What Is an Elo Rating and Where Did It Come From?

An Elo rating is a numerical system for calculating the relative skill levels of players or teams in competitive games and sports. Named after its creator, Arpad Elo, the system assigns each competitor a rating number that fluctuates based on match results. Unlike absolute measures of skill, an Elo rating is fundamentally a predictive tool—it estimates the probability that one player will defeat another based on their respective ratings.

The system operates on a zero-sum principle: when one player gains rating points, their opponent loses an equivalent amount. This fairness ensures that the total rating pool remains balanced and that rating changes accurately reflect competitive outcomes.

Who Was Arpad Elo and Why Did He Create This System?

Arpad Emmerich Elo (1903–1992) was a Hungarian-American physics professor and chess master who revolutionized competitive rating in the 1960s. Before Elo's innovation, the U.S. Chess Federation used the Harkness rating system, which relied on subjective tournament performance assessments. Elo recognized the flaws in this approach and developed a mathematically rigorous alternative based on statistical probability theory.

In 1960, the U.S. Chess Federation officially adopted Elo's system. By 1970, the International Chess Federation (FIDE) followed suit, making it the global standard for chess ratings. Elo's genius lay in creating a system that was simultaneously simple to understand, mathematically sound, and highly predictive of actual game outcomes.

Arpad Elo himself acknowledged the limitations of any rating system, famously stating: "The measurement of the rating of an individual might well be compared with the measurement of the position of a cork bobbing up and down on the surface of agitated water with a yardstick tied to a rope and which is swaying in the wind." This humble recognition underscores that Elo ratings, while powerful, are probabilistic estimates rather than absolute measures.

How Has the Elo Rating System Evolved Since 1960?

Since its inception, the Elo system has expanded far beyond chess. While the original chess application remains the most refined and widely recognized use, modern adaptations have emerged to address specific competitive contexts:

• Glicko and Glicko-2 (1995+): Chess.com and other platforms use these variants, which incorporate a "Rating Deviation" (RD) to account for uncertainty in a player's true skill level.
• TrueSkill (2006): Microsoft's Bayesian approach, used in Xbox Live and esports, handles team-based competition and complex multiplayer scenarios.
• Sports Betting Models: FiveThirtyEight's NBA Elo model and soccer prediction systems adapt Elo with margin-of-victory adjustments and home-field advantages.
• Esports and Gaming: League of Legends, Dota 2, and competitive programming platforms use Elo-based ranking systems.
• AI Evaluation: The ChatGPT Arena uses Elo ratings to rank and compare large language models in competitive evaluation.

This evolution reflects the system's versatility and enduring appeal across diverse competitive domains.

How Does the Elo Rating System Actually Work?

The elegance of the Elo system lies in its core mechanism: rating adjustments are proportional to the difference between expected and actual outcomes. Understanding this principle unlocks the entire system.

The Core Mechanism: Expected Score and Rating Adjustment

The fundamental insight of Elo ratings is that a player's rating predicts their win probability against any opponent. A rating gap of 100 points corresponds to approximately a 64% win probability for the higher-rated player. A 200-point gap translates to roughly a 75% win probability.

This relationship is captured by the expected score formula:

Expected Score = 1 / (1 + 10^((Opponent Rating - Your Rating) / 400))

For example, if you're rated 1600 and face a 1700-rated opponent:

• Expected Score = 1 / (1 + 10^((1700 - 1600) / 400)) = 1 / (1 + 10^0.25) ≈ 0.36

This means you're expected to score 36% of a point (roughly a 36% win probability). If you win (actual score = 1), you've exceeded expectations and gain rating points. If you lose (actual score = 0), you've met or fallen short of expectations and lose fewer points than you would against a lower-rated opponent.

What Is the K-Factor and Why Does It Matter?

The K-factor is a constant that determines the maximum number of rating points a player can gain or lose in a single game. It's the "volatility control" of the Elo system.

Rating Change = K × (Actual Score - Expected Score)

The K-factor varies based on a player's rating and experience level. According to FIDE (International Chess Federation):

Why does K-factor matter?

A higher K-factor increases volatility, allowing ratings to change more dramatically with each game. New players have K=40 because their true skill level is uncertain; large rating swings help them find their accurate level quickly. Once a player reaches elite status (K=10), their rating stabilizes, preventing wild fluctuations that would undermine confidence in the rating.

Some betting models and esports platforms use different K-factors. For instance, FiveThirtyEight's NBA Elo model uses K=20 for regular season games and K=32 for playoff games, reflecting the higher importance of playoff outcomes.

Step-by-Step: How to Calculate an Elo Rating Change

Let's work through a complete example to demystify Elo calculations.

Scenario 1: Upset Victory

• You: 1600-rated player, K-factor = 20
• Opponent: 1800-rated player, K-factor = 20
• Result: You win

Step 1: Calculate your expected score.

• Expected Score = 1 / (1 + 10^((1800 - 1600) / 400)) = 1 / (1 + 10^0.5) ≈ 0.24

Step 2: Determine actual score.

• Actual Score = 1 (you won)

Step 3: Calculate rating change.

• Rating Change = 20 × (1 - 0.24) = 20 × 0.76 = +15.2 points

Your new rating: 1600 + 15 = 1615

Opponent's new rating: 1800 - 15 = 1785 (they lost to a lower-rated player, so they lose the same points)

Scenario 2: Expected Loss

• You: 2100-rated player, K-factor = 20
• Opponent: 1900-rated player, K-factor = 20
• Result: You lose

Step 1: Calculate your expected score.

• Expected Score = 1 / (1 + 10^((1900 - 2100) / 400)) = 1 / (1 + 10^-0.5) ≈ 0.76

Step 2: Actual score.

• Actual Score = 0 (you lost)

Step 3: Calculate rating change.

• Rating Change = 20 × (0 - 0.76) = 20 × (-0.76) = -15.2 points

Your new rating: 2100 - 15 = 2085

Opponent's new rating: 1900 + 15 = 1915

In this case, you lost to a lower-rated player, so you lose more points than you'd lose to an equal-rated opponent. This penalty incentivizes playing stronger competition.

What Are the Practical Applications of Elo Ratings?

The Elo system's versatility has made it the de facto standard across multiple competitive domains. Each application adapts the core formula to suit its context.

Elo Ratings in Chess: The Original and Most Refined Application

Chess remains the most sophisticated application of Elo ratings. The system has been refined over six decades, with FIDE maintaining official ratings for millions of players worldwide.

FIDE Official Ratings: The International Chess Federation publishes ratings monthly, based on games played in FIDE-rated tournaments. These ratings determine player titles, tournament invitations, and competitive rankings.

Chess.com Glicko Variant: While FIDE uses traditional Elo, Chess.com implements a modified version called Glicko-2, which incorporates rating uncertainty. This accounts for the fact that a player's rating after a long break should be considered less reliable than one based on recent games.

Rating Categories and Titles:

Real Example: Magnus Carlsen, the 16th World Chess Champion, achieved a peak Elo rating of 2882 in 2014, the highest rating ever attained by a human player. This rating reflects not just his victories, but his consistent performance against the world's strongest competition.

Using Elo Ratings for Sports Betting and Predictive Models

Beyond chess, Elo ratings have become invaluable tools for sports bettors and data analysts seeking to predict match outcomes and identify value in betting markets.

FiveThirtyEight's NBA Elo Model: The renowned statistical publication uses Elo ratings to predict NBA game outcomes. Their model incorporates several refinements to the basic Elo formula:

• Margin of Victory Adjustment: The rating change is scaled based on how decisively a team won or lost. A 20-point victory has greater significance than a 2-point victory.
• Playoff Multiplier: Playoff games carry higher K-factors, reflecting their greater importance.
• Home-Court Advantage: The model adjusts for the typical 3-point advantage of playing at home.

Historical data shows FiveThirtyEight's NBA Elo model achieves approximately 57–58% accuracy in predicting game outcomes—significantly better than random guessing (50%) and often more accurate than betting market lines.

Soccer and Football Betting: Analysts adapted Elo for team sports by assigning ratings to teams rather than individuals. Home-field advantage, player injuries, and recent form are incorporated through K-factor adjustments. Soccer Elo models typically achieve 52–55% accuracy, making them profitable for identifying mispriced odds when combined with proper bankroll management.

Tennis Predictions: Elo models for tennis account for surface preferences (clay, grass, hard court), player form, and injury status. Some bettors use Elo-based models to identify discrepancies between their predictions and betting market odds, generating positive expected value (EV) over time.

Elo in Esports, Video Games, and Beyond

The competitive gaming industry has embraced Elo-based ranking systems due to their simplicity and fairness.

League of Legends Ranked System: Riot Games' popular MOBA uses an Elo-inspired system where players climb from Bronze to Challenger tiers. While the exact formula is proprietary, the underlying principle is identical: rating gains and losses are proportional to expected outcome.

Dota 2 Ranking: Valve's Dota 2 employs a similar system, with players earning or losing "Matchmaking Rating" (MMR) based on game outcomes. This enables fair team formation and competitive balance.

AI Model Evaluation: The ChatGPT Arena, developed by LMSYS, uses Elo ratings to rank and compare large language models. When users submit prompts and choose which model's response is better, the system updates Elo ratings for both models. This crowdsourced evaluation method has identified emerging models and measured improvement over time.

Other Applications: Competitive programming platforms, online poker sites, and multiplayer game lobbies increasingly use Elo or Elo-derived systems to ensure fair matchmaking and transparent skill assessment.

What Are the Strengths and Limitations of Elo Ratings?

No rating system is perfect. Understanding Elo's strengths and weaknesses is essential for proper application and interpretation.

Why Elo Ratings Are So Popular: Key Advantages

1. Simplicity: The Elo formula is straightforward enough for anyone to understand and calculate by hand. This accessibility has been crucial to its widespread adoption.

2. Objectivity: Elo ratings are based purely on game results, not subjective judgments. A win is a win, regardless of style or perceived quality of play. This eliminates bias and favoritism.

3. Predictive Power: Elo ratings accurately predict the probability of future outcomes. A 100-point rating gap consistently corresponds to approximately 64% win probability, making the system reliable for forecasting.

4. Scalability: The system works for any two-player or team-based game. It has been successfully adapted to chess, esports, sports betting, and even AI model comparison.

5. Dynamic Adjustment: Ratings update after each game, allowing the system to respond to changes in player form, skill development, or decline.

6. Fair Point Distribution: The zero-sum nature ensures that rating points are conserved. No inflation or deflation of the overall rating pool occurs.

What Are the Main Criticisms and Limitations of Elo?

1. Doesn't Measure Absolute Skill: Elo ratings are relative and probabilistic, not absolute. A 1600-rated player today may be stronger than a 1600-rated player in 1980, because the overall player pool has improved. Elo measures relative standing, not intrinsic ability.

2. Rating Inflation: Over time, player pools improve due to better training, resources, and competition. Ratings tend to climb across the board, making historical comparisons difficult. For example, a 2400 rating in 1980 was rarer and arguably more impressive than a 2400 rating today.

3. Rating Volatility: A player's rating fluctuates around their "true skill level," especially early in their career or after a break. This volatility can be frustrating and doesn't always reflect actual performance changes. A bad tournament streak doesn't mean the player has genuinely declined.

4. Doesn't Account for Draws (in some variants): In chess, draws are common, but traditional Elo treats a draw as 0.5 points for each player. This can undervalue the achievement of holding a much stronger opponent to a draw.

5. Limited Context: Elo doesn't capture external factors like:

• Player injuries or fatigue
• Psychological state or confidence
• Preparation quality for specific opponents
• Home-field advantage (in team sports)
• Equipment or environmental factors

6. K-Factor Arbitrariness: The choice of K-factor is somewhat arbitrary and varies by organization. Different K-factors produce different rating trajectories for the same results.

Common Misconceptions About Elo Ratings

Misconception 1: "A higher Elo rating means you'll always win."

Reality: Elo ratings predict probability, not certainty. A 2000-rated player has a ~64% chance against a 1900-rated opponent, but the 1900-rated player will still win about 36% of the time. Upsets happen.

Misconception 2: "Elo measures absolute skill."

Reality: Elo measures relative strength within a specific player pool at a specific time. A 1800-rated online player is not necessarily equivalent to an 1800-rated FIDE-rated player. Context matters.

Misconception 3: "Rating volatility means the system is broken."

Reality: Rating volatility is expected and normal. It reflects the inherent uncertainty in estimating true skill from a limited sample of games. As more games are played, ratings stabilize.

Misconception 4: "Two players with the same rating are equally skilled."

Reality: Two players with the same rating may have different strengths and weaknesses. One might excel against aggressive opponents while struggling against defensive players. Rating is a single number; it can't capture all dimensions of skill.

Misconception 5: "Elo ratings are perfect predictors."

Reality: While Elo is highly predictive, it's not perfect. Factors like preparation, psychological state, and luck influence individual games. Elo is best used to predict outcomes across many games, not individual contests.

How Do Elo Ratings Compare to Other Rating Systems?

Arpad Elo's system remains the most popular, but several alternatives address specific limitations and offer different trade-offs.

Glicko and Glicko-2: Improvements on the Original Elo

Developed by Mark Glickman in the mid-1990s, the Glicko system enhances Elo by incorporating Rating Deviation (RD), a measure of rating uncertainty.

Key Innovation: Instead of a single rating number, Glicko provides both a rating and a confidence interval around that rating. If a player hasn't played in months, their RD (uncertainty) increases, reflecting the fact that their rating may no longer be accurate.

How It Works:

• Rating Deviation (RD) is high for new players and increases over time without games.
• The K-factor is dynamically adjusted based on RD. Higher RD = higher K-factor (more volatility).
• This allows the system to account for inactive players and rating uncertainty more elegantly than traditional Elo.

Chess.com uses Glicko-2, a refined version that further improves rating accuracy. This is why a player's Chess.com rating may differ from their FIDE rating—different systems, different K-factors, different rating pools.

Advantages over Elo:

• Better handles inactive players
• Accounts for rating uncertainty explicitly
• More accurate predictions in some contexts

Disadvantages:

• More complex to calculate and explain
• Requires more data to initialize

TrueSkill: Microsoft's Modern Alternative

TrueSkill is a Bayesian rating system developed by Microsoft for Xbox Live and esports tournaments. It represents each player's skill as a probability distribution rather than a single number.

How It Works:

• Each player has a mean skill rating and a standard deviation (uncertainty).
• When games are played, both the mean and standard deviation update based on the outcome.
• The system naturally handles team games, draws, and complex matchups.

Advantages:

• Sophisticated uncertainty modeling
• Excellent for team-based games
• Handles draws and ties elegantly
• More mathematically rigorous

Disadvantages:

• Much more complex to calculate
• Difficult to explain to non-technical users
• Requires more computational resources
• Less intuitive than Elo's simple number

Bradley-Terry Model and Other Statistical Approaches

The Bradley-Terry model is a statistical framework that Elo ratings actually simplify. It models the probability of one competitor beating another as a function of their relative strength parameters.

Elo is essentially a practical, computationally efficient implementation of Bradley-Terry for sequential game updates. The Bradley-Terry model is more general and can handle tournaments with many simultaneous games, but it's more complex to compute.

Other alternatives include:

• Thurstone-Mosteller Model: Similar to Bradley-Terry but uses normal distributions instead of logistic distributions.
• Plackett-Luce Model: Extends Bradley-Terry to rank more than two competitors simultaneously.

Why does Elo persist despite these alternatives?

Elo wins the simplicity-vs.-accuracy trade-off. It's accurate enough for most purposes, easy to compute, and intuitive to understand. In competitive gaming and betting, this combination of traits makes Elo the default choice.

Frequently Asked Questions About Elo Ratings

Q: What is a good Elo rating?

A: It depends on the context. In chess:

• Below 1200: Beginner
• 1200–1600: Casual player
• 1600–2000: Serious amateur
• 2000–2400: Expert to professional
• 2400+: Elite / titled player

In sports betting or esports, "good" depends on the player pool and the system used.

Q: How long does it take to reach a certain Elo rating?

A: This varies dramatically based on talent, effort, and competition level. A naturally gifted player might reach 1600 in chess in 1–2 years of serious study. Reaching 2000 typically requires 3–5 years. Reaching 2400 (International Master level) is the work of a lifetime for most.

Q: Can Elo ratings be negative?

A: Technically, yes, but it's extremely rare. The FIDE minimum rating is typically around 1000. Some online platforms allow ratings to drop below 1000, but negative ratings are practically unheard of.

Q: Why does my Elo rating fluctuate so much?

A: Rating volatility is normal, especially for newer players with higher K-factors. Your rating is an estimate of your true skill; as you play more games, the estimate becomes more accurate and stable. Short-term variance is expected.

Q: Is Elo rating the same across all platforms?

A: No. Chess.com uses Glicko-2, FIDE uses traditional Elo, and different esports platforms use custom variations. A 1600 on Chess.com is not directly comparable to a 1600 FIDE rating. Always check which system is being used.

Q: How is Elo used in sports betting?

A: Bettors calculate Elo ratings for teams, then use the rating difference to estimate win probability. If their calculated probability differs from the betting market's implied probability, they identify value bets. For example, if Elo predicts 60% win probability but the market offers 2.5 odds (40% implied probability), that's a value bet.

Q: Can Elo ratings predict individual game outcomes with certainty?

A: No. Elo predicts probability, not certainty. Even a heavily favored player (say, 80% probability) will lose 20% of the time. Elo is best used to predict outcomes across many games, not individual contests.

Q: How do draws affect Elo ratings in chess?

A: In traditional Elo, a draw is scored as 0.5 points for each player. This means the expected score for both players is recalculated as 0.5 each, and rating adjustments are smaller than for decisive games. Some variants adjust this differently.

Q: What's the difference between Elo and Glicko?

A: Glicko adds a "Rating Deviation" (RD) to measure uncertainty in the rating. A player with high RD might have an accurate rating or a misleading one; more games are needed to be confident. Glicko adjusts K-factors based on RD, making it more sophisticated than traditional Elo.

Q: Can I use Elo ratings to guarantee profits in sports betting?

A: No. Elo is a tool for identifying value, not a guaranteed profit machine. Even if Elo is accurate 57% of the time, you need favorable odds to generate positive expected value. Additionally, betting margins, juice, and variance all affect long-term profitability. Elo is best used as one component of a comprehensive betting strategy.

Example

An Elo-based model assigns each team a strength rating that updates after each result, driving market predictions. For instance, if a 1700-rated soccer team beats a 1800-rated team, the model recognizes this upset and adjusts both teams' ratings accordingly. The 1700-rated team gains significant points (exceeding expectations), while the 1800-rated team loses more points than usual (underperforming expectations). Over a season, these rating adjustments create a dynamic ranking that reflects recent performance and strength.

Related Terms

• Power Rating — A subjective or statistical measure of team strength used in sports analysis
• Model Betting — Using mathematical models like Elo to identify betting value
• Statistical Betting — Betting strategy based on statistical analysis and probability models