Rate of price change, how to calculate - page 7

New comment
 
Avals:
pf>3 at least 500 transactions)))

This has nothing to do with determining market efficiency.
 
MetaDriver:
Hmmm. Strongly echoesthe"tachyon breakdown theorem". ;) Is this the birth of a new religion?


Sort of :). What we lack in life is the concept of sin :)
MetaDriver:

I don't even know how to estimate. Do you think that profitability is a bad criterion? Seems to fit well with your cornerstone brick.

I didn't understand the mutual information, I got lost reading articles and wikipedia - I didn't understand shit, I just got confused with terminology. I ended up with another unfortunate self-diagnosis and a series of "I'm a sucker for arithmetic" and "you shouldn't have left university". I was waiting for Lyokha the mathematician to slowly grasp the matter on the forums (complex notions are sometimes better understood and mastered by examples, rather than by reading definitions).


Profitability is a good criterion, but it is only defined for a specific system.

Mutual information is not too abstract, though. I myself have already forgotten almost everything, but now I looked - the formulas are not so complicated. There the problem is created by conditional entropy, because even for a modest number of characters in the alphabet you need quite big statistics to estimate conditional probabilities. As a concrete example, I once started to reason this way: suppose we want to see if Friday's bar depends on Thursday's bar. We need to code the bars somehow, so let's assume we have three binary signs, so we will have 8 letters. So the number of possible pairs is 8*7=56. In a year we have 52 weeks, the daily history, say for the pound, I was able to ask the terminal for 20 years, this means that on average each pair combination of letters 1040/56 = 18.57 times fell out. If the probabilities are highly unequal, then the frequency estimates will be good for popular pairs, and unpopular ones can probably be neglected. But if the patterns are weak, the statistics will be spread pretty evenly over the pairs, and then the available history is already kind of small, because at 18 events the statistical uncertainty will be 20 percent .

I don't claim that this is a good example of course, but in my time this very reasoning discouraged me from looking for abstract correlations :). Because there are enough statistics for correct estimations only in case of very simple "questions".

 
Avals:
nf>3 at least 500 deals)))

The problem here is that rare events also happen. So we need to show that this is not a rare event :).
 
avtomat:

It has nothing to do with determining market efficiency.
It has. An efficient market is mathematically a Markov chain and the best price prediction at any point in time and on any horizon is the last available quote. Consequently, any strategy has an equity Cb with mo=spread, as the equity is the sum of the individual pieces of the price series. Therefore, a stable difference between mo and zero is evidence of inefficiency.
 
Candid:

The problem here is that rare events also happen. So you have to show that it's not a rare event :).
Yes, robustness is a complicated issue. So for simplicity 500)))
 
avtomat:


Is it possible to formally express this very efficiency of the market? How do you define it? Can you give us a formula for calculating it?

If it can be formally expressed, then the assumption of inefficiency in terms of some numerical characteristic will also make sense.

If it cannot be formally expressed, then the hypothesis is no more than a mere speculation, having no practical utility whatsoever.

.

===================

Moreover, I believe that this "rubbish" is put into circulation for disorientation -- that is its true practical purpose.


Well, with Vladimir we are talking about some specific assessments. There are no objective arguments at the moment. Your position is understandable, but you have no real arguments either. So no one can convince anyone at the moment.
 
Candid:

Sort of :). What we lack in life is the concept of sin :)

No, I think there is an overabundance of such concepts. What is missing is a systemic feeling. I don't know how to explain it (I have it sometimes, although it is very unstable and poorly predictable). In this state, intellectual morality and the concept of sin are completely absent, but systemically ineffective (sinful?) actions just don't happen. "Not happening" is not quite the right term, for in this state my own actions are happening rather than being done by me (the will and volitional behaviour seem to be absent for lack of use). It's not very well explained, but anyone who is familiar with it will probably recognise it.

Profitability is a good criterion, but it is only defined for a particular system.

Although mutual information is not very abstract either. I myself have already forgotten almost everything, but now I looked - the formulas are not so difficult. There the problem is created by conditional entropy, because even for a modest number of characters in the alphabet you need quite big statistics to estimate conditional probabilities. As a concrete example, I once started to reason this way: suppose we want to see if Friday's bar depends on Thursday's bar. We need to code the bars somehow, so let's assume we have three binary signs, so we will have 8 letters. So the number of possible pairs is 8*7=56. In a year we have 52 weeks, the daily history, say for the pound, I was able to ask the terminal for 20 years, this means that on average each pair combination of letters 1040/56 = 18.57 times fell out. If the probabilities are highly unequal, then the frequency estimates will be good for popular pairs, and unpopular ones can probably be neglected. But if the patterns are weak, the statistics will be spread pretty evenly over the pairs, and then the available history is already kind of small, because at 18 events the statistical uncertainty will be 20 percent.

I don't claim that this is a good example of course, but in my time this very reasoning discouraged me from looking for abstract correlations :). Because there are only enough statistics for correct estimations in the case of very simple "questions".

I quite understand such logic. However, simple questions (dichotomous) may be enough to build a base of "primary oracles". They can then be mutually contextualised by pairwise analysis. In this case "statistical" of a sample with each iteration is halved, but there is a possibility to see not only "non-randomness" of regularities (or to diagnose a fit - a trifle), but also some general criteria of their identification, which is important for automation of the process of searching for regularities. The fact that the MO of the "twin oracle" is higher than the "elementary" is sort of understood by default. Sorry for writing too abstractly (this is deliberate because the technology of concentrating raw ore is even more expensive than a particular deposit. the forum is not the place for such revelations, I can discuss more specifically in skype).
 
Candid:

Well, with Vladimir we are talking about some specific assessments. There are no objective arguments at the moment. Your position is understandable, but you have no real arguments either. So no one can convince anyone now.

But I also do not have unsubstantiated assertions (judgments, hypotheses, assumptions) about the efficiency/ineffectiveness of the market ;)))
 
This hypothesis of market efficiency/inefficiency is a shaky pillar. Or should I say a fallacy.
 
avtomat:

But I also have no unsubstantiated assertions (judgments, hypotheses, conjectures) about market efficiency/inefficiency ;)))

--

This hypothesis of market efficiency/inefficiency is a shaky pillar. Or should I say a fallacy.

Whatever the hell with inefficiency. The scientific nature of the term attracts intellectuals like honey to a fly. Though in the very first lines the Ministry of Health wikipedia warns about unprovable and even unprovable...

What do you not like about profitability as a substitute?

1234567891011121314...23
New comment