As pointed out by Shiller (“Irrational Exuberance”, 2000), equity risk premia won’t always be positive. However, that is only a straw man; nobody actually claims that.
These use four sets of random numbers, representing the periods “whole”, “early”, “later” and “mixed”. During the later period (1985-2006), equity returns were considerably lower than during the early period, materially affecting the risk premia. As shown in the table below, over periods of 15 years, based upon compound UK return differences between equities and long conventional gilts, the risk premium would be negative 42% of the time for the later period. Taking the mixed column (80% chance of being in the later period), the risk premium (% pa) would be higher than 1% 64.8% of the time, higher than 2% 50.2% of the time and higher than 3% 42.3% of the time. So long as one has the time, these seem pretty good odds.