Ive got a fairly robust system - 2 parameters - test it over 25 years and 90 markets and you get an MAR of 0.9 - thats not bad IMHO for 2 parameters
However its run raw with no risk reduction blocks - i.e. correlated units / thermal / group risks
When I add in these blocks it simply craps out on performance - down to 0.5-0.75 MAR
There are some systems I have where risk reduction seemingly improves performance but its noticeable on these very robust systems it doesnt - of course the odd combination of risk reduction greatly enhances performance but not in a smooth parmater spacing ?
I'm guessing run it raw on half the amount of capital
Just see if anyone else had same experiences or any thoughts
Best
C
What would you do ?
Get two identical copies of Blox software on two identical computers and supply them with identical price data.
On computer A, run the unmodified system and NOTE ITS POSITIONS, SIZES, AND NEW ORDERS for the last day.
On computer B, run the modified system with all the blocks added in. NOTE ITS POSITIONS, SIZES, AND NEW ORDERS for the last day.
Since you are convinced that B is inferior to A, the difference between A and B is the "advantage" or the "superiority" of A, compared to B.
Measure the difference (i.e. subtract the positionsizes: (sizeA - sizeB)) between A and B. This is the superiority of A; this is its edge.
Now here's the cute part: improve A's results by adding in some extra amount of superiority.
Start: sizeA
Superiority = (sizeA - sizeB)
Finish: sizeA + (some constant)*Superiority
Which is: sizeA + (some constant)*(sizeA - sizeB)
If you are very timid, set (some constant) to be a very tiny number like 0.01. If you are confident and bold, set (some constant) to a bigger fraction like 0.75 or 1.50.
Best of all, since this is a mechanical system trading idea, you can backtest it. You can try all possible values of (some constant) and see which one looks best to you.
Math aficionados will recognize this as "interpolation" along the straight line in N dimensional space (N = # portfolio instruments) between sizeA and sizeB. The interpolation parameter "(some constant)" controls your position along this line. When (some constant) = -1.0, you are at sizeB. When (some constant) = 0.0, you are at sizeA. I am suggesting that you extrapolate this line beyond the sizeA point, by choosing (some constant) > 0. Using the commonsense notion that if A is superior to B, then beyond-A is likely to be even more superior to B. Then backtest and find out.
On computer A, run the unmodified system and NOTE ITS POSITIONS, SIZES, AND NEW ORDERS for the last day.
On computer B, run the modified system with all the blocks added in. NOTE ITS POSITIONS, SIZES, AND NEW ORDERS for the last day.
Since you are convinced that B is inferior to A, the difference between A and B is the "advantage" or the "superiority" of A, compared to B.
Measure the difference (i.e. subtract the positionsizes: (sizeA - sizeB)) between A and B. This is the superiority of A; this is its edge.
Now here's the cute part: improve A's results by adding in some extra amount of superiority.
Start: sizeA
Superiority = (sizeA - sizeB)
Finish: sizeA + (some constant)*Superiority
Which is: sizeA + (some constant)*(sizeA - sizeB)
If you are very timid, set (some constant) to be a very tiny number like 0.01. If you are confident and bold, set (some constant) to a bigger fraction like 0.75 or 1.50.
Best of all, since this is a mechanical system trading idea, you can backtest it. You can try all possible values of (some constant) and see which one looks best to you.
Math aficionados will recognize this as "interpolation" along the straight line in N dimensional space (N = # portfolio instruments) between sizeA and sizeB. The interpolation parameter "(some constant)" controls your position along this line. When (some constant) = -1.0, you are at sizeB. When (some constant) = 0.0, you are at sizeA. I am suggesting that you extrapolate this line beyond the sizeA point, by choosing (some constant) > 0. Using the commonsense notion that if A is superior to B, then beyond-A is likely to be even more superior to B. Then backtest and find out.