Extract difficult market situation timeframes
#1
Hey crypto trading guys,

I am experimenting a bit with japonicus (brute force strategy parameters with a genetic evolution approach). 
There often seem to be problems with overfitting while using gekko in live trading (I am still only backtesting).


Now I had the idea (link to japonicus issue on github) to improve japonicus to do the following:
not only backtest for a specific timeframe, but also mixing in 5% or 10% into the tests with
timeframes that are particular difficult to handle for strategies. 

For this, we have to get our hands on market times that match this criteria. How do we get there?

Say, we have a strategy relying on the RSI indicator, and parameters for a specific timeframe, three month, one year, whatever.
It is doing extremely good (overfitted?) in that specific timeframe, and when we look at the trading history, 
there are some days (or a week) in the timeframe where it does particularly bad, no matter how good this strategy/parameter pair was
at the end. 

Then we have a second strategy, based on Bollinger Bands, and parameters that are very good for a specific timeframe, too.
But even there are times when the strategy isn't working very good with the market situation.

Perhaps the code of those strategies could be modified to write time periods to CSV when it is not doing a good job.

Then, when some CSVs are collected, they could be run through a script which overlays the data of all CSVs and looks for equality.
The result of this could then be fed to japonicus as particular difficult market times.

Therefore, I am interested in thoughts on this approach and how a strategy could be modified to write out such a CSV.
And perhaps some strategy/parameter pairs that did very well in the past, with the specific timeframe they were used and measured at. 
Can be from 2018/2017, so no one has to reveal his strategy portfolio that he uses today.

Of course it could be that my thinking is too complicated and there is an easier solution for this. I am just beginning with gekko and getting to a
viable live trading solution, so bear with me.

Have a nice week!
  Reply
#2
I 'm an A student because of your assistance!
Thank you for saving me during difficult times and helping me maintain my present GPA standing. More gradesfixer reviews here.
  Reply
#3
time spans that are specific hard to deal with for procedures. For this, we need to get our hands on market times that coordinate this models.
  Reply


Forum Jump:


Users browsing this thread: