Comment by viraghrr | Hacker News Reader

Comment by viraghrr | Hacker News Reader

I'm a crank. I found this article very valuable.

In 2010 I put online a brief refutation of the EMH, the efficient market hypothesis. Here it is: https://arxiv.org/abs/1011.0423

If I had seen this then, I think I would have followed the directions in the inset box, where the article suggests saying something like: "The reason I succeeded in finding an algorithm for P=NP is that I noticed that... No one else seems to see that this insight is very powerful. It is very useful since it implies ..."

I could have simply written: "Nobody has bothered to test the Efficient Market Hypothesis, but since I have some background in computer science I made an unusual mental connection, and it follows that it doesn't need to be tested. It cannot be true."

(My proof outlined a way to show it has to be false.) I remember my thinking at the time was that I didn't need to explain it at all, just make it correct and brief.

At the time, I didn't want to include anything personal in the paper. Perhaps I should have. It's not a very good paper.

Isn't this equivalent to saying "I have incorporated a company, it owns a lockbox, and only I know what's inside." Only an insider knows the real value. But even Wikipedia's summation of the strongest version of EMH says that it's impossible if insider information can be hidden: "If there are legal barriers to private information becoming public, as with insider trading laws, strong-form efficiency is impossible"

https://en.wikipedia.org/wiki/Efficient-market_hypothesis#St...

So it seems that your contention- that insider information that hasn't leaked yet obviously won't be reflected in the price- is spelled out already.

okay so going with your analogy it's like publishing a picture of what's inside the vault but you publish the picture encrypted. imagine the encryption is a 20-letter random password in winzip (imagine winzip is secure) with the picture of what is in the vault. Although only an insider knows what's in the vault, the public gets the encrypted picture. Every day you publish one more letter of the passwrod. So if the password is p?jbAL;nQpg-37(z(feH then on day 1 you publish just "p" (the first letter) so now it's as strong as 19-letter password since you gave them the first letter.

I don't know what will happen exactly between day 1 and day 20, but on day 19 anyone can just try every 20-leter password with the first 19 letters you've published by that date. So if by then you've published p?jbAL;nQpg-37(z(fe already they'll just try

p?jbAL;nQpg-37(z(feA p?jbAL;nQpg-37(z(feB p?jbAL;nQpg-37(z(feC p?jbAL;nQpg-37(z(feD

in winzip. Anyone can just try all of them until they find the one that unlocks it. So on day 19 if it's worth knowing, for sure someone will take a minute or two to try it. Not that hard if there's just 1 missing letter.

So for sure if you tell people it's a 20-letter password and give them 19 letters they will figure it out before you've given them the last letter.

All right, but on day 1 nobody can guess all 20 characters in the password. That's too hard. it's impossible.

So every day you make it easier, but it'll be "broken" sometime between day 1 and day 20, assuming that the insider information doesn't leak in any other way. that means that someone will break it but not on day 1. on some day between day 1 and day 20. The person who analyzes it "best" will have the insider information first, but for sure we don't have to wait until day 20 because by then it's ridiculously easy to analyze.

meanwhile you're making the analysis easier and easier. the person who can do the analysis best will have it first, and this proves that there's no "infinitely great analysis" already being done all the time by public markets. it's a simple demonstration of this fact. I don't think you actually need to do the experiment to now the results. hope this helps.

Your explanation does not disprove the EMH as far as I can tell.

Sure, there's some people who will have information sooner than others. In fact, the EMH says that "people who outperform the market do so based on chance" and it's expected that some people do win some of the time.

What better test of chance than who guesses the password soonest by luck?

Okay, fine, it's not luck you say, it's whoever has the best supercomputer to crack the password each time that happens. However, in reality, contrived examples like brute-forcing passwords aren't frequent parts of the stock market. Rather, it's events that traders predict the potential impact of, which is much closer to a random affair.

What you've shown in your comment is that it's difficult to clearly define private and public information in some contrived cases.

I think difficult analysis is a very frequent part of the stock market.

My argument was not so much about luck as about speed of analysis. EMH doesn't allow some people to analyze information faster than others. For example if Intel always releases a 450 page earnings at 11 o'clock, and it takes 4 hours for the market to really analyze it because it's written really abstrusely and that's the fastest anyone can read and figure out its meaning, then under the EMH there's no scope for 1 specific trader to read it in 1 hour and perform the analysis faster, before the price has adjusted to the new information. In fact, under the EMH the price has to instantly fully reflect the implication of the information of the report the moment it is released. If two hours in, a famous analyst (Warren Buffett let's say) tweets about it, and that tweet moves the price, under the EMH it must have moved the price because of new information. It can't just be because the famous analyst is smarter and analyzes better and faster than the rest of the market. Not possible.

That doesn't mean that until the tweet the analyst's better-than-market analysis is somehow "private" (rather than public) for 4 hours. there's no new information, it's just that the public price doesn't fully reflect the completed analysis. The market can be a bit dumb.

But how do you know the facts I just gave are true? How do you know the completed analysis can't be factored into the price the instant the report is released? (The price does jump instantly.) Maybe it jumps to the new fully completed analysis price? Maybe everything I wrote is not true and in fact the market is perfectly efficient, and only new information, or stochastic events, change the price? Maybe it fully reflects all public information all the time?

So I showed how to show that analysts can't all work instantly, yes, under a contrived example. The EMH doesn't require that markets only be efficient under conditions that aren't "contrived". But rather, all the time. you can run the experiment and show it if you want. you can literally see the stock market figure it out. They could also figure it our wrong. Maybe early on #117 will spike (so that prices look like this: https://imgur.com/a/C782Fgt ) - it could be for any reason. Maybe a false rumor that it was leaked (a false leak), whatever. But the market gives the rumor just a 10% chance of being true. But regardless of the reason for the spikes, at the end there must be a single spike, in the value of the stock that actually wins. it must be around $1 billion since it'll have that in cash. (It's important that people actually believe this will happen.)

Hi! I am a person who reads crankery.

First, thank you for self-assessment. It's rare.

This paper constructively suggests that P=NP iff EMH: https://arxiv.org/abs/1002.2284

It may be worth reading up on PPAD and Nash's thoughts on P=NP.

Do not despair, but do not expect to be correct, either. Explore and be wrong, but look for proofs, too. Above all, do not forget that crankery is fundamentally about being unwilling to see existing proofs and results. Always try to open your mind.

Thank you for the link. you say, "do not expect to be correct" but the abstract of the paper you linked says "Since P probably does not equal NP, markets are probably not efficient". do you agree with the paper?

There are two very distinct statements which say entirely different things: "markets are probably not efficient" is a fine statement and I agree with it easily.

"Markets are provably not efficient because xyz" is much harder to agree with, especially given that paper, since it's the same as saying you've actually proven that P!=NP and thus requires an extraordinary proof.

You're claiming to have a solid proof. Just because someone agrees that your result is likely correct, that doesn't mean they agree with your proof.

Right, two different things. I meant: 1) those results wouldn't be at odds with my results. 2) but I wonder if myWindoonn judges the paper they linked to be correct? (unflawed, complete etc - does it prove its titular claim?)

if the paper is flawed, nothing follows from it, there's no reason to spend time on it except to understand its flaws. Then it's just another crank paper, besides and unrelated to mine.

Not GP, but I do not. The paper is built on a pretty basic error:

> For simplicity but without loss of generality, assume there are n past price changes, each of which is either UP (1) or DOWN (0). How many possible trading strategies are there? Suppose we allow a strategy to either be long, short, or neutral to the market. Then the list of all possible strategies includes the strategy which would have been always neutral, the ones that would have been always neutral but for the last day when it would have been either long or short, and so on. In other words, there are three possibilities for each of the n past price changes; there are 3^n possible strategies.

The number of possible n-move histories is 2^n, not n. That means there are 3^(2^n) possible strategies. Nondeterministically searching that space in polynomial time does not work. The claim that a strategy could be verified in O(n) time is dubious as well since the size of a full description of a strategy is (in the worst case) exponential in the length of history it considers. So their claim that finding a technical strategy which performs well in backtesting is in NP is not proven.

They repeat the same error later:

> there are 2^(t + 1) possible strategies (because every possible strategy is a mapping from 2^t different possible sequences of 1’s and 0’s of length t into two possible choices, LONG or OUT)

There are other issues, like assuming a strategy that backtests well will continue performing well in the future.

Thanks, great analysis. (I take your point.) How about my paper I started this thread with:

https://news.ycombinator.com/item?id=19722455

Any thoughts?

Reply on news.ycombinator.com