Error: TestGenerator: unmatched data error (volume limit 97 at 2008.09.16 15:25 exceeded)

[PipCop]

Ever get this error? This means you have invalid data in MetaTrader. Follow these directions to resolve it:

In MetaTrader, click Tools > History Center > your desired currency pair > your desired time frame > Download

Do this for each currency and timeframe you backtest with. When you're done, click Close and try again.

[BarrowBoy]

This does indeed get you data - but only MetaTrader history data.
It wont give matching results to any single-broker history.
Even single-broker demo accounts dont exactly match their live (i.e. real money) feeds.
The differences may not be great but are certainly big enough to upset an EA...

You can have an EA that is 'amazing' on a MetaTrader account (or MT history), good on a brokers demo account, but loses on the brokers live account

Factors involved include:-
- Different OHLC
- Tick count
- Tick frequency
- Spread changes
- Server reponse times (particularly at busy times)

[PipCop]

So what you're saying is that to have real results on any of these EAs, we'd have to test on a live account?

Scary.

[BarrowBoy]

Whatever is done on MT or broker demo accounts, you should back-test & re-optimize on the brokers live data (expect slightly different results).
Live trade on minimum lot settings & watch for any statistical departure from the PF or win/loss % seen in back-testing.
EA's that enter trades mid-bar can live-trade waaaay different from back-test results - and not usually in a good way

[PipCop]

Correct me if I am wrong - to get live data from the broker I simply need to backtest in MetaTrader by using my live account and history data, correct?

[BarrowBoy]

Yes, you can test as far back as account opening day without any trouble.
To go back further, get data off the chart - see notes file that came with the TrendRider EA

[PipCop]

Why does this seem to happen all of the sudden? For example, I was testing an EA and have made several passes back to 01-01-2008, and then after making a small change I try to test it again and now I get that error? That doesn't make any sense to me...

[BarrowBoy]

In spite of the Model being set to 'every tick ' it isnt!
It is an extremely good approximation, even at M1 level
So... when the data for timeframes >M1 is regenerated for a backtest or optimization, occasional errors occur in reading or writing to disk, the error reveals a mismatch between your M1 data and tested timeframe.
The error can be in a 'new' incoming tick where OHLC data is wrong or where the volume numbers on the lower timeframe, when added up, don't match the volume numbers of the higher timeframe.
You may find a different date range is OK, or it may fix itself if a test for another EA/pair is doen & the original test redone.
If the results are in reasonable sync with expected, I wouldnt worry about it - especially if 'Modelling Quality' on a back-test is 90%-ish
Just keep in mind that any back-test is an approximation - especially for EA's that open mid-bar

[PipCop]

That's a little scary. I need to find something that uses exact data... any suggestions?

[BarrowBoy]

Regret to say that exact data is hard to come by!
Dont want to shatter any illusions but any concept of scientific accuracy or consistency is just not present in what is a staggeringly huge and very fluid market.
There is no central clearing portal for Forex, so at any given moment, the price of a pair can vary by up to 20 pips between feeds, more after big news hits.
You have different banking groups supplying wholesale rates, other portals offering a choice of wholesalers prices to brokers, some brokers are 'straight through', others operate an internal market (aka 'dealing desk'), some are a mixture depending on market conditions.
So against all that, exact data isnt going to happen, whether for live or historical purposes.
Trading and (successful) EA writing is about managing chaos!!!