Probability Concepts — CFA Level I (Quantitative Methods)

"If numbers are a language, probability is the grammar that stops your sentences from turning into lies." — That’s me, dramatizing your morning coffee.

---

Opening: Why probability matters (and why your portfolio begs you to care)

You just came from Basic Statistics and Time Value of Money — so you know how to summarize historical returns and discount future cash flows. Probability is the bridge that connects those facts to decisions under uncertainty: estimating default likelihoods, pricing contingent claims, stress-testing portfolios, and thinking ethically about the confidence you communicate to clients.

Remember Ethics and Professional Standards? Misstating probability — or cherry-picking priors — is not just sloppy math, it’s a professional sin. So let’s get the concepts right, clearly, and with enough flair that we’ll never mislead anyone again.

---

1) The building blocks: experiment, outcome, sample space, event

• Experiment: any process that generates an outcome (e.g., next-quarter return on a stock).
• Outcome: a single possible result (e.g., +5%).
• Sample space (S): the set of all possible outcomes (e.g., up, down, unchanged—or all real numbers for continuous returns).
• Event: any subset of S (e.g., ‘stock loss > 10%’).

Why care? Because every probability question you face — credit default, option exercise, recession — is just an event hiding in plain sight.

---

2) Core rules (axioms) — the law of the land

1. For any event A: 0 ≤ P(A) ≤ 1.
2. P(S) = 1.
3. If A and B are mutually exclusive (disjoint), P(A ∪ B) = P(A) + P(B).

Quick corollary: P(A') = 1 − P(A) (where A' is the complement). Useful when "at least one" problems look annoying — use the complement.

Example: If each bond in a small basket has a 2% independent default probability, the probability none default = (0.98)^n. So probability at least one defaults = 1 − (0.98)^n.

---

3) Addition and multiplication rules

• Addition (general): P(A ∪ B) = P(A) + P(B) − P(A ∩ B).
• Multiplication (conditional): P(A ∩ B) = P(A) × P(B|A) = P(B) × P(A|B).

These let you build from simple probabilities to joint and conditional ones — essential for multi-factor credit models, scenario analysis, and any chain of dependent events.

---

4) Conditional probability & Bayes' theorem — the courtroom drama of statistics

Conditional probability: P(A|B) = P(A ∩ B) / P(B).

Bayes' theorem (useful for updating beliefs when new info arrives):

P(A|B) = [P(B|A) * P(A)] / P(B)

Finance example (credit rating): Suppose 5% of firms are originally rated BB. If a firm is downgraded this year, what is the updated probability it was BB last year? Bayes lets you reverse known forward probabilities. Important ethical note: pick priors (P(A)) honestly — you’re not allowed to invent priors to justify a model that looks nice.

---

5) Independence vs mutual exclusivity — roommates, not the same thing

• Mutually exclusive: A and B cannot both occur (e.g., a loan both fully repays and defaults at the same instant). If exclusive, P(A ∩ B) = 0.
• Independent: occurrence of A gives no information about B; P(A|B) = P(A). If independent, P(A ∩ B) = P(A)P(B).

Meme metaphor: "Mutually exclusive" is like choosing pizza or sushi for dinner — you can’t eat both at the exact same time. "Independent" is like two roommates ordering takeout independently; one chooses pizza and the other’s choice doesn’t affect the other.

In finance, independence assumptions are convenient but dangerous. Correlations spike under stress — don’t be the analyst who assumed independence during a crisis.

---

6) Discrete vs continuous distributions — pick your weapon

Characteristic	Discrete	Continuous
Outcomes	Countable (e.g., defaults)	Uncountable (returns)
Probability	P(X = x)	f(x) dx
CDF	F(x) = P(X ≤ x)	Same idea, continuous
Examples	Bernoulli, Binomial, Poisson	Normal, Lognormal

Practical note: Stock returns often modeled as continuous (lognormal); counts of defaults are discrete (binomial/Poisson). Choose the right model for the question.

---

7) Expected value and variance — your probabilistic rulers

• Expected value (mean): E[X] = Σ x P(x) (discrete) or ∫ x f(x) dx (continuous). In portfolio land, it’s the expected return.

• Variance: Var(X) = E[(X − E[X])^2]. For two assets, Var(aX + bY) = a^2Var(X) + b^2Var(Y) + 2abCov(X,Y).

Tie-back to Basic Statistics: you already met mean and variance; now view them through probability as properties of a random variable. Combine with TVM when you discount expected cash flows.

Code-like reminder for expected return of discrete outcomes:

E[R] = Σ_i (R_i * P(R_i))

---

8) A quick worked example — small and glorious

Scenario: A corporate bond has a 3% annual default probability. If default, recovery is 40% of face. What is the expected loss rate this year?

Calculation: Expected loss = P(default) × (1 − recovery) = 0.03 × 0.60 = 0.018 = 1.8%.

If two bonds independent with same profile, probability both default = 0.03^2 = 0.0009 (0.09%). But if correlation exists, this changes — and that’s where models and judgment meet (and ethics must step in).

---

Closing: Key takeaways and the ethical mic drop

• Probability is the toolset for turning uncertain futures into actionable analysis: expected values, risks, joint events, and updates from new data.
• Be explicit about assumptions (priors, independence, distributions). Hiding assumptions is an ethical failure, not a math shortcut.
• Use complements and Bayes — they often make messy problems elegant.
• Match model to question: discrete for counts, continuous for returns, and never assume independence without stress-testing it.

Final thought: Probability doesn't eliminate uncertainty — it organizes it. Your job as a future CFA is to communicate probabilities honestly, stress-test them relentlessly, and resist the seductive siren of pretty outputs that rest on dishonest priors. That, my friend, is both good quant practice and sound ethics.

"Probability is not prophecy; it's a careful, humble map of our ignorance." — go map responsibly.

Version: If you want, next we can do a short problem set: conditional probability, Bayes with market signals, and a binomial pricing mini-derivative. Snacks optional but encouraged.