Topic: What do you think about this idea of finding future Eas

Hi,

did someone have tried such idea of generating strategies?

- E.g i use data from 2018-2019 let them generate

- than we use from the collection before and change data from 2018-2020

- so than we will have simulated 1 year forward testing of each startegie and can sort them out and can see the robustness

What do you think?

Re: What do you think about this idea of finding future Eas

use out of sample data. and also consider number of trades . that will be good...

Re: What do you think about this idea of finding future Eas

Roughey wrote:

Hi,

did someone have tried such idea of generating strategies?

- E.g i use data from 2018-2019 let them generate

- than we use from the collection before and change data from 2018-2020

- so than we will have simulated 1 year forward testing of each startegie and can sort them out and can see the robustness

What do you think?

Lots of people do this as a way to eliminate forward testing. From my personal testing this will only get you part of the way there.

For e.g. create some strategies using the default data from JFD. Trade those strategies on a JFD demo account (same data as they were generated on) for 1 month and then re-import them back into EA Studio and change the data horizon to match the demo account dates. You will find that the results will not match, and in some cases and be quite different between what EA Studio thinks would have happened and what actually happened. This is due to many things, the data feed itself, execution issues, using indicators that use weighted metrics or volume etc.

I find the only reliable result is to actually trade the strategies on the account they are going to be used on. Anything other than that is best guess.

You can improve the best guess though by using WFA, OOS etc.

Re: What do you think about this idea of finding future Eas

burrup.lambert wrote:

Trade those strategies on a JFD demo account (same data as they were generated on) for 1 month and then re-import them back into EA Studio and change the data horizon to match the demo account dates. You will find that the results will not match

What's the point in using these programs if everything boils down to gambling? Because if you have developed strats, which do not trade according to their development idea, then it is simply gambling. I have to disagree on non-matching issue. There are clear answers for these problems and once eradicated and solved, it is plain sailing in terms of matching trades.

Re: What do you think about this idea of finding future Eas

footon wrote:
burrup.lambert wrote:

Trade those strategies on a JFD demo account (same data as they were generated on) for 1 month and then re-import them back into EA Studio and change the data horizon to match the demo account dates. You will find that the results will not match

What's the point in using these programs if everything boils down to gambling? Because if you have developed strats, which do not trade according to their development idea, then it is simply gambling. I have to disagree on non-matching issue. There are clear answers for these problems and once eradicated and solved, it is plain sailing in terms of matching trades.

I wouldn't call it gambling. Gambling implies you don't have an edge. And given that the edge in trading is so small every bit counts, so for me, I calculate my metrics/statistics straight from my incubator account itself instead of re-importing them back into EA Studio. This way I can be sure that the statistics are correct, and I can also apply some filtering and other "offsets", like commission, to incorporate some of the costs of trade copying like (lag/delay) or copying trades from an incubator account that has a low commission to a live account that has higher commission: functions that are outside of the scope of EA Studio.

Each to their own though, do whatever works. I do it this way because I've found the results the most reliable and it works for my processes.

Re: What do you think about this idea of finding future Eas

I'm not talking about edges, processes and methodologies. I'm talking about your statement that "the results will not match" between trading and after-the-fact backtesting.