Strategy MAEX – experiment with MFE & MAE

• 10/26/2020 at 7:13 PM #148568
  Paul

  This strategy Maex (Maximum Excursion) tries to find a good opportunity to buy the lows.

  So I experimented with MFE & MAE. I ‘ve some doubt if it’s programmed right, but it gives interesting results.

  The algo is based on the dow 1 minute, no position overnight and a stoploss of 0.5%.

  As basis, I wanted to minimise the optimisation of parameters sl/pt/ts/be and all are based in a rough ratio on the stoploss (except the atr trailingstop).

  The goal is to improve on this preferably without changing the ratio’s but through new methods which could replace the current one or add to it.

   

  8 users thanked author for this post.

  TempusFugit, Midlanddave, Kovit, eckaw, GraHal, boonet, Nicolas, swedshare

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  10/27/2020 at 2:47 PM #148660
  Roger

  Intersting! Maybe im wrong but i would write it :

   

  maerec=0 mferec=0 maerec2=0 mferec2=0 for i=1 to n maerec=maerec+(high[i]-open[i]) mferec=mferec+(open[i]-low[i]) next for j=1 to nn maerec2=maerec2+(open[j]-low[j]) mferec2=mferec2+(high[j]-open[j]) next averecentmael=close-(maerec2/nn) averecentmfel=close+(mferec2/nn) avgl=(averecentmael+averecentmfel)/2 averecentmaes=close+(maerec/n) averecentmfes=close-(mferec/n) avgs=(averecentmaes+averecentmfes)/2 avgmfemaes=average[25](avgs)/2 avgmfemael=average[25](avgl)/2

  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

  maerec=0 mferec=0 maerec2=0 mferec2=0   for i=1 to n maerec=maerec+(high[i]-open[i]) mferec=mferec+(open[i]-low[i]) next   for j=1 to nn maerec2=maerec2+(open[j]-low[j]) mferec2=mferec2+(high[j]-open[j]) next   averecentmael=close-(maerec2/nn) averecentmfel=close+(mferec2/nn) avgl=(averecentmael+averecentmfel)/2   averecentmaes=close+(maerec/n) averecentmfes=close-(mferec/n) avgs=(averecentmaes+averecentmfes)/2   avgmfemaes=average[25](avgs)/2 avgmfemael=average[25](avgl)/2

   

  

  Roger

  Participant

  Topics: 9

  Replies: 93

  Been thanked: 30 times
  10/27/2020 at 3:07 PM #148664
  Paul

  Yes your right and that’s the part I had doubts about! I did write it like above but then there too many signals and the strategy breaks. But looking where the signals come, it’s in general very good but needs more work.

  I prefer it the way you wrote it, but how to make it better and outperform the way I wrote it?

   

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  10/27/2020 at 5:51 PM #148705
  Roger

  I have tried to optimize my version but all results are very bad… I dont know why but your version performs far better than mine

  

  Roger

  Participant

  Topics: 9

  Replies: 93

  Been thanked: 30 times
  11/12/2020 at 2:42 PM #150293
  nonetheless

  Hi, just looking at this part of your breakeven code

  if besensitivity=1 then besensitivity=close besensitivity=close elsif besensitivity=2 then besensitivity=high besensitivity=low elsif besensitivity=3 then besensitivity=low besensitivity=high endif if longonmarket then if besensitivity-positionprice>=((positionprice/100)*besg)*pointsize then benewsl=positionprice+((positionprice/100)*besl)*pointsize endif endif if shortonmarket then if positionprice-besensitivity>=((positionprice/100)*besg)*pointsize then benewsl=positionprice-((positionprice/100)*besl)*pointsize endif endif

  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

  if besensitivity=1 then besensitivity=close besensitivity=close elsif besensitivity=2 then besensitivity=high besensitivity=low elsif besensitivity=3 then besensitivity=low besensitivity=high endif if longonmarket then if besensitivity-positionprice>=((positionprice/100)*besg)*pointsize then benewsl=positionprice+((positionprice/100)*besl)*pointsize endif endif if shortonmarket then if positionprice-besensitivity>=((positionprice/100)*besg)*pointsize then benewsl=positionprice-((positionprice/100)*besl)*pointsize endif endif

  shouldn’t it be ?

  if besensitivity=1 then besensitivitylong=close besensitivityshort=close elsif besensitivity=2 then besensitivitylong=high besensitivityshort=low elsif besensitivity=3 then besensitivitylong=low besensitivityshort=high endif if longonmarket then if besensitivitylong-positionprice>=((positionprice/100)*besg)*pointsize then benewsl=positionprice+((positionprice/100)*besl)*pointsize endif endif if shortonmarket then if positionprice-besensitivityshort>=((positionprice/100)*besg)*pointsize then benewsl=positionprice-((positionprice/100)*besl)*pointsize endif endif

  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

  if besensitivity=1 then besensitivitylong=close besensitivityshort=close elsif besensitivity=2 then besensitivitylong=high besensitivityshort=low elsif besensitivity=3 then besensitivitylong=low besensitivityshort=high endif if longonmarket then if besensitivitylong-positionprice>=((positionprice/100)*besg)*pointsize then benewsl=positionprice+((positionprice/100)*besl)*pointsize endif endif if shortonmarket then if positionprice-besensitivityshort>=((positionprice/100)*besg)*pointsize then benewsl=positionprice-((positionprice/100)*besl)*pointsize endif endif

   

  

  nonetheless

  Participant

  Topics: 83

  Replies: 1237

  Been thanked: 978 times
  11/12/2020 at 3:09 PM #150302
  GraHal

  Attached shows the differences (green being None’s proposed version).

   

  

  GraHal

  Participant

  Topics: 211

  Replies: 8773

  Been thanked: 2179 times
  11/12/2020 at 3:24 PM #150304
  Paul

  Thnx GraHal, that made it easy, had to blink twice first 🙂 Nonetheless y’re right. Thnx for pointing it out.

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  11/13/2020 at 10:04 AM #150358
  nonetheless

  Your sensitivity options are such a great innovation for TS if you get it right.

  2 seems to work best for scalping algos where you want to lock it up as quick as poss. 3 (or typicalprice) works better if you’re playing a longer game.

  I want to have another look at this MAEX algo – could have potential. I know you wanted to keep everything related to the sl but I’m not entirely convinced. Freeing that up made an immediate improvement, but I’m sure you looked at that already.

  Also think it could be suited to a longer TF – will try that over the weekend.

  1 user thanked author for this post.

  Kovit

  

  nonetheless

  Participant

  Topics: 83

  Replies: 1237

  Been thanked: 978 times
  11/14/2020 at 10:57 AM #150464
  ullle73

  interesting, following 🙂

  

  ullle73

  Participant

  Topics: 51

  Replies: 178

  Been thanked: 7 times
  11/14/2020 at 2:15 PM #150475
  Paul

  Looking foward to your tests nonetheless. I can’t find good ways to improve on this. Maybe define a bigger trend with hull average, i.e. weekly, and use that to buy in the trend the lows? It maybe works for selling the highs too. I will give it a try later.

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  11/14/2020 at 5:38 PM #150498
  snucke

  Applied a simple filter for trend

   

  see line 74-76

   

  can only test on 100 k bars though

  

  snucke

  Participant

  Topics: 9

  Replies: 143

  Been thanked: 21 times
  11/14/2020 at 9:38 PM #150521
  Paul

  Thnx snucke

  here are both 200k 1m.

  I also have struggled to get good OOS results with WF

  Included a WF test of 50k bars

   

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  11/14/2020 at 10:09 PM #150525
  snucke

  looks decent i like it more with the filter i used.

  and it gave similar results prior to the dates i tested it at

  

  snucke

  Participant

  Topics: 9

  Replies: 143

  Been thanked: 21 times
  11/14/2020 at 11:15 PM #150531
  Paul

  Yeah the equitycurve is indeed better. I included it in the new version but removed vectorial for now, focussed on the 2 methods and added short.

  For testing probably it’s best to use reenter=1 as that’s normally used.

  If optimising for long or short, you can exit on opposite signal but not reversing.

  The parameters aren’t optimised for any timeframe so it looks bad.

  2 users thanked author for this post.

  nonetheless, GraHal

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  11/16/2020 at 1:23 AM #150585
  nonetheless

  Here’s a version of the 1m long. My additions are optimized at 70/30

  I also tried to rework your v3 on a 5m TF, it sort of worked but not so profitable

  5 users thanked author for this post.

  Midlanddave, Paul, swedshare, josef1604, GraHal

  

  nonetheless

  Participant

  Topics: 83

  Replies: 1237

  Been thanked: 978 times
• Author
  Posts