Advanced Handicapping: Overcoming Standard Box Score Analytics Constraints
For the first three years I bet NBA games, I based my analysis on points scored, rebounds, assists, and turnovers. Basic box-score numbers. And I was profitable — barely. Then I started incorporating advanced metrics into my spread model, and my win rate jumped from 52.8% to 54.6% in a single season. Nearly two full percentage points from nothing more than measuring the same game with better tools.
A betting simulation on NBA spread data across the 2013-17 seasons produced a best result of +8.96% ROI at a 52.60% win rate. The models that achieved those returns were not using raw box-score data. They were built on efficiency metrics, pace adjustments, and net rating — the advanced statistics that strip away the noise of volume and reveal the underlying quality of play. Raw points scored tells you nothing about how efficiently a team produced those points. A team scoring 110 in 105 possessions is performing worse than a team scoring 105 in 95 possessions — but the box score says the opposite.
The gap between box-score analysis and advanced-stat analysis is the gap between a recreational bettor and a systematic one. Every metric in this piece can be accessed for free, requires no programming, and can be plugged directly into a spread or totals model. There is no excuse for ignoring them.
eFG% and True Shooting: Measuring Scoring Efficiency
Field goal percentage was the original efficiency stat, and it is deeply misleading for modern basketball. It treats a made two-pointer and a made three-pointer identically, even though one is worth 50% more. A player who shoots 40% from three is creating more value per attempt than a player who shoots 47% from two, but traditional FG% ranks the second player higher.
Effective field goal percentage (eFG%) corrects this by weighting three-pointers at 1.5 times the value of two-pointers. The formula is (FG + 0.5 * 3PM) / FGA. A player who goes 5-for-12 with three made threes has a raw FG% of 41.7% but an eFG% of 54.2% — a massive difference that reflects the actual value of those made shots. For team-level betting analysis, the eFG% gap between two teams is a more reliable predictor of scoring margin than raw shooting percentage.
True shooting percentage (TS%) goes one step further by incorporating free throws. The formula is Total Points / (2 * (FGA + 0.44 * FTA)). The 0.44 multiplier estimates the number of possessions used by free throw trips (accounting for and-ones, technical foul shots, and three-shot fouls). TS% gives the most complete picture of how efficiently a player or team converts possessions into points.
I use TS% at the team level as a primary input for spread modelling. When Team A’s offensive TS% exceeds Team B’s defensive TS% allowed by more than three percentage points, the spread typically favours Team A by 5-8 points depending on pace. The reverse gap — Team A’s defence allowing a low TS% against an opponent with a weak offensive TS% — is a strong under signal for totals. Efficiency mismatches are the raw material of spread prediction, and TS% captures them more accurately than any traditional stat.
Net Rating: the Single Best Team-Level Predictor
If I could use only one statistic to set an NBA spread, it would be net rating. Not win-loss record, not point differential, not any individual player stat. Net rating — the difference between a team’s offensive rating (points scored per 100 possessions) and defensive rating (points allowed per 100 possessions) — is the single best predictor of future performance at the team level.
The average home margin in the 2024-25 NBA season was just 1.62 points. Net rating cuts through that thin margin to identify which teams are genuinely good versus which are lucky. A team that wins close games at an unsustainable rate will show a record better than its net rating predicts. Historically, record regresses toward net rating rather than the reverse. This creates one of the most reliable betting edges available: fading teams whose records outperform their net rating and backing teams whose records underperform it.
I calculate net rating on rolling 15-game windows rather than full-season aggregates. The full-season number includes data from October that may no longer be relevant in February — injuries, trades, rotation changes, and player development all shift the true net rating over time. A 15-game window is long enough to smooth out single-game variance and short enough to reflect recent reality. When my rolling net rating diverges from the bookmaker’s power rating by more than 2.0 points, I flag the game for further analysis.
Net rating also decomposes into its two components. Offensive rating and defensive rating move independently. A team might improve offensively by five points per 100 possessions after a trade while its defensive rating stays flat. That specific shift affects spreads differently from totals — it pushes the spread in the team’s favour and pushes the total higher. Modelling the two halves separately rather than relying on the combined net rating produces a more nuanced and more accurate projection. The betting model guide walks through how to build net rating into a full predictive framework.
Pace and Usage Rate: Volume Context for Every Stat
Pace is the number of possessions a team uses per 48 minutes. It is not a quality metric — it does not tell you whether a team is good or bad. What it tells you is the context in which every other stat should be interpreted. A team that scores 115 points per game at a pace of 105 is less efficient offensively than a team scoring 108 at a pace of 96. Without pace context, the first team looks better. With it, the second team is clearly superior per possession.
For totals, pace is the most important variable, full stop. Game pace — the average of the two teams’ paces, adjusted for strength of schedule — determines the expected number of possessions. Multiply possessions by expected points per possession (derived from the teams’ offensive and defensive ratings), and you have a total projection. If that projection diverges from the posted line by more than three points, you have a bet. The formula is simple, the data is free, and the method outperforms gut-feel totals betting by a wide margin.
Usage rate applies the same volume-context principle at the player level. Usage rate measures the percentage of team possessions a player uses while on the floor. A player with a 30% usage rate is involved in nearly a third of his team’s possessions — taking shots, drawing fouls, or turning the ball over. When that player sits, his usage is redistributed to teammates. Understanding usage redistribution is critical for player props: when a 30% usage player is ruled out, his teammates’ usage rates spike by predictable amounts, and their prop lines may not fully adjust.
I maintain a usage redistribution matrix for every NBA team. When a high-usage player is listed as out, I calculate the expected usage increase for each remaining starter and plug those numbers into my prop model. The resulting projections frequently diverge from the bookmaker’s adjusted lines by enough to generate bettable edges. This is the kind of analysis that takes 20 minutes per game and produces consistent value — the unsexy grind that separates systematic bettors from the crowd.
