Thursday, May 31, 2018

UFC Fight Night Rivera vs Moraes Fight Projections

*Disclaimer: Betting advice is just for fun, I am not responsible for any bets placed*
*All bet amounts are based on percent of bankroll, per the Kelly criterion*

Jimmie Rivera (53.64%) vs Marlon Moraes (46.36%)

 53.64%
Overall
46.36% 
 10.1%0%
KO
12.58% 
 3.71%
Sub
5.24% 
 39.83%
Decision
28.54% 


Plays:
1.82% on Rivera on +100

Jake Ellenberger (45.98%) vs Ben Saunders (54.02%)

 45.98%
Overall
54.02% 
 34.42%
KO
34.87% 
 0.34%
Sub
9.26% 
 11.22%
Decision
9.89% 


Plays:
7.46% on Saunders +155

Gian Villante (77.18%) vs Sam Alvey (22.82%)

 77.18%
Overall
22.82% 
 28.81%
KO
12.73% 
 0.72%
Sub
0.26% 
 47.65%
Decision
9.83% 


Plays:
34.20% on Villante at +130

Nik Lentz (24.02%) vs David Teymur (75.98%)

 24.02%
Overall
75.98% 
 4.35%
KO
11.49% 
 3.19%
Sub
0.85% 
 16.47%
Decision
63.64% 


Plays:
1.56% on Teymur at -290

Desmond Green (77.90%) vs Gleison Tibau (22.10%)

 77.90%
Overall
22.10% 
 16.59%
KO
7.05% 
 0.77%
Sub
0.83% 
 60.54%
Decision
14.22% 


Plays:
3.37% on Green at -295

Saturday, May 26, 2018

UFC Fight Night Thompson vs Till Fight Projections

*Disclaimer: Betting advice is just for fun, I am not responsible for any bets placed*
*All bet amounts are based on percent of bankroll, per the Kelly criterion*

Stephen Thompson (49.82%) vs Darren Till (50.18%)

 49.82%
Overall
50.18% 
 25.94%
KO
33.01% 
 0.11%
Sub
0.02% 
 23.78%
Decision
17.15% 


Plays:
3.94% on Till at +115

Jason Knight (56.78%) vs Makwan Amirkhani (43.22%)

 56.78%
Overall
43.22% 
 9.91%
KO
10.61% 
 27.27%
Sub
6.75% 
 19.60%
Decision
25.85% 


Plays:
3.03% on Amirkhani at +150

Eric Spicely (51.80%) vs Darren Stewart (48.20%)

51.80% 
Overall
48.20% 
 7.23%
KO
4.46% 
 22.47%
Sub
0.83% 
 22.10%
Decision
42.91% 


Plays:
7.63% on Stewart at +145

Dan Kelly (22.42%) vs Tom Breese (77.58%)

 22.42%
Overall
77.58% 
 4.31%
KO
20.24% 
 0.04%
Sub
10.29% 
 18.07%
Decision
47.05% 


Plays:
None

Elias Theodorou (85.70%) vs Trevor Smith (14.30%)

 85.70%
Overall
14.30% 
 9.12%
KO
3.79% 
 0.15%
Sub
0.49% 
 76.42%
Decision
10.01% 


Plays:
23.64% on Theodorou at -385

Friday, May 18, 2018

UFC Fight Night Usman vs Maia Fight Projections

*Disclaimer: Betting advice is just for fun, I am not responsible for any bets placed*
*All bet amounts are based on percent of bankroll, per the Kelly criterion*
*Note: Start of new random forest model*

Diego Rivas (54.74%) vs Guido Cannetti (45.26%)

 54.74%
Overall
45.26% 
 16.15%
KO
16.99% 
 14.34%
Sub
5.49% 
 24.25%
Decision
24.12% 


Plays:
6.34% on Cannetti at +155

Vincente Luque (69.56%) vs Chad Laprise (30.44%)

 69.56%
Overall
30.44% 
 17.33%
KO
4.02% 
 28.11%
Sub
10.75% 
 24.12%
Decision
15.67% 


Plays:
2.46% on Luque at -215

Zak Cummings (50.98%) vs Michel Prazeres (49.02%)

 50.98%
Overall
49.02% 
 11.49%
KO
8.32% 
 4.32%
Sub
6.22% 
 35.17%
Decision
34.49% 


Plays:
12.71% on Prazeres at +165

Brandon Moreno (50.12%) vs Alexandre Pantoja (49.88%)

 50.12%
Overall
49.88% 
 7.49%
KO
6.50% 
 8.11%
Sub
18.13% 
 34.51%
Decision
25.25% 


Plays:
2.88% on Moreno at +110

Henry Briones (20.30%) vs Frankie Saenz (79.70%)

 20.30%
Overall
79.70% 
 5.47%
KO
3.72% 
 0.54%
Sub
1.72% 
 14.29%
Decision
74.26% 


Plays:
9.92% on Saenz at -320

Thursday, May 17, 2018

Relative Yards per Carry: Evaluating Running Backs with Context

When evaluating players in the NFL, one of the chief concerns should be separating the contributions of each player rather than evaluating those players as a unit. For example, how can we find a way to separate a wide receiver's ability from the ability of the quarterback and their relative ability as compared to the receivers around them? Well, one of the ways we can do that is to look at dominator rating, which has been widely used in the football analytics community to evaluate receivers coming out of college. Today, I'm going to propose a similar metric, one designed to adjust for offensive line, running scheme, and other running backs in their committee. It's called Relative Yards per Carry or rYPC for short.

rYPC is defined as the ratio between a player's yards per carry and the rest of the team's yards per carry, excluding QB runs. Doing this gives us a number that tells us how efficient that player was at rushing in comparison to the other runners on the team. This helps controls for offensive line and scheme, as the other runners in the offense are also benefiting or being held back by those factors.

As an example, let's look at Alvin Kamara from last year's draft class. His yards per carry number was 5.79. An above average number (average YPC in my data set is 5.41), but nothing that jumps off the page. However, the runners around him accumulated 958 yards on 215 carries, good for only a 4.46 yards per carry. This gives Kamara a 1.30 rYPC, a fairly high end number for a running back prospect. By contrast, Donnel Pumphery averaged a higher yards per carry (6.11) than Kamara, yet came in a much lower rYPC (0.80) based on the performance of players around him.

Now, the idea sounds good in a vacuum, but is it actually any good for projecting how these running backs will play in the NFL? Well, let's take a look at a few plots, using data from 2007 to 2013:

This graph plots approximate value per season against the natural log of draft position. As we can see, the relationship is pretty clear, other than the two at the very end (Trent Richardson and Darren McFadden).

As we can see, the relationship is much smaller, but there is some signal here. Only two running backs with a rYPC below 1.0 have ever averaged more than 5 AV per season, while more than a dozen above 1.0 have. At the very high end, the top two scorers in rYPC were great players in the NFL (Matt Forte and Jamaal Charles).

When we combine both draft position and rYPC, the relationship between them increases than using either of them alone. As we can see, Trent Richardson and Darren McFadden are not nearly as big of outliers, because both of them had well below average rYPC numbers (0.89 and 0.77 respectively).

Things get quite a bit interesting when we look at just the top 3 rounds of the draft:

When removing the last 4 rounds of the draft (who are by definition, extreme long shots), the relationship between draft position drastically lowers. The r^2 goes from 0.322 to 0.079.

While the relationship between draft position and NFL production got weaker, the relationship between rYPC and NFL production gets stronger when only the top 3 rounds are considered. In fact, the r^ 2 value of rYPC is actually BIGGER than draft position in the first 3 rounds (0.079 vs 0.084).

As in the full sample, combining both draft position and rYPC results in a better correlation than using either factor alone. 


Looking at the data, at the very least rYPC is an interesting tool to use to evaluate prospects, and a vastly superior tool compared to regular YPC (YPC had a r^ of 0.02 and 0.005 in the full sample and top 100 picks sample respectively, no correlation). 

When we look at the draft classes from 2014 to 2017, the top four rYPC numbers are Jordan Howard, Aaron Jones, Dri Archer, and David Johnson. While Archer never broke out (and one would argue was never a true RB), Howard, Jones, and Johnson all look to be at the very least much better than their draft position.

Here's how the running backs selected in the top 100 picks of this year's draft ranked in rYPC:

 Name
rYPC 
 Rashaad Penny
1.41
 Sony Michel
1.36
 Ronald Jones
1.23
 Saquon Barkley
1.03
 Royce Freeman
0.99 
 Derrius Guice
0.96
 Nick Chubb
0.89
 Kerryon Johnson
0.80
Surprise first round picks Rashaad Penny and Sony Michel form the first tier of runners, while Ronald Jones follows close behind. Perhaps surprisingly, consensus top RB Saquon Barkley looks fairly pedestrian by this measure. However, it's important to note that rYPC only looks at a running back as a runner, and Saquon offers significant value in the passing game. Freeman and Guice are just about at a 1.0 rYPC, which shouldn't be a giant red flag but also shouldn't inspire great confidence. Chubb is below average, but he also had to share a backfield with Michel, which could dilute how good Chubb really is. However, we had a similar situation with Mixon and Perine last year in Oklahoma and the vastly inferior player in rYPC (Perine) seems to be on his way to the bench after an ineffective year on the ground. 

The guy with the biggest red flag is Kerryon Johnson. Johnson was less efficient than other Auburn runners like Kamryn Pettway, Kam Martin, and hybrid player Eli Stove. This trend also held true in 2017, where Johnson was the least effective back in the rotation. If I had to pick one player from this year's top 100 backs to bust, I would choose Kerryon without much of a second thought.