The first weekend of Madness is over and we are on to the sweet 16! Let's review how my models held up in Vegas over the first week of games (and see if it is worth continuing to offer advice). As a reminder, I posted all my picks on twitter to make sure you don't think I'm pulling a "Chad Ford".
In my last post I made some suggestions for what people should bet in Vegas against the spread (ATS). This included 9 games from Thursday/Friday action. I was only able to add one additional ATS pick over the weekend as I was too busy watching the Seattle regional action live (which included Mamadou costing me one of my picks). In the end, I made 10 ATS picks and my overall record was…7-3! Not bad, not bad at all…
Here is a chart showing my predictions:
I got some good fortune…it took a missed free-throw in the final seconds by a career .805 shooter for Indiana to beat the spread. Yet I also got a bad break…a garbage layup by VCU to beat the spread which apparently impacted more than just my picks (Vegas must have been reading my blog!).
Since my last post I have also added a second ATS model (which I will not be revealing the methodology for). Here are both models projections1 for all tourney games up until this point along with results:
Here are some observations:
- If you would have bet on every single game using the models, both would have individually gone 27-24 (52.9%).
- If you would have bet all games where both models made the same pick you would have went 15-12 (55.5%).
- If you would have taken the models and combined with some human intelligence you could have went 7-3 (70%) like me! Ok…ok…small sample size, let's see if that holds up.
I didn't create any sort of blog post or anything describing my methodology but if you followed my twitter you would have noticed some order/under picks throughout the weekend. However, I jumped the gun a bit on publically revealing anything from my over/under models. It wasn't until Sunday that I found out they were pretty flawed causing me to overhaul (which basically negated all the picks I had made on twitter). The only "true" picks from my revised models were the two Sunday picks which went 1-1. Even then, by my count the "flawed" picks from Friday and Saturday still went 6-5.
Here are the over/under model projections for all tourney games up until this point using the revised models:
Here are some observations:
- If you would have bet on every single game, M1 would have went 28-23 (54.9%) and M2 would have went 33-18 (64.7%)
- If you would have bet all games where both models made the same pick you would have went 18-8 (69.2%)! This sounds too good to be true, I must have over fit somehow…
- In addition to the ATS and Over/Under models, I would like to add a money line model to suggest the best opportunities for money line action relative to payout.
- If these models continue to perform as well as they have been it would probably be a good idea to back test them to see if they are getting lucky or if I might have a hidden gem on my hands!
- More picks! That is the fun part after all!
Next up: Sweet 16 model projections and picks
If you look carefully at the first model you might see some of my numbers have changed since the first post. That is because I have made some minor improvements which I won't reveal. Note that all tweaking I have done has had 2015 results as out of sample so there shouldn't be over-fitting concerns here. ↩