Beyond Greed and Fear
Page 6
I have organized the remainder of the book around specific applications. I devote part II to applications that feature predictions about the overall market, stock returns, and earnings. Most of these applications concern different forms of heuristic-driven bias and the effect of these errors on market efficiency.
Part III presents applications that involve individual investors, such as selling at a loss, portfolio selection, and retirement saving. These applications deal mostly with frame dependence and heuristic-driven bias.
The applications discussed in the remainder of the book involve all three themes. In part IV I focus on the relationship among the money management industry and the investors they serve. This part deals with institutional investors: open-ended mutual funds, closed-end mutual funds, the management of fixed income securities, and the tax-exempt money management industry.
In part V I look at corporate executives and their relationships with analysts and investors. I discuss these relationships in several contexts: corporate takeovers, initial public offerings, seasoned equity offerings, and analysts’ earnings forecasts and stock recommendations. Part VI is devoted to special topics in investment: options, futures, and foreign exchange.
Parts II through VI follow a particular format. Each chapter begins with a short case that illustrates the main message of that chapter. Then, in the remainder of the chapter, I present the general findings from the behavioral finance literature, as typified by the case. The upside of this technique is that it makes the application of behavioral concepts to finance very easy to see. The downside of the technique is hind-sight bias, a behavioral error. Someone susceptible to hindsight bias views events, after the fact, as being almost inevitable. In presenting these cases, by no means do I wish to suggest that they needed to turn out the way they did. Rather, they happened to do so.
What’s Next?
I have arranged the topics in a particular order. Readers will thus find some advantage in reading the chapters consecutively; however, this should not deter those who want to follow a different order. I say this with one caveat. Since the next three chapters focus on main themes of behavioral finance—heuristic-driven bias, frame dependence, and inefficient markets—which constitute the core concepts in the book, I strongly suggest that readers complete these chapters before moving on to the applications. With that, let’s take up the first theme.
Chapter 2 Heuristic-Driven Bias: The First Theme
This chapter discusses the following:
• availability bias
• representativeness, grade point average, winners, and losers
• regression to the mean and stock market prediction
• gambler’s fallacy and stock market prediction
• overconfidence and expert judgment
• anchoring-and-adjustment and earnings forecasts
• aversion to ambiguity
The dictionary definition for the word heuristic refers to the process by which people find things out for themselves, usually by trial and error. Trial and error often leads people to develop rules of thumb, but this process often leads to other errors. One of the great advances of behavioral psychology is the identification of the principles underlying these rules of thumb and the systematic errors associated with them. In turn, these rules of thumb have themselves come to be called heuristics.
An Illustrative Example
Consider this question: Which is the more frequent cause of death in the United States, homicide or stroke? How do most people go about answering this question? The majority rely on recall, that is, by seeing how many events of each type come readily to mind. If people more readily recall instances of homicide than of stroke, then they will answer “homicide.” This simple rule conforms to the principle known as availability—the degree to which information is readily available. A rule based on this principle is called an availability heuristic.
Heuristics are like back-of-the-envelope calculations that sometimes come close to providing the right answer. But heuristics may involve bias, meaning they may tend to be off target in a particular direction, and this can apply to an availability heuristic also. Most people rely on the media for their information about homicides and strokes. Suppose that the media tends to report one cause of death more than the other, because one is newsworthy and the other is not. Then people who rely on an availability heuristic may recall instances related to one type of death more readily than the other. Therefore, media coverage biases a rule based on recall.
What about error? Which is the more frequent cause of death, homicide or stroke? The answer is stroke. In fact, strokes occur eleven times as often as homicides (Slovic, Fischoff, and Lichtenstein 1979). People who rely on an availability heuristic tend to be amazed by this fact.
Let’s look at these steps from a broader perspective:
• People develop general principles as they find things out for themselves;
• They rely on heuristics, rules of thumb, to draw inferences from the information at their disposal;
• People are susceptible to particular errors because the heuristics they use are imperfect; and
• People actually commit errors in particular situations.
Taken together, these four statements define heuristic-driven bias.1
Representativeness
One of the most important principles affecting financial decisions is known as representativeness. Representativeness refers to judgments based on stereotypes. The principle of representativeness was proposed by psychologists Daniel Kahneman and Amos Tversky (1972), and analyzed in a series of papers reproduced in the collection edited by Kahneman, Slovic, and Tversky (1982).
Consider an example involving admissions officers in universities. One measure of successful admission decisions is that students who are admitted perform well scholastically. Therefore, imagine a situation where an admissions officer is attempting to predict the grade point average (GPA) of some prospective students based upon their high school GPA levels.
Here are some actual data for undergraduates at Santa Clara University, based on students who entered the university in the years 1990, 1991, and 1992.2 During this period, the mean high school GPA of students who entered as freshmen and graduated was 3.44 (standard deviation was 0.36). The mean college GPA of those same students was 3.08 (standard deviation 0.40). Suppose you are given the task of predicting the graduating GPA for three undergraduate students, based solely on their high school GPA scores. The three high school GPA scores are 2.20, 3.00, and 3.80. What are your predictions for the college GPAs of these students upon graduation?
In administering this question to large groups, I have obtained very consistent mean responses. Table 2-1 contains the mean predictions along with the actual results. The average predictions for the question are 2.03, 2.77, and 3.46, whereas the actual results are 2.70, 2.93, and 3.30, respectively. Notice that at both the low end and the high end, the predictions are too far from the mean of 3.08. That is, both the low (2.20) and high (3.80) high school GPAs result in college GPAs that are much closer to the mean than the predictions. These responses illustrate that people do not appreciate the extent to which there is regression to the mean.
Table 2-1 Actual GPAs Are Closer to the Mean Than Predicted GPAs.
Representativeness is about reliance on stereotypes. The simplest example based on this principle is to predict that college GPA will be the same as high school GPA. Now most people do not use as simple a rule as this one. But they do base their predictions on how representative a student appears to be. Thus a student with a high GPA in high school is seen as representative of a good student. Notice that they are especially hard on students with low high school GPAs. What most people fail to appreciate is that students with the lowest high school GPAs may have experienced bad luck, and consequently will, on average, do better in college.3 So, the heuristic involves bias; representativeness can be misleading. Again, people fail to recognize regression to the mean. Therefore, they are predi
sposed to making errors when they predict the future GPA of particular individuals.
A financial example illustrating representativeness is the winner-loser effect documented by Werner De Bondt and Richard Thaler (1985, 1987). De Bondt and Thaler find that stocks that have been extreme past losers in the preceding three years do much better than extreme past winners over the subsequent three years. De Bondt (1992) shows that the long-term earnings forecasts made by security analysts tend to be biased in the direction of recent success. Specifically, analysts over-react in that they are much more optimistic about recent winners than they are about recent losers.
Do you recognize any similarities with the GPA question above? De Bondt and Thaler base their argument on the misapplication of representativeness. In effect, I suggest that investors treat past losers like high school students with low GPAs, and past winners as high school students with high GPAs. Notice that the predictions are particularly pessimistic when it comes to the low GPA students. People tend to predict that a student with a low high school GPA will end up with an even lower college GPA, indicative of a “kick ’em when they’re down” perspective.4 As we shall see in chapter 4, the same phenomenon also appears to be at work when it comes to stocks. The returns to past losers are exceptionally high, suggesting that investors become unduly pessimistic about the prospects of these stocks.
Before leaving representativeness, let us consider one more example showing that although financial professionals may recognize regression to the mean, they may not apply it properly. Below is an excerpt from an interview that appeared in the August 18, 1997 issue of Fortune magazine, with global strategist Barton Biggs of Morgan Stanley and senior investment adviser Robert Farrell of Merrill Lynch (Armour, 1997). This interview occurred after two-and-one-half years of spectacular stock market returns. I have divided the excerpt into two parts. The first part sets the stage for a discussion about regression to the mean, and also for an issue that comes up in chapter 5 (on skewed confidence intervals). Here is the first part of the excerpt:
Biggs: My view is that we’re at the very tag end of a super bull market. That means the prudent person who’s thinking ahead toward retirement should assume that over the next five to ten years the total return from his equity portfolio is going to be in the 5%-to 6%-a-year range.
Fortune: NOT THE 15% TO 20% WE’VE COME TO LOVE AND EXPECT?
Biggs: Right. It’s very late in the game.
Farrell: Trouble is, it’s looked that way for a long time.
Biggs: Yes, but it’s never looked as much that way as it does right now.
We will come back to the “late-in-the-game issue” a little later. For now, consider regression to the mean.
Farrell: It’s been better to have been a novice than a professional the past few years, because people with the most experience have been the most cautious. But markets do regress back to the mean return to their long-term average performance, and I agree we are late in the ball game. This is the longest period we’ve ever had with such high returns from equities, and I can’t believe it’s a new era that will just keep going forever. I don’t know if returns going forward will be 7% or 8%, but I’m pretty sure they will be below average.
This interview raises a number of very important issues. Look first at the last three sentences in Robert Farrell’s remarks, where he predicts below-average returns. What’s his rationale? Well, he says markets “regress back to the mean” and points out that this “is the longest period we’ve ever had with such high returns.”
Is a prediction of below-average returns appropriate? Take another look at table 2-1, the GPA example. Would we predict that the student with the 3.80 high school GPA would end up with a college GPA below the mean of 3.08? I don’t think so. Regression to the mean suggests that future returns will be closer to their historical average. But it doesn’t say they will be below their historical average.5
Farrell’s error, too low a prediction, stems from gambler’s fallacy. If five tosses of a fair coin all turn out to be heads, what is the probability that the sixth toss will be tails? If the coin is fair, the correct answer is one-half. Yet many people have a mental picture that when a fair coin is tossed a few times in a row, the resulting pattern will feature about the same number of heads as tails. In other words, the representative pattern features about the same number of heads and tails. So, after a run of five heads, people tend to predict tails on the sixth toss, because of the representativeness heuristic. From their perspective, “a tail is due.” But this reasoning is wrong, just as below-average returns are no more likely after “the longest period we’ve ever had with such high returns.”
Gambler’s fallacy arises because people misinterpret the law of averages, technically known as the “law of large numbers.” They think the law of large numbers applies to small samples as well as to large samples. This led Tversky and Kahneman (1971) to facetiously describe gambler’s fallacy as the “law of small numbers.”
Let’s go back to Farrell’s remarks about future returns. Notice that he tells us he is “pretty sure they will be below average.” Time will tell if he ultimately is right. I say ultimately because in the twenty-one months that followed the Fortune magazine interview, the S&P 500 returned more than 41 percent. But his statement that he is “pretty sure” leads us to the next issue—overconfidence.
Overconfidence
Here is a question for you.
The Dow Jones Industrial Average closed 1998 at 9181. As a price index, the Dow does not include reinvested dividends. If the Dow were redefined to reflect the reinvestment of all dividends since May 1896, when it commenced at a value of 40, what would its value have been at the end of 1998? In addition to writing down your best guess, also write down a low guess and a high guess, so that you feel 90 percent confident that the true answer will lie between your low guess and your high guess.
Ready? The answer to the preceding question is found in the title of a paper by Roger Clarke and Meir Statman (1999): “The DJIA Crossed 652,230 (in 1998).” If people were well calibrated, then 90 out of every 100 would find that the correct answer lay between their low and high guesses. But when I ask this question as part of a survey, virtually nobody finds that the true answer lies between his or her low and high guesses. For most, their high guesses are much too low. So most people are not well calibrated. Instead, they are overconfident.
When people are overconfident, they set overly narrow confidence bands. They set their high guess too low (and their low guess too high).
Hence they get surprised more frequently than they anticipated. Later in this volume we will come across Wall Street strategists who, in the course of reviewing their predictions in the light of actual events, speak about being “humbled.” In other words, they were overconfident in their predictions.
Anchoring-and-Adjustment, Conservatism
Next is a textbook problem in probability, designed by psychologist Ward Edwards (1964) that provides some insight into analysts’ earnings revisions.
Imagine 100 book bags, each of which contains 1,000 poker chips. Forty-five bags contain 700 black chips and 300 red chips. The other 55 bags contain 300 black chips and 700 red chips. You cannot see inside any of the bags. One of the bags is selected at random by means of a coin toss. Consider the following two questions about the selected book bag.
1. What probability would you assign to the event that the selected bag contains predominantly black chips?
2. Now imagine that 12 chips are drawn, with replacement, from the selected bag. These twelve draws produce 8 blacks and 4 reds. Would you use the new information about the drawing of chips to revise your probability that the selected bag contains predominantly black chips? If so, what new probability would you assign?
This problem is analogous to the tasks faced by financial analysts. The bag is like a company that in the future may operate in the black or in the red. So in accordance with generally accepted accounting colors, black chips stand for good futur
e earnings, red for poor future earnings. Analysts start out with information that leads them to form their initial beliefs. In this case, beliefs concern the probability that the bag contains predominantly black chips. The most frequent answer given to the first of the two preceding questions is 45 percent. So, the bag of chips is like a company that appears more likely to generate poor future earnings than good future earnings.
The second question is a lot more difficult than the first. The drawing of 8 black chips and 4 red chips is akin to a positive earnings announcement. So now the question is how to react to a positive earnings announcement made by a company that has not been performing all that well.
When I administer these questions, I find that the two most frequent responses to the second question are 45 percent and 67 percent—the two most salient numbers in the problem—with 45 percent being the number of bags containing predominantly black chips, and 67 percent the fraction of black chips drawn with replacement.
Those who respond with 45 percent essentially do not know how to incorporate the new information. So, they stick with their initial beliefs. Since the “earnings announcement” is favorable, they underreact.
People who answer 67 percent (or thereabouts) focus on the fact that two thirds of the chips drawn with replacement are black. They ignore their prior information, in accordance with the representativeness heuristic. Do they overreact, underreact, or get it just right?
The correct answer to the second question is 96.04 percent. About 55 percent of those responding choose either 45 percent or 67 percent; The remaining responses are scattered. But most are well below 96 percent. In fact, most are below 75 percent. In other words, most people respond too conservatively to the new information in this problem. Perhaps they get anchored on to 45 percent and do not adjust sufficiently to the new information.