Ideas on Backtesting

Someone say that 1 month of live trade is better than 10 years of back-test.Is that true?

Maybe backtest is bad and let's discuss why.

Q1.the data you use for back test is not same as the really market.
A1.that's ture.But you can download the high-quility data form some web like dukascopy,as you trade only 1or2 lots one time and your slippage is not huge with a good broker,I think the back-test can 99% = the live data.

Q2.the over optimization stuffs.let's me explan this first.think about if you go back to 1 year ago and let you trade this year again,will your trade result be great?sure it will be great,yeah?So how can an ea be bad if you optimizate it with the data of last 1 year and back-test with the same data?but as you can never optimizate with data of future,this backtest result do not mean anything.
A2.that's ture as well.But in one hand,we can use backtest to test our system but not to optimizate it,in another hand,we can use WFT to make good use of back test.check this to read detels about WFT:
http://www.multicharts.com/walk-forward-testing/ http://www.tradestation.com/trading-...ward-optimizer

my question:anyone know a easy way to do WFT with MT4?it takes alot of time if done by hand.

I have only done WFT once on MT4, it didn't seem that difficult to do, but I didn't see the advantage. I did it as a bit of an experiment and probably didn't do it right, but I couldn't see any point in spending more effort.

With the usual crappy historical data, WFT can't give you an advantage over customary back testing. Its still garbage in, garbage out. I think the focus should be on data quality, not a fancier testing scheme.

1. I doubt that anybody is going to be profitable after just 1 month of live trading!!

Re back-testing, I agree with you that data must be accurate. I'd also suggest that you need to somehow find the right balance between having enough data to ensure statistical validity (too little data runs the risk of being 'fooled by randomness') and obsolescence (old data may be inaccurate simply because market cycles, patterns, daily ranges, fundamentals/economics and other drivers, etc have changed in the interim). Also, your underlying criteria need to be robust enough to avoid 'curve fitting'.

Much also depends on how much of your trading methodology is 'mechanical' (can be coded into an EA) and 'discretionary' (rules are either too complex, or too vague, to be coded). To state the obvious, if it can't be coded, then it can't be back-tested using automation.

If your bias is toward mechanical trading, here are two excellent posts that might provide insight as to the processes required: here and here. And also this article by Merlin Jeffries. Merlin makes the great (IMO) point that the developer should use the data to prove a theory, rather than using the data to try to create a theory.

2. Here are links to two simulators: vHands and Quantem. I haven't used either recently, although I did experiment with vHands briefly back in 2008. You might like to give them a try. Instructions are in the linked threads.

Hope that helps you to get started.

David

thanks for the posts URLs,that's really great and helpful.
I agree with you that NOBODY can be profitable after 1 month of live trade.
thanks

Hello,
There is a rumor in Forex that says Back Testing is not good and it proves nothing.
I think this rumor began when some vendors tried to prove their products via showing some awesome back tests, but in real their customers were not satisfied from the results.
Back testing would be so useful if you use it properly,
First your input data must be accurate, second choose the correct model for that.
I would recommend you to use Open Price Only and do not use optimization, instead you can use some temporarily drawing code in your EA to see what does your EA really do and then back test with your eyes and find out in what situations your EA had bad results, finally maybe you decide to add some other indicator to your EA to rise its quality. It is what I always do.

Very helpful post Hanover

Thought I'd add another simulator that might be of interest: Forex Tester

It's a standalone app that kinda looks and feels like MT4.

I haven't used it for anything automated, although it does support such testing as well, like this thread suggests, but I've used it to replay market data of news events in the past to practice manually trading/scalping high volatility post-news releases. (The trick of course is to not look at the price history before testing yourself...)

It's a good tool for people who want to get simulated screen time on weekends and such.. Though, they've started a subscription fee for decent tick data and won't let you import your own tick level data. You can only import M1 data and higher without a subscription service.