Cross validation too uses training sets and test sets. This sort of time-ordered data will not be independent, so the prequential approach seems to be a more suitable approach to measuring forecast accuracy. (I haven't read the paper.)
Anyways, as somebody who has spent GIANT amounts of time experimenting with machine learning using market data (mainly stocks), I can give you TONS algorithms I've created that would show similar, even much higher, gains when you only test on 6 weeks of data. Been there done that. For example 6 weeks you make a 100% return, then in the next 3 weeks you suffer a 50% loss. Reality sets in...
