|
PureBytes Links
Trading Reference Links
|
> A possible question is: what can be done fairly soon, and at
> reasonable expense? Can we prioritize the desires?
Aww, we have to be reasonable?? :-)
I'd say running N systems on M markets is a must.
That's what portfolio trading IS.
Good stats, at least minimal ones, are a must,
else you can't tell which portfolio to trade.
Portfolio-level money management (e.g. risking X% of the
**entire account** on one trade) is a VERY high want.
Handling multiple timeframes is a VERY high want.
Linking parameter values (e.g. using the same/different params
on all markets, or in several systems) is a want.
OOS testing would be great but not necessary.
You have to have some kind of "goodness" measure for optimizations.
Sharpe is the best IMHO, and I could survive with only that.
A pre-set list of other measures would be next best,
followed by user-specifiable.
One thing a user might specify in the "goodness" measure is "choose
the set that has the highest *average* Sharpe when you include its
neighbors" -- i.e. don't accept an isolated spike, but require that
all nearby parameter sets be pretty good too. This would be an
excellent option to build into the optimization engine.
A barely-acceptable subset, for what I want anyway, would be to
support 1 system on M markets, all using the same parameters, and
support only one timeframe. If I have to, I can combine the N
systems together after the fact, though that kills the portfolio-
level money management. But I suspect N systems on M markets would
be a fairly simple extension of the basic model, if you design it
that way from the start.
BTW a tool already exists that does most or all this -- Trading
Recipes. But it has no charting module and (so far) no intraday or
realtime abilities. I want to be able to SEE my system as I develop
it, AND execute it on the same platform, the way I can in TS.
Basically I want TS with portfolio capabilities.
Gary
|