3 ms·

> But if a model with more complex assumptions is a better approximation to reality than a model with simpler assumptions, would you not adopt the former? Not

• morningseagulls 7y ago
  >Maybe the simpler model is enough to understand what is going on. Maybe a more complex model suffers from over-fitting and performs worse than the simpler model. It really depends. If only physics departments everywhere would adopt this philosophy! Physics undergrads would then be liberated from the yoke of Einstein's relativity, let alone the tyranny of the mind-blowing quantum field theory with its insufferably complicated Feynman diagrams. It does not depend. The simpler model in economics is fatally flawed, and is the one suffering from over-fitting (to a straight line, no less) and contributing to the bad reputation of economics as a "dismal science". >> The SMD theorem guarantees, for example, that most market demand curves will NEVER satisfy the fundamental neoclassical "law" of demand. >Does it really say that? Or simply that weird market demand curves are possible in a few exceptional situations? Yes, it does.[0] The exceptional situations are precisely those that neoclassical economics assumes are the norm, namely, that there is only one agent and one commodity in the market.[1] As soon as you have more than one agent/consumer in an economy, the mathematics will undermine the "law" of demand, because the inter-agent interactions would generate non-linear terms in the market demand function. And once you have terms of higher order, the demand curve generated by that function will have sections that slope upwards, exactly what is being forbidden by the "law" of demand. And we haven't yet taken into account the very real fact that multiple consumers cannot possibly have the same preferences. Or that these preferences can change with different levels of income.[2] In short, neoclassical microeconomics only works in a communist economy of identical clones with identical preferences and incomes. Oh, the irony. [0] https://en.wikipedia.org/wiki/Sonnenschein%E2%80%93Mantel%E2%80%93Debreu_theorem https://en.wikipedia.org/wiki/Sonnenschein%E2%80%93Mantel%E2... [1] https://en.wikipedia.org/wiki/Representative_agent https://en.wikipedia.org/wiki/Representative_agent [2] https://en.wikipedia.org/wiki/Homothetic_preferences https://en.wikipedia.org/wiki/Homothetic_preferences
  • taffer 7y ago
    > If only physics departments everywhere would adopt this philosophy! Let's take electrical engineering as an example. If you model an electrical circuit, you assume that there simply is a voltage source where electricity comes from. That assumption is of course complete nonsense: In reality, electrical energy is not just there, it has to be generated somewhere, for example in a power plant. The electricity also flows through a grid that is shared with many other consumers. So, according to your requirement to always use the more realistic model, electrical engineers should always model a whole power plant and an entire power grid in each circuit, because otherwise their model would be bullshit. > Yes, it does.[0] Do you know of any empirical studies that show that the models currently used in macroeconomics actually cause problems with real data?
    • morningseagulls 7y ago
      >Let's take electrical engineering as an example. I'd usually say you're shifting the goal posts, perhaps muse about how dehumanising and simplistic it is to think of humans as components in an electrical circuit, and move on. Then I thought that it'd be more illuminating to address your analogy, because your reasoning suffers from the same problem that I've been talking about. >If you model an electrical circuit, you assume that there simply is a voltage source where electricity comes from. That assumption is of course complete nonsense: In reality, electrical energy is not just there, it has to be generated somewhere, for example in a power plant. The electricity also flows through a grid that is shared with many other consumers. I think it might help to read Musgrave's paper[0] (the abstract will do if you can't get the paper itself) on the three kinds of unrealistic assumptions, in which he attacked Friedman's twist. The assumption you pointed out is the most benign type: a negligibility assumption. It's like assuming that a object falling through air experiences no friction: complete nonsense, but air friction exerts only a negligible effect on most falling objects, unless the object is, say, a feather. In the case of your example, taking into account the power plant adds nothing to the analysis, because the electrical circuit you're concerned with usually doesn't output electricity. Usually. But in the 21st century, that can change... >So, according to your requirement to always use the more realistic model, electrical engineers should always model a whole power plant and an entire power grid in each circuit, because otherwise their model would be bullshit. Firstly, I did not say the more realistic model should "always" be used. The situation in economics is more like the more realistic model is ignored in favour of the incorrect unrealistic model, even when the more realistic model is appropriate. The analogy with Newtonian/Einsteinian mechanics is quite shaky here, since as I've said, Newtonian mechanics is vastly more accurate than neoclassical microeconomics. To take another example from engineering and continue the theme of relativity, error corrections required for GPS depend on an understanding of relativity.[1] If, as you've previously suggested, the simpler and more easily understandable model should be preferred, then we would not have GPS and allied technologies today. Secondly, in the 21st century when we have renewable energy, electrical engineers are now required to take into account a power grid, because consumers are now also power plants if they have solar power, for example, and want to feed the excess power generated into the power grid. The point is this: if the circuit you're modelling is a passive one that doesn't generate electricity, sure, make use of the unrealistic assumption that there's an unknown voltage source. However, if your circuit is also a voltage source, that assumption is then false, and you need to update your model. The same should apply in economics. Unfortunately, it turns out that markets and economies are far more complex systems than power grids: as soon as you have more than one agent (e.g. humans, corporate entities, HFT algorithms, etc.), the simplistic analysis derived from marginal utility theory fails. So, outside of the laboratory perhaps, the models used in neoclassical microeconomics are probably almost never appropriate for the problems they're treating. >> Yes, it does.[2] >Do you know of any empirical studies that show that the models currently used in macroeconomics actually cause problems with real data? This is a really garbled question. Firstly, nowhere did I assert that economic models "cause problems with real data". I really don't understand this assertion and would welcome any clarifications. Secondly, I think you misunderstand the point of my argument about the SMD theorem. The point is that, as soon as you have more than one agent/consumer (human, firm, algorithm, what-have-you), they will interact in a non-linear way, giving rise to a nonlinear market demand function that gives a "weird" market demand curve. This has nothing to do with any effect that the belief in a certain economic model would have on the economic behaviour of an agent. That is an entirely separate question altogether. Back to your question about whether "weird market demand curves are possible" only "in a few exceptional situations". The SMD theorem basically says no: in all but the most exceptional cases, multiple market equilibrium points can exist. The implication is that the aggregation problem[3] is not solved by the neoclassical assertion that the market demand curve for a well-defined commodity has the same pleasant mathematical properties -- in particular, that it is monotonically decreasing ("downward-sloping"), i.e. obeys the "law" of demand -- as the demand curve for that commodity of a rational agent. And the reason why the aggregation problem isn't solved is because, as soon as you have more than one agent or commodity (like in most IRL markets), you get a complex system. And when you have a complex system, you can't draw line diagrams like you do in the textbooks. You need to do modelling and numerical analysis. There's a field called complexity economics[4] that's only beginning to do that, some 40 years after the SMD theorem appeared. Better late than never, but it's still a lot later than the other less "dismal" sciences, which have come to realised that they're dealing with complex systems and have updated their thinking accordingly. [0] https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1467-6435.1981.tb01195.x https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1467-6435.... [1] https://en.wikipedia.org/wiki/Error_analysis_for_the_Global_Positioning_System https://en.wikipedia.org/wiki/Error_analysis_for_the_Global_... [2] https://en.wikipedia.org/wiki/Sonnenschein%E2%80%93Mantel%E2%80%93Debreu_theorem https://en.wikipedia.org/wiki/Sonnenschein%E2%80%93Mantel%E2... [3] https://en.wikipedia.org/wiki/Aggregation_problem https://en.wikipedia.org/wiki/Aggregation_problem [4] https://en.wikipedia.org/wiki/Complexity_economics https://en.wikipedia.org/wiki/Complexity_economics