Quote:
Originally Posted by SwoopAE
I'm pretty sure if you disregard odds you should be able to pick 60% of fights correctly at least. That's only a -150 novig breakeven point and there are plenty of fights where one guy is -700 or whatever, auto-picking all favourites alone should yield 60%+ if you're just talking about picking a winner and not making profitable bets. This next card has a lot of competitive fights where neither fighter is -150 or bigger fav, but a lot of cards have 50%+ of the fights where one fighter is a bigger than +150 dog at which point you should be getting 60%+ on those fights and a skilled handicapper should still get a pickem fight more than 50% of the time
Yes. % correct picks is definitely not the best measure for betting success, but it is the easiest to communicate, and is probably very highly correlated with betting success. I don't have historical betting odds in my database, but that's definitely something that would be useful.
I decided to put in the extra effort and coded a process which applies the model I'm using to data only before year x, then predicts year x, and gives me the result. I started in 2001. That means I used only data up to and including 2000, and predicted all of 2001 with that model, and so on for every year up until now. This means I'm doing perfectly out of sample predictions. Every parameter is re-estimated based only on the data available before the year which I predict. This method is fairly punishing, but should be closer to the truth. The results were as follows:
Year All UFC
2001 71% 83%
2002 71% 78%
2003 70% 80%
2004 71% 79%
2005 74% 88%
2006 72% 79%
2007 73% 73%
2008 74% 75%
2009 74% 70%
2010 71% 73%
2011 71% 71%
2012 73% 70%
2013 73% 72%
2014 73% 73%
2015 73% 68%
2016 72% 67%
2017 75% 71%
Again, probably not enough to beat the books in 2017, but its something. It is worth noting this has no special UFC adjustments. It is the exact same fight prediction model, just filtered for all vs only UFC fights.