|
PureBytes Links
Trading Reference Links
|
> Just for kicks, why would not WORST CASE DRAWDOWN
> to date be something to base the computation on rather
> than average loss or total loss in a string.
Might work. Let us know what you find!
> That's what the Monte Carlo does. It assumes there is no
> correlation between the results of one trade and the next and that
> your realtime historical trade order was just a random chance
> (might not be true for all systems but it's true for mine).
That's my biggest gripe about MC testing. For my system at least,
and I suspect for many/most, the trade order is NOT random chance.
E.g. my system has been behaving VERY differently for the last 3
months than it did in the 2 years of testing before that. (And I
**love** it!! :-) The market is acting fundamentally differently
than it did before -- more volatility, etc -- and that affects the
trades that happen during this period.
I think many systems have "memory" like this. E.g. take a simple
breakout system. It's going to tend to have a string of losses
(fakeouts) followed by a big win when the market finally trends.
That's not a random series. MC testing of that will produce
questionable results that do not necessarily reflect reality.
Gary
|