This is importance to evaluate the EA.. But you must think that market cycle is vary, depend on currency pairs. I test many good EA with good results and fail in one day at Crisis and destroy all fund.
test it for 3/4 years AND on different pairs that's important because some EAs are curve fitted to the data of a specific pair but will fail on a new pair
Just my point of view.
The thing is the longer time setting you went through, the more possibilities of market condition your EA will face.
If your EA can survive with low deposits amount for a long time (maybe 10 years) in the tester, it simply means it can adapt to many market possibilities.
However, there are several factors that make me no longer depending on the tester such as controlled volume and spread by a broker.
You can test as long as you want but there will always be a chance of new market possibilities in the future which render the tester useless.
What you really need is an EA who can read the market condition and predict its closest possibilities, and at the same time can adjust to your broker settings.
Ahmad is right, backtest data will never be as precise as real ones. If you don't know what your EA is doing (I presume this is your case in front of your question) then forget such solution. Backtests are usually used when developping with a particular focus in mind. If you want to test an EA then plan a forward testing on the instruments you will finally use. For how long? that depends of the trading style, of what the EA is ready for (various situations that can misguide an EA, they can occur only time to time, only experienced ones will go through safely), of your own appreciation (your expertise).
See if there are forward test available or ask for one.
You want more than 30 total number of trades, or you do not have a valid statistical sample.
I usually don't go for fewer than 100 total number of trades for a back test.
I run the real and then back test its. Result could be opposite! I run in demo account and I run in real. Result is not same, real account always have much noise.. really .. chaiya
I´m looking for conditions time for several years.
I see all thye comments. I know that several years submit EA to several conditions and possibilities. But I think that 5/6 years is a good period. The world changes, quantity of people on earth grows, volume of trades in world grows. Today, everyone can trade at home by internet (in back times, it was restrict to phisical places). So, I imagine that 10 years is something very difficult to reach good results on EA.
Another question is: I downloaded historical data from http://www.histdata.com. I have an script that convert m1 to all timeframes. But, I made a test and checked that differences goes around 15 points or more sometimes if I compare price with another broker. Is it possible download historical data with more precision on levels price?
I checked that the same strategie needed change volumes to runs well on others pairs...
I have a strategie focus on mind. I get good results for 6 years. But, I was checking what traders at community thinks about it. I agree the longest test is better, but the world changes definetelly at 10 years as example.