It did used to be exactly 2 held to 2-hours before; obviously, it has long since changed. There is definitely no rule-of-thumb now other than the algorithms are absolutely different & based on each flight's specifics, but (in general) it is usually always a number higher than 4 being held unless the loads (& historicals) are really, really light...

Remember, for each cabin of travel they are factoring the number of seats available combined with: days/hours to departure, the number of seats sold/ticketed, the number of seats on hold but not-yet-ticketed, the load factors on adjacent flights and they are doing this all by fare bucket (i.e. yield); the algorithms are also overlaying historical route data to help forecast the percentage/# of no-shows, walk-ups and so on...all of this dictates how many seats are sold or available in any given fare bucket for any given flight along with if/when to release specific inventory (upgrade-type inventory included) on what is (or certainly will be) a near real-time basis . Sometimes the algorithms get it right & sometimes they don't...As the airlines view it (per mile), their end goal is to balance generating both traffic & yield at their maximum possible values with the most efficient cost/capacity structure.

Another key thing that I believe is (&/or will be) factored into the algorithms as technology continues to allow advances in yield management structures is: Individual flier (particularly frequent flier) historical travel patterns to further refine the yield management process. More specifically (& typically): what type of fare(s) do you buy, what route(s) do you fly, your no show/itinerary change/upgrade patterns, etc, etc, etc...No absolute certainties, but lots of possibilities that could positively impact their forecasting/strategic models & enhance revenue...