Neoclassical economics is built on the assumption that the agents in the economy are self-interested and rational. These assumptions are what distinguishes economics from other social science disciplines, such as psychology and sociology; as such, they are simultaneously the source of the power in economic theorizing and the fundamental weakness in the method. The assumption of rationality is what permits economists to use their most powerful theoretical tool: optimization. But, if agents are not rational, what then? My research over the last 15 years has attempted to answer this question.
Of course, critiques of the assumptions of economics are as old as the use of those assumptions. For many economists, Milton Friedman provided the definitive response to such criticism in his famous essay on positive economics.(1) There he argued that theories should not be evaluated on the basis of the validity of their assumptions, but rather on the accuracy of their predictions. An expert billiards player, he noted, may not know the laws of physics, but acts as if he knows such laws. I completely share Friedman's view that theories should be judged on the basis of predictive power; in my research, I have used this criterion to evaluate alternative models. However, as I have tried to show in my series of "Anomalies" articles in the Journal of Economic Perspectives,(2) the theory fails on precisely these grounds. In such circumstances it makes sense to take a careful look at the assumptions. Perhaps most economic agents make decisions the way most of us play pool: badly.
At some level, of course, the rationality assumption has to be wrong. As noted by Herbert Simon, people are only boundedly rational. How does a boundedly rational agent differ from a rational agent? If the differences are random then the rational model still produces unbiased predictions of behavior. However, as the work of Daniel Kahneman and Amos Tversky(3) has shown, actual behavior differs from rational choice in systematic ways. Their research program of discovering the heuristics people use to make judgments, and the biases inherent in those heuristics, provided my motivation in exploring behavioral economics.
If economic agents make judgments and choices that differ systematically from those prescribed by the rational model, then we can improve the models by incorporating these factors into our theories. My first paper in this domain(4) described several ways in which most people fail to act like "Homo Economicus": for example, they fail to ignore sunk costs, they undervalue opportunity costs relative to out-of-pocket costs, and they have trouble exerting self-control. Sometimes, agents know about their own biases. Thus, people who know they have self-control problems will, like Odysseus, tie themselves to the mast to prevent future transgressions. They join Christmas dubs (or used to before credit cards eliminated the need to be liquid at Christmas time), go to fat farms (resorts that, for a high fee, agree to starve their guests), and pay in advance to join a health club, knowing that the sunk cost will help motivate them to go more often.
One objection to models with less than fully rational agents is the claim that in markets, such agents either will be eliminated or rendered irrelevant. Russell and I investigated this claim in a paper(5) that considers a world with two kinds of agents: the fully rational agents that populate standard economic models, and what we call "quasi-rational agents" who make predictable, systematic errors. We then looked for the conditions necessary for such a world to produce the same equilibriums that would obtain if all the agents were fully rational. We find that these conditions rarely are met, even in markets that function very well, such as financial markets. Often, if I insist on choosing a less than optimal choice for me (say a dominated alternative), there will not be any way for you to make money from my mistake, either by exploiting it or by educating me. Quasi-rationality is neither fatal nor immediately self-defeating.
How can we begin to model "Homo Behavioral Economicus"? A small list of factors can go a long way toward better descriptive models. The following concepts have proven most useful so far:
1. Overreaction. If people make judgments using what Kahneman and Tversky call the representativeness heuristic, they judge the likelihood of an event by how similar the event is to the typical or stereotype of that event. This heuristic, like all those that are used widely, is often accurate, but it leads to systematic biases. For example, forecasts stemming from the representativeness heuristic violate Bayes's rule because recent evidence is given too much weight relative to prior odds. Forecasts tend to be too extreme, relative to the norm.
2. Loss aversion. It is well accepted that people tend to adapt to current circumstances and to react to changes relative to the recent norm rather than to levels. Furthermore, the sensitivity to losses is greater than the sensitivity to gains. Roughly speaking, the loss of $1 is about twice as painful as the gain of $1 is pleasurable.
3. Mental accounting. Standard accounting principles, and all economic theory, assumes that money is fungible. Behavior should not depend on the label associated with any pot of money. However, both individuals and organizations violate this principle. This implies that the rules people use to aggregate transactions and wealth holdings are important in understanding their behavior. I call these rules "mental accounting."(6)
4. Fairness. All other things equal, people prefer to be treated fairly and like to treat other people fairly. This is not to say that people do not care first and foremost about their own household's welfare, but rather that concerns about fairness also matter and compete with purely selfish motives for scarce resources. In another paper,(7) Kahneman, Knetsch, and I explored lay perceptions of what is fair. We found that the determinants of such perceptions were explained in turn by other behavioral factors such as loss aversion. For example, people thought it was much less fair for a dealer to impose a $200 surcharge above list price for a scarce car model than to eliminate a $200 discount. Perceptions of fairness also displayed money illusion. For example, cutting wages 5 percent in an economy with no inflation was judged to be much less fair than offering a 7 percent raise in an economy with 12 percent inflation.
I have applied these and other behavioral factors in a variety of settings, always with the goal of trying to improve the explanatory power of economic analysis. One domain that has taken a lot of my attention over the last ten years is financial markets. This field is an attractive place to do behavioral economics for two reasons. First, the data are extraordinarily good. Second, many economists have the "prior" (or did a decade ago) that behavioral factors are least likely to surface in these, the most efficient of all markets.
My first papers in this area were written with Werner De Bondt. In the first one,(8) we reported the successful prediction of a new anomaly in asset markets, namely long-term mean reversion for individual stocks. David Dreman(9) offered a behavioral explanation for the "price/earnings" anomaly: namely, that stocks with low p/e ratios systematically outperform those with high p/e's. His explanation was: because investors use the representativeness heuristic, they would make excessively optimistic earnings forecasts for firms growing rapidly and excessively pessimistic forecasts for those in trouble, that is, high and low p/e firms, respectively. When these forecasts proved wrong in the predictable ways, prices would adjust, causing high p/e firms to have low returns and low p/e firms to have high returns.
We thought that if this argument was correct we should be able to observe the same phenomenon for firms chosen by past performance rather than p/e. We therefore ranked firms by 3-5-year past returns and selected the most extreme performers. We then tracked these extreme "winners" and "losers" for the next 3-5 years. We found substantial mean reversion, as predicted by the behavioral analysis.(10)
De Bondt and I also looked for evidence of the representativeness heuristic in professional analysts' forecasts of earnings.(11) Recall that when this heuristic is used, forecasts are too extreme. We regressed the actual change in earnings for a year on the consensus forecasted change in earnings (made in April for December 31 fiscal year firms). If the forecasts were rational, the intercept in this regression would be zero and the slope would be one. Instead we found that the forecasts were biased in two ways. First, the intercept was significantly negative, implying that the forecasts were optimistic (consistent with past research). Second, the slope coefficient was only 0.65. This means that the forecasted changes in earnings were too extreme. One could improve the analysts' forecasts simply by subtracting a constant (to correct for the optimism bias) and reducing the forecasted change by a third (to make the forecasted changes less extreme).
Saving behavior is a domain in which behavioral factors have proven to be extremely important. In the standard life-cycle model, households compute their lifetime wealth constraint, optimize their spending stream, and save accordingly. Components of wealth are not distinguished since all money is assumed to be fungible. In the behavioral life-cycle model,(12) the standard framework is modified to incorporate two important behavioral factors: self-control and mental accounting.
First, we recognize that most households other than the very rich are constrained by self-control considerations. Even if they could and did solve for their optimal spending path, they would find it difficult to stay on this path. Mental accounting comes into play because the location of wealth influences spending choices, since some mental accounts are more "tempting" than others. For example, cash on hand is more tempting than money in a savings account, which in turn is more tempting than money in an IRA, pension plan, or home equity.
The difference between the standard model and the behavioral model is particularly striking in analyzing programs designed to increase saving, such as IRAs and 401(k)s. According to the standard theoretical analysis, these plans have no effect on saving because most contributors already had some assets saved up that simply could be shifted to the retirement accounts. In this scenario, there is no incentive to save at the margin, so the programs should have no effect.
In a mental accounting framework, however, saving is expected to increase because the programs help put money into less tempting accounts. Consider a child who takes money from a leaky piggy-bank and puts it into a bank. According to the standard theory, assets simply have been shifted, but a behavioral analysis predicts that spending will fall. The same is true for IRAs and other pension saving; my reading of the evidence is that this view is supported.(13)
Loss aversion is perhaps the most pervasive of all the behavioral factors uncovered to date. In a sample experiment Kahneman, Knetsch, and I ran,(14) subjects traded tokens, whose value to each subject was private information. Half of the subjects were then given a token, and markets were "conducted" in which reservation prices were elicited from both token owners and buyers. Both price and volume in these markets were exactly as simple supply and demand analyses would predict, showing that transactions costs were negligible in this market.
Then, another set of markets was conducted, this time for coffee mugs imprinted with the local university insignia. Again, half the subjects had a mug and half did not. The Coase theorem predicts that in this situation half the mugs on average should trade: mugs should end up in the hands of the subjects who value them most, only half of whom would have received one in the random assignment of mugs at the beginning of the experiment. Counter to this prediction, only about 15 percent of the mugs traded; the median reservation prices of mug owners were roughly twice the reservation prices of the mug buyers. In other words, we found "loss aversion" for mugs. Obviously, this result calls into question the Coasean claim that in the absence of transactions costs (known to be negligible in this case) the initial assignment of property rights will not affect the ultimate allocation of resources.
In a paper with Benartzi,(15) the concepts of loss aversion and mental accounting are combined to offer an explanation for the well-known equity premium puzzle.(16) The puzzle refers to the fact that over very long periods of time, equities have earned much higher rates of return than bonds or other fixed-income assets (roughly 6-7 percent real annual return versus 1 percent). This difference is too large to be explained easily within the standard framework. Our explanation is based on the following intuition: suppose investors are loss averse, implying that losses are weighted more heavily than gains.(17) Then, their willingness to hold risky assets will depend on the frequency with which they evaluate their holdings. If loss-averse investors compute the value of their portfolios daily, they will hate equities, since on a daily basis stocks decline nearly as often as they rise. In contrast, investors with a horizon of say 20 years would find stocks very attractive since the risk of loss is tiny. Thus motivated, we estimate the frequency with which loss-averse investors would have to be evaluating their portfolios in order to make the investors indifferent between stocks and bonds. We find this to be approximately once a year, a highly plausible result. Therefore we dub our explanation for the equity premium puzzle "myopic loss aversion," since most investors (for example, pension plans, endowments, and those saving for retirement) should have horizons of much more than one year.
As I hope these examples have made clear, I view behavioral economics as an entirely constructive enterprise. The goal is simply to make economic models better at explaining economic activity. In this sense, behavioral economics is simply economics with a higher R(2).
1 M. Friedman, "The Methodology of Positive Economics," in Essays on Positive Economics, Chicago: University of Chicago Press, 1953.
2 These are reprinted in R. H. Thaler, The Winner's Curse, Princeton: Princeton University Press, 1992.
3 D. Kahneman, P. Slovic, and A. Tversky, Judgment Under Uncertainty: Heuristics and Biases, New York: Cambridge University Press, 1982; and D. Kahneman and A. Tversky, "Prospect Theory: An Analysis of Decision Under Risk," Econometrica 47, 2, (1979), pp. 363-391.
4 R. H. Thaler, "Toward a Positive Theory of Consumer Choice, "Journal of Economic Behavior and Organization 1 (1980) pp. 39-60. This paper, and most of the others cited here, are reprinted in R. H. Thaler, Quasi-Rational Economics, New York: Russell Sage Foundation, 1991.
5 T. Russell and R. H. Thaler, "The Relevance of Quasi-Rationality in Competitive Markets," American Economic Review 75 (December 1985), pp. 1071-1082.
6 R. H. Thaler, "Mental Accounting and Consumer Choice," Marketing Science 4 (Summer 1985), pp. 199-214.
7 D. Kahneman, L. Knetsch, and R. H. Thaler, "Fairness as a Constraint on Profit Seeking: Entitlements in the Market," American Economic Review 76 (September 1986), pp. 728-741.
8 W. F. M. De Bondt and R. H. Thaler, "Does the Stock Market Overreact?" Journal of Finance 40 (July 1985), pp. 793-805.
9 D. Dreman, The New Contrarian Investment Strategy, New York: Random House, 1982.
10 This pattern has now been well documented in numerous additional studies, such as W. F. M. De Bondt and R. H. Thaler, "Further Evidence on Investor Overreaction and Stock Market Seasonality, "Journal of Finance 42 (July 1987), pp. 557-581; E. Fama and K. French, "The Cross-Section of Stock Returns, "Journal of Finance 46 (1992), pp. 427-466. and J. Lakonishok, A. Shleifer, and R. W. Vishny, "Contrarian Investment, Extrapolation, and Risk," Journal of Finance 49 (December 1994), pp. 1541-1578.
11 W. F. M. De Bondt and R. H. Thaler, "Do Security Analysts Overreact?" American Economic Review (May 1990), pp. 52-57.
12 R. H. Thaler and H. M. Shefrin, "An Economic Theory of Self-Control, "Journal of Political Economy 89 (1981), pp. 392-401; and H. M. Shefrin and R. H. Thaler, "The Behavioral Life-Cycle Hypothesis," Economic Inquiry 26 (October 1988), pp. 609-643.
13 See, for example, J. Skinner "Individual Retirement Accounts: A Review of the Evidence, "Tax Notes 54, 2 (January 1992), pp. 201-212.
14 D. Kahneman, L. Knetsch, and R. H. Thaler, "Experimental Tests of the Endowment Effect and the Coase Theorem, "Journal of Political Economy 98 (December 1990), pp. 1325-1348.
15 S. Benartzi and R. H. Thaler, "Myopic Loss Aversion and the Equity Premium Puzzle," Quarterly Journal of Economics 110 (February 1995), pp. 73-92.
16 R. Mehra and E. C. Prescott, "The Equity Premium Puzzle, "Journal of Monetary Economics 15 (1985), pp. 145-161.
17 Specifically, we assume investors act in accordance with cumulative prospect theory. See A. Tversky and D. Kahneman, "Advances in Prospect Theory: Cumulative Representation of Uncertainty, "Journal of Risk and Uncertainty 5 (1992), pp. 297-323.