Ftse 100 trend following & candlestick pattern 1hr strategy.

• 07/10/2017 at 3:31 PM #40177
  Francesco78

  Hello everyone.

  I have found this strategy on the FTSE 100 by playing with candlestick pattern.

  In particular here I assumed that 2 doji bull/bear candle are followed by an hammerdown/hammerup candle.

  I then added some momentum filter condition based on 3 simple moving average.

  The optimization involve:

  • the parameter x that define how much doji are the candle
  • the profit target
  • the trailing stop

  Moving averages are not optimized, and long and short trailing stops are set to be the same to avoid overfitting.

  Hope you like it.

  Francesco

  defparam cumulateorders = false ///money management reinv = 0 leverage= 3 IF REINV = 1 THEN // réinvestissement capital = 10000 + strategyprofit n = (capital / 10000) *leverage else n = 1 ENDIF ///market conditions av1 = average[5](close) av2 = average[10](close) av3 = average[50](close) bull = close> av1 and close >av2 and close > av3 bear = close <av1 and close <av2 and close <av3 ///optimized variables x = 0.6 PR = 50 timestart = 070000 timeend = 230000 TG = 40 ///candle pattern defirnitions timeok = time>timestart and time < timeend dojibull2 = open[2]<close[2] and abs(open[2]-close[2])/(high[2]-low[2])<x dojibull1 = open[1]<close[1] and abs(open[1]-close[1])/(high[1]-low[1])<x dojibear2 = open[2]>close[2] and abs(open[2]-close[2])/(high[2]-low[2])<x dojibear1 = open[1]>close[1] and abs(open[1]-close[1])/(high[1]-low[1])<x hammerup = min(open,close)>high-(high-low)/2 //and max(open,close) < min(open[1],close[1]) hammerdown = max(open,close)<low+(high-low)/2 //and max(open,close)<min(open[1],close[1])//and timeok //trade conditions cl = (dojibull2 and dojibull1) and hammerdown and bull cs= (dojibear2 and dojibear1) and hammerup and bear if cs and timeok then sellshort n contract at market endif if cl and timeok then buy n contract at market endif ///ttailing stops if not onmarket then MAXPRICE = 0 MINPRICE = close PREZZOUSCITA = 0 ENDIF if longonmarket then MAXPRICE = MAX(MAXPRICE,close) if MAXPRICE-tradeprice(1)>=TG*pointsize then PREZZOUSCITA = MAXPRICE-TG*pointsize ENDIF ENDIF if shortonmarket then MINPRICE = MIN(MINPRICE,close) if tradeprice(1)-MINPRICE>=TG*pointsize then PREZZOUSCITA = MINPRICE+TG*pointsize ENDIF ENDIF if onmarket and PREZZOUSCITA>0 then EXITSHORT AT PREZZOUSCITA STOP SELL AT PREZZOUSCITA STOP ENDIF //fixed profit target set target pprofit PR

  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78

  defparam cumulateorders = false ///money management reinv = 0 leverage= 3   IF REINV = 1 THEN  // réinvestissement capital = 10000 + strategyprofit n = (capital / 10000) *leverage else n = 1 ENDIF   ///market conditions   av1 = average[5](close) av2 = average[10](close) av3 = average[50](close) bull = close> av1 and close >av2 and close > av3 bear = close <av1 and close <av2 and close <av3   ///optimized variables x = 0.6 PR = 50 timestart = 070000 timeend = 230000 TG = 40     ///candle pattern defirnitions   timeok = time>timestart and time < timeend dojibull2 = open[2]<close[2] and abs(open[2]-close[2])/(high[2]-low[2])<x dojibull1 = open[1]<close[1] and abs(open[1]-close[1])/(high[1]-low[1])<x dojibear2 = open[2]>close[2] and abs(open[2]-close[2])/(high[2]-low[2])<x dojibear1 = open[1]>close[1] and abs(open[1]-close[1])/(high[1]-low[1])<x hammerup = min(open,close)>high-(high-low)/2 //and max(open,close) < min(open[1],close[1]) hammerdown = max(open,close)<low+(high-low)/2 //and max(open,close)<min(open[1],close[1])//and timeok   //trade conditions cl = (dojibull2 and dojibull1) and hammerdown and bull cs=  (dojibear2 and dojibear1) and hammerup and bear         if cs  and timeok then sellshort n contract at market endif   if cl  and timeok then buy n contract at market endif   ///ttailing stops if not onmarket then MAXPRICE = 0 MINPRICE = close PREZZOUSCITA = 0 ENDIF if longonmarket then MAXPRICE = MAX(MAXPRICE,close) if MAXPRICE-tradeprice(1)>=TG*pointsize then PREZZOUSCITA = MAXPRICE-TG*pointsize ENDIF ENDIF if shortonmarket then MINPRICE = MIN(MINPRICE,close) if tradeprice(1)-MINPRICE>=TG*pointsize then PREZZOUSCITA = MINPRICE+TG*pointsize ENDIF ENDIF if onmarket and PREZZOUSCITA>0 then EXITSHORT AT PREZZOUSCITA STOP SELL AT PREZZOUSCITA STOP ENDIF   //fixed profit target set target pprofit PR

   

  2 users thanked author for this post.

  Despair, jebus89

  

  Francesco78

  Participant

  Topics: 43

  Replies: 242

  Been thanked: 77 times
  07/10/2017 at 3:41 PM #40268
  Nicolas

  Hi Francesco, another great addition, thank you.

  I just made a WF analysis on a 200k bars basis and this is what I found (picture attached).

  Even if the overall result is so-so, there are things that annoy me, please consider them as advises and don’t be offend 🙂

  1. I think I should use more than 5 WFA on this longer period, at least 10 (double size of your own test, made on 100k bars).
  2. the first IN SAMPLE  result is not good
  3. some of the OOS results are giving huge Walk Forward Ratio (+500% !)
  4. I used 10-100 for PR and TG with a 10 points step and all the best OOS optimized variables are set to 10 (?!)

  Beside this, the strategy has performed well before the start of your own 100k bar backtest, that’s why I’m a bit perplexed of the whole analysis..

  What were the settings of your optimisation for PR and PG please?

   

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20867

  Been thanked: 4783 times
  07/10/2017 at 4:10 PM #40270
  Francesco78

  Thank you Nicolas, no offense at all! I am actually happy to see that you spent some time to analyse the strategy in more details.

  My settings for the WF were the same as yours, I also considered x as a variables ranging from 0.4 to 0.6 with steps 0.1.

  Im running it again in order to give you the details of the optimal value for the variables in the different IN SAMPLE and will let you know

   

  

  Francesco78

  Participant

  Topics: 43

  Replies: 242

  Been thanked: 77 times
  07/10/2017 at 4:21 PM #40272
  Francesco78

  Not anchored 75/25

  x  / TG/ PR

  • 0.4 /100/100
  • 0.4 /80/ 90
  • 0.4 /70 /70
  • 0.4 /50/ 60
  • 0.4/ 60 /70

  

  Francesco78

  Participant

  Topics: 43

  Replies: 242

  Been thanked: 77 times
  07/11/2017 at 11:24 AM #40339
  Francesco78

  HI Nicolas, do you get the same results as I get on the 100k bar?

  Thanks

  Francesco

  

  Francesco78

  Participant

  Topics: 43

  Replies: 242

  Been thanked: 77 times
  07/13/2017 at 10:57 AM #40536
  Nicolas

  Did not test on 100k bar but as far as I could do in tick mode with another try here with 9 OOS periods.

  1 user thanked author for this post.

  Francesco78

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20867

  Been thanked: 4783 times
  07/13/2017 at 11:41 AM #40544
  Francesco78

  Thank you Nicolas, just for my info if you optimize without WF on the 200k bars how does it look?

  Many thanks

  

  Francesco78

  Participant

  Topics: 43

  Replies: 242

  Been thanked: 77 times
  10/10/2017 at 12:29 AM #48838
  ozz87

  I started this one one demo but got the “zero division” problem or what you call it. Is there any fix for it?

  

  ozz87

  Participant

  Topics: 0

  Replies: 168

  Been thanked: 127 times
  10/15/2017 at 7:29 AM #49387
  ozz87

  Not sure the strategy will work because of above error, but I optimized it for OMX if anyone is interested.

  5 users thanked author for this post.

  Francesco78, jebus89, Nicolas, AsgerK, Berra

  

  ozz87

  Participant

  Topics: 0

  Replies: 168

  Been thanked: 127 times
  10/16/2017 at 8:33 AM #49489
  Nicolas

  @ozz87

  I think you can get rid of the zero division error by ensuring there are enough bars loaded to complete the candlesticks patterns calculation:

  defparam preloadbars=1000

  1	defparam preloadbars=1000

   

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20867

  Been thanked: 4783 times
  10/16/2017 at 8:36 AM #49491
  jebus89

  Another great idea Francesco! And many thanks to you Nicolas. Will keep working on this with WF testing and such. The shorting looks insane in ftse!

  Interesting results nicolas.

   

  Also im interested in hearing about your thoughts about this: If a strategy is optimized for 1 market, lets say FTSE. Then you copy/paste that same strat on lets say DAX. If it shows positive results and steady curve, does this make the strategy more robust and sturdy? Or is this a useless thing to look for?

   

  Im sorry if you guys think this is too off topic! Great strategy, great work.

  Another question for you Francesco, i’ve seen a screenshot of some of your portfolio numbers, mighty impressive 🙂

  I was just wondering, how long did it take you to make your first profitable strategy, and are you still running it?

  and follow up regarding the question over: Have you had a strategy that was profitable, but stopped being profitable? How did u notice and when did you draw the line? Maybe when the max drawdown on backtest was beaten? (as in you’ve lost more than your backtested max drawdown?

   

  I mean, when can you say that your strategy was too curve-fitted and not working in the real market anymore?

   

  In other words: The more universal strat = better?

  Please share your knowledge on this if possible! What does your experience tell you? 🙂

  Also i got a question about running the same strategy on the same timeframe but on different markets with different optimized variables. Is this generally a stupid idea? Is it better to run the one strategy on just 1 market, then work on a new strategy?

  

  jebus89

  Participant

  Topics: 118

  Replies: 587

  Been thanked: 239 times
  10/16/2017 at 3:51 PM #49584
  ozz87

  @Nicolas: Thanks, I’ll try it 🙂 Got the error this morning also. And my optimization for OMX is gone, it seems PRT removes my codes when I log out sometimes, so got to make sure to export them from now on… 😉

  

  ozz87

  Participant

  Topics: 0

  Replies: 168

  Been thanked: 127 times
  10/19/2017 at 1:44 PM #49942
  ozz87

  @nicolas: The preloadbars code didn’t help. Still cannot run this code for unknown reason.

  

  ozz87

  Participant

  Topics: 0

  Replies: 168

  Been thanked: 127 times
  10/19/2017 at 1:51 PM #49945
  Nicolas

  @ozz87

  The only divisions I see that could generate this error are between lines 32 to 35. A rough solution would be to include these lines with a condition that make sure the differences between highs and lows are strictly superior to 0:

  if (high[2]-low[2])>0 and (high[1]-low[1])>0 then dojibull2 = open[2]<close[2] and abs(open[2]-close[2])/(high[2]-low[2])<x dojibull1 = open[1]<close[1] and abs(open[1]-close[1])/(high[1]-low[1])<x dojibear2 = open[2]>close[2] and abs(open[2]-close[2])/(high[2]-low[2])<x dojibear1 = open[1]>close[1] and abs(open[1]-close[1])/(high[1]-low[1])<x endif

  1 2 3 4 5 6

  if (high[2]-low[2])>0 and (high[1]-low[1])>0 then dojibull2 = open[2]<close[2] and abs(open[2]-close[2])/(high[2]-low[2])<x dojibull1 = open[1]<close[1] and abs(open[1]-close[1])/(high[1]-low[1])<x dojibear2 = open[2]>close[2] and abs(open[2]-close[2])/(high[2]-low[2])<x dojibear1 = open[1]>close[1] and abs(open[1]-close[1])/(high[1]-low[1])<x endif

   

  2 users thanked author for this post.

  ozz87, Francesco78

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20867

  Been thanked: 4783 times
  10/19/2017 at 2:10 PM #49949
  Francesco78

  what I normally do is the following:

  dojibull2 = open[2]<close[2] and abs(open[2]–close[2])/(high[2]–low[2]+0.00000001)<x

  dojibull1= open[1]<close[1] and abs(open[1]–close[1])/(high[1]–low[1]+0.00000001)<x

  dojibear2= open[2]>close[2] and abs(open[2]–close[2])/(high[2]–low[2]+0.00000001)<x

  dojibear1= open[1]>close[1] and abs(open[1]–close[1])/(high[1]–low[1]+0.00000001)<x

   

  1 user thanked author for this post.

  ozz87

  

  Francesco78

  Participant

  Topics: 43

  Replies: 242

  Been thanked: 77 times
• Author
  Posts