Power Ratings 101: Using Elo and SPI to Price World Cup Matches

A single number — France's Elo of roughly 2080 against Australia's 1720 — tells you more about Saturday's match than any pundit's preview. That 360-point gap isn't an opinion. It plugs into a formula that spits out a win probability of about 89%, and that probability is the only thing you can actually trade against the price on Kalshi or Polymarket. With the 2026 World Cup kicking off June 11 in Mexico City and 104 matches repricing across 39 days, the traders who win will be the ones who can turn a soccer power ratings Elo SPI betting model into a fair price faster and more honestly than the market.

Power ratings are the cheapest edge in sports. They compress every result a team has ever played into one strength number, and the gap between two numbers converts cleanly into a probability. This article is the conversion mechanics: how Elo and SPI work, the exact logistic formula that turns a rating gap into a win probability, the home and neutral adjustments that matter at a World Cup, and how to use the output as a prior against the market rather than a prediction you marry.

What Elo and SPI actually measure

Elo is the older, simpler system — borrowed from chess and adapted for football. Every team carries a rating; after each match the winner takes points from the loser, with the amount scaled by the result's surprise and the margin of victory. Beat a team you were expected to beat and you gain little. Beat a team rated far above you and your number jumps. Over hundreds of matches the ratings settle into a stable strength ranking where the gap between any two teams is directly meaningful.

SPI (Soccer Power Index, the FiveThirtyEight-style approach) is the more modern cousin. Instead of one rating, it carries two ratings per team — an offensive rating (goals you'd expect to score against average opposition) and a defensive rating (goals you'd expect to concede). It blends recent results with underlying performance signals like expected goals, then converts the two ratings into a projected score line, and from there into win/draw/loss probabilities.

Both systems share the same virtue: they are self-correcting and opinion-free. They don't care that a team "feels due" or has a famous badge. They care only about results, weighted by who those results came against.

The logistic formula: rating gap to win probability

Here is the heart of it. For two teams with Elo ratings Ra and Rb, the expected result (win probability, treating a draw as half a win) for team A is:

P(A) = 1 / (1 + 10^(−(Ra − Rb) / 400))

That 400 is the Elo scale constant. It's what makes the system readable: a 400-point gap means the stronger team is expected to score 10-to-1 in the long run, a 200-point gap is about 76%, and a 100-point gap is about 64%. Equal ratings give exactly 50%.

Work the France-Australia example. The gap is 2080 − 1720 = 360. Plug in: 10^(−360/400) = 10^(−0.9) ≈ 0.126, so P = 1 / 1.126 ≈ 0.888, or about 89%. That single line of arithmetic is your fair price before you've read a word of preview.

The shape matters for trading. The curve is steepest in the middle, near a coin-flip, and flattens hard at the edges. That means a 50-point rating revision near an even matchup swings the probability by 7 points, but the same 50-point move when one team is already a 90% favourite barely registers. Your edge lives where the curve is steep — in the close, knockout-style matchups where small rating differences create exploitable price gaps.

One caution: the raw logistic gives a two-way (win-or-lose) probability. Football has draws. For a three-way moneyline you have to carve a draw probability out of that number — typically by assigning more draw mass to closer matchups — before you can price all three outcomes. That conversion is its own craft, covered in the three-way moneyline trading guide.

Adjusting for home advantage and neutral sites

A raw rating gap assumes a neutral pitch. Real matches aren't neutral, and home advantage in international football is worth roughly 70–100 Elo points — historically the equivalent of about two-thirds of a goal. You add that bonus to the home side's rating before running the formula.

At a World Cup, this matters more than usual because of the hosts. The 2026 tournament has three: the USA, Mexico, and Canada all play group matches at home or near-home, and the crowd, travel, and familiarity are real. A USA side rated 1820 on neutral ground might play at an effective 1900+ in front of a home crowd at MetLife or SoFi — enough to flip a 45% underdog into a near coin-flip. The host nations odds breakdown leans hard on exactly this adjustment.

For everyone else, most World Cup matches are functionally neutral — no home bonus for either side. But "neutral" still hides altitude and climate. Mexico City sits at 2,240 metres; teams unaccustomed to the altitude fade late, which a pure rating ignores. Treat those as manual overlays on top of the rating, not as something the base number captures.

Turning a rating gap into a tradeable price

Once you have a clean win probability, the rest is market mechanics. A Kalshi or Polymarket contract resolves to $1 if it hits and $0 if it doesn't, so the price in cents is the implied probability. Your model says France beats Australia 89% of the time; if the market is selling France at 84¢, you have a 5-point edge to buy.

Run your own number through the converter below. Type your model probability, see the fair American and decimal odds it implies, then compare to the screen price.

Now line your model up against the board visually. Here's an illustrative slate of group-stage matchups with the Elo-implied probability beside the market's price for the favourite to win — the gaps are where you trade.

Where the model bar is taller than the market bar — France, Spain, the host-boosted USA — the contract is cheap relative to your read. Where it's shorter, like Brazil-Serbia here, the market is overcharging and you'd rather fade than back. That gap, model minus price, is your raw edge before vig.

Use the rating as a prior, not gospel

Here's the discipline that separates a model-builder from a mark: your rating is a prior, and the market is data too. Power ratings are excellent at baseline strength and useless at last-minute information. They don't know that France rested four starters because they've already qualified, or that the opposing keeper is suspended.

So treat the workflow as Bayesian. Start from your Elo/SPI fair value. Then ask what the market knows that your rating doesn't — and only deviate from the no-vig consensus when you can name the reason. If your model says 89% and the market says 84%, the honest question isn't "the market is wrong," it's "what does the market see that my rating can't?" Sometimes the answer is nothing, and you have a clean edge. Often the answer is a team-news edge the rating is blind to.

Ratings also decay and drift. A team that overhauled its squad since the last competitive window is mis-rated until new results recalibrate it. World Cup squads are full of these — late-blooming attackers, a new manager's system. Lean less on the rating where the roster has churned, more where it's stable.

The real power of ratings shows up when you chain them across the bracket. A single match probability is one trade; feeding all 104 of them into a simulation gives you outright and stage-of-elimination prices for all 48 teams. That's the next step up — building the full Monte Carlo World Cup model — and for goal-level markets you'll want to pair Elo with an xG model so your totals and both-teams-to-score prices have the same rigour as your moneylines.

The team with the better badge wins the argument in the pub. The team with the better rating, priced against an honest market, wins you money over 104 matches. Build the number, respect the market, and only fire when the gap is real.