In our last post of the series, I left off saying that I planned on discussing the risks behind blindly using evolutionary optimization models. I ran across a post on the blog of an engineering student named Jordan Burgess this morning, however, that is going to cause a brief divergence in our discussion.
Below, I have recreated an image that demonstrating an algorithm creating the stiffest shape possible for a given amount of material.
(As Mr. Burgess points out on his blog, it is rather re-assuring that the shape we end up with is a shape we often see on the bridges we drive over.)
This is what we wish optimization always was. I can’t say it any better than Mr. Burgess did in the title of his post; we want:
“algorithms that design structures better than engineers.”
Unfortunately, when it comes to portfolio management, we cannot blindly apply an optimization scheme and cross our fingers that an algorithm will find a better structure than ourselves, the portfolio engineers. This is because physics, at least at the macro-level, has well defined laws and behaviors that we can rely on always being true. Gas molecules trapped in a box don’t all rush for the exit when you fling open the top; people do. The laws that govern financial markets and economies are not fixed and defined: they are fluid. While we may attempt to define macro principles that serve as guides for economic policy, even those are hotly debated.
Without defined rules, we can never be guaranteed that the solution we find is actually optimal; after all, we may be basing a portfolio on the wrong set of rules. Consider building a portfolio assuming negative correlation between stocks and bonds, and suddenly the correlation relationship is inverted due to changes in inflation expectations. In mechanical engineering, it would be like building a bridge where we assume gravity pulls us down only to wake up the next day to find that suddenly it pushes us away. If the very fundamental laws and relationships we rely on to “optimize” are not fixed, then we cannot be guaranteed that our outcome is truly optimal for all future situations.
This example demonstrates what we wish optimization was in portfolio engineering, and serves as a reminder of what it will never be.