Conjugacy as a Distinctive Feature of the Dirichlet Process

Abstract.  Recently the class of normalized random measures with independent increments, which contains the Dirichlet process as a particular case, has been introduced. Here a new technique for deriving moments of these random probability measures is proposed. It is shown that, a priori , most of the appealing properties featured by the Dirichlet process are preserved. When passing to posterior computations, we obtain a characterization of the Dirichlet process as the only conjugate member of the whole class of normalized random measures with independent increments.     

Initializing EM using the properties of its trajectories in Gaussian mixtures

A strategy is proposed to initialize the EM algorithm in the multivariate Gaussian mixture context. It consists in randomly drawing, with a low computational cost in many situations, initial mixture parameters in an appropriate space including all possible EM trajectories. This space is simply defined by two relations between the two first empirical moments and the mixture parameters satisfied by any EM iteration. An experimental study on simulated and real data sets clearly shows that this strategy outperforms classical methods, since it has the nice property to widely explore local maxima of the likelihood function.     

VARIANCE ESTIMATION WITH HIGHLY STRATIFIED SAMPLING DESIGNS WITH UNEQUAL PROBABILITIES

It is common practice to design a survey with a large number of strata. However, in this case the usual techniques for variance estimation can be inaccurate. This paper proposes a variance estimator for estimators of totals. The method proposed can be implemented with standard statistical packages without any specific programming, as it involves simple techniques of estimation, such as regression fitting.     

正态随机变量线性组合的分布探讨

文章构造两例,分别说明两个正态随机变量之和不是正态随机变量,而n个(n≥3)正态随机变量之和亦不是正态随机变量,对教学具有一定的指导意义.     

Automatic choice of driving values in Monte Carlo likelihood approximation via posterior simulations

For models with random effects or missing data, the likelihood function is sometimes intractable analytically but amenable to Monte Carlo approximation. To get a good approximation, the parameter value that drives the simulations should be sufficiently close to the maximum likelihood estimate (MLE) which unfortunately is unknown. Introducing a working prior distribution, we express the likelihood function as a posterior expectation and approximate it using posterior simulations. If the sample size is large, the sample information is likely to outweigh the prior specification and the posterior simulations will be concentrated around the MLE automatically, leading to good approximation of the likelihood near the MLE. For smaller samples, we propose to use the current posterior as the next prior distribution to make the posterior simulations closer to the MLE and hence improve the likelihood approximation. By using the technique of data duplication, we can simulate from the sharpened posterior distribution without actually updating the prior distribution. The suggested method works well in several test cases. A more complex example involving censored spatial data is also discussed.     

The coefficient of variation asymptotic distribution in the case of non-iid random variables

Due to the widespread use of the coefficient of variation in empirical finance, we derive its asymptotic sampling distribution in the case of non-iid random variables to deal with autocorrelation and/or conditional heteroskedasticity stylized facts of financial returns. We also propose statistical tests for the comparison of two coefficients of variation based on asymptotic normality and studentized time-series bootstrap. In an illustrative example, we analyze the monthly return volatility of six stock market indexes during the years 1990–2007.     

Sampling from the posterior distribution in generalized linear mixed models

Generalized linear mixed models provide a unified framework for treatment of exponential family regression models, overdispersed data and longitudinal studies. These problems typically involve the presence of random effects and this paper presents a new methodology for making Bayesian inference about them. The approach is simulation-based and involves the use of Markov chain Monte Carlo techniques. The usual iterative weighted least squares algorithm is extended to include a sampling step based on the Metropolis–Hastings algorithm thus providing a unified iterative scheme. Non-normal prior distributions for the regression coefficients and for the random effects distribution are considered. Random effect structures with nesting required by longitudinal studies are also considered. Particular interests concern the significance of regression coefficients and assessment of the form of the random effects. Extensions to unknown scale parameters, unknown link functions, survival and frailty models are outlined.     

B值同分布鞅随机列矩完全收敛性的注记

讨论了B值同分布鞅随机变量的矩完全收敛性,在一定矩条件下,利用切尾法和下鞅的极大值不等式等分析技巧,得到了同分布鞅随机变量的矩完全收敛性,将Chow实值独立同分布随机变量的矩完全收敛的结果在B值鞅的情况下进一步推广,在补充了B值鞅随机变量收敛的条件下,得到了平方矩存在的条件与同分布一样的结果。     

Resampling Methods for Testing a Semiparametric Random Censorship Model

This paper presents a goodness-of-fit test for a semiparametric random censorship model proposed by Dikta (1998 ). The test statistic is derived from a model-based process which is asymptotically Gaussian. In addition to test consistency, the proposed test can detect local alternatives distinct n ^-1/2 from the null hypothesis. Due to the intractability of the asymptotic null distribution of the test statistic, we turn to two resampling approximations. We first use the well-known bootstrap method to approximate critical values of the test. We then introduce a so-called random symmetrization method for carrying out the test. Both methods perform very well with a sample of moderate size. A simulation study shows that the latter possesses better empirical powers and sizes for small samples.     

Even Risk-Averters may Love Risk

A decision maker bets on the outcomes of a sequence of coin-tossings. At the beginning of the game the decision maker can choose one of two coins to play the game. This initial choice is irreversible. The coins can be biased and the player is uncertain about the nature of one (or possibly both) coin(s). If the player is an expected-utility maximizer, her choice of the coin will depend on different elements: the nature of the game (namely, whether she can observe the outcomes of the previous tosses before making her next decision), her utility function, the prior distribution on the bias of the coin. We will show that even a risk averter might optimally choose a riskier coin when learning is allowed. We will express most of our results in the language of stochastic orderings, allowing comparisons that are valid for large classes of utility functions.