4 ms·
I have only had time to skim through this very quickly, but it seems entirely based on false definitions. He begins so: The efficient market hypothesis claims
by pron 14y ago
I have only had time to skim through this very quickly, but it seems entirely based on false definitions.
He begins so:
The efficient market hypothesis claims that all information relevant to future prices is immediately reflected in the current prices of assets. In other words, you cannot consistently make money using publicly available information.
Then, he encodes a 3-SAT problem into market orders and claims:
So what should the market do? If it is truly efficient, and there exists some way to execute all of those separate OCO orders in such a way that an overall profit is guaranteed ... then the market, by its assumed efficiency, ought to be able to find a way to do so.
So he defines market efficiency to be all-knowing, and then gives at an NP-hard problem to solve. It seems like he could have given it the halting problem as well, only it might be harder to encode as market orders. But the point is that an efficient market is "all knowing" only that which is "knowable". The market isn't an oracle that can answer any question about the past, but a true machine that can answer all questions about the past that are answerable using the machine.
What he's doing seems similar to saying "if God exists then P=NP". After all, if God is omniscient and omnipotent, He could solve all NP problems in polynomial time. :)
- ubernostrum 14y agoI think this demonstrates, in a clever way, the types of absurdities that result from assumptions economists make about markets and the people who act in them :)
- uvdiv 14y agoI skimmed it quickly too, but I think you've misread. The "interesting" claim isn't the 3-SAT encoding; it's earlier where he claims that markets "naturally" have to solve NP-hard problems, in particular that to optimize an investment strategy you have to solve the Knapsack problem. (?!) It's hand-waving nonsense.
- pron 14y agoIt doesn't matter. When people say efficient market, they mean that the market knows whatever is knowable by the present time using current technologies. They don't mean that the market knows everything that's theoretically knowable using unknown/impossible algorithms. Before there were computers in an efficient market (supposing one exists), the current prices would reflect everything that is known and can be computed/deduced without computers by the present time. Anyway, the paper misunderstands "known" to mean "anything which can theoretically be known", which obviously includes solutions NP-hard problems. It's just that nobody claims that's what efficient markets are.
- 001sky 14y agoRE: When people say efficient market The hypothesis is predicated on perfect rationality. http://en.wikipedia.org/wiki/Perfect_rationality http://en.wikipedia.org/wiki/Perfect_rationality vs http://en.wikipedia.org/wiki/Bounded_rationality http://en.wikipedia.org/wiki/Bounded_rationality
- Dn_Ab 14y agoI'm surprised he decided to use 3-SAT instead of pointing out the non-convexities which arise in optimization for many interesting markets. Or the problem with digital goods say. That these are exploitable is what leads to excess returns and monopolies. These are failings due to reasons one can argue with computational compexity (non convex optimization is in NP).
- csense 14y ago> Or the problem with digital goods say. That these are exploitable is what leads to excess returns and monopolies. Copyright and network effects are the main reasons monopolies exist in digital goods.
- Dn_Ab 14y agoThe issue with digital goods is also the matter of uncertainty, the issue to do with scarcity or lack thereof and the trickiness of information as a good itself. In essense there is a built in inefficiency for information goods. But the points you bring up are correct and extend beyond digital goods.
- aurelianito 14y ago'If God exists then P=NP. After all, if God is omniscient and omnipotent, He could solve all NP problems in polynomial time. :)' Great quote! I will take it.
- gizmo686 14y agoBut it is not entierly true. P=NP means that the set of problems that can be solved in polynomial time on a turing machine is the same set as could be solved on a non-deterministic turing machine. Assuming God exists, I think it would be safe to say that he is capable of using atleast a non-deterministic turing machine, so his ability has no bearing on P=NP. Similarly, it is possible (theoretically at least) that the market has access to some means of computation not available to a turing machine. Given that most of the market is run by traditional, turing, computers, that does not seem to be the case.
- pron 14y agoExactly! The definition of an efficient market makes no claim to the market's computational power whatsoever. Computation and efficiency are unrelated. All it means is that the market would make the best use of whatever computations it can carry out. The paper's author, however, quite funnily understands efficiency to imply magical computation, and then "proves" that this would require solving NP-hard problems. He first understands efficient to mean godlike, and then shows that for God to work, He would need to solve NP-hard problems quickly. Well, duh.
- gizmo686 14y agoEffiecient market, as I understand it, means that the market would make the best possible desicions given the information it has available. Working with this definition, the market would need to solve NP-complete problems quickly. As I think is obvious, the market cannot (heck, at the speeds the market runs, I doubt it can even solve most P problems in time). It seems like the definition of effiecent market you are using would mean the market makes the best decisions given the information and computational resources it has. I would make an even weaker claim, saying the market makes the best possible decision, given the information, computational resources, and known algorithms to utilize the information and computational resources, and given players whose goal is not achieving an optimal outcome from the market.
- Evbn 14y agoOmniscience and omnipotence don't say anything about algorithmic runtime.
- lobo_tuerto 14y agoIf you are omniscient and omnipotent then there is no runtime.