unshared information on decisions. Much of this work was pioneered by Stasser and Titus (1985) and is referred as hidden profile decision problems. 19

20 To illustrate this type of research, imagine that Kate, Ken, and Keith are members of a three person investment committee who have the task of selecting between one of three candidates (A, B, and C) for a position as a money manger. There are eight items of information (x1 to x8) about the candidates on job dimensions considered relevant to the selection task, e.g., a rating of technical skill. To make things easier, assume that each item of information takes on just two possible values, 1) positive or 2) negative/unknown. Now consider two information conditions shared and unshared. In the shared information condition, all three members of the committee, Kate, Ken, and Keith, know all the information about each candidate. It is known by all members that candidate A has positive ratings on all eight job dimensions. It is known by all members that candidate B has positive ratings on dimensions x1 to x5 and negative ratings on dimensions x6 to x8. Similarly, it is known by all committee members that candidate C has negative ratings on dimensions x1 to x4 and positive ratings on the remaining four dimensions. Knowing all the information, it is clear that candidate A is the superior choice. Often however information is not fully known to all members of a committee. Instead some members know more about some things and other members know more about other things. This is called the unshared condition. For example, imagine that Kate knows the following information: Candidate A has positive ratings on three dimensions x1, x2, and x3. However, Kate knows nothing about the ratings for candidate A on the other dimensions. Kate does know that candidate B has positive ratings on five dimensions x1, x2, x3, x4, and x5. Kate also knows that candidate C has positive ratings on four dimensions, x5, x6, x7, and x8. Committee member Ken, on the other hand, knows the same information about candidates B and C as everyone else but only knows that candidate A has positive 20

21 ratings on three dimensions: x4, x5, and x6. Finally, committee member Keith shares the information on candidates B and C but knows only that candidate A has positive ratings on the three dimensions: x6, x7, and x8. If the committee members in the unshared condition described above were to fully share all their information, then the group would be back in a shared information condition. Again the choice of candidate A would be obvious. Unfortunately that is not what often happens. Instead, choice of candidate A becomes much less likely in the unshared condition. Instead candidate B, and sometimes candidate C, become more likely to be selected by the group. The reason is that people tend to first share the information that is already known to everyone. Shared information is also discussed more often. Another reason is that people seem to adopt a position of advocating the candidate they think was best initially, i.e., candidate B, and present the information in support of that position rather than sharing all their information. The bottom line is that all the information known to the members of the group is frequently not shared during the decision making process. This failure to completely share unique information is an important source of process loss. The search for confirming information. One of the important, and most common, biases in human judgment is the tendency of people to search out information that tends to confirm (rather than disconfirm) previously held beliefs. As Francis Bacon noted, The human understanding when it has once adopted an opinion (either as being the received opinion or as being agreeable to itself) draws all things else to support and agree with it. And though there be a greater number and weight of instances to be found 21

22 on the other side, yet these it either neglects and despises, or else by some distinction sets aside and rejects, in order that by this great and pernicious predetermination the authority of its former conclusions may remain inviolate. Information that supports a previously held position tends to be subject to a can I believe it test. Information that does not support a position tends to be subject to a much more stringent must I believe it test. Further, an item of information that is ambiguous in its meaning tends to be interpreted as supporting the already held opinion. Confirmation bias is a form of path-dependence in judgment that characterizes behavioral models of human judgment and choice. Unfortunately, this tendency towards a confirmation bias seems to be amplified during group decisions. Schulz-Hardt, Frey, Luthgens, and Moscovici (2000), for instance, found that the confirmation bias in the search for supporting rather than conflicting information was significantly stronger for groups than for individuals. Individuals did show the confirmation bias but to a lesser extent. Further, the larger the majority in a group that was in favor of the initially preferred option, the stronger the confirmation bias. More recently, Kerschreiter, Schulz-Hardt, Mojzisch, and Frey (2008) demonstrated that one factor contributing to a confirmation bias in group information seeking is the high confidence that groups can develop in the correctness of their decision. Highly confident groups show a strong confirmation bias. Further, groups that are more homogeneous in their initial preferences are more confident. Remember, that a group s high confidence in the correctness of their decision may reflect overconfidence 22

23 not accuracy. Consequently, the impact of confidence on confirmatory information seeking is likely to amplify poor decision making. At an even more general level, Benabou (2009) argues that groups, organizations, and markets can be reinforce individual errors in beliefs where contagious exuberance can take hold and collective delusions occur. One factor that can contribute to groups reinforcing rather than dampening errors in beliefs is social conformity, discussed next. Social conformity. Poor group decision making can result from social conformity pressures and other factors related to social interactions. Often, for example, people will modify their opinions in the direction perceived to be consistent with opinions held by others in a group. The classic demonstration of a social conformity effect is by Asch (1956). In that work Asch showed that even a straightforward perceptual judgment (the line of a line) can be influenced by the wrong opinions expressed by other people in a group. There is also some evidence that social conformity pressures can lead group members to suppress divergent opinions, decide quickly in order to avoid unpleasant tensions within a group, and defer to a respected leader s position. McCauley (1998) has argued in a similar vein that poor group decision making results from seeking to preserve friendly relations in a group based on the personal attractiveness of group members. Sunstein and Hastie (2008) refer to this as reputational cascades in that information that is expressed earlier in deliberations tends to get reinforced over time because people are reluctant to express a contrary opinion in order to avoid losing the good opinion of the person who spoke earlier. Poor decision making can also result from members of a group seeking to support an attractive leader. Often criticism of ideas is seen as criticism of the individuals behind those ideas, and thus something to avoid. This social interaction 23

24 action effect also can be seen as a leading to a confirmation bias in information processing. More generally, see the special issue in Organizational Behavior and Human Decision Processes, Vol. 7, 1998, on the topic of Groupthink (Janis, 1972) and the social interaction determinants of good and bad group decision making. Benabou (2008) offers a formal model of when collective denial of reality in more or less likely in groups. Summary of Quality of Group Judgments This chapter has provided a few examples of when groups do better than individuals in judgment and choice tasks, and when groups do poorer. For other examples see Kerr and Tindale (2004) and De Dreu, Nijstad, and van Knippenberg (2008). A summary of the results from the extensive literature on group performance follows. First, when groups are faced with tasks where once a correct solution is proposed the answer is clear, groups do better than individuals. That is, groups do better in solving Eureka tasks. The major danger in such tasks is when the shared conceptual model of what is a correct is flawed. Further, sometimes people treat tasks as more intellectual than they really are, forgetting the critical role of assumptions. During the late 1990s, for instance, many people shared the view that a new internet economy had emerged in which old rules no longer applied. Today we are dealing with the results of a shared view that housing prices could only go up. We are also dealing with the shared belief that the wealth of information now available to financial managers meant that risk could be understood (modeled) and priced much better than before (see Breeden, 2009, for a discussion of the use and misuse of models in investment management). Second, on more judgmental tasks such as estimation, where the primary source of error is random noise, then groups tend to be better than the average individual. 24

25 However, for estimation and forecast tasks you are often better off by simply taking a statistical average of a collection of individual forecasts or estimates. Having the members of a group meet and reach a consensus forecast or estimate may be worse than statistical averaging. Interacting groups, compared to statistical groups, tend to add noise, which lowers the overall validity (truthfulness) of judgment. That is, interaction among group members can produce more unpredictably in judgments (see, for example, Schkade, Sunstein, and Kahneman, 2000). An exception to the recommendation for statistical averaging is when you have individuals with very different levels of forecasting ability, and you are able to identify the better forecasters. In that case, adding the opinions of the poorer forecasters through statistical averaging will hurt. However, evidence suggests that people are not that good at identifying the better versus poorer forecasts. People who are the most overconfident in their forecasts may also be seen as better. In addition, we often tend to view that people whose forecasts are more in agreement with our own are better. People also tend to view themselves as more accurate than the average judgment, and consequently tend to overweight their own opinion when revising opinions in light of the opinion of an advisor. For more on how, and how well, identify and use other expert opinions, see Soll and Larrick (2009). Suggestions for Improvement While there are clearly reasons to be skeptical regarding the promise of investment committee decision making it is unlikely that investment committees will disappear. One recommendation for improving investment committee decision making is simply to become aware of the pitfalls (traps) that can affect group decisions. Awareness 25

26 of decision traps can help (Russo & Schoemaker, 2002). In addition, the literature on group decision making suggests a number of other ways in which groups might make better judgments and choices. In this section of the chapter a few of the best established suggestions for improving committee or group decision making are discussed. Selection of committee members. Arnold Wood and I did a survey a few years ago of investments committees and found that investment committees were very homogeneous in membership. Over 90% of the members were white males, with most of those 60 years or more old. An interesting question, for which we do not have data, is how homogeneous the committee members were in educational backgrounds and other experiences. However, given that group formation tends to be guided by the principle of similarity among potential group members, a high degree of shared backgrounds (e.g., same graduate schools), experiences, and viewpoints is likely to be the case. The more the backgrounds and experiences of a committee s members are shared, the more likely it is that their judgments will be correlated. That is, they will exhibit shared biases in their judgments. This has implications for how to go about selecting the members of an investment committee. For instance, there are some modeling results regarding group judgments suggesting that a smaller committee (n = 4) with less correlation in judgments (e.g., r = 0) would perform better than a much larger committee (n=16) with even modest levels of correlations (e.g., r =.30). Thus, one suggestion is to put resources into trying to find a small number of relatively independent judges rather than spend resources on getting a lot of people on a committee. This suggestion is also consistent with the old idea that an increased size of a group can be a curse in terms of coordination costs. Note, this suggestion is in terms of the accuracy of judgments that are made. One might want a 26

27 larger committee if that is related to fund raising, not the quality of the decisions to be made. Because deliberations tend to reduce the internal diversity of groups over time, another suggestion is to systematically change the membership of an investment committee over time. This procedure will incur some greater management costs but will increase the diversity of thought that is one of the reasons for using committee decision making. In the section above it is argued that diversity in group membership (information, perspectives, skills, and values) enhances the quality of decisions. Research supports the benefits of diversity of group membership in terms of knowledge, etc. And, as argued above, investment committees may not be diverse enough in terms of group decision making. Nonetheless, recent research does suggest that preference diversity across group members may come at a decision cost. Nijstad and Kaps (2008) have shown that preference diversity, particularly in terms of disliking different options, can lead groups to become more decision averse. That is, increased preference diversity can lead groups to be more likely to refuse to decide. Training of committee members. If the majority of the members of a group exhibit a judgmental bias, then group decisions tend to amplify that bias. This is particularly true if the group uses some form of majority rule decision scheme. Consequently, individuals need to be trained to avoid decision biases, rather than counting on groups to overcome those biases. That is, do not count on committees to correct for systematic biases in judgment. 27

28 An idea that has been suggested for improving group behavior is to have each individual member of a group think independently about a problem before meeting in a group. To the extent that such a procedure is adopted it also suggests the importance of the training of each individual committee member. This suggestion also relates to the need to actively manage the exchange of information, as discussed next. Herzog and Hertwig (2009) have suggested that individuals can be trained to capture some of the benefits of collective wisdom by using an approach they call dialectical bootstrapping. The idea is to reduce an individual s error in quantitative estimates by averaging a person s first estimate with a second one that harks back to somewhat different knowledge. One way to encourage accessing different knowledge is to present people with the first estimates and then ask them to assume that your first estimate is off the mark think about a few reasons why that could be. Which assumptions and considerations could have been wrong? what do these new considerations imply? Was the first estimate rather too high or too low? (p.234). Herzog and Hertwig report that this procedure boasts accuracy but does not quite emulate the wisdom of many independent judges. Management of information sharing. Poor information sharing is one of the key reasons for poor group decisions. An important role for any group leader is to actively manage the information sharing process. Manzoni, J, Strebel, P, and Barsoux, J. (2010), for instance, argue that a key factor in making diversity membership on boards a good thing is the role of the leader as a facilitator of discussion. The leader needs to make it clear that people were selected for membership in the group, at least in part, because of the unique information they were expected to bring to various decisions. The sharing of 28

29 unique information is to be encouraged. Also, leaders need to actively work at balancing the participation of the group members in the sharing of information. One important way to encourage more information sharing is to separate the discussion of the pros and cons of a proposed solution to a problem from the person who first proposed the solution. Critical thought about proposed solutions should be encouraged by the leader of a group. Criticism of a person behind a solution should be discouraged. Leaders should also emphasize the importance of information exchange between the members of a committee rather than on persuading others about the correctness of one s position. Finally, leaders should try to prevent a premature consensus on decision refusal (Nijstad & Kaps, 2008). Two other leadership suggestions by Russo and Schoemaker (2002) for avoiding poorer group decisions are 1) the leader should make sure that the group listens to minority views, and 2) the leader should withhold his or her ideas at first. They also mention a custom in some Japanese companies that the member of a group with the lowest status should speak first, followed by the next lowest, etc. The idea is that this will mitigate a tendency for the lower ranked person to withhold his or her opinion from a fear of contradicting a higher ranked person. Whether the causes of poor group decision making are cognitive, motivational, or due to social interaction effects, there is a common thread of a confirmatory bias in judgments and choices. As noted above, one of the most significant individual biases in judgment is the search for and interpretation of information in a way that confirms a prior or favored belief. Many of the process losses of group decision making result from an 29

30 amplification of this confirmatory bias. That is, groups may be even more susceptible to a confirmation bias in judgment than individual judges. Accountability for process. Groups should agree upon what constitutes good decision processing early in deliberations. For example, one should get agreement that the group will include base-rate information in probabilistic forecasts, that the group will seek disconfirming as well as confirming information, and that sunk-costs are not relevant to future decisions. Unfortunately, too often agreement on process is viewed as luxury item that should be minimized when groups meet due to time pressures. This is a mistake. Time spent on getting agreement on processes of committee decision making is seldom wasted. Further, the group should take joint responsibility for monitoring the groups decision processing. Every member of the group should value the contribution of unique, and sometimes conflicting, information and beliefs, and seek such information during deliberations. That is, the group should be accountable for the process, not just the outcome, of a decision. A simple procedure for increasing process accountability that has been used in experimental studies is to set up (or pretend to) a meeting in which the decision-making processes used in a prior decision would be reviewed and discussed. Evidence that process accountability can alleviate the poor sharing of information is provided by Scholten, van Knippenberg, Nijstad, & De Dreu (2007). Russo and & Schoemaker (2002) suggest an audit of group decision making processes. Finally, the focus on good decision reasoning, information sharing, etc., must be a continuing activity. Most of the pitfalls of group or investment committee decision making are human errors that occur very naturally. Even very smart people who are 30

31 well motivated can be part of a poor group or committee decision making process. While a group might start off with the best of intentions, and with good practices, it is easy to fall back into more natural human tendency that can hurt group decisions. Group, and leader, accountability for good decision processing must be a continuous activity. Conclusion There are good reasons why investment committees (teams) are responsible for many billions of dollars in investments worldwide. As the world of financial decision making becomes more complex and dynamic, the potential value of committee based decisions will increase. Making decisions as a group also increases feelings of participation (ownership) of the decision, confidence in the decision, and consequently willingness to implement the decision. Feelings of decision ownership maybe particularly important in an increasingly uncertain world where even good investments can be expected to have bad outcomes from time to time. The quality of the decisions can also be higher when the majority source of error in judgments is due to noise. Collective decision making does tend to cancel out random judgmental errors particularly when individuals are more diverse in the knowledge and thought processes and therefore their individual judgments are likely to bracket truth. Unfortunately, the potential pitfalls of poor group decision making are also likely to increase in size and frequency as an investment committee deals with more complex, uncertain, and dynamic judgmental tasks. The purpose of this chapter is to increase awareness of some of the potential pitfalls (traps) of group decision making. Hopefully, an awareness of traps will help people avoid those traps. Awareness of a few of the 31

32 many ways in which the processes of group decision making can be enhanced should also help to improve investment committee decisions. There are methods available to increase the potential benefits of investment committee decisions while mitigating the potential pitfalls of group decisions. To close, if committees are going to be used to manage investments, the processes of good committee decision making must also be managed, and managed continually. 32

33 References Ariely, D. (2008). Predictably irrational: The hidden forces that shape our decisions. Harper Collins. New York. Asch, S. E. (1956). Studies of independence and submission to group pressure. Psychological Monograms, 70. Barber, b. & Odean, T. (2000). Too many cooks spoil the profits: The performance of investment clubs. Financial Analyst Journal Benabou, R. (2009). Groupthink: Collective delusions in organizations and markets. Working paper 14764, National Bureau of Economic Research. Benartzi, S., &Thaler, R. (1995). Myopic Loss-Aversion and the equity premium puzzle, Quarterly Journal of Economics, 110, Ben-David, I., Graham, J. R., & Harvey, C. R. (2007). Managerial overconfidence and corporate policies. Working paper. Duke University. Binder, A. S., & Morgan, J. (2005). Are two heads better than one? Monetary policy by committee. Journal of Money, Credit, and Banking, 37, Bliss, R. T., Potter, M. E., & Schwarz, C. (2008). Performance characteristics of individually-managed versus team-management mutual funds. The Journal of Portfolio Management, Breeden, D. T. (2009). The use and misuse of models in investment management. CFA Institute Conference Proceedings Quarterly, Buehler, R., Griffin, D., & Ross, M. (1994). Exploring the "planning fallacy": Why people underestimate their task completion times. Journal of Personality and Social Psychology, 67,

34 Buehler, R., Messervey, D., & Griffin, D. (2005). Collaborative planning and prediction: Does group discussion affect optimistic biases in time estimation? Organizational Behavior and Human Decision Processes, 97, De Dreu, C. K. W., Nijstad, B. A., & van Knippenberg, D. (2008). Motivated information processing in group judgment and decision making. Personality and Social Psychology Review, 12, Herzog, S. M. & Hertwig, R. (2009). The wisdom of many in one mind: Improving individual judgments with dialectical bootstraping. Psychological Science, 20, Hogarth, R. M. (1978). A note on aggregating opinions. Organizational Behavior and Human Performance, 21, Kerr, N. L., & Tindale, R. S. (2004). Group performance and decision making. Annual Review of Psychology, 55, Kerschreiter, R., Schulz-Hardt, S., Mojzisch, A., & Frey, D. (2008). Biased information search in homogeneous groups: Confidence as a moderator for the effect of anticipated task requirements. Personlity and Social Psychology Bulletin, 34, Kovalesky, D. (2000). More mutual fund companies take a team approach. Pensions and Investments. 32. Lo, A. W., & Mueller, M. T. (2010). WARNING: Physics envy may be hazardous to your wealth. Working paper, March 19, 2010, MIT. Manzoni, J, Strebel, P, & Barsoux, J. (2010). Why diversity can backfire on company boards. Wall Street Journal, R3. McCauley, C. (1998). Group dynamics in Janis s theory of groupthink: Backward and forward. Organizational Behavior and Human Decision Processes, 73,

35 Nijstad, B., A., & Kaps, S. C. (2008). Taking the easy way out: Preference diversity, decision strategies, and decision refusal in groups. Journal of Personality and Social Psychology, 94, Plous, S. (1995). A comparison of strategies for reducing interval overconfidence in group judgments. Journal of Applied Psychology, 80, Russo, J. E., & Schoemaker, P. J. H. (2002). Winning Decisions. Doubleday, New York, NY. Schkade, D. A., Sunstein, C. R., & Kahneman, D. (2000). Deliberating about dollars: The severity shift. Columbia Law Review, 100. Schkade, D., Sunstein, C. R., & Hastie, R. (2007). What happened on deliberation day? California Law Review, 95, Scholten, L., van Knippenberg, D., Nijstad, B. A., & De Dreu, C. K. W. (2007). Motivated information processing and group decision making: Effects of process accountability on information processing and decision quality. Journal of Experimental Social Psychology, 43, Schulz-Hardt, S., Frey, D., Luthgens, C., & Moscovici, S. (2000). Biased information search in group decision making. Journal of Personality and Social Psychology, 78, Simonson, I & Tversky, A. (1992). Choice in context: Tradeoff contrast and extremeness aversion. Journal of Marketing Research, 29, Slaughter, J. E., Bagger, J., & Li, A. (2006). Context effects on group-based employee selection decisions. Organizational Behavior and Human Decision Processes. 100,

36 Soll, J B., & Larrick, R. P. (2009). Strategies for revising judgment: How (and how well) people use others opinions. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35, Stasser, G. & Titus, W. (1985). Pooling of unshared information in group decision making: Biased information sampling during discussion. Journal of Personality and Social Psychology, 48, Steiner, I. D. (1972). Group process and productivity. Academic Press, New York. Sunstein, C. R., & Hastie, R. (2008). Four failures of deliberating groups. Working Paper No. 401, University of Chicago Law School. Surowiecki, J. (2004). The wisdom of crowds. Random House, New York. Sutter, M. (2007). Are teams prone to myopic loss aversion? An experimental study of individual versus team investment behavior. Economic Letters, 97, Tsai, C. I., Klayman, J., & Hastie, R. (2008). Effects of amount of information on judgment accuracy and confidence. Organizational Behavior and Human Decision Processes, 107, Weber, E. U. & Johnson, E. J. (2009). Mindful judgment and decision making. Annual Review of Psychology, 60. Whyte, G. (1993). Escalating commitment in individual and group decision making. Organizational Behavior and Human Decision Processes. 73,

37 Figure 1 37