Regression to the Mean: The Most Profitable Idea in Football Betting

Aston Villa took 65 points last season. Going by the chances they created and conceded, Understat's expected points model says they "deserved" about 51. That gap of almost 14 points was the biggest over-performance in the 2025/26 Premier League, bigger even than Sunderland's remarkable first season back.

Wolves were at the other end. They finished on 20 points from chances worth around 35 and went down. Luck didn't relegate them, but it turned a bad season into something that looked like a catastrophe.

The space between what happened and what the chances said should happen is where regression to the mean lives, and we think it's the most useful idea a football bettor can own. Saturday's price is built partly on the table and partly on what punters believe, and a run of luck can fool both. Learn to see through the run and you stop paying for luck that has already been spent.

It starts with Sir Francis Galton and his 1885-86 study of heights. Very tall parents tended to have tall children, but on average those children were a bit closer to normal height than their parents. Very short parents had children who were a bit less short. The extremes pulled back towards the middle, and statistics has used the word "regression" ever since.

Once you see why, it's obvious. Every result is part skill and part chance. Whoever tops a test on day one is good, but probably had a lucky day as well. On day two the skill turns up again and the luck mostly doesn't. The more luck went into an extreme result, the less likely that result is to repeat.

The whole rule fits in one line:

expected next value = average + r × (observed value − average)

r = how much of the spread is real, repeatable skill
r = 1  → nothing regresses, the result was all skill
r = 0  → straight back to the average, the result was all luck

Football sits somewhere in between. For some measures, finishing luck above all, r is much closer to 0 than most people assume.

The jinx that never was

Sport is full of superstitions that are really regression in fancy dress. The Sports Illustrated cover jinx is the famous one: an athlete makes the cover after an extraordinary spell, then has a more ordinary one, and the magazine gets the blame. Baseball has the sophomore slump, where a brilliant rookie season is followed by a duller second year. Nobody cursed anyone. The cover and the rookie award were both handed out at a peak, and peaks don't last.

The flip side is the regression fallacy, where whatever you did at the low point gets the credit for a recovery that was coming anyway. Speed cameras go up after a spike in crashes, the crashes fall, and the cameras take the bow. Football's version is the manager sacked after a terrible run. Hold that thought; we'll come back to it.

It all comes back to football being a low-scoring game. Premier League matches produced 3.28 goals a game in 2023/24, 2.93 in 2024/25 and 2.75 in 2025/26. The 2026/27 season has started at 2.82 (141 goals in the first 50 matches). When fewer than three goals settle most games, a deflection, the woodwork or a keeper's inspired afternoon can swing the result.

You can put a rough number on that noise. Think of every shot as a coin weighted by its expected goals (xG) value: a 0.11 xG chance goes in about 11% of the time over many repeats. Add up the randomness across a run of shots and you get this:

Goals from N shots, each scoring with probability p (its xG)
Luck variance = sum of p × (1 − p)

Illustrative 10-game spell: ~135 shots at ~0.11 xG each
  expected goals            ≈ 14.9
  one standard deviation    ≈ 3.6 goals of pure luck

Illustrative 38-game season: ~520 shots
  one standard deviation    ≈ 7 goals of pure luck

Now hold that up against reality. Across 60 Premier League team-seasons from 2023/24 to 2025/26, the actual spread of goals minus xG (after stripping out each season's league-wide offset) was 6.2 goals. That's roughly what luck alone would produce. Over a season, most of the difference between how teams finished their chances and how well they created them looks like noise rather than a shooting skill they can repeat.

Points say the same. The spread of points minus expected points across those 60 team-seasons was 7.7. A typical side ends the season about eight points away from what its chances were worth, and around a quarter of the league ends up further adrift than that. Eight points is the difference between Europe and mid-table.

These were last season's biggest gaps between the real table and Understat's expected points (xPTS). A positive number means the team took more points than its chances deserved.

Team	Points	xPTS	Points minus xPTS	Goals minus xG
Aston Villa	65	51.07	+13.9	−0.2
Sunderland	54	42.03	+12.0	−1.1
Fulham	52	45.08	+6.9	−2.6
Manchester United	71	64.45	+6.5	−3.2
Chelsea	52	58.85	−6.9	−14.2
Tottenham	41	49.25	−8.2	+1.2
Crystal Palace	45	53.88	−8.9	−21.5
Leeds	47	56.50	−9.5	−10.8
Wolves	20	35.44	−15.4	−11.4

Understat data, 2025/26 Premier League. Arsenal won the title on 85 points from 79.87 xPTS.

One quirk to keep in mind: Understat's xG runs high against actual goals across the whole league. In 2025/26 the average team scored 52.2 goals from 58.1 xG, so nearly everyone shows a negative "goals minus xG". Judge each team against that league offset of roughly six goals rather than against zero. Palace's −21.5 is still dreadful, but in fair terms it's closer to −15.6.

Now look at Villa and Sunderland. Neither finished especially well against their xG, so this wasn't a striker on fire. The extra points came from elsewhere: tight games won, late goals, opponents missing big chances. Those are exactly the things least likely to repeat. If you want a refresher on how the chance-quality numbers are built, our guide to using xG for smarter football betting walks you through it.

The test is straightforward. Take every club that played consecutive Premier League seasons between 2023/24 and 2025/26 (34 team-season pairs, with promoted and relegated sides left out) and ask whether one year's over-performance tells you anything about the next.

The falls after the highs

• Manchester United 2023/24: 60 points from 44.42 xPTS, a +15.6 gap and the biggest in the sample. The following season they took 42 points and finished 15th.
• Nottingham Forest 2024/25: 65 points and 7th from 50.01 xPTS (+15.0). In 2025/26 they managed 44 points and 16th, almost exactly on their xPTS of 42.24.
• Tottenham 2023/24: 66 points from 57.42 xPTS (+8.6). The next season, 38 points.
• West Ham 2023/24: 52 points from 41.28 xPTS (+10.7). The next season, 43 points.

It works the other way too. Brentford took 39 points from 52.94 xPTS in 2023/24 and bounced up to 56 the year after. Forest's 2023/24 haul was 14 points short of their chances before that jump to 7th. Squads and managers change, so not every swing is pure luck, but the table told a misleading story in both directions.

The numbers behind it

The eight biggest over-performers averaged +10.5 points above their xPTS, then −1.8 the following season. The eight biggest under-performers went from −9.6 to +1.1. Both groups snapped back to roughly zero.

Season-to-season correlation (34 Premier League team pairs)
  points minus xPTS          −0.13   → no persistence
  goals minus xG             +0.18   → slight at most
  goals conceded minus xGA   −0.07   → no persistence

Forecasting next season's points
                         correlation   avg error   RMS error
  using actual points        0.53       10.8 pts    13.9 pts
  using xPTS                 0.71        8.2 pts    10.2 pts

That bottom block is the punchline. Last season's expected points predict next season's real points better than last season's real points do. The table carries luck, and xPTS strips a good chunk of it out. It's a small sample built on one provider's model, but the direction is clear and it lines up with the theory.

Run Villa through the formula and the picture is stark. With r for the points gap close to zero, the best estimate of their underlying 2025/26 level is about 51 points' worth of chances, not 65. Villa aren't a bad side. But anyone pricing them as a 65-point team is paying for luck.

Regression isn't a "they're due" argument. A coin that lands heads five times in a row is no more likely to land tails next. What actually happens is that every new game is played at the team's true level, and over enough games that level drowns out the early freak results. That's the crucial difference from the gambler's fallacy.

Newly promoted sides

Sunderland were last season's feel-good story: 54 points as a promoted side from chances worth 42. Five games into 2026/27 they have 4 points from 7.69 xPTS, scoring 6 from 10.07 xG. The luck has flipped, and a side that looked overrated in the summer may now be underrated. Promoted teams throw up these swings all the time, so read our guide to betting on newly promoted Premier League teams alongside this one.

The new manager bounce

Clubs sack managers when results are at their worst, which is exactly when regression is about to kick in. Results pick up, the new man gets the credit and the board feels vindicated. Some of that improvement is genuine coaching. Plenty of it is a team drifting back to its level, and that would have happened whoever was in the dugout. We go through the evidence in does the new manager bounce really exist.

Goalkeepers and luck gauges

Over a short run, a keeper's save rate depends heavily on where the shots happen to be struck. A goalkeeper "in the form of his life" over six games deserves the same scepticism as a striker on a hot streak. Some analysts borrow PDO from ice hockey, which adds a team's shooting percentage to its save percentage. One side's shots are the other side's saves, so the league averages out around 100, and a team running well above or below that is probably riding luck one way or the other.

When the gap is real: Crystal Palace

Not everything regresses. Palace scored 3.1 goals above their xG in 2023/24, then fell 16.8 short in 2024/25 and 21.5 short in 2025/26. Two seasons in a row that far adrift, even after allowing for Understat's league-wide offset, tells you either the chances were worse than the model saw or the finishing genuinely lagged. Before you bet on a bounce, ask why the gap is there.

Leicester are the other reminder. The 5000-1 outsiders won the 2015/16 title under Claudio Ranieri, sealing it on 2 May 2016, with Jamie Vardy scoring 24 league goals and setting a Premier League record by scoring in 11 consecutive matches. Extremes rarely repeat, but every so often the extreme is the story.

Five rounds in, with the international break giving everyone a breather, these are the clubs whose gaps between the table and the chances matter most for the markets (Understat data to 20 September).

Team	Points	xPTS	Points minus xPTS	Goals / xG
Manchester City	15	9.59	+5.4	13 / 11.04
Newcastle	8	4.29	+3.7	9 / 5.54
Arsenal	12	9.81	+2.2	8 / 9.41
Tottenham	2	5.45	−3.5	2 / 5.82
Sunderland	4	7.69	−3.7	6 / 10.07
Nottingham Forest	5	9.17	−4.2	4 / 8.48
Bournemouth	3	7.76	−4.8	6 / 8.10

Manchester City have won all five, with Erling Haaland top of the scoring charts on five goals, and they've genuinely been good. They've also banked about five points more than their chances. That doesn't make them a lay. Regression pulls a team back towards its own true level, and City's true level is still very high. Their perfect start just flatters them a little.

Newcastle are the starker case: 8 points from chances worth barely 4. At the other end, Bournemouth, Forest and Sunderland have created enough to expect far more than they've got, and Manchester United's 5 points from 8.35 xPTS belongs in the same bracket. Bournemouth, Fulham and Spurs are all still winless after five, and Coventry, back in the top flight for the first time in 25 years, have scored once from 4.98 xG.

Five games is five coin flips. Use this table as a watch list, not a tip sheet. By the 10 to 15 game mark the xG picture firms up, and that's when the gaps become properly bettable.

Bookmakers know all about xG, and the market already prices some of this in. The edge sits where the public's view of form still moves the money. The aim is simple: don't pay a premium for a run the chances don't support, and don't give up on a good team because of a bad one.

Where the idea earns its keep

Match odds and Asian handicaps. A side that has won five on the bounce without outcreating its opponents is just the sort of team that gets shortened beyond its merits. The mirror image is a team winless in five but winning the xG battle, which is often the better-value side of a handicap line.

Goals markets. A team on a dry run despite healthy xG is still making chances. Totals and both-teams-to-score lines that lean on recent scorelines rather than chance quality can drift too low.

Outrights and finishing positions. Top-four, top-half and relegation markets in the first eight to twelve weeks lean heavily on the points table. That's where a 2025/26 Villa or a 2026/27 Newcastle can be overpriced, and where a slow-starting Bournemouth or Forest can be underpriced.

Goalscorer markets. A striker scoring well above his xG over a few weeks will usually come back to earth. Back the chances he's getting, not the goals he's already scored.

A simple process

1. Compare points with xPTS and goals with xG over the last six to ten matches and across the previous season.
2. Discount hot finishing heavily. Discount cold finishing less if the chances are still coming.
3. Ask why before you bet. A sold or injured striker, a new system, a set-piece specialist or a Palace-style persistent gap all change the picture.
4. Give last season's xPTS more weight than most punters do. It forecast the following season better than the actual table.

For the bigger picture on separating signal from noise, our guide to form, variance and luck in sports betting is the natural companion to this one.

• Trust xPTS over the table. Last season's expected points were a better guide to the next season than the points themselves, so build your view of a team from its chances first.
• Be wary of last season's overachievers. Villa (+13.9) and Sunderland (+12.0) banked far more than their chances in 2025/26. Price them on what they created, not where they finished.
• Fade the five-game wonders, carefully. Newcastle's 8 points from 4.29 xPTS is the kind of start that tends to cool. Pick your spots on handicaps rather than opposing them blindly.
• Back the creators who aren't scoring. Bournemouth, Forest and Sunderland have under-delivered on their chances so far. If the xG holds up over the next few weeks, the results usually follow.
• Don't confuse a good team with a lucky one. City are both right now. Regression drags them back towards their own high level, not towards mid-table.
• Ignore the bounce narrative. When a manager is sacked at rock bottom, part of the improvement that follows is plain regression, so don't pay extra for the "new manager" headline.
• Always check the why. Palace show that some gaps persist. Look for a real cause before assuming the luck will turn.

Keep your stakes level and modest while you test these ideas over a meaningful number of bets. Regression rewards patience, not big swings, and the same maths explains how many bets it takes to tell skill from a lucky run.

• Form vs variance vs luck in sports betting
• How to use xG for smarter football betting
• Does the new manager bounce really exist?
• Betting on newly promoted Premier League teams
• Luck vs skill: which sports are the most unpredictable?
• How to tell if a tipster is actually good

18+. Please gamble responsibly. For free, confidential support visit BeGambleAware.org.