In those datasets you could improve compression by adding apriori knowledge about their context in the real world - if the compression program knows about historical S&P500 prices, it will be pretty good at compressing all historical stock info. Kolmogorov complexity is indistinguishable from AI