Topic: Some ideas for improving the strategy generator

I have a few ideas for improving the strategy generator.  These would be somewhat slow on some computers so they would have to be optional.

1.  The ability to optionally go through every strategy FSB can create rather than the pseudo random method used now.

2.  Checking and marking over optimized indicators.

3.  Over optimization checks.  The basic idea would be to plot the equity  curve of the selected strategy with normalized values between 0 and 100 then check the various possible indicator values to check for deviation from that curve.

Just my 2 cents
-Allen

Re: Some ideas for improving the strategy generator

Hi Allen,

Nice ideas, but 2 of them are actually being implemented and further developed already, in the newest beta you can set it to write optimization reports, in generator you can set the deviation curve for OOS etc. Only thing which is not there is the first idea you mentioned, but the more I think about it, the less I understand it. Can you elaborate your idea please?

If you have more thoughts about improving FSB concept, keep them coming  big_smile

Re: Some ideas for improving the strategy generator

footon wrote:

Can you elaborate your idea please?

Certainly.  It is my understanding that the current way that the generator functions is to randomly generate a strategy withing the given limitations and compare the results with the previously generated strategy.  It should be a simple matter to have it run through an iteration loop and generate a strategy for each possible indicator with every possible setting and compare that with the previously generated strategy.  It is understood that it would take longer than the current implementation as I understand it so it would have to be something the user could turn on or off and would also require disabling the maximum time option.

Re: Some ideas for improving the strategy generator

It would also be nice to have sanity checks for some indicators too.  for example I have seen cases where the macd fast ma is a longer period than the slow ma.

Re: Some ideas for improving the strategy generator

I should add that it would also be nice to be able to tell the generator not to use certain indicators that end up with insane values I.E. Previous high/Low with a vertical shift of -1000 on a 5 min time frame. This would NOT be so simple to implement.

Re: Some ideas for improving the strategy generator

DragonEagle wrote:

the user could turn on or off and would also require disabling the maximum time option.

type a zero and it runs indefinitely

Re: Some ideas for improving the strategy generator

footon wrote:
DragonEagle wrote:

the user could turn on or off and would also require disabling the maximum time option.

type a zero and it runs indefinitely

Yes that's in the documentation.  I was just saying that the 2 options would not work well together.

Re: Some ideas for improving the strategy generator

all in all I understand now, but the limits have to be there - for example it is a must to define the number of slots, as I understand you want to use as many slots as there are indicators, but this is not practically achieveable.

About sanity check, I don't know; if the result is what one is after, then it doesn't really matter if it is in opposition to indicator's principle. However what I'd like to see is more methodological approach, for instance I want to improve the strategy by one indicator slot at a time, and that generator would go through all indicators/logics/base parameters but NOT optimizing them in the process. When I try to do this in current versions, it pops up "Day of Month" or similar indicator with insane values (clearly a sign of curve fitting), which is... stupid if I'm frank.

Re: Some ideas for improving the strategy generator

footon wrote:

as I understand you want to use as many slots as there are indicators, but this is not practically achieveable.

Actually no.  I just want the option to cycle through the indicators up to a maximum of the selected number of slots rather than choosing them randomly.  As for sanity checks, it would require adding an additional method to the indicator to verify that the values were good e.g.

bool hasSaneValues(){
    return slowMA > fastMA;
}

an inverted indicator like that could also be a sign of curve fitting.

Re: Some ideas for improving the strategy generator

If this may not be effective for some traders because it could slow computers, then there should be a way to improve this strategy generator to make it more effective for all.

Re: Some ideas for improving the strategy generator

I had an idea which I thought may improve the strategy builder,   if there was an option,  where the option is for you to pick on next signal to do "nothing", "add" or "winner", if there was another option for modify,  as in the add feature where it modify's all orders in a grid type strategy, but instead of opening another trade it just changes the current open order

Thanks for the great product and all of the work!

Re: Some ideas for improving the strategy generator

Hi i have idea, if is it posible ofcours.
Automaticly save the generatet strategy at every "X" minutes
For example start the generator at 8 am, make it at every 10 minets auto save the generatet stragey, and start automaticly next strategy whit the first parameretes, then at 8 pm we will have 72 generatet strategy and we will save alot of taim.
Best regards! smile

Re: Some ideas for improving the strategy generator

Actually you should have more than that.  after you run the generator, use ctrl+h and ctrl+j to cycle through the generated strategies.

Re: Some ideas for improving the strategy generator

I noticed this, too. And when I looked at the FSB GUI interface, I noticed that the Fast MA was set to a longer timeframe and the Slow MA was set to a shorter timeframe. FSB inverted the numbers after I ran a generated strategy. But when I looked at the the default, the MACD had the Fast and Slow parameters correct (the fast was the small value than the slow)  Is there any information as to why this is the case?

I assume its possible to implement the inverted strategy, but I use Oanda, and I their graphical chart cannot implement a longer "fast" MA over a shorter "slow" MA. So any generated strategy using MACD with the numbers inverted is useless.


DragonEagle wrote:

It would also be nice to have sanity checks for some indicators too.  for example I have seen cases where the macd fast ma is a longer period than the slow ma.

Re: Some ideas for improving the strategy generator

binaryoptions wrote:

I noticed this, too. And when I looked at the FSB GUI interface, I noticed that the Fast MA was set to a longer timeframe and the Slow MA was set to a shorter timeframe. FSB inverted the numbers after I ran a generated strategy. But when I looked at the the default, the MACD had the Fast and Slow parameters correct (the fast was the small value than the slow)  Is there any information as to why this is the case?

I assume its possible to implement the inverted strategy, but I use Oanda, and I their graphical chart cannot implement a longer "fast" MA over a shorter "slow" MA. So any generated strategy using MACD with the numbers inverted is useless.


DragonEagle wrote:

It would also be nice to have sanity checks for some indicators too.  for example I have seen cases where the macd fast ma is a longer period than the slow ma.

Allen made a custom MACD indicator with sanity check: http://forexsb.com/forum/topic/2030/tes … ty-checks/

Is there any information as to why this is the case?

It's the pure result FSB is after, and often strategies give a better result if fast "becomes" slow and vice versa, principally it can be set right by reversing the logic as well, but this idea needs to be tested and confirmed.

Re: Some ideas for improving the strategy generator

Well, this is quite better! Thank you very much!

footon wrote:

Allen made a custom MACD indicator with sanity check: http://forexsb.com/forum/topic/2030/tes … ty-checks/

It's the pure result FSB is after, and often strategies give a better result if fast "becomes" slow and vice versa, principally it can be set right by reversing the logic as well, but this idea needs to be tested and confirmed.