|
PureBytes Links
Trading Reference Links
|
Dans un courrier daté du 18/09/98 12:41:05 , vous avez écrit :
<<
Hello again Pierre,
What you say makes sense. My approach differs concerning
relationships in "equity curves" though.
Basically your approach seeks to eradicate the idea of "maximum
excursion" from the normal result. I see this a lot in systems
designed by NN fellows. By eliminating undue volatility to the up
or downside, the equity is smoothed and (hopefully) ascendant.
>>
I do not exactly eradicate.
I think that I must live with it, so I must take this in account in a
realistic way.
Increasing max DD is a simple method that works if you compare it to the
theoretical profit obtained at the time where max DD occured.
<<
However, there *are* ways to retain *good* volatility and lose
the bad. Ergo you are left with a marginal upward slope in EQ
during quiet times followed by sharp increases in equity during
volatile times. When designing a system this is always the
goal I am shooting for. Granted this is tough to do mechanically
unless you have the capability of creating and testing futures/
stocks/options mixes. But it can be done.
>>
Right, but complicated to implement and backtest.
<<
As an example, let's say your NN say's "No trade" here due to
projected volatility beyond acceptable parameters of the system.
Very often, such a market call is derived from an estimation against
acceptable stop-loss protocols. On the other hand my system would
say "Trade Like Normal, and defend with options".
>>
We do not use NN, but neurofuzzy logic.
The system are always in the market, no stop loss (stop and reverse in fact).
So, we do not care of stop loss, that I have always found worse than nothing.
<<
Giving this some thought, you actually can do this using the straight
underlying using your system, provided that you have an extremely
large equity base. In this case, a maximum market excursion against
the system could still logically be deemed within your 1/20th limit.
Given these cases and scenarios, I am curious as to your opinion of
any marked advantages shown by neural nets over any linear regressive
(and much simpler) process.
regards,
Walt Downs
CIS Trading
http://cistrader.com
>>
The NN advantage is not involved in this , because the system that we develop
are very similar to the classical systems. The only difference is in the
complexity of the rule decision tree.
But basically, it's like a moving average crossover (that features only 2
rules), with the difference that hundred or thousands of rules may be active
even with 3 or 4 indicators.
The goal being to increase the net profit /(2.5*max dd+ deposit) function (we
have others) regardless any external consideration.
In fact, the rules are so complicated that it's impossible to figure out what
they globally exactly do (at least for my small brain), so I try to be
satisfied with the results obtained on unseen data, without trying to
understand.
A reason more why I' enforced to stick to the max DD rule that I test on a
huge intraday database (thousands of trades).
Sincerely,
Pierre Orphelin
www.sirtrade.com
|