Topic: Demo and Real historical data.

I notice that the strategy generated form either demo or real data source even from save behave differently when the data source is switch.

I think while price between the 2 source are similar, I think they have subtle difference. It will be very disadvantage to FSB Pro when it is so sensitive to such small change.

Can the raw data be prepossessed before strategy is generated ? The aim is to lessen the impact of the raw data fluctuation so its more predictable in live trading.

Re: Demo and Real historical data.

I notice that the strategy generated form either demo or real data source even from save behave differently when the data source is switch.

You can have different settings in the data sources: Symbols' settings, Data Horizon, Account Exchange rate. You use a Data Source for fast switching between the settings. You can easily check if both Data Source has equal settings if you point them both to one data folder.

It will be very disadvantage to FSB Pro when it is so sensitive to such small change.

You just prove how accurate is FSB Pro. What I'm doing the last 10 years is to make the world most reliable backtester.  It must be uncompromising to any fluctuations in the market, account and broker parameters.

Can the raw data be prepossessed before strategy is generated?

I think this is counterproductive.

Re: Demo and Real historical data.

We have to know that the data in FSBPro is clean and complete as possible and that the data flow from the MT4 broker will have garbage bars and missing bars and delayed bars and messed up bars.

So...... we have to be sure to have our strategy able to withstand the mess from the broker, we have to allow for variable spreads, broker manipulations, internet malfumctions; our stop losses have to be large enough and we have to accommodate misquotes. Latency will cause problems even if the bars are the same between FSBPro and the trader.

Especially on lower time frames internet trading is a crap shoot and we have to be prepared for data discrepancies. The best strategy in FSBPro can be destroyed by the data from the broker.

These MT4 brokers do not have the reputation for integrity...... and we have to be prepared for the lousy data feeds.

My 'secret' goal is to push EA Studio until I can net 3000 pips per day....

Re: Demo and Real historical data.

Hi mel8331,
If you check the strategy with two different data series and the result is different, than the strategy is curvefitted.

This one of the biggest problems in automatic strategy generation.
To improve the curvefitting tests should be the biggest target for Forex Strategie builder pro.

At the moment it is not solved.
All tests to find non curvefitted strategies failed.

I check differnt timeframes for the same strategy. Different Dataseries. Different currencies.
Robustnesstest.
This tests are not able to filter all curvefitted strategies.

thomas

http://www.tradessystem.com/

Re: Demo and Real historical data.

Thomas

What do you recommend to avoid curve fitting...?

What have you seen elsewhere to avoid curve fitting..?

Please elaborate so that Popov can consider your ideas.

thanks

My 'secret' goal is to push EA Studio until I can net 3000 pips per day....

6 (edited by mel8331 2014-12-02 17:54:24)

Re: Demo and Real historical data.

tnickel wrote:

Hi mel8331,
If you check the strategy with two different data series and the result is different, than the strategy is curvefitted.

This one of the biggest problems in automatic strategy generation.
To improve the curvefitting tests should be the biggest target for Forex Strategie builder pro.

At the moment it is not solved.
All tests to find non curvefitted strategies failed.

I check differnt timeframes for the same strategy. Different Dataseries. Different currencies.
Robustnesstest.
This tests are not able to filter all curvefitted strategies.

thomas

Can the problem of over fitting be over come by FSB Pro using multiple data source during the generation process, ie strategy that is profitable from source A but not profitable from source B are not selected but strategy that survive multiple data source is allow to survive.

Can strategy generation using switching multiple data source switching during the generation process prevent curve fitted strategy ? If it can then that reduce the testing time on which generated strategy is robust.