Issue: Real-Time Vs Backtesting

[pstrusi]

I'm pretty aware about your explanation-warning in the help guide about the typical discrepancies when the same strategy works different depending if it's under backtest or real-time. I've been working in a strategy that is having this problem. The Strategy is pretty simple with basic calculations BUT it must work with two DataSeries simultaneously: Bid and Ask ( in this case the $EURUSD ) . Let me give you a simple description of its logic:

if (BarsInProgress == 0)
			{	
				// Simple math calculations	and get a signal 
			}
if (BarsInProgress == 1)
			{	
				// Simple math calculations	and get a signal
			}
// Evaluate signals from DataSeries 0 and DataSeries 1. 
// If the same Signal then set an order needed otherwise wait for new OnBarUpdate

When I backtest this Strategy, the calculations, signals, evaluations and executions match the expected behavior.
When I run it in Real-Time, that's another story. I've set all kind of evaluations to research it. such as: Output windows with key variable prints to see where are the differences.
After thorough research for possible causes of these discrepancies, I can tell:

DISCARD CAUSES:
- Executions filling prices or fillings quantities
- Both processes ( Backtest or Real-Time ) are under Tick-by-Tick

POSSIBLE CAUSE
- Bad filtering of data. It seems that there must be some trouble receiving, filtering and processing events that come pretty fast ( few milliseconds of difference )
The most clear example that I can give is this:
for the same event, registered at the same time, in the backtest there's just 1 but in Real-Time there're more, sometimes the double or more, without any logic.

This issue has stopped all my work, and I'd like some help, new ideas, suggestions...etc

Thanks in advance

[NinjaTrader_PatrickH]

Hello pstrusi,

May we test the strategy or a version that reproduces the matter?
Can you also detail some specifics on what is occurring?

[pstrusi]

Hi Patrick, thanks for your attention.

This is simple:
- As you can see in the code, I filter the event for DataSeries: bid or ask;
- it's done the simple calculations to that event to get a signal value for that DataSeries;
- the Strategy evaluates the very last current signal values for bid and ask
- If signals are equal , an order is submitted, otherwise return

In order to monitoring how events are occurring and their calculations, I print in the Output windows the main variables with the exact time, so I can see if all events are in sync with that chart. Here is when I realize that for example, without any occurence, I have printouts that don't appears in the chart, basically price variation that don't exist.

I'm trying to do some trials looking for filtering events that don't belong to index 0 or 1 throught this:

if ( BarsInProgress != 0 && BarsInProgress != 1 )
			{
				return;
			}

So far, printouts are matching. If this solves the issue, this means logically that OnBarUpdate is triggering with events out of index 0 or 1.

What do you think?

[pstrusi]

I've found a solution for the issue, filtering OnBarUpdates events that don't belong to DataSeries 0 and 1 ( Last and Ask in my example ) of the Strategy, as I wrote in the earlier post.
in spite of that in Forex, Last=Bid, OnBarUpdates triggers with any Last, Bid, or Ask coming tick. I supposed wrongly that since Last=Bid it should be just one OnBarUpdate event and it's not during a Real-Time.