> While using 10-level deep data will give different backtesting results
> (obviously), it should *not* be too much different if we apply the
> right transformation.  If all is well, having more samples for the
> depth should just reduce the variability of the our estimate for
> supply--demand, which 'should' translate into a reduced variability of
> a trading strategy (although some strategies are hyper-sensitive to
> even tiny fluctuations, but these ones are probably not worth trading
> anyway).
>
> Right now, we track a normalized difference between the bid and ask
> depths.  If we have a depth of N, all we need to do is find the best
> scaling factor that keeps our estimate approximately the same as a
> function of N.  This could be done empirically for a few days to find
> an appropriate scaling.  This would be impossible to check on our old
> dataset, but I believe it should work well enough for using our old
> dataset.
>
> If someone has the full recorded data for a few 10-deep days, we could
> simply try this out; that is, plot the distributions of our
> 'normalized difference' estimate as a function of N.  If this doesn't
> work, I suspect that our assumption of collapsing 2N numbers (from the
> bid and ask) to a single number is not well justified ...

You are assuming that the difference between the 5-level deep balances
and the 10-level deep balances is just variability which can be
compensated by a scaling factor. My guess is that there is more to it.
However, I agree that it would be useful to plot both side by side.
I'll modify my code to record both 5-level deep and 10-level deep
balances for a day and will post the results.

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups 
"JBookTrader" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to 
[email protected]
For more options, visit this group at 
http://groups.google.com/group/jbooktrader?hl=en
-~----------~----~----~----~------~----~------~--~---

Reply via email to