Thank you - it makes sense to me now (I think!). If the returns for both portfolios were samples, then I think test would make sense. The idea of taking a sample of returns for a portfolio doesn't make sense though, so I wonder is this test appropriate for comparing portfolio returns. Thank you for your answer.

Does the fact that a stock can move from one portfolio to another break the independence of observations ? Which would mean that I shouldn't use Kruskal-Wallis ANOVA ?

Hi - thanks for replying. This is for college.