[R-SIG-Finance] blotter update MM & weights without recalculation all

[Immanuel]

Hello Brian,

thanks for helping to clear thinks up!
This is how I understand things now.

I create an strategy object using quantstrat and use blotter to manage
portfolios and accounts.
I ran the strategy on a fix unit basis and can then obtain (from
blotter) a list of trades for each strategy & underlying.
Using this trades I can calculate position sizing rules (optimal f)
independent for each strategy and then rerun the strategies
using costumized position sizing rules for each strategy.

Now I pull the returns for each strategy & underlying from blotter, and
calculate the optimal weights (using portfolio.optim(),
the weights tell me how much money gets allocated to each individual
strategy.)
A weighted sum over all strategy returns will now give me the return for
the portfolio of trading strategies.
If I want to accomplish this in blotter I would have to adjust the
position sizing rules for each strategy to the weights.
Finally I would run the strategies again. The account results would now
be the results of the portfolio of strategies using optimal position
sizing and optimal allocation to each strategy. (well, optimal in the
sens how I decided to optimize).

> I'll note that Markowitz optimization is not likely what you want.  It
> will penalize positive skew and a long right tail on the distribution,
> which are precisely the features you would want to emphasize in a
> portfolio of trading strategies or products traded in a single strategy.
>
I was planing to calculate the covariances that go into the Markowitz
optimization only on the pairwise loosing returns, since there is no point
in penalizing variation at the points where both strategies make profit.
I want to penalize variation in pairwise loosing returns especially
large variations.
I'm aware that the Markowitz optimization might not be the right
approximation for my real objective , but for educational reasons
I have to use Markowitz first bevor I move on to other methods. (I'm
planing to look into LSPM)

regards,
Immanuel