University:
Massachusetts Institute of Technology
Course:
1.151 | Probability and Statistics in Engineering
Academic year:

2019
Views:

40

Pages:

1
Author:

Lucille Fox

Brief Notes #7 Conditional Second-Moment Analysis • Important result for jointly normally distributed variables X1 and X2 If X1 and X2 are jointly normally distributed with mean values m1 and m2, variances σ12 and σ22, and correlation coefficient ρ, then (X1 | X2 = x2) is also normally distributed with mean and variance: σ1 ⎧ ⎪m1|2 (x 2 ) = m1 + ρ σ (x 2 − m 2 ) 2 ⎪ ⎨ 2 ⎪ (x 2 ) = σ12 (1 − ρ 2 ) σ1|2 ⎪ ⎩ Notice that the conditional variance does not depend on x2. (1) The results in Eq. 1 hold strictly when X1 and X2 are jointly normal, but may be used in approximation for other distributions or when one knows only the first two ⎡X ⎤ moments of the vector X = ⎢ 1 ⎥ . ⎣X 2 ⎦ • Extension to many observations and many predictions ⎡X ⎤ Let X = ⎢ 1 ⎥ , where X1 and X2 are sub-vectors of X. Suppose X has multivariate ⎣X 2 ⎦ normal distribution with mean value vector and covariance matrix: Σ12 ⎤ ⎡Σ ⎡m ⎤ Σ = ⎢ 11 m = ⎢ 1 ⎥ , and (Σ12 = Σ21T). ⎥ Σ Σ m 22 ⎦ ⎣ 2⎦ ⎣ 21 Then, given X2 = x2, the conditional vector (X1 | X2 = x2) has jointly normal distributions with parameters: ⎧m1|2 (x 2 ) = m1 + Σ12 Σ −221 (x 2 − m 2 ) ⎪ ⎨ ⎪ Σ (x ) = Σ − Σ Σ −1 Σ T 1|2 2 11 12 22 12 ⎩ (2) Notice again that Σ1|2 does not depend on x2. As for the scalar case, Eq. 2 may be used in approximation when X does not have multivariate normal distribution or when the distribution of X is not known, except for the mean vector m and covariance matrix Σ.

Brief Notes #7: Conditional Second-Moment Analysis

Related Documents

Recommended Documents

Description

Related Documents

Almost There!

Free up your schedule!

Take 5 seconds to unlock