Quote:
Originally Posted by omelette
newdigital, could you be more specific about why trading at the Ask and Bid (bar 0) is, 'cheating' so to speak?
Attached are the equity curves of one of my Scalpers (10pip, 1:1 r/r) that shows spectacular results in Backtests but performs absymally when forward tested so it clearly is not profitable.
What I want to know is why exactly, so I can stop wasting time on stuff that does not work in reality.
Also, one of the 'tell-tale' sign that an EA's results are bogus seems to be what happens during the last 6-9 months of the backtest - usually the equity curve turns down abruptly. This occurs with all of my 'profitable' EA's unfortunately and I have yet to hear a convincing reason as to why this happens - more specifically, why does an inherently unprofitable EA seem very profitable up until the final months of the test? Anyone who has been using Metatrader for a reasonable period will have noticed this last '6-9 month' problem - note that 1 year ago, you had the same poor results for the last 6-9 months even though this period now shows up as profitable in backtests.....
The idea that tick data gathered realtime over the poorly-performing interval is the reason doesn't stand up as it also occurs when Metatrader is freshly installed. Could it be that all the historical data, apart form the last 6-9 months, that the Brokers supplies is 'artificial', generated in much the same way that the backtester interpolates its tick data? Also, I found that running the same EA, settings etc. on different broker platforms produces almost identical results so the all seem to be supplying the same crappy historical data...
I should have more on this soon as I am in the process of getting my hands on 'quality' historical data.....
|
I am not talking about 'cheating'. That is why I moved those posts to this thread to 'take it easy'.
Nothing wrong in coding on open bar. I know that some coders are coding and there is some financial funds using EA coded on open bar also.
But there are some particularities.
There are very different results for backtesting and forward testing for EAs coded on open bar, or on high/low of the close bar. Sorry I am not a coder. I just experimented for many years and made some conclusion.
Do you know MaChannel by Codersguru? If I remember it was coded on high/low of the close bar (some ma mode or something - don't remember). So, backtesting results do not match forward testing. I don't know why. May be it is something with strategy tester in Metatrader ...
For example I am having very good results with this EA for North Finance broker. Forward testing. Backtesting on same period are completely different. And Codersguru used this EA for last Metaquotes contest and EA did not win: I used North Finance broker and metaquotes used IBFX broker for last contest. And high/low data is very different for NF and IBFX sometimes ... And we got opposite results: IBFX contest was lost and I got very good results for same period of time (I repeated the trades with NF broker as Codersguru and Igorad asked me about it during that time).
If somebody is selling EA based on backtesting results and not telling the people how it was coded (open bar, high low of the bar) so it is not good.
The other example is Firebird. As I see it was coded on close bar, high/low of the close bar and open bar (simultaniously). i am not a coder so it is difficult for me to understand how this EA is selecting the order and which type (close bar, open bar) should be opened in which day of the week. But I know that forward testing results are very diferent from backtesting.
The other case is Brainwashing EAs. Those versions are using iTrend indicator as icustom. And this indicator is showing differently with different broker. May be because of that, may be some other cases, but backtesting and forward testing of this EA is very different. EA was coded on close bar only.
It is nothing with losses. Simple diferently. Not the same (backtesting and forward testing).
The other cases are MTF EAs. There are 2 MTF EAs in elite section. All of them were coded on close bar so that's fine. But if MTF EA was coded on open bar concerning more highier timeframe so the backtesting will not match forward testing.
I only wanted to say that it is very difficult to understand about EAs performance based on backtesting results only because backtesting is not telling the true in many cases. And if the seller is not telling us some particularities (without disclosing the system of course) so backtesting is nothing in this case.
Nothing wrong on open bar coding. I know one coder (he is not a member of our forum yet - he is posting his EAs on metaquotes codebase) and he is coding on open bar only. And there are some very very good ideas about tick indicators, tick MACD, tick MA and so on: i mean indicators to trade on open bar by ticks. And in this case it may be very very good scalpers and other EAs based on open bar.
But it should be mentioned always: coded on close bar, open bar, MTF with open bar for highier timeframe, high/low of the bar and so on.
Otherwise people will be confused and will consider backtesting as useless for all EAs. But backtesting is not useless if we know how it was coded.
And there is some other case.: gaps in data. Especially during the news events. It is not often case but today I found two gaps for august on GBPJPY M1 data when I was checking some indicators for re-paintng. It may be also the reason why backtesting is not the same with forward testing.
But generally backtesting must be the same with forward testing in most of the cases. Shoud be ...