Contrary to some of the comments in this old thread, I think it would be very useful to do a statistical analysis of the predictors of both profits and of system failure. How important is longevity? How important are low drawdowns? How important is a win % below 90%?
In my PhD statistical training, one of my professors noted that smart people were often good at guessing which variables are associated with outcomes and the direction of the relationship, but they are very poor at guessing the MAGNITUDE of the effects, and which effects persist after adding control variables.
Also, there are different sorts of outcomes that one could try to model/predict. For example, if one were predicting which strategies might fail over the next six months, my guess is that current high-flyers would be over-represented. But if one were predicting which systems would get the highest average returns over the next six months, my guess is that current high-flyers would also be over-represented in that group as well. In part, of course, these two contrary outcomes would be expected because of the higher risk of most extremely high return strategies.
To do a good study, one would have to overcome the very serious survivor bias problem, probably by downloading the GRID and the LEADER BOARD today and then coming back later to test results.
As David Stephens has argued, this community thrives when investors here actually make money–and stick around. Both Collective2 and System Developers probably make more money when there is a larger pool of investors to draw from, some of them looking for additional strategies for diversification.
If there are some predictors that are much better than others for predicting future returns or avoiding collapse, these should be moved to the Leader Board. Personally, I’d like to see “Returns since autotrading began,” and 3 month or 6 month returns with blanks for strategies not old enough to qualify.