One of the things I've fooled around with, is simultaneously trading several extremely-similar systems at once. I wanted to see whether small diversification in entry date/price and exit date/price, would smooth out the equity curve.In another topic area, kmulford wrote:Using two or more different trading systems, each with positive expectancy, should provide a trader with more opportunities to trade and, thus, realize on the positive expectancy of the systems.
Has anybody had any experiences they would like to share regarding the construction, testing and trading of multiple systems methods? How does one go about picking the systems to trade in conjunction with one another? What characteristics of the second system are most useful in complementing (offsetting) the first? Long term and short term?
How many systems can/should be optimally traded?
This is stuff like (A. enter on the open the day after PGO exceeds 3.0), (B. enter at market-on-close on the day whose close pushes PGO beyond 3.0), (C. enter on a limit order 5 ticks below the close of the day that pushed PGO to 3.0) and so forth. Trade 1/3 of your bankroll on system A, 1/3rd on system B, 1/3rd on system C. Spread your entries around a little bit.
Here's a concrete example with a surprising result. Take a "Standard deviation bands" system whose time window is X days. Add to it a decrementing trailing stop (such as George Pruitt suggested for Trendchannel and Aberration a few issues ago). I picked an exponential moving average whose #days decrements by one for every day the position is held. On initiation day (#days=0), it's the X day EMA. Four days after initiation (#days=4), it's the (X-4) day EMA. And so forth.
This modified system had about the same performance ratios as the original without the trailing stops. (Sharpe, MAR, Lake, RRR, and so forth). Changing the exit, by itself, didn't improve the system.
However when these two variants are blended together in a 50-50 ratio (half your bankroll on the std dev bands without decrementing trailstop, the other half on std dev bands with decrementing trailstop), the performance measurements shot upwards. Apparently for the time periods I chose and the markets I chose and the system I chose, exiting half of your position when hitting the decrementing EMA, and letting the other half of your position ride onward to the final exit signal, boosted profits significantly. All the measures shot up, the one I remember right at the moment was that MAR increased by 1.24X (MAR of 50-50 blend = 1.24 * (MAR of original)).
This might be a programming error (I haven't double and triple checked the results yet) and it might be an accidental anomaly that's good for this one well-chosen example but lousy in general. But I doubt it. I kinda think it's often helpful to diversify entries and exits, even on the same base system.