League of Legends Betting Model: Glicko-2 Monte Carlo

Engine Overview

Esports betting is the fastest-growing segment of the sports betting industry, and League of Legends is its flagship title. The LoL competitive scene features hundreds of professional matches per week across dozens of leagues worldwide, creating a rich dataset for quantitative modeling. However, esports presents unique challenges that traditional sports models cannot handle: frequent game patches that alter champion balance, roster swaps that change team composition mid-season, and regional meta differences that make cross-league comparisons unreliable.

The Championship v2.1 engine is purpose-built for these challenges. It uses a multi-layer Glicko-2 rating system that separates team skill from meta adaptation, applies regression when teams have been inactive, and blends model probabilities with market odds to produce calibrated predictions. The result is a model that captures the rapid pace of change in competitive LoL while maintaining the statistical rigor of a quantitative sports model.

How the LoL Simulation Works

1. 5-Layer Glicko-2 Rating System

Glicko-2 is an advanced rating system that extends the classic Elo approach by tracking not just a team's strength (rating), but also the uncertainty in that rating (rating deviation, or RD) and the consistency of performance (volatility). The Championship engine applies Glicko-2 across five separate layers:

• Overall team rating — The primary Glicko-2 score reflecting general team strength
• Regional rating — Separate rating within each league (LCK, LPL, LEC, LCS) that captures region-specific performance context
• Best-of-format rating — Separate ratings for Bo1, Bo3, and Bo5 formats, since team performance varies significantly by series length
• Side rating — Blue side vs. red side performance, reflecting draft advantage and strategic preferences
• Recency-weighted rating — A rolling 30-day window that captures current form vs. season-long average

These five layers are combined into a composite rating using empirically-optimized weights. The composite captures both a team's fundamental skill and their adaptation to specific competitive contexts. A team might have a strong overall rating but a weak Bo5 rating, indicating they excel in regular-season Bo1/Bo3 matches but struggle in playoff series.

2. Market Blend

Esports betting markets are less efficient than traditional sports markets, which means there are more edges available — but the model also has more room for error. The engine blends model probabilities with market-implied probabilities using an empirically-derived alpha weight. This blend serves as a form of regularization: when the model strongly disagrees with the market, the blend pulls the probability toward the market, reducing the risk of acting on a flawed model signal.

The blend weight is calibrated on historical data. In leagues with more liquid markets (LCK, LPL), the market receives higher weight because these markets are more efficient. In leagues with thinner markets (smaller regional leagues, play-ins), the model receives higher weight because the market is less informed.

3. Staleness Regression

Esports teams frequently have gaps between competitive matches: international breaks, mid-season pauses, roster announcements that disrupt practice. When a team has not played a competitive match in 14+ days, the engine applies staleness regression that increases the Glicko-2 rating deviation (uncertainty). This means the model becomes less confident in the team's current strength, which reduces bet sizes and prevents overcommitting to stale data.

The regression is calibrated by observed performance degradation. Historical data shows that teams returning from breaks of 2+ weeks perform approximately 3-5% worse than their pre-break rating suggests, likely due to meta drift and practice disruption. The staleness regression captures this empirical finding.

4. Patch-Aware Meta Analysis

Riot Games updates League of Legends approximately every two weeks with balance patches that change champion strengths, item statistics, and map objectives. These patches can dramatically shift the competitive meta. A team that dominates on a tank-focused meta may struggle when a patch shifts the game toward early-game aggression.

The engine tracks the current patch and applies meta-awareness adjustments. When a new patch drops, the model reduces confidence in all predictions for the first few days (until enough competitive matches have been played on the new patch to recalibrate). Teams known for champion pool depth and fast adaptation receive smaller confidence penalties than teams with narrow playstyles.

5. League Coverage

Data Sources

• Official match data — Game results, game duration, kill/death/assist statistics from Riot and third-party APIs
• Patch notes — Champion and item changes tracked for meta-shift detection
• Odds providers — Real-time moneyline and map betting odds for market blend calibration
• Roster tracking — Player swaps, loan agreements, and emergency substitutions from official announcements

LoL Performance

Why Glicko-2 Beats Elo for Esports

Classic Elo ratings assume that a team's uncertainty is constant. A team that has played 100 games and a team that has played 5 games both receive the same weight when updating ratings after a match. This is a serious flaw in esports where roster changes, meta shifts, and competitive breaks create genuine uncertainty about a team's current strength.

Glicko-2 solves this by tracking uncertainty explicitly. A team that has not played in three weeks has a higher RD, meaning their rating is less certain. When they play again, the outcome updates their rating more aggressively (because there is more to learn). A team that plays every day has a low RD, and individual match outcomes cause smaller rating changes (because the system is already confident in their level).

The volatility parameter adds another dimension: it captures whether a team's results are consistent or erratic. A team that alternates between dominant wins and surprising losses will have high volatility, leading to wider confidence intervals in predictions. This is particularly useful for identifying teams that are risky bets despite strong average ratings.

Explore Other Models