Forget random quotes - page 63
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
Where did 903.50 come from? Is that the file you're looking at? The first number is 325.25:
1 2 3 4 5 6 7
1992.11.01,00:00,355.75,364.00,352.75,359.25,1901
Otherwise, yes, hours and minutes are different columns, all columns are comma separated (standard export from MT4).
In the second file, column 16 should correspond to the net operators position: Net Operators.
One data for Tuesday, and the other for Sunday... 2012.07.15 corresponds to 2012.07.17 or does it correspond to 2012.07.22?
Some data for Tuesday and others for Sunday... Does 2012.07.15 correspond to 2012.07.17 or does it correspond to 2012.07.22?
The data are recorded for Tuesday, but are only published (made available to us) on Friday evening. Accordingly, in order not to fall into the trap of self-deception, it is necessary to make an offset by one week. I.e. it is even better to use the opening price of the following week:
1992.11.01,00:00,355.75,364.00,352.75,359.25,1901
Why 1992.11.01 is Sunday, I have no idea, it should be 1992.11.02. These are the only quotes I have. But not the point. If the opening price was on Sunday or as it should be on Monday, then for it we use the COT data for the previous Tuesday, i.e. for 1992.11.01 we should use the COT data for 1992.10.27
Yes, a couple of other comments.
It is customary to consider the price, and therefore the indicators, relative to time. But who said that this is the only correct methodology?
It is known that operators' positions depend on the price (the higher the price - the higher their net hedge). But what does time have to do with it? Therefore it would not be bad to draw a pointwise diagram of dependence between the price and positions of operators. It seems to me that it will be more or less similar to the linear, and it is possible to apply the Granger test.
The results are as follows.
Synchronised the dates. In Zw I had to insert 6 empty naludes to align. Then linearly interpolated the NA values obtained.
Got the following data.
Net
2000.01.04 -4479 2000.01.09 265.5
2000.01.11 -15963 2000.01.16 264.25
2000.01.18 -22316 2000.01.23 259.75
2000.01.25 -26656 2000.01.30 257
2000.02.01 -19041 2000.02.06 269.5
2000.02.08 -19564 2000.02.13 265
/
/
/ end
648 2012.05.29 11227 2012.06.03 629.25
649 2012.06.05 34075 2012.06.10 612.25
650 2012.06.12 36996 2012.06.17 673
651 2012.06.19 34250 2012.06.24 741.75
652 2012.06.26 -4088 2012.07.01 790.25
653 2012.07.03 -13515 2012.07.08 834
654 2012.07.10 -23508 2012.07.15 944.75
655 2012.07.17 -38701 2012.07.22 903
Causal test on the whole sample of 655 observations with a shift of 2 lags and 10 lags (the result is the same)
Pairwise Granger Causality Tests
Date: 08/03/12 Time: 15:21
Sample: 1 655
Lags: 10
Null Hypothesis: .............................................................................Obs F-Statistic Prob.
NET_OPERATORS no Granger reason for SER06_INTERPOLATE 645 2.66043 0.0035
SER06_INTERPOLATE is not a Granger reasonfor NET_OPERATORS 20.9059 1.E-33
.
the idea of testing causality on the whole sample is not correct.
Take the first 30 observations:
Pairwise Granger Causality Tests
Date: 08/03/12 Time: 15:44
Sample: 1 30
Lags: 2
Null Hypothesis: Obs F-Statistic Prob.
SER06_INTERPOLATE does not Granger Cause NET_OPERATORS 28 10.8494 0.0005
NET_OPERATORS does not Granger Cause SER06_INTERPOLATE 0.39443 0.6785
The picture is different. the second line sounds like this
We reject with 67% probability the hypothesis that NET_OPERATORS is not a Granger cause for SER06_INTERPOLATE.
Conversely, we cannot reject the hypothesis that NET_OPERATORS is not a cause!
Yes, a couple of other comments.
It is customary to consider the price, and therefore the indicators, relative to time. But who said that this is the only correct methodology?
It is known that operators' positions depend on the price (the higher the price - the higher their net hedge). But what does time have to do with it? Therefore it would not be bad to draw a pointwise diagram of dependence between the price and positions of operators. It seems to me that it will be more or less similar to the line chart, and it may apply the Granger test.
Observations are bound to time. What happens between observations we do not know. We need to interpolate, which can be either a very simple task (done above) or a very complicated one, as you suggest.
I think one has to look for the window size on which to determine causality and work with that, the window size found. I flatly deny the claim that the larger the sample, the better. This is asserted by people who have not gone further than the theorem. We are interested in the trend a few steps ahead. We are interested in market fluctuations on which to win or lose. And the average lack of causality between variables over 12 years says nothing. 30 weeks is six months. And the question should be put like this: is this 30 weeks enough to predict a couple of weeks ahead?
The results are as follows.
We reject with 67% probability the hypothesis that NET_OPERATORS is not a Granger cause for SER06_INTERPOLATE.
Conversely, we cannot reject the hypothesis that it is not a cause!
So is a yes-cause or a yes-no-cause?
! (Net_Operators != Ser06_Interpolar) = true;
I.e. Net Operators is most likely the cause of Ser06_Interpolate?
Oh I see, for the first 30 lags, yes, operators are the cause. For the whole sample, no, operators are not the cause. Not an important result at all. Still, you need a confirmation for the whole story, otherwise logically it is either the cause or not. And sometimes it is sometimes not - it is the same 50/50.
Are the increments or moments compared?
So is yes a yes, or yes a no?
! (Net_Operators != Ser06_Interpolar) = true;
I.e. is it more likely that Net Operators is the cause of Ser06_Interpolate?
That's the thing, not only is there a lot of shades of grey, so the world is also coloured on top of everything else!
It's just an eye-opener. And what to do is another question.
Oh I see, for the first 30 lags, yes, operators are the cause. For the whole sample, no, operators are not the cause. Not an important result at all. Still, you need a confirmation for the whole story, otherwise logically it is either the cause or not. And sometimes it is sometimes not - it is the same 50/50.
Are the increments or moments compared?
You get a very interesting picture if you run a window along the sample. It's good if the calculated causality value doesn't change, but most likely it does!
I first saw this on the coefficients of a weighted moving average - I was just horrified. You can't work with anything in TA. Everything I got from the tester cannot be trusted, etc.