logo
Welcome Guest! To enable all features please Login or Register.

Notification

Icon
Error

Options
Go to last post Go to first unread
guara_riua  
#1 Posted : Monday, November 21, 2005 4:01:50 AM(UTC)
guara_riua

Rank: Advanced Member

Groups: Registered, Registered Users
Joined: 4/24/2005(UTC)
Posts: 77
Location: Canada

First I want everybody know that me and Henry took this issue off line and we worked it out, I would say pretty well. I did a substantial more thorough tests and I looked into additional details about how this add-on is working. Overall, I will conclude that the 2 external functions tested (TDataCreate/TDataLocalize and TrendDirection) are performing in accordance to the ways the PP+ internal algorithms work and how the pivot types are defined and evolve (again as per PP+ internal algorithm). However, there is a limitation for this algorithm that is it works like a low pass filter for the prices that are moving above a determined speed. In this particular case I found that almost all the pivots are flagged as Minor. I have a couple of screen shots to post, but I have reached my limit upload quota for this forum. So I am going to delete my previous uploaded documents regarding this matter and considering them as no longer significant. Bellow it's a more detailed explanation of my findings. The reason the plot of the "Median Price" for a TF=4 or higher was so much distorted is mainly because of the Downloader. Main Reason 1) The Downloader is CHANGING THE TIME FORMAT when it is used to translate XLS files to MS data files. After the translation, in the "Time" column can be seen that the "seconds" field has 3 digits and the time for 3 successive rows may look like this: 1:44:00; 1:44:059; 1:45:060 . I have already flagged this as a bug in the Downloader and also opened a technical service request with Equis. By using a "Cut and Paste" approach to fix the "Time" column values from the CSV file, the erroneous plot for the Median Price in higher time frames was eliminated. Reason #2) The correct test is supposed to be done TWICE and in two different time frames, which means the plots for the chart with 1 min periodicity when using TF=4 must match EXACTLY the plots from the chart using the same data but with a 4 min periodicity and TF=1, that in fact shows a great consistency. I am pretty sure that everybody who used my "SineSumA1.CSV" file could see the TrendDirection "discrepancies" for a TF>=4 in a 1 min chart, that are showing in the first 13 days or through the whole length of the chart. Well, when I compared the plots for the TrendDirection for 1 min and 4 min charts, with TF=4 and respectively TF=1, I understood why the plots are showing the way they do: - in the 4 min periodicity chart the time segment that defines the higher frequency price movements (lower bars per cycle) shows a slow transition from ONLY Minor pivots to a mix of Minor and Intermediate pivots as the data price advances toward slower frequencies. The TrendDirection for TF=1 in the 4 min chart is identical to the TrendDirection plot from the 1 min chart with TF=4. - another reason why the TrendDirection looks "asymetrical" instead of a more regular output which we would expect, is that at higher frequencies there are less samples for the fastest Sine wave. This will implicitly generate bars that will have some variations in amplitude between the largest cycle periods (the 3rd sine wave that modulates the other 2 and has the lowest frequency) at the sine wave min and max values (+/-1). This is happening because the sampling resolution is very low and the sampled bars values are "missing" the exactly max/min values for that sine wave frequency, but are somewhat close to the peak/bottom. This is the reason why there is a slight variation in the data amplitude for the higher frequencies data segment, that directly generate the TrendDirection to react in a "non uniform" way; also this slight amplitude variations generate correspondent Intermediate pivots in the higher time frame that will be at higher or lower amplitudes from their previous "brothers" (at extreme values only). - however, seem that at higher frequencies for the stimulus data, there is a hindsight in the algorithm that identifies the pivot types, and almost all the pivots for these frequencies are flagged as Minor. Almost for sure there is a low pass filter type in front of the function that defines the pivot types or the algorithm itself acts like a low pass filter, because otherwise the time segment I am talking about will have to have also a mix of Minor and Intermediate pivots. So I have found that the tested functions are consistent to the way the PP+ defines the pivot types and how its external functions are defined to work in relation to the Pivot Types in higher time frames. Even if I am questioning the quality of the pivot type definition algorithm for data that has higher frequency variations, at least the function behavior is consistent with the way the PP+ defines pivot types in higher time frames. This finally concludes that for the vast majority of the real time data cases, the higher time frame functions (TDataCreate/TDataLocalize) and the TrendDirection are performing quite well. I attached the 2 screen shots for the "SineSuma1_Inv", that is the opposite of the "SineSumA1", meaning it ends with the higher frequency data variations. I have used this file in order to avoid the risk that higher frequency signals could "fall" into the tested functions "stabilizing" region. I hope this will clear the clouds my previous and less thorough tests may have laid over this add-on. Guara Riua
smg  
#2 Posted : Wednesday, November 30, 2005 3:19:31 PM(UTC)
smg

Rank: Advanced Member

Groups: Registered, Registered Users, Subscribers
Joined: 8/13/2005(UTC)
Posts: 90

Dear member, I understand that intraday data in downloader does not have a SECONDS portion, the three digit portion that we see is basically the TICK portion. Downloader sequentially numbers the ticks received during a minute from 001 to 999 and if there are more ticks than 999 during a minute, then the extra ones are all marked 999. I think I had read above somewhere in Downloader manual some years ago. Please check the Downloader help to get the latest about it. Let's hope we can find some solution to the small problem, if my memory is right. Regards SMG
guara_riua  
#3 Posted : Wednesday, November 30, 2005 4:29:32 PM(UTC)
guara_riua

Rank: Advanced Member

Groups: Registered, Registered Users
Joined: 4/24/2005(UTC)
Posts: 77
Location: Canada

SMG Thank you for the detail. However, even if this is true, the time format is generating errors for the tested external function. The proof is that using "Cut & Paste" from the original (not translated within the Downloader) CSV/XLS file the errors are going away. I am stil waiting for the Equis Tech. support reply. Guara
Users browsing this topic
Similar Topics
Re: PP+ Test Conclusions = Better Later than Never! (Data Related Questions)
by FXLearning 1/1/2006 6:11:15 PM(UTC)
Forum Jump  
You cannot post new topics in this forum.
You cannot reply to topics in this forum.
You cannot delete your posts in this forum.
You cannot edit your posts in this forum.
You cannot create polls in this forum.
You cannot vote in polls in this forum.