YMori

Posted on Mar 1 • Edited on Mar 7 • Originally published at zenn.dev

Last Place to Champions: What Marcel Projection Reveals About 2021 NPB Yakult and Orix

#python #baseball #machinelearning #datascience

Background

Marcel projection predicts next season's performance by taking a weighted average of the past 3 years of individual stats. It's designed to show the continuation of current trends — which means it struggles when teams change dramatically. Projections include only players registered on a team's roster for the target year (players who left for MLB or retired are excluded).

2021 was exactly that kind of year. Both pennant winners came from last place the season before:

Tokyo Yakult Swallows: Last in 2020 → Central League champions in 2021
Orix Buffaloes: Last in 2020 → Pacific League champions in 2021

Using data from my NPB prediction app, I traced how Marcel handled these two teams — and what it couldn't see.

→ GitHub: https://github.com/yasumorishima/npb-prediction

Key Terms (for first-time readers)

Term	Meaning
Marcel method	A simple projection method using a 3-year weighted average of past stats (weights: 5:4:3, recent years weighted higher)
Pythagorean win%	Estimates win% from runs scored (RS) and allowed (RA): `RS^1.83 / (RS^1.83 + RA^1.83)`
wOBA	Weighted On-Base Average — a batting metric that weights different outcomes (walk, single, double, HR)
ERA	Earned Run Average — earned runs allowed per 9 innings
MAE	Mean Absolute Error — average prediction miss. Lower is better

What Marcel Said Before the Season

"Marcel projected wins" here means a pre-season prediction based on aggregating individual player projections into team run scoring and run prevention, then converting to win% via Pythagorean expectation.

Yakult

Year	Marcel Proj.	Actual W	Gap
2019	69.5 W	59	-10.5
2020	56.7 W	41	-15.7
2021	66.6 W	73	+6.4

Orix

Year	Marcel Proj.	Actual W	Gap
2019	66.2 W	61	-5.2
2020	55.8 W	45	-10.8
2021	69.7 W	70	+0.3

2020 was a shortened 120-game season

Two things stand out.

① Marcel's projection barely changed

Comparing 2020 to 2021 Marcel projections:

Yakult: 56.7 W → 66.6 W (~10 win increase reflects 120→143 game schedule recovery)
Orix: 55.8 W → 69.7 W (same)

The underlying win% estimate was essentially unchanged. Marcel uses a 3-year weighted average, so one last-place season doesn't move the needle much.

② What changed was the direction of the gap

Last-place year: fell well short of Marcel projection (-10 to -16 wins)
Championship year: matched or exceeded Marcel projection (Yakult +6.4, Orix +0.3)

The "last place to champions" story wasn't Marcel's projection suddenly rising — it was the year they actually achieved the level Marcel projected.

2021: A League-Wide Reversal Marcel Couldn't See

This wasn't just about two teams. Across all 12 NPB teams in 2021, the teams Marcel rated as strong collapsed, while teams Marcel rated as weak rose up. A league-wide reversal happened in both leagues simultaneously.

Central League (by win percentage)

Rank	Team	Marcel Proj.	Actual (W-L-D)	Win%	Gap
1st	Yakult	66.6 W	73-52-18	.584	+6.4
2nd	Hanshin	72.9 W	77-56-10	.579	+4.1
3rd	Giants	81.2 W	61-62-20	.496	-20.2
4th	Hiroshima	75.2 W	63-68-12	.481	-12.2
5th	Chunichi	69.1 W	55-71-17	.437	-14.1
6th	DeNA	69.9 W	54-73-16	.425	-15.9

Note: Yakult won fewer games than Hanshin (73 vs 77) but had 8 more draws and 4 fewer losses, giving them the higher win percentage for the pennant.

Pacific League (by win percentage)

Rank	Team	Marcel Proj.	Actual (W-L-D)	Win%	Gap
1st	Orix	69.7 W	70-55-18	.560	+0.3
2nd	Lotte	68.3 W	67-57-19	.540	-1.3
3rd	Rakuten	71.3 W	66-62-15	.516	-5.3
4th	SoftBank	75.1 W	60-62-21	.492	-15.1
5th	Nippon Ham	69.8 W	55-68-20	.447	-14.8
6th	Seibu	68.6 W	55-70-18	.440	-13.6

Marcel's top-rated team — Giants at 81.2 projected wins — finished with 61 wins (-20.2). Marcel's lowest-rated CL team — Yakult at 66.6 — won the pennant. The top 4 CL teams by Marcel projection all collapsed; the bottom 2 rose.

The Pacific told the same story: SoftBank, Nippon Ham, and Seibu (all rated as strong) fell 13–15 wins short. Orix essentially matched their Marcel projection (+0.3) to win the pennant.

"Last place to champions" wasn't a random fluke — it was part of a systematic league-wide reversal that Marcel couldn't anticipate.

Why the Gap? Player-Level Factors Marcel Can't See

Yakult 2021

Pitching (team ERA: 4.61 → 3.48)

Player	Change
Okugawa Yasunori	0 innings in 2020 → 105 innings, ERA 3.26 in 2021 ← outside Marcel's scope
Takahashi Keiji	ERA 3.94 → 2.87 (major improvement)
Shimizu Noboru	ERA 3.54 → 2.39 (settled as closer)
McGuff	ERA 3.91 → 2.52 (peak performance)

Offense (team RS: 468 → 625)

Player	Change
Yamada Tetsuto	OPS (OBP+SLG) .766 → .885 (comeback season)
Murakami Munetaka	HR 28→39, maintained OPS
Santana	New signing, OPS .877 ← outside Marcel's scope
Osuna	New signing, OPS .694 ← outside Marcel's scope

Orix 2021

Pitching (team RA: 502 → 500, but quality improved dramatically)

Player	Change
Yamamoto Yoshinobu	ERA 2.20 → 1.39, innings 127→194 (full breakout)
Miyagi Hiroya	0 innings in 2020 → 147 innings, ERA 2.51 in 2021 ← outside Marcel's scope
Tajima Daiki	143.1 innings of steady contribution

Offense (team RS: 442 → 551)

Player	Change
Sugimoto Yutaro	141 plate appearances, OPS .695 → 542 plate appearances, 32 HR, OPS .931 ← Marcel underestimated
Yoshida Masataka	OPS .992 — consistently elite
So Yuma	Expanded to 543 plate appearances

The Other Side: What Happened to the Teams That Collapsed

For the collapsed teams, Marcel projected a certain level of aggregate production from its top-rated players. Here's how much of that production actually materialized.

Team	Batters: Marcel proj. → Actual	Pitchers: Marcel proj. → Actual
Nippon Ham	3,617 → 2,371 PA (-1,246)	591 → 625 IP (+34)
Seibu	4,292 → 3,165 PA (-1,127)	886 → 721 IP (-165)
Giants	4,812 → 4,058 PA (-754)	840 → 744 IP (-96)
DeNA	2,480 → 2,752 PA (+272)	950 → 642 IP (-308)
SoftBank	4,312 → 4,188 PA (-124)	908 → 756 IP (-152)

Batters projected at 200+ PA and pitchers projected at 50+ IP

Marcel projects forward from registered-roster performance. When aggregate playing time for projected contributors falls short, the team-level projection ends up above reality.

The failure mode also differed by team: Nippon Ham and Seibu saw massive batter shortfalls while pitching roughly held. DeNA's batters exceeded projection, but the pitching staff fell far short. SoftBank and the Giants saw modest-to-moderate shortfalls across both.

Three Patterns Marcel Can't See

Pattern	Examples	How Marcel handles it
Young ace sudden emergence	Okugawa, Miyagi	Zero prior innings → excluded from calculation
Foreign player hitting it big	Santana, Osuna	No NPB data → excluded from calculation
Mid-career breakout	Sugimoto, Takahashi	Dragged down by prior low numbers → underestimated

Marcel projects the continuation of current trends. All three of these patterns represent changes that are invisible in prior-year data — Marcel's greatest weakness.

What This Means for 2026

Two teams in 2025 show a similar pattern:

Team	Marcel 2025 Proj.	Actual 2025	Gap
Yakult	72.2 W	57	-15.2
Lotte	66.2 W	56	-10.2

Yakult's 2025 gap (-15.2) is nearly identical to their 2020 gap (-15.7).

Marcel 2026 projections:

Team	Marcel 2026 Proj.
Yakult	64.3 W
Lotte	67.1 W

In 2020→2021, Yakult exceeded their 66.6-win projection by 6.4 wins. For 2026 Yakult to do the same, they'd need to outperform their 64.3-win projection by roughly 6–10 wins.

What would need to happen — outside Marcel's scope:

Young starter with near-zero 2025 innings becomes a rotation anchor (Okugawa pattern)
Foreign position player has a breakout first season (Santana pattern)
Existing player surges +200 plate appearances with dramatically improved numbers (Sugimoto pattern)

Marcel shows where teams stand based on current trends. When multiple "invisible" factors align, actual results can significantly exceed the projection.

Summary

Marcel projected both teams at roughly 65–70 wins regardless of whether they finished last or won the pennant
Last-place years: fell well short of Marcel projection (players underperformed, playing time dropped)
Championship years: matched or exceeded Marcel projection (factors Marcel couldn't see all aligned)
"Last place to champions" is less about the projection rising — it's about whether the invisible factors showed up
In 2021, both leagues saw a systematic reversal: Marcel's highest-rated teams collapsed, lower-rated teams rose

Marcel projection reveals the baseline. What it can't show is the breakout rookie, the new foreign star, or the player who finally puts it all together — the exact variables that turn a last-place team into a champion.