NBA Strategy: Using Spreads To Predict Quality Of Matchup
Using Sportsbook spreads is an integral part of most DFS sports, because in most sports, points, runs, or goals practically equal fantasy points. NBA is unique as a DFS sport because the majority of fantasy points are scored through non-point categories such as defensive statistics and assists, and therefore are not captured very well by the spread. But aside from points, which the spread should predict very well, how well can we predict fantasy statistics based on Sportsbook projections?
Margin Of Victory and Fantasy Points
All the statistics I’m using were downloaded from the website NBAStuffer, which sells raw data for a very reasonable price.
My assumption going into this research was that the margin of victory, something the spread attempts to predict, would be positively correlated to both offensive and defensive fantasy stats. I had to extract victory margins from the data, but after doing so I simply had to run a few correlations. First, let’s look at individual defensive statistics. To avoid confusion, TOT refers to total rebounds.
> cor(gameLog$ST, gameLog$victoryMargin)
> cor(gameLog$BL, gameLog$victoryMargin)
> cor(gameLog$TOT, gameLog$victoryMargin)
The correlations are quite small, with rebounding unsurprisingly being the most correlated. This makes a lot of sense because rebounding is directly related to missed shots, which occur more often for the opposing team the worse they do offensively.
> cor(DFP, gameLog$victoryMargin)
When we look at defensive fantasy points, we see a slightly stronger correlation than rebounds alone. It seems that victory margin has a small amount of predictive power for defensive fantasy stats. But what about offensive?
> cor(OFP, gameLog$victoryMargin)
We see a much stronger correlation with offensive fantasy points, which makes sense because points comprise the majority of offensive fantasy points, and a large margin of victory implies a large amount of points scored. But is it only predictive of points? Turns out it does a relatively good job of predicting assists as well.
> cor(gameLog$A, gameLog$victoryMargin)
> cor(gameLog$PTS, gameLog$victoryMargin)
> cor(gameLog$TO, gameLog$victoryMargin)
Overall, victory margin does not seem like a very good predictor of fantasy output, although it does a much better job with offensive fantasy points than defensive.
Efficiency and Fantasy Points
Offensive and defensive efficiency is a measure of how many points and team scores and allows on a per possession basis. It is essentially a measure of how successful a team is offensively and defensively with taking pace out of the equation. We already know that offensive and defensive efficiency are directly correlated with offensive and defensive fantasy points allowed. So shouldn’t the same be true in each individual game? Because we can predict the offensive and defensive efficiency of each team for any given day through the spreads, it makes sense to look at the predictive power of efficiency statistics.
> cor(OFP, gameLog$OEFF)
Here we get a much stronger correlation, but is it all just point predictions?
> cor(gameLog$PTS, gameLog$OEFF)
> cor(gameLog$A, gameLog$OEFF)
> cor(gameLog$TO, gameLog$OEFF)
While points does have by far the strongest correlation, turnovers and assists do have some relationship with offensive efficiency. It would make sense then to look at offensive efficiency projections to help determine expected assists and turnovers. But what about defensive efficiency and defensive fantasy points?
> cor(DFP, gameLog$DEFF)
Here we get a much better correlation than with victory margin.
> cor(gameLog$ST, gameLog$DEFF)
> cor(gameLog$BL, gameLog$DEFF)
> cor(gameLog$TOT, gameLog$DEFF)
As we can see, defensive efficiency is drastically better at predicting rebounds than victory margin. Steals and Blocks are still not predicted very well by defensive efficiency.
The small predictive power of victory margin likely has to do with it’s own correlation with offensive and defensive efficiency. Overall, efficiency does a decent job at predicting fantasy points, moreso with rebounds and assists than blocks, steals, or turnovers. Still, other factors such as fantasy points allowed and defense versus position statistics probably are more important to consider when looking at matchup.
However, it makes sense then to use Sportsbook spreads to help predict offensive and defensive efficiency and project expected defensive and offensive fantasy points. In general, you’re going to want to pick guards from teams with high point projections, and pick big men from teams with low opponent point projections.
View all posts by Daniel Steinberg