[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Real-time trading Vs Simulation



PureBytes Links

Trading Reference Links

Hello Adrian,

There are two issues here, one being slippage, and the other the
validity of your equity curves.

If your equity curves consistently
blow up as soon as you walk them forward in realtime, Optimisation is a culprit,
especially if you ignore the other candidates to see
whether you chose the best of a robust collection of profitable
scenarios, rather than some profitable outlier.  Excluding wins or
anything else to make your equity curves pretty, are just forms of artificial optimisation which
will not increase the confidence about the future.  Running your systems in
the future is merely a final test to give you confidence that your
good looking equity curve is not a random phenomenon, like the
clustering of shotgun pellets.

The other issue is slippage.  If your data is only trade ticks, then
it's difficult to find the points of liquidity.  Your systems should
be able to "see" the depth in level II data, in order to come up with
a better representation of slippage.  Otherwise, you can always deduct
slippage for each trade.  In either case, it's foolish to ignore
slippage and not have it, as well as your commissions, factored into the
equity curve results.


Friday, August 1, 2003, 5:12:55 AM, you wrote:


AP> Hi,

AP> Most of the time real time results don't live up to simulation results
AP> as there is always a certain amount of artificial optimisation going on.
AP> My question is directed to those who have compared the two.  In your
AP> opinion what is the best method of making simulated results look similar
AP> to real-time results?  There are many adjustments one can make but does
AP> anyone have any direct experience in what they find is best?  For
AP> example we could make a RINA Index type adjustment, or we could just
AP> exclude wins, or increase slippage etc.  If people have strong views on
AP> this, what are your suggestions for the extent we should degrade testing
AP> results to see if they have a fair chance of holding up in real time?
AP> Assuming of course we already have what we believe is a robust trading
AP> idea before costs.

AP> Regards,
AP> Adrian