This article provides a brief recap on how our NBA playoff predictions have performed so far in 2017. To see a more detailed background on the models used for these predictions, see this article.

Total Trends

So far in the 2017 playoffs, points continue to be piled on. Recalling from our first article, playoff games since 1991 have an average points scored of 192, and games since 2013 have an average that rises to 198. Considering all 65 first and second round games in this year’s playoffs, the mean actual total was 208.3, the mean Vegas Total was 207.4, and our mean predicted 284 Total was 206.4. Perhaps not a surprise, the over has done well, going 37-28 (56.9%) so far. Even if we compare to 2014-2016, this year has seen an increase in points scored (see the chart below). This could adjust as the playoffs wrap up, as the NBA finals typically produce lower scores than earlier rounds.

Spread Trends

Considering all 65 first and second round games in this year’s playoffs, the mean home team margin of victory (mov) was -2.5, the mean Vegas Spread was -3.8, our mean predicted 284 Spread was -5.3, and home teams went 25-40 (38.5%) ATS along with 38-27 (58.5%) straight-up. Following along with this trend, favorites are 37-28 (56.9%) ATS so far in 2017, and road favorites are 15-2 (88%) both ATS and straight up. As there have been some dominant teams in early rounds, these figures are simply reflecting the fact that the higher seeded teams are having no problem winning on the road. For purposes of the plot below, we have changed “home team mov” to “winning team mov” which illustrates that 2016 and 2017 have seen a jump in the margin of victory (i.e., Cavs and Warriors dominating everybody).

Model Accuracy

Here is how each of our models has fared in predicting all 2017 playoff games. With the exception of the 284 Spread model, the models have been close to their historical ATS %s (shown in the right-most column of the table below). As far as the 284 Spread results, one contributor to these poor predictions is that this model typically favors the home team to cover, and home teams have only covered 38.5% of games in 2017.

Optimized Model Accuracy

As outlined in our first article, the accuracy of our predictions sees a big jump when we apply a few simple criteria to the games we are choosing. To recap that criteria: for spreads, we use the criteria of:

(1) our 284 Spread and Cover Prob agree on which team will cover,
(2) our Cover Prob is 70% or more, and
(3) our 284 Spread is at least 2 points different from the Vegas Spread

and for totals, we use the criteria of:

(1) our 284 Total and Over Prob are on the same side of the total,
(2) our Over Prob is 70% or more (or, in the case of an under, 30% or less), and
(3) our 284 Total is at least 2 points different from the Vegas Total

So far, 2017 games meeting one of these criteria are a combined 25-6 (80.6%), with most of those games meeting the total criteria (rather than the spread). Maybe this means we are due for some Ls? Or, maybe it means our models have Adam Silver right where we want him… Either way, they are off to a hot start.

If you are interested in seeing our predictions for the remaining playoff games this year, follow us on Twitter. If you are looking for more info on these models, predictions, or our data, check out this article, or feel free to reach out on Twitter or Facebook.