Even though utility maximisation is ubiquitous in economic textbooks to model consumer behaviour, its usefulness is rarely demonstrated by evidence.

  1. Is there any evidence that some consumers do behave that way on at least some markets?

Utility maximisation applied to a representative agent is also often used to model the demand side of a whole market. However, here also, this is hardly ever justified by evidence. On the other hand, alternative models seem to have good properties too: for instance in Complex Economics by Alan Kirman, it is said that random consumption + heterogeneity of income suffices to construct an aggregate demand that decreases as price increases. Random consumption does not seem more stupid than perfectly informed utility-maximisation... but sounds definitely simpler.

  1. Is there evidence that utility maximisation used as a model of the demand side of a market is actually better than other (simpler) models of consumer behaviour?
  • 1
    $\begingroup$ Look for empirical tests of GARP. $\endgroup$ Jul 12, 2020 at 17:44

2 Answers 2


Since utility by definition measures something very innate or subjective, it is difficult to obtain any non-experimental data on it. Hence researchers focus on choice, which is something observed in practice. This is where WARP, GARP and SARP comes in. Of which, GARP is the most relevant.

Any data generated by a utility maximising agent (with well behaved preferences) will satisfy GARP. The converse is true as well through Afriat's theorem - for any data satisfying GARP, there exists a utility function that rationalises it.

Note that it is difficult to study aggregate demand for such conclusions as well. The Sonnenchein-Mantel-Debreu Theorem basically kills any predictive power in this regard. So I would suggest you look up papers that empirically test GARP from choice data. Here's one such paper: Echenique et al

P.S. Mentioning @MichaelGreinecker here since I just noticed his comment that gave the same answer. :)

  • $\begingroup$ "[A]ny data satisfying GARP must have been generated by a utility maximising agent" -- this is not true. Perhaps what you meant to say is that any data satisfying GARP is consistent with having been generated via utility maximisation. After all, one could randomly generate data that just happens to satisfy GARP. $\endgroup$ Jul 14, 2020 at 10:18
  • $\begingroup$ If by "consistent with having been generated via utility maximisation" you mean there exists a utility function that rationalizes the data, then yes - thats what I mean. If you could somehow randomly generate a data that satisfies GARP, Afriat's theorem will still spit out a utility function that rationalizes the data. If I remember correctly, the theorem remains mum about the true DGP. $\endgroup$
    – user28372
    Jul 14, 2020 at 14:50
  • $\begingroup$ You should probably edit your answer, then, as the statement in your answer is much stronger than "there exists a utility function that rationalises the data". In particular, your answer takes a stand on the true DGP. $\endgroup$ Jul 14, 2020 at 14:53
  • $\begingroup$ I don't see how it makes a stance on the true DGP, but I'll edit it to avoid confusion either way. Thanks. :) $\endgroup$
    – user28372
    Jul 14, 2020 at 16:39
  • $\begingroup$ Well, you do say that it "must have been generated by a utility maximising agent". That, to me, rules out that the data was generated randomly and just happens to satisfy GARP. $\endgroup$ Jul 14, 2020 at 16:40

There is plenty of evidence. Many successful policy interventions have been designed assuming that people are utility maximizers and deriving its consequences. To the extent that the model can correctly anticipate the effects of the intervention, and then this effects can be measured in the data, the model is justified.

The list is long, but many procurement auctions are studied this way, antitrust policy, social programs to alleviate poverty, optimal ways to design contracts, education policy, monetary policy to control inflation, the effect of taxes and subsidies, etc, etc, etc.

You touch on an important point. If there is another theory like the one in Complex Economics, that produces the same predictions (for example a downward sloping demand), then the data may not be able to help us figure out what model is better. The key is to find a model that can predict better.

I would also like to push back on which model is simpler. I do not know Kirman's model, but I think that assuming that people maximize their happiness given their knowledge and their means is pretty simple too.

  • 1
    $\begingroup$ With all due respect, I am not sure you are answering my question. You seem to be concerned with my point (1) but don't provide evidence that individuals do act as perfectly-informed utility-maximizers. (I would actually expect quite the contrary, given the what I know of for instance the work of Kahneman & Tversky.) $\endgroup$
    – Arthur
    Oct 16, 2019 at 22:33
  • $\begingroup$ I should probably be more specific about the individual level vs. market level distinction. The fact that many people use utility-maximizing representative agents does not make the model valid at the individual level. Textbooks generally present the problem at individual level, even if validity is probably more at the market level. $\endgroup$
    – Arthur
    Oct 16, 2019 at 22:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.