Not to buck the trend, macroeconomists have followed the same type of pattern. Over the last half century, macroeconomists have built models that employ either macrovariables, representative agents, or both (Hirschleifer. 1970; Kohn 1981; Lucas 1972). The move to DSGE simulations represents an extension of this equilibrium paradigm (Wagner 2011). The relaxation of some assumptions, such as the assumption of perfect information, still leaves economics with the same core. Agents in this paradigm are automatons driven by a singular desire to maximize a utility vector. The motives of these little resemble that of agents in the real world. To the extent that we are wrong, we would prefer not to associate with such avaricious, monotonously motivated agents.
The social world is far richer than the description provided by what has come to be accepted as the neoclassical paradigm. While a perfect model of reality would be redundant, the dominant paradigm lacks the fidelity necessary to be considered a simplified replication of a reality within an open system. Action of economic agents are not driven by solely by price and quantity vectors, although they do interact with prices and quantities. To the extent that they are, a subjective interpretation of price must be employed (Mises 1949). The world which economic agents inhabit is a subjective one. In this world, agents are not only imperfectly informed; they imperfectly perceive the world. They are certainly in no position to optimize (Chaitain, Doria, and da Costa 2011). Even when the world is defined in terms of prices and quantities, agent optimization according to the typical calculus is computationally intractable. In the least this is an insurmountable problem for the Walrasian auctioneer (Axtell 2005).
It is necessary, then, that economic theory not only drop its assumption of perfect information, but that economists altogether reevaluate the perceptual framework that modeled agents inherit. Theory must identify and employ only the elements most significant for the domain of study, but with the realization that agents interact with particular objects at a given place and time. An appropriate framework allows for the appearance of general contexts that represents the details of that contexts as general types. A ecological, or realist, framework of perception and rationality fulfills these requirements.
An ecological perspective can be described as containing heterogeneous and interacting agents. These agents inherently interact with the world in a manner consistent with Bayesian updating.
In the ecological view, thinking does not happen simply in the mind, but in interaction between the mind and its environment. This opens up a second more efficient way to solve the problem: to change the environment. The relevant part of the environment is the representation of the information, because the representation does part of the Bayesian computation [emphasis ours]. (Gigerenzer 2008, 17)Agents offload computation onto objects and systems of objects that they perceive in the course of their existence. Agents do not optimize, they adopt patterns of behavior that tend to promote predicatable outcomes. One such family of decision-making rules is what Gigerenzer and Goldman term “Fast and Frugal” (1996). Fast and frugal rules allow the agent to make a decision based only on a single piece of information. (Are dark clouds forming to the immediate west? It is probably a good idea to wear a coat until they either subside or until a storm has come and gone. One does not need to check the weather report under such conditions.) Agents do not have the time to collect all available information when making decisions. In a world that is part of an open system and where variables of interest are not independent of one another, a fast and frugal rule actually outperforms regression analysis in predictive power (Gigerenzer 2008, 41). As agents grow accustomed to these rules, they become ingrained in habit and thereby reduce the computational work required by the agent.
Rules that guide agent action take a vast variety of forms. For example, an agent investor may only invest according to fundamental measures. Others may invest according to past data. Still others may copy the investment decisions of investors who consistently beat the market. An obvious consequence of action guided by these rules is agent interaction. Agents influence one another by influencing conditions of scarcity and by interacting with and copying one another directly. The first of these is accounted for, if only imperfectly, in the modern formulation of economic theory. The latter is non-existent in that realm. A convenient approach to framing this latter problem is in terms of an information cascade where the information transferred is a decision making rule (Earl, Peng, and Potts 2006). The rules that appear to promote an agent’s ends are copied by other agents (Hayek 1962; Bikhchandani, Hirshleifer, and Welch 1998).
Notice that this framework for decision-making fits nicely into a theory of expectations that does not assume the end it is supposed to prove. Consider the Lucas critique (1972). Lucas noted that any relationship among macrovariables will tend to disappear one it has been recognized. Proponents of rational expectation would argue that agents make predictions about the future using all available information and that agent predictions of the future are conveniently distributed around a median outcome that, absent information shocks, represents an accurate prediction (Muth 1961, Fama 1970). This model looks nothing like social reality. It suffers from the same problem that Gigerenzer identifies above concerning Bayesian computation. It has in its favor a modest degree of predictive power, but lacks the fidelity that is required for deeper understanding. The alternative contained within an ecological framework suggests that agents form rules that conform to their interpretation of reality. If agents expect that there will be a high degree of inflation, they will substitute assets in their portfolio in lieu of cash. They only need to know what sort of action to take given expectation of a particular circumstance. Those agents who tend to be better at predicting will tend to maintain larger stocks of wealth than those who predict poorly. As agents receive feedback concerning their actions, they will update if they believe they were following an inferior pattern of action. Over time, this allows strategies to be developed, tested, and either discarded or duplicated.
Roger Koppl describes this process in his theory of expectations. Borrowing from Schutz and Hayek, Koppl posits two types of expectations formed by agents: cognitive and acognitive (2002). Cognitive expectations represent conscious predictions of the world given some information. Conscious predictions, however, are limited in their scope inasmuch as they only affect agent action a single time. Agents also form habits over time that ideally promote their continued existence and prosperity. The old maxim, “early to bed, early to rise,” for example, encourages the formation of acognitive expectations with respect to ones bedtime and waketime. Those who follow must assume that action in concordance with the maxim promotes a state of affairs that is superior to a world where the agent lacks such a habit. Returning to the inflation example, agents may learn to immediately purchase assets – maybe real estate, stocks and commodities – whenever they hear a trusted source of information suggest that there will be inflation in the future. Any misallocations that occur in this process will be smoothed out over time by competing arbitrageurs in the long-run. This does not exclude the possibility of economic volatility in the meantime as relatively ignorant agents compete with one another, collectively discovering the true conditions of the market by a process of trial and error (Hayek 1942; [1968] 2002).
Over the last few decades, substantial progress has been made in understanding of human perception and that perception's interaction with the environment. Reliance on rational expectations by economists have prevented them from taking advantage of this progress. While rational expectations is useful in justifying econometric work which represents the bulk of applied research in the last half-century, there exists an opportunity to return to the mode of realist theorizing that dominated economics before World War II. Finally, I have not mentioned agent based modeling above. Econometric analysis was the workhorse of the most recent epoch of economic thought. I expect agent-based modeling to become the workhorse of economists working with pure analysis. Many of the components for this already exist (see Simon 1996; Hayek 1962; Crawford and Ostrom 1996; North and Denzau 1994; North 2005). It is only a matter of time before these methodologies are employed to create agent-based simulations across our field.