unexplained buying frenzie

New comment
 

This sounds like a server issue, but there's enough inconsistencies, I wonder if anyone else has seen something like this.

InterBank FX server. Demo account. Running several EAs, but only two of them (but two, not just one) had this happen:
at 2006.25.06 0521-0524 server time, they each bought GBPUSD 20 times closing them almost immediately for a rash of 1 to 8 pip losses. The EAs were running on the H1 chart. Other EAs were on H1, 15m, 5m charts, and did not exhibit this behavior. Both EAs had open positions at the time of the flurry. Unsure about the other EAs, but estimate they did not have open positions at the time of the flurry. This region is in the upper 'wick' of that bar on the chart. The market at that point was well away from any signal associated with either EA. Each EA had very different logic. Other EAs running on that pair didn't have similar buying frenzies.

So, even if this was a data problem, what would cause this? Is it possible for an EA to be 'credited' with an order by some other means? Meaning, if an error occurs, can an EA make a trade that has nothing to do with the internal mechanism? These orders had the EA's comments in them, so they 'belong' to that EA, but yet no-where near their triggering mechanism. Is it possible for a data error to trigger the EA but the evidence not appear on the chart? Though this was on a demo, understanding how this could happen would go a long way to allaying fears that this is possible in a live account.

Thanks
-Matt

 
Please provide us with codes of these wrong EAs.
 
First, I'd like to know how such an error is even possible. Pick any EA design. Mine aren't that important. Can you design an EA that would place 20 orders in 4 minutes immediately closing them without designing them to do that on purpose? What data is allowed to come off a server that is not refelected in the history? Can the ask price be allowed to fluctuate so wildely that it would cause an EA to fire, but the bid-line wouldn't reflect a price change? Or is ask ever allowed to define the Close[0] for an open bar? Or how would those be closed? You can't even get a 1-trade-per-bar EA to close that fast, or any crossover always in method for that matter, as the closing should be accompanied by a sell. The point is there is a lump of bad data, and I'm trying to understand if anyone else saw it, what sort of bad data is possible, how to possibly trap for such errors, in general, independant of the specific EAs that happened to detect the bad data this time.

There's at least one other report at sbfx of such a frenzy associated with data stream. ibfx has not responded to queries on their data during that period, which is typical. Demos don't count for much with them it seems.
 

The matter is that we haven't ever faced such a problem by now. To reconstruct your situation, we would like to have the log (the "Journal" tab in the "Terminal" window) and the codes of those wrong EAs. Could you provide us with this?

New comment