Common Mistakes, Advanced Topics & Practice Exercises
This final lesson covers the pitfalls that trip up even experienced bettors, introduces advanced modeling concepts, and provides comprehensive exercises to cement your understanding.
Five Common Mistakes to Avoid
Mistake 1: Treating Correlated Legs as Independent
The Error: You see Williams Over 1.5 TDs at 47.8% and Loveland Over 0.5 TDs at 34.5%. You multiply them to get 16.5% and think the SGP should pay +506.
The Problem: These legs are strongly correlated (r ≈ +0.47). The true joint probability is much higher—around 29%. The independence calculation is way off.
Warning
Same-game props are almost never independent. Always check for correlation before using simple multiplication.
Solution: Always check for correlation using conditional probability or the r-adjustment formula.
Mistake 2: Using Correlation on Raw Stats Instead of Hit/Miss Indicators
The Error: You calculate r = +0.47 for Williams passing TDs (0, 1, 2, 3, 4...) and Loveland receiving TDs (0, 1, 2...) and think that's the correlation you need for SGP pricing.
The Problem: The SGP legs are binary (hit/miss), not the raw TD counts. The correlation between the binary indicators can be different.
Solution: Use conditional probability directly from game history. Count how often both legs hit together, not how the raw stats correlate.
Mistake 3: Ignoring Sample Size
The Error: You condition on 4 high-usage games and get P(B|A,T) = 75%. You bet big because "the math says so."
The Problem: 3 out of 4 games is a tiny sample. One extra TD or miss would change 75% to 50% or 100%.
Warning
Be cautious with estimates based on fewer than 10 observations. Widen your confidence intervals or use a method that incorporates more data.
Solution:
- Minimum 15 games for reliable conditional probability
- For smaller samples, use the r-adjustment method with conservative assumptions
- Never bet big on small-sample estimates
Mistake 4: Forgetting to De-Vig Individual Legs
The Error: You use the market-implied probabilities (with vig) to calculate independence, then compare to the SGP price.
The Problem: You're comparing vigged inputs to a vigged output. Your edge calculation is distorted.
Solution: Always de-vig the individual legs before building joint probabilities.
De-Vig Reminder
Fair P = q_leg / (q_over + q_under)=A1/(A1+B1)Mistake 5: Chasing the Story
The Error: You see a 3-leg SGP that tells a great story: "Big passing game, favorite receiver scores, team wins." The payout looks juicy at +800. You bet it because it "makes sense."
The Problem: Compelling stories are exactly what sportsbooks want you to bet. They price these aggressively because they know recreational bettors love them.
Key Insight
If the story is obvious, the book has already priced it in—and probably overpriced it.
Solution: Run the math. Calculate the implied r. Compare to your own estimate. Never bet because a narrative "feels right."
Advanced Topic: Gaussian Copulas
With 2-leg SGPs, you can often get by with conditional hit rates or simple correlation adjustments. But with 3+ legs and hundreds of combinations, serious quants need something more systematic.
What Is a Copula?
A copula separates the modeling problem into two layers:
- Marginals: How each leg behaves on its own (its single-leg distribution)
- Dependence: How the legs move together across games
For example, you might model:
- Williams passing TDs with a count distribution (Poisson or Negative Binomial)
- Loveland targets with another count distribution
- Loveland TDs as a function of targets
The copula ties these marginal pieces together so that simulated games respect the observed dependence structure.
Why Gaussian Copulas Are Popular
| Advantage | Description |
|---|---|
| Interpretable | Dependence parameters closely related to correlation |
| Simulation-friendly | Easy to draw correlated "games" in software |
| Well-studied | Widely implemented in statistics libraries (Python, R) |
The Catch: Not an Excel Formula
A Gaussian copula is not a single formula you can drop into a cell. You need:
- A paired dataset of the stats or legs you care about
- Software (Python, R) to fit marginal distributions and dependence parameters
- Simulation code to generate thousands of correlated game outcomes
Tip
Copulas are overkill for a single 2-leg SGP but invaluable if you're building a full SGP pricing engine.
When to Use Each Method
| Situation | Best Method |
|---|---|
| 2-leg SGP with 15+ paired games | Conditional probability |
| 2-leg SGP with limited history | Correlation adjustment formula |
| 3-leg SGP with 15+ paired games | Chain rule |
| Many SGPs across many players | Gaussian copula + simulation |
Chapter Summary
Covariance and correlation give you the language to describe how props move together:
- Positive correlation: Legs share the same story (more scoring, more usage, more volume)
- Negative correlation: Trade-offs exist (limited possessions, competing roles)
For pricing SGPs, correlation is a useful diagnostic, but the number you actually need is the joint probability—best estimated with conditional probabilities when you have paired history, and with correlation-informed models when you don't.
The sportsbook's SGP prices are not magic. They're built on the same math you just learned. Your edge comes from:
- Better data
- Better context adjustments
- Faster execution
Comprehensive Practice Exercises
Exercise 11.1: Correlation in Excel (Warm-Up)
📝 Exercise
Instructions
You have 10 games of data for two props:
Prop A (QB passing TDs): [2, 1, 3, 2, 0, 2, 3, 1, 2, 4]
Prop B (WR receiving TDs): [1, 0, 1, 1, 0, 0, 2, 0, 1, 1]
- Enter the data into Excel
- Compute r with
=CORREL(A2:A11, B2:B11) - Classify the relationship as weak/moderate/strong
Exercise 11.2: Implied r From an SGP Price
📝 Exercise
Instructions
You have two legs with fair probabilities:
- pA = 0.55
- pB = 0.48
- SGP market price: +250
Tasks:
- Convert +250 to p_mkt
- Compute the independence baseline (pA × pB)
- Compute the book implied r using:
r = (p_mkt - pA×pB) / √(pA(1-pA) × pB(1-pB)) - Interpret the sign and magnitude of r
Exercise 11.3: Price It Yourself With Your Own r
📝 Exercise
Instructions
Using the same pA = 0.55 and pB = 0.48, compute fair SGP odds under three scenarios:
- r = 0.00 (independence)
- r = +0.20 (weak positive)
- r = +0.40 (moderate positive)
Formula: pAB(r) = pA × pB + r × √(pA(1-pA) × pB(1-pB))
Convert each pAB to American odds and compare to the market +250.
Exercise 11.4: Pricing a 3-Leg SGP With the Chain Rule
📝 Exercise
Instructions
You have 15 games of paired data:
- Leg A hits in 9 of 15 games
- A ∩ B hits in 6 of 15 games
- A ∩ B ∩ C hits in 4 of 15 games
Tasks:
- Compute P(A)
- Compute P(B|A)
- Compute P(C|A ∩ B)
- Compute P(A ∩ B ∩ C) and convert to fair American odds
Exercise 11.5: Evaluating a Same-Player SGP
📝 Exercise
Instructions
A same-player NBA SGP is offered:
- Leg A: Player Over 25.5 points (-120)
- Leg B: Player Over 5.5 rebounds (-110)
- SGP: +140
Assume opposite sides are +100 for both legs (for de-vigging).
Tasks:
- De-vig the singles to get fair pA and pB
- Compute the independence baseline
- Back out the book implied r
- If you believe points and rebounds for this player are correlated at r ≈ 0.35, is the SGP +EV or -EV?
Quick Reference: Formula Summary
Converting Odds
| Conversion | Formula |
|---|---|
| Negative odds → Probability | abs(odds) / (abs(odds) + 100) |
| Positive odds → Probability | 100 / (odds + 100) |
| Probability → Positive odds | 100 × (1-P) / P |
| Probability → Negative odds | -100 × P / (1-P) |
Correlation Formulas
Correlation Coefficient
r = Cov(X,Y) / (σₓ × σᵧ)=CORREL(range1, range2)r-Adjustment Joint Probability
P(A ∩ B) ≈ pA × pB + r × √(pA(1-pA) × pB(1-pB))=A1*B1 + C1*SQRT(A1*(1-A1)*B1*(1-B1))Implied r from Market
r = (p_mkt - pA×pB) / √(pA(1-pA) × pB(1-pB))=(D1-A1*B1)/SQRT(A1*(1-A1)*B1*(1-B1))Chain Rule (3 Legs)
P(A ∩ B ∩ C) = P(A) × P(B|A) × P(C|A ∩ B)=(A/total) × (AB/A) × (ABC/AB)Final Takeaways
Key Insight
The SGP Pricing Checklist
- De-vig individual legs before any calculations
- Check for correlation — same-game props are rarely independent
- Use conditional probability when you have 15+ paired games
- Use r-adjustment when data is limited
- Back out the book's implied r to see what you're betting against
- For 3+ legs, use the chain rule — errors compound quickly
- Always consider context — weather, injuries, matchups matter
- If the story is obvious, the book has priced it in
You now have a complete toolkit for pricing SGPs. The sportsbook's prices aren't magic—they're built on the same math you just learned. Your edge comes from better data, better context adjustments, and faster execution.
Go find those mispriced parlays! 🎯