Defining Replacement Level in CPBL

Defining replacement level in the CPBL is an exceptionally challenging task. Before we dive into the numbers, we need to ask a fundamental question: does the concept of "Replacement" even make sense in the CPBL?

Consider this: in 2025, a total of 432 players appeared in the CPBL system. Of those, approximately 75%—310 players—played at the Major League level. The player pool simply isn't deep enough to directly transplant the MLB-born concept of "Replacement." One might reasonably argue we should use WAA (Wins Above Average) instead of WAR (Wins Above Replacement).

But WAA has its own problems. League-average starters can show negative values, which feels counterintuitive when evaluating player worth. So despite the conceptual challenges, we decided to calculate a CPBL-specific Replacement Level.

In 2010, FanGraphs' Piper Slowinski defined Replacement Level as:

"The level of production you could get from a player that would cost you nothing but the league minimum salary to acquire."

This doesn't mean taking the average of minimum-salary players. It refers to minor league free agents, waiver claims, and what baseball people call "AAAA players"—those who dominate the minors but can't stick in the majors.

That phrase—"AAAA players"—caught my attention. In CPBL terms, these would be players who put up solid numbers in the Minor League but struggle when promoted. If we can identify this group, their Major League performance becomes our Replacement Level.

Methodology

The methodology was straightforward. For every local player (excluding foreign players, who don't fit the "freely available" criterion), I calculated their wRC+ in both the Major and Minor Leagues. Using these two metrics plus each player's Major/Minor PA ratio as a third dimension, I ran K-Means clustering to find natural groupings in the data.

The optimal number of clusters turned out to be four. The results are visualized below—you can toggle between 2D and 3D views, and click on any cluster to see which players belong to it.

Results

The red cluster represents our replacement-level players: 59 individuals who posted solid Minor League numbers but struggled at the top level. Their median Major League wRC+ was 70.7—about 29% below league average.

For pitchers, I applied the same logic using FIP- instead of wRC+. Starters landed at 137.5 FIP- (37.5% worse than average), while relievers came in at 150.6 FIP- (50.6% worse). The larger gap for relievers makes intuitive sense—teams are more willing to give struggling starters a longer leash than relievers who can be replaced on short notice.

Validation

To validate these numbers, I constructed a hypothetical team of replacement-level players. Using our findings, this team would score 2.82 runs per game while allowing 5.65. The Pythagorean expectation gives them a 21.9% win rate—a 26-94 record over 120 games.

The match is remarkably close. The worst teams in CPBL history have won around 22% of their games—almost exactly what our model predicts for a replacement-level squad. This gives us confidence that our methodology is capturing something real.

A Note on Yearly Adjustments

One might wonder: shouldn't replacement level vary by year? After all, the CPBL expanded from 4 teams (2008-2019) to 5 teams (2020-2023) to 6 teams (2024-present). League difficulty should theoretically shift with such structural changes.

I explored this question extensively. I tried year-by-year clustering, rolling windows, EqA-style same-player tracking, percentile-based adjustments, and even machine learning approaches. The results were inconclusive—small sample sizes (as few as 8-30 players per cluster per year) produced wildly inconsistent estimates ranging from 56 to 82 wRC+.

Interestingly, the 5-team era (2020-2023) showed anomalously high replacement levels (around 79-82 wRC+). This appears related to COVID-19 travel restrictions that limited foreign player recruitment, combined with the Wei Chuan Dragons' re-entry creating unusual roster dynamics. When the league expanded to 6 teams in 2024, replacement level returned to historical norms (65-67 wRC+).

Given these complications and the lack of statistically robust year-over-year adjustments, I've chosen to use a single replacement level across all seasons. The pooled 2005-2025 data provides a stable baseline: 70.7 wRC+ for batters, 137.5 FIP- for starters, and 150.6 FIP- for relievers.

Conclusion

One final observation: CPBL's replacement level (21.9%) is significantly lower than MLB's (29.4%). This means the talent gap between replacement and average players is wider in Taiwan. It reflects what we'd expect from a smaller player pool and a less developed minor league system. In the CPBL, the drop-off from average to replacement is steeper—making every win above replacement that much more valuable.

These findings are now integrated into myCPBL's WAR calculations. Position players are measured against a wRC+ 70.7 baseline, starters against FIP- 137.5, and relievers against FIP- 150.6. It's not a perfect translation of an MLB concept to a different context—but it's grounded in CPBL-specific data and validated against historical performance.

For a league where three-quarters of all players reach the majors, that's about as close to "replacement level" as we can get.