Monday, November 20, 2017

What CoolStandings told us about the 2017 Astros

Every morning while I put the Hot Links together during the 2017 season, I would check FanGraphs' Playoff Odds page. Even back in April. I wouldn't always talk about it, because I didn't want to be the guy scoreboard watching in May. Once a week, usually on an off-day, I would post a run-down of where things stood, projection-wise. It's like getting on a scale, you're not really supposed to do it every day, so you don't overreact to things like water weight and normal ebbs and flows of...things that determine how much you weigh. But I do it anyway. Getting on the scale, and checking the Playoff Odds everyday. I'd check it against FiveThirtyEight and BaseballProspectus, and I'd scoff at the projection that didn't tell me what I wanted to know - that the Astros were locks to win the World Series. Go ahead, click that link above. You'll like what you see.

So how does it work? Reading the About section tells us this:
The FanGraphs Playoff Odds are an adaptation of the playoff odds originally created by Greg Agami and Sean Walsh. To generate the playoff odds we simulate each season 10,000 times. 

It's FanGraphs' tweaking of pythagorean theorem, recent results, depth charts, ZIPS projections, etc. It's updated every day in the middle of the night, so by the time I get to this part of the internet (usually around 5:30am), it's ready to release its fresh horrors.

So let's take a look, through the smug lens of hindsight, and see how the Playoff Odds reacted to the 2017 Astros' season. Here's how FanGraphs projected the season on the Eve of Opening Day:

AL West:
Houston: 90-72
Anaheim: 83-79
Seattle: 82-80
Arlington: 82-80
Oakland: 77-85

That's pretty much exactly how the AL West shook out. Anaheim actually won three fewer games, but I doubt the projections figured in how much time Mike Trout missed in 2017. The Rangers and Mariners tied, but with 78 wins, and the projections were off on Oakland by two wins. But that's pretty impressive.

Where else in the pre-season projections was FanGraphs remarkably wrong? They were -11 on the Astros. Who else either out- or under-performed by 10+ wins?

San Francisco's FG projection: 89-73
San Francisco's actual record: 64-98 (-25)

Detroit's FG projection: 82-80
Detroit's actual record: 64-98 (-18)

Mets' FG projection: 87-75
Mets' actual record: 70-92 (-17)

Arizona's FG projection: 77-85
Arizona's actual record: 93-69 (+16)

Milwaukee's FG projection: 70-92
Milwaukee's actual record: 86-76 (+16)

Yankees' FG projection: 79-83
Yankees' actual record: 91-71 (+12)

Houston's FG projection: 90-72
Houston's actual record: 101-61 (+11)

Toronto's FG projection: 86-76
Toronto's actual record: 76-86 (-10)

Minnesota's FG projection: 75-87
Minnesota's actual record: 85-77 (+10)

There are so many stories as to why the projections missed. Nine teams were off by 10 or more wins, four of them were in the National League, five in the AL. There were a group of teams that were within three wins of their pre-season projection. But injuries (cough BUMGARNER cough), regression, playing out of your mind for a month, etc. But let's just stop and note that, of the six divisions in MLB, FanGraphs' projection system got all six division winners correct.

So let's take a look at the Astros, specifically. 

Making the playoffs: 78.4%
Winning the division: 60.1%
Winning the Pennant: 20.0%
Winning the World Series: 10.3%

The Astros obviously had a good team coming out of Spring Training, eschewing Chris Carter for Yuli Gurriel; jettisoning Colby Rasmus for Josh Reddick; wife-swapping Jason Castro for Brian McCann. They'd be good, we knew that. If you at least make the playoffs in 7840 out of 10,000 simulations, things look alright. Then you see what happens in October.

But after sweeping the Mariners to start the season, the Astros' odds of making the playoffs jumped to 85%. Thanks to going 10-3 from April 10-25, their odds jumped from 77.6% to 92.3%. On the morning of May 18 the Astros' playoff odds were 98%, having won 15 of 19 games.

May 29: Astros' postseason chances hit 99% for the first time. So after the most incredible win of the season, the Memorial Day Miracle, the Astros make the postseason 9,930 out of 10,000 times. Which, after that win, seems low.

June 2: Having won 10 of 11, the Astros' projection to win the division hits 99% for the first time. October is an inevitability.

July 3: The Astros are a 100% lock to make the ALDS. The Astros have a 15-game lead after taking two of three from the Yankees at home (Ed. note: I was at the one loss. With The Wife, Freakin' Pat, and Tyler.)

July 16: The Astros' odds of winning the ALDS and making it to the ALCS peaked. The Astros were 62-30, projected for 103.5 wins, after taking two of three from Minnesota at home after the All-Star Break.

The trade deadline came and went and were described as "a disappointment would be an understatement." The Astros woke up on August 1 and were 57.4% to win their ALDS, 28.7% to win the Pennant (the exact same as Cleveland), and 15.2% to win the World Series. They were projected to go 102-60.

A 3-10 stretch through August 14 had them at 58.2% to win their ALDS, 28.5% to win the Pennant, and 15.3% to win the World Series. They would finish the month 8-7, so the Astros' 11-17 August didn't affect them too heavily. The numbers knew who the Astros were, and could account for a 28-game stretch (17% of a season) where they didn't play to the level they had shown prior.

(Side note: Cleveland's 22-game win streak didn't do as much as you would think for their postseason success odds. The streak started on August 24, when their odds of winning an ALDS series was 50.5%, 27.5% to win the Pennant, and 15.5% to win the World Series. By the time it ended on September 14 it was 62.4% to win their ALDS, 32.3% to win the Pennant, and 19.6% to win the World Series. Maybe that's significant, I don't know. But when the simulations take recent results into account, a 22-game win streak - winning 13.6% of the season in a row - shifted things, but not dramatically. I guess the take away is that we overreact too much on a game-by-game basis. This is not exactly breaking news.)

September 2: The Astros introduced Justin Verlander at MMP and won both ends of a double-header against the Mets in their first game back at Minute Maid Park following Harvey, pushing their record to 82-53. They were already at 100.0% to win the division, 61.3% to win the ALDS (highest since their 62.4% on July 16), 32.9% to win the Pennant, and 19.8% to win the World Series.

September 4: Following the Astros' 5th-straight win they post their highest odds of winning the Pennant and the World Series at 35.8% and 22.0%, respectively. The following day Verlander would make his Astros debut and start a string of 10 straight Astros wins in games he starts.

October 2: The Astros woke up three days out from starting the ALDS against the Red Sox in the ALDS with simulations showing them beating the Red Sox by a narrow 52.2-47.8 margin, lowest it had been since May 25th's 50.7% chance of advancing to the ALCS. They were AL Pennant winners in 26.3% of the simulations, which had not been that low since May 11's 25.8% chance. Houston won the World Series in 15.1% of FanGraphs' simulations, the lowest it had been since August 16, after the Astros had lost eight of 13 August games.

But let's take stock of ten playoff teams' "chances" of winning each round of the playoffs, at the outset of each round:

Win ALDS (as of October 2, 2017):

Cleveland: 64.0%
Houston: 52.2%
Boston: 47.8%
New York: 25.6%
Minnesota: 10.3%

Obviously FanGraphs favored Cleveland to win the ALDS over either New York or Minnesota (whoops). Houston was a narrow favorite over Boston, likely due to facing Chris Sale twice and Craig Kimbrel however many times it was necessary.

Win NLDS (as of October 2):

Los Angeles: 65.8%
Washington: 50.4%
Chicago: 49.6%
Arizona: 19.0%
Colorado: 15.2%

Arizona and Colorado were the Wild Card teams, so they had a steeper hill to climb. But the Dodgers are the clear favorites here.

Win ALDS (as of October 5):

Cleveland: 59.2%
Houston: 52.7%
Boston: 47.3%
New York: 40.8%

I think everyone in the world thought it would be a Houston/Cleveland ALCS, but the simulations said Cleveland would easily handle the Yankees.

WIN ALCS (as of October 12):

Houston: 59.3%
New York: 40.7%

Houston was the clear favorite, but obviously FanGraphs didn't plan on the Astros dropping all three in New York.

WIN NLCS (as of October 13):

Los Angeles: 53.8%
Chicago: 46.2%

Interesting to see this series as narrow as it was projected, given that the Dodgers handled the Cubs easily while the Astros actively tried to take years off our lives in the ALCS.

WIN World Series (as of October 22, after both pennants had been won):

Houston: 52.6%
Los Angeles: 47.4%

That said, after Clayton Kershaw won Game 1 - a game nobody really thought he was going to lose - the simulations shifted to 62.5-37.5 in favor of the Dodgers. After dominating Game 1, the Dodgers won four of the next six games (three of them at home) in 6250 of 10000 simulations.

So, after Game 2 (tied 1-1):

Houston: 51.6%
Los Angeles: 48.4%

When the Astros took Game 2 in LA, it essentially made the series a best-of-five with home-field advantage. The simulations shifted slightly in Houston's favor.

After Game 3 (2-1 Houston):

Houston: 70.8%
Los Angeles: 29.2%

I guess that Game 3 was crucial.

After Game 4 (Tied 2-2):

Houston: 53.7%
Los Angeles: 46.3%

The bullpen's blown save gave the Dodgers about 1700 extra series simulation wins, needing to win two home games to take the World Series.

After Game 5 (3-2 Houston):

Houston: 79.0%
Los Angeles: 21.0%

Biggest swing of the series, naturally. The simulations figured Verlander and Keuchel could get one win in Los Angeles.

After Game 6 (Tied 3-3)

Houston: 53.2%
Los Angeles: 46.8%

Maybe it was Keuchel vs. Darvish, but I was surprised that the Astros were favored in this game. The only time - again, in the simulations - throughout the World Series that the Astros didn't win the World Series in the majority of simulations was after Game 1.

This certainly does not mean that you should check CoolStandings next March and roll out to Vegas and bet everything on CoolStandings being right. I mean, go back and check the Indians' chances in the simulations at getting through the ALDS. But still, things turned out generally how they were simulated.