Clarification for Conrad and group
I wrote: This could also include rules for allowing or calling for judgement and intervention, ie when to bail, when to reset, etc.
Well, that is only so if you have a working model for judgement etc.
Yes, that's true, but I had something less high-concept in mind. The algorithm has bounds; when bounds are crossed, you can either continue or intervene. Thes might be called "decision points." There're only two well-behaved ends for an AIM series that I can think of, either the series ends or the stock value zeros. Running out of cash is a decision point; I can imagine there could be more ways for the algorithm to behave badly if we add more complexity than plain AIM.
There are ways to deal with the complexity that you alluded to in your previous post. It's true that, for such problems, the number of possible solutions increases exponentially with a linear increase in the number of variables. To deal with this problem, the so-called combinatoric explosion, typically you pick some random combinations of variable values to characterize the solution space, and then assume that the solution space has no discontinuities or small, local maxima. Some possible approaches are the method of steepest descent, genetic algorithms, simulated annealing, and others.
I'm getting way ahead of the story I suppose.
best regards Tim
AI
Market Data 
Markets 
