Backtesting with bad data. What am I doing wrong?

I have two MT4 platforms, one for live trading, and one for backtesting, but both have historical data loaded into it. I have followed THIS guide to getting old price data for backtesting. I am pretty sure I did everything correctly, but when I backtest between the two platforms I get different results on my trading. This obviously not good, how can I trust my backtest results if the data is bad to begin with?

I have attached two charts, one from each terminal running the same system on the same dates with the same parameters, etc. The "Report" tab is open, look at how different the numbers are (profit factor, etc). On the chart itself you can see some price discrepancy between the two, where the trades play out differently.

What do you recommend? How come the data is so different?

When you say you have two platforms, are you saying you have one on Aplari and one on IBFX or, you have both on IBFX one is a demo you use for backtesting and the second on a Live account?

Or do you have two platforms on the same exact account and one you use for backtesting and the second is your Live platform?

Both have shitty inaccurate data, unless you are testing on 1m data. To test accurately in small timeframes you need to use tick data which is one heck of a pain to get working.

Do this:
Open an MT4 account with any broker (you can also use an existing demo acc, etc).
Take your MT4 folder, copy it, and paste it somewhere that doesn't confuse you and is completely seperate from your main trading MT4.
Login on the copied MT4, make sure settings are saved on close and change the server/account login data. This should fuck up MT4 and you won't be able to get quotes or place trades and you should only be able to open charts with open offline. Be sure to close/open MT4 to make sure it does not automatically login or connect to the server.
Close MT4, delete your chart history (EURUSD60.hst, EURUSD5.hst, etc).
Now download accurate 1m data. You need to download a file that you can import, don't use the download button from MT4 client. Make sure you use a site with good data (preferably the broker that you use). Make sure it's 1M data. Not tick data, not 5M data.
Go to History Center. It should be empty (because you deleted all your chart history). Double click USD and double click 1M, then import the data you just downloaded.
Now use the PeriodConverter script (should come with MT4 install) to change the 1M chart (use open offline to open your pair 1M chart). You can convert to 5 min, 60 mins, 1440 mins, etc.
Now run a backtest, use generate ticks and it will generate a 90% accurate history. How accurate this is exactly can't really be said, but it's much better then using a normal MT4 install because it completely screws up all history whenever you connect to a server to save diskspace and such. NEVER let your tester MT4 installation login to your account or connect to the server, else you can do this all over again (removing all history, importing history, converting history).

Hope this made it clear.

I have two Oanda MT4 platforms. Should be same feed/data.

I am testing 1M data. I am using pending orders that are well outside of 1m data range, so I do not care if price is off by a pip or two, but the restults should not be this dramatically different.

I will look into your way of importing data, hopefully that will work. Thanks.

We are not talking about a pip or 2 in difference. It's something like 1M bars with only open or close. Spikes are eliminated, any movement from open to close is eliminated. You can still generate decent data in say 1H, but testing 1M or 5M is just useless and it's going to give results that have zero reliability.

Thank you, I did not know that this is how it worked. Let me explain a little more about how my EA works, but keep in mind I am new to this so this is likely not a conventional way to do things.

My EA opens orders once a week, then modifies them on a weekly basis. They way I set that up is that on Monday morning at time 0000, I first delete all pending orders, since they didnt get triggered and are no longer valid. 1 minute later I modify active orders, 1 minute after that I open new pending orders for the week. I have tested several different ways to go about this, but this allowed each trade internvention to run separate and not interfere with other things. (for ex I have to separate my deletePendingOrder() from placePendingOrder() or else I get a million order opened/deleted/opened/deleted/etc. I separated them by 1 minute and it runs).


Sooo, all that is to say that the price flucuations for every 1M are not THAT important to me, I just need the platform to run on 1M so that it runs my function on that minute. If I run it on 5M data or higher, it misses the function call at 0001, etc. I know I am probably just programming this in the most awkward way, but the differences in the EA results show that price is different on a 4hr+ basis, enough to change performance quite dramatically between platforms.

Now I guess I probably need to just reset all data and do like you said, so that will be my next attempt. I am just worried about bad data. Prior to me opening my 2nd platform, I did everything on 1 platform and thought life was great. Now I make a separate platform for backtesting and find out several of my systems are failing on it. Kinda worries me.