BTC being totally transparent, you can just calculate:
If the miners want to make the same amount they do today, at a price of 500$/BTC it would work out like this:
1 block (25 bitcoins) created every 10 minutes -> for each minute, 2.5 bitcoins are created X $500/bitcoin = $1250/minute.
So every minute $1250 in fees needs to come from transactions (instead of the block rewards today).
Using 5 transactions/sec as maximum speed (with current block size), 300 transactions each minute.
$1250/300 transactions = $4,16 per transaction.
Of course this changes as the metrics change:
2x higher BTC price -> 2x higher fees
2x more transactions per block -> 1/2 fees
Are you sure it's that easy? Clearly the metrics you state do influence transaction fees. But there are also a number of other factors that make calculation a bit more complicated or outright impossible.
For example, one has to consider investment costs in hardware and the efficiency of the hardware. Also cost of electricity is important. I think as we head into the future, efficiency will improve further on all fronts due to increased competition - that might even include use of heat generated by miners for industrial purposes. Therefore I expect that the cost per transaction will increase far less than a simple calculation based on number of transactions, reward, and blocksize might suggest.
(Btw. I think that 2x higher BTC price leads to half the fees in BTC and the same fees in fiat.)
ya.ya.yo!
No, of course it's not that easy, I mostly want to help to get this discussion started. Most complicating thing is the relation between mining costs, difficulty and bitcoin price. You can write pages full of theories about it, and knowing how that relation works might make you understand what will happen to the price after the halving. I for one agree with r0ach who says that the mining ecosystem works as a decentralised exchange.