Conditional Expectation

This is the most important article in the module.

Conditional expectation $E[X\mid {\mathcal{F}}_t]$ is not a number --- it is a random variable. This subtlety is where most people’s intuition breaks down, and it is precisely the concept that makes derivative pricing, martingale theory, and dynamic risk management mathematically coherent.

The idea is deceptively simple: given partial information (modeled by a $\sigma$-algebra ${\mathcal{F}}_t$, as introduced in sample-spaces-and-sigma-algebras), what is the best forecast of $X$? The answer depends on which partial information you have --- different realizations of the world up to time $t$ give different forecasts. Hence $E[X\mid {\mathcal{F}}_t]$ is a function of $\omega$, not a scalar.

This article starts from Bayes’ theorem (the discrete warm-up), builds through conditional distributions, arrives at the measure-theoretic definition of conditional expectation, and culminates in the tower property --- the single result that underpins all of dynamic pricing theory.

Key Topics

• Bayes’ theorem: $P(A\mid B)=\frac{P(B\mid A)P(A)}{P(B)}$ --- the discrete starting point
• Conditional distributions: $f_{X|Y}(x\mid y)=\frac{f_{X,Y}(x,y)}{f_Y(y)}$
• $E[X\mid Y]$ as a random variable: it is ${\mathcal{F}}_Y$-measurable, meaning it depends on $\omega$ only through $Y(\omega )$. This is the subtle leap from elementary to measure-theoretic probability
• Formal definition: $E[X\mid {\mathcal{F}}_t]$ is the unique (a.s.) ${\mathcal{F}}_t$-measurable random variable $Z$ satisfying $E[Z\cdot 1_A]=E[X\cdot 1_A]$ for all $A\in {\mathcal{F}}_t$
• Tower property (law of iterated expectations): $E\left[E[X\mid {\mathcal{F}}_t]\mid {\mathcal{F}}_s\right]=E[X\mid {\mathcal{F}}_s]$ for $s\le t$. In the special case: $E\left[E[X\mid Y]\right]=E[X]$
• Conditional expectation as best $L^2$ predictor: $E[X\mid {\mathcal{F}}_t]$ minimizes $E\left[(X-Z{)}^2\right]$ over all ${\mathcal{F}}_t$-measurable $Z$. This is the bridge to regression: OLS is conditional expectation under a linearity constraint

Finance Connections

• Martingales: $M_t$ is a martingale iff $E[M_{t+1}\mid {\mathcal{F}}_t]=M_t$. Under the risk-neutral measure $Q$, discounted asset prices are martingales. This is the Fundamental Theorem of Asset Pricing in action
• Dynamic pricing: the price of a derivative at time $t$ is $V_t=e^{-r(T-t)}{E}^Q[V_T\mid {\mathcal{F}}_t]$. This formula IS conditional expectation
• Tower property in pricing: price today = discounted expectation of tomorrow’s price. This lets you build pricing backward through a tree (binomial model) or forward through simulation (Monte Carlo)
• Bayesian credit scoring: updating default probability as new information arrives is conditioning on a growing $\sigma$-algebra --- exactly what you were doing at Credimi, implicitly
• Regression: $E[Y\mid X]=\alpha +\beta X$ under linearity. The multivariate regression at Credimi was estimating conditional expectations of credit outcomes given financial ratios

Prerequisites

• joint-distributions-and-independence

---

This article is a roadmap --- content to be developed in future sessions.