Likelihood Ratio Type Research Articles

Robust hypothesis testing in functional linear models

We extend three robust tests – Wald-type, the likelihood ratio-type and F-type in functional linear models with the scalar dependent variable and the functional covariate. Based on the percentage of variance explained criterion, we use the functional principal components analysis and re-express a functional linear model to a finite regression. We investigate the theoretical properties of these robust testing procedures and assess the finite sample properties through the numerical simulation. In our experiments, the power performance and Type I error rates are studied separately in the sparsely and densely functional linear models. The simulation results show that the robust test procedures are more stable and less sensitive to heavy-tailed distributed errors than the classical ones. Two real datasets are analysed to compare the classical and robust testing procedures.

Goodness of fit tests for random multigraph models

Goodness of fit tests for two probabilistic multigraph models are presented. The first model is random stub matching given fixed degrees (RSM) so that edge assignments to vertex pair sites are dependent, and the second is independent edge assignments (IEA) according to a common probability distribution. Tests are performed using goodness of fit measures between the edge multiplicity sequence of an observed multigraph, and the expected one according to a simple or composite hypothesis. Test statistics of Pearson type and of likelihood ratio type are used, and the expected values of the Pearson statistic under the different models are derived. Test performances based on simulations indicate that even for small number of edges, the null distributions of both statistics are well approximated by their asymptotic -distribution. The non-null distributions of the test statistics can be well approximated by proposed adjusted -distributions used for power approximations. The influence of RSM on both test statistics is substantial for small number of edges and implies a shift of their distributions towards smaller values compared to what holds true for the null distributions under IEA. Two applications on social networks are included to illustrate how the tests can guide in the analysis of social structure.

ESTIMATION OF A HIGH-DIMENSIONAL COUNTING PROCESS WITHOUT PENALTY FOR HIGH-FREQUENCY EVENTS

This paper introduces a counting process for event arrivals in high-frequency trading, based on high-dimensional covariates. The novelty is that, under sparsity conditions on the true model, we do not need to impose any model penalty or parameters shrinkage, unlike Lasso. The procedure allows us to derive a central limit theorem to test restrictions in a two-stage estimator. We achieve this by the use of a sign constraint on the intensity which necessarily needs to be positive. In particular, we introduce an additive model to extract the nonlinear impact of order book variables on buy and sell trade arrivals. In the empirical application, we show that the shape and dynamics of the order book are fundamental in determining the arrival of buy and sell trades in the crude oil futures market. We establish our empirical results mapping the covariates into a higher-dimensional space. Consistently with the theoretical results, the estimated models are sparse in the number of parameters. Using this approach, we are also able to compare competing model hypotheses on the basis of an out-of-sample likelihood ratio type of test.

A likelihood-ratio type test for stochastic block models with bounded degrees

A fundamental problem in network data analysis is to test Erdös–Rényi model Gn,a+b2n versus a bisection stochastic block model Gn,an,bn, where a,b>0 are constants that represent the expected degrees of the graphs and n denotes the number of nodes. This problem serves as the foundation of many other problems such as testing-based methods for determining the number of communities (Bickel and Sarkar, 2016; Lei, 2016) and community detection (Montanari and Sen, 2016). Existing work has been focusing on growing-degree regime a,b→∞ (Bickel and Sarkar, 2016; Lei, 2016; Montanari and Sen, 2016; Banerjee and Ma, 2017; Banerjee, 2018; Gao and Lafferty, 2017a,b) while leaving the bounded-degree regime untreated. In this paper, we propose a likelihood-ratio (LR) type procedure based on regularization to test stochastic block models with bounded degrees. We derive the limit distributions as power Poisson laws under both null and alternative hypotheses, based on which the limit power of the test is carefully analyzed. We also examine a Monte-Carlo method that partly resolves the computational cost issue. The proposed procedures are examined by both simulated and real-world data. The proof depends on a contiguity theory developed by Janson (1995).

Statistics of SPDEs: From Linear to Nonlinear

We study statistical inference for stochastic partial differential equations (SPDEs). Though inference linear SPDEs have been studied well (with lot of problems still remain to be investigated) in the last two decades, inference for nonlinear SPDEs is in its infancy. The inference methods use both inference for finite-dimensional diffusions and inference for classical i.i.d. sequences. Solving 2D Navier-Stokes equation is one of the challenging problem of the last century. However, with additive white noise, the equation has a strong solution. We estimate the viscosity coefficient of the 2D stochastic Navier-Stokes (SNS) equation by minimum contrast method. We show $n$ consistency in contrast to $\sqrt n$ consistency in the classical i.i.d. case where $n$ is the number of observations. We consider both continuous and discrete observations in time. We also obtain the Berry-Esseen bounds. Then we estimate and control the Type I and Type II error of a simple hypothesis testing problem of the viscosity coefficient of the SNS equation. We study a class of rejection regions and provide thresholds that guarantee that the statistical errors are smaller than the given upper bound. The tests are of likelihood ratio type. The proofs are based on the large deviation bounds. Finally we give Monte Carlo test procedure for simulated data.

Adaptive testing method for ergodic diffusion processes based on high frequency data

We consider parametric tests for multidimensional ergodic diffusions based on high frequency data. We propose two-step testing method for diffusion parameters and drift parameters. To construct test statistics of the tests, we utilize the adaptive estimator and provide three types of test statistics: likelihood ratio type test, Wald type test and Rao’s score type test. It is proved that these test statistics converge in distribution to the chi-squared distribution under null hypothesis and have consistency of the tests against alternatives. Moreover, these test statistics converge in distribution to the non-central chi-squared distribution under local alternatives. We also give some simulation studies of the behavior of the three types of test statistics.

On the conditional test of order restricted multivariate normal mean vectors: simulation study and application

Multivariate isotonic regression theory plays an important role in the field of testing statistical hypotheses under order restriction for several vector valued parameters and this problem may arise in several situations. Sasabuchi et al. derived a likelihood ratio type test for homogeneity of order restricted mean vectors under a multivariate normality assumption with unknown and common covariance matrices; however, its null distribution depends on the unknown covariance matrix. In order to overcome this difficulty, in the present paper we propose a test that conditions on the sufficient statistic for the covariance matrix. We remove the dependence of the null distribution of test statistic proposed by Sasabuchi et al. on the unknown value of covariance matrix and present a condition on a sufficient statistic for covariance matrix under the null hypothesis. A method based on Markov Chain Monte Carlo sampling is used to calculate the value of the test. We evaluate the operating characteristics of the proposed test through simulation studies. Our extensive simulation studies demonstrate that the power of this new test is generally higher than the test proposed by Sasabuchi et al. and also by Sasabuchi. A real data example is presented.

Estimation and Testing in M‐quantile Regression with Applications to Small Area Estimation

SummaryIn recent years, M‐quantile regression has been applied to small area estimation to obtain reliable and outlier robust estimators without recourse to strong parametric assumptions. In this paper, after a review of M‐quantile regression and its application to small area estimation, we cover several topics related to model specification and selection for M‐quantile regression that received little attention so far. Specifically, a pseudo‐R2 goodness‐of‐fit measure is proposed, along with likelihood ratio and Wald type tests for model specification. A test to assess the presence of actual area heterogeneity in the data is also proposed. Finally, we introduce a new estimator of the scale of the regression residuals, motivated by a representation of the M‐quantile regression estimation as a regression model with Generalised Asymmetric Least Informative distributed error terms. The Generalised Asymmetric Least Informative distribution, introduced in this paper, generalises the asymmetric Laplace distribution often associated to quantile regression. As the testing procedures discussed in the paper are motivated asymptotically, their finite sample properties are empirically assessed in Monte Carlo simulations. Although the proposed methods apply generally to M‐quantile regression, in this paper, their use ar illustrated by means of an application to Small Area Estimation using a well known real dataset.

A continuous threshold expectile model

Expectile regression is a useful tool for exploring the relation between the response and the explanatory variables beyond the conditional mean. A continuous threshold expectile regression is developed for modeling data in which the effect of a covariate on the response variable is linear but varies below and above an unknown threshold in a continuous way. The estimators for the threshold and the regression coefficients are obtained using a grid search approach. The asymptotic properties for all the estimators are derived, and the estimator for the threshold is shown to achieve root-n consistency. A weighted CUSUM type test statistic is proposed for the existence of a threshold at a given expectile, and its asymptotic properties are derived under both the null and the local alternative models. This test only requires fitting the model under the null hypothesis in the absence of a threshold, thus it is computationally more efficient than the likelihood-ratio type tests. Simulation studies show that the proposed estimators and test have desirable finite sample performance in both homoscedastic and heteroscedastic cases. The application of the proposed method on a Dutch growth data and a baseball pitcher salary data reveals interesting insights. The proposed method is implemented in the R package cthreshER.

Robust bent line regression

We introduce a rank-based bent linear regression with an unknown change point. Using a linear reparameterization technique, we propose a rank-based estimate that can make simultaneous inference on all model parameters, including the location of the change point, in a computationally efficient manner. We also develop a score-like test for the existence of a change point, based on a weighted CUSUM process. This test only requires fitting the model under the null hypothesis in absence of a change point, thus it is computationally more efficient than likelihood-ratio type tests. The asymptotic properties of the test are derived under both the null and the local alternative models. Simulation studies and two real data examples show that the proposed methods are robust against outliers and heavy-tailed errors in both parameter estimation and hypothesis testing.

Meta-analytic cointegrating rank tests for dependent panels

Two new panel cointegrating rank tests which are robust to cross-sectional dependence are proposed. The dependence in the data generating process is modeled using unobserved common factors. The new tests are based on a meta-analytic approach, in which the p-values of the individual likelihood-ratio (LR) type test statistics computed from defactored data are combined into the panel statistics. A simulation study shows that the tests have reasonable size and power properties in finite samples. The application of the tests is illustrated by investigating the monetary exchange rate model for a panel data of 19 countries.

Change-point analysis with bathtub shape for the exponential distribution

ABSTRACTLikelihood ratio type test statistic and Schwarz information criterion statistics are proposed for detecting possible bathtub-shaped changes in the parameter in a sequence of exponential distributions. The asymptotic distribution of likelihood ratio type statistic under the null hypothesis and the testing procedure based on Schwarz information criterion are derived. Numerical critical values and powers of two methods are tabulated for certain selected values of the parameters. The tests are applied to detect the change points for the predator data and Stanford heart transplant data.

Stochastic averaging and sensitivity analysis for two scale reaction networks

In the presence of multiscale dynamics in a reaction network, direct simulation methods become inefficient as they can only advance the system on the smallest scale. This work presents stochastic averaging techniques to accelerate computations for obtaining estimates of expected values and sensitivities with respect to the steady state distribution. A two-time-scale formulation is used to establish bounds on the bias induced by the averaging method. Further, this formulation provides a framework to create an accelerated "averaged" version of most single-scale sensitivity estimation methods. In particular, we propose the use of a centered ergodic likelihood ratio method for steady state estimation and show how one can adapt it to accelerated simulations of multiscale systems. Finally, we develop an adaptive "batch-means" stopping rule for determining when to terminate the micro-equilibration process.

On Comparing Cumulative Incidence Functions Using an Empirical Likelihood Ratio Type Test

We consider the competing risks problem with two risks and develop empirical likelihood ratio type tests for testing the null hypothesis that the cumulative incidence functions corresponding to these two risks are equal against the alternatives: (a) they are not equal and (b) they are linearly ordered. The asymptotic null distributions of the proposed test statistics are shown to have simple distribution-free representations in terms of a standard Brownian motion process. The results of a simulation study indicate that the proposed test for testing for the presence of the linear order is more powerful than a test designed for the same situation in Aly et al. (1994). To illustrate the theoretical results, we discuss an example involving survival times of mice exposed to radiation.

Inference from PseudoLikelihoods with Plug‐In Estimates

SummaryEffective implementation of likelihood inference in models for high‐dimensional data often requires a simplified treatment of nuisance parameters, with these having to be replaced by handy estimates. In addition, the likelihood function may have been simplified by means of a partial specification of the model, as is the case when composite likelihood is used. In such circumstances tests and confidence regions for the parameter of interest may be constructed using Wald type and score type statistics, defined so as to account for nuisance parameter estimation or partial specification of the likelihood. In this paper a general analytical expression for the required asymptotic covariance matrices is derived, and suggestions for obtaining Monte Carlo approximations are presented. The same matrices are involved in a rescaling adjustment of the log likelihood ratio type statistic that we propose. This adjustment restores the usual chi‐squared asymptotic distribution, which is generally invalid after the simplifications considered. The practical implication is that, for a wide variety of likelihoods and nuisance parameter estimates, confidence regions for the parameters of interest are readily computable from the rescaled log likelihood ratio type statistic as well as from the Wald type and score type statistics. Two examples, a measurement error model with full likelihood and a spatial correlation model with pairwise likelihood, illustrate and compare the procedures. Wald type and score type statistics may give rise to confidence regions with unsatisfactory shape in small and moderate samples. In addition to having satisfactory shape, regions based on the rescaled log likelihood ratio type statistic show empirical coverage in reasonable agreement with nominal confidence levels.

Detecting non-simultaneous changes in means of vectors

Likelihood ratio type test statistics are suggested for detecting changes in means of coordinates of observed random vectors. It is supposed that changes in different coordinates need not to occur at the same time. Under the assumption of no change, asymptotic distributions of the proposed test statistics are given by distributions of maxima of \(\chi ^2\) random fields. High-level exceedance probabilities of non-homogeneous \(\chi ^2\) fields may be applied to get approximate asymptotic critical values.

A Comparison of Sequential and Information‐based Methods for Determining the Co‐integration Rank in Heteroskedastic VAR Models

AbstractIn this article, we investigate the behaviour of a number of methods for estimating the co‐integration rank in VAR systems characterized by heteroskedastic innovation processes. In particular, we compare the efficacy of the most widely used information criteria, such as Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) , with the commonly used sequential approach of Johansen [Likelihood‐based Inference in Cointegrated Vector Autoregressive Models(1996)] based around the use of either asymptotic or wild bootstrap‐based likelihood ratio type tests. Complementing recent work done for the latter in Cavaliere, Rahbek and Taylor [Econometric Reviews(2014) forthcoming], we establish the asymptotic properties of the procedures based on information criteria in the presence of heteroskedasticity (conditional or unconditional) of a quite general and unknown form. The relative finite‐sample properties of the different methods are investigated by means of a Monte Carlo simulation study. For the simulation DGPs considered in the analysis, we find that the BIC‐based procedure and the bootstrap sequential test procedure deliver the best overall performance in terms of their frequency of selecting the correct co‐integration rank across different values of the co‐integration rank, sample size, stationary dynamics and models of heteroskedasticity. Of these, the wild bootstrap procedure is perhaps the more reliable overall as it avoids a significant tendency seen in the BIC‐based method to over‐estimate the co‐integration rank in relatively small sample sizes.

Nearly optimal sequential tests of composite hypotheses revisited

We revisit the problem of sequential testing composite hypotheses, considering multiple hypotheses and very general non-i.i.d. stochastic models. Two sequential tests are studied: the multihypothesis generalized sequential likelihood ratio test and the multihypothesis adaptive sequential likelihood ratio test with one-stage delayed estimators. While the latter loses information compared to the former, it has an advantage in designing thresholds to guarantee given upper bounds for probabilities of errors, which is practically impossible for the generalized likelihood ratio type tests. It is shown that both tests have asymptotic optimality properties minimizing the expected sample size or even more generally higher moments of the stopping time as probabilities of errors vanish. Two examples that illustrate the general theory are presented.

Robust tests for linear regression models based on [formula omitted]-estimates

ANOVA tests are the standard tests to compare nested linear models fitted by least squares. These tests are equivalent to likelihood ratio tests, so they have high power. However, least squares estimators are very vulnerable to outliers in the data, and thus the related ANOVA type tests are also extremely sensitive to outliers. Therefore, robust estimators can be considered to obtain a robust alternative to the ANOVA tests. Regression τ-estimators combine high robustness with high efficiency which makes them suitable for robust inference beyond parameter estimation. Robust likelihood ratio type test statistics based on the τ-estimates of the error scale in the linear model are a natural alternative to the classical ANOVA tests. The higher efficiency of the τ-scale estimates compared with other robust alternatives is expected to yield tests with good power. Their null distribution can be estimated using either an asymptotic approximation or the fast and robust bootstrap. The robustness and power of the resulting robust likelihood ratio type tests for nested linear models is studied.

A note on error estimation for hypothesis testing problems for some linear SPDEs

The aim of the present paper is to estimate and control the Type I and Type II errors of a simple hypothesis testing problem of the drift/viscosity coefficient for stochastic fractional heat equation driven by additive noise. Assuming that one path of the first $$N$$ Fourier modes of the solution is observed continuously over a finite time interval $$[0,T]$$ , we propose a new class of rejection regions and provide computable thresholds for $$T$$ , and $$N$$ , that guarantee that the statistical errors are smaller than a given upper bound. The considered tests are of likelihood ratio type. The main ideas, and the proofs, are based on sharp large deviation bounds. Finally, we illustrate the theoretical results by numerical simulations.