Asta-Advances in Statistical Analysis最新文献

Markov chain Monte Carlo (MCMC)-based simulation approaches are by far the most common method in Bayesian inference to access the posterior distribution. Recently, motivated by successes in machine learning, variational inference (VI) has gained in interest in statistics since it promises a computationally efficient alternative to MCMC enabling approximate access to the posterior. Classical approaches such as mean-field VI (MFVI), however, are based on the strong mean-field assumption for the approximate posterior where parameters or parameter blocks are assumed to be mutually independent. As a consequence, parameter uncertainties are often underestimated and alternatives such as semi-implicit VI (SIVI) have been suggested to avoid the mean-field assumption and to improve uncertainty estimates. SIVI uses a hierarchical construction of the variational parameters to restore parameter dependencies and relies on a highly flexible implicit mixing distribution whose probability density function is not analytic but samples can be taken via a stochastic procedure. With this paper, we investigate how different forms of VI perform in semiparametric additive regression models as one of the most important fields of application of Bayesian inference in statistics. A particular focus is on the ability of the rivalling approaches to quantify uncertainty, especially with correlated covariates that are likely to aggravate the difficulties of simplifying VI assumptions. Moreover, we propose a method, where we combine both advantages of MFVI and SIVI and compare its performance. The different VI approaches are studied in comparison with MCMC in simulations and an application to tree height models of douglas fir based on a large-scale forestry data set.

This work proposes a new method for building an explanatory spatial autoregressive model in a multicollinearity context. We use Ridge regularization to bypass the collinearity issue. We present new estimation algorithms that allow for the estimation of the regression coefficients as well as the spatial dependence parameter. A spatial cross-validation procedure is used to tune the regularization parameter. In fact, ordinary cross-validation techniques are not applicable to spatially dependent observations. Variable importance is assessed by permutation tests since classical tests are not valid after Ridge regularization. We assess the performance of our methodology through numerical experiments conducted on simulated synthetic data. Finally, we apply our method to a real data set and evaluate the impact of some socioeconomic variables on the COVID-19 intensity in France.

Hyperparameter tuning is one of the most time-consuming parts in machine learning. Despite the existence of modern optimization algorithms that minimize the number of evaluations needed, evaluations of a single setting may still be expensive. Usually a resampling technique is used, where the machine learning method has to be fitted a fixed number of k times on different training datasets. The respective mean performance of the k fits is then used as performance estimator. Many hyperparameter settings could be discarded after less than k resampling iterations if they are clearly inferior to high-performing settings. However, resampling is often performed until the very end, wasting a lot of computational effort. To this end, we propose the sequential random search (SQRS) which extends the regular random search algorithm by a sequential testing procedure aimed at detecting and eliminating inferior parameter configurations early. We compared our SQRS with regular random search using multiple publicly available regression and classification datasets. Our simulation study showed that the SQRS is able to find similarly well-performing parameter settings while requiring noticeably fewer evaluations. Our results underscore the potential for integrating sequential tests into hyperparameter tuning.

We consider, robust estimation of wrapped models to multivariate circular data that are points on the surface of a p-torus based on the weighted likelihood methodology. Robust model fitting is achieved by a set of weighted likelihood estimating equations, based on the computation of data dependent weights aimed to down-weight anomalous values, such as unexpected directions that do not share the main pattern of the bulk of the data. Weighted likelihood estimating equations with weights evaluated on the torus or obtained after unwrapping the data onto the Euclidean space are proposed and compared. Asymptotic properties and robustness features of the estimators under study have been studied, whereas their finite sample behavior has been investigated by Monte Carlo numerical experiment and real data examples.

In small area estimation, the sample size is so small that direct estimators have seldom enough adequate precision. Therefore, it is common to use auxiliary data via covariates and produce estimators that combine them with direct data. Nevertheless, it is not uncommon for covariates to be measured with error, leading to inconsistent estimators. Area-level models accounting for measurement error (ME) in covariates have been proposed, and they usually assume that the errors are an i.i.d. Gaussian model. However, there might be situations in which this assumption is violated especially when covariates present severe outlying values that cannot be cached by the Gaussian distribution. To overcome this problem, we propose to model the ME through sub-Gaussian (alpha)-stable (SG(alpha)S) distribution, a flexible distribution that accommodates different types of outlying observations and also Gaussian data as a special case when (alpha =2). The SG(alpha)S distribution is a generalization of the Gaussian distribution that allows for skewness and heavy tails by adding an extra parameter, (alpha in (0,2]), to control tail behaviour. The model parameters are estimated in a fully Bayesian framework. The performance of the proposal is illustrated by applying to real data and some simulation studies.

Orthonormality constraints are common in reduced rank models. They imply that matrix-variate parameters are given as orthonormal column vectors. However, these orthonormality restrictions do not provide identification for all parameters. For this setup, we show how the remaining identification issue can be handled in a Bayesian analysis via post-processing the sampling output according to an appropriately specified loss function. This extends the possibilities for Bayesian inference in reduced rank regression models with a part of the parameter space restricted to the Stiefel manifold. Besides inference, we also discuss model selection in terms of posterior predictive assessment. We illustrate the proposed approach with a simulation study and an empirical application.

We use Bayesian model selection paradigms, such as group least absolute shrinkage and selection operator priors, to facilitate generalized additive model selection. Our approach allows for the effects of continuous predictors to be categorized as either zero, linear or non-linear. Employment of carefully tailored auxiliary variables results in Gibbsian Markov chain Monte Carlo schemes for practical implementation of the approach. In addition, mean field variational algorithms with closed form updates are obtained. Whilst not as accurate, this fast variational option enhances scalability to very large data sets. A package in the R language aids use in practice.

Sufficient dimension reduction is a widely used tool to extract core information hidden in high-dimensional data for classifying, clustering, and predicting response variables. Various dimension reduction methods and their applications have been introduced in the past decades. Data analysis using sufficient dimension reduction involves two steps: dimension reduction and model estimation. However, when we implement the two-step modeling process, we consider the estimated sufficient predictor as a true predictor variable and proceed to the model development step, which includes statistical inference such as estimating confidence intervals and performing hypothesis tests. However, the outcome obtained using this method is by no means complete because it contains errors only from the model estimation step. Therefore, post dimension reduction inference is an important topic because it is essential to consider errors from sufficient dimension reduction. In this paper, we review the fundamentals of sufficient dimension reduction methods. Then, we introduce an intuitive and heuristic approach for the recently developed post dimension reduction statistical inference.

The past few decades have seen considerable interest in modeling time series of counts, with applications in many domains. Classical and Bayesian modeling have primarily focused on conditional Poisson sampling distributions at each time. There is very little research on modeling time series involving Zero-Modified (i.e., Zero Deflated or Inflated) distributions. This paper aims to fill this gap and develop models for count time series involving Zero-Modified distributions, which belong to the Power Series family and are suitable for time series exhibiting both zero-inflation and zero-deflation. A full Bayesian approach via the Hamiltonian Monte Carlo (HMC) technique enables accurate modeling and inference. The paper illustrates our approach using time series on the number of deaths from the influenza virus in the city of São Paulo, Brazil.

Environmental, social and governance (ESG) criteria are increasingly integrated into investment process to contribute to overcoming global sustainability challenges. Focusing on the reaction to turmoil periods, this work analyses returns and volatility of several ESG indices and makes a comparison with their traditional counterparts from 2016 to 2022. These indices comprise the following markets: Global, the US, Europe and emerging markets. Firstly, the two-component mixture of generalized normal distribution was exploited to objectively detect financial market turmoil periods with the Naïve Bayes’ classifier. Secondly, the EGARCH-in-mean model with exogenous dummy variables was applied to capture the turmoil period impact. Results show that returns and volatility are both affected by turmoil periods. The return–risk performance differs by index type and market: the European ESG index is less volatile than its traditional market benchmark, while in the other markets, the estimated volatility is approximately the same. Moreover, ESG and non-ESG indices differ in terms of turmoil periods impact, risk premium and leverage effect.