Behavioral Rational Choice

Economics is often interpreted as the study of how people make choices in the allocation of their resources. Microeconomics hinges on this process through which agents come to make decisions and then act on these decisions. Indeed as much of contemporary economics is considered to be derivable from micro foundations, choice theory plays a foundational role in the whole contemporary enterprise of economics. However, how we combine our beliefs and desires to make decisions and then act on them is widely considered to be a major unresolved puzzle.  The standard approach to interpreting human choices within an economic context became formalized into the idea of rational choice theory. The limitations of rational choice theory have long since been noted, but today they are coming under increasing critique and once again leaving open the question of how people really make their choices in an economic context.

In this article, we explore both the standard approach of rational choice theory and newly emerging approaches based on behavioral economics and agent-based modeling. We will look at each aspect that supports the rational choice model, including consistent choice, utility, model expectations, the representative agent, methodological individualism and the macro-level equilibrium outcomes that it leads to. In parallel to this, we will examine how these assumptions only really hold in simpler environments and have significant constraints when dealing with complexity. We look at the idea of heterogeneous value, adaptive choice, heterogeneous agents and agent-based modeling, the limitations to methodological individualism and non-equilibrium macro-level outcomes.

Rational Choice

Rational choice theory is the idea that economic agents have perfect information, both in space and time. They can know all prices and products available across an entire market. They are also perfectly informed of all the events that have previously occurred or will occur in the future through the use of probability, and they have an infinite capacity to compute all this information. From this they can, on aggregate, derive information that is directly correlated to some kind of underlying objective reality and then will act in a logically consistent manner upon this information. This is the so-called rational expectations hypothesis, which basically says that the information that agents act upon is known and cannot be systematically inaccurate or random. It must be correct on average. An important thing to note here is that standard economics does not say that individuals never make mistakes. It recognizes that individual people sometimes make mistakes but states that on aggregate they do not.


The concept of value will, of course, be central to any model for describing choice. Fundamentally there are two models to value within an economic context, that of subjective value and that of objective value. Subjective value is the idea that economic value exists only in relation to some subject and is a function of the usefulness that the object under evaluation has in achieving their desired ends. Objective value is the idea that value may exist independently from the subjective evaluation of an entity, instead being derived from the objects functional role within a broader social, environmental or cultural context. For example, the value of a tree may exist in its role within a broader ecosystem it is a part of, and the ecosystem services that the whole system provides.

Rational choice theory is based on a subjective formulation of value, also called the extrinsic theory of value. Which is a theory that advances the idea that the value of a good is determined by the importance an acting individual places on a good for the achievement of his or her desired ends; what is called utility. The theory of extrinsic value posits that value cannot be measured or observed directly. So instead, economists devised a way to infer underlying relative utilities from observed choice, called ‘revealed preferences.’ Utility always defines value in relation to someone. Thus, within this model, something cannot have value independent from someone desiring it. When value is defined by the interaction between different well-defined utility functions, this means that the thing does not have an intrinsic value.  Within this model, value is mono-dimensional and seen to be homogeneous. That is to say, that all forms of value can be reduced to a single form of exchangeable monetary value. Thus a comparison of different options an agent has is a simple comparison or optimization according to one single metric.

Heterogenous Value

In contrast to this behavioral economics, psychology and much empirical data will present the case that value is not a homogeneous thing but instead people value different things, social capital, natural capital, financial capital, cultural capital and they do not perceive these different forms of value to be reducible to each other. Thus actors in the real economy act in a way that is taking into account and making trade-offs between these different forms of value.

For example, we can go to a restaurant and pay for the meal that we consume, this is considered appropriate in that it is accepted that the cooks and waiters work is exchange for money. However, we perceive that it is not acceptable to go to a meal with one’s girl friend’s parents and then pay her mother at the end of the meal for her work. What would be acceptable though is that one might invite them out to a meal in exchange, or do some other activity for them. This is because we make a clear differentiation between the different forms of value and they can not be reduced to a homogenous form. Sometimes we can reduce all forms of value to a single form and make direct comparisons between them. But just as often we recognize their difference and then have to make a more complex trade-off between different forms of value.

The result of this is a more complex model to how people make choices, where they are never quite sure of the value of things and they are continuously redefining what things they value and the relationship between these different forms. When we want to go on holiday we may first consider the price of that and its expected utility to us. But we also may consider, how many holidays our friends have taken and whether they will think we are lazy taking another. Whether it fits with our self-image. We keep in mind the picture that we saw on the advertisement of the golden beach. We perhaps may consider the cost to the environment, etc. These are different costs and benefits that can’t be reduced to a single metric but in fact, in order to model them, we need to define different forms of value, social, cultural, financial, ecological and then try to optimize that network of values and how they interact. This is a more complex model that requires us to incorporate many more factors into people’s preference.

Consistent Choice

An important component of rational choice theory is what is called consistent choice. Once an agent’s preferences have been revealed, they are not allowed to change unless the properties of a good they are evaluating changes. If I prefer an apple to an orange now, I cannot change my preference unless some property of the good changes. But if all the properties stay the same and instead we add a banana to this set of choices and I then make the choice again, I am not allowed to choose the orange this time because in both cases I did not choose the banana. It is considered not part of the optimization equation and thus cannot alter my choices. The banana is exogenous to the equation. It is part of the context, and within this model, that context cannot alter what happens in the equation.

Put more formally, if an agent chooses X over Y then whenever X and Y are available she will always choose X over Y. What this makes explicit is that the context cannot change the evaluation of things. The context cannot add or subtract any value. As long as this condition of consistent choice holds, we can then model agents’ behavior in terms of an optimization over a set of choices and we will get a closed form solution. In our previous example, because the banana should not have affected my choice, we can abstract away from the particular context within which the choice is being made, assuming that this context does not add or subtract value to the properties of the goods under evaluation, and thus we don’t need it in our model.

What this means essentially is that the value of something is in the properties of that entity and people’s perception of them. You are not allowed to change your evaluation of it unless its properties change. If you suddenly notice that the apple has a little blemish on it, then its properties have changed and you can change your evaluation of it and choose another option, this is considered rational. But it is considered non-rational to change one’s mind based on what are considered exogenous factors.

Context Dependent              

Much empirical data from behavioral economics has shown that people’s evaluation of things is framed by the context. This context involves many social, cultural or even environmental factors, all of which can add or subtract value to a good, service or activity. Thus, in order to capture this concept of objective value, we need a much more complex multidimensional conception of value, one that incorporates all of these factors. One that includes the different dimensions to people’s conception of value, social capital, cultural capital, industrial capital and ecological capital.

Many experiments from behavioral economics will tell us that value does in fact change depending upon the context, and not only this but value may in fact sometimes be fully defined by the context. For example, the value of a color is dependent upon what color is placed next to it. The perceived value of a product in a magazine is dependent upon the other products in that magazine that they were never going to purchase, but the price and quality of those other products did create the context for their evaluation of the item that they did purchase. With hyperbolic discounting, our preference for different payoffs is dependent upon the context of time.

This phenomenon that value is not just dependent upon the properties of something in isolation but also upon its context is everywhere and very intuitive to us, but incorporating context as a defining factor in value would lead us to a whole new paradigm of objective value, where the value of something can be at least partially independent from the value ascribed to it by some agent in isolation. With objective value theory, the change in value may derive from the item’s relations to other things, and thus dependent upon those other things, what we can call the context. Thus, this context can, in fact, have some value and it can add or subtract value to any individual item within that context. Allowing for this empirical fact will give us very different models, where value and preferences are dynamically changing depending on the connections and context.

Out of this more complex conception of value, we get something that is able to approximate the idea of well-being. In that we know that well being is not captured in a single value such as GDP, but in fact is a much more subtle thing that emerges out of one’s connections with the things that one values – friendship, sense of purpose, respect from others, security, health etc. all of which this more complex metric of intrinsic value tries to capture. What behavioral economics and the empirical data coming out of it have shown, not surprisingly, is that people are in fact, people. They are complex creatures. They don’t just value one thing. They value lots of different things. What people really want is well-being, and well-being is not just one thing and people are then making a trade-off between different forms of value.

Unstable Preference

Behavioral economics draws upon neuroscience and evolutionary biology to present a picture to human decision making that is driven much more by irrational instincts, primordial motives such as hope, fear, and greed, that all totally bypass any kind of abstract isolated reasoning based on objective information.

Agents are driven by motives and these motives frame our whole point of reference. So if your care motivation or fear motivation is activated, then you will interpret information through this context. You will interpret signs differently, seeing cues that symbolize these things more readily. Motivation organizing one’s perception is very different from the computation model to how humans interpret and process information. This agent with limited cognitive capabilities is placed in the world with a single location at a single point in time. In this scenario, information is scarce. Agents may only have access to local information and the future represents a deep uncertainty. With all of this lack of information and incapacity to process it all, we use all sorts of shortcuts that allow us to cope in complex environments. We make many irrational associations between things that aren’t always apparent. We make reference to the context and our environment, such as simply copying other people. We think in scenarios and narrative. Everything has to fit into a context for us to make sense of it and that context can alter the meaning and value of anything within it.

Adaptive Choice

Much of how people make choices in the real world is the product of an adaptive strategy. Instead of always having the same preference and making the same choice, people’s choices evolve over time, the choices we make adapt to past events and information. We take actions in the world, see how those actions affect the world, and then make choices based upon that. Continuously updating and adapting our strategy in an evolutionary fashion, where we select from those choice patterns that worked best in the past. Much of human behavior that is non-rational is a set of evolved adaptive strategies for survival and homeostasis.

In economics, adaptive expectations is a hypothesized process by which people form their expectations about what will happen in the future based on what has happened in the past. For example, if inflation has been higher than expected in the past, people would revise expectations for the future. In such a way people take a limited amount of past data to predict future events. Adaptive expectations economize on the need for information as the agent only needs a limited amount of information on past period values. Adaptive expectation is time-dependent in that it puts more weight on the near past as opposed to information from the distant past. As one moves farther into the past, the effect on the current choice is reduced.

Adaptive expectations can be seen during periods of financial crises where actors have lost significant value and institutions have been shown to fail. In such a case markets can seize up as agents use their recent information naive. Adaptive choices are typically based on simple heuristics, such as naive expectations, where the agent’s forecast is simply a continuation of the last state that they observed or experienced. Or it may be a simple change rule heuristic, for example following previous changes, if economic growth has been growing by one percent in the previous cycles then extend this out into the future etc.

Homogenous Choices        

The standard economic theory of action is built on the idea of the representative agent. An economic model is said to have a representative agent if all agents are of the same kind and will thus act identically. Many macroeconomic dynamics stochastic general equilibrium models today are based on this idea of the representative agent.

The homogenous agent models work only as long as the deviations from the representative agent are not correlated. Typically this requires that the actors are acting independently, without interdependence between their actions and thus deviations can cancel each other out with negligible effects on the aggregate. When economists study a representative agent, this is because it is simpler to consider one ‘standard’ or ‘normal’ decision maker instead of looking at many different decisions at the same time. It is an analytical convenient shortcut required to achieve closed form equation based models within a complex world. Of course, this assumption must be left aside when differences between individuals are central to the question at hand as we will discuss.

A corollary to this is what is called model expectations. Model expectations underpins the rational agent model and means that people will make mistakes and will get misinformed, but because this is random noise it will on aggregate cancel itself out, and thus on aggregate people will act according to the model’s expectations. Because it is random, meaning not correlated, one person’s misinformation that goes in one direction will be canceled out by another person’s misinformation that goes in another direction. If we add all these up, we will get some kind of equilibrium that is a correct representation of the actual underlying information.  

This is part of how we can get very abstract clean models out of very complicated and noisy data. It works by assuming that the individual has the same properties as the average, and this will work on aggregate as long as the system is linear. This is why we will never get a model for any specific individual agent. We will always be talking about averages and aggregates because we have to take everything to the aggregate level to cancel out the noise and then assume any individual is identical to this average individual that is derived from the aggregation. The net result of all of this is that we can on average expect people to behave as if they have perfect information and act rationally upon it. Rational expectations hypothesis is a necessary condition to obtain internal consistency in stochastic dynamic aggregate models in economics. All of this will of course only work if we are dealing with a closed linear system.

Thus, it is assumed that outcomes that are being forecasted do not differ systematically from the market equilibrium results. As a result, rational expectations do not differ systematically or predictably from equilibrium results. That is, it assumes that people do not make systematic errors when predicting the future, and deviations from perfect foresight are only random.

A model that contains many different agents whose choices cannot be aggregated in this way is called a heterogeneous agent model. In such circumstances, it becomes more the correlations between actors choices rather than the choice of the representative agent that comes to define the aggregate outcome. In such circumstances, one must look at both the network of interactions and the differences among the actors in order to form models that will describe the macro system. Methodologically this requires a combination of both agent-based modelling and network analysis.

Agent-based models are the primary alternative to the limiting assumptions and model of the representative agent approach. Computers can deal with thousands and even millions of free parameters, and thus it is possible to code models with different expectations, preferences, information, propensities, psychological framing and tendencies etc. Agent-based simulation allows the explicit representation and exploration of the complex relationship between individual behavior and society; the micro-macro link between them.

“Agent-based modeling is a computer simulation technique, that is you write a computer program so that the computer program when it runs does similar things to what you’re simulating. In this case, we’re trying to simulate people interacting in a social way. Agent-based modeling is a particular kind of that technique because in the simulation you have different bits of the simulation to represent separate people so that the simulation keeps track of the different people separately in the simulation. It doesn’t think of them as a complete bunch so to speak, it thinks of them individually. The interaction between the individuals is, of course, the key important part from a social science point of view and that is represented by interaction between the bits in the simulation… this is of interest because it allows for different group level phenomena to emerge out of the interaction of all the bits just like we seem to observe with real people… It is this link, this ability to link the individual behavior of people to the society level effects of their interactions that makes this technique particularly of interest to social scientists” – Bruce Edmonds Manchester Metropolitan University


The premise of rational choice theory as a social science methodology is that the aggregate behavior in society reflects the sum of the choices made by individuals. Each individual, in turn, makes their choice based on their own preferences and the constraints they are presented with. This approach is called methodological individualism, an approach where the whole social system is seen to be nothing more than the sum of its parts. All higher level socioeconomic phenomena are seen to be traced back to micro-level elementary parts – more generally this is the idea of reductionism. Thus no higher level phenomena can exist that can not be derived in a linear additive fashion from the micro elementary parts. The net result of this will be an equilibrium on the macro-level.

A key part of the rational choice theory is that agents are making choices in isolation and it is out of this that we are able to get stable equilibrium outcomes. In reality, though, people more often make choice in the context of others, so that how people act become synchronized.

When the choices people make becomes interconnected and synchronized things can all move in one direction and we get positive feedback instead of simply balancing negative feedback. The classical example of this being a stock market bubble. For example, if you look at the dot com bubble, what happens is that due to a few positive shocks new technology, the internet technology, pushes the prices up for fundamental economic reasons and these trend following rules then take over to reinforce this trend as possitive feedback takes over. This is an explanation of the dot com bubble due to trend following heuristics.

Instead of there being an economy of agents making choices in isolation leading to stable balancing negative feedback, when we allow for the interconnectivity between agents we get positive feedback, things can all move in one direction and we get non-equilibrium outcomes that we see during financial crises and other forms of herd mentality where more begets more.