I was looking at si1ly's system for any other patterns, so I took a look at the games where the si1ly's system had one team favored more than the actual line (e.g. Team A favored by -5.0 in the power rankings calculation, while favored by only -4.5 in the real vegas odds). I was hoping to find potential ML value in betting those games.
Using these parameters, I found 38 games (since his All-Star post of this system) that fit the bill. Then I separated "si1ly's system plays" and non-system plays to see if I could see any other patterns. I did noticed some trends so--though keep in mind, it's a terribly small sample size (I did this pretty quickly and didn't double-check thoroughly.)
When it says "switch" or "no switch" it means, that there were games where the power ranking calculation resulted in a favored team that wasn't actually favored in the actual betting line. (e.g. Team X favored by -1 via system, but actually an +2 underdog).
si1ly's system play (>=3.5 diff) were 7-7 ATS, 6-8 SU
si1ly's system play (>=3.5 diff, no switch) were 1-1 ATS, 2-0 SU
si1ly's system play (>=3.5 diff, switch or PK) were 6-6, 4-8 SU
Non-Sys Play (<3.5 diff) were 17-6-1 ATS and 19-5 SU
Non-Sys Play (No switch) were 17-5-1 and 19-4 SU
Non-Sys Play (Favored >1.5 in real line) were 14-2-1 and 16-1 SU
Total Plays were 24-13-1 ATS and 25-13 SU.
So far it seems like there is something about the "switch" thing where the predictions become murky, especially in the SU category--that is, maybe the team really isn't the favorite as the calculations suggest.
However, when there is a team actually favored in both (17 games sample size lol), the SU become a lot stronger. Also, those same teams do seem to cover at a higher percentage as suggested by si1ly's rankings, but
For example, today the only example was Cleveland. The system had CLE -2.3, yet the real line was -2, for a 0.3 value (not a system play). Using these trends, maybe you could take both the ATS and money line bet gaining "some" value (they both hit, fwiw).
Unfortunately, it's a terribly small sample size, so it may end up being nothing more than a great run, but I'll keep tracking and see what comes up.
Love to hear anyone's thoughts, especially si1ly's. Maybe something with his power ranking calculations have to do with it.