Interesting post. The tricky part is there are infinitie variables one could dissect -- maybe exceptions could be proven on games with a 3-point spread (haven't tested it, just a possible variable that might matter).
I did similar math trying to figure out if I should buy points. It seems one is better off just never buying the points, but I still do it from time to time. But playing a dog on the ML is more like selling points.
Anyway, I suspect there is probably some extensive research on this somewhere -- probably purchased/commissioned by large books where they have resources to test on huge sample sizes and account for so many variables. Would love to read it.
I wonder if you could find some of this in quantitative analysis student thesis papers. I am sure more than a few have written about vegas, and tested models.
