A detailed analysis of what's advantageous here would be relatively complicated as the problem depends on a multitude of variables:
- number of gold-working tiles in your empire
- growth rate of the same when analyzing whether it's better to save up the cheaper early GA to pop it later
- number of cities in your empire
- growth rate of the same for the same reason as above
- the gold you can get for trading
- whether you have any modifiers to golden age length
- weighing of the production and cultural bonus accrued during GA
I can give you a very simple rule of thumb:
The number of turns the four happiness would take to trigger a golden age is
t = H/h
where H is the happiness required for the next golden age and h=4 the happiness of the luxury. This assumes selling the luxury puts you at 0 happiness, or, equivalently, that you weigh the golden age yield by the fraction of h/hE where hE is your total excess happiness.
If n is the number of gold-yielding tiles in your empire and m your average gold multiplier (1 if you have nothing in the way of markets and the like), the total gold yield during a golden age of length l is
G = n*l*m
The total amount of money you can make by selling the luxury during that time is
G' = t*s/30 (= 8 usually)
where s is the amount of money you sell the luxury for and the 30 is from the 30 turns, assuming you don't cheese out by declaring war or somesuch.
Saving for the golden age will be better if G > G' so we can check
G = G'
n*l*m = H/h * s/30
n = H/h * s/30 *1/(l*m)
If H is, for example, 500 (first GA and only one city), s/30 = 8 (240 gold) and l=10 (no modifiers), h=4
n = 100
So your one city would have to provide a hundred gold-working tiles for this to be better from a purely financial point of view, discounting culture and hammer benefits.
If H = 1000 (larger empire and maybe second GA), and assuming m=1.1, you'll need 180 tiles that work gold to yield the same amount of cash during a golden age.
Judging from this, it's typically advantageous to always sell your luxuries as long as you stay happy. Once you get unhappy, lack of growth has to be taken into account. The problem for GA is that 8 gpt is a lot and a golden age takes a ton of happiness to trigger.