Elke U. Weber
Professor of Psychology, Columbia Business School

"The Determinants of Risk-Taking"

It is a real pleasure to be here. I want to give you a different disciplinary perspective on risk taking. In particular, I want to decompose risk taking into two components: risk perception and true risk attitude. Over the next 45 minutes, I want to convince you of the utility of making that distinction.

In this talk, I will start with the economic and finance theory of risk taking. We will then deviate from this theory in several ways. First, we will look at the power of the situation. I will argue that risk is not an immutable attribute of a risky choice or investment option but often varies as a function of when the decision is made and by whom it is made. The question of who made the decision also brings up the topic of individual and cultural differences. We will talk about the best ways to capture, model and explain those differences. In the context of risk, I will touch on rare events and how we as individuals react to rare events in terms of how information has been conveyed to us. I will conclude with some possible applications to your areas of interest.

Let's start with economic and finance theory. There are two traditional economic and finance models. The first is expected utility theory (originally axiomatized by Von Neumann and Morgenstern), which assumes that people do transform outcomes subjectively from outcomes to utility of outcomes. This subjective transformation can be a concave or convex function. A concave utility function is evidence for risk aversion. People will settle for equivalents that are smaller-than-expected value. A convex function is evidence for risk-seeking behavior. Risk-return models like the capital asset pricing models are essentially a description of different types of utility functions. Even though psychologically people talk about a trade-off between greed and fear, operationally we talk about a trade-off between expected value and variance. Both of these approaches treat the riskiness of an option as an inherent and invariant attribute. The only variable that we can assign to individuals or groups in this scenario is risk attitude or risk tolerance.

Technically, risk attitude is a way to describe a utility function, and the utility function is only a way to describe a behavior. However, people often assign different attributes to the term "risk attitude." Often, risk attitudes are interpreted as personality traits, and we use them to make decisions in our organizations. We have psychometric scales that measure your degree of risk aversion. Risk attitude is often used when making hiring decisions. We send people to business schools to try to make them more risk neutral in their decision making. We often use risk attitudes to match clients to individual advisors. Banks often match their top clients with an advisor who has the same risk attitude, and I am working with a farm coop in Argentina that is trying to match the risk attitudes of their farmers with those of their technical advisors.

What is wrong with viewing risk attitude as a personality trait? Personality traits by definition are supposed to be characteristics of an individual that are invariant across situations. If you describe someone as timid, you expect them to behave timidly across situations. If a person is risk seeking, you expect that person to exhibit risk-seeking behavior across situations. If something is not invariant across situations, it is not a "trait."

The problem is that risk-taking behavior in an individual does vary across situations. There are many examples of this. Managers may be risk neutral when investing company money, but when they get home, they may be more risk averse with their personal investments. Rock climbers apparently take a lot of recreational risk, but they may shy away from taking social risks. In fact, the technical advisors from Argentina that I mentioned earlier are often exasperated with the risk aversion of the farmers that they advise. Some of these advisors have started farming themselves because they think that they can make more money by being risk neutral or even risk seeking, but after two or three years, they find themselves becoming just as risk averse as the other farmers.

In pari-mutuel betting, very few people bet on long shots at the beginning of the day. By implication, a risky horse seems to be one with a small probability of winning. At the end of the racing day, more and more people bet on the long shots. There are a number of different possible explanations for this. Do the risk-seeking bettors only come at the end of the day? Another explanation is that the perception of what constitutes a "risky" horse has changed during the day. At the end of the day, money has shifted from your pockets to those of the race track. You might have to go home and confront your spouse with the fact that the bonus is gone. The only horse that gives you some possibility of recouping those losses might be the long shots. You are still betting on what you think is the less-risky horse, but your perception of what is less or more risky has changed as a function of the situation.

When you look at this systematically, when you put people in lots of different risky situations and measure their risk-taking behavior, you find that there is very little consistency in risk taking in the same individuals across different domains from gambling and investing to health, recreation and social situations.

Let me digress to talk about a topic that Michael Mauboussin has written about eloquently: fundamental attribution error. When you see a behavior occurring-for example, your secretary is often late for work-you can attribute that behavior either to the person or the situation. If a person is late for work, she might be unreliable or not very conscientious, or it could be that she has situational reasons like child care problems. Given that we can attribute a behavior either to the person or to the situation-even though it is often a combination of the two-we tend to attribute the behavior far too much towards personality. We will tend to think that our secretary who is late is unreliable, and we might even fire her and hire someone new, while all along, the real problem was an easy-to-fix child care problem.

When this tendency was first talked about in the 1970s, it was described as the "fundamental attribution error." Even though the experiments were done using undergraduate students at the University of Michigan, we do find this behavior in other groups within Western cultures. When we looked at groups in Eastern cultures, however, we found a much smaller tendency to over-attribute to personality rather than the situation. This tendency, then, is not quite as "fundamental" as we used to think.

For my purposes, however, the fundamental attribution error absolutely applies. Economic analysis was developed by Western white males, and economic analysis tries to explain all instances of risk aversion or risk taking in terms of personality traits. We have been ignoring the power of the situation. The situation encourages an individual to act in risk seeking ways in one situation and risk-averse ways in other situations. If you really want to explain and ultimately to predict behavior for macroeconomic policy reasons or otherwise, you really need to understand both situational and chronic determinants of risk taking. Obviously, it is true that there are constant and stable variations in individual risk taking, including age, gender and sensation-seeking. These factors underlie some of the variability that we see in risk-taking situations, but we cannot forget the power of the situation. It is very hard to do this with a single parameter.

Risk attitude is not a constant. An individual does not always have the same reaction to a perceived risk (either a pleasurable rush of excitement or an unpleasant sensation that we want to get away from). Some people enjoy the stress response more than others, but we must also take into account situational factors. How can we do that?

Behavioral models of risk taking have tried to add more variables and parameters to the menu in order to allow us to do that. The successor to expected utility theory is prospect theory, for which Danny Kahneman got the Nobel Prize a few years ago. Prospect theory adds the notion of reference points. We use a reference point to judge the anticipated outcome of a decision. If the anticipated outcome is better than our reference point, then we perceive it as a gain. If the outcome is worse than the reference point, it is a loss. We have different attitudes towards losses than to gains (loss aversion). We also have concave and convex functions that describe our risk attitudes. The new concept here is "loss aversion." Loss aversion can differ as a function of the situation because the situation can reset our reference point. In one situation, we might perceive a given outcome as a gain, but in another situation, we might perceive the same outcome as a loss. If your attitude towards gains is different than your attitude towards losses, you will act in different ways. Prospect theory is one way, then, of adding situation determinants to our study of risk taking.

The behavioral successor in finance of the capital asset pricing model-type of risk-return models are more generalized versions of those models, where we now allow the perceptions of returns and of risks to also become psychological variables. In addition, the trade-off coefficient between the perceived returns and the perceived risks becomes the true measure of attitudinal differences.

Given that we want to add risk perception to the menu of psychological variables, allow me to say a few words about perception in general, and about risk perceptions in particular. Asking people for absolute judgments is very hard. Well, asking is easy, but getting an answer is very hard. If someone had asked you after Robert Sapolsky's presentation, "How bad is it to get 25 electric shocks?", you would have had to think about that for a while. The subjects in that experiment were in two situations-one group had previously received 50 shocks and one group had previously received no shocks. Each group responded quite differently. The relative answer is quite simple-25 shocks are much better than 50 shocks, and a lot worse than no shocks at all.

We see examples of people turning absolute questions into relative questions all the time. I lived in Ohio for four years, and since there is not much to do in Columbus, people tell lots of stories about famous residents. James Thurber is one of them, so I learned a lot of James Thurber stories. In one of them, he is holding a press conference shortly after his wedding. One of the reporters asked him how he liked his new wife. Thurber answered, "Compared to what?" This example demonstrates how automatically we turn absolute questions into relative questions because they are easier to answer.

Let me give you a couple other thought experiments. The first one involves three buckets of water: one is very hot, one is very cold, and one is room temperature. Leave one hand in the hot water and the other hand in the cold water for two or three minutes; then place both hands into the room-temperature water. You will swear that the hot-water hand is now on ice and that your ice-water hand is on fire. Even though you know that this cannot be the case since your two hands are in the same bucket of water, this is, nevertheless, your subjective experience. Our perception of differences is encoded in terms of relative comparisons to prior conditions. This kind of relative encoding is hardwired into our neurons. If we have these kinds of reactions to relatively simple sensory continua, will this also apply to more complex evaluations, like the perceptions of differences or of volatility?

There is a much more famous Weber from the 19th century who formalized "Weber's Law," which encapsulates a series of psycho-physical experiments that were conducted that have to do with our impressions of differences on sensory continua. The experimenter played a series of tones for a subject to listen to. Each tone was incrementally louder than the previous. The subject was asked to indicate the first time he noticed the increase in volume from one tone to the next. If the initial tone is very soft, it does not take long for the subject to notice a difference. If the initial tone is very high, however, the volume would have to increase significantly before the subjects could notice a difference. For psycho-physical continua, our ability to notice differences depends entirely on where we are at the starting point. This is formalized by taking the difference between the two tones and standardizing by where you are (the average or expected value of those two different scores). The ratio of the standard deviation over the expected value is a measure of statistical variation that is widely used in the applied literature. It is used in medicine, engineering, and cultural economics.

It is also a measure that is very useful in animal behavior. While I was at Ohio State, I got in touch with Sharoni Shafir who studies bee behavior. He told me about some of his recent experiments that were quite revolutionary in animal ecology. There had been a model called the energy budget rule developed in the 1980s that derived what animal risk taking ought to look like for basic evolutionary reasons. In this experiment, animals were allowed to choose between a constant foraging environment-where they would get, say, 4 milliliters of nectar if they are a bumble bee-or a variable environment in which sometimes they would get more and sometimes less. Through experimentation, animals can figure out how often the different states arise. The energy budget rule states that in situations of gains when the animal is not starving, greater variability should be less attractive. The animals should go to the predictable foraging environment. In the situation of losses, when the animal is starving, the animals should not pick the predictable foraging environment if that assures them of not getting enough food. In this situation, the animals should prefer the variable environment, and the greater the variability, the more attractive it should be because that increases the probability of finding enough food to stay above survival level.

This was a beautiful theory, and everyone accepted its premises, but it did not predict the data at all. Eighty-four studies had been done, and none of them supported this theory. The variability of the uncertain option had been plotted in terms of variance of standard deviation because that is what the model predicted. On the y-axis, you see the percentage of animals that choose the "sure" option. If you plot this data in terms of variance of standard deviation, it is a mess. There is no regularity. There is no difference between the domains of gains and the domains of losses. There was no clear trend as a function of variability. Sharoni did a very simple thing-he divided the standard deviation by the expected value. In all of these studies, the expected values were very different because of the animals involved. Because you're dividing the standard deviation by the expected value, the units drop out and you can put all of this data into the same graph. Almost magically, trends arose. We see an increase in risk aversion in the domain of gains, just as predicted. We see greater risk seeking as variability increases in the domain of losses. The only change we have introduced is to standardize the conventional measure of variability by the expected value. We have relativized the variability by the situation, and order emerged. This is a step in the right direction.

I would like to share with you several examples in which the perception of risk is subjective. I conducted a study with a Chinese colleague, Christopher Hsee, at the University of Chicago a number of years ago. We compared the investment decisions of business students in Chicago and Shanghai. We had become interested in cultural differences in risk taking, and we asked a variety of people to make predictions about this experiment. There are two types of stereotypes about Chinese business students and investors. The first is that the Chinese are quite risk averse, based on the John Wayne stereotype that the white people are the risk seekers and the Chinese cooks are more timid. There is also a stereotype of the Chinese as gamblers. The folklore does not give us a consistent picture of cultural differences. We decided to test it. We gave business students in Chicago and Shanghai a series of investment choices to make using limited resources to invest. The investment choices varied in terms of variability and returns. When you look at their revealed risk attitudes based on their willingness to pay for risky options, the Chinese were actually greater risk seekers than the Americans. In fact, they were closer to risk neutrality.

When we looked at the data, we developed a "cushion hypothesis." This has to do with different cultural constraints in the two cultures. If you live in a more collectivized society and something bad happens, you can turn to your family or friends to help you out financially. You will also do the same thing for them. This gives you "collectivist insurance" against the worst-case scenario. Quite objectively, your risk is lower in those financial situations. We asked our students not just to pay for various financial options, but also to rate the riskiness of each. The Chinese students actually gave us lower risk judgments-they perceived the risk to be lower than the Americans.

For any set of results, of course, there is a wealth of possible explanations. This is one of the fun aspects of doing research-it's a little bit like detective work without any of the physical danger. We are trying to establish a result; in this case, the cushion hypothesis was our "prime suspect." We needed to eliminate several other suspects. One of the options we needed to eliminate was risk attitudes. It is possible for some kinds of evolutionary reasons that some cultures have different attitudes towards risk. Perhaps the Chinese like the feeling of the stress response they feel in response to risk. How can we distinguish between a risk-attitude argument and a risk-perception argument that is mediated by cultural differences?

We decided to offer the students choices not just on transferable financial decisions, but also on non-transferable decisions like healthcare treatments. You can ask students to gamble for grades-one type of essay will guarantee you a B, while a more controversial topic may get you either a much better or a much worse grade. We expanded our subject populations from the business students to all of the people visiting the science museums in Chicago and Shanghai, and we offered them risky decisions in different domains. For decisions involving healthcare and grades, the Chinese and Americans were indistinguishable in terms of their approaches to risk. That is not to say that there were not individual differences, but the distribution of risk attitudes for those domains that were not transferable were exactly the same in the two cultures. For the distribution of apparent risk attitudes with respect to financial stimuli, the Chinese population shifted towards risk neutrality. This argues for the utility of distinguishing between attitudes towards risk and the perception of risk.

This example shows how the risk perceptions are different in two different groups for very justifiable reasons. It is difficult to argue that the risks are not lower for the Chinese. I also want to show you some examples in which the different risk perceptions are not justifiable, or what economists would call "rational." One such difference has to do with women and financial risk taking. Women's investments tend to be far more conservative than those of men. This is quite a problem for retirement planning because of the risk-return relationship. This means that women who already make less money than men will be even more impoverished in their old age. The question is why these differences exist. One could argue that this risk aversion in women is hardwired into our reaction to the stress response-men might like it more and women might like it less. When we have investigated gender differences in financial decision making or in other domains, women consistently rate the risks to be larger except in social situations. Socially risky situations are the only domain in which women appear to be more risk seeking than men. So this is an example in which differences in risk perception drive differences in risk taking, but it is not very rational.

Why do we have gender differences in perceived risk? It is not biological, because this only happens in majority groups. White women and white men in Western societies show differences in risk perception and risk taking. If you look at minority groups, minority males look like white females. There is a poster hanging at the National Science Foundation that talks about the "White Male Effect"-this phenomenon is sociological in nature and has to do with majority versus minority status. Another colleague of mine interviewed business executives and asked them about taking risks. The executives corrected him, saying that they "manage" risks; they do not "take" risks. If you are in the majority of the population, you make the rules. And if a situation appears too risky to you, you change the rules. If you are in a minority situation, you do not have this latitude. Minorities may generalize this phenomenon in irrational ways. It is very difficult to argue that the risk for males in a pension fund is any different than the risk for females in a pension fund, yet the chronic stress response has trained women to perceive risk differently.

Let me say a little bit more about these psychological risk dimensions. Perceived controllability is very important. Earlier today, Greg Berns called the brain a "prediction engine." We have been successful evolutionarily because we can predict what is going to happen. That is one of the strengths that we bring to the table compared to other species. This makes us one of the most adaptive and successful species on earth. Perceived controllability is so important to us in part because of its role in our survival.

If you ask people lots of questions about different types of risk and analyze their responses, you will see that outside of the outcomes themselves, controllability plays a large role. It is statistically more dangerous to drive than to fly, but we feel as though we can avoid the accident if we are in control of the car. The other important dimension is dread. This is that uneasy feeling you get after you've made a risky decision and you are waiting to see what the outcome will be. This feeling of dread is not closely tied to the outcomes of the decisions, but it strongly influences how we decide. This leads to objectively irrational decisions.

Many of the factors that drive us to make irrational decisions are very emotional in nature. They come from the older part of our brain-the associative, affective brain-and not the analytic brain that we do not share with lower animals. There may be connections between gender differences in risk perception and risk taking and gender differences in overconfidence and trading volume.

Let me summarize what we know about determinants of perceived riskiness. Obviously, perceived riskiness is a function of expected outcome volatility. This, however, is not the only determinant, and this often may not even be the most important determinant. Even objective outcomes are being generalized and relativized by us relative to goals, expectations and other reference points. Given that we have different risk attitudes towards gains versus losses, whether we are comparing a decision to a recent experience or a hoped-for outcome will play a large role in our perception of that volatility or difference.

I also made a point about familiarity and control. If we have been in an environment for a while and nothing bad has happened, it must mean that this environment is less dangerous. Familiarity-exposing people repeatedly to the same stimuli-causes people to rate the risks of their choice options lower. Familiarity also increases expertise, which in turn increases perceived controllability, which is another reason to perceive the risks to be lower. Home bias effects in investment decisions seem to be mediated by these feelings of familiarity and the reduced perceptions of riskiness.

In addition to perceived controllability, we also have other affective responses. I already talked about dread-that uneasy feeling we get between making a decision and having it resolved. The simulations of adverse consequences that people play out in their minds can often explain why people make the choices they do. Myopic loss aversion is probably mediated by personal experiences of loss or by simulations of possible consequences of loss. Given that stocks provide you with many more opportunities for actual losses every quarter than bonds do, that explains why we perceive the risks to be higher for stocks and why we expect higher returns.

Insufficient diversification is also mediated by these emotional reactions. If you have a diversified portfolio, you will find that some of your holdings are underperforming every time you look at them. The immediate reaction that most of us feel at these moments is to quickly sell off the underperformers to invest more in those holdings doing well. This completely defeats the purpose of diversification, but it is also a perfectly natural reaction to seeing something underperform relative to another option.

So let's go back to the discussion about the animals choosing a predictable or volatile foraging environment. We sent a graduate student to the library to find all of the studies that had been done similar to this one using humans. We expected the human results to look very similar to the animal results using the coefficient of variation as the measure of perceived riskiness rather than the standard variation. It is a nice relative measure of risk. We identified hundreds of such studies using humans. When we plotted this data, the results were a mess. What is different about humans compared to the lower animals used in the foraging studies?

The human studies exclusively involved decisions by description. We show people what the outcomes will be numerically or graphically. For example, a subject could choose to win $100 for sure, or he could choose a 50-50 chance of winning $200 or nothing. When we present the decisions in this way, we engage the analytic brain far more than the emotional and affective parts of the brain. To test this hypothesis, we turned our undergraduate subjects into human foragers. They could come to the lab and take samples from two "foraging environments"-two decks of cards. They got to take enough samples to get a sense of the variability in each deck. Then they had the option to choose one deck or the other to draw a card from "for real." Whatever the card showed was how much money they walked out of the lab with. When we did this, our data looked even more beautiful than for the animal studies. There was decreased risk aversion as a function of the coefficient of variation and increased risk seeking for losses. Humans look indistinguishable from lower animals when they learn from experience. For animals, this is the only mode of acquiring information. We are partly so successful as a species because we have acquired a neocortex, symbolic processing, and the ability to learn from other people's mistakes and experience. These functions are incorporated into the format of problems that are presented to us by descriptions. When you look at the prospectus, it tells you how this particular risky option has performed in the past numerically and graphically.

Why does this make a difference? The coefficient of variation is a beautiful description of human and animal decision making when it comes to experience, but not for decisions from description. Different models explain our behavior in these different situations. Expected utility theory or prospect theory explains decisions from description beautifully. It says that when a rare event is involved in a decision, people will overreact to that outcome when it is described to them-they give it more weight than it deserves by probability alone. This is often described as an "attentional regression towards the mean." You will give equal attentional weight to all of the possible outcomes no matter what the different probabilities are.

On the other hand, when we make decisions from experience, rare events are experienced very differently. If you don't have very many samples in your experience, then it is quite likely that you will never even experience that rare event. In this case, you will underweight the probability of the rare event in your decision-making process. When you force people to sample sufficiently so that they should get a sense of the rare event and how rare it really is, they still do not act accordingly. Most of the time, we still underweight the small probability of the rare event, because our decision-making process is basically an average of our initial impression of an option with our most recent feedback about the option.

There is a strong recency bias-whatever we have experienced most recently will determine how we update our impression of this risky option. With strong recency bias, we tend to underweight rare events naturally because rare events have a very small probability of having occurred recently. Overall, we underweight rare events when we make decisions based on experience, except in those rare occasions when the rare event has happened recently. In this case, we overweight the rare event tremendously, even more so than in decisions from description. We have much higher volatility in our actions when our decisions are based on experience, not description. We also tend to learn from yesterday's war. We base our decisions on what has happened most recently, even if the most recent event has a very small probability of re-occurring. The main take-away is that decision making can be very different based on how we find out about the probabilities and outcomes of a decision-by description or by experience.

Most investment decisions are based on a combination of those two types of learning. We do look at past returns, but we also often have personal experience with a given investment instrument. It turns out that for most people, their experience is a greater determinant of decision making than the statistical data available by description.

Let me start to wrap-up. Why is it useful to distinguish between risk attitude and risk perception when it comes to decision making? This distinction helps us get a better understanding of differences in individual and group risk taking. Not all differences are due to attitudes towards risk, but some are. These attitudinal differences are probably based in biology and optimal arousal set points. Those differences that are due to perceived risk, however, are due either to stable differences in our social and cultural environment or to transient situational characteristics. Differences in perceived risk can sometimes be justifiable due to differences in opportunities and constraints, but they can also reflect less-justifiable, myopic, emotional reactions or overgeneralizations.

I would like to argue with my colleague Gary Becker's statement that we should not argue about taste. With true attitudes towards risk, it is hard to argue that someone should be more risk averse or risk seeking, especially if it just involves a personal decision. But I do want to argue that if risk taking is determined not just by risk attitude, but also by perceptions of risk, and that those perceptions can sometimes be irrational, perhaps we should argue not about taste, but about risk perceptions.

As another brief digression, the same also holds for types of risk for which we are not evolutionarily prepared. We developed our stress response to help us when we encounter a tiger at a watering hole-this reaction is hardwired. We have automatic reactions to certain types of risks, often immediate risks. We do not have those reactions to other types of risks that have important consequences for us, but which were not around in evolutionary times when our brain was developing to warn us about risks. In particular, we are not adept at perceiving risks at long time scales. We all know that we should be saving more for our retirement, yet as a country and as individuals, we are not losing any sleep over it. It is so boring that we do not get an emotional stress response when we think about it. Because the perceived risk is not there, we don't do anything about it. Similarly, we could argue that climate change is something we should be concerned about, but because it is so removed from us in time, perhaps by generations, it does not worry us. If things do not worry us, we don't do anything about them. It is perceived risk-or in this case, the absence of a perceived risk-that is responsible for our inaction, not our risk attitudes.

What does this buy us? There might be a way to improve communications to different market segments if you are more aware of not just the analytic determinants of risk perception but also the emotional determinants. While it might be true that institutional investors might be immune to the types of biases exhibited by private investors (like the home bias, insufficient diversification, or excessively high trading volume), I would not bet too much that institutional investors are immune from the influence of recent events. This argues that you might want to communicate risk differently to first-time clients who do not have personal experience of being in a fund than to clients who have recently experienced a particular loss or gain.

Knowledge of these regularities can improve investor education. You might be able to design risk communication or educational strategies that take advantage of the emotional bases of most of the biases exhibited by private investors. This could represent a behavioral competitive advantage. For example, if the home bias is due to familiar perceptions of risk, it might be wise to familiarize your investors with foreign investment options rather than providing them matrices of covariants and rates of return. The descriptive elements are probably completely irrelevant to their decision making.

If insufficient diversification is due to myopic, emotional overreactions to seeing losses, then it might be wise to develop reporting formats that minimize losses by highlighting the long-term performance of funds and portfolios.

In summary, my talk tried to introduce you to a different perspective on risk taking, one influenced by psychology and behavioral research. I want to argue that the psychological component of risk taking is not just limited to personality traits. A richer framework that introduces perceptions of riskiness as another psychological variable might help us to resolve a number of apparent paradoxes. The first apparent paradox is the inconsistency of risk attitudes across domains. When a person appears risk seeking in one domain and risk averse in another domain, the key difference between domains may be that person's perception of the different risks. Risk perception is often a function of familiarity and controllability-the rock climber does not think he's doing anything crazy because he has been doing it for years and he is good at it. The situation may seem risky from the outside, but it may not appear risky from the inside out. Matt Raben showed that people's risk attitudes inferred from people's small-stake decisions did not predict how they would approach large-stake decisions. By relativizing the perceived riskiness by something like the coefficient of variation, we can explain some of the apparent irrationalities.

The take-away is a better appreciation of the non-analytic processes on investment behavior, and perhaps a better appreciation of the malleability of the perceptions of riskiness. Risk is not an immutable attribute of an investment instrument. In fact, the mode by which you inform people about past volatility makes a big difference. People will react very differently to experiences vs. descriptions. Familiarity, comfort and other relative comparisons can play a large role in people's decisions. If these things influence how we perceive risk, then we can also manipulate these factors to influence how other people perceive riskiness, hopefully in a way that makes their perceptions more in line with the actual, objective contingencies. Knowledge of these mechanisms can provide input into the design of more effective information formats and investor education. Thank you.

Q&A Session

Question: Among the factors that influence risk perception and decision making, you have mentioned expected outcome volatility, expectations, other reference points-familiarity, perceived control, dread and fear. Has there been any research done to try to prioritize these factors?

Answer: The short answer is no. A lot of these ideas are relatively new, and a lot of them have been undervalued in part because of the economic view of risk taking being determined only by risk attitude. A broader appreciation of the importance of risk perception will result in studies of the type that you are talking about. Now, we do know one thing about these factors. In some situations, one factor will be more salient than in other situations. Often doing an analysis of the type of decision being made will tell you which factors will play a larger role. If all of the options are equal in terms of, say, controllability, then that factor will not play a roll in the decision. This will be a function of situational and individual differences. Obviously expertise plays a large role-we put different weights on the outputs of our analytic brains and our emotional brains. If you have been trained in an analytic domain, you will probably tend to think about it in those terms. In other domains, you will act in different ways. It will depend, too, on who you are in terms of your preparedness for the situation and also what the situation brings to you.

Michael Mauboussin asked us three questions at the beginning of the day. I would like to comment on those questions. They come from something called the Frederick Cognitive Impulsivity Scale (FCIT). It is true that people have differences on those scales, and even the people at Cal Tech often perform very badly on it. Mike mentioned that the scale is significantly correlated with the degree of risk taking. One way of explaining this test is that the types of answers that occur to us quickly are normally wrong-these are the outputs of the associative brain. The correct answer must be derived analytically, and this takes longer. We need to wait long enough and not give in to the emotional or affective answer. To answer these questions correctly, we have to control our impulsivity. The more we control our impulsivity, the more analytical our approach to risk taking will be.

Question: I have a question about the pari-mutuel example you described earlier in your talk. If I understood you correctly, bettors tend to be less risky early in the day and more risky later in the day. They may do this in order to "catch up" before they go home and face their spouse. Has it also been shown that a successful betting strategy would be to bet against the crowd, and bet on the long shots in the morning and the favorites in the late afternoon?

Answer: Yes, this strategy does work. When I was an undergrad, I took a decision-making course from a professor who went on sabbatical to study behavior at the race track. He came back to write a book called How Gambling Saved Me From a Misspent Sabbatical. The short answer to your question is that yes, there are successful strategies, and there are people who make money this way. It is an arduous way to make money.

Question: I have a corollary question. If expected outcome volatility increased the perceived riskiness, then you would expect the risk premium on risky assets to be a lot higher so your aggregate expected return would be higher. Is that actually proven out, or do people treat risky stocks like lottery tickets and have a negative expected return?

Answer: You are a better person to answer that question than I am. It is absolutely true that there is a risk-return relationship. The point I was trying to make was that it is not just volatility that makes a difference-it has to do with perceived volatility standardized by previous experiences. Some volatility is not so bad if you have experienced a lot of volatility just prior. If you perceive volatility over and over again, you become very familiar with it and it just becomes baseline. It is not perceived as risky anymore because it is just perceived as normal. While the economic and finance-based factors obviously play a role, they are not the only ones.

Question: Do you have any thoughts about how people set themselves up for catastrophic failures like Amaranth or Long-Term Capital Management?

Answer: It is very similar to the racetrack phenomenon. These companies pursued a particular strategy that got them into trouble. To recoup their initial losses, they had to do something that is even more risky, and it becomes a spiral. Sometimes risky strategies work. For every trader who loses, there are others who are pursuing equally risky strategies, but they get lucky and recoup their losses on the second or third try. But every once in a while, the small probabilities add up and you get a catastrophic failure. I would argue that the trader in question did not perceive what he was doing as very risky-this was the only way that he could prevent the loss of his career.

Question: In terms of risk perception, the mutual fund industry has a $1T draw down in the bear market ending in 2003, which is a number that gets no publicity, but Amaranth has a $5B draw down (a much smaller magnitude), and the public perceives a much greater risk. Why does the public perceive an isolated risk in Greenwich, Connecticut, as a much bigger risk than a systemic disaster such as a bear market?

Answer: One of the answers to that has to do with the psychological risk factors that I discussed earlier. I talked about dread and controllability. There are some other smaller ones, like disaster potential. This also explains why we worry less about car accidents than plane crashes. Anonymous people die on the roads. Anonymous people lose money in certain kinds of funds. But it is the small probability, but high magnitude events that get media attention and our attention. A lot of these phenomena are mediated by selective exposure and selective reporting. Because you have different psychological reactions to different types of events, these large catastrophic events get a lot more weight than they deserve on an actuarial basis alone.

The comments, opinions and any forward predictions presented about any particular security, the economy and "the market" are based on the analysis of the speaker. These are not necessarily the opinion of, and should not be construed as a recommendation on the part of Legg Mason Capital Management or any of its affiliates.