I'm going to start with a comment contrary to where you are going, and then help address your point.
I think the cost of a 16 is /you can't start there/.
1. The balance point of ASI vs. feat needs to be a real tradeoff to have meaning.
2. Some builds can excel with a single very high stat (rogue, various archers) that covers the majority of what they do while others require multiple abilitiy scores and would be well behind if someone could just focus on one.
3. One of the more common "X is overpowered" threads are when people are using rolled scores and starting with high abilities score. I think there's one now complaining that Bladesinger AC is overpowered because at low levels their character has 20 INT and 20 DEX.
4. Starting with a +4 and going to a +5 breaks a bit out of bounded accuracy for those levels.
5. Spending a lot of point buy on a 16 then ends up with lots of mediocre or dump stats. "Well, I've got an 18 DEX after racials, but everything else gives +0 or worse.
6. This encourages only playing classes with a +2 racial ability to the score you want to maximize, so you'll further restrict class/race combinations.
Okay, those are my reasons why I discourage it. Now onto helping figure out the right price for people who want it.
Breaking down the point buy costs compared to the distribution of 4d6 drop the lowest. A 13 is the most likely result, I'd like to treat that those all cost 1 as a given and not worry about it. (By the distribution, 8 to 9 and 9 to 10 should probably cost 2 points, but that would encourage more dump stats by saving twice as many points leaving at 8, so I'm on board with it.)
I would strongly recommend having two abilities scores with the same price increase - 14 and 15 are the same, 16 and 17 would be the same. This is to prevent the issue with races that only give +1 racial becoming too far behind the curve. And anyway, to get to 17 you need to have already paid the higher cost for 16 as well.
I'm torn. Looking at the distribution, I'd got up by 3 for 16 and 17, but I think that that makes 16 too accessible because of the possibility of starting with an 18 so I'd rather go for increasing the cost by 4.
16 ... 13 points
17 ... 17 points
Progression: output [highest 3 of 4d6]
http://anydice.com/program/9b26