Expected degree of accuracy

Discussion of the latest application betas including reports of bugs, issues and suggestions before it is made available as a general release.
Post Reply
Emersonian
Posts: 2
Joined: Fri Mar 03, 2023 11:27 pm

Expected degree of accuracy

Post by Emersonian »

Can someone advise how accurate I should expect the results to be, comparing backtest ATM to forward test in same period and settings please.
tickstory
Posts: 5040
Joined: Sun Jan 06, 2013 12:27 am

Re: Expected degree of accuracy

Post by tickstory »

Hi Emersonian,

The question of "accuracy" is a bit of an open-ended question as it will depend a lot on the nature of your EA. If, for example, your EA is only making one trade every day and the order entry trigger is a simple moving average, then you are likely to get much more similar results in your forward test than a scalper.

What you should be comparing in your back-testing versus forward test comparison is that the performance is largely the same. That is, are you getting similar a similar profit curve to what you back-tested? Are the draw-downs similar? Are you getting roughly the same amount of trades? You'll need to compare these critical metrics to understand if your forward-testing is matching your back-test results.

The settings you ultimately choose should be the most "stable" set across all your forward-tests (i.e. the one that represents the most stable draw-downs, profit curve, etc).

Hope this helps.
Emersonian
Posts: 2
Joined: Fri Mar 03, 2023 11:27 pm

Re: Expected degree of accuracy

Post by Emersonian »

"The settings you ultimately choose should be the most "stable" set across all your forward-tests (i.e. the one that represents the most stable draw-downs, profit curve, etc)...."

Are you suggesting that back testing is not a suitable tool to choose EA settings ???

What purpose does backtesting with ATM serve ?
tickstory
Posts: 5040
Joined: Sun Jan 06, 2013 12:27 am

Re: Expected degree of accuracy

Post by tickstory »

Hi Emersonian,

The statement is saying that you should choose the best settings that you have found through the ATM. Your back-tests will get a set of candidate settings and the forward testing you do with the ATM will validate those settings are working on "unseen" data.

There are few a sections on "Walk Forward Analysis" in the ATM manual which go into more detail about how walk-forward analysis can help your testing (see the "A Primer on WFA" and "How does the WFA work?").

Hope this helps.
Post Reply