- Joined
- Jul 11, 2008
- Messages
- 1,905
- Reaction score
- 185
- Points
- 63
This guide is for bettors who understand xG as a concept and want to use it at the team level to identify specific divergences between current market pricing and underlying performance quality.
What xPoints Actually Is
Expected points is a straightforward concept with a slightly involved calculation. For each match, you take the pre-match win, draw, and loss probabilities - derived from the xG the two teams generated in that specific game - and calculate the expected points outcome. A team that created 2.1 xG against 0.8 xG in a match had roughly a 70% chance of winning, 18% chance of drawing, 12% chance of losing. Their expected points from that match are approximately (0.70 × 3) + (0.18 × 1) + (0.12 × 0) = 2.28 points. Sum those expected points across all matches played and you have the xPoints total.The actual points table is the sum of what actually happened. The expected points table is the sum of what the underlying performance deserved. The difference between them - across a large enough sample - is primarily variance. Goals scored from high-quality chances that happened not to go in. Clean sheets maintained despite allowing high-quality chances that happened not to convert. The goalkeeper who saved three one-on-ones in October. The striker who hit the post twice in November. All of this variance, accumulated over weeks, produces divergences between where teams sit in the actual table and where their underlying performance quality says they should sit.
The predictive logic follows from the statistical principle of regression toward the mean. Teams running significantly above their xPoints total are - on average, over a large enough sample - more likely to see their actual results converge downward toward their underlying performance level. Teams running below their xPoints total are more likely to see their results improve. The actual table is the noisy signal. The xPoints table is the cleaner one.
How Large Does the Gap Need to Be to Matter
This is the question that determines whether a specific divergence is exploitable or just within normal variance, and it's where most xPoints analysis goes wrong by either treating any gap as meaningful or setting thresholds so high that nothing qualifies.A rough calibration from the data: a gap of one to two points between actual and expected over a ten-game sample is within normal variance and doesn't carry strong predictive signal on its own. A gap of three to four points over the same sample starts to be meaningful. A gap of five or more points over ten games - a team five points above or below where their xG performance suggests they should be - is a strong signal that variance has accumulated to a degree where reversion is more likely than continuation.
The sample size matters significantly. A five-point xPoints gap over six games is interesting but could still be a small-sample fluke. The same gap over fourteen games is harder to explain as pure variance and carries stronger predictive weight. I'd generally want at least ten games before treating an xPoints divergence as a betting input, and I'm most confident in the signal between games twelve and twenty-five when the sample is large enough to be meaningful but the season still has enough remaining for the reversion to show up in bets.
Worth noting: the reversion is probabilistic, not guaranteed or immediate. A team five points above their xPoints total doesn't have their results mechanically corrected in the next five games. What changes is the probability distribution of their upcoming results - it tilts toward under-performance of the market's expectation more than it otherwise would. This is a tool for adjusting probability assessments, not a formula that produces certain outcomes.
The Teams Where Divergence Is Most Predictive
The xPoints gap is more predictive for some team profiles than others, and identifying where the signal is strongest focuses the analysis on the situations with most betting value.Teams whose goal record is dominated by finishing variance are the clearest case. A team with a striker who has been converting 25% of his shots when the historical average for that shot quality is 12% is running significantly above expected. The xPoints gap for that team reflects the unsustainable finishing rate of a single player. When the finishing rate normalises - which it almost always does over a large enough sample - the actual points will converge toward xPoints rapidly. This is the individual skill version of the variance story, and it's both common and consistently underpriced because the market responds to goals scored rather than to shot quality and conversion rate against expected.
Teams with unusual goalkeeper performance above expected are the mirror image. A goalkeeper saving 85% of shots while the post-shot xG against suggests he should be saving around 72% has been producing clean sheets and minimal conceding that his performance quality doesn't fully support. The team's defensive xPoints outperformance reflects goalkeeper variance more than defensive system quality. The market prices the defensive record without adequately discounting for the keeper's unsustainable save percentage contribution.
The teams where the xPoints signal is weaker are those with genuine quality asymmetries that the xG model captures imperfectly. A team with an elite set piece system - generating goals from corners and free kicks at a rate significantly above average - might legitimately outperform their open-play xG because the set piece xG is a real quality edge, not variance. Similarly, a team with an elite counter-attacking striker who regularly creates shots from breaks that don't look dangerous in xG terms but have unusually high conversion value for that specific player might have a genuine xPoints outperformance that's partly talent rather than luck.
These are exceptions to the general regression principle rather than reasons to abandon it. The work is in identifying which outperformers are genuine quality advantages and which are variance, and applying the reversion logic specifically to the latter.
The Expected League Table as a Ranking Tool
Beyond individual team analysis, the full xPoints league table provides a ranking of team quality that's more stable and more predictive than the actual table - particularly early in a season when the small sample makes the actual table especially noisy.In October, after eight or nine games, the actual Premier League table is often significantly reshuffled from what underlying quality would suggest. Three or four teams are sitting significantly above their xPoints rank because they've been finishing well or keeping unexpected clean sheets. Three or four others are below their xPoints rank for the opposite reasons. A bettor using only the actual table is pricing these teams incorrectly in opposite directions.
The xPoints table in October, by contrast, is a better reflection of the underlying quality ordering that was roughly established in pre-season assessments. It hasn't yet fully separated the genuinely good teams from the variance beneficiaries. But it's closer to the true quality ranking than the actual table, which means betting decisions anchored to the xPoints ranking rather than the actual ranking produce better calibrated probability assessments.
The practical application: before pricing a specific fixture, check both the actual league position and the xPoints position of both teams. If they're roughly the same, proceed with normal analysis. If they diverge significantly - one team ranked fifth in actual points but twelfth in xPoints, their opponent ranked twelfth in actual points but fifth in xPoints - you have a specific signal that the fixture is priced from a distorted quality picture and the true competitive balance is different from what the table implies.
Betting on the Divergence Specifically
Translating xPoints divergence into specific betting positions requires some care about which markets and which timings are most appropriate.Asian Handicap is the primary market. When you've identified a team running significantly above their xPoints total facing a team running significantly below theirs, the Asian Handicap for that fixture may be priced from the actual table positions rather than the underlying quality positions. A team that's third in the actual table but ninth in the xPoints table playing a team that's ninth in the actual table but third in the xPoints table is a matchup where the market's quality assessment diverges from the underlying data. The handicap for the actual-table leaders probably underprices the opposition.
Outright markets are the longer-horizon application, already discussed in the outright markets article earlier in this series. The xPoints table is the specific data that drives the "quality anchor" versus "narrative adjustment" distinction described there. A title contender whose actual points lead has been built significantly above their xPoints total is the definition of a team whose price has moved on narrative rather than genuine quality improvement.
Total goals markets are the third application. Teams running above their xPoints total are often doing so because they've been scoring more than expected from their xG - which means their attacking play has been more efficient than the underlying shot quality suggests. That efficiency is unlikely to continue. Their expected goal output in upcoming matches is closer to their xG than to their actual goal rate. Unders for a team overperforming their xG significantly can be a consistent directional input when the line is built around their actual goal average rather than their xG.
The timing is important. The xPoints divergence signal is most actionable in the period after enough games have passed to make the divergence meaningful - roughly from matchday ten onward - and before so many games have elapsed that the reversion has already begun to show up in results and the market has started incorporating it. The window between matchday twelve and twenty-five is where the analysis is typically at its most useful.
Is xPoints Actually Better Than Form
This is the specific comparison worth making directly because form - the last five or six results - is the most commonly used market input and the most obvious alternative to xPoints as a predictive signal.Form is a proxy for recent performance that has two specific weaknesses. First, it's outcome-based rather than performance-based. Five wins from five looks identical in form terms whether those wins were built on dominant performances with high xG or on scrappy results with minimal xG and exceptional goalkeeper performances. The quality of the wins is invisible in the form record.
Second, form uses a window that's too short to eliminate the variance that makes it misleading. Six games is not enough to reliably separate quality from variance. A team genuinely in terrible form looks similar in a six-game sample to a team having a normal variance run. A team genuinely hitting peak quality looks similar to a team finishing at an unsustainable rate. The signal-to-noise ratio in six-game form is low.
xPoints uses all available season data, weights each game equally regardless of when it occurred, and measures the quality of performance rather than the quality of the outcome. These properties make it a more stable and more predictive measure than form for the purpose of assessing a team's true quality level.
The research comparing xG-based performance metrics to actual results as predictors of future results consistently finds that xG-based metrics outperform actual results over medium-term samples of twenty or more games. Over shorter samples, the evidence is more mixed - which is partly why I'd caution against applying xPoints divergence analysis before matchday ten when the sample is too small.
Actually - I want to be careful about overstating this. xPoints is better than form as a quality anchor. It's not a complete system on its own. Squad changes, injury situations, tactical shifts, managerial changes - all of these affect underlying quality in ways that season-long xPoints doesn't capture. The value of xPoints is as the primary quality anchor that gets updated by genuinely new information rather than by result noise. Treating every result deviation from the xPoints prediction as an update to the xPoints assessment defeats the purpose.
Where to Get the Data
The practical accessibility of xPoints data has improved significantly in recent years and it's now achievable without paid subscriptions for most major European leagues.Understat is the most comprehensive free source. It publishes expected points tables alongside actual tables for the major European leagues, updated after each matchday. The specific comparison between actual rank and xPoints rank is directly visible on the site without any calculation required. It goes back several seasons which allows historical analysis of how well xPoints divergences predicted subsequent results in past seasons.
FBref carries xG data at the team level with sufficient detail to calculate xPoints manually if you want to weight or adjust the calculation differently from Understat's methodology. The team-level xG for and against by match is available for most competitions going back five or more seasons.
The Analyst and various football analytics Twitter accounts publish regular xPoints table comparisons during the season - useful for cross-referencing your own analysis and for tracking the divergences without building your own database.
The one practical limitation: different xG models produce slightly different xPoints results because they weight shot characteristics differently. Understat's model, FBref's model, and the models used by commercial analytics providers don't produce identical xPoints tables. The divergences between actual and expected are consistent in direction across models but differ in precise magnitude. Building your analysis around a consistent single source is better than mixing models, which can produce artificial divergences that reflect methodological differences rather than genuine performance gaps.
FAQ
Q1: How quickly does the market incorporate xPoints divergence information once it becomes visible in the data?More slowly than you might expect, for two reasons. First, the market's primary inputs are form, head-to-head, and league position - xG-based metrics are incorporated in the models of sophisticated operators but weighted less heavily than results in the standard pricing. Second, the general betting public uses the actual table as their quality reference, which means recreational money continues to flow in directions that the xPoints table would suggest are wrong. The combination of incomplete model incorporation and recreational money anchored to the actual table means xPoints-derived value takes longer to be arbitraged away than CLV-based value in main markets. This makes it a more durable edge than most mainstream analytical tools, though not an infinite one.
Q2: Does xPoints work better for predicting results against specific opponent types or is it general across all matchups?
The predictive value is most consistent when applied to matchups between teams of roughly comparable quality, where the match result market is genuinely competitive rather than reflecting a large quality gap. For heavy favourite versus heavy underdog fixtures, the favourite's xPoints position matters less because the quality differential dominates the variance regardless. The xPoints divergence signal produces its best signal-to-noise ratio in fixtures where both teams are within a few positions of each other in actual table terms - specifically because those are the matches where the market's pricing is most sensitive to the table position input that the xPoints data is correcting.
Q3: Should you adjust xPoints for strength of schedule, or is the raw comparison sufficient?
The raw comparison is a reasonable starting point but schedule-adjusted xPoints is more accurate for teams whose fixture lists have been unusually easy or hard in the period generating the divergence. A team five points above their xPoints total after playing five of their first ten games against bottom-half opposition has a different implication from a team five points above their xPoints total after a genuinely mixed fixture list. Understat's data lets you cross-reference the opponents faced in the period of over or underperformance. A team overperforming against weak opposition carries a stronger reversion expectation than one overperforming against a genuinely hard schedule. It adds complexity to the analysis but produces a better-calibrated output for the borderline cases.
