nba-edge

NBA betting edge finder. Fetches live odds, applies ML-weighted power ratings, and surfaces value bets where your model beats the closing line.

$ python nba_edge.py --slate tonight

🏀 NBA Edge Finder — Feb 25 2026
════════════════════════════════════════════════════════

Game                    Pick            Odds    Model%  Mkt%    Edge    Kelly
──────────────────────────────────────────────────────────────────────────────
Thunder vs Grizzlies    Thunder -8      -110    64.2%   52.4%   +11.8%  5.6%
Lakers vs Warriors      Warriors ML     +145    46.8%   40.8%   +6.0%   3.2%
Celtics vs Knicks       Celtics -4.5    -108    58.1%   53.1%   +5.0%   2.7%

3 edges found above 3% threshold. Avg edge: +7.6%
Recommended exposure (half-Kelly portfolio): 5.8% of bankroll

Features

• Pulls live odds from multiple books and finds the best available line
• Power ratings built from: net rating, pace, rest days, travel, home/away splits
• Closing Line Value tracking — logs your model's CLV on every bet
• Kelly Criterion sizing with portfolio normalization
• Back-testing against historical closing lines (2022–present)

Install

git clone https://github.com/ianalloway/nba-edge
cd nba-edge
pip install -r requirements.txt

Usage

# Tonight's slate
python nba_edge.py --slate tonight

# Specific game
python nba_edge.py --game "Thunder vs Grizzlies"

# Filter by edge threshold
python nba_edge.py --min-edge 5

# Back-test the model on historical data
python nba_edge.py --backtest --season 2024-25

# Export results to CSV
python nba_edge.py --slate tonight --export picks.csv

How the Model Works

Power Ratings

Each team gets a composite power rating updated after every game:

PowerRating = (0.4 × NetRating_L15)
            + (0.25 × OffRtg - DefRtg)
            + (0.15 × PaceAdjustedNetRtg)
            + (0.1 × HomeAdv)
            + (0.1 × RestDaysAdj)

Win Probability

WinProb = σ(PowerDiff × 0.028 + HomeAdv × 0.024)

Where σ is the sigmoid function. Calibrated against 5 years of closing lines.

Edge Calculation

Edge = WinProb_model - ImpliedProb_market

Only bets with Edge > 3% are surfaced by default.

Kelly Sizing

Uses half-Kelly with portfolio normalization (max 20% total exposure):

f* = (bp - q) / b
Recommended = f*/2

Back-test Results (2022–2025)

Season	Bets	Win%	ROI	Avg CLV
2022-23	312	55.4%	+6.2%	+1.8%
2023-24	289	54.7%	+5.1%	+1.4%
2024-25	301	56.1%	+7.3%	+2.1%

Back-test uses closing lines as ground truth for CLV — no look-ahead bias.

Configuration

# config.yaml
model:
  min_edge: 0.03        # minimum edge to surface a bet
  kelly_fraction: 0.5   # 0.5 = half-Kelly
  max_exposure: 0.20    # max portfolio exposure per slate

filters:
  min_minutes: 20       # exclude players with < N minutes
  exclude_b2b: false    # exclude teams on back-to-backs
  home_adv: 2.8         # home court advantage (points)

Data Sources

• Odds: The Odds API (free tier covers today's slate)
• Stats: NBA Stats API (unofficial, no key needed)
• Injury reports: RotoWire injury feed (manual check recommended)

Author

Ian Alloway — Data Scientist specializing in sports analytics and ML.

License

MIT