Chapter 3: Key Takeaways

Summary Card

Core Idea

Probability theory is the mathematical language of prediction markets. Every market price is a probability, every trade is a disagreement about probability, and every profit or loss is the resolution of probabilistic uncertainty. Mastering the fundamentals --- from sample spaces to Bayesian updating --- gives you the tools to analyze any prediction market contract and make mathematically grounded decisions.


The Five Essential Concepts

Concept One-Sentence Summary Key Formula
Conditional Probability The probability of an event changes when you learn new information P(A|B) = P(A and B) / P(B)
Bayes' Theorem The exact recipe for updating beliefs with evidence P(H|E) = P(E|H) P(H) / P(E)
Expected Value The probability-weighted average outcome; your "edge" in a trade EV = q - p
Variance How spread out outcomes are; your "risk" in a trade Var = p(1-p) for binary
Law of Large Numbers Average outcomes converge to EV over many trials Sample mean converges to population mean

Key Formulas at a Glance

Complement:       P(not A) = 1 - P(A)
Addition:         P(A or B) = P(A) + P(B) - P(A and B)
Conditional:      P(A|B) = P(A and B) / P(B)
Multiplication:   P(A and B) = P(A|B) * P(B)
Bayes:            P(H|E) = P(E|H) * P(H) / P(E)
Bayes (odds):     posterior odds = likelihood ratio * prior odds
Independence:     P(A and B) = P(A) * P(B)  [iff independent]
Binary EV:        EV = q - p  (your prob minus market price)
Bernoulli Var:    Var = p * (1 - p)
Beta update:      Beta(a,b) + s successes, f failures -> Beta(a+s, b+f)
CLT std error:    SE = sigma / sqrt(n)

Key Distributions for Prediction Markets

Distribution Use Case Parameters
Bernoulli Single binary contract outcome p (probability)
Binomial Number of wins in n independent trades n (trials), p (win prob)
Beta Uncertainty about a probability itself alpha, beta (shape)
Normal Profit distribution over many trades (via CLT) mu (mean), sigma (std)

Decision Checklist for Evaluating a Trade

  1. What is the sample space? What are all possible outcomes?
  2. What is my estimated probability q for this event?
  3. What is the market price p?
  4. Is there an edge? (q - p > 0 for a buy)
  5. How strong is my evidence? (Compute the likelihood ratio)
  6. What is the risk? (Variance, max loss, drawdown potential)
  7. How does this trade correlate with my other positions?
  8. Do I have enough bankroll to survive the variance?
  9. Am I making enough trades for the LLN to work in my favor?

Common Mistakes to Avoid

Mistake Why It Is Wrong The Fix
Confusing P(A|B) with P(B|A) Base rate fallacy / prosecutor's fallacy Always apply full Bayes' theorem
Assuming independence Correlated events amplify portfolio risk Test for independence; model correlations
Ignoring variance Positive EV does not mean guaranteed profit Calculate risk alongside EV
Gambler's fallacy Past losses do not make future wins more likely Each independent trade is a fresh event
Assigning P = 0 or P = 1 No evidence can update a certainty (Cromwell's rule) Keep probabilities in [0.001, 0.999]
Overconfidence at extremes Small calibration errors at 95%+ are very costly Extra scrutiny for extreme probabilities

Code Patterns to Remember

Quick Bayes update (odds form):

def quick_bayes(prior_prob, likelihood_ratio):
    prior_odds = prior_prob / (1 - prior_prob)
    posterior_odds = likelihood_ratio * prior_odds
    return posterior_odds / (1 + posterior_odds)

Binary trade EV:

def trade_ev(market_price, your_prob, position_size=1.0):
    return (your_prob - market_price) * position_size

Beta-Binomial update:

def beta_update(alpha, beta, successes, failures):
    return alpha + successes, beta + failures

Bridge to Next Chapter

Chapter 3 gave you the mathematical theory of probability. Chapter 4 will give you the statistical practice --- how to estimate probabilities from data, test whether your edge is real, build confidence intervals, and assess whether prediction market prices are well-calibrated. The tools you built here (Bayesian updater, EV calculator, Monte Carlo simulator) will be extended and applied to real market data.