NPB Elo ratings 2026, and how to turn them into predictions

Updated 14 September 2026, built from 815 games with a recorded result this season, 20 of them tied. Average combined runs per game: 7.17.

Below are power ratings for all twelve Nippon Professional Baseball teams, built from this season's results and nothing else. No picks on this page, no odds, no affiliate links. The formula for turning any two of these numbers into a win probability is here as well, so you can price a matchup yourself rather than take our word for one.

A power rating answers a narrower question than the standings do. The standings tell you how a season has gone. A rating tries to tell you how good a team is right now, which is the only version of the question that helps before tomorrow's game.

The ratings

1500 is the starting point every team begins the season on, so a rating above it means a team has taken points off the field and below it means it has given them away. Run differential is there because it is the single most useful sanity check on a record: a team well above .500 with a flat run differential has usually been winning close games, and close games are the least repeatable thing in baseball.

Turning two ratings into a win probability

This is the whole formula. There is nothing hidden behind it.

We use 25 points of home advantage for baseball, which corresponds to a home team winning about 54% of otherwise even matchups. Worked example with two of the ratings above, a 1716 team visiting a 1538 team:

So the home side wins roughly 29% of the time, which is a fair price of about 3.41 and an implied 71% for the visitor. Compare that with the price a bookmaker is offering and you have the beginning of an opinion. Only the beginning: a rating built on results alone knows nothing about who is pitching tomorrow, and in baseball the starting pitcher is the single largest known variable before first pitch.

A rating is a prior, not a prediction. It tells you what the season says before anyone has looked at the lineup card.

Where this rating disagrees with the standings, as of 14 September

The commentary in this section was written on 14 September against the table as it stood that day. The tables above and below refresh from live results; these paragraphs do not. Check them against the current numbers rather than assuming they still hold.

Two rows are worth pulling out. The Yokohama BayStars sit third on rating with a record barely over .500, and the Hanshin Tigers sit sixth on rating with the second best run differential in the league. Those are the two largest disagreements between our number and the table.

The BayStars case is the clean one. They were 44-53 at the start of August and 23-12 since. A rating updates game by game, so a team playing well now outranks its own season record, which is exactly what you want if you are pricing tomorrow. Their rating has moved 108 points since 1 August, the largest swing in the league in either direction.

Where we think our own rating is wrong

Now the part that argues against the number we just published. Sort the same teams by the average rating of the opponents they have actually faced.

As of 14 September the BayStars had the weakest schedule in the league by this measure, and against the teams rated 1550 or better they were 2-7. So the surge that moved them into third came largely against the part of the league that is not very good, and the rating has not discounted it as hard as we think it should.

We are saying this about our own output because a rating you cannot catch being wrong is not worth much. If you use the ratings above, use the schedule column next to them. A number built on a soft month is a weaker number than the same figure built against the top of the table, and nothing in a single rating column tells you which one you are looking at.

How the rating is calculated

• Elo, updated after every finished game, from results only. No hand set ratings, no preseason opinion, no adjustment we make by feel.
• Margin of victory is weighted, so a six run win moves a rating more than a one run win. This is why a team can outrank its own record, and it is also why a hot month against weak opposition can overshoot.
• Every team starts at 1500 at the beginning of the season. Nothing carries over from last year.
• All Star games are excluded. Our data provider lists the July exhibition series with "Pacific league" and "Central league" as the two teams, which would otherwise put two fictional clubs in the table with a 1-1 record. We drop them by appearance count rather than by name, because the same field that marks those games also marks the Japan Series, and those we want to count.

What we actually bet in this league

Since this is a betting site, the fair question is whether any of it pays. Our NPB record is 56 settled bets. Thirty eight of them are totals, and those are the ones that have worked. The rest are moneylines in ones and twos per team, which is not a sample anyone should reason from.

Some honest context around that. Across every league we publish, our yield right now is slightly negative over 310 settled bets. NPB totals being our strongest slice is a statement about 38 bets, and 38 bets is not enough to prove a thing. It is the sort of number that turns into noise in either direction over the next month, which is why it sits on a public record that we do not edit rather than in a headline.

Worth one more line, because it shapes every totals line in this league: NPB is a low scoring environment. The average combined runs per game is in the header of this page, and it runs roughly three runs below the Korean league. A totals number that looks low next to a KBO game is usually just Japanese baseball.