23
0

Non-standard boundary behaviour in binary mixture models

Abstract

Consider a binary mixture model of the form Fθ=(1θ)F0+θF1F_\theta = (1-\theta)F_0 + \theta F_1, where F0F_0 is standard Gaussian and F1F_1 is a completely specified heavy-tailed distribution with the same support. For a sample of nn independent and identically distributed values XiFθX_i \sim F_\theta, the maximum likelihood estimator θ^n\hat\theta_n is asymptotically normal provided that 0<θ<10 < \theta < 1 is an interior point. This paper investigates the large-sample behaviour for boundary points, which is entirely different and strikingly asymmetric for θ=0\theta=0 and θ=1\theta=1. The reason for the asymmetry has to do with typical choices such that F0F_0 is an extreme boundary point and F1F_1 is usually not extreme. On the right boundary, well known results on boundary parameter problems are recovered, giving limP1(θ^n<1)=1/2\lim \mathbb{P}_1(\hat\theta_n < 1)=1/2. On the left boundary, limP0(θ^n>0)=11/α\lim\mathbb{P}_0(\hat\theta_n > 0)=1-1/\alpha, where 1α21\leq \alpha \leq 2 indexes the domain of attraction of the density ratio f1(X)/f0(X)f_1(X)/f_0(X) when XF0X\sim F_0. For α=1\alpha=1, which is the most important case in practice, we show how the tail behaviour of F1F_1 governs the rate at which P0(θ^n>0)\mathbb{P}_0(\hat\theta_n > 0) tends to zero. A new limit theorem for the joint distribution of the sample maximum and sample mean conditional on positivity establishes multiple inferential anomalies. Most notably, given θ^n>0\hat\theta_n > 0, the likelihood ratio statistic has a conditional null limit distribution Gχ12G\neq\chi^2_1 determined by the joint limit theorem. We show through this route that no advantage is gained by extending the single distribution F1F_1 to the nonparametric composite mixture generated by the same tail-equivalence class.

View on arXiv
Comments on this paper