Pathfinder swing TS

Viewing 15 posts - 1,651 through 1,665 (of 2,005 total)

← 1 2 3 … 110 111 112 … 132 133 134 →

Author

Posts
10/16/2017 at 8:38 AM #49492 quote

dajvop
Participant

Master

@Patrick

Yes, you are right of course and I know why the lines are there. I just pointed out what makes the differences between the codes.

But, if you want to optimise only one period, then that is the difference between the different results.

Have you found out a smart and easy way how to optimise the whole year?

Best regards, David

10/16/2017 at 9:16 AM #49500 quote

wp01
Participant

Master

@David,

You are probably a couple of steps ahead on me. It took me longer to understand the issue.

If you what to optimize the whole year in one go i do not think that is recommendable.

You can of course take the roadmap and fill in the green positions with 1, 2 and 3 like the map suggested. On the red ones you

can fill in -1 or zero. The default settings you can optimize than. Than you are probably finished in half an hour per instrument.

But that is not the idea behind it i think. What i started with the Nasdaq Okt2 (which i posted last week) is that that one is prepaired

for optimization per period. I left every period untoched but set the seasonalPatternMultiplier to zero for the periods i do not use.

I also added at every period “tradeInPeriod = 0” to prevent you get results other than the period you optimize. And i also added accumulationModeLong

and Short to every period and set them to zero for the periods i do not use. If theze accumulationline are not in the period you can only use them in the

overall line above and that is not recommended because you do not use them in all periods. And when used above it is just on or off.

After optimization the period you want you move to the next period. Leave everything untouched with one difference: you set SeasonalPatternMultiplier to 0//1

(or 2 or 3 depending the positionsize you have used) and you set TradeInPeriod in the old period to zero and you move to next period and start over again.

After a year you have a full year optimized. I don’t think there is a quicker way because the whole idea is to get it optimized per period with different optimizationparameters.

If you throw that overboard you just have an optimization with average parameters.

One point to keep in mind. When you have a huge drawdown per period and you have for example 24 times €2K

drawdown, it adds to an amount per year that makes the code for the whole year unusable. My idea is to keep the DD per period as low as possible and when the DD is high,

try the opposite. (short instead of long or the other way around) or set the period to zero. Just my thought of course.

kind regards,

Patrick

dajvop thanked this post

10/16/2017 at 10:34 AM #49516 quote

dajvop
Participant

Master

Hmm, what I thought…

Then we have to do everything all over again but with v3 up to October and then we can add it all up 🙂

wp01 thanked this post

10/16/2017 at 5:04 PM #49600 quote

dajvop
Participant

Master

I need some help here.

I adjusted v3 to make it easier to optimise each period (less scrolling) and moved the lines higher up, more like v2.

Now I have a good enough result and copied the numbers to the original v3 to check and I get a slightly different result.

Why?

Both are attached, instrument is Platinum 10 JAN-18:

PF-00-TEST-v3.itf PF-00-TEST-v3_.itf

10/16/2017 at 7:37 PM #49618 quote

wp01
Participant

Master

@David,

One of them has an initial capital of $15K. The other $10K.

Line 63 in TEST V3_ says: IF CurrentMonth = 12 THEN; but you only optimize just 1 month. (I understand that the figures for January 1 are for december 1)

It is at least related because when i change this you get different results (and different DD).

I think you removed too much. Also the lines ONCE periodThirdMAJanuary1 = 2 and so on for all periods have some kind of function even if not used.

If you set everything to zero besides December 1 to 5, you get no results.

With these two changes the difference is more or less $ 300 in the backtest.

The scrolling is indeed an issue. I noticed it to. A suggestion might be to leave the original code unharmed and move the period you want to optimize somewhere above January 1 and than do

you thing. The next period the same. You move the new period above the last optimized period. In the end only the periods are the other way around. But that is easier to change and if you do not want

to also fine. Because the should work the same also when they are the other way around.

Kind regards,

Patrick

dajvop thanked this post

10/17/2017 at 9:40 PM #49744 quote

dajvop
Participant

Master

@Patrick

Have you had any issue with MaxCandles and it doesn’t seem to matter which number you set, you still get the same result?

10/18/2017 at 5:22 AM #49753 quote

wp01
Participant

Master

@David,

You mean with optimizing i guess. Yes that even happens frequently. These variables i do as last and it is also just around 10/12 nummers.

It gives often the same result and it does not seem to matter which one you take.

I’m also not stretching them up anymore as i did in the past. If you use for example 5-25 the numbers from 15-25 give in most cases better results

but these was also a bit of the problem that causes the losses last months. If the short positions were closed earlier it would have saved money and

that could have been achieved with less Maxcandles. Probably overoptimized. If you start by the way with Maxcandles

and you work upwards to the other variables you also get different results. The firstMa and secondMa (5 and 10) i also leave as it is. In the past we also

change them from time to time to get better results but in V3 Reiner wrote “center of gravity” next to it. He probably thought about that and keep them fixed at

5 and 10. The more you change it gets more and more optimized and curved fitted.

Best regards,

10/18/2017 at 6:58 AM #49763 quote

dajvop
Participant

Master

I have experienced several times now when optimising v3 that when I get to MaxCandles it doesn’t matter at all if I put 1 or 30, I get the same results anyway.

I also start with ThirdMA and MA, then StopLoss and TakeProfit and finish with MaxCandles.

One thing I thought about last night was that we can take a full year v2 just like v3, but use 3 different versions. One starting Jan1, the next Jan2 and the last Feb1 and then repeat every third periods. Then we will avoid the problem with the MaxCandles changing every period.

10/18/2017 at 8:20 AM #49768 quote

wp01
Participant

Master

Better backtest with higher results doesn’t mean automaticly higher profits.

That is why the defaultsettings is added and why i think we should put more effort in it. In a case that it doesn’t matter the amount of MaxCandles it should

be the default setting. And that also applies for the other variables. I’m only not sure how to tackle the optimization of the default settings. In a case

where the differences are more or less the same you should fall back on the default setting. The less you alter the less of risk of curve fitting.

10/18/2017 at 8:30 AM #49771 quote

dajvop
Participant

Master

I agree with you Patrick, but I think something is broken in v3.

Try to get any other result in this algo by changing the MaxCandles:

PF-FY-PL-v3.itf

10/18/2017 at 9:10 AM #49782 quote

wp01
Participant

Master

Yes it looks like it. In V3 i hardly get any results when i change others than the MaxCandle.

Differences in MaxCandle also did not make a change. I first thought it is the instrument so i did the V2 Dec2 on

the same instrument to test it.

Attached the results.

Pathfinder-PL-Dec2-V2-TEST.itf report-PL-Dec2-V2-TEST.png

10/18/2017 at 9:13 AM #49785 quote

dajvop
Participant

Master

Yes, I have been trying to recreate the first swing algo that Reiner made, but with v3 and it is not possible.

10/18/2017 at 9:17 AM #49787 quote

wp01
Participant

Master

I’ve created two in V3 and they look fine to me. I didn’t do them in V2 so i can not compare them.

I will post them so you can see for yourself.

10/18/2017 at 9:19 AM #49788 quote

wp01
Participant

Master

Dow and NQ Nov1 in V3 (NQ with accumulation):

dajvop thanked this post

Pathfinder-Dow-Nov1-V3.itf Pathfinder-NQ-Nov1-V3.itf report-Dow-Nov1-V3.png report-NQ-Nov1-wa-V3.png

10/18/2017 at 10:10 AM #49803 quote

O-jay8
Participant

Veteran

Hi Dajvop,

The idea with the V2 version and only doing one full year with the first period in a month (Jan1, Feb1, Mar1, …) and another with the second period in the month (Jan2, Feb2, Mar2 …) occured to me as well couple month ago and I posted it somewhere here as well far back on page 40 probably.

I think the originally Full year got posted by Reiner or Michi with Silver.
There it was close with the individual BT for one period but not identical.

Best regards
Author

Posts