Regression to the mean in betting

Regression to the mean is the tendency for extreme performances to be followed by more ordinary ones. Hot streaks cool and cold streaks warm, without anything changing.

Any observed result is part skill and part luck. When the luck component is unusually favourable, the next observation is likely to be closer to the underlying skill level, because the luck does not repeat. In football this shows up most sharply in finishing: a side converting 22% of shots will drift back toward the league norm near 10% to 12%, dragging its goal output with it.

Predicted next value ≈ mean + reliability x (observed - mean), where reliability rises with sample size.

A team has 14 points from six matches with 4.9 xG against 9 goals scored. The market prices their next fixture off the results. A bettor who prices it off the underlying xG expects the goals to fall back and takes the opposition or the Under — that disagreement is the whole trade.

It is the mechanism behind most contrarian value in football betting. Public money follows recent results; regression says recent extremes are the least repeatable thing in the data.

A newly promoted NPFL side on a five-match winning run is exactly the profile where local sentiment and price diverge most from the underlying numbers.

• Assuming regression means a reversal is "due" on the very next match.
• Applying it to a team that genuinely improved through a signing or managerial change.
• Regressing goals without checking whether the shot quality changed too.