0 Daumen
458 Aufrufe

Ich bräuchte Hilfe bei folgender Frage:

$$x = (x_{i}, ..., x_{n})'$$ sei ein Vektor von Zufallsvariablen. y sei ein Vektor mit Elementen $$y_{1} = x_{1}, y_{i} = x_{i}-x_{i-1}$$ mit $$i = 2,3,...n$$.

Angenommen $$y_{i}$$ sind gegenseitig unabhängige Zufallsvariablen, jede mit Einheitsvarianz so das Cov[y]=$$I$$ (Einheitsmatrix). Was ist Cov[x]?

Besten Dank!

Avatar von

1 Antwort

0 Daumen
 
Beste Antwort

Behauptung:

$$ E( x_i x_j ) = \min(i,j) $$

Induktionsanfang mit \( i = 1 \)

Die erste Zeile der Kovaraianzmatrix für \( y \) sieht so aus

$$ E \begin{pmatrix} x_1^2 && x_1 x_2 - x_1^2 && x_1 x_3 - x_1 x_2 && ... \end{pmatrix} $$

Wegen \(  E(x_1^2) = 1 \) folgt auch \( E(x_1 x_i) = 1 \)

Die i-te Zeile der Kovarianzmatrix für \( y \) sieht so aus

$$ E \begin{pmatrix}   .. &&  (i,i) && .. && (i,j) && .. \\ .. && x_i^2 - 2 x_{i-1} x_i + x_{i-1}^2 && .. &&  x_i x_j - x_i x_{j-1} - x_{i-1} x_j + x_{i-1} x_{j-1} && .. \end{pmatrix}  $$

Unter der Induktionsvoraussetzung für \( i - 1 \) folgt $$ E ( x_i^2 ) = 2 E(x_{i-1} x_i) -E(x_{i-1}^2)+1 = 2 (i - 1) - (i - 1) + 1 = i  $$ und weil \( j > i \) gilt, folgt $$ E(x_i x_j) = E(x_i x_{j-1}) + E(x_{i-1} x_j) - E(x_{i-1} x_{j-1})  = i + (i - 1) - (i-1) = i $$

D.h. die Kovarianzmatrix sieht so aus

$$ \begin{pmatrix}  1 && 1 &&  ... && 1 && 1 \\ 1 && 2 &&  ... && 2 && 2 \\ ... && ... && ... &&  ... && ... \\  1 && 2 &&  ... && n-1 && n-1 \\ 1 && 2 &&  ... && n-1 && n \end{pmatrix} $$

Avatar von 39 k

Es geht auch einfacher. Sei

$$ y = A x $$ mit $$ A = \begin{pmatrix}  1 &&  0 && ... && 0 && 0 && 0 \\ -1 && 1 && .. &&  0 && 0 && 0 \\  ... && ... && ... && ... && ... \\  0 && 0 && ... &&  -1 && 1 && 0 \\ 0 && 0 && ... && 0 && -1 && 1 \end{pmatrix}  $$

Dann gilt weiter

$$ E = \text{Cov}(y) = E \{ ( y-\overline{y} ) ( ( y - \overline{y} )^T \} = $$

$$ E \{ A ( x - \overline{x} ) ( x - \overline{x} )^T A^T \} = A \cdot \text{Cov}(x) \cdot A^T $$

Daraus folgt $$  \text{Cov}(x) = A^{-1} A^{-T} = (A^T A)^{-1} $$

Ergebnis ist das gleiche.

Ein anderes Problem?

Stell deine Frage

Ähnliche Fragen

Willkommen bei der Mathelounge! Stell deine Frage einfach und kostenlos

x
Made by a lovely community