Optimization vs. Adaptation

A word on optimization. This is feasible for static problem domains, like airplane wings, since the problem domain (laws of aerodynamics) doesn't change. In dynamic problem domains, such as traffic or societies, you can't really optimize, because the "optimum" is changing constantly (if it is knowable). In these circumstances, indeed the system tries to find the "best" solution for the current situation (optimize), but since the optimization process neither reaches an optimum nor stabilizes, it would be better described as an adaptation process. Like this you can understand why short term decisions lead to long term failures.

More on my paper: Self-Organizing Traffic Lights. Complex Systems 16(1): 29-53. [preprint]

Comments

spec said…
It seems the stock market is a dynamic system where many try to develop trading systems based on optimization. Most trading systems based on this approach fail because what is optimum changes over time. Can you develop a trading system based on adaptation instead?
Carlos said…
Answer to spec:

My gut feeling is that an adaptive market strategy would beat an optimizing one. Certainly, if you "optimize" your predictions every minute, this counts as adaptation...
However, if everybody would use adaptive strategies I wouldn't be able to say what would happen, but it would be terribly interesting to find out...
spec said…
Optimizing a system every minute could take hours of computer time as the possible combination of parameter settings increases exponentially when you have more than a trivial number of parameters.

Since time is of the essence when trading, there must be a means to approximate continuous optimization. Don't scientist often rely on approximation when the domain set is intractable?

If everybody uses the same adaptive strategy, my guess is that the markets would wind up in a stale mate, just as when two computers of equal strength play chess against each other.
spec said…
Or if everyone were to use an adaptive trading strategy, it could lead to flocking type behavior. By the way, could flocking algorithms be used to approximate continuous optimization?
Carlos said…
Answer to spec:

I think that in principle you can use swarm optimization to approximate solutions continuously, but I'm not familiar with the literature.

If everybody used the same strategy, not necessarily they would reach a stale mate, since timing is also an important factor, e.g. the first or last trader might have a slight advantage that on the long run would lead to outperforming the other traders with the same strategy... also not all traders have the same resources, so usually a trader with more resources can outperform those with less...

Popular posts from this blog

flocking memes...

Complex Systems Society Seminars

The Past, Present and Future of Cybernetics and Systems Research