With the help of several computers, I simulated the 2023-2024 NHL regular season a million times in order to estimate what is likely to happen. To estimate the probability of the home team winning each game I used my prediction model, Magnus. Curious readers will find lots of detail following that link, but, very briefly:
- I estimate the likely rosters for each team, starting from their contract list on
October 4, 2023, and then weighting by past playing time. For definiteness:
- All as-yet unsigned restricted free agents are expected to play a full season for their current teams;
- All players signed to tryouts are expected to not play;
- I estimate the individual impact of each player and head coach on:
- Shot generation and suppression, at even-strength and special teams, taking account of teammates, competition, and the effects of score, zone, coaching, and home-ice deployment;
- Rates of taking and drawing penalties
- Individual tendencies towards shooting or passing; and
- Ability to shoot or stop the puck, as appropriate.
- An adjustment for rest.
The method I used is strongly similar to the one I used last year, with some key improvements; some of the explanation is copied from last year's preview.
Team Summaries
Using the isolated abilities of their players and their coach, I can form an estimate of how each team will perform "in a vaccuum", that is, before considering the schedule of games, which determines opponents and fatigue. As is my usual habit, positive values mean "more shots" or "more goals" as appropriate, and thus on offence red (more than average) is desirable and on defence blue (fewer than average) is desired. Similarly, positive shooting impact is better for shooters and negative impact is preferable for goaltending; the axis for the goaltending distributions is flipped so that "better" appears to the right for both shooting and goaltending.
Stat Breakdowns
Using the above, I can make some summary graphics for the league as a whole.
Even Strength Shot Rates
The most basic result in the sport is the shot. To win more, a team will shoot often and from dangerous locations and with some measure of finishing skill. First, we look at even-strength shot rates, taking into account the shot's type and location. This gives an idea of "total offence" or "total defence", before we consider finishing talent and goaltending talent. I call this estimate expected goals, or xG for short. The zero point is set to what we saw in 2022-2023, of around 2.6 goals per hour of 5v5 play. The "NHL" button indicates the average of the team rosters as constructed currently. As usual, the league as a whole appears to be better than the average of the previous season; as injuries take roost in in their usual random way, the average this season should probably settle into an average close to last year, although the league has been gradually becoming more dense with offence in the past few years.
The better offensive teams are further to the right, the stronger defensive teams are near the top. The best two-way teams (at five-on-five) are in the top corner. The range of offensive talent, even at the team level, is broader than the range of defensive talent.
Special Teams Shot Rates
The same measurement above, but for special teams: power-play offence on the horizontal axis, and penalty-kill defence on the vertical axis. Power-play defence and penalty-kill offence are not considered here.
This season (like last season), the Oilers will have an exceptionally strong chance generation on the power-play.
The overall distribution of talent in power-play ability is a fair bit larger than the distribution of penalty-kills.
Penalties
Expected penalty differentials for teams are computed from expected icetimes for each player, multiplied by individual tendencies to cause their team to take or draw penalties, not merely individual rates.
The total magnitude of the pair of distributions is not very wide; the strongest team (Vegas) playing the weakest team (Arizona) can expect to obtain, on average, about one extra power-play in a typical game.
Shooting and Saving
In addition to the shot volume and quality, some skaters (by passing and by shooting) are better at converting chances than others, and, even more obviously, some goaltenders are better than others at preventing those chances from becoming goals.
Several changes that I made to my xG model, which I use to measure setting, finishing, and goaltending abilities, have the effect of making the team-level goal threat distributions similar to the team-level goaltending distributions, which I find curious. (There is no particular relationship between the two distributions that one might expect ahead of time, as far as I can see.)
For the sake of fair comparison, the abilities are here extrapolated to five thousand league-average shots, that is, to five thousand shots containing a distribution of shots that matches what was observed in the entire league last season. In fact each team will take and allow its own idiosyncratic pattern of shots, possibly accentuating team strengths or perhaps blunting them. Those interactions are handled in simulation below.
Notice in particular the striking confluence of weak finishing with excellent goaltending in NYI, they are likely to play very low-scoring games, allowing random chance to have a larger impact than one would expect from more run-and-gun teams.
The breakdown of minutes among the goaltenders themselves is somewhere more ad hoc, the proportions here are my own (somewhat) educated guesses, with some guidance from trusted confederates elsewhere who in exchange for their frank opinions have been provided with anonymity here.
Fatigue
All of the above is taken entirely from the team's roster and coach, without regard to the schedule. Of course, not every team has the same schedule, which affects how well they are likely to perform. Not all teams are equally affected by fatigue. The strongest effects from rest are seen when teams play after playing the night before against an opponent who has not done so. The table below shows how many times each team plays 'tired' in this sense, as well as how many times they play against a team that is tired. Games played between teams which are both tired in this sense are not counted, since neither team has a rest advantage.
This season, like last season, the Ducks have the most-favourable rest schedule, with a net-seven games more playing at rest advantage versus rest disadvantage. The Dallas Stars have the least-favourable rest schedule, playing at rest disadvantage seven more times than at rest advantage. The combined advantage (in terms of shot rates) of playing rested vs a tired opponent is approximately 11.5%, compared to a league-average rate.
Points and Playoff Chances
The most important factor of the schedule, though, is not rest but the fact that teams do not all play against the same opposition. To measure the effect of this imbalance, only simulation will suffice, the results of which are presented below.
Each team's bar is centred on the average point total obtained in the simulations for each team, sorted with the highest averages to the right, with the divisions indicated by colour. The changing colour intensities indicate "stanines", that is, each coloured square is half of one standard deviation high.
Thus, we expect:
- six or seven of the thirty-two teams to finish in the darkest box (very close to their name);
- ten or eleven teams to fall into the adjacent boxes;
- seven or eight to fall into the next pair of boxes;
- four or five into the next pair, and
- two or three teams to fall into the faintest boxes or even beyond.
The numbers in the table are the team's most likely finishing total (rounded to the nearest integer), as well as one standard deviation above and one standard deviation below this mean. The colours correspond to the four divisions that the league currently uses; Atlantic in maroon, Metropolitan in green, Central in purple, and Pacific in dodger blue.
This year I have added a wrinkle to my season simulations: a crude representation of the trades that are commonly observed near the trade deadline. Rather than trying to simulate specific trades, I assume that, in total, players with a net total impact of 2.5 wins will be traded from the league's weaker teams to the stronger teams during the period between February 24 and March 8 of 2024. Every team is simulated to be a buyer with a chance proportional to \( \exp(-t/2) \), where \(t\) is the rank of the team counted from the top of the league, and simulated to be a seller with a chance proportional to \( \exp(-b/5) \), where \(b\) is the rank of the team counted from the bottom of the leauge. This asymmetry means that the sellers, on average, are the very bottom teams of the league, while the buyers include a larger chunk of the middling teams as well as the usual top teams. The specific numeric choices here (2.5 wins, coefficients of 2 and 5) are taken "out of a hat"; a more careful modeller might derive them from more elementary principles somehow.
Divisional Breakdowns
| Oilers | 99.6 | 8.5 | 87% |
| Golden Knights | 95.3 | 8.6 | 73% |
| Flames | 93.9 | 8.6 | 68% |
| Kings | 93.2 | 8.6 | 65% |
| Kraken | 89.3 | 8.7 | 46% |
| Canucks | 88.8 | 8.7 | 43% |
| Ducks | 79.8 | 8.9 | 12% |
| Sharks | 76.1 | 9.0 | 6% |
The weakest division in the league this season is the Pacific; with only one team in the top quarter of the league and two very near the bottom. San Jose, in particular, look dire, having bereft themselves of most of the players who could have been considered good last season. Calgary are very strong defensively but will struggle to score; both things are true to a lesser degree about Los Angeles.
Every team's most likely finishing position is marked in the Position Chances graph. The Oilers are the presumptive favourites in the division, even ahead of the defending champion Knights; they have an 87% chance of making the playoffs, joint-highest in the league.
Playoff cutoff: 89.7 points
| Team | Mean points | Standard Deviation | Playoff Chance |
|---|---|---|---|
| Stars | 100.2 | 8.5 | 87% |
| Avalanche | 95.8 | 8.6 | 74% |
| Jets | 95.3 | 8.6 | 71% |
| Wild | 94.8 | 8.6 | 69% |
| Predators | 89.8 | 8.7 | 46% |
| Blues | 87.2 | 8.8 | 34% |
| Coyotes | 81.1 | 8.9 | 14% |
| Chicago | 76.1 | 9.0 | 5% |
Dallas are the class of the Central, with their exceptionally strong forward corps and goaltending. Colorado are broadly solid in every respect, Winnipeg and Minnesota not far behind but much more specialized (specifically, defensive) in style. Chicago should remain well down the bottom of the table, despite the addition of Connor Bedard.
Playoff cutoff: 92.3 points.
| Team | Mean points | Standard Deviation | Playoff Chance |
|---|---|---|---|
| Hurricanes | 100.4 | 8.5 | 84% |
| Devils | 98.1 | 8.6 | 77% |
| Rangers | 96.1 | 8.6 | 70% |
| Islanders | 94.7 | 8.6 | 63% |
| Penguins | 92.7 | 8.7 | 54% |
| Capitals | 87.0 | 8.8 | 28% |
| Flyers | 81.9 | 8.8 | 12% |
| Blue Jackets | 81.7 | 8.9 | 12% |
The metropolitan division is not quite as competitive as it was in recent years, with Philadelphia weakening and Columbus remaining stagnant, while Washington and Pittsburgh's strongest players are all firmly established on the declining side of the aging curve. The Rangers make an interesting stylistic comparison to Carolina at the top of the division; where the Hurricanes rely on forechecking and generating offensive volume at 5v5, New York rely on finishing talent and their very strong special teams. The Islanders will play exceptionally low-scoring games, with an excellent defensive roster, excellent goaltending, and very weak goal threat talents.
Playoff cutoff: 93.7 points.
| Team | Mean points | Standard Deviation | Playoff Chance |
|---|---|---|---|
| Bruins | 99.9 | 8.5 | 82% |
| Leafs | 95.8 | 8.6 | 67% |
| Lightning | 94.8 | 8.6 | 63% |
| Panthers | 94.2 | 8.6 | 60% |
| Senators | 92.4 | 8.7 | 52% |
| Sabres | 90.8 | 8.7 | 44% |
| Wings | 85.7 | 8.8 | 23% |
| Canadiens | 80.1 | 8.9 | 9% |
The strongest division in the league this year, again, is the atlantic—only two teams in the division have playoff chances substantially lower than 50%. Even assuming a significant step back from their performance last regular season, the Bruins remain favourites. An interesting age-dynamic plays out here also, where Toronto and (even more so) Tampa Bay have veterans for their best players, while Ottawa and Buffalo are instead led by their younger players. I do not directly project age-related changes in these previews, so perhaps the already-chaotic middle of the division will be even more tightly packed.
Playoff cutoff: 93.3 points.