Topic: Crazy Strategie

Hello.
I have found a Strategie, which i have run live.

Account 1 for 2 Months = 97 Trades and in this 95 Profit Trades and only 2 Loss Trades.

Account 2 live for 3 Weeks = 27 Trades and all profit Trades.

Now i have backtest this Strategie with MT4 and EAStudio

and in EA Studio and MT4 Tester all backtests are loss. That is crazy or is that possible?

All the same Broker Datas from live.

Greetings, Marcus smile

Re: Crazy Strategie

Hello Marcus,

Where did you find the strats? From Studio?

Re: Crazy Strategie

footon wrote:

Hello Marcus,

Where did you find the strats? From Studio?

Hi Footon,

yes all in EAStudio found.

I have now a second Strat found, in this is the same crazy Situation big_smile

Re: Crazy Strategie

Yes, this is possible. My first explanation would be that the sample size is small and it magnifies the result. Backtesting on 20k bars and trading for ~1k bar is not that comparable, it might be genuinely profitable strat or market conditions played into its hand. I'm not saying it is all doom and gloom but one has to be prepared for it, either way my fingers are crossed and wishing you good luck smile If you can, keep us posted how the live trading will advance.

Re: Crazy Strategie

footon wrote:

Yes, this is possible. My first explanation would be that the sample size is small and it magnifies the result. Backtesting on 20k bars and trading for ~1k bar is not that comparable, it might be genuinely profitable strat or market conditions played into its hand. I'm not saying it is all doom and gloom but one has to be prepared for it, either way my fingers are crossed and wishing you good luck smile If you can, keep us posted how the live trading will advance.

Ok, thank you smile

Yes i post here, no Problem.

Lets hope that it remains good smile

Re: Crazy Strategie

Excellent news -- that means there is still hope...

I have similar observations to add.  I have seen strategies with great statistics trade poorly, and also strategies with poor stats trade well.  So -- statistics alone don't tell the whole story.  In my experience, the data set used for back testing makes a *big* difference.  If you were to use a different data set then you might get better statistics that more closely correspond to how the strategy is currently trading in your live account.  I don't know -- I'm just guessing...

I had a question -- if the strategies had such poor statistics, then how did they make it into your collection in the first place?  If you were using EA Studio, then EA Studio must have thought they were good enough to make it into a collection in the first place.  Is your recent back testing similar or different than what was used when your strategies were originally generated by EA Studio?

7 (edited by rantampla 2017-09-17 00:26:43)

Re: Crazy Strategie

sleytus wrote:

Excellent news -- that means there is still hope...

I have similar observations to add.  I have seen strategies with great statistics trade poorly, and also strategies with poor stats trade well.  So -- statistics alone don't tell the whole story.  In my experience, the data set used for back testing makes a *big* difference.  If you were to use a different data set then you might get better statistics that more closely correspond to how the strategy is currently trading in your live account.  I don't know -- I'm just guessing...

I had a question -- if the strategies had such poor statistics, then how did they make it into your collection in the first place?  If you were using EA Studio, then EA Studio must have thought they were good enough to make it into a collection in the first place.  Is your recent back testing similar or different than what was used when your strategies were originally generated by EA Studio?

Let me explain.

First i search at least 7.200 Minutes in EA Studio for Strategies. When not 100, then again 7.200 Minutes.

When this finish, i take it on Demo Account, After 100 Trades and Profitfaktor over 1,00 i put it on Live.

This was with this the same.

I have found very good Strats, have 100 on Demo and after the Testtime i have from this 100 ~15 on Live.

After 1-3 months or ~ 100 Trades i look the Stats, that i can maybe go over 0.01 lot.

This i have make every weekend. But today i have 2 perfect Strategies in MT4-Tracker see and i wanted to see what for Settings in Strategy, but i have forgot save collection (.json) and then i have in Portfolio MQ4 the Strat take and load in other EA to reload in EAStudio.

Then i was totally bewildered. EAStudio very bad backtest and after this i have load in MT4 Tester, the same very bad. But in live very very good.

Now i think, what is when we use bad Strategies from begin....but i dont know maybe it was coincidence.

I excited for next weekend, to see that goes continue in the same or not.

I mean with this text i have found good Strategies in Backtest few weeks back and now the Backtest very bad but live very very good.

Sorry for my bad Englisch, i hope you can read it smile

Re: Crazy Strategie

Thanks for the explanation -- I think I understand.

Since the strategy was created 1-3 months ago, then the DataHorizon (i.e. the data set) used to generate, optimize and back test is different than the one you used just recently when you reloaded the strategy into EA Studio.  Since the data set can make a big difference, then it could be the data set used 1-3 months ago did a good job of generating and optimizing your strategy and that is why it trades well. 

People use different terms that mean the same thing -- "optimize", "curve-fit", "train".  I prefer "train".  The data set that you use to "train" a strategy is used to compute the best settings for each indicator in your strategy.  So, if you use a good data set then the indicator settings are more likely to work well when live trading.

Re: Crazy Strategie

sleytus wrote:

Thanks for the explanation -- I think I understand.

Since the strategy was created 1-3 months ago, then the DataHorizon (i.e. the data set) used to generate, optimize and back test is different than the one you used just recently when you reloaded the strategy into EA Studio.  Since the data set can make a big difference, then it could be the data set used 1-3 months ago did a good job of generating and optimizing your strategy and that is why it trades well. 

People use different terms that mean the same thing -- "optimize", "curve-fit", "train".  I prefer "train".  The data set that you use to "train" a strategy is used to compute the best settings for each indicator in your strategy.  So, if you use a good data set then the indicator settings are more likely to work well when live trading.

Yes, i think you are right.

This Strategies are full optimized. With lower optimization i have not so good results.

Re: Crazy Strategie

But i am now convinced. For me, it is the best way, produce 100 Strategies. Run it and after 100 Trades take the best with MT4-Tracker analysed.

With this i think, i can maybe win this game smile

Re: Crazy Strategie

and in EA Studio and MT4 Tester all backtests are loss. That is crazy or is that possible?

EA Studio was designed to simulate the MT4 result as close as possible.
Even more, it produces the exactly same results from the last update (I'll post a video these days how we can do that).

However, neither MT4 nor EA Studio are capable to recreate the exact real trading behaviour during a backtest. In order to prevent showing overestimated results, EA Studio uses the so called Pessimistic method of bar interpolation. That means it is possible to have strategies with ambiguous bars and lower stats that trades well on the market.

If we want to recreate the exact trading results, we need tick data for the hole data set. Unfortunately this is almost impossible and neither MT4 nor EA Studio do that. Please note that using tick data from Dukascopy is not a solution.

Our purpose is to create the best possible tester and Generator for the real world.

Re: Crazy Strategie

Popov wrote:

and in EA Studio and MT4 Tester all backtests are loss. That is crazy or is that possible?

EA Studio was designed to simulate the MT4 result as close as possible.
Even more, it produces the exactly same results from the last update (I'll post a video these days how we can do that).

However, neither MT4 nor EA Studio are capable to recreate the exact real trading behaviour during a backtest. In order to prevent showing overestimated results, EA Studio uses the so called Pessimistic method of bar interpolation. That means it is possible to have strategies with ambiguous bars and lower stats that trades well on the market.

If we want to recreate the exact trading results, we need tick data for the hole data set. Unfortunately this is almost impossible and neither MT4 nor EA Studio do that. Please note that using tick data from Dukascopy is not a solution.

Our purpose is to create the best possible tester and Generator for the real world.

Thank you for this good Info Miroslav smile