Machine Learning in ProOrder ProRealTime

• 04/18/2020 at 1:17 PM #126735
  GraHal

  Here’s the link to where I got excited again! 🙂

  GraHal wrote:

  Even better (2nd image) with the settings shown for Renko Type = 2 and Box size = 95.

  Bard wrote:

  Do you use Monte Carlo or Vonasi’s Robustness Tester first and then monitor systems on the demo before going live?

  I have done Monte Carlo and the VRT and yes I have loads on Demo, but I seem to have got back into my old (bad) habit of putting some Real Live straight off!  After decent backtesting performance stats of course!

  I enjoy getting straight on to an Auto-Trade as soon as it executes and sometimes will manual stop / exit with a loss or profit of £2 ish. Reason: I find trades often open at 61.8% Fib retrace and Resistance or Support levels etc etc  and I judge they would then go the wrong way and I’d be in loss for hours!

  The loss I got early yesterday was due to me on the deck chair in the sun, phone in pocket set to get a notification when a trade opened! Anyway my daughter and grandkids arrived and were shouting and squealing around everywhere, bless em (in the garden > 2 meters away! 🙂 )  I never heard the alert sound or felt the vibration in my pocket and when they’d gone I was in 3 losing trades … due to me not monitoring etc.  Lesson learnt!!

  

  GraHal

  Participant

  Topics: 203

  Replies: 8374

  Been thanked: 2024 times
  04/18/2020 at 11:48 PM #126825
  dnystrom

  Thanks for great code!
  I have one question. When i put this in my existing algo and with a starting value. When go live, should i change the starting value to the latest value in the backtest? And if so, how do i see what the latest value is?

  

  dnystrom

  Participant

  Topics: 1

  Replies: 18

  Been thanked: 6 times
  04/19/2020 at 12:30 AM #126828
  Paul

  it would make sense to do that I guess.  Add to the code graph valuex and run the backtest again.

  1 user thanked author for this post.

  dnystrom

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  04/19/2020 at 9:16 AM #126838
  dnystrom

  Have a little test going on Jans MA-algo and implemented this algo with an entry signal and looks promesing.
  wow and thanks for the code!

  

  dnystrom

  Participant

  Topics: 1

  Replies: 18

  Been thanked: 6 times
  04/19/2020 at 1:45 PM #126887
  Bard

  Hi GraHal, thanks but I don’t see the link or any 2nd image?

  I remember reading about someone who algo traded, it was in Jack Schwager’s Market Wizards, it could have been Richard Dennis (Turtles fame) and they said they would over-ride the algos’ decisions sometimes but when they did a study it turned out the algos’ performance beat their over-ride decisions. Did you ever study that, I mean let the algo run entirely on it’s own (unless extreme market conditions presented themselves) and compare it with the performance when you manually over-ride?

  Why is being in a loss for hours a problem, are you a short term scalper? I remember the “baby sitting” of large discretionary currency positions that went against me but hadn’t hit my wide 10% of capital stop. It’s that stop method which is why I think I was as so profitable back then, because I stuck to that rule like glue). I’d monitor the position every few hours overnight and by day three I’d be so tired and didn’t care if the position made money or was a loss! So I’d kick it in to touch, typically only to watch the position hit breakeven and then go into profit within a day…

  Grandkids? You don’t look old enough for grandkids! Do you have some special diet/exercise regime? 😃

  

  Bard

  Participant

  Topics: 131

  Replies: 809

  Been thanked: 99 times
  04/19/2020 at 2:44 PM #126896
  GraHal

  Bard wrote:

  I don’t see the link or any 2nd image?

  You needed to click on the words ‘GraHal wrote’ then that links you straight back to the post where the 2nd image is.

  Bard wrote:

  Did you ever study that,

  Not a definitive study no, but I always look back a few mins later and most times think … wow I’m glad I got out of that one!

  Bard wrote:

  Why is being in a loss for hours a problem, are you a short term scalper?

  Yeah I guess I must be … more suited to my personality!  Also I like to go to sleep at night with my money back in my pot! Doesn’t always happen and I’ve had a few nights, as you describe, with phone at side of my bed set to alert me if price drops to x levels etc.   I’d rather make less profit and be stress free / sleep well??

  If a trade is in loss for hours then in my opinion the entry was bad, like not a logical entry point?  If in loss for a few minutes even … I feel like I am living in hope and kidding myself that price is going to turn around?

  I guess I must think different than most? A trade entry can’t be correct if it then goes to a loss??  Do we code ‘loss journeys’ into our strategies?  Even a Long entry at a 50% fib retrace and then be happy for it to go to 61.8% retrace and then go back up again?  Surely not even that? We would code for the turning back up after the retrace has finished and even safer after a failed 2nd attempt to go lower … then we code for an entry point as price turns back up again ??

  We should have put these ‘chats’ on the Trading Forum as others may benefit or add more points of view.

  Maybe I’ll ask the Mods to move a few posts?

  Bard wrote:

  Do you have some special diet/exercise regime

  Thank you! I keep myself at same weight as in my early 20’s and eat healthy … not vegan or veggie, but not loads of meat either!  I eat very few ’empty calories’ (junk) but I do have treats  … like a chocolate eclair / profliteroles! 🙂 I take lots of nutrition supplements … if I read this or that is good, I get on ebay and order some … after researching as to safety etc.

  Off for daily walk now … no beauty spots, no shaking hands, no snogging!! 🙂

  

  GraHal

  Participant

  Topics: 203

  Replies: 8374

  Been thanked: 2024 times
  04/19/2020 at 6:33 PM #126929
  Bard

  Got it, thanks. I don’t see any Renko “change types” in my settings: pls see image.

  I’d rather make less profit and be stress free / sleep well??

  Amen to that!

  I don’t think anyone will ever be successful having a trading style or system that doesn’t fit their personality type. I’m like you. I want to get in and get out, normally on the second wave. I don’t want to marry the position. I think a trade entry can be correct even if it initially goes against you. I think the markets are subject to many irrational (sudden fear based moves, or a hedge fund unloading a losing position!) that doesn’t and shouldn’t invalidate the signal. Particularly in markets that are “noisy with news.” I also think the entry is the least important part of a trading style. It took me a long time to realise that. But of course bad timing means sitting and waiting with a unrealised loss. Jimmy Rogers has admitted his entry timing is poor, in the example he gave about a silver trade, I worked out that he must have had to sit on millions in losses in silver for months before his bull theory worked. This was before the 2008 crash. It’s really about having giving the market time to move within a context of knowing where the market has proved you wrong. By all means if you wish to move this – although it may be difficult due to the mixture of theory/ideas and then actual comments about ML within the same comment.

  Ditto, I’m the same weight, eat very healthily, vegan is not healthy is my mind but then neither is excess meat! Ditto supplements.

  1 user thanked author for this post.

  GraHal

  

  Bard

  Participant

  Topics: 131

  Replies: 809

  Been thanked: 99 times
  04/19/2020 at 9:05 PM #126950
  Bard

  Here is the Code for Renko ML1 — 1 x Machine Learning algo ValueX set to determine the Boxsize.
  It beat the original system, just: https://www.prorealcode.com/topic/discussion-re-pure-renko-strategy/
  2% less drawdown than the non ML version, but more trades taken. 

  If you try changing:

  Position Size

  N = 5*Capital / Close

  1	N = 5*Capital / Close

  to the value 10*Capital… The drawdown still stays low at 11%, whilst profits explode going from £85k to £609k, although clearly, ML settings were changed and “optimised” during the 5* Capital testing phase.

  I think next test should be to apply the machine learning ValueX to the boxsize and ValueY to the stop loss perhaps? (i.e. 2 x machine learning algos). I’m also wondering if it’ll perform better without the ResetPeriod?

  Feel free to take the itf code and see if it can be improved. Note, none of these tests were done with tick by tick mode, which surely makes these results academic, even with the original Renko system?

  
  @Juanj @Noneththeless @DocTrading @Gabri @Paul @GraHal

   

  1 user thanked author for this post.

  eckaw

  

  Bard

  Participant

  Topics: 131

  Replies: 809

  Been thanked: 99 times
  04/19/2020 at 10:19 PM #126960
  dnystrom

  My first try on the Renko system with ML, TF 30s

  ValueX for renkosystem

  ValueY for Boxsize, value below. But it don’t make sense with starting value 90, minvalue 85 and decrease steps with 10? But still get the best results. The Renkoversion is 2.3pR and in 30s TF.

  ValueY

  StartingValue2 = 90 graph valuey ResetPeriod2 = 12 //Specify no of months after which to reset optimization Increment2 = 85 MaxIncrement2 = 10//Limit of no of increments either up or down Reps2 = 1 //Number of trades to use for analysis MaxValue2 = 100 //Maximum allowed value MinValue2 = increment2 //Minimum allowed value

  1 2 3 4 5 6 7 8

  StartingValue2 = 90 graph valuey ResetPeriod2 = 12 //Specify no of months after which to reset optimization Increment2 = 85 MaxIncrement2 = 10//Limit of no of increments either up or down Reps2 = 1 //Number of trades to use for analysis MaxValue2 = 100 //Maximum allowed value MinValue2 = increment2 //Minimum allowed value

   

  

  dnystrom

  Participant

  Topics: 1

  Replies: 18

  Been thanked: 6 times
  04/19/2020 at 10:28 PM #126963
  eckaw

  I’m doing forward testing on my verison of renko ML from tomorrow morning too. Backtest looks promising!

   

  

  eckaw

  Participant

  Topics: 13

  Replies: 118

  Been thanked: 61 times
  04/19/2020 at 11:39 PM #126968
  Bard

  A large improvement to the performance issues using “tick by tick” with Nneless-Renko ML1 itf:
  
  Let your profits run: If you use tick by tick backtesting, set ValueX set to stop loss (originally it was set to Boxsize), then change the Target Profit from 150 to 500 and set Boxsize to 100:

  £/$ Daily: Money Mgt, 30 x Capital/Close | Aug 2015 – Ap 2020 | 305 Trades | £53k 434% | Win 42% | Gainloss 1.39 | Ddown 13.8%

  ML Settings:
  StartingValue = 10 //10 Stop Loss
  ResetPeriod = 0.5 //0.5 Specify no of months after which to reset optimisation
  Increment = 5 //20
  MaxIncrement = 10 //10 Limit of no of increments either up or down
  Reps = 2 //Number of trades to use for analysis //2
  MaxValue = 150 //150 //Maximum allowed value
  MinValue = increment //Minimum allowed value

   

   

  

  Bard

  Participant

  Topics: 131

  Replies: 809

  Been thanked: 99 times
  04/19/2020 at 11:48 PM #126971
  Bard

  Hi dnystrom, What is “RenkoSystem” with ValueX’s? What is being optimised?

  Re: Steps. From earlier in this long thread:

  Grahal: So if I set increment = 2 then in any one bar, ValueX or Value Y should not change by more than ‘2’ (up or down)?

  Junaj: Correct, unless the increment limit value has been reached, after which ValueX will revert to the new Average best performing value.
  

  https://www.prorealcode.com/topic/machine-learning-in-proorder/page/5/#post-121432

   

  

  Bard

  Participant

  Topics: 131

  Replies: 809

  Been thanked: 99 times
  04/19/2020 at 11:56 PM #126972
  Bard

  When everyone was having a s&*% time trading the Dow but you were trading the Renko ML1, Lol. (tick by tick backtest and exact same settings as £/$ above https://www.prorealcode.com/topic/machine-learning-in-proorder/page/11/#post-126968)

  

  Bard

  Participant

  Topics: 131

  Replies: 809

  Been thanked: 99 times
  04/20/2020 at 12:11 AM #126975
  Bard

  Ditto Brent Crude:

   

  

  Bard

  Participant

  Topics: 131

  Replies: 809

  Been thanked: 99 times
  04/20/2020 at 12:21 AM #126978
  Bard

  £55 million on coffee, that’s got to be a mistake surely? Settings were again, exactly the same as above.

  The maximum Qty in Detailed Report is 142. If you said £10/tick was one Contract or Lot (Not sure as I haven’t researched/traded coffee and I haven’t looked up lot sizes or volumes in the coffee market but basing it on what the £/$ used to be), then 14 Lots sounds small and not so much to trade that you’d ever tip off the market/brokers and get front run, but still… £55m?

  

  Bard

  Participant

  Topics: 131

  Replies: 809

  Been thanked: 99 times
• Author
  Posts