## Portfolio-Level Simulation Tools

Discussions about Money Management and Risk Control.
Roger Rines
Roundtable Knight
Posts: 1964
Joined: Wed Oct 06, 2004 10:52 am
Location: San Jose, CA

### Portfolio-Level Simulation Tools

viewtopic.php?t=1926

The topic of distributing money to each of the systems in a trading plan was especially interesting.

In the past, I have found that the draw down to gain results can be reduced if the distribution of funds in a trading program isn't equal across all the systems in a trading plan. In addition I've found that within each system's portfolio the percentage of draw down to gain within that system can be improved if the percentage of funds made available to each market in the system isn't the same.

While simple and crude approaches have been encouraging me to look in this direction, being able to understand how this would work using the results from large-scale Monte Carlo style simulations would certainly help me believe in what I think I'm seeing.

I've found tools like:
MvoPlus - A mean Variance Optimizer

Portfolio MCS - Portfolio risk analysis software using Monte Carlo simulation for systematic traders using TradeStation

Each of these products provide some answers and have been reinforcing my belief of what I've been finding with my simpleton approaches. However none of these tools allow the user to work with all the various aspects needed for understanding the larger picture. While I could dream that a blend of the above packages might be a better approach for developing useful information, I have yet to see any software that would be better than what is listed above.

Are there better tools available for doing this kind of simulation work?

sluggo
Roundtable Knight
Posts: 2986
Joined: Fri Jun 11, 2004 2:50 pm
Suppose you have three systems (S1, S2, S3) that each trade three markets (Ma, Mb, Mc). Your approach creates twelve new optimizable parameters (p1 thru p12) that are the weights of the individual markets within the systems, and the weights of the systems themselves:

Code: Select all

``````System S1 trades (p1*Ma) and (p2*Mb) and (p3*Mc)
System S2 trades (p4*Ma) and (p5*Mb) and (p6*Mc)
System S3 trades (p7*Ma) and (p8*Mb) and (p9*Mc)

You blend the systems together according to (p10*S1) + (p11*S2) + (P12*S3)``````
It seems to me that granting yourself another twelve degrees of freedom to play with and optimize, jolly well should increase whatever figure of merit you're seeking to maximize.

Finding the optimum point in a 12-dimensional space will be a time consuming task; if you only investivage 3 settings per parameter that's stilll 3^12 = 531,441 simulations. Very likely you'll want to use some kind of stochastic or probabilistic algorithm, like genetic optimization or simulated annealing, to get a near-opimum answer in finite runtime.

Roger Rines
Roundtable Knight
Posts: 1964
Joined: Wed Oct 06, 2004 10:52 am
Location: San Jose, CA
Hello Sluggo,

Your scenario does capture some of the complexity of what I'm attempting, and may be the reason why I haven't found any software that will help me get the data I think is needed to prove my idea has merit.

As for the calculation versus a Monte Carlo simulation, proven calculations certainly take a lot of effort out of the process. However, from my perspective that approach is reserved to those who have expert math experience in field of asset allocation and advanced statistics. Unfortunately I don't have either background, which is why I'm looking for the brute-force approach of a Monte Carlo simulation that generates visual outputs and data tables.

On the thread that I referenced above, c.f. made mention that there might be tools available, so if there are some better solutions, I'm hoping this thread will surface them. If nothing is available, then I'll attempt to write something. I find programming easy when I know the science needed for what I'm doing, but this task is looking for more than bit-twiddling prowess, meaning a collaboration with competence in advance statistics and asset allocation will be required.

Thank you for responding.