Hi Beepsi,
I think the post leaves too many variables in question to perform a comparison. When you are using two different historical data sets as this test appears to be doing, you are going to get very marked differences – especially if making trading decisions on a tick-by-tick basis. From looking at the EA code, it appears that the first time one EA takes a trade that the other doesn’t, it will inevitably lead to a chain of out-of-sync trades which will cause the two equity graphs to diverge. If the purpose of the comparison is to see how the two EAs perform on two different platforms with two different sets of data, then it should at least ensure the trades are synchronised with some sort of common trigger – by MA, dates or the like.
If you are trying to assess which system to “trust”, the only way to achieve a level of confidence in your results is by trying to model your trading environment as accurately as possible. Many EAs that are tested on Dukascopy data do not necessarily produce the same results live due the fact that the broker’s live feed does not “match”. Data quality & network lag are major factors that will cause your results to vary with the same EA.
Here are some factors to consider when trying to get as accurate back-tests as possible:
– Does your broker’s data feed match with your test data? If they are markedly different, expect markedly different results for HFT/tick-based strategies.
– What is your ping to the broker’s server? Remember that back-testing assumes a ping of 0ms where this is not reasonable in reality. If you have a HFT trading strategy, you should code assume some delay when placing your trades in the back-tester.
– Is your strategy reliant on spread? If so, test with varying spread values to make sure the back-test results aren’t adversely impacted by small changes to the spread.
I’m sure there are other items that perhaps other users could comment on as well.
Hope this helps.
