Robustness in Automated Algorithmic Trading Systems

• 06/03/2016 at 10:27 AM #8715
  Nicolas

  Hey David, thanks for your updates on these tests. Because I don’t know how are calculated your random trades PnL, I can’t obviously make any comment on the results of your test 🙂 But on the pictures, the optimized one seems more relevant indeed.

  About the TS that work on all european indexes.. I cannot say more than:  if you trade the almost same exact curve twice with the same criterias to enter and exit, you should expect to have at least things pretty similar.

   

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20863

  Been thanked: 4773 times
  06/04/2016 at 7:17 AM #8785
  David

  I would like to continue the discussion but I feel I need to become more experience and knowledgeable in the subject and I believe that this might be very subjective subject.  Just like asking economists opinion of rate decision sometimes: 50% will say A and 50% will say B.

  Currently due to life commitments I need to decide where to spend my efforts to gain skills, in programming to write better TS (i’m not a trained programmer) so I can quit my day job or develop a framework for testing robustness or at least more research into the area. For the moment more programming skills.

  Nevertheless my opinion is that if 2, 3 , etc… indexes have a high correlation in price action then the TS should work across all of them but then what’s your management approach. Spread your investment across this indexes and get somewhat similar results which requires more management?

  In other words, you have a TS and it works on 5 indexes does that many you feel comfortable that it is robust and will last the test of time? It won’t make me feel comfortable.

  General investment theory is to spread your risk across assets which are not too correlated. So why not develop many TS for non correlated pairs/indexes to spread your risk?

  

  David

  Participant

  Topics: 24

  Replies: 170

  Been thanked: 31 times
  06/04/2016 at 11:02 AM #8787
  deleted190722

  I agree… but… I would be more confertable to ha one TS on 4/5 index another one totally different on other 4/5 index. etx… this would be my best practice…

  If you look into my supertrend TS on 5 different indexs (MIB, DAX, CAC , GOLD, EUROSTOX) if you would start the TS in only one of them you will have big drawdown… But if you start it in 5 different indexs the drawdown will be much less and in my opinion it helps you to do not stop the TS.

  

  deleted190722

  Participant

  Topics: 8

  Replies: 80

  Been thanked: 6 times
  06/05/2016 at 9:38 AM #8807
  Nicolas

  This is something to discuss about : portfolio diversification with averaged weight for all assets. There are billions of ways to do it .. One simple rule is to weight more the strategy that have better PnL over a recent momentum period, and decrease ones that not perform so well. If you have 5 strategies running, you can establish a ranking method that affect your money management. While it’s not possible to trade multiple instruments with ProOrder for the moment, it is possible to calculate the momentum of the strategy profit and to change the lot size accordingly to its momentum at least individually in each one. It’s just another simple function to code 🙂

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20863

  Been thanked: 4773 times
  06/05/2016 at 10:40 AM #8818
  David

  David-1984 I agree with you that a TS running on 4/5 is better than 1/5 (I think that’s what you meant) but I’m not sure I would run it on the 4 indexes because if they are highly correlated and something changes then they will all start to lose. I saw your plotted returns graph but I would be cautious on this as an assurance. When I run it on Australia Cash 200 index it just loses money.

  I backtested your Supertrend DAX 30 without any spread because I’m not sure what it should be? Or is it already factored in the code somewhere – I can’t see that? Also I can only get 5EUR as the smallest contract maybe because I’m not in Europe.

  Overall your TS is profitable and I like:

  Avg Gain >  Avg Loss

  but the areas that concern me are:

  2012 September to 2014 August is effective 2 years of losses.

  Losing trades > Winning trades

  Maximum losses: 9 (this makes me really uneasy)

  What happens if the you start getting the same results as 2012 September to 2014 August?

  My suggestion is that you try to find a way that the TS will not run during this negative period. So even though you won’t have trades, you won’t have losses.

  Nicolas would you have portfolio diversification on this situation with highly correlated indexes (I’m not sure that they are GOLD shouldn’t be but if they are in the EURO zone let’s assume they are more correlated than other indexes) even with different weightings – would you diversify like this?

  

  David

  Participant

  Topics: 24

  Replies: 170

  Been thanked: 31 times
  06/05/2016 at 2:02 PM #8829
  Nicolas

  I were talking about strategies diversification in my last reply, but of course instruments diversification come first. So IMO, instruments that are highly correlated should not be traded at the same time with the very same method.

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20863

  Been thanked: 4773 times
  01/26/2020 at 8:08 AM #117855
  jmf125

  Nicolas wrote:

  I preferred first to test your result on a non-optimized basis, since if I remember correctly, it were the one with indicators default value, isn’t it?

  Results of the analysis are not bad, but the distribution of the random strategies results are quiet wide, so the confidence that a probabilistic future result of this strategy traded in the future would be around the mean profit are poor.

  But the mean profit is still positive though.

   

  Hi Nicolas

   

  Is the monte Carlo Excel sheet available to download on the site or on the web somewhere ? Or is this your hard work ?

  

  jmf125

  Participant

  Topics: 8

  Replies: 111

  Been thanked: 14 times
  01/26/2020 at 10:25 AM #117858
  GraHal

  jmf125 wrote:

  Is the monte Carlo Excel sheet available to download

  Yes here you are …

  https://docs.google.com/spreadsheets/d/166j305XG7Vz4AP4zDFfC4OPxwtXRn8ElImLa3awsDFE/edit?usp=sharing

  1 user thanked author for this post.

  jmf125

  

  GraHal

  Participant

  Topics: 211

  Replies: 8761

  Been thanked: 2176 times
  01/26/2020 at 12:02 PM #117871
  jmf125

  Thanks Grahal

  I have this one, but what Nicolas is using is different. It seems more comprehensive in terms of MC simulation.

  

  jmf125

  Participant

  Topics: 8

  Replies: 111

  Been thanked: 14 times
  01/26/2020 at 3:45 PM #117906
  GraHal

  david-1984 wrote:

  here the result with the montecarlo testing of the 2 TS.

  If you mean the Monte Carlo Sim shown on the post link above then I think david-1984  made it so you need to ask him if he will share it?

  

  GraHal

  Participant

  Topics: 211

  Replies: 8761

  Been thanked: 2176 times
  01/27/2020 at 12:19 PM #118016
  jmf125

  Nicolas wrote:

  I preferred first to test your result on a non-optimized basis, since if I remember correctly, it were the one with indicators default value, isn’t it?

  Results of the analysis are not bad, but the distribution of the random strategies results are quiet wide, so the confidence that a probabilistic future result of this strategy traded in the future would be around the mean profit are poor.

  But the mean profit is still positive though.

  Hi Nicolas,

  Is the Spreadsheet shown in the attachment available somehwere or is this your personal product ?

  

  jmf125

  Participant

  Topics: 8

  Replies: 111

  Been thanked: 14 times
• Author
  Posts