Appreciate any comments on this
Posted: Wed Dec 21, 2011 7:33 am
Hi
I am building a portfolio of systems and would appreciate any comments. Testing has been run on 3 different component systems, which have then been optimised (hopefully in a robust way) and put together into a portfolio that seems to backtest quite ok by my standards.
Quick system descriptions:
1. A volatility breakout system
2. A moving average system with profit targets
3. A lowest low/highest high of last X days type system
Two of the systems trade futures long and short, and the last stocks long only. Entry day retracement has been set to 100%. All my other global parameter settings are realistic as far as I have been able to determine though my own trading experience and through consultations with a "live" systems trader with whom I share an office. I have chosen not to include any dividends or earned interest in my testing calculations.
The systems are put together aproxximately even in terms of weighting. The systems all works ok over a wide range of parameter settings.
The CAGR is about 20% and the Max DD is about 15%.
My initial test period incorporates both bull and bear markets in stocks and commodities alike, as I wanted to include lots of different market conditions in the in-sample period. Out of sample testing has started but has so far been limited as I am in the process of setting up new data feeds. So far so good.
My primary concern is that the system seems to work extraordinarily well in periods with strong bullish trends in the stock market (no surprise there), however, I would have liked to incorporate an element that may perform quite badly in bull markets but fairly well in bear markets to smooth these downturns better. The positive expectation would'nt have to be all that to be useful to me. I have experimented with a large number of systems to achieve this but I am coming up short.
My question is, what is your experience in trying to find elements like this - that is short biased systems that have a positive expectation - do they in fact exist - or should I rather go the route of testing further on a larger portfolio of instruments, test for historical correlation among these instruments and cherry pick the ones that historically do not correlate that much (I should probably widen the portfolio of instruments anyway).
Ok...Any comments or suggestions are much appreciated.
I am building a portfolio of systems and would appreciate any comments. Testing has been run on 3 different component systems, which have then been optimised (hopefully in a robust way) and put together into a portfolio that seems to backtest quite ok by my standards.
Quick system descriptions:
1. A volatility breakout system
2. A moving average system with profit targets
3. A lowest low/highest high of last X days type system
Two of the systems trade futures long and short, and the last stocks long only. Entry day retracement has been set to 100%. All my other global parameter settings are realistic as far as I have been able to determine though my own trading experience and through consultations with a "live" systems trader with whom I share an office. I have chosen not to include any dividends or earned interest in my testing calculations.
The systems are put together aproxximately even in terms of weighting. The systems all works ok over a wide range of parameter settings.
The CAGR is about 20% and the Max DD is about 15%.
My initial test period incorporates both bull and bear markets in stocks and commodities alike, as I wanted to include lots of different market conditions in the in-sample period. Out of sample testing has started but has so far been limited as I am in the process of setting up new data feeds. So far so good.
My primary concern is that the system seems to work extraordinarily well in periods with strong bullish trends in the stock market (no surprise there), however, I would have liked to incorporate an element that may perform quite badly in bull markets but fairly well in bear markets to smooth these downturns better. The positive expectation would'nt have to be all that to be useful to me. I have experimented with a large number of systems to achieve this but I am coming up short.
My question is, what is your experience in trying to find elements like this - that is short biased systems that have a positive expectation - do they in fact exist - or should I rather go the route of testing further on a larger portfolio of instruments, test for historical correlation among these instruments and cherry pick the ones that historically do not correlate that much (I should probably widen the portfolio of instruments anyway).
Ok...Any comments or suggestions are much appreciated.