One week of games are in the book and I wanted to take a quick look at this week’s action to see what our old friend #math might be able to tell us. As I have discussed in some of my previous pieces, I have developed a series of mathematical tools over the years that I use analyze sports data. One of those tools is a simple algorithm that does a pretty good job of projecting point spreads, from which win probabilities can be estimated. It is sort of like the garage-band version of ESPN’s FPI. Throughout the season, I plan to use the tools at my disposal to provide some advice (that you would be wise to ignore) as it relates to the Vegas point spreads for MSU and around the country.
As for MSU, the Spartans opened at -17.5 for this weekend’s game under the (neon green?) lights of Spartan Stadium. Although the line is slowing falling, my numbers suggest that MSU has only a 10.7% chance of getting upset, based on the opening spread. While that number is pretty small, this approximate magnitude of upset does occur almost once a week (~10 times per year).
As for MSU covering the spread, the metrics I use are a bit of a mixed bag. My power rankings (which are currently based on an equal blend of the preseason magazine rankings and Week 1 results) suggest MSU will cover slightly, while the FPI is pulling in the opposite direction. So, I cannot in good faith issue a strong suggestion either way for MSU. It’s probably best not to bet on the Spartans anyway.
As for the over/under in Week 1, I don’t feel confident that I have enough data to make a solid prediction right now on that one either, but when I asked my computer for the final score, it tells me to expect a 35-13 MSU victory. That point total of 48 is slightly over the 46.5 that I am seeing, but do we honestly think we are going to see 35 points out of MSU this weekend? You know in your heart what to do...
To put this into perspective, I like to visualize the data for the entire country in a single plot that compares all of my spreadsheet picks to the actual opening lines. That plot for Week 2 is shown below.
As you can see, my algorithm does a fairly good job in predicting the Vegas spread, with some notable differences. Each week there is usually a handful of games where my prediction is significantly above or below the Vegas line. This year, after reviewing my picks over the past 4-5 years, I concluded that if my picks differ from Vegas by more than 12 points, more likely than not (53.6% of the time, to be exact), my pick is correct against the spread. The dashed lines on the figure are at the 12-point threshold.
For reasons that are still a bit perplexing, my spreadsheet tends to trend towards the favored team covering. As such, the only data points that are outside of the main window this week are all on the high side. My algorithm likes the following teams:
- Penn State (-24) to cover vs. Buffalo
- Wisconsin (-34) to cover vs. Central Michigan
- Boise State (-12) to cover vs. Marshall
- Syracuse (-2.5) to cover vs. Maryland
- Appalachian State (-21) to cover vs. Charlotte
- Iowa (-20.5) to cover vs. Rutgers
- Baylor (-26) to cover vs. UTSA
The figure above also shows one other notable category, and that is the section on the left side of the vertical red line. The games listed in this section of ones where my spreadsheet disagrees with the final result of the game, i.e. my straight-up upset picks. This week, I have three: Stanford over USC, BYU over Tennessee, and Arkansas State over UNLV.
I should note, however, that I historically reference all of my data to the opening spread, and my analysis is really only valid relative to that value. Considering that the line has already moved in most cases, I can’t actually promise that my accuracy of 53% is at all valid after that line has movement. (Actually, I am not promising anything at all. That’s why I am calling this “bad betting advice...”) But, my stats suggest that the line could move by up to 4 points and I still finish over 50%.
I also find it amusing to compare my methods to ESPN’s FPI model. To be honest, the FPI does tend to perform a little better than my model overall. But, as we will see, we tend to predict different things. If nothing else, my model tends to make bolder picks, some of which turn out to be true. For comparison, I have also plotted the predictions from the FPI to the Vegas lines, which is shown below:
As you can see, the FPI matches the Vegas spread quite well. My analysis of the FPI suggests that a team is also likely to cover (about 59% of the time) when the FPI deviates from the opening line by more than 6 points. So, that is how I set the dotted line boundaries on this chart.
For Week 2 the FPI data suggests the following bets:
- Penn State (-24) to cover vs. Buffalo
- UCLA (-7) to cover vs. San Diego State
- Maryland (+2.5) to cover (actually upset) Syracuse
- VA Tech (-26.5) to cover vs. Old Dominion
- Vanderbilt (+9.5) to cover at Purdue
- LSU (-4.5) to cover vs. Texas
- Cincinnati (+17) to cover at Ohio State
- Kentucky (-14) to cover vs. EMU
Interestingly, those two lists only overlap with the Penn State game, and the models actually disagree about the Maryland / Syracuse game, so I guess that is a bit of a push. Either way my suggested bets for the two models combined went 8-3 last week (which I posted in Twitter). Just sayin’.
For convenience, I have also summarized my machine upset picks and the FPI upset picks in the table below. The FPI interestingly likes Maryland to upset Syracuse and Colorado to husk Nebraska.
Finally, based on the full set of Vegas lines, I can also perform a series of Monte Carlo simulations on the week’s slate of games. (Basically, this is a very large number of weighted coin flips). This simulation tells me that we should expect to see 9.3 ± 2.6 straight-up upsets in Week 2.
That’s all for now. As a reminder, taking my advice is most likely a terrible idea. But, just for fun, I plan to report back each week with how my model and the FPI model performed each week. Until next time, Go Green.