Abstract
While it is common knowledge that portfolio separation in a continuous-time lognormal market is due to the basic properties of the normal distribution, the usual exposition found in text books relies on dynamic programming and therefore invokes Itô stochastic calculus. Khanna & Kulldorff (1999) gives a rigorous proof which essentially reduces to the elementary properties assuming a risk free asset exists, an assumption we drop. Further simplifications are given, and generalizations to (symmetric and non-symmetric) alpha-stable driving noise.