By Scott Sheffield, as credited on the PDF title slide. No date is assigned from the URL alone.
Cited passages: PDF pp. 4–8 define conditional expectation and its uniqueness up to almost surely equality; pp. 9–11 discuss integrability and uniqueness.
Once a nonnegative martingale hits zero, it stays zero almost surely. Conditional expectation on gives ; nonnegativity forces the next value to vanish there. A countable intersection makes these assertions simultaneous for all times, proving absorption at the first zero.
If the parametric score function has zero conditional expectation given a covariate , it is orthogonal to the centered functions of that form the covariate-density nuisance tangent space. Its efficient score then equals its parametric score function, and the unknown covariate distribution causes no loss of Fisher information. This applies to Gaussian regression scores and more generally to regular conditional models with unrestricted covariate distribution and no additional nuisance components.
Aggregate claims model Created 2026-10-06 Updated 2026-11-05
An aggregate claims model separates the number of claims from their sizes , with total and the empty sum equal to zero. Under independent and identically distributed random variables for claim sizes, independent of , conditional expectation gives tractable moments and transforms.
Autocorrelation Created 2026-09-28 Updated 2026-10-14
The autocorrelation function of a real weakly stationary process with positive variance is its normalized autocovariance:
It is dimensionless, with , , and by the Cauchy-Schwarz inequality. It is unchanged by replacing with for fixed . This makes autocorrelation useful when comparing records measured on different scales. The ACF is a property of the model; the sample autocorrelation function estimates it and will generally not reproduce theoretical zeros exactly.
An AR(1) with coefficient has autocorrelations at lags one through three. With these become : the same diminishing magnitude, alternating in sign. An MA(1) with instead has lag-one autocorrelation and zero at every higher lag. Equal numerical coefficients in an autoregressive model and a moving-average model do not mean equal dependence. See Bartlett, lecture 3: theoretical ACFs and prediction.
Figure 1.
Theoretical autocorrelation: AR(1) with coefficients 0.7 and -0.7, compared with MA(1) with coefficient 0.7
. Details.
The bars show exact model autocorrelations, not estimates from the illustrated sample paths in time series. The AR(1) retains a diminishing trace of a disturbance at arbitrarily large lags; the MA(1) stops sharing driving noise after lag one. A zero autocorrelation means uncorrelated random variables, and does not generally imply independent random variables. For a Gaussian process it does imply independence of the two coordinates.
There is a direct prediction interpretation. Among affine predictions based on just , the best linear predictor of is , with mean squared prediction error . Thus negative autocorrelation is also useful predictive information. A prediction using the whole observed past can do better; best linear prediction from a finite past describes that problem. Interpreting the affine prediction as a conditional expectation requires additional assumptions, such as a joint normal distribution.
Autoregressive process of order one Created 2026-09-28 Updated 2026-10-14
An autoregressive process of order one obeys
where is real and is centered white noise of variance . The zero-mean version is ; in the affine form , the stationary mean satisfies .
For , its unique weakly stationary process solution on all integer times is the causal time-series representation
The sum converges in the sense of mean-square convergence because . It is an infinite moving-average representation using only present and past driving noise. Weak white noise suffices for this second-order conclusion; strong white noise also gives strict stationarity. The autocovariance of an AR(1) process is , so its variance is and its ACF is . See Bartlett, lecture 5: AR(1) representation and causality.
The appeal of an AR(1) is that one coefficient expresses how a deviation persists. For , disturbances propagate with weights . For , these weights alternate sign. With , , and , the recursive forecasts of an autoregressive process give and one and two steps ahead. If , the stationary variance is , while the one-step forecast error variance is only . Under weak white noise these are best linear forecasts; independent driving noise also makes them conditional expectations given the observed past.
The condition describes a stationary causal solution, not every finite recursion with those coefficients. Starting from a prescribed value gives the transient in finite initialization of an AR(1) process. Nor is it a condition for all possible two-sided stationary solutions: when , noncausal stationary autoregression uses future driving noise. See stationary versus causal solution of a two-sided AR(1) equation for the boundary cases.
Bayes formula for conditional expectation Created 2026-09-28 Updated 2026-11-05
Suppose is a probability measure with , so and . For a sub-sigma-algebra , write . If is integrable under , then on ,
This is Bayes formula for conditional expectation. It changes the underlying probability measure as well as conditioning on the available information. The set where has zero -probability, so a version there can be chosen arbitrarily.
To verify the formula, let denote the ratio. For , measurability of and the tower property give . These are exactly the defining conditional-expectation identities. For a filtration, choosing gives the usual time-indexed form. The denominator is the likelihood weight after reducing the full information to , not the original pointwise weight .
Bayesian credibility Created 2026-10-14 Updated 2026-11-05
Bayesian credibility estimates a risk's conditional expectation by its Bayesian posterior expected value under squared-error loss. When that posterior mean is an affine combination of exposure-weighted individual experience and a prior mean, the coefficient of the experience is its credibility factor. Exact Bayesian credibility is distinct from merely selecting the best affine estimator when the true posterior mean is nonlinear.
Bellman equation Created 2026-09-24 Updated 2026-10-14
A Bellman equation expresses the dynamic programming principle as a recurrence relation for a value function: optimize the current reward plus the best attainable continuation. Let be a dynamic programming state, the allowed actions at time , the immediate reward, and the next dynamic programming state. With terminal time and terminal reward , the deterministic finite-horizon Bellman equation is
Here measures total reward from time through the terminal reward. Start with and evaluate backwards; the terminal condition anchors the recurrence relation. For finite nonempty action sets, finite rewards, and finite terminal rewards, a maximum exists at each step. If an optimum need not be attained, use a supremum. A minimization problem instead uses a minimum or infimum. The 0-1 knapsack problem above is the same comparison between an immediate item value plus a continuation and the continuation after skipping. See Bertsekas, MIT lecture 2: the backwards recursion.
In a Markov decision process, the next dynamic programming state is random. If denotes the expected immediate reward and is a discount factor, the finite-horizon Bellman equation becomes
The conditional expectation averages over the next dynamic programming state before choosing the current action; that action cannot use a future outcome that has not yet been observed. For a finite state space with transition probabilities , the continuation term is . Taking gives the undiscounted finite-horizon problem. With a discount factor , rewards steps ahead carry a factor , including the terminal reward. A maximizing action for each specifies a Markov policy, which contains decisions rather than just optimal scores.
For an infinite horizon with a finite state space, finite nonempty action sets, time-independent transition probabilities, bounded rewards , and a discount factor , the optimal value function satisfies
In this setting the maximizing dynamic programming operator is a contraction mapping in the supremum norm, so the Banach fixed-point theorem gives a unique bounded fixed point and convergence of repeated application to it. See Bertsekas, MIT lecture 15: bounded discounted problems. These hypotheses matter: with one state, one action, zero reward, and , the Bellman equation is merely , so every finite constant is a fixed point, although the actual total-reward value function is zero. The average-reward Bellman equation and Hamilton-Jacobi-Bellman equation address other settings with their own conditions.
Bernoulli polarization martingale Created 2026-10-25 Updated 2026-10-31
Let , , and let be a binary random variable satisfying . Suppose is -measurable and is -measurable. The update
remains in and defines a bounded martingale. For example, take independent uniform random variables independent of the current filtration and set .
The binary identity gives
and hence
The L2-bounded martingale convergence theorem gives a terminal random variable with almost sure convergence and convergence in L2. Iteration followed by the bounded convergence theorem yields
Thus takes only and , and preservation of its expected value gives a Bernoulli distribution with parameter . The terminal conditional expectation representation gives the exact mean squared error
The finite-time paths can stay strictly inside while their limit lies entirely at the endpoints. The positive conditional variance in the interior makes permanent convergence to an interior value impossible.
A best affine state estimate minimizes mean squared error among estimates formed from the available observation vector . With finite second moments and invertible observation covariance matrix, it is
This follows from the orthogonal projection of centered state components onto the linear span of centered observation components. For Jointly Gaussian state and observation it equals the conditional expectation; for general laws a nonlinear estimate can have lower mean squared error.
Binomial call-price recursion Created 2026-09-24 Updated 2026-11-05
Let a binomial options pricing model multiply the stock price by or in each period, with and bank account gross return . The unique one-step risk-neutral probability is . If is the time-zero price of an -period call option starting at with strike , then
This is conditional expectation after the first move, with one period of discounting. The terminal payoff's positive homogeneity gives for , hence
The coefficients are positive because the no-arbitrage inequalities put strictly between zero and one. This strike-rescaling recursion is useful when the shorter-maturity call curve is available. The physical up probability does not enter the recursion.
For every there is a positive integer such that every finite bipartite graph has an -regular pair of partitions with at most cells on each side.
To prove this, give the uniform probability measure, let be the indicator function of the edge set, and for partitions define their energy by
This conditional expectation is an orthogonal projection, so . If the partitions are not -regular, choose witnesses and in every irregular cell pair and refine each by all its and each by all its . The refined energy exceeds the old energy by
On the absolute value function of the mean of the difference is greater than , while this rectangle occupies at least an fraction of . The Cauchy-Schwarz inequality therefore contributes more than from that cell pair. Irregular pairs have total weight greater than , so every refinement raises the energy by more than . The process stops after at most refinements. If the current cell counts are , the construction gives at most new cells, so a finite iterated bound depending only on supplies .
Birkhoff ergodic theorem Created 2026-09-24 Updated 2026-11-05
Let preserve a probability measure , let , and let be the invariant sigma-algebra. The Birkhoff ergodic theorem states
almost everywhere. The convergence also holds in , and the limit has the same integral as . A nonzero finite measure can first be normalized to probability. Invertibility and ergodicity are unnecessary for this conditional-expectation form.[ref]
If is ergodic, the limit is the constant . If is the identity, the averages equal at every time and is the whole sigma-algebra, so the limit need not be constant. The conditional expectation is exactly the portion of the observable that remains visible in invariant components.
A proof first handles invariant functions and bounded differences , whose averages telescope to . A dense approximation argument plus the maximal ergodic lemma controls the error and yields pointwise convergence for integrable data. Bounded truncations and the averaging operators' contraction give the L1 convergence in the Birkhoff ergodic theorem on a finite measure space.
Finiteness is essential to the stated same-integral conclusion. On with counting measure and , take . Every time average tends to zero, although . The infinite-measure pointwise theorem therefore requires a different formulation, rather than the unchanged probability statement.
An adapted process with integrable coordinates is a martingale exactly when the displayed equality holds for every bounded stopping time. For the forward implication, expand and condition each term on . Conversely, for choose on and elsewhere. Subtract the equality for the constant stopping time to obtain , which is the defining conditional expectation identity.
The conditional expectation of at time is , by independent Brownian increment. Itô's formula cancels its time derivative against its second spatial derivative and leaves the displayed stochastic integral. The integrand is bounded, so Itô isometry verifies its square integrability.
Bühlmann–Straub credibility estimate Created 2026-10-06 Updated 2026-11-05
For total exposure and weighted average , the best affine mean squared error estimate of the conditional expectation is , where . If the coefficients sum to , their error is . Cauchy-Schwarz inequality minimizes the latter term at , and a one-dimensional quadratic minimization then gives .
Bühlmann–Straub credibility factor Created 2026-10-06 Updated 2026-11-05
The experience weight is , where is total observed exposure, the variance of hypothetical means, and the expected process variance. With positive , it is . More exposure raises the weight; larger process noise lowers it. If , the unknown conditional expectation is constant and one can set .
Bühlmann–Straub model Created 2026-10-06 Updated 2026-11-05
Conditional on a common latent variable , annual averages are independent, have common conditional expectation , and have conditional variance for known positive exposures . Its structural parameters are , expected process variance and variance of hypothetical means . It generalizes the equal-exposure Bühlmann model.