Usefulness of Monte Carlo
Posted: Fri Feb 12, 2010 10:32 am
For some reason I've been pondering the usefulness of the MC simulation lately. It seems to me that if a person developed the ultimate robust system that has a realistic jagged equity curve with lots of long and deep drawdowns, the MC stats for that system will look terrible, but the system will likely perform as tested.
On the other hand, the person that develops the ultimate curve fit system with a straight line equity curve to the moon will have great MC stats, but the system that will not hold up in the future.
So whats the point/value?
As an alternative, I wonder if using monte carlo process to create synthetic data from real data would be useful? You could take fairly large slices across the same time frame for each instrument in your portfolio (to retain both serial & intermarket correlations) and then chain the data together in the same order for each instrument for 1,000 years. Then you can run your system on that to see how it holds up on "out of sample" data. This could be interesting...although fake data is not ideal...
I know these topics have been discussed in the past, i'm just rambling...
On the other hand, the person that develops the ultimate curve fit system with a straight line equity curve to the moon will have great MC stats, but the system that will not hold up in the future.
So whats the point/value?
As an alternative, I wonder if using monte carlo process to create synthetic data from real data would be useful? You could take fairly large slices across the same time frame for each instrument in your portfolio (to retain both serial & intermarket correlations) and then chain the data together in the same order for each instrument for 1,000 years. Then you can run your system on that to see how it holds up on "out of sample" data. This could be interesting...although fake data is not ideal...
I know these topics have been discussed in the past, i'm just rambling...