This preview shows page 1. Sign up to view the full content.
Unformatted text preview: 2, . . . , p Example: Normal Distribution • Suppose X1, ...., Xn are i.i.d. observations from N(a, b2). We want to ﬁnd the MoM estimators of a and b.
n 1 m1 = Xi n i=1 n 1 2 m2 = (Xi ) n i=1 Estimated Sample Moments: First two moments of a normal 2 2 N(a, b2) random variable: µ2 = E X = Var(X ) + (E[X ]) = b2 + a2 MoM estimators are computed based on the following system of equation:
µ1 = m1 µ2 = m2
n 1 a= ˆ Xi n i=1 n ˆ2 + a2 = 1 and b ˆ (Xi )2 n i=1 µ1 = E [X ] = a 1 a= ˆ n n i=1 Xi ˆ2 = 1 and b n n i=1 (Xi ) −
2 1 n n i=1 Xi 2
20 Example: Gamma Distribution • • Suppose X1, ...., Xn are i.i.d. observations from a gamma distribution with scale parameter λ and shape parameter α. We want to compute the MoM estimators of λ and α. Fact: If X has a gamma distribution with scale parameter λ and shape parameter α, then α α α −1 −λx E[X ] = λx e 2 if x ≥ 0, α α2 λ Γ(α) f (x) = E X = 2+ 2 α 0 otherwise. λ λ Var[X ] = 2 λ MoM estimators are computed based on the following system of equation:
α ˆ 1 = m1 = ˆ n λ α ˆ α2 ˆ 1 + = m2 = ˆ ˆ n λ2 λ2
n i=1 n i=1 Xi (Xi )
2 n 2 ( i=1 Xi ) α= ˆ n n 2 2= 2−( m2 − m1 n i=1 Xi i=1 Xi ) n n i=1 Xi m1 ˆ= λ n n 2 2= 2−( m2 − m1 n i=1 Xi i=1 Xi ) m2 1
21 Example: Potential Problems with MoM • Suppose X1, ...., Xn are i.i.d. observations from a uniform distribution over the interval [0,α]. We want to ﬁnd the MoM estimator of α.
n 1 m1 = Xi n i=1 α µ1 = E[X ] = 2 α ˆ = m1 2 ⇔ • • By LLN, as n →∞, the MoM estimator converges to 2 E[X] = α, which seems reasonable Major Flaw: Suppose the data are 0.2, 0.1, and 0.9. Then, MoM gives n 2 α = 2m1 = ˆ Xi n i=1 0.2 + 0.1 + 0.9 α = 2m1 = 2 × ˆ = 2 × 0.4 = 0.8 3 Our estimate of α is smaller than one of the sample values! • • In many cases, MoM estimators will have undesirable properties. So, MoM should NOT be the method of choice. Use MoM only when more sophisticated methods (like maxim...
View Full Document