Lumpiness, Uncertainty and 'Infinite' Choice: Three Elements of the 'Utility Maximization' Fraud
January 3rd, 2015
by Unlearning Economics
This appeared originally at The Institute for Dynamic Economic Analysis Blog.
Agents in neoclassical economics typically maximise some function, such as utility (‘satisfaction’) in consumption, or profits in production. Taken at face value, this is a transparently unrealistic depiction of how the world works, since it’s pretty difficult for people or firms to truly ‘maximise’ anything – especially something as intangible as utility – in the real world. But economists have shown that if you start with some not-too-unreasonable assumptions about human behaviour, then you can derive relationships that are functionally equivalent to maximising behaviour.
Thus the names of the theories and their pop interpretation should not be taken too literally, as these are not the criteria on which the theories stand or fall. Unfortunately ‘not-too-unreasonable’ is not the same as ‘correct’. I believe that three relatively simple violations in the core assumptions of standard Walrasian theory prevent it from being a workable model of individual behaviour.
These share a common characteristic: they strip away possible choices – or information about these choices – and reduce the number of feasible actions an agent can take. It thus becomes generally infeasible for agents to take an ‘optimal’ action. Below I discuss the implications of this in each case, and then go on to suggest an alternative path for modelling the behaviour of economic agents.
Computational constraints: Utility maximisation may seem a plausible approximation to consumer behaviour in the 2-good models of undergraduate classes, or even in the ‘n-good’ models of more advanced classes (since n is generally left unspecified). Both cases are generally taught with straightforward examples using oranges, bananas and so forth. But once we start to look at how utility maximisation might work when people face choices in the real world, such as shopping in a supermarket, the limitations of the theory become clear.
A good demonstration of this was given by Steve Keen, drawing on an experiment by Reihard Sippel. The number of different plausible commodity bundles one faces in a supermarket is well off into the trillion of trillions. Even assuming one had super- human calculation speeds, it would take billions of years to find the optimal bundle. It is no wonder that people appeared to violate ‘rationality’ as defined by economists even when faced with only 8 goods and choose instead sub-optimal bundles, and change their decisions in repeated experiments.
A comprehensive review of the literature finds people systematically violate utility maximisation, in particular by using a wide variety of different discovery or decision-making methods (heuristics) to condense down their decision-making processes. For example, consumers usemental accounting to split their budget into categories, hence avoiding the trouble of considering the relative price of, say, rent, while buying flour. They use the ‘availability heuristic’ – the tendency of the mind to reach for what is nearest – and are attracted to offers or items placed in easy-to-access locations. They copy one another. Finally, with the ‘status quo bias’, after they do decide what to buy, consumers tend to stick to the same routines, brands, shops and types of good, thereby avoiding costly and exhausting recalculations. So consumer behaviour may bebetter predicted by the environment in which consumers are placed than any measure of individual preferences; the environment being how the shop is set up, what is ‘normal’ in society, what information and advice is available to them, and default options.
Knightian Uncertainty: Mainstream economics often relies on modelling ‘uncertainty’ by considering all possible future states of the world and then attaching a probability to each one. But this is something of an abuse of the term and is more aptly defined as ‘risk’, a measurable uncertainty, which is not uncertainty at all in the true sense (which is named Knightian uncertainty after Frank Knight). ‘Uncertainty’ is properly applied to the kinds of future events which we cannot predict and to which we cannot attach any probability. Examples of this include a new invention or the eventual displacement of Google. The information about these events cannot be modelled, because it simply does not exist. Knightian uncertainty may not only refer to far off eventualities; it could apply to the consumer’s knowledge of whether they will get food poisoning at a restaurant, for which there is no general basis for estimating a probability.
This may seem to imply nihilism for economic modelling, but it needn’t. As John Maynard Keynes suggested, agents themselves intuitively know the future is unknowable, but still have to form some decision-making rule to cope. They may follow the crowd, extrapolate from past trends, rely on rules of thumb, or simply guess. Subsequent work on behavioural economics and financial markets has shown Keynes was right. Investors tend to herd, rely on psychological heuristics, or guess in the face of uncertainty.
Lumpiness: In neoclassical economics, firms and consumers select the optimal point in a mathematical set or function. The key assumptions required for this is divisibility, which posits that you can always change any input, consumption good, or other variable (such as location) by a tiny amount, and thus converge toward the optimal outcome. A moment’s reflection reveals that this is rarely the case. A farmer cannot buy 3/8 of a sheep, and a consumer cannot purchase ½ a car. So firms and consumers often cannot select the optimal point due to lumpiness. Even if they were fully rational and well-behaved as defined by neoclassical theory, the optimal point may not be feasible, so they may have to select a choice which is ‘close enough’.
What’s more, firms must often purchase factors of production together. Adding another worker to an office when there are no extra computers will not result in a marginally smaller increase in total production, but in no increase at all. A decision to invest in a new factory may entail mass hiring, buying new equipment and more raw materials. The individual decisions being inseparable from one another, the sum of them will be a ‘leap’ across the production space in all dimensions, the. This results in a degree of indeterminacy. If the optimal choice is not feasible, firms may make use of other criteria in their decision making, since many different decisions may place their output levels into an acceptable ‘zone’. One of INSIDE’s regular contributors, Cameron Murray, has used this idea in his theory of return-seeking firms, where the term ‘seek’ is deliberately chosen rather than ‘maximise’, since firms may not be able to select the optimal point, but merely the closest feasible choice.
Given the necessary limitations of individual maximization, or optimisation, in the real world, the most resilient aspects of human behaviour may lie outside the individual. Faced with a complex, unpredictable and irregular world, the most rational way for individual agents to behave is to adopt habits, follow rules already in place, or copy their peers. Analysis incorporating this insight is not only more empirically relevant, it is also the best way to shield theories from the kind of reflexivity highlighted in the Lucas Critique. Long-standing cultural practices and institutions are unlikely to change at quickly as, say, inflation expectations during a period of stagflation. A realistic economics should acknowledge the critical limitations of these constraints and refocus on the context of institutions, environment and cultural practices in which agents are embedded. It would allow us to form a workable model of individual behaviour and to make reliable predictions.