Saturday, December 24, 2016

Guess the Lines - Final Results

Here are the week 16 results for the final installment of "Guess the Lines". This is a weekly feature in which I put my Vegas NFL rankings to the test by predicting the point spread of upcoming NFL games. I compare the accuracy of my prediction against those of Bill Simmons and Cousin Sal, as featured on their weekly NFL podcast. In addition, I also calculate an implied point spread from ESPN Chalk's weekly Vegas NFL rankings.

I'm not sure if Bill and Sal intend to do a Week 17 podcast, but I've decided to halt the prediction contest at Week 16. Week 17 lines can be a bit wonky due to teams that don't have anything to play for.

Here are the week 16 results. Bill Simmons wins this week with a mean absolute prediction error of 0.84 points.

Friday, December 23, 2016

Turnover Index Week 16

Here are the Turnover Index picks and results for week 16. The Turnover Index is a simple betting strategy based on the theory that the market overvalues defensive turnovers when judging team strength. See here and here for more background.

We went 0-1 with our single Week 14 bet, and our bankroll dropped by $17 from $1,023 to $1,006. We are now 6-7 against the spread on the season, with an ever so slightly positive ROI of 0.5%. Bets are sized according to the Kelly criterion.

Here are the 2016 season to date results:

weekbetswonstarting
bankroll
amount betprofitsending
bankroll
6 1 0 $1,000 $36 (3.6%) ($36) $963
7 1 1 $963 $10 (1.1%) $9 $973
8 1 1 $973 $38 (4.0%) $35 $1,008
9 3 2 $1,008 $64 (6.4%) $51 $1,060
10 2 1 $1,060 $15 (1.5%) ($7) $1,052
12 2 1 $1,052 $13 (1.3%) ($6) $1,046
13 2 0 $1,046 $23 (2.2%) ($23) $1,023
14 1 0 $1,023 $17 (1.7%) ($17) $1,005

Saturday, December 17, 2016

Guess the Lines Week 16

Here are the week 15 results and week 16 predictions for this installment of "Guess the Lines". This is a weekly feature in which I put my Vegas NFL rankings to the test by predicting the point spread of upcoming NFL games. I compare the accuracy of my prediction against those of Bill Simmons and Cousin Sal, as featured on their weekly NFL podcast. In addition, I also calculate an implied point spread from ESPN Chalk's weekly Vegas NFL rankings.

Here are the week 15 results. Cousin Sal wins this week with a mean absolute prediction error of 0.94 points.

Saturday, December 10, 2016

Turnover Index Week 14

Here are the Turnover Index picks and results for week 14. The Turnover Index is a simple betting strategy based on the theory that the market overvalues defensive turnovers when judging team strength. See here and here for more background.

We went 0-2 with our Week 13 bets, and our bankroll dropped by $23 from $1,046 to $1,023. We are now just 6-6 against the spread on the season, with an ROI of 2.3%. Bets are sized according to the Kelly criterion.

Here are the 2016 season to date results:

weekbetswonstarting
bankroll
amount betprofitsending
bankroll
6 1 0 $1,000 $36 (3.6%) ($36) $963
7 1 1 $963 $10 (1.1%) $9 $973
8 1 1 $973 $38 (4.0%) $35 $1,008
9 3 2 $1,008 $64 (6.4%) $51 $1,060
10 2 1 $1,060 $15 (1.5%) ($7) $1,052
12 2 1 $1,052 $13 (1.3%) ($6) $1,046
13 2 0 $1,046 $23 (2.2%) ($23) $1,023

Guess the Lines - Week 15

Here are the week 14 results and week 15 predictions for this installment of "Guess the Lines". This is a weekly feature in which I put my Vegas NFL rankings to the test by predicting the point spread of upcoming NFL games. I compare the accuracy of my prediction against those of Bill Simmons and Cousin Sal, as featured on their weekly NFL podcast. In addition, I also calculate an implied point spread from ESPN Chalk's weekly Vegas NFL rankings.

Here are the week 14 results. Bill Simmons wins for the second week in a row, with a mean absolute error of 1.00 points.

Saturday, December 3, 2016

Guess the Lines Week 14

Here are the week 13 results and week 14 predictions for this installment of "Guess the Lines". This is a weekly feature in which I put my Vegas NFL rankings to the test by predicting the point spread of upcoming NFL games. I compare the accuracy of my prediction against those of Bill Simmons and Cousin Sal, as featured on their weekly NFL podcast. In addition, I also calculate an implied point spread from ESPN Chalk's weekly Vegas NFL rankings.

Here are the week 13 results. Bill Simmons wins this week, with an impressive mean absolute error of 0.87 points. ESPN Chalk rankings continue to bring up the rear.

Turnover Index Week 13

Here are the Turnover Index picks and results for week 13. The Turnover Index is a simple betting strategy based on the theory that the market overvalues defensive turnovers when judging team strength. See here and here for more background.

Once again, we went 1-1 with our Week 12 bets, and our bankroll dropped by $6 from $1,052 to $1,046. We are now 6-4 against the spread on the season, with an ROI of 4.6%. Bets are sized according to the Kelly criterion.

Here are the 2016 season to date results:

weekbetswonstarting
bankroll
amount betprofitsending
bankroll
6 1 0 $1,000 $36 (3.6%) ($36) $963
7 1 1 $963 $10 (1.1%) $9 $973
8 1 1 $973 $38 (4.0%) $35 $1,008
9 3 2 $1,008 $64 (6.4%) $51 $1,060
10 2 1 $1,060 $15 (1.5%) ($7) $1,052
12 2 1 $1,052 $13 (1.3%) ($6) $1,046

Wednesday, November 23, 2016

Guess the Lines - Week 13

Here are the week 12 results and week 13 predictions for this installment of "Guess the Lines". This is a weekly feature in which I put my Vegas NFL rankings to the test by predicting the point spread of upcoming NFL games. I compare the accuracy of my prediction against those of Bill Simmons and Cousin Sal, as featured on their weekly NFL podcast. In addition, I also calculate an implied point spread from ESPN Chalk's weekly Vegas NFL rankings.

Here are the week 12 results. Cousin Sal was the winner for the second straight week, with a mean absolute prediction error of 1.22 points. My rankings and Bill Simmons' guesses were close behind. The ESPN Chalk rankings came in a relatively distant 4th.

Turnover Index - Week 12

Here are the Turnover Index picks and results for week 12. The Turnover Index is a simple betting strategy based on the theory that the market overvalues defensive turnovers when judging team strength. See here and here for more background.

There were no bets that satisfied our criteria for Week 11. We went 1-1 with our Week 10 bets, and our bankroll dropped by $7 from $1,060 to $1,052. We are now 5-3 against the spread on the season, with an ROI of 5.2%. Bets are sized according to the Kelly criterion.

Here are the 2016 season to date results:

weekbetswonstarting
bankroll
amount betprofitsending
bankroll
6 1 0 $1,000 $36 (3.6%) ($36) $963
7 1 1 $963 $10 (1.1%) $9 $973
8 1 1 $973 $38 (4.0%) $35 $1,008
9 3 2 $1,008 $64 (6.4%) $51 $1,060
10 2 1 $1,060 $15 (1.5%) ($7) $1,052

Wednesday, November 16, 2016

Guess the Lines - Week 12

Here are the week 11 results and week 12 predictions for the "Guess the Lines" prediction contest. This is a weekly feature in which I put my Vegas NFL rankings to the test by predicting the point spread of upcoming NFL games. I compare the accuracy of my prediction against the predictions of Bill Simmons and Cousin Sal, as featured on their weekly NFL podcast. In addition, I also calculate an implied point spread from ESPN Chalk's weekly Vegas NFL rankings.

Here are the week 11 results. Cousin Sal was the winner this week, narrowly edging out my rankings with a mean absolute prediction error of 1.07 points (compared to my 1.11). The ESPN Chalk rankings came in last, with an error of 1.46.

Saturday, November 12, 2016

Turnover Index Week 10

Here are the Turnover Index picks and results for week 10. The Turnover Index is a simple betting strategy based on the theory that the market overvalues defensive turnovers when judging team strength. See here and here for more background.

We went 2-1 with last week's bets, bringing us to 4-2 on the season, and an ROI of 6.0%. Bets are sized according to the Kelly criterion. Fortunately, our one loss was on a game that was just barely above the breakeven point, so it only cost us 0.4% of our bankroll. The model was more confident on our two winning picks, allowing us for a nice return on our 6.1% investment.

Here are the 2016 season to date results:

weekbetswonstarting
bankroll
amount betprofitsending
bankroll
6 1 0 $1,000 $36 (3.6%) ($36) $963
7 1 1 $963 $10 (1.1%) $9 $973
8 1 1 $973 $38 (4.0%) $35 $1,008
9 3 2$1,008 $64 (6.4%) $51 $1,060

Guess the Lines - Week 11 (plus Week 10 results)

The results are in. Who was right, the pundits or the models?

Last week, I launched a weekly feature that pits my Vegas NFL rankings against the instincts of Bill Simmons and Cousin Sal, as featured in The Ringer's "Guess the Lines" podcast. The podcast aired Monday, and I had results ready to go Tuesday, but other events interceded that night.

As a reminder, I am using my Vegas rankings for the NFL to predict the point spreads for future games. The rankings use point spreads from this season's games to reverse engineer an implied power ranking.

Opening lines tend to come out a week and a half in advance, and my rankings automatically update when they do. In order to make the test properly out of sample, I need to make my predictions almost a week in advance of The Ringer's Monday podcast. The predictions below are based on my rankings as November 1. The aGPF and hGPF columns are the away and home "Generic Points Favored" numbers from my rankings. The predicted point spread is just the difference between the two numbers plus a 2.5 point adjustment for home field advantage.

Sunday, November 6, 2016

Turnover Index Week 9

Here are the Turnover Index picks and results for week 9. I distributed prior picks this season via tweet, but now have time to put these in a proper blog post. The Turnover Index is a simple betting strategy based on the theory that the market overvalues defensive turnovers when judging team strength. See here and here for more background.

Our single week 8 pick was successful (Bears over Vikings), bringing us to 2-1 on the season, and an ROI of 0.8%. Bets are sized according to the Kelly criterion. We finished last season with somewhat mixed results. After factoring in results from our week 16 picks, we were 11-7-1 against the spread, but with a negative ROI of -4.8%, meaning we lost on bets where we were risking a greater fraction of our bankroll.

Here are the 2016 season to date results:

weekbetswonstarting
bankroll
amount betprofitsending
bankroll
6 1 0 $1,000 $36 (3.6%) ($36) $963
7 1 1 $963 $10 (1.1%) $9 $973
8 1 1 $973 $38 (4.0%) $35 $1,008


Tuesday, November 1, 2016

Guess the lines - Machine vs Men

I spend a lot of time in my car these days, thanks to a commute that takes me through the worst stretches of LA's 105 and 110 freeways. To fill that idle time, I listen to a variety of podcasts (and a Waze app that likes to talk over them at the worst possible times). A staple of my regular pilgrimage between Pasadena and El Segundo is the "Guess the Lines" podcast from The Ringer, in which Bill Simmons and "Cousin Sal" engage in a friendly prediction contest while previewing the week's upcoming NFL matchups. The goal of the contest is to see who is better at predicting how Vegas will set the line for each game.

I make my own guesses as I listen to the podcast, but it turns out my Vegas intuition is pretty poor. However, while I'm not a natural bookmaker, I do have a ranking system that was explicitly designed for this purpose. My betting market rankings for the NFL (and the NBA, WNBA, MLB, College Football, and College Basketball) are trained and calibrated on the Vegas point spreads, which I use to reverse engineer an implied power ranking.

So, I am going to put my ranking system to the test, and see how it fares against the instincts of Simmons and Cousin Sal, starting with week 10 of the NFL season.

Sunday, October 9, 2016

NFL Playoff Implications - Week 5

Here are playoff implications for week five of the NFL season. The purpose of this feature is to highlight games that have a significant impact on the playoff picture (see this post for background).

The playoff implications below are derived from a 50,000 round simulation of the remainder of the NFL season. I use my daily NFL rankings to simulate future games. I can group the simulation runs by the outcome of each game and then see how a team's playoff chances vary between the two groups. The interactive table at the bottom of the post will allow you to see corresponding results for any game or team. The results of the Thursday night game have already been taken into account.

Ranking Week 5 Games by Leverage

The table below ranks the week 3 games by total leverage. Leverage in this context is a measure of both how uncertain a game's outcome is (games between evenly matched teams have higher leverage) and how much the playoff picture swings as a result of that outcome.

Wednesday, October 5, 2016

NFL Week Four Power Ranking Roundup

10/7/2016 update: The original version of this post had the incorrect FPI rankings. The table has now been updated and some of the text as well. Thanks to Steve in the comments for pointing out the error.

For the past several years, I have archived various NFL power rankings following week four of the season. When the regular season is finished, I compare how accurate each ranking was at predicting wins for weeks 5 through 17. Over the past nine seasons, the most accurate ranking system was the Vegas-based ranking system I publish here on my site.

Here are the rankings:


  • SRS - What's known as the Simple Ranking System, which is based solely on margin of victory and strength of schedule.
  • DVOA - Football Outsiders' DVOA rankings, a proprietary ranking system based on analysis of detailed play by play data.
  • 538 - FiveThirtyEight's Elo rankings. The Elo ranking system was originally developed for chess, but has been extended to the NFL by Nate Silver.
  • ESPN - ESPN's weekly NFL power rankings.
  • Market - The betting market rankings found here at inpredictable.
  • FPI - ESPN's Football Power Index (stat-based model developed by Brian Burke).

  • In addition to these rankings, the table below has two additional columns:
    • Consensus: The consensus ranking derived by taking a straight average of each team's ranking.
    • Consistency: How consistently a team is ranked by the various systems.
    Here is the ranking table:

    team SRS DVOA 538 ESPN MARKET FPIConsensusConsistency
    DEN 4 4 1 1 5 1 1 5
    MIN 3 3 3 5 4 2 2 2
    PIT 2 5 4 3 3 4 3 3
    SEA 10 2 2 2 2 7 4 14
    PHI 1 1 9 7 9 6 5 17
    GB 8 6 6 6 6 3 6 4
    NE 12 17 5 4 1 5 7 30
    ATL 5 8 13 8 12 11 8 9
    CIN 11 12 10 10 10 12 9 1
    CAR 9 21 8 15 8 9 10 28
    BUF 6 10 11 20 16 10 11 26
    ARZ 15 16 12 17 7 8 12 21
    DAL 7 9 17 11 17 14 12 20
    OAK 13 7 21 9 14 13 14 25
    KC 19 15 7 18 11 15 15 24
    BAL 14 11 16 14 19 21 16 15
    NYG 16 20 22 12 20 20 17 16
    SD 18 13 24 22 21 16 18 19
    HOU 29 29 14 13 13 18 19 32
    WAS 20 14 20 19 22 24 20 13
    LA 25 24 15 16 23 29 21 29
    NYJ 27 32 18 21 15 22 22 31
    NO 22 18 26 24 28 19 23 18
    IND 30 27 23 26 18 17 24 27
    DET 21 26 19 25 24 28 25 12
    MIA 17 22 25 30 26 23 25 22
    JAC 31 19 30 23 27 26 27 23
    CHI 26 23 28 28 29 27 28 6
    TEN 28 25 32 27 31 25 29 8
    SF 23 28 27 31 30 31 30 10
    TB 32 30 29 29 25 30 31 7
    CLE 24 31 31 32 32 32 32 11

    The consensus #1 team is the Denver Broncos. The team ranked most consistently is the Cincinnati Bengals, with a ranking that varies from just #10 to #12. The least consistently rated team is the Houston Texans. ESPN and the market have them ranked 13th, while DVOA and SRS have them at #29.

    Overall, the rankings with the least disagreement are my market rankings and FiveThirtyEight's Elo rankings (Spearman correlation of 94%). The two rankings with the most disagreement are Football Outsiders' DVOA and the market rankings (Spearman correlation of 66%).

    Check back at the end of the regular season for the results. 

    Saturday, September 24, 2016

    NFL Playoff Implications - Week 3

    Here are playoff implications for week three of the NFL season. The purpose of this feature is to highlight games that have a significant impact on the playoff picture (see this post for background).

    The playoff implications below are derived from a 50,000 round simulation of the remainder of the NFL season. I use my daily NFL rankings to simulate future games. I can group the simulation runs by the outcome of each game and then see how a team's playoff chances vary between the two groups. The interactive table at the bottom of the post will allow you to see corresponding results for any game or team. The results of the Thursday night game have already been taken into account.

    Ranking Week 3 Games by Leverage

    The table below ranks the week 3 games by total leverage. Leverage in this context is a measure of both how uncertain a game's outcome is (games between evenly matched teams have higher leverage) and how much the playoff picture swings as a result of that outcome.

    Saturday, September 17, 2016

    NFL Playoff Implications - Week 2

    Here are playoff implications for week two of the NFL season. The purpose of this feature is to highlight games that have a significant impact on the playoff picture (see this post for background).

    The playoff implications below are derived from a 50,000 round simulation of the remainder of the NFL season. I use my daily NFL rankings to simulate future games. I can group the simulation runs by the outcome of each game and then see how a team's playoff chances vary between the two groups. The interactive table at the bottom of the post will allow you to see corresponding results for any game or team. The results of the Thursday night game have already been taken into account.

    Ranking Week 2 Games by Leverage

    The table below ranks the week 2 games by total leverage. Leverage in this context is a measure of both how uncertain a game's outcome is (games between evenly matched teams have higher leverage) and how much the playoff picture swings as a result of that outcome.

    Sunday, September 11, 2016

    NFL Playoff Implications - Week 1

    Playoff implications return for the 2016 season. The purpose of this feature is to highlight games that have a significant impact on the playoff picture (see this post for background).

    The playoff implications below are derived from a 50,000 round simulation of the remainder of the NFL season. I use my daily NFL rankings to simulate future games. I can group the simulation runs by the outcome of each game and then see how a team's playoff chances vary between the two groups. The interactive table at the bottom of the post will allow you to see corresponding results for any game or team. The results of the Thursday night season opener have already been taken into account.

    Ranking Week 1 Games by Leverage

    The table below ranks the week 1 games by total leverage. Leverage in this context is a measure of both how uncertain a game's outcome is (games between evenly matched teams have higher leverage) and how much the playoff picture swings as a result of that outcome.

    Wednesday, September 7, 2016

    NFL Market Rankings are Live

    NFL Vegas rankings are live and will update daily for the remainder of the season. These rankings are an attempt to reverse engineer what the market "thinks". Markets tend to be efficient, and betting markets are no exception. It is difficult to beat the accuracy of the Vegas point spread over any extended period of time. Based on my own research, these Vegas rankings outperform a broad cross-section of NFL power rankings when it comes to predicting future wins.

    The current version of the rankings reflect both the latest Week 1 lines and the opening lines for Week 2. Because there are not enough connections between teams with just two weeks of point spreads, I use the week 1-16 point spreads published by CG Technology (formerly Cantor Gaming) earlier this year to fill in the gaps. But once we have week 3 opening lines, the teams should all be connected and we can dispense with the outdated CG lines.

    Tuesday, September 6, 2016

    20 Years of WNBA Win Probability Graphs

    The NBA season is still some two months away, but the WNBA season is in full swing. Regular season play has now resumed following the Olympics shutdown, and the playoffs are just a few weeks away.

    One of my offseason goals this year was to extend many of the tools and analysis I had developed for the men's game to the WNBA. I started small last month with the addition of the WNBA to my suite of Vegas team rankings. Today, I have a much more substantial update to announce:

    The following tools and stats are available for the entire 20 year history of the WNBA:


    Building a Win Probability Model for the WNBA

    The nice thing about this project is that I didn't have to start from scratch. WNBA data is structured very similarly to NBA data, so in a lot of ways, I could just point my existing code and methodology to a new dataset. The WNBA win probability model was built using the same approach as the NBA win probability model. It is based on play by play data for over 2,000 WNBA games, going back to the 2007 season.

    Sunday, July 17, 2016

    Betting Market Rankings for the WNBA

    The WNB-Ays
    I have added the WNBA to my suite of betting market rankings, to go alongside those for the NBA, NFL, MLB, College Football, and College Basketball. The purpose of these rankings is to reverse engineer an implied power ranking from the Vegas point spreads, essentially distilling the combined wisdom of the market.

    Here are the rankings as of July 17:


    GPF stands for "Generic Points Favored". It is what you would expect a team to be favored by against a league average opponent on a neutral court. By combining the betting over/under with the point spread, I can decompose GPF into its offensive and defensive components, oGPF and dGPF (note: offense and defense are on a points allowed per game basis, rather than points per possession - there is no way to derive implied per possession metrics from the betting data). GOU stands for "Generic Over/Under" and it is what you would expect the betting over/under to be for that team when playing an average opponent.

    Saturday, July 16, 2016

    Bonus Tim Duncan Chart - Bank Shots

    On the occasion of Tim Duncan's retirement earlier this week, I used SportVU motion tracking data to call attention to an unnoticed element of his game: his low and tight, line drive shot arc.

    Using that same data, we can also delve into a more well known aspect of Duncan's shooting game: his bank shot. For the bank shots I am able to identify, the chart below shows where Duncan's bank shots hit the glass. For comparison purposes, I also have a chart for all NBA bank shots.
    Duncan seemed to favor the upper left portion of the glass, especially compared to the league average, which is clustered far more in the center, just at the top of the backboard's inner square.

    On an unrelated note, I noticed that there appears to be a bias to when a scorekeeper will classify a shot as a bank shot, and that bias is skewed towards made shots. I have found several examples where the motion tracking data clearly shows a missed bank shot, but the official shot description does not call it out. For that reason, I would be skeptical of any stat that shows a particular player is far more effective when shooting bank shots.

    Once I have the methodology cleaned up, I hope to use the SportVU create a deeper dive into the bank shot and the underlying physics (similar to last summer's post on the effects of air drag on a basketball's trajectory).

    Monday, July 11, 2016

    The weirdest thing about Tim Duncan

    Tim Duncan announced his retirement today, after 19 seasons in the NBA. On Duncan and his impact to the game, there is no shortage of articles, retrospectives, and in-depth analyses (statistical and otherwise) from which to choose today, most pre-written, no doubt.

    Duncan was famous, paradoxically, for not attracting attention to himself, and his retirement announcement was no exception. Contrast Duncan's brief, matter of fact press release (not even a press conference) to Kobe's season long, air ball ridden farewell tour. I wonder how far off the Onion's version of Tim Duncan is from the real thing: Tim Duncan Raving About Health Benefits of Standing Bench, Tim Duncan Around if any Spurs Have Questions About Sequester, Tim Duncan Sends Teammates Google+ Invitations for Fifth Straight Day (I will miss the Onion's Tim Duncan almost as much as Uncle Joe Biden).

    But behind Tim Duncan's staid, middle of the road public persona lies a hidden deviancy: his shooting arc.

    Using location data tracked by the NBA's SportVU camera system, we can analyze player shooting mechanics in exhaustive detail. For more background, see my introductory post on this topic from last year, as well as some more recent research on free throw shooting.

    Monday, June 20, 2016

    Inspired by last night's game 7, a new metric: The Tension Index

    Tightrope walkingPrior to Sunday night's game 7, I lamented the lack of in-game drama in this year's NBA finals. The average excitement index of those six games was 4.77, which at the time was the lowest average for the finals in the past 10 years. Game 7, however, delivered on the hype, registering an excitement index of 8.6, and bringing the series average up to 5.32.

    While 8.6 is well above the typical figure for an NBA game, it ranked 185th out of 1,316 NBA games this season, and was just the 13th most "exciting" game of these playoffs. But that doesn't feel quite right. There is a type of "excitement" that isn't necessarily captured by the excitement index.

    Sunday, June 19, 2016

    Exciting Finals, Boring Games

    Regardless the outcome in tonight's game 7, we are guaranteed a compelling story. Either the "jump shooting" Warriors put a capstone on their record setting 73 win regular season, or Lebron James and the Cavaliers end the longest championship drought of any major American city.

    At a macro level, the 2016 Finals has had more than its share of drama and excitement, allowing for the hot take cannon to swing wildly in all directions, firing indiscriminately as the outcome of each game seemed to flip the prevailing narrative.

    But when it comes to action on the court, the 2016 NBA finals have been the most "boring" of the past 10 years, at least by one measure. For each NBA game, we can use this site's win probability model to calculate an "excitement index". The index measures how much the win probability graph "travels" over the course of the game. It's a concept I stole adapted from the now-defunct win probability graphs from Advanced Football Analytics.

    The 2016 NBA finals have been largely devoid of any late game heroics that can lead to wild win probability swings. The average excitement index for this year's Finals is 4.77, which, barring a more exciting game 7, would be the lowest in the last ten years, beating the 2014 NBA Finals in which the Spurs beat the Heat handily in five games (average excitement index: 4.79). The most exciting championship round of the past 10 years was the 2011 Finals between the Heat and the Mavericks, with an average excitement index of 7.17.

    The most exciting NBA finals game of the past 10 years was game two of the 2015 NBA Finals, in which the Cavaliers stole home court advantage from the Warriors in overtime. The second most exciting Finals game was game 6 of the 2013 Finals, featuring Ray Allen's buzzer beater, amongst many other memorable plays.

    The chart below shows excitement index for each NBA Finals game from the past 10 seasons. We are definitely due for some late game drama.


    Monday, May 30, 2016

    Is pace contagious?

    Possessions, despite being analytical bedrock, are not an officially tracked NBA statistic. As a result, the counting of possessions in an NBA game has historically been indirect. We tease them out of the box score by counting the ways a possession can end, like physicists searching for the Higgs boson by tracking the particles into which it immediately decays[1].

    This indirect measurement has limited our ability to understand how pace works in the NBA, and who controls it. The box score can tell us that the Golden State Warriors play at a very fast pace, and squeeze in an above average amount of possessions into 48 minutes of regulation play. The Warriors are clearly fast on offense, but does that spillover into their defense? Do their opponents get caught up in the Warriors' hectic flow?

    Sunday, May 8, 2016

    Endgame strategy in the NBA

    Or how I learned to stop worrying and love the #quick2 (sometimes).

    When a team is trailing by three late in a game (e.g. less than 30 seconds left), are they better off going for a tie with the three pointer? Or is the superior strategy to attempt a quick, high percentage two point shot, and hoping for a turnover or missed free throws on the ensuing possession?

    The "quick two" approach draws plenty of derision from my analytics-heavy Twitter feed. Probably because it's emblematic of the conservative, risk averse thinking that mars strategic decisions across a number of sports. Football coaches punt too often on 4th down. Baseball managers still call for the sacrifice bunt, even though it reduces run expectancy. And NBA teams were historically slow to embrace the three point shot.

    But punts, bunts, and three pointers have all been thoroughly analyzed from a statistical perspective. As far as I know, no one has run the numbers on whether that quick two really is a suboptimal strategy. In this post, I will examine which strategy leads to victory more often. I'll start with teams that trail late by three points, and then look at the same situation, but when trailing by four.

    Sunday, April 24, 2016

    New box score feature: Pace and Efficiency Report

    I have added a new feature to my NBA Win Probability Graphs and Box Scores. If you click the "pace" tab, you will see a table that looks like this:


    Using the play by play data, this table summarizes offensive pace, as measured by seconds per possession. The "season" row shows each team's average offensive pace for the season. The "opponent" row shows how each team's opposing defense has controlled the offensive pace this season. Similar statistics are shown for scoring efficiency, as measured by points per 100 possessions. This is basically a game-specific version of the team summary pace tool I rolled out last season.

    The table above is from Game 3 of the 2015 NBA Finals, in which the Cavaliers took a (fleeting) 2-1 series lead over the Warriors. The Cavaliers' success in that game, and the game prior, was often credited to slowing the tempo of the notoriously fast-paced Warriors. But what the table shows is that while Cleveland certainly slowed their own pace on offense (from 15.9 seconds to 17.0), the Warriors were still playing their game on offense, mostly. They averaged 13.9 seconds per possession, just slightly above their season average, and still well below the league average of 15.1 seconds.

    Just a note on the possession counts: These do not precisely align with those you find on sites like basketball-reference.com. In general, I am counting more possessions on average than what is usually tallied using the box score, which results in slightly lower efficiencies. The main reason for this, I believe, is that my method counts "end of quarter" possessions that do not result in a typical box score "possession marker", like a made/missed shot or a rebound. If a team gets the ball with 10 seconds left and fails to get a shot off, that will likely not count as a possession using the box score stats, but is counted as such by my method.

    Eventually, I hope to add pace and efficiency for the other possession types to this feature: after made shot, after defensive rebounds, and after turnovers.

    Saturday, April 16, 2016

    The 2015-16 NBA Regular Season Review

    Using the various NBA tools I have built for this site over the years (see the sidebar on the top right), here are the top performers, top performances, top games, and team superlatives from the 2015-16 season.

    Player awards

    For the most part, I'm judging players according to win probability added (WPA), which is an admittedly limited metric (it only accounts for made/missed shots, free throws, and turnovers).
    • Most Valuable Player - Stephen Curry led the league with +10.59 win probability added. Kevin Durant came in a distant second with +8.01, an MVP-worthy total in just about any other season. Here is how Curry's progress compares against the top WPA finishers of the past seven seasons (an update to this post):
    • Least Valuable Player - Rajon Rondo. -4.96 WPA. Granted, this excludes Rondo's other box score contributions, such as assists (where Rondo ranks second in assist WPA) and steals, where he ranks in the top ten.
    • Most Improved Player - The player with the single biggest leap in WPA from 2014-15 to 2015-16 was Kevin Durant, but that "improvement" was due to his injury shortened season last year in which he played just 27 games. If we set Durant aside, the most improved player this year was.......Steph Curry. Curry won the MVP award last year with +5.75 WPA, and nearly doubled down on that mark this season.

    Saturday, April 2, 2016

    Playoff Seed Probability Motion Charts

    With an 82 game season, an NBA team's fortunes then to ebb and flow in increments, rather than huge leaps. This gradual evolution can make for some interesting motion-chart visualizations. For example, there is Aaron Barzilai's animation of the Warriors pursuit of the single season wins record. Or this visualization of the evolution of the win percentage of the NBA's top four teams.

    And as I have done the last couple seasons, here are motion charts that show how each team's playoff seed probabilities have evolved and shifted over the 2015-16 season. The probabilities are calculated using my NBA Vegas rankings, which update daily and re-project the remainder of the season and resulting playoff seeds. Time permitting, I will update the chart with the latest results, up until the end of the regular season. Just check back at this same post.

    Sunday, March 27, 2016

    How long does a rebound take?

    A deep dive into the minutiae of NBA timekeeping.

    I'm working on a post on endgame strategies in the NBA that I hope to roll out soon. As part of that work, I needed to figure out approximately how much time runs off the clock between a missed shot and a rebound. Rather than keep this scintillating information to myself, or perhaps placing it behind a paywall and charging exorbitant sums, I am providing my findings here, free of charge. And for those of you unfamiliar with the concept of a rebound, and what it entails, I refer you to this excellent tutorial from Baylor's Taurean Prince.

    Using play by play data from the current 2015-16 season, here are some basic statistics:
    • An average rebound takes 1.15 seconds. This is the average elapsed game time between the missed shot and the rebound (according to the play by play game logs)
    • 19% of missed shots are rebounded in "zero" seconds (i.e. the rebound is recorded in the same second as the miss)
    • 56% of missed shots are rebounded in one second
    • 19% of missed shots are rebounded in two seconds
    • 4% of missed shots are rebounded in three seconds
    • 2% of missed shots are rebounded in four or more seconds

    Sunday, March 20, 2016

    Gregg Popovich is a Timeout Trendsetter

    In a post on the length of NBA games, I noted a trend in the length of each minute. The average length of the 5:00 minute in the 1st quarter (between 5:59 and 5:00 on the game clock) has been dropping steadily, from an average of 2.9 minutes to 2.6 minutes. And on a related note, the average length of the 6:00 minute has been increasing.


    Thursday, March 17, 2016

    The odds for a perfect bracket this year are 1 in 12 billion

    As I did last year, I have used my betting market rankings to calculate an "optimal" NCAA tournament bracket. My ranking system attempts to harness the combined wisdom of the betting market, as revealed by the Vegas point spreads and totals. The rankings can also be used to calculate the odds that this so-called optimal bracket picks every game correctly. Last year, the odds were 6 billion to 1.

    This year, the odds are slightly less favorable, at 12 billion to 1, but that is still better than the pre-2015 average of 50 billion to 1. Here is how those odds break down by region and the final four:
    • Midwest: 163 to 1
    • West: 297 to 1
    • East: 191 to 1
    • South: 227 to 1
    • Final Four: 5 to 1
    I have created two versions of a populated bracket using my rankings:
    • inpredictable optimal - This bracket picks the best team in each matchup, according to my rankings. It's fairly chalk-y, though it does pick a couple 11 seeds, Gonzaga and Wichita State, to make it further than their seed would suggest. Kansas is the predicted champion.
    • inpredictable upsets - This bracket picks more upsets, but in a strategic way. The lower seed is picked as long as they are expected to be no worse than a two point underdog in the matchup. Michigan State is the predicted champion. 
    I have also used my ranking system to enter Kaggle's March Machine Learning Mania contest. Go team boooeee!

    Sunday, March 13, 2016

    Steph Curry is the MVP even if he doesn't play another game this season

    MVP debates across all sports tend to devolve into tiresome semantics. What does "most valuable" mean? Is it the best player? The player most valuable to his team? The best player on the best team? The player you'd most like to build a franchise around?

    From my admittedly biased perspective, I think a stat like win probability added is ideally suited for determining a season MVP. It is a narrative stat for what is a narrative award. It explicitly rewards clutch play and ignores garbage time contributions. Last December, I showed how Steph Curry's win probability added pace was well ahead of any recent precedent.

    Wednesday, March 9, 2016

    Free Throw Deep Dives: Launch Angle

    This is the first post in a planned series of deep dives into free throw shooting (and shooting in general). Using SportVU data, which tracks the position of the basketball 25 times per second in all three dimensions, and combining that with a simple physics model, I have built a database of some 250,000+ field goal attempts and 100,000+ free throws from the past three seasons of NBA play. From this database, I can create a variety of new statistics with which to assess shooting. These include:
    • Release angle (vertical) - at what vertical angle the ball leaves the players hand (higher angle = more arc)
    • Release angle (horizontal) - can tell you whether a shot was on target (e.g. wide left, wide right, or on target).
    • Release velocity - how fast the ball is going upon release
    • Release position  - where precisely on the court the player releases the ball. 
    • Release height - how high the ball is when released by the player
    • Approach velocity - how fast the ball is moving when it reaches the hoop
    • Approach angle - At what angle, relative to horizontal, the ball approaches the hoop.
    • Effective hoop area - A function of approach angle, it shows how big the hoop appears to the ball on approach
    • Approach position - where the ball crosses the plane of the hoop (i.e. a PitchF/x view of shooting).
    This initial post will focus largely on the first metric: vertical release angle.

    Sunday, March 6, 2016

    Hero Ball follow up - Inequality and Efficiency

    As a follow up on last week's hero ball post, here is how inequality correlates with shooting efficiency. In that original post, I used Gini coefficients to quantify how equally, or unequally, teams share shot attempts in clutch situations. This season, the Cleveland Cavaliers lead the league in hero ball with Lebron James accounting for 71% of the Cavs shot attempts in "double clutch" situations.

    But is that necessarily a bad thing? The chart below shows how team shooting efficiency (as measured by effective field goal percentage) correlates with inequality.


    Saturday, February 27, 2016

    The Hero Ball Index

    In last year's conference semis, Lebron James, with just 1.5 seconds left and the game tied, hit a 21 foot shot from the corner to give the Cavaliers a victory over the Bulls, 86-84. Said James after the game:
    I was supposed to take the ball out. I told coach, 'There's no way I'm taking the ball out unless I can shoot it over the backboard and go in.' So I told him to have somebody else take the ball out, give me the ball and everybody get out the way.
    In clutch situations, teams tend to run their offense through their stars (or in Lebron's case, the star runs the offense). At its best, you get iconic moments like Michael Jordan's game winner in the '98 Finals and Lebron's aforementioned heroics. At its worst, you get low efficiency desperation, clock draining iso play, and heavily contested, off the mark mid-range jumpers (sarcastically referred to as "hero ball").

    Earlier this season, I launched clutch shooting reports. This interactive report uses win probability to classify shots into four basic categories: garbage time, "normal" basketball, clutch, and "double" clutch. That fourth category, double clutch, represent the top 1% of shots in terms of potential impact on a team's chances of victory. The following chart summarizes these four phases as a function of score difference and time remaining.