FiveThirtyEight told readers to take the gap between two NFL Elo ratings and divide by 25 to get a point spread. The rule is easy to use. Test the rule against the betting market and the spreads run about 17% short.

Here is the claim in numbers. A 100-point Elo gap converts to a 4.0 point spread under the 25 rule. A market pricing NFL margins as a normal curve with a 13.5 point standard deviation puts the same gap at 4.8 points. The rule compresses every spread you build from Elo, and the error grows with the size of the gap.

This guide gives you the Elo formula, the FiveThirtyEight parameters, a conversion table, a derivation of the divisor the market implies, a worked rating update and a calculator. The guide closes with how far to trust any Elo spread.

The Elo Formula in One Line

Arpad Elo built his rating system for chess. The expected score for a player rated R_A against a player rated R_B is a logistic curve:

E_A = 1 / (1 + 10^((R_B - R_A) / 400))

The scale constant is a convention. A 400 point gap means the stronger side is ten times more likely to win than to lose, which is 91% (Chance, Statistical Analysis of the Elo Rating System in Chess). At a 200 point gap the stronger side scores about 0.76. ChessBase reports FIDE treats any gap above 400 as exactly 400 when computing expected scores (ChessBase, The Elo rating system: correcting the expectancy tables).

Chess has no home field and no point margin. Football has both, so the NFL version adds two pieces. A home bonus goes onto the home rating before you compute E. A margin of victory multiplier scales each update.

The FiveThirtyEight NFL Parameters

FiveThirtyEight published the NFL Elo setup most bettors copy. These are the parts you need.

  • Spread conversion. Take the difference in the ratings and divide by 25 (FiveThirtyEight, Introducing NFL Elo ratings). The later methodology page repeats the rule (FiveThirtyEight, How our NFL predictions work).
  • K-factor. The research found K = 20 predicts future games best.
  • Margin multiplier. LN(ABS(PD) + 1) × (2.2 / ((ELO_W - ELO_L) × 0.001 + 2.2)), where PD is the point differential and ELO_W and ELO_L are the winner and loser ratings before the game. The page adds an adjustment for autocorrelation, so bigger favorites earn a smaller multiplier.
  • Offseason. Teams keep two thirds of the prior rating and revert one third toward the mean.
  • Home field. Neil Paine, who built the model, put home field at about 65 Elo points, or 2.6 NFL points (Paine, NFL Elo ratings are back).

Look at the last item. Sixty-five points divided by 25 equals 2.6. The home bonus and the 25 rule come from the same scale. The market prices home field near 1.5 points today, as the home field advantage guide covers. If you copy the 65 point bonus into a model you compare to market spreads, you start 1.1 points above the market before any team is rated.

Table 1: Elo Gap to Win Probability and Spread

Every cell comes from the formulas shown. The third column applies the 25 rule. The fourth column asks what win probability a normal market with a 13.5 point standard deviation assigns to the 25 rule spread. Stern fitted a standard deviation of 13.86 to NFL margins in 1991, and Sides et al. report 13.5 as a recent value (Sides et al., Converting College Football Point Spread Differentials to Probabilities).

Elo gapElo win probabilitySpread by the 25 ruleNormal win probability at the spreadFair price from Elo
2553.6%1.053.0%-115
5057.1%2.055.9%-133
10064.0%4.061.6%-178
15070.3%6.067.2%-237
17573.3%7.069.8%-274
20076.0%8.072.3%-316
30084.9%12.081.3%-562

Read the 175 row. Elo gives the favorite 73.3%. A 7.0 point spread under a normal market gives 69.8%. The two models disagree by 3.5 points of win probability, and neither one has seen a game. The Elo curve is steeper than the market curve at every gap.

The Divisor the Market Implies

Reverse the table. Start with a market spread S and a standard deviation of 13.5. The favorite wins with probability p = Φ(S / 13.5), where Φ is the standard normal cumulative distribution. Then invert the Elo formula:

Implied Elo gap = 400 × log10( p / (1 - p) )

Divide the gap by S and you have the Elo points the market charges for one point of spread.

Market spreadNormal win probabilityImplied Elo gapElo points per spread point
153.0%20.520.5
358.8%61.720.6
461.6%82.520.6
769.8%145.520.8
1077.1%210.521.0
1485.0%301.521.5

The ratio holds between 20.5 and 21.5 across the whole range. A single divisor near 21 fits the market. The 25 rule does not.

The slopes confirm the gap. Near an even game the Elo curve at 25 Elo per point moves win probability 0.0360 per spread point. The normal curve at a 13.5 standard deviation moves 0.0296. The Elo curve is 22% steeper, so each rating gap converts to a spread 17% smaller than the market shows.

Take two teams rated 1580 and 1520 with the home team on the stronger side. The gap is 60 plus the 65 home bonus, or 125. The 25 rule gives a 5.0 point spread. A divisor of 20.8 gives 6.0. If the market opens at -6, the 25 rule tells you the home team is overpriced by a full point and the dog has value. The market and the model agree. Only the divisor disagrees.

How the Rule Misleads You

  1. Phantom underdog edges. Any favorite you rate with the 25 rule shows a smaller spread than the market. You see the dog as +EV on most games. The error scales with the spread, so the biggest phantom edges sit on the biggest numbers.
  2. The home bonus. A 65 point home field at 25 per point is 2.6. At 20.8 per point the bonus is 3.1. The bonus and the divisor must move together or your fair spread drifts.
  3. Key numbers. A normal curve has no spike at 3 or 7. The spread to moneyline conversion guide shows where the normal model fails at 7. Treat the 20.8 as a smooth average and expect real prices to bend around 3 and 7.
  4. The standard deviation. At 13.86 the ratio at a 7 point spread is 20.2. At 13.5 the ratio is 20.8. Pick a standard deviation from a dataset you trust and keep the value fixed.

A Worked Rating Update

Ratings drift unless you update them with the same scale you forecast with. Use the FiveThirtyEight update on one game.

Home team rated 1600. Away team rated 1500. The home bonus is 65.

  1. Adjusted gap: 1600 + 65 - 1500 = 165.
  2. Expected score: 1 / (1 + 10^(-165/400)) = 0.721.
  3. The home team wins by 7. Margin multiplier: LN(8) × 2.2 / (0.165 + 2.2) = 2.079 × 0.930 = 1.934.
  4. Rating change: 20 × 1.934 × (1 - 0.721) = 10.8 points. The home team moves to 1610.8 and the away team drops to 1489.2.

Now run the upset. The away team wins by 7. The multiplier uses the winner gap of -165 and gives LN(8) × 2.2 / (-0.165 + 2.2) = 2.248. The change is 20 × 2.248 × 0.721 = 32.4 points. A 7 point upset moves the ratings three times as far as a 7 point win by the favorite. The model punishes the upset three times harder.

How Far to Trust an Elo Spread

Elo has a real record and a limit. Hvattum and Arntzen tested Elo ratings for match prediction in association football and compared them with six other methods. Elo beat four of them. Two models built on bookmaker odds beat Elo (Hvattum and Arntzen, International Journal of Forecasting, 2010).

Glickman and Stern built a state-space model for NFL scores on 1988 to 1993 data. The authors report the model beat the Las Vegas line over the last 110 games of the 1993 season (Glickman and Stern, A state-space model for National Football League scores). The sample is small. Treat the result as proof the idea is sound, not proof of a durable edge.

Use Elo as a baseline, not a bet generator. Your edge comes from what Elo omits, which is quarterback status, injuries, rest and weather. Elo tells you where the market should sit with no news. The gap between your adjusted number and the price is the bet.

Calculator

Enter two ratings, your home bonus, a divisor and a market spread. Use 20.8 for a market-matched scale and 25 to see the FiveThirtyEight rule.

How to Use This

  • Divide Elo gaps by 21, not 25, before you compare them with market spreads. The market charges 20.5 to 21.5 Elo per point under a 13.5 point standard deviation.
  • Scale the home bonus with the divisor. 65 Elo at 20.8 per point is 3.1 points. Cut the bonus toward 31 Elo if you want the 1.5 points the market has used.
  • Subtract the vig before you read a win probability off a moneyline. The no-vig conversion guide shows the steps.
  • Adjust every Elo spread for the quarterback, injuries and weather before you act. Bet the adjusted gap, not the raw rating.
  • Log every Elo spread next to the closing line. After 100 games, fit your own divisor from the pairs.

Sources