Question from Trent:
Quote:
In relation to testing on in sample/out of sample data. As you mention in the course documents we already have a bit of an understanding of what our system results will be over a certain period. Say for example if we divide our dataset into three from 1995 to present. If we choose the middle set of data for our in sample phase we will be testing over a bullish environment. If you are testing a long only system on this set of data you would expect it to perform fairly well.
I tested one of the systems that I am trying to develop a period between 2001 to 2010 and the results were reasonable (22% cagr/ 18%dd). I then tested it from 2010 to present it had been in drawdown the whole time. I know that you wouldn’t expect great returns from the second period but that doesn’t quite seem good enough. So from these two is different sets of data would you discard the entry and keep exploring other options?
Great question.
A lot of trend systems suffered between 2010 and 2015 and this is actually a good thing – it shows the vulnerability of that specific system from which you can now build upon.
The foundation question start with, “Am I happy with that period of underperformance (2010 – 2015)?. ” If which the answer is invariably ‘no’ then you need to either work with something else, figure out what was causing the issue and decide if it could’ve been avoided, or diversify it out using a secondary system.
