2017-18 NHL Season Projection

2017-18 NHL Season Projection

In my previous post, I introduced the stat Player Overall Value, which is a metric that is intended to be an all-encompassing stat, establishing the value of a player. Using an all-encompassing stat paves the way to do a season projection for the coming season, using POV as the main factor. 

Method

The model is relatively simple. We use each player's POV from the previous two seasons, weighting more heavily for the more recent season. This is also combined with an age curve introduced by Evolving Wild in this post, in order to project what their value will be for the coming season. Once the values for the coming season are determined, they are plugged into projected lineups for the season. These lineups are primarily drawn from the projected lineups done by The Hockey News in their season preview, with a few adjustments based on recent roster moves. Each line is weighted differently for the projected amount of ice time they will receive, and these added together equal the team's total value. The formula for the team's total value is the following: 

.35*F1 + .3*F2 + .2*F3 + .15*F4 +.4*D1 + .35*D2 +.25*D3 + .5*G1

Once a team's total value is determined, win probabilities for each game in the season are established. These account for home/away, as well as the amount of rest each team has had. Then, once each game has established game probabilities, each team's season is simulated 1,000 times, and from there we get their expected points and playoff probabilities. 

Results

A few observations about the results of the simulation

• A few teams jump out at being modeled differently than how good they are perceived to be. The Los Angeles Kings missed out on the playoffs last year, and made no improvements to their roster for the upcoming season, besides the addition of an aging Mike Cammalleri. However, because of them being the best possession team in the NHL by the Corsi metric this year, they are projected to be much better than most expect them to be. The New York Rangers are the inverse of that, a team that is not generally one of the better teams by possession metrics, but are generally considered to be a safe bet to make the playoffs.
• The playoff odds for the Bruins, Kings, and Penguins are considerably higher than the other teams, and this is pretty easily explainable. The Bruins and Kings were the top teams by Corsi from last year, which explains a large portion of the POV. The Penguins are a top team by possession, but they also boast the best offense in the NHL and are favorites to repeat as champions. 
• Teams that boast defenses of defensive-minded defensemen suffer by this model. Since these players don't contribute much on the offensive end of the ice, they do not generally have high POV's. Therefore, players like Nicklas Hjalmarsson, Josh Manson, Mark Giordano, and Ryan McDonagh aren't fairly represented by this metric and model. 
• The biggest surprise in the simulation is the Carolina Hurricanes, a team that has been on the upswing but was on the outside looking in at the playoffs last season. The addition of Scott Darling, who is regarded very highly by POV, and a good defensive core could push this team into the playoffs for the first time since 2009. 

Limitations

As with every statistical model ever created, there are many limitations that cannot be solved by a statistical model. 

• Players with small sample sizes are difficult to project. While only two years of data is used, for players such as Auston Matthews, Patrik Laine, and Jakob Chychrun, are difficult to project because we only have one year of data to look at. Other players, such as Jake Guentzel, Ivan Barbashev, and Charlie McAvoy, have even smaller sample sizes than a year, making this task even more difficult. And with rookies, such as Nolan Patrick and Nico Hirschier, we have no data to look at. This essentially makes it a guessing game, which can have mixed results.
• This model cannot take into account injuries which are inevitable to happen over the course of the season. The only injuries I was able to take into account are injuries we already know of- Ryan Ellis of the Predators and Ryan Kesler of the Ducks being out until Christmas, and Jay Bouwmeester of the Blues being out for 3 weeks. Otherwise, there is no way of predicting who will get injured and what games they will miss, which will obviously have an impact on the accuracy of these. 
• These seasons are all ran independently of each other, which is why playoff chances don't exactly line up. To determine playoff chances, I took the median amount of wins it took to make the playoffs and figured out how many times each time eclipsed that amount of wins. If I was able to run the season with each team's results dependent on each other, then the playoff chances would line up. However, with a limited programming capability, this wasn't possible which is why we have the numbers we do.
• Another disadvantage to running each team independent of each other is the seemingly large amount of parity in the league. Each team is pretty close to each other, which we would not expect to happen in the actual season, even with the false parity the NHL's system creates. One factor in this is the incapability of predicting overtime games. Overtime losses in the NHL are worth one point, but strictly based off of win probabilities, we can't predict what games will go into overtime, and turning that into a guessing game would drastically alter the accuracy of the model. 

However, even with these limitations, there are major benefits to doing a model based on this. On a player by player level, we can individually assess the performance of players and what they will add to the team. Over the course of the season, when we look at game probabilities the day of games, we can adjust for injuries and player substitutions, and while that isn't reflected in this model, it will be evident in game-by-game probabilities. 

Conclusion

Every statistical model is flawed, and this is no exception. However, using statistics to predict an entire season is a trend that we should expect to grow, as statistics are becoming more and more popular. Time will only tell to see if the Los Angeles Kings can buck their downward spiral, and instead follow the results of this model to lead themselves to a President's Trophy at the end of the season. With the season starting on Wednesday, however, we will find out sooner rather than later.