Modern portfolio theory

I've touched a little bit of so-called modern portfolio theory. The basic philosophy behind (both single-index model and multi-variable model) is to decompose an extremely unpredictable variable (expected return) to one (beta) or several less unpredictable variables (GDP, interest...,etc.) The previous problem still exists: how to measure the difference between expection and real value? Well, the theory assume these expectations are easier to measure... Another fundmantal issue of the theory is to assume there's a linear relationship between expected returen and all kinds of decomposed variables. One of the shortcomming of regression analysis is overlooking of movement patterns. We need using more advanced mathmatical tools like wavelet analysis, fraction, game theory, fuzzy logic or evolutionary model (my favorite) to address the complexity we face today.