As the fans of the Colorado Avalanche can probably
attest, a team can ice a very similar lineup for two consecutive seasons, and
yet finish at a very different position in the standings. On average, since
2005-2006, the season-to-season variation of the number of points obtained by
NHL teams is of 11. A model of consistency like the Red Wings, which have not
finished two consecutive seasons during the last decade with a greater
difference than 10 points, represents the exception, not the norm.
When we spoke a couple of weeks ago, we quantified
that approximately 40% of the season-to-season variation of the number of
points obtained by an NHL team is attributed to that big and abstract concept
we call luck. This luck factor includes everything that can hardly be
controlled by teams, such as injuries, a bad schedule and missed calls by the
officials. The remaining 60% of the season-to-season variation of the number of
points is caused by a real change in the underlying quality of the team.
Now that we have split that luck vs quality contribution, there is one nice little thing we can do:
from the number of points that each team obtained last season, estimate the
probability that they will make the playoffs in 2016-2017. We will quantify
that probability using a technique called Monte Carlo simulation.
Let’s simulate the future… many,
A Monte Carlo simulation is pretty simple and is used
in a multitude of applications for risk assessment. In a nutshell, a Monte
Carlo simulation aims to, using historical data, predict all possible outcomes
and estimate how probable each of these outcomes is. This same methodology is
used by sportsclubstats.com to follow,
along the season, the probability that each team will make the playoffs based
on their current record and remaining schedule.
In our case, we want to estimate all the probable
number of points that each team can obtain next season based on how they performed last season and the historical season-to-season
variations in luck and skill.
To do so, we start with the number of points that each
NHL team obtained last season. Then, we randomly estimate the impact that luck
had on each team last season, the season-to-season variation of their
quality, and the impact that luck will have on the number of points that they
will obtain next season, which we accomplish using the normal distributions that we have calculated with our time-series models. We identify which NHL
teams make the playoffs considering these values of luck and variation of their
quality, and repeat these simulations a thousand times. Each of these
simulations represents a plausible outcome of the 2016-2017 season. If a given
team makes the playoffs in 250 of these 1000 predicted scenarios, we can
conclude that, based on recent NHL history, it has a 250/1000 = 25% chance of
making the playoffs in 2016-2017.
Without further ado, here is the probability that each
team will make the playoffs next season according to our new Monte Carlo friend:
At the top of the list come the Capitals, who made the
playoffs in 99% of the 2016-2017 scenarios simulated by Monte Carlo. To miss
the playoffs, they would have to finish 25-30 pts below last season. It is not
impossible (the Flyers finished the 2006-2007 season 45 pts below their
previous one, after all) but, given historical NHL season-to-season variations
in number of points, it remains highly unlikely.
At the other end, we have the Leafs, with a mere 2%
chance to improve their record sufficiently to reach the playoffs. The playoff
hope of the Oilers is a bit better, at 6%, explained by the inferior number of teams
in the western conference and probably its slightly weaker division.
The Canadian teams do not fare well by this metric. The
Senators lead the pack at 28%. Overall, if we play a bit with the numbers, we
can estimate that there is a 29% probability that, for two consecutive seasons, no
Canadian team reaches the playoffs.
As an added bonus, here is, for each team, the
probability that they will lead their division:
As expected, given how much harder winning a division
is than making the playoffs, less than half of the teams can realistically expect
to accomplish such a feat. For the others, all they can do is hope for a
2005-2006 to 2006-2007 Pittsburgh-like improvement, a once in a decade
In the end, the advantage of Monte Carlo simulation is
that it is entirely unbiased. It starts with last season records, and simply uses
recent history to identify which season-to-season variations of the number of
points are likely and unlikely. Therefore, it
provides a nice line in the sand for each team on which realistic expectations
of success can be established.
Obviously, if you feel strongly about some of the
transactions made during the summer or on the extent of good or bad luck that a
team faced last season, you may adjust some of these probabilities
accordingly. Personally, I see the 14% playoff probability for the Canadians as
a bit on the low side, as a healthy Carey Price should bump the playoff hope of
any team by a significant margin. Perhaps it is a side effect of being an
Oilers fan, but I find the 64% playoff probability of the Flyers to be a bit
high, as a team leaning on Nick Schultz appears to me as wanting a decent lottery
What we can be sure is that, beginning tomorrow night,
no more simulation will be required, as the real thing will begin. Have a good