Classic distance-based pairs trading earned steady returns in the 1990s and has been fading since. The spread still mean-reverts; it just does so faster than a daily strategy can capture.
Cointegration tests help less than expected. A pair that passes on five years of history fails out of sample about as often as one picked by sector alone, so a sensible stop on spread divergence is not optional.
The versions that still hold up use baskets rather than pairs, shorter holding periods, and an explicit cost model for the short leg.