Teach a Man to Fish – "Rise of Machines" version

ag_fx

Well-Known Member
#91
EOD Update 29-Jun-20

If any trend follower "GURU" or any intraday trading "GURU" is telling you that they made a killing today, stop and run away from them. They are likely lying and only interested in your money. Days like today are nemesis of intraday trend following systems.

However, we stick to the system. We don't go back to the drawing board to see if we can tweak any rules. Accepting losses is part of trading profitably.


1593429489164.png
 

ag_fx

Well-Known Member
#92
Hi
How far we should go back with historical intraday data for back testing a intraday 5 minute timeframe strategy? Whether is it enough to back test with 6 months to 1year of 5 minute bars or we should look back for more than a year of historical bars to test for any intraday systems?

Thanks
It depends on the system you are testing. Ideally, you should divide your data into two parts - "In-Sample(IS)" and "Out of Sample(OOS)". You should use IS data to optimize your system. As in decide on look back window, breakout rules, SL% etc. Then test those rules on OOS data.

Never use entire data set to optimize your strategy or make small tweaks here and there. Always use splits.

Regarding a 5 min TimeFrame system, I'd say 2+ years data should be fine as long as it is an intraday system. It will give you roughly 500+ days to optimize and test. Markets evolve constantly and changes regime often. However, I would always suggest having larger data set to test the system. More data means more for you to learn and adapt.
 
#93
It depends on the system you are testing. Ideally, you should divide your data into two parts - "In-Sample(IS)" and "Out of Sample(OOS)". You should use IS data to optimize your system. As in decide on look back window, breakout rules, SL% etc. Then test those rules on OOS data.

Never use entire data set to optimize your strategy or make small tweaks here and there. Always use splits.

Regarding a 5 min TimeFrame system, I'd say 2+ years data should be fine as long as it is an intraday system. It will give you roughly 500+ days to optimize and test. Markets evolve constantly and changes regime often. However, I would always suggest having larger data set to test the system. More data means more for you to learn and adapt.
Thank you for your suggestion.
 

raju.vzm

Well-Known Member
#98
This is a wonderful thread. So good to see the algo-trading/systematic trading version of "Teach a man .." thread. I will definitely try the approach on my historical dataset and backtest it. I'm currently trying a similar approach but more from a swing-trade perspective. So my current long-short triggers are happening on an average of 3-4 days. Downside is that I can't initiate a short->long trade. So opening of my trades is always long.
Any one backtested this system? If so please provide those details

Sent from my Realme1 using Tapatalk
 

ag_fx

Well-Known Member
#99
Any one backtested this system? If so please provide those details

Sent from my Realme1 using Tapatalk
Not yet. There seems to be no way to get pre-open market data for past days. If someone knows, let me know and i can set up.
 

Similar threads