pointzerofive.

· Sports data · 116 coach changes · 2012/13–2025/26

Does Firing the Coach Work?

Bundesliga clubs that change coach mid-season usually improve afterwards. So do clubs in the same slump that keep their coach. We compared 116 coach changes since 2012 with those teams.

After a change 0.68 → 1.28 points per game, 6 games each side
Same slump, coach stayed 0.68 → 1.21 matched on points before
Difference +0.07 95% CI −0.05 to +0.18
Part 1 The bounce

Borussia Dortmund, 2017/18

2 points from 6 games. Then a new coach, and 12 points from the next 6.

Peter Bosz was replaced by Peter Stöger after game 15 of the season. Each dot is one league game: 3 points for a win, 1 for a draw, 0 for a loss.

The question

Did the new coach cause that, or would the team have recovered anyway?

A turnaround right after a coach change is easy to read as proof that the change worked. To test it, you need to know what happens to teams in the same position that keep their coach.

The data

116 coach changes in 4,284 Bundesliga games, 2012/13–2025/26.

We counted a change at the first league game a club played under a new manager during a season. 10 cases where the previous coach came back later the same season (suspensions, illness) are excluded. 91 changes have six league games on both sides within the season; the chart shows their average.

The bounce

Before the change: 0.68 points per game. After: 1.28.

That is close to double. In 80% of the 91 cases, the team took more points in the six games after the change than in the six before.

Timing

In the last game before the change, teams averaged 0.15 points.

Clubs tend to change coach right after a defeat, at the low point of a bad run. That timing matters for what comes next.

Part 2 Slumps end anyway

If form continued

If recent form simply continued, every team would sit on the dashed line.

A team with 4 points from 6 games would keep taking 0.67 per game. The league average over these seasons is 1.38 points per game.

Coach stayed

Teams with 4 points or fewer from 6 games went on to take 1.18 per game, without a coach change.

Each dot summarizes all six-game stretches with that many points where the coach stayed in charge for the next six games too (4,779 stretches in total, overlapping). Teams with no points at all from six games took 0.93 per game next.

The other end

Teams that won six in a row took 2.23 points per game in the next six.

The same pull works at the top. A team that wins six in a row is good, but rarely good enough to keep taking three points every week, so its results drift back down.

Regression to the mean

Extreme runs are followed by less extreme ones.

Six games are a short sample. A run of results reflects how good a team is and also luck: a deflected goal, a missed penalty, a red card. A very bad run usually includes some bad luck, and luck does not carry over, so the next six games tend to be closer to the team’s normal level.

Coach changes

The coach changes, in red, sit in the same cloud.

Each red dot is one change, placed at its points before and its points per game after (jittered slightly so they do not overlap). 77% of the changes came after 5 points or fewer, the range where teams recover anyway.

Part 3 The fair comparison

The comparison

Matched teams that kept their coach went from 0.68 to 1.21 points per game.

For each change we took every six-game stretch with the same number of points in which no coach change happened, and averaged what came next. Before the change the two lines have the same average by construction.

The difference

The coach changes did +0.07 points per game better.

That is about 0.4 points over six games. The 95% confidence interval runs from −0.05 to +0.18 points per game, so the data are consistent with no effect and with a small positive one. Measured as a share, 80% of teams improved after a change, against 76% of matched teams that kept their coach.

Other choices

Other windows and matching rules give estimates between +0.03 and +0.12.

We repeated the comparison with windows of 4, 8 and 10 games, matched on the season’s form so far as well as the last six games, and kept only changes made in a deep crisis. Every confidence interval includes zero.

Opponents

The schedule does not explain it either.

Opponents in the six games after a change took 1.37 points per game over their season, against 1.38 for the matched teams. New coaches did not get an easier run of games.

Summary

Most of the improvement after a coach change would have happened anyway.

If changing the coach helps in the short run, the effect is small: the upper end of the confidence interval is 0.18 points per game. Two earlier studies, on the Bundesliga and on the Dutch league, also found no measurable effect.

What the data shows

After a mid-season coach change, Bundesliga teams improve on average from 0.68 to 1.28 points per game. Teams that hit the same low point and kept their coach improve almost as much, to 1.21. The gap between the two, +0.07 points per game, is within what chance alone produces in a sample of this size.

The recovery itself is real. What it mostly reflects is timing: clubs change coach at the bottom of a run, and runs that bad rarely last. Any decision taken at that moment, including doing nothing, is followed by better results on average.

Earlier research

Heuer et al. (PLoS ONE, 2011) modelled Bundesliga team strength and found that firing a coach mid-season had essentially no effect on later results. Ter Weel (De Economist, 2011) reached the same conclusion for the Dutch league, 1986–2004, using a difference-in-differences design similar to ours. Our data cover a later period, 2012/13–2025/26.

Limitations

The data record who was in charge for each game, not why a coach left, so the 116 changes include resignations and departures by mutual agreement as well as dismissals. Clubs decide when to change coach, and they may know things the results do not show, such as injuries returning or a squad that has stopped listening; matching on recent points cannot account for that. The comparison covers six games after the change. It says nothing about whether the new coach was better over a full season or several. Changes in the first or last six games of a season are left out because one of the windows would be incomplete.

Data and methods

Sources. Match results and the manager of each team in every Bundesliga game come from the games table of transfermarkt-datasets (CC0 1.0; release of 5 September 2026; data from Transfermarkt.de). All 4,284 results were checked against openfootball/deutschland (CC0 1.0) and agree on every match date.

Coach changes. A change is the first league game under a different manager within a season. If the previous manager returns later in the season, the gap is treated as a stand-in and not counted (10 cases). A caretaker followed by a new permanent coach counts as one change. Of 116 changes, 91 have six games on each side within the season.

Comparison. Controls are all six-game stretches with no coach change in the six games before or after, matched exactly on points in the previous six games. The effect is the average difference in points over the next six games, divided by six. The 95% confidence interval comes from a cluster bootstrap with 2,000 replicates, resampling changes and resampling control stretches by club-season.

Reproducibility. Pipelines: notebooks/bundesliga_matches_pipeline.py and notebooks/coach_firing_pipeline.py. The notebook linked above recomputes every number in this story from notebooks/raw/bundesliga/.

Data: transfermarkt-datasets (CC0 1.0, data from Transfermarkt.de) · openfootball (CC0 1.0) · Heuer, A. et al., PLoS ONE 6(3), e17664 (2011) · ter Weel, B., De Economist 159, 279–303 (2011) · Last updated: 2026-10-14