Zero sample correlation does not necessarily mean there is no linear relationship - page 14
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
So, your main thesis is that there must be a strong correlation (i.e. correlation coefficient close to +1/-1) between any traded assets; not finding such a relationship has disappointed you in the statistical/mathematical methods traditionally used and made you look for new approaches.....
You are attributing your selves to me.
In this thread, it has already been repeatedly pointed out by many forum participants to the need to first have a logical relationship between the two series, and only after that, try to apply mathematical methods of analysis, but not vice versa!
What do you mean by logical relationships?
So, I ask you to explain how, on a fundamental or technical level, the existence of a mandatory pronounced linear interdependence between any assets should exist or should be justified (you can use the example of the assets cited above from your post...).
A linear relationship between two BPs is present when there is such a sum of them (the sum of the absolute values of the weights is equal to one) that its variance is less than the variance of the original BPs. Correlation is not always able to track this. An example is given at the beginning of the branch. Even so, correlation is a bearable (simple) tool for estimating a linear relationship.
If you generate price BPs, there will always be a linear relationship between them. Which will "walk" as the sliding sample window moves. An analysis of these "walks" will give you an estimate of the linear relationship of your BPs.
I'll get into it. The yen is known to depend on the price of oil. oil up - yen down. the yen and the dollar are kind of in the same boat... :)
I.e. FA level linkages are easy to justify.
;)
Agreed, besides: energy futures to each other, S&P500 and DJ, national currencies and national stock indices given the balance of trade, oil futures and oil stocks, commodity futures and commodity stocks, etc. etc....... The list has a logical connection and if we suddenly find "multi-trends" in the movement of such assets, it is a reason to take a trading reflection (famous example with S&P500 and DJ)....
The topikstarter brings up the need for total linear correlation of all assets with each other.
"Wow, there is almost no linear correlation between the Norwegian krone and the Signpura dollar - nonsense!", "Even funnier, there is almost no linear correlation between gold and the Canadian dollar - dick!" ............... If he is able to explain (or maybe you can explain) on the examples given, a dependence that derives from some macroeconomic constituents or something else - I would be grateful.....
.......... ...
If we are discussing correlation without a substantive basis for the eligibility of the connection as well as without giving a numerical estimate of the confidence in the result obtained - all this is profanity, which does not allow for practically meaningful results.
to hrenfx:
try to understand what people write to you!
try to get a feel for what people are writing to you!
The answer was:
What confidence assessment can we talk about when the calculations before analysing the sliding window results are incorrect...
You analyse the dynamics of QC on a sliding window to give a confidence estimate of QC.
The answer was:
Analyse the dynamics of QC on a sliding window to give confidence in the QC characteristics.
Look, do you really have no idea what you are talking about, or are you just messing around?
Look, do you really have no idea what you're talking about, or are you just messing around?
Shit, I'll spell it out for you:
Took the M5 timeframe. We took a window size of 288 bars (24 hours). Calculated QC (EURUSD and GBPUSD) for, say, 75000 bars (a year). Obtained BP from the CC with 75000 values.
So, the MO of this BP is an estimate of the EURUSD and GBPUSD relationship, and the variance of this BP is a confidence estimate of the EURUSD and GBPUSD relationship over a year with a 24-hour window.
Is this a trend you have detected from a historical start?
My understanding is that any application of statistics involves detrending BP, as the presence of a trend "clogs" the BP statistics.
In my particular case the trend is a linear regression on all the data in the table (over 2000 hourly candles), the formula is shown in the chart.
Detrending is not a simple matter for me, both in terms of the trend formula and the time horizon. I pick both by gauge, trying to get the noise component closest to normal. This is the criterion. But this criterion is taken from the ARPSS model, in which BP = trend + cycle + noise. The limitations are known, so I don't insist.
But Prival rightly points out the problems with computational accuracy at all stages of the calculations.
The problem of accuracy in statistics is well known and I have nothing to add.
You have an uncanny ability to get off topic. And it's not the first time.
The point of my posts in this thread: you can't solve the wrong problems with the right methods. And the accuracy of calculations cannot solve this very problem. The problem of accuracy is far from the first place. A lot of other problems must be solved before it, and then maybe the problem of accuracy will arise. I, for example, have never encountered this problem, as increasing the accuracy is not capable of changing the confidence interval for some BP statistic. I realise that my experience is not proof. In Prival's calculations, this is probably important, but he never indicated that changing the precision changes the confidence interval.
My understanding is that any application of statistics involves detrending BP, as the presence of a trend "clogs" the BP statistics.
In my particular case the trend is a linear regression on all data in the table (over 2000 hourly candles), the formula is shown in the graph.
Detrending is not a simple matter for me, both in terms of the trend formula and the time horizon. I pick both by gauge, trying to get the noise component closest to normal. This is the criterion. But this criterion is taken from the ARPSS model, in which BP = trend + cycle + noise. The limitations are known, so I don't insist.
I personally believe that one can look for trends on forex with the same success as on HGC implementations.
There shouldn't be any over the whole historical span. If the exit/country is anything to go by.
Well, unless you count as a trend movement within knees 33...
Another case is funds or commodities. There the removal of seasonality and trends is appropriate.
;)
So, the MO of this BP is an estimate of the EURUSD and GBPUSD relationship, and the variance of this BP is a confidence estimate of the EURUSD and GBPUSD relationship over a year with a 24-hour window.