Thread: How do i use optimization!

Originally Posted by wannaberich

I am very new to metatrader and have just come accross this trading system forex website which is fantastic and very helpfull. I want to optimize some of these EAs to give me the best results.

How do i go about this using metatrader and once it is optimized what does that acctually mean and how do i use those optimized results to improve the EA.

Thanks
Jason

Hi wannaberich and welcome,

First of all, you have to download a good tick data :

Tools > History Center > the wanted currency > Download.

Then you do Ctrl+R, you check Optimization, you select Every Tick Model, you take the Symbol, the Period of your choice, you can also check "Use Date".

Finally you click on Expert Properties :

Here, depending on the backtesting (if it is done fast or not) and the power of your computer, you can check one or several boxes in Inputs. You have to enter for each a Start a Step and a Stop. If you enter 10-5-500, it will begin by 10, then it will try with 15, 20, 25, 30, ... up to 500.

When an optimization is done, you go on Optimization Results. Here you can select the best parameters (Set Inputs Parameters) according to what you consider to be a good result, for example you can accept 30% of max drawdown, or 20%. And you continue until you find the best ratio Profit/Drawdown.

Note that you need a 90% modelling quality, you get it with a good tick data.

Originally Posted by wannaberich

Hi

Thanks for the reply, really appreciated, i will try out what you have said and see what happens. One other thing i have done some basic backtests of several EAs before but never get a 90% quality, in fact most times its either na, or something more like 24% if i am lucky etc.

If it is not 90% does that mean the EA wont make money on a live account, or is having 90% just better. Sometimes i backtest an EA and get really poor backtest quality, yet it gives me a really high profit, why would this be. It is either something i am doing wrong or it is my data supplier, i guess it probably depends on how good you data is in the first place

Do you have any ideas why this could be

Thanks
Jason

A 90% modelling quality will generate theorical results closest to the real results on the respective period than with a 24% modelling quality. But then whatever is the modelling quality, you have this rule : past results do not guarantee future performance.

Originally Posted by simpleis

mel8331, can you explain more on how you optimize and test? Thanks

Basically, I leave about 1-2 weeks of the latest data out of the optimization period depending on how many trade the system trades in a week or two.

So I would optimize for period say 1st Jan 2008 - 31st Dec. 2008. After completion of the optimization, I would change the test period to 1st Jan 2009 -- 18th Jan 2009.

Now I goes back to the optimized parameters looking for low drawdown, high profit factor and high number of trades. After selecting this parameters, I would test them for the period that it was not optimized, in this example, it would be 1/1/09-18/1/09.

I would do it for a few sets of parameter and look at the equity curve, basically looking for a as smooth as possible upward trend and then look at the report for the profit factor, drawdown and the profit.

After doing it for a few parameters selection, it is your judgement call on which set of optimized parameters you will use in your forward trading.

If you do that sometime you may find that the most profitable optimized parameter turn out to be losing money in the unoptimized period. So if you have use that in actual trading, you would end up losing money.