17 Limit theorems and simulations Flashcards

1
Q

(T) Strong law of large numbers

A

(Xi)i∈N iid RV’s such that E(|Xi|) < ∞. With µ := E(X1) we then have

P(lim.n→∞ X(n)– = µ) = 1 resp. X(n)– → µ (n→∞, almost surely)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

(T) Glivenko-Cantelli (fundam. theorem of stat.)

A

(Xi)i∈N sequence of independent RV’s having the same CDF F. Then almost surely

||Fn - F||∞ → 0 as n→∞.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

(T) Kolmogorov

A

(Xi)i∈N independent sequence of RV’s having continuous CDF F. Then we have uniformly in t ∈ R

P(√n Dn(X1, …, Xn) ≤ t) → K(t) (n→∞)

with K : R → [0, 1] the continuous Kolmogorov CDF
K(t) = ∑k=-∞^∞ (-1)^k * e^(-2k^2t^2) if t>0 // 0 if t≤0

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

(D) Kolmogorov-Smirnov-GOF: Teststatistik, Verteilung, Idee

A
  1. Idee
    Für großen Stichprobenumfang liegt ECDF der Stichprobe sehr nahe an zugrundeliegender CDF (Glivenko-Cantelli). Man nimmt H0:F=G an, wenn maximaler Abstand zwischen den beiden nicht zu groß.
  2. Teststatistik
    Dn(X1, …, Xn) = sup_t∈R |Fn(t;X1, …, Xn) - F(t)|
  3. Verteilung
    Kolmogorov-CDF K ist CDF von √n Dn
    (dh, H0 annehmen, wenn √n Dn ≤ K1-a)
  4. Hypothesenpaar
    H0: F=G, H1: F≠G
  5. Annahmebereich
    I = [0, K1-a/√n]
How well did you know this?
1
Not at all
2
3
4
5
Perfectly