2017-18 team win projection contest

[RyanRiot]

Obviously a great return to form for 538's preseason predictions, but how do we feel about their playoff predictions? Some of the probabilities they're giving look a little funky, probably due to underrating teams lile the Warriors and Cavs who've lost a bunch of regular season games to bad teams because they don't really care about the regular season.

[sndesai1]

yeah, their playoff predictions are not good and i think they know that. i don't know if there's a good way to get around the issue when they want to base their stuff on elo...

[shadow]

I don't think you can reasonably expect anyone to come up with a model that consistently beats Vegas when they're predominantly using team level data, regardless of the specific ratings methodology they utilize. Even the opening line routinely beats just about every computer team level rating at ThePredictionTracker in terms of absolute and MSE each season.

[sndesai1]

i didn't expect them to beat vegas, but they're wildly off-market with the warriors ~5% to win a title. it seems that's partially because their method is so recency driven

[shadow]

Gotcha. I agree. The recency weighting they use does seem a bit extreme. You would think the CARMELO portion of the CARM-ELO ratings, which project team strength based on their individual player RPM/BPM blend ratings, would boost the Warriors overall rating a bit more, but that doesn't appear to be happening.