The idea is that most of what looks like stock picking is exposure to a small number of shared characteristics. If a manager's returns can be reproduced by tilting a cheap portfolio toward value and small caps, the manager was selling factor exposure rather than alpha.
Each documented factor comes with a story for why it might persist: compensation for a risk that shows up at bad times, or a behavioural pattern investors repeat. Both stories are contested, and a factor with neither explanation is hard to distinguish from data mining across the thousands of variables researchers have tested.
Factor premia are slow and irregular. Multi-year and occasionally decade-long stretches of underperformance are normal even for the best-documented factors, which is the practical reason most factor programmes are abandoned before they pay. See value-factor and factor-crowding.
Related: value-factor, size-factor, momentum-factor, quality-factor, fama-french-three-factor, smart-beta