Marc-Andre Fleury is having an unbelievable postseason. His current Sv% of .947 doesn’t just lead all goalies in these playoffs, it’s actually the highest Sv% of any goalie in a playoff year since the 1960s (min 8 games) …with one important caveat: he has one round yet to play. I think the biggest question heading into the finals is, can we expect Fleury’s dominance to continue?
I’ve previously looked at the impact that Save Percentages have on team’s success in the playoffs. Tl;dr: it’s huge. One third of playoff series since 1984 featured a team with a PDO (their Sv% + their Sh%) over 1.035, and those teams are undefeated: 151W-0L. However, in the next round, these teams were just as likely to post a PDO below .965 (guaranteeing defeat) as they were to again top 1.035 (guaranteeing victory). Sv% and Sh% can have a huge impact on the results of playoff series, but they are highly influenced by luck over 4 to 7 games.
Now, the 2018 Knights are a very unusual team: they’ve had a PDO above 1.035 in each of their first 3 rounds. If PDO were completely random (it’s not completely random, but stay with me) and every team had a roughly 1 in 6 chance of a PDO over 1.035 in each series (the historical rate), the odds of any individual team doing this 3 times in a row would be less than 0.5%. Since the NHL moved to four best-of-7 rounds in 1987, there have been 31 playoff years (including this one), and 496 total teams competing. Every team with a PDO above 1.035 in a series has advanced, so we would expect about 2.3 teams since 1987 to reach the finals with a PDO over 1.035 in each of the first 3 rounds.
How many have there actually been? Vegas is the second. The first was Edmonton in 2006, which missed the chance to see if Roloson could sustain his dominance when he suffered a Game 1 injury.1 So again, I’m not saying PDO is 100% luck, but we’ve had about as many teams reach the finals with consistently dominant PDOs as we would expect if it were 100% luck. A hot or cold goalie will often decide a playoff series, but in the last 30 years, no goalie has ever sustained that level of dominance through 4 rounds. That’s an argument against expecting Vegas to keep riding their PDO high.
Fleury’s had a weird playoff history
After winning the cup in 2009 (with a mediocre .908), Fleury had 4 straight years of playoff Sv% below .900, and developed a reputation as a playoff under-performer. It’s perhaps poetic justice that he’s now having an all-time great postseason, but it also speaks to the unpredictability of playoff goaltending. After 2013, Fleury looked like an inveterate choker, and the last guy you’d want your playoff hopes to depend on. I would not say that today. But I also would not expect Fleury to repeat this performance in years (or even series) to come.
I wanted to get a sense of how Fleury’s Sv% has changed over the course of his career. Since he’s played 15 games so far this postseason, I tracked his rolling 15 game average Sv% throughout his career.
You can see it’s pretty volatile. It’s not uncommon for his rolling average to rise or fall by more than .030 within a 15-game stretch (the faint grey vertical lines are at 15-game intervals).2 That said, you can see that Fleury’s Sv% has been trending slightly up in recent years, and dips in the average below .900 are rare in the past 5 seasons. He also had the best regular season of his career this year. His current high is the third best non-overlapping 15-game stretch of his career, behind early this season, and early in the 2008 playoffs. Of course, peaks by definition are followed by declines, but you can see that when he last hit .947, he then dropped below .910 within 15 games.
I haven’t charted this out for other goalies, but I’m confident this kind of up and down is pretty typical. Years ago, Eric T found that goalies with a reputation for being either dependable or erratic are actually pretty similar in terms of the (in)consistency of their Sv%, and a goalie with a career average of .913 actually looks pretty similar in his ups and downs to a robot that has a 91.3% chance of stopping every shot.
Since what we want to know is whether Fleury’s last 15 games predict anything about his next 4 to 7, I sampled every season or postseason in which he’s played at least 22 games, and compared his Sv% over the first 15 games to his Sv% over the next 7. For years in which he played at least 44 or 66 games, I included a second or third sample.
You can see that historically, Fleury’s performance over 15 games tells us nothing about how he’ll preform in the next 7. This is expected: a goalie’s baseline skill level generally doesn’t change too radically over time, and 7 games (or even 15) is just a really small sample, highly influenced by luck. So my best guess for Fleury’s Sv% in the finals would be about his career average: .913. But it could be much higher! Or much lower.
The Knights have been very lucky
What does this say about the quality of the Vegas team? Well, they were pretty good in the regular season. They won 51 games, were 9th in Score-Adjusted Corsi, scored 43 more goals than they allowed, and had a modestly-positive PDO of 1.010. (Meanwhile, Washington had a below-average Corsi, with their 49 wins boosted by a PDO of 1.017.)
So in spite of their expansion draft pedigree, Vegas is a good hockey team. The fact that they’ve been cruising on a sky-high PDO doesn’t necessarily mean they would have lost one of the previous rounds without it. They’ve been lucky, no doubt, but that doesn’t mean they’re not also good. They just haven’t really been tested yet.
Update 30/07/18: What happened in the Finals
This is old news, but I wanted to record for posterity (hi, posterity!) how things turned out for Fleury. In a word: poorly. Vegas lost to Washington in 5 games, and Fleury’s Sv% was below .900 in all 5. By contrast, in his first 15 playoff games, he’d been below .900 only twice. The Knights’ PDO in the finals was .937 – well below worst PDO ever posted by a series-winner. They’re actually the first team in NHL history to be outside the Competitive PDO Zone in 4 straight best-of-seven series. Unfortunately for them, the last series was outside on the guaranteed-loss side.
If you like building narratives around sports results, you’ll say that Fleury choked under the pressure. If you believe in the gambler’s fallacy, you might say that Fleury was due to underperform. If you’ve spent as much time as I have staring at the charts above, you’ll conclude that Fleury and the Knights were really lucky for a while, and they they were unlucky.
1 The 1985 Oilers also were above 1.035 in the first 3 rounds, but their round 1 was a best-of-5. Also, the extreme concentration of talent on the dynasties of the 70s and 80s enabled them to regularly post PDOs above 1.020, in a way no modern team can match.
2 Note that the 15-game rolling average doesn’t represent his performance in brief (<15-game) playoff appearances very well. These are better represented just by looking at his Sv% in each playoff year.