Tennis Betting & Analytics Glossary

The Brier Score measures the accuracy of probabilistic forecasts. It is calculated as the mean squared difference between predicted probabilities and actual outcomes. Crucially, it punishes overconfidence: a model that says 95% when the true probability is 60% scores far worse than one that expresses calibrated uncertainty. Tennis Glicko's current Brier Score is 0.178, compared to ~0.22 for simple Elo models.

Closing Line Value measures whether the odds you obtained at bet placement were better than Pinnacle's closing price — the line just before the match starts, after all sharp money has been absorbed. Because the closing line represents the market's most efficient estimate of true probability, consistently beating it is strong evidence that your model has genuine edge, not just variance. A model that wins money but consistently gets worse odds than the close is likely profiting from luck. CLV is the metric professional bettors use to distinguish skill from a lucky run.

Developed by Arpad Elo for chess, the Elo system assigns each player a numerical rating that updates after every match based on the expected vs actual result and the opponent's rating. A win over a highly-rated opponent gains more points than a win over a weaker one. While significantly better than ATP rankings for betting, Elo treats skill as a single static point with no uncertainty measure. Tennis Glicko uses Elo ratings as a feature input to its XGBoost model, alongside Glicko-2. Elo provides a surface-adjusted prior; the final probability is the model output, not a direct average of Elo and Glicko-2.

Expected Value is the probability-weighted average of all possible outcomes. In betting: EV = (Win Probability × Net Win) − (Loss Probability × Stake). If your model says a player has a 65% chance of winning and the market offers odds of 1.70 (implying 58.8%), the bet has positive expected value (+EV). Consistently betting +EV opportunities is the only mathematically defensible path to profitability. Negative EV bets lose money in the long run, regardless of short-term variance.

Developed by Mark Glickman as an extension of Elo, Glicko-2 defines each player through three parameters: Rating (r) — the central skill estimate; Rating Deviation (RD) — confidence in that estimate (high RD = uncertain, low RD = reliable); and Volatility (σ) — consistency across results. This three-dimensional profile enables the model to recognize 'rust' in players returning from injury (their RD expands during inactivity), identify genuine upset candidates, and produce better-calibrated probabilities. Tennis Glicko maintains independent Glicko-2 models for hard, clay, and grass courts.

Green EV is Tennis Glicko's high-conviction value alert. It triggers when two conditions are met simultaneously: VOPO exceeds 12% (the market is significantly undervaluing the player) and the internal win probability is above 50% (the model considers this player the actual favorite). Both conditions must be true — a 14% VOPO on a 45% underdog does not qualify. Green EV matches are surfaced prominently in the matches table and trigger push notifications for PRO subscribers.

Every betting odd implies a probability. At 1.50, the implied probability is 1/1.50 = 66.7%. However, bookmakers build in a margin (vig or overround), so the sum of implied probabilities across all outcomes exceeds 100%. To get the 'true' implied probability from Pinnacle (the sharpest market), we remove the vig by normalizing the raw implied probabilities. Tennis Glicko uses Pinnacle's no-vig implied probability as the market benchmark for VOPO calculation.

The Kelly Criterion is a mathematical formula for determining the optimal fraction of your bankroll to stake on a positive-EV bet: f = (bp − q) / b, where b is the decimal odds minus 1 (net profit per unit), p is your estimated win probability, and q = 1 − p. Full Kelly maximises long-run growth rate but produces aggressive stake sizes and high variance. Most professional value bettors use Fractional Kelly (0.25× or 0.5×) to reduce variance while preserving most of the growth advantage. Kelly requires accurate probability estimates — if your model overstates edge, Kelly will over-stake and accelerate ruin.

Pinnacle is widely regarded as the sharpest sportsbook in professional sports betting. Unlike recreational-facing bookmakers, Pinnacle does not limit or ban winning players — it welcomes sharp action, which means its lines are continuously corrected by professional bettors. The result is that Pinnacle's odds are considered the closest available approximation to the true market probability. Tennis Glicko uses Pinnacle's current market odds as the reference line for all VOPO calculations.

Prediction accuracy is the classification rate — the proportion of matches where the model's favorite (the player with >50% win probability) wins. In professional tennis, random guessing gives ~50%. Simple ATP-ranking-based models achieve ~65–68%. Tennis Glicko's XGBoost model — trained on ~440k ATP/WTA matches with Glicko-2 and Elo as feature inputs — achieves 72.8%. A 1–2% improvement in accuracy in a sharp market is the difference between sustainable profitability and long-term loss, because it shifts expected value across thousands of bets.

Rating Deviation is the standard deviation of the Gaussian distribution around a player's Glicko-2 rating. A player with RD = 30 is well-characterized — their true skill is within a narrow band of their stated rating. A player with RD = 200 could be significantly better or worse than their rating suggests. RD increases automatically when a player is inactive (simulating 'rust'), and decreases as more matches provide data. It is the key feature that makes Glicko-2 superior to Elo for detecting true upset potential.

Sharp money refers to wagers placed by professional bettors (sharps) whose long-term track record is profitable. Because sharps bet based on models and statistical edge rather than opinion, their action is informative about where the true probability lies. At sharp-accepting books like Pinnacle, sharp action moves the line quickly. Line movement opposite to public betting volume is often attributed to sharp money and can signal genuine pricing errors in the initial line.

The vig (also called juice, vigorish, or overround) is how bookmakers guarantee profit regardless of outcome. If a true 50/50 match paid 2.00 on both sides, the bookmaker would break even. Instead, they price it at 1.91 / 1.91 — implying 52.36% + 52.36% = 104.72%. The 4.72% excess is the vig. Pinnacle operates with the lowest vig in the industry (~2–3%), which is why it is used as the benchmark. To compare model probabilities to market odds fairly, the vig must first be removed.

Volatility is the third parameter in the Glicko-2 system (alongside rating and RD). It measures how erratically a player's performance fluctuates beyond what their rating would predict. A player with high volatility wins and loses against players they shouldn't — they are the archetype of the dangerous lower-ranked opponent. A player with low volatility performs consistently within the bounds of their rating. Volatility adjusts how much weight is given to recent results when updating a player's rating.

VOPO is the central metric of Tennis Glicko. It is calculated as: VOPO = XGBoost win probability − Pinnacle no-vig implied probability. The XGBoost model uses Glicko-2 and Elo ratings as feature inputs to produce the internal probability. A positive VOPO means the market is undervaluing the player relative to our model. A negative VOPO means the market rates them higher than our model does. The larger the positive VOPO, the more the market is mispricing the match. VOPO is calculated for every player in every upcoming tennis match, updated live as odds move.