Premier League predictions

The starting point is the market, not an opinion. Consensus probability is derived from roughly twenty-four bookmakers' 1X2 prices with the margin removed. The Premier League is the most heavily traded football competition in the world, which makes its odds unusually efficient: staring at statistics buys you nothing, because the same data is already in the price.

The model receives the table, recent results, head-to-head records and home advantage, but it is never allowed to guess a probability. It answers exactly one question: by how many percentage points is the market mispricing the home side, capped at ±6 points. The cap is deliberate - language models are measurably overconfident, and a free-form estimate produces systematically inflated probabilities.

The side, the best available price, the expected value and the stake are then computed in code. A tip appears only when expected value clears the publication threshold. In practice a Premier League round typically yields zero, one or two tips.

Home advantage is one of the few measurable regularities in football, and it has been shrinking. Before the 2000s home sides won roughly 46 per cent of Premier League matches; in recent seasons the figure has sat between 41 and 44 per cent. The empty-stadium season of 2019/20 made the crowd's contribution explicit: home wins dropped sharply once the noise and the refereeing pressure disappeared.

The practical consequence is that old rules of thumb overstate home advantage - and the market knows it. Edges are not found there. They appear elsewhere: in asymmetric team-news, in fixture congestion, and in how quickly money reacts to information.

• Overreaction to the last result: a heavy win or loss moves prices further than a three-match sample statistically justifies.
• European fatigue is priced crudely - a Thursday Europa League tie before a Sunday kickoff is not the same as a Wednesday Champions League night.
• Promoted clubs in the opening rounds: Championship data is a poor predictor and the market's own spread is wider than usual.
• Draws are rarely underpriced but frequently overpriced in evenly matched fixtures where both sides have a reason to play safe.

Every tip carries four numbers. Estimated probability is the market consensus adjusted by the model's deviation. Expected value states how much of a €100 stake is returned on average if the same situation repeated a thousand times. The suggested stake is a quarter-Kelly fraction, capped between 0.5 and 3 per cent of bankroll. Risk describes the uncertainty of the estimate, not the size of the price.

An expected value of +3% does not mean the bet wins. It means that over the long run bets like it return three euros per hundred on average - and the individual bet loses exactly as often as the probability says it will.