Champions League league phase predictions: 36 teams, one table, unequal fixtures
By FootInsights · Published · 6 min read
A domestic league table is a fair document because everyone plays everyone, home and away. The Champions League league phase is not that document. Thirty-six clubs sit in a single standings table, but each of them plays only eight matches against eight different opponents drawn from a pool of thirty-five. Two clubs on the same points have not completed the same task. Any approach to Champions League league phase predictions that reads the table the way it reads a domestic one is measuring the draw as much as the football.
That single structural fact — an unbalanced schedule scored in a shared table — changes which methods work and which quietly break.
Why the league phase table is not a league table
In a 38-game domestic season, schedule differences wash out. Everyone faces the champions twice and the bottom club twice, so points become directly comparable and a table can be read at face value.
Over eight matches against eight distinct opponents, nothing washes out. The pool of thirty-five possible opponents spans continental champions and clubs that arrived through the qualifying rounds. Drawing four of the stronger group and four of the weaker one is a materially different assignment from the reverse, and both assignments are legal outcomes of the same draw.
The consequence is that points totals in this competition carry an error term the domestic table does not have. Before comparing two clubs, you have to subtract the schedule.
Strength of schedule: a worked example
Take two clubs that both finish the league phase on 12 points. For each of their eight fixtures, suppose a model outputs an expected points value — the probability-weighted average of 3, 1 and 0 for that specific match-up, home or away.
Club A’s eight fixtures produce expected points of 2.1, 1.9, 1.8, 1.5, 1.4, 1.1, 0.9 and 0.7. That sums to 11.4 expected points.
Club B’s eight fixtures produce 1.6, 1.5, 1.3, 1.2, 1.0, 0.9, 0.8 and 0.6, summing to 8.9 expected points.
Both clubs banked 12. Club A beat its schedule by 0.6 points; Club B beat its own by 3.1. The table shows them level. The schedule-adjusted reading does not: Club B extracted far more from a harder draw, and a model carrying that information into the knockout rounds should rate Club B the stronger side despite identical standings.
This is the everyday work of forecasting the competition. The raw table is an outcome; expected points against the actual opponents faced is the estimate of ability underneath it. The gap between the two is the part the draw contributed.
How the draw structure shapes Champions League league phase predictions
The schedule is not random. The thirty-six clubs are divided into four pots of nine, seeded by coefficient, and each club draws two opponents from every pot — one at home, one away. That constraint is useful, because it bounds the inequality.
No club can be handed eight fixtures against the weakest pot, and none can be buried under eight against the strongest. Every schedule contains exactly two opponents from each seeding tier. What varies is the spread within each pot, which is wide: the ninth-seeded club in a pot and the first-seeded club in the same pot can be a long way apart in real strength, and coefficient seeding is a historical measure that lags current ability.
So the pot structure caps the damage without removing it. A useful model treats each of the eight fixtures on its own merits — team ratings, venue, and the specific match-up — rather than trusting that “two from each pot” makes the draws equivalent. The same logic that separates a rating from a league position in our league coverage applies here with more force, because there is less football to average over.
The two cut lines that decide everything
The league phase sorts thirty-six clubs into three groups. The top eight advance directly to the round of 16. Clubs finishing ninth to twenty-fourth enter a knockout play-off round. Twenty-fifth and below are eliminated outright.
For forecasting, this means the interesting quantity is almost never a club’s predicted finishing position. It is the probability of clearing a cut line — and those two lines sit in very different parts of the distribution.
Around eighth place, the table is dense. With eight matches played, a large number of clubs cluster within a couple of points of each other, so a single result reshuffles a wide band of positions. Small probability differences matter there. Around twenty-fourth, the spacing is usually wider and the outcome is less sensitive.
Points thresholds for each line get quoted as though they were constants. They are not. They depend on how the draw distributed strength and how the results actually fell, and they have to be recomputed from the current standings rather than carried over. A forecast that reports “X points is enough” without saying which state of the table it was computed from is stating an opinion, not a probability.
Simulate the table, don’t extrapolate it
The right tool here is simulation rather than arithmetic. Extrapolating a club’s current points rate over its remaining fixtures assumes the remaining fixtures resemble the completed ones, which is exactly the assumption an unbalanced schedule violates.
The procedure that respects the structure is straightforward: estimate match probabilities for every unplayed fixture in the league phase, play out the entire set of remaining matches thousands of times, sort the full 36-team table under the competition’s own tiebreakers on each run, and count how often each club lands above each cut line. That yields a probability per club per outcome rather than a projected points total, and it naturally handles the cases arithmetic cannot — a club needing other results to go its way, or a tiebreaker decided by goal difference accumulated against opponents nobody else faced. The Monte Carlo approach to simulating a season is the same machinery, applied to a table where the fixtures differ by club.
Goal difference as the first tiebreaker deserves particular caution. In a domestic league it is a rough proxy for dominance because the fixture list is shared. Across unequal schedules it partly measures who drew the weaker opponents, and a heavy win over a qualifying-round entrant counts the same as a narrow one over a title contender.
Reading a published league-phase forecast
Three questions separate a forecast that has done the work from one that has repackaged the table.
Does it adjust for the opponents actually faced, or does it treat points as comparable across clubs? Does it publish probabilities for the cut lines rather than a single projected finishing position? And does it say what its numbers looked like against results afterwards?
The last one is the one to insist on. A probability is only meaningful next to a record of how previous probabilities performed, which is why our own forecasts are scored in public on the track record rather than summarised as a win rate. A competition with eight matches per club and two hard cut lines is exactly the setting where confident-sounding forecasts are hardest to check and easiest to publish — small samples produce dramatic-looking tables, and dramatic tables invite strong claims.
Treat the league phase as what it structurally is: thirty-six clubs, eight fixtures each, and a table that needs correcting before it can be read.