I created an ELO model for win probabilities in college football last night. It's bare bones, and doesn't take in any advanced stats yet, just simply looks at historical win/loss/spreads over the last 3 years and into this season. It doesn't have anything opponent adjusted or even home field advantage yet. For now I just manually adjusted 3 points off the away team. Going to implement those later to really dial in what I want the model to pay attention to. In other words, very untrustworthy at the moment. That being said, this is what it gave me for UT today:
-1 Tennessee at Missouri (-4 before manually adjusting 3 points since we are away)
That's close to what the real line is, DraftKings is at -2.5 currently, but other ones I ran for notable match ups today seemed further off. For example, Washington vs Utah, at Washington, was essentially a pick em in my model but DK has Washington at -8. I think the main reason is that the model is weighing too heavy on prior seasons currently. Kind of like FPI with their rolling 3 year model being slow to catch up to a major uptick for a team in a season. Will need to look ways to counter that. Plus, the lack of any advanced stats so far doesn't help either. That will help.