What is a good back testing period/sample size in your opinion

• 12/20/2019 at 6:44 PM #115315
  J.

  I’ve developed a strategy for the 15 minute time frame of the GBP/JPY. I back tested it on as many units as possible (100000). This gives me approximately 2 years worth of back testing data with 116 trades carried out and a 69,26% time in market statistic.

  The strategy carries out both short and long trades and the long signal has the opposite entry conditions as the short signal so that it wouldn’t matter if the market is bullish or bearish in the entire back testing period, it should be able to be profitable either way.

  Is this enough data to say something about the trustworthiness of this strategy in the Forex market, or would one need more years to back test this on to say something about that?

  PRT is great, the only thing I don’t like about it is that the data to back test on in the lower time frame models is quite limited.

  

  J.

  Participant

  Topics: 15

  Replies: 27

  Been thanked: 6 times
  12/20/2019 at 7:23 PM #115318
  robertogozzi

  Upcoming v11 will grant up to 1M bars, thus enhancing backtesting a lot!

  To become acquainted with optimization I suggest that you read:

  https://www.prorealcode.com/blog/learning/strategy-optimisation-walk-analysis/

  https://www.prorealcode.com/blog/video-tutorials/analyse-walk-forward-prorealtime/

  https://www.prorealcode.com/topic/walk-forward-efficiency-ratio/

  https://www.prorealcode.com/topic/walk-forward-but-more-versatile/

   

  2 users thanked author for this post.

  J.

  

  robertogozzi

  Moderator

  Topics: 69

  Replies: 16464

  Been thanked: 3640 times
  12/21/2019 at 12:30 AM #115328
  J.

  Thank you Robert, do you have any information on when v11 will come out?

  

  J.

  Participant

  Topics: 15

  Replies: 27

  Been thanked: 6 times
  12/21/2019 at 9:39 AM #115337
  robertogozzi

  For manual trading it is available (even for free with end-of-day data) at https://www.prorealtime.com/nl/,

  As to IG… well it depends when they will have finished updating their interface for the new version. We all hoped it would have been made available before the end of this year, but rumors say we’ll have to wait a few more months.

  I have found this youtube channel which could be of some interest https://www.youtube.com/channel/UCIu1dQIvIwVJ1K9_hVfXgTA.

   

  1 user thanked author for this post.

  J.

  

  robertogozzi

  Moderator

  Topics: 69

  Replies: 16464

  Been thanked: 3640 times
  12/25/2019 at 10:16 PM #115461
  jebus89

  from my own experience i would say 5-10 years is alot better than 2 years.

  In theory you could make a very robust and profitable system with 1 week or 1 day data even. But I would argue that the chances that your system is actually robust and profitable in the future get smaller and smaller the less data you have! Even if you had 100% of the data it might still not be what you need to create a robust enough system for future results.

  In my opinion: More data = easier to make a robust system because you obviously have more data to look at and test against. But depending on how and what your system is trying to do, you can also test it against similar products (similar pairs f example) and get some extra “data”. But if your system is all about trading between 12.00-16.00 in eurusd, it might not be a very similar fit to test against usd/jpy for example. Something to think about.

  Also another think to have in mind is that (again my opinions only) the more trades you see in a backtest system, the less data is needed for confidence. Lets say you have a 10 year backtest with 300 trades in total, vs a 2 year backtest with 2000 trades, I would probably be looking more into the 2 year backtest than the 10 year backtest for further investigation.

  

  jebus89

  Participant

  Topics: 118

  Replies: 586

  Been thanked: 238 times
  01/05/2020 at 10:42 PM #115975
  J.

  Good point @jebus89. My strategy is on the 15 minute chart, has 113 trades for backtest and works on the GBP/JPY.

  Unfortunately I can only backtest 100k units, but Nicolas said this will go to 1M for IG users soon.

  

  J.

  Participant

  Topics: 15

  Replies: 27

  Been thanked: 6 times
• Author
  Posts