Hacker News Reader

Comment by lidHanteyk | Hacker News Reader

This is unsurprising. P likely is not equivalent to NP [0], and predicting the market is NP-hard [1]. It's nice to see empirical work in the field, though, and especially nice to see reproductions of published papers.

[0] https://www.scottaaronson.com/papers/pnp.pdf

[1] https://arxiv.org/abs/1002.2284

Edit for downvoters and repliers: If enough market participants are irrational, then it can still be possible for people to predict other people, instead of the market, and make money that way.

NP-hardness indeed doesn't rule out heuristic approaches, but experience with 3-SAT and other NP-complete problems suggest that there will be arbitrarily bad times, and that in those times, the amount of loss can be exponential in the length of time that the heuristic poorly predicts the market.

View on HN

Your second link doesn't say predicting the market is NP-hard. It says the opposite: that the market is only efficient if P=NP.

According to the paper, if you don't believe that P=NP then you believe that the market is inefficient, which means there's profit to be made. The paper even suggests how.

What do you mean? People are successfully making millions or even billions on predicting the market. Seems like a stretch to relate this to P=NP

You are talking about writing an algorithm that has a 100% accuracy on an NP-hard problem, and taking the impossibility of this to discard an approach that may yield 58% accuracy.

Philip Maymin seems like a serious guy... but that EMH ↔ P=NP paper is absolutely not even remotely a proof. Was genuinely very curious and it's at best an intuition. Some claims, e.g. Knapsack and 3SAT are (almost?) isomorphic to the efficient market hypothesis, are pretty bold. And the justification is hand-wavy at best.

It's not just hand-wavy. It's outright wrong. The size of the search space isn't just exponential in the amount of history -- that's the number of possible histories. The number of mappings from histories to positions is doubly exponential, so they can't even be uniquely described in less than exponential space.

He's not a serious guy at all, he's a nutjob who likes listening to himself talk.

Why must serious guy and nut job be my only two options. And why must they be mutually exclusive.

Warren Buffet must have a secret proof of P=NP then, that would explain how he has made billions trading.

Warren Buffet is an interesting case, because his approach seems largely to consist of ways to reduce the N of what he is considering: choose a few bets, bet big, hold for the long term, choose bets where you already understand the market. Sampling, rarely recalculating and selecting subsets of N where the related data is already in cache are all well-understood techniques for addressing computationally-hard problems.

He's done a good job of avoiding expensive mistakes, not being pushed to invest cash in something marginal just because it's sitting there.

Half the battle is not doing something stupid.

NP-hardness is irrelevant here because algotrading is not 100% formalizable.

Reply on news.ycombinator.com