Menu
Sister Sites

Options

Bookmark Thread

First Page First Unread Last Page Last Post

Print Thread

Similar Threads

Is Grid Trading (combined trend following and counter trend) Profitable? 3 replies

Making a good diversified portfolio 10 replies

Diversified EA Money Management Code 10 replies

HAMA PAD - A Simple Trading Approach 7 replies

Trading a (semi) automatic approach...but want more 11 replies

Attachments: Diversified Trend Trading Approach
Exit Attachments

Diversified Trend Trading Approach

News and advice from behemoth in trend trading space:

Attached Image (click to enlarge)
Click to Enlarge Name: rRrHuf6.png Size: 203 KB


https://www.bloomberg.com/news/artic...roller-coaster
Cut short your losses. Let your profits run on. David Ricardo (1772-1823)
Quoting Ramadas
News and advice from behemoth in trend trading space: {image} https://www.bloomberg.com/news/artic...roller-coaster
They are legends at the top of their game in the divergent space aren't they mate? Incredible risk-weighted returns and look how they are skyrocketing at the moment.

Attached Image (click to enlarge)
Click to Enlarge Name: Capture.JPG Size: 129 KB


Attached Image (click to enlarge)
Click to Enlarge Name: Capture4.JPG Size: 111 KB


https://www.transtrend.com/en/approach/philosophy/

Attached Image (click to enlarge)
Click to Enlarge Name: Capture3.JPG Size: 20 KB


Attached Image (click to enlarge)
Click to Enlarge Name: Capture2.JPG Size: 52 KB


What drives the cocoa market.
For the Backtest Geek in us...that inhabit this thread

Bert has come a long way. Magnificent piece of coding.

Inserted Video
With all the philosophical discussions you had with VEEFX I lost the thread about the instant karma strategy. I'd like to replicate the results you posted. I read that you added a DC3000 as a filter but it is unclear how you use it. I quickly built a simple KAMA crossover to see how it behaves. The results are terrible, as expected. Only A/U H1 gives a positive result that bootstrapping can't distinguish from random luck.
Attached Image (click to enlarge)
Click to Enlarge Name: equity.png Size: 62 KB

Attached Image (click to enlarge)
Click to Enlarge Name: bootstrap.png Size: 192 KB


I limited the backtest to Dec 2011 to Jun 2012 to visualize what happens. The period was selected because the first part is an up trend and the second part is a down trend. Dispite the trending periods and dispite the strategy is mainly long when the market is rising (green background) and vice-versa (red) it loses with a WR lower than 20%! It takes many not so small losses. The screenshot is taken on H4 to see the full range.
Attached Image (click to enlarge)
Click to Enlarge Name: AUDUSD-H4.png Size: 58 KB
No greed. No fear. Just maths.
Quoting Copernicus
For the Backtest Geek in us...that inhabit this thread Bert has come a long way. Magnificent piece of coding. https://www.youtube.com/watch?v=qcd_VBaR1P0
Interesting.
Did you try this and did your BT results actually match forward tests ?

That would be Spooky, wouldn't it ;-)

Inserted Video


LuXing
Wisdom begins in wonder
Quoting PipMeUp
With all the philosophical discussions you had with VEEFX I lost the thread about the instant karma strategy. I'd like to replicate the results you posted. I read that you added a DC3000 as a filter but it is unclear how you use it. I quickly built a simple KAMA crossover to see how it behaves. The results are terrible, as expected. Only A/U H1 gives a positive result that bootstrapping can't distinguish from random luck. {image} {image} I limited the backtest to Dec 2011 to Jun 2012 to visualize what happens. The period was selected because the...
Hi Pip

Just give us a bit of time mate as we are developing the EA at the moment for long range testing. It is too early for me to make any definitive statements at the moment but will keep you posted on our progress. Once we have honed in a data-set that we are happy with I will post it on this thread with our EA specs so you can also put it through the crunch test.

We applied a KAMA crossover with no Donchian condition over 24 instruments between 1 Aug 2017 to 30 April 2018 which was producing such terrible results in attempting to catch the trends....... that when we applied the trade flip idea of yours based on the prior trades result, the results were spectacular. We are going to look a bit closer at this but need to finalize the auto-correlator EA.

Cheers

C
Quoting LuXing
{quote} Interesting. Did you try this and did your BT results actually match forward tests ? That would be Spooky, wouldn't it ;-) https://www.youtube.com/watch?v=DgxS...tzxKSO&index=7 LuXing
Hi Lux

Just into it now as I picked it up this weekend. It is a dream for backtesting once you have the Dukascopy data downloaded. Prior versions of Tickstory were too clunky for my liking and required significant server space to house the data. This is a great total rewrite where the past problems and issues appear to be resolved.

It is a far less painstaking process. I am now flying through long term tests with variable spread using tick data and it is a far faster process than earlier versions.
Quoting Copernicus
{quote} Hi Lux Just into it now as I picked it up this weekend. It is a dream for backtesting once you have the Dukascopy data downloaded. Prior versions of Tickstory were too clunky for my liking and required significant server space to house the data. This is a great total rewrite where the past problems and issues appear to be resolved. It is a far less painstaking process. I am now flying through long term tests with variable spread using tick data and it is a far faster process than earlier versions.
Excellent. Thanks again for the heads up mate !
Will take a crack at it and see if it can do a year's USDJPY before this track's over :-)

Inserted Video
Wisdom begins in wonder
Quoting PipMeUp
With all the philosophical discussions you had with VEEFX I lost the thread about the instant karma strategy. I'd like to replicate the results you posted. I read that you added a DC3000 as a filter but it is unclear how you use it. I quickly built a simple KAMA crossover to see how it behaves. The results are terrible, as expected. Only A/U H1 gives a positive result that bootstrapping can't distinguish from random luck. {image} {image} I limited the backtest to Dec 2011 to Jun 2012 to visualize what happens. The period was selected because the...
Pip while we work on finalizing the AutoCorrelator EA I have attached trade details of the Baseline strategy and our method of flipping trades in Excel..... including details of the baseline strategy we are using that seems to possess strong serial correlation for the period 1 Aug 2017 to 30 Apr 2018.

We want to test this strategy under EA (not by spreadsheet flip) and over a long term test to see if it still stacks up. In the meantime this hopefully will give you sufficient info to mimic with your testing and answer some of your queries. It is still very simple mate...and we can apply some rules to improve results under steroid treatment such as extreme outlier stops etc. to improve results further.

Note the rules in the flip through the formulas applied.

If this supports your findings we are pretty happy mate as we have plenty of solutions for these devastating baseline strategies that could be deployed this way. I am pretty excited mate....but just don't want to get too cocky yet.

Cheers

C
Attached File(s)
File Type: xlsx Test Ref 7 Baseline for Pip.xlsx   1.2 MB | 277 downloads
Quoting PipMeUp
Despite the trending periods and despite the strategy is mainly long when the market is rising (green background) and vice-versa (red) it loses with a WR lower than 20%! It takes many not so small losses. The screenshot is taken on H4 to see the full range. {image}
That's what we are finding as well in the testing. Imagine if your solution is that easy...it boggles the mind :-)
Quoting Copernicus
{quote} That's what we are finding as well in the testing. Imagine if your solution is that easy...it boggles the mind :-)
Yep, we just need terrible strategy with a lot of consecutive losses or wins. But losses are easier to come by

Interesting article about cloak and dagger warfare going on in algo world. Is this where we are heading with algo development? To be automatically reverse engineered?

Attached Image (click to enlarge)
Click to Enlarge Name: 0kgAr2Y.png Size: 164 KB


https://www.babelfishanalytics.com/n...-in-four-notes

I wonder if they could discover my systematic manual trading
Cut short your losses. Let your profits run on. David Ricardo (1772-1823)
Quoting Copernicus
{quote} Pip while we work on finalizing the AutoCorrelator EA I have attached trade details of the Baseline strategy and our method of flipping trades in Excel..... including details of the baseline strategy we are using that seems to possess strong serial correlation for the period 1 Aug 2017 to 30 Apr 2018. We want to test this strategy under EA (not by spreadsheet flip) and over a long term test to see if it still stacks up. In the meantime this hopefully will give you sufficient info to mimic with your testing and answer some of your queries....
I see you use a higher TF KAMA cross to filter the trades. I also tried this and the results were even worse because the strat can no longer catch many of the big fishes: they get filtered out. This one is rejected!
Attached Image (click to enlarge)
Click to Enlarge Name: missed.png Size: 31 KB
No greed. No fear. Just maths.
Quoting Ramadas
{quote} Yep, we just need terrible strategy with a lot of consecutive losses or wins. But losses are easier to come by Interesting article about cloak and dagger warfare going on in algo world. Is this where we are heading with algo development? To be automatically reverse engineered? {image} https://www.babelfishanalytics.com/n...-in-four-notes I wonder if they could discover my systematic manual trading
LOL Nice R. This is where we do appear to be heading. Into a world of twisted human logic that unravels the biases. A world where 'if you can't beat em'...then beat em'.

I am having those moments when you slap your forehead saying....'perhaps it was hidden in plain sight all along'. It takes an algorithmic mindset to find this stuff and peel away the biases. You would think that looking for 'signals in the noise' would be an intuitive concept. Just goes to show how much noise actually biases the mindset.

Still too early to call at this stage....but looking good so far.
Quoting PipMeUp
{quote} I see you use a higher TF KAMA cross to filter the trades. I also tried this and the results were even worse because the strat can no longer catch many of the big fishes: they get filtered out. This one is rejected! {image}
Thanks Pip. I am still struggling with the interpretation part as it requires a flip in my 'divergent' thinking...but your input is really helping here. Once this gets bedded down then I may be able to enhance the biased result intuitively so we can step the juice up further. Sorry for my lagging mush :-)
Quoting PipMeUp
{quote} I see you use a higher TF KAMA cross to filter the trades. I also tried this and the results were even worse because the strat can no longer catch many of the big fishes: they get filtered out. This one is rejected! {image}
A negative outlier catcher such as an extreme $ soft stop on steroid is a worthwhile addition that doesn't disrupt the sequencing. That certainly pushes things further in a positive direction filtering out the few that get away.
KAMA cross has got three majors flaws.

The first is a human cognitive bias apparent in all the MA crossover I encountered. "A ⇒ B" is not the same as "A ⇔ B". Yet a lot of people confuse. If it is raining (A) the grass is wet (B): A ⇒ B. But the grass may be wet because it is so dry the sprinkler is on (A ⇎ B). You can only deduce that if the grass is not wet (¬B) then it is not raining (¬A): ¬B ⇒ ¬A

If the price is raising (A) the fast MA is above (B) is not the same as the fast MA is above therefore the price is raising. The fast MA can be above in a ranging market (sprinkler). You can only deduce that if the fast MA is above then the price is not falling. In the context of a divergent strategy, the fast MA above doesn't mean LONG it means NOT SHORT. MA below doesn't mean SHORT it means NOT LONG.

The second issue is inherent to the calculation of the KAMA. When the price is in a congestion KAMA flattens quickly and tends to settle in the middle of the range. So a pair of KAMA will always cross during a range because they overlaps in the middle. If the range is volatile enough it makes the faster MA oscillate and cross many times.

The last problem comes from the market itself: the trends are too short lived in general to overcome the two points above. When the price trends in stair steps the losses accumulates in the ranges and aren't covered during the directional moves.
No greed. No fear. Just maths.
Quoting PipMeUp
KAMA cross has got three majors flaws. The first is a human cognitive bias apparent in all the MA crossover I encountered. "A ⇒ B" is not the same as "A ⇔ B". Yet a lot of people confuse. If it is raining (A) the grass is wet (B): A ⇒ B. But the grass may be wet because it is so dry the sprinkler is on (A ⇎ B). You can only deduce that if the grass is not wet (¬B) then it is not raining (¬A): ¬B ⇒ ¬A If the price is raising (A) the fast MA is above (B) is not the same as the fast MA is above therefore the price is raising....
Thanks Pip

All those issues clearly are building a bias towards extreme failure which we can capitalise on. The approx 20% win rate is clearly supporting your valid points

I am tending not to interpret via whether it is trending or not but rather punching through using your auto-correlation referring to a benchmark...being the failed trending solution. Perhaps we should not be interpreting according to a common trading logic of trending versus non-trending but an autocorrelated one related to an inefficient benchmark. Clearly the inefficiency as a trending model (as you have stated) is what is generating this significantly auto-correlated result. I will see where this leads us.

There is no doubt that we can replicate these results in EA through benchmark referral and I am starting to pull some great long range results. They are very un-correlated to our other systems in the portfolio given their underlying basis of failure which could be a very handy string to the bow in terms of ironing out remaining volatility of the portfolio return streams.
3 Instruments tested 1/1/2010 to 30/4/2018......21 to go

Tested:

  1. ASX200
  2. AUDCAD
  3. AUDJPY


Composite Steroid results

Attached Image (click to enlarge)
Click to Enlarge Name: Capture.JPG Size: 71 KB


Once the testing is done I will put the results through the portfolio cruncher so we can see the individual contributions of each instrument.
Quoting Copernicus
{quote} Thanks Pip All those issues clearly are building a bias towards extreme failure which we can capitalise on. The approx 20% win rate is clearly supporting your valid points I am tending not to interpret via whether it is trending or not but rather punching through using your auto-correlation referring to a benchmark...being the failed trending solution. Perhaps we should not be interpreting according to a common trading logic of trending versus non-trending but an autocorrelated one related to an inefficient benchmark. Clearly the inefficiency...
Of course you realize that mostly reversing a divergent system makes it a convergent one! R and I will have to set a stake to burn you on public place. Heretic!
No greed. No fear. Just maths.
Resurrection of John Calvin
1 trader viewing now