Why is the normal distribution not normal? - page 10
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
I'll bet you $5,000 that this claim is false ?
Well yes, it is a somewhat bold assertion by Sergei that there is a significant statistical significance to the increments. It's objectively speaking below the plinth there.
Well, yes, it is a bit bold of Sergei to say that there is a significant statistical significance to the increments. It's objectively speaking below the plinth there.
Have there been any studies? Maybe there is a link to another thread. Would really like to understand what patterns have been tested and how.to IlyaA
I understand that we are talking about the relationship of current increments to future increments. And I strongly suspect that the ACF of the first readings will be the proof, but a value like 0.4 (or so), to put it mildly, is questionable to demonstrate such an advantage. But maybe Sergei has something new in mind. And there were a lot of studies, so it's hard to remember at once.
PS: And there is one more subtlety, the formula of ACF estimation is not quite suitable for these series, besides it is necessary to estimate the reliability of ACF itself and error estimation for each lag found. It's not that simple there.
Bet $5,000 that this statement is false ?
Do you doubt the autocorrelation of volatility? And it's a relationship between price increments, albeit modulo.
My understanding is that it's about the relationship between current increments and future increments. And I strongly suspect that the proof will be the ACF of the first readings, but a value like 0.4 (or so), to put it mildly, is questionable to demonstrate such an advantage. But maybe Sergei has something new in mind. And there were a lot of researches, so it's hard to remember at once.
For example, candlestick analysis uses not only increments but also lock extrema and volume. I remember a month ago there was a thread about an EA based on candlestick analysis. I think it started positively. But I did not take part in it. And I wonder if anyone has investigated not only closing prices but also other characteristics of candlesticks in aggregate.Incidentally, the sum of non-stationary series can be stationary and have HP. Such series are called cointegrated series. Clive Granger got a Nobel Prize in economics for this. I haven't checked it myself, but elite for example claims that the yen exchange rate is cointegrated with the consumer price index (CPI). More precisely with the difference between consumer prices in the USA and Japan http://monetarism.ru/cointegration.shtml
to Avals
Not that I'm getting into other people's squabbles :o), but Sergei clearly wrote: "I can cite as much evidence as I like for the existence of statistically significant dependencies between price increments". Your arguments can hardly be recognized as evidence for this case, since it is not yet clear what "happens" directly to the increments (colleagues have not demonstrated anything). But if, suppose that there is no such dependence for the "source" (i.e. directly the object of research), it is rather strange to insist on its existence. Well, if we assume that volatility = RMS, the correlation of first lags of volatility for a number of increments with quite long windows (about 500 samples) is roughly in the region of 0.114-0.2 (which, to put it mildly, is a bit out of "presence").
to IlyaA
I did not, but I think not, it's just the nature of it. I, for example, have always been interested in (H+L)/2, because the error in predicting the "middle" is somewhat compensated by the range of "fluctuations" of price around this very middle.
And I like this kind of pressure, the main thing is that if Risk loses, he will learn how to beat the market in return,
but Neutron's expectation of winning in this bet is the same as losing.
And it all depends on the honesty of the disputers, I think Neutron will pay in case of a deal, but what about Risk?
So if I were Neutron, I wouldn't even respond to that.
Are you a guarantor?
Then we go to the bank, put $10,000 in a safe deposit box.
Write Neutron's claim, and if I refute it, I take the money.
to IlyaA
I understand that we are talking about the relationship of current increments to future increments. And I strongly suspect that the ACF of the first readings will be the proof, but a value like 0.4 (or so), to put it mildly, is questionable to demonstrate such an advantage. But maybe Sergei has something new in mind. And there were a lot of studies, so it is hard to remember at once.
PS: And there is one more subtlety, the formula for estimating ACF is not quite suitable for these series, besides it is necessary to estimate the reliability of ACF itself and error estimates for each lag found. It's not so simple there.
If we look at the correlation between adjacent increments, the correlation is very low indeed. If you take "groups" of increments, the picture starts to look very different. Don't ask me for a rigorous calculation, the conclusions are purely my own and purely empirical.
to Avals
Not that I'm getting into other people's squabbles :o), but Sergei clearly wrote: "I can cite as much evidence as I like for the existence of statistically significant dependencies between price increments". Your arguments can hardly be recognized as evidence for this case, since it is not yet clear what "happens" directly to the increments (colleagues have not demonstrated anything). But if, suppose that there is no such dependence for the "source" (i.e. directly the object of research), it is rather strange to insist on its existence. Well, if we assume that volatility = RMS, the correlation of first lags of volatility for a number of increments with rather long windows (about 500 samples) is roughly in the region of 0.114-0.2 (which, to put it mildly, is not much of a "presence").
The fact that volatility is autocorrelated is proven by Robert Engle who incidentally received a Nobel prize in economics the same year as Granger (2003). Mostly for the ARCH model, which is exactly based on autocorrelation of variance. Which is widely used in risk management. Briefly http://www.dengi-info.com/archive/article.php?aid=312