The single number that explains everything
Edge — expected return per unit risked — is the foundational concept that ties together every other element of professional sports betting. Kelly sizing depends on edge. Risk-of-ruin calculations depend on edge. The decision whether to place any given bet depends on edge. Account longevity depends on edge (books detect persistent edge and limit accordingly). Bankroll growth depends on edge × volume.
The math is simple. The hard part is having real edge that survives book counter-actions, variance, model degradation, and the operational friction of executing thousands of bets per year.
The formula
# Edge as a fraction of stake edge_% = (your_estimated_probability × decimal_odds) − 1 # Worked examples bet at -110 (decimal 1.909), your_p = 0.55: edge = (0.55 × 1.909) − 1 = +5.0% bet at +150 (decimal 2.50), your_p = 0.45: edge = (0.45 × 2.50) − 1 = +12.5% bet at -200 (decimal 1.50), your_p = 0.65: edge = (0.65 × 1.50) − 1 = -2.5% # DON'T BET
Edge can be positive (+EV bet, worth placing if sized correctly) or negative (-EV bet, should not be placed). The number directly equals expected long-run percentage return per bet. A consistent +3% edge bettor expects $30 return per $1,000 bet on average across hundreds of bets — boring numbers that compound through volume.
Sources of edge — where it actually comes from
| Edge source | Typical magnitude | Sustainability | Required skills |
|---|---|---|---|
| Quantitative model on major sport | 0.5-3% | Years; degrades as books improve | Statistical modeling, data infrastructure |
| Player prop modeling | 2-6% | Months-to-years; book models weakest here | Sport-specific knowledge, distribution modeling |
| Correlation arbitrage (SGP) | 1-5% | 1-3 years; books improving | Copula modeling, conditional probability |
| Steam-chasing | 1-3% | Years; window compressing | Fast tooling, multi-book ops |
| Arbitrage between books | 0.5-2.5% | Indefinite; capital-bounded | Scanning tools, multi-account ops |
| Bonus exploitation | 5-15% | One-shot per account; cycles | Promotional math, account rotation |
| Live (in-play) modeling | 3-8% | Years; book models weakest | Real-time modeling, fast execution |
| Insider information | variable | Illegal in most jurisdictions; high risk | Legal/ethical hazards |
| Niche market specialization | 3-8% | 1-5 years per niche | Deep domain expertise |
The edge degradation problem
No edge lives forever. Books invest in risk management to compress edges that bettors exploit. Compression timeline:
# NFL spread modeling edge — historical compression 2010: sharp model edge ~ 4-6% # pre-public sharp tooling 2015: sharp model edge ~ 2-4% # public sharp data emerged 2018: sharp model edge ~ 1.5-3% # PASPA repeal, more competition 2022: sharp model edge ~ 1-2.5% # US book ML investment 2026: sharp model edge ~ 0.5-2% # mature US market
Survivors continuously refresh edge sources. Bettors who built single-model strategies in 2018 and didn't evolve are running at break-even or worse in 2026. The pros who continue to profit are those who: ① diversify edge sources; ② invest in new data and modeling approaches; ③ pivot to less competitive markets (specific player props, niche sports, live betting) as major markets compress.
Edge × Volume = Income
The math of compounded edge through volume:
# Pro bettor with documented edge edge = +2.5% avg stake = $500 bets / year = 1,500 expected_annual_profit = 1,500 × $500 × 0.025 = $18,750 # Same edge, larger scale avg stake = $2,000; bets/year = 2,500 expected_annual_profit = $125,000 # Larger edge, same scale edge = +4%; avg stake = $2,000; bets/year = 2,500 expected_annual_profit = $200,000
The arithmetic is straightforward. The hard parts are: ① having real +EV at the documented magnitude; ② being allowed to bet at the assumed stake level (books limit sharps); ③ executing the assumed volume (sourcing 2,500 +EV bets per year requires sophisticated tooling and book coverage). Most pros operate at smaller scale than the math suggests is possible, because operational reality compresses theoretical numbers.
Edge measurement — the documentation problem
You can't manage what you don't measure. Pros track:
- Edge by bet — at placement, log estimated edge based on model.
- CLV by bet — at game start, log closing line value (post-vig-strip).
- Result by bet — at game end, log win/loss/push/void.
- Edge by market segment — aggregate edges by sport, bet type, book, day of week.
- Edge over time — rolling 50, 100, 500-bet windows to spot degradation.
- Edge vs. realized — does the bettor's edge estimate match actual returns? Systematic over/underestimates indicate model bias.
Most recreational bettors don't track. Many tracking systems exist (Pikkit, BetSquare, custom spreadsheets) but discipline matters more than tools. The pro who knows their edge by market segment can rationally allocate capital; the recreational bettor without data optimizes by gut feel and loses to themselves over time.
The "edge inflation" trap
Bettors systematically over-estimate their edges. Reasons:
- Confirmation bias — remembering hits, forgetting misses.
- Selective sample — only counting bets you remember as +EV.
- Model overfit — strong backtest results that don't generalize to live betting.
- Tilt accommodations — recasting bets you placed for entertainment as 'edge bets.'
- Optimistic probability estimates — your model says 55%, but the calibration check shows you predict 55% events that actually happen 51% of the time.
Calibration checks are essential. Periodically (every 200 bets), bucket your bets by your stated edge (0-1%, 1-2%, 2-3%, etc.) and check realized ROI in each bucket. If you bet many "+3% edges" but your realized ROI is +0.5%, you're systematically over-estimating by ~2 percentage points. Adjust future estimates accordingly. Most bettors discover their actual edge is 30-50% smaller than their estimated edge after honest calibration.
Edge vs. yield vs. ROI — terminology
| Metric | Calculation | What it measures |
|---|---|---|
| Edge per bet (EV%) | (p × decimal) − 1 | Expected return per unit risked, per bet |
| Yield | total_profit / total_stake | Realized cumulative edge across all bets |
| ROI (return on invested capital) | net_profit / bankroll_deployed | Return on capital, accounts for re-betting |
| Sharpe ratio | (return − risk_free) / std_dev | Risk-adjusted return; variance-normalized |
Pro publications usually quote yield (typical sharp: 3-5% yield) and ROI on bankroll (typical sharp: 10-25% annual ROI on bankroll, vs. 3-5% yield on turnover). Confusing yield with ROI is a common mistake — they're related but different. Yield is per-bet; ROI is per-bankroll-cycle.
Sources & further reading
- Kelly, John L. Jr. "A New Interpretation of Information Rate." Bell System Technical Journal, 1956 — original derivation of optimal sizing for edge bettors.
- Thorp, Edward O. Beat the Dealer. Vintage, 1962 — foundational text on edge identification and management.
- Hausch, Donald B., Ziemba, William T. (eds.). Handbook of Sports and Lottery Markets. Elsevier, 2008.
- Buchdahl, Joseph. Squares and Sharps, Suckers and Sharks. High Stakes Publishing, 2016.
- Pinnacle Betting Resources — "What is value betting? A complete guide to finding edge" (open documentation).