here are a commentary about backtesting of a business company which are coding EA:
At least the big difference between Backtest and Forward-Test is noticeable for system developers when they activate a system after a successful development in Live-Trading. Quite often the excellent performance bend in Backtest turns out to be a completely unpleasant bend in the live-operation. So it could happen that a profitable system becomes a loss –maker. We have had this experience as well.
Well, what are the reasons for this?
1. MetaTrader doesn’t recognize tick-data
All the developed steps and decisions are basing on the available and historic data if you are developing a system. But the available data are not tick-data. Many developers believe that they are developing on the basis of historic real passed benchmark data. That’s not the case because MetaTrader calculates Pseudo-Ticks and how they could have been on the basis of 1minute candle with the appropriate High/Low/Open/Close.
Even Scalping systems which appear virtually fantastic in Backtest, fail regularly on this fact. Although of course we are developing our own systems on this basis of available data. Then, after gathering the appropriate forward-test data we either make improvements on that system or decide to reject it.
2. All Backtests are based off the data which had been loaded by Metaquotes Server!
It doesn’t matter which Broker you got. The data in the development is based on the provided data by Metaquotes. The „correct“ data is not available at Forex-Markt but every Broker / Dealing-Desk makes his own prices or rather conveys each prices of the associated banks. In reality this leads to the phenomenon "3 Broker - 3 exchange rates". A system which delivers in Forward-Test at Broker 1 x trades and at Broker 2 y trades is going to deliver at backtest a totally different number of trades.
3. They work with an established spread in backtest
The spread each Broker has looks, quite often, completely different and is even swaying.
Therefor, I use a micro account and test my EA among live conditions.