Risk/return in the work I've done is the expected return divided by the expected maximum drawdown. My work has indicated that using a diverse set of systems improves this ratio. I've also found that even though saturation can be reached, further improvement can be achieved by using a treed apporach. When I have some time I'll post some data and examples for further discussion.
does not tell you anything about the smoothness of the equity curve, which is the only interesting thing for a trader. like profit factor and others.
Almost. He is defining risk/reward as "expected gain" versus "expected loss". Expectation of the system is a bit different. I use the term 'expected' loosely given that no trader actually takes the time to model their trades as a random distribution -- nor do I even think that is possible (unless you believe that either your system is random or the market is, in which case, why bother even trading?)...so 'expected' doesn't really make sense in the mathematical sense. To me this is an incomplete 'risk' versus 'reward' metric. Why doesn't anybody ever take opportunity cost into consideration? But then again, it doesn't matter what I think ... it's all about the clients, isn't it?
Lets say strategy A is like investing in asset class I and strategy B is like investing in asset class II. Strategy A either has or has not perfect correlation with strategy B, if not is the case then they are uncorrelated. If they are uncorrelated and assuming volatility for both strategies are the same X and expected reward for both strategies are 10% each year, if everything goes as expected after 20 years both strategies will yield 673%. Lets assume no capital is added along the way to simplify. Asset class I do not go to zero, asset class II do not go to zero, risk is zero. Reward is the same for both 673%. Obviously what happens along the way does not matter, no risk is taken and a sure gain is captured if the expected return is gotten. The same is gained and the same is risked no matter if 100% is invested in A or 100% is invested in B or any other possible mix of both like 50/50 is one example of. Since they are uncorrelated the mix is swinging less along the way but this does not affect risk/reward. However if leverage were to be applied thus making blowup a possibility, things will look differentâ¦.. Correlation and the degree of correlation will matter⦠The mix will less likely in theory blow up, in practice however the market can chose to be insane in a very correlated way..
Blow up is defined as uncorrelated classes becoming corrolated and no remediation being in place nor recognition being processed. In non stationary time series almost no effort is made to deal with the requirements to keep in place what is required to keep the window up to snuff. Check out some of Mark Brown's and Steenbarger's screw ups.
This got me into constant thinking. Alan made a clear distinction between strategy with edge and strategy fit to market character. What does the above quote imply? Is it that he abandoned edge-based strategies, and replaced them with simple, market character based strategies, scattered all over different timeframes and markets, and has a single model based on immune system model, which, based on the recent information input (results of the strategies), tells which strategies are currently worth to trade. I can only imagine how this model works, since I don't know the details. I've got a hold of a quite detailed science paper about immune system simulation. Probably will read it some time this month.
I don't understand where our difference of opinion comes from. If "smoother" means less volatility in a return stream, intuitively (to me at least) this means less risk / more stability. Assume a single strategy and a portfolio strategy return the same %gain for a given risk per trade over a fixed period. Assume the shape of the single strategy return stream is more jagged than the portfolio strategy. Ceteris Paribus, in order to match the smoothness of returns streams beween a (jagged) single system and a (smoother) portfolio of systems, one increases the risk per trade in the portfolio strategy and in doing so increases the returns beyond those of the single strategy - thus the "risk reward ratio" is boosted. More reward, same risk. Comments?