Speakers and abstracts
Alan Agresti - University of Florida, USA
Title: A historical overview of textbook presentations of statistical science
Friday, January 27, 2023 - 14:50 – 15:20
We discuss the evolution in the presentation of statistical science in textbooks during the first half of the twentieth century as the field became better defined by advances due to R. A. Fisher and Jerzy Neyman. An early influential book with 14 editions was authored by G. Udny Yule. Method books authored by Fisher and George Snedecor showed scientists how to implement Fisher's advances. Later books from the World War 2 era authored by Maurice Kendall, Samuel Wilks, and Harald Cramer had stronger emphasis on the theoretical foundations. The Bayesian approach emerged somewhat later in textbooks, influenced strongly by books by Harold Jeffreys and Leonard Savage. We conclude by discussing the future of textbooks on the foundations of statistical science in the emerging, ever-broader, era of data science.
Tomoyuki Amano - The University of Electro-Communications, Tokyo, Japan
Title: Asymptotics of the conditional least squares estimator and the estimating function estimator for nonlinear time series models
Saturday, January 28, 2023 - 11:10 – 11:50
There have been proposed many nonlinear time series models in order to represent behaviours of data and many researchers have investigated these models. One of the most fundamental estimators for nonlinear time series models is the conditional least squares estimator (CL estimator). However, Amano and Taniguchi (2008) showed CL estimator is not asymptotically optimal in general for ARCH model. On the other hand, Chandra and Taniguchi (2001) constructed the optimal estimating function estimator (G estimator) for ARCH model based on Godambe’s optimal estimating function and showed G estimator is better than CL estimator in the sense of the sample mean squared error by simulation. In this talk we apply CL and G estimators to famous nonlinear time series models and show that the G estimator is better than CL estimator in the sense of efficiency. Furthermore, CL and G estimators are applied to vector valued nonlinear time series models.
Chiara Amorino - Université du Luxembourg, Luxembourg
Title: On estimating interacting particle systems from discrete data
Friday, January 27, 2023 - 10:10 – 11:50
We consider the problem of joint parameter estimation for drift and volatility coefficients of a stochastic McKean-Vlasov equation and for the associated system of interacting particles. The analysis is provided in a general framework, as both coefficients depend on the solution of the process and on the law of the solution itself. Starting from discrete observations of the interacting particle system over a fixed interval [0, T], we propose a contrast function based on a pseudo likelihood approach. We show that the associated estimator is consistent when the discretization step ($\Delta_n$) goes to 0 and the number of particles N goes to $\infty$, and asymptotically normal when additionally the condition $\Delta_n N \rightarrow 0$ holds. We will also compare our results (and our condition on the decay of the discretization step) with the results known for classical SDEs.
The talk is based on a joint work with A. Heidari, V. Pilipauskaite and M. Podolskij.
Raffale Argiento - University of Bergamo, Italy
Title: Model-based clustering for categorical data via Hamming distance
Thursday, January 26, 2023 - 11.30 – 12:10
In this work a model-based approach for clustering categorical data with no natural ordering is introduced. The proposed method exploits the Hamming distance to define a family of probability mass functions to model categorical data. The elements of this family are considered as kernels of a finite mixture model with unknown number of components. Fully Bayesian inference is provided using a sampling strategy based on a trans-dimensional blocked Gibbs-sampler, facilitating the computation with respect to the customary reversible-jump algorithm. Model performances are assessed via a simulation study, showing improvements in clustering recovery over existing approaches. Finally, our method is illustrated with an application to reference datasets. (Joint work with Lucia Paci and Edoardo Filippi-Mazzola).
Francesco Bartolucci - University of Perugia, Italy
Title: A causal latent transition model with multivariate outcomes and unobserved heterogeneity: Application to human capital development
Friday, January 27, 2023 - 15:20 – 15:50
In order to evaluate the effect of a policy or treatment with pre- and post-treatment outcomes, we propose an approach based on a transition model, which may be applied with multivariate outcomes and accounts for unobserved heterogeneity. This model is based on potential versions of discrete latent variables representing the individual characteristic of interest and may be cast in the hidden (latent) Markov literature for panel data. Therefore, it can be estimated by maximum likelihood in a relatively simple way. The approach extends the Difference-in-Difference method as it is possible to deal with multivariate outcomes. Moreover, causal effects may be expressed with respect to transition probabilities. The proposal is validated through a simulation study, and it is applied to evaluate educational programs administered to pupils in the 6th and 7th grades during their middle school period. These programs are carried out in an Italian region to improve non-cognitive skills. We study if they impact also on students’ cognitive skills in Italian and Mathematics in the 8th grade, exploiting the pre-treatment test scores available in the 5th grade. The main conclusion is that the educational programs aimed to develop non-cognitive abilities help the best students to maintain their higher cognitive abilities over time.
Christian Bontemps - ENAC and Toulouse School of Economics, France
Title: Optimal moment-based tests for distributional assumptions
Thursday, January 26, 2023 - 14:20 – 15:00
In this paper we aim at testing a distribution against a specific one, when observations are (serially) dependent according to a general unknown scheme. To allow for dependence, we consider tests based on moments (or estimating functions) which may involve estimates of unknown parameters (hence parameter uncertainty). Since this setup problem falls outside the conditions under which the Neyman-Pearson (NP) Lemma applies, we consider the problem of building optimal moment-based tests. Two categories of procedures are studied: (1) tests based on the assumption that empirical moments follow a normal distribution (at least asymptotically) along a (long-run) which can be consistently estimated; (2) a recasting of the NP test as a moment-based test. To deal with the effect of nuisance parameter estimation, we propose to use an appropriate orthogonalization of the estimating function used, similar to the one proposed for score tests by Neyman (1959). For NP moments, this yields $C(\alpha )$\emph{-LR tests}, which can be viewed as extensions of the standard NP tests, meant to allow for serial dependence of unknown form and for the presence of nuisance parameters. The size and power properties of the proposed tests are assessed in a Monte Carlo simulation. The results demonstrate excellent power properties. In particular, in the i.i.d case, the proposed procedures have power remarkably close to the power of the NP test, and vastly superior to the one usual goodness-of-fit tests. (joint work with Jean-Marie Dufour and Nour Meddahi).
Michela Cameletti - University of Bergamo, Italy
Title: A spatiotemporal analysis of NO2 concentrations during the Italian 2020 COVID- 19 lockdown
Thursday, January 26, 2023 - 15:00 – 15:40
The lockdown measures taken worldwide in 2020 to reduce the spread of the SARS- CoV-2 virus can be envisioned as a policy intervention with an indirect effect on air quality. In this talk a statistical spatiotemporal model will be presented as a tool for intervention analysis. In particular, it is able to take into account the effect of weather and other confounding factors, as well as the spatial and temporal correlation existing in the data. The model was applied to the 2019/2020 relative change in nitrogen dioxide (NO2) concentrations in the north of Italy, for the period of March and April during which the lockdown measure was in force. It was found that during March and April 2020 most of the studied area is characterized by negative relative changes (median values around −25%), with the exception of the first week of March and the fourth week of April (median values around 5%). As these changes cannot be attributed to a weather effect, it is likely that they are a byproduct of the lockdown measures.
Serguei Dachian - Université de Lille, France
Title: On Smooth Change-Point Location Estimation for Poisson Processes and Skorokhod Topologies
Friday, January 27, 2023 - 11:50 – 12:30
We consider the problem of estimation of the location of what we call smooth change- point from $n$ independent observations of an inhomogeneous Poisson process. The smooth change-point is a transition of the intensity function of the process from one level to another which happens smoothly, but over such a small interval, that its length $\delta_n$ is considered to be decreasing to $0$ as $n\to+\infty$. We study the maximum likelihood estimator (MLE) and the Bayesian estimators (BEs), and show that there is a ``phase transition'' in the asymptotic behavior of the estimators depending on the rate at which $\delta_n$ goes to $0$. More precisely, if $\delta_n$ goes to zero slower than the ``critical'' rate $1/n$ (slow case), the behavior resembles that of the smooth case, and if $\delta_n$ goes to zero faster than $1/n$ (fast case), the behavior is exactly the same as in the change-point case. It should be noted that these results were obtained using the likelihood ratio analysis method of Ibragimov and Khasminskii, which equally yields the convergence of polynomial moments of the considered estimators. On the other hand, for the study of the MLE, this method needs the convergence of the normalized likelihood ratio in some functional space, and up to the best of our knowledge, until now it was only applied using either the space of continuous functions equipped with the topology induced by the $\sup$ norm, or the space of càdlàg functions equipped with the usual Skorokhod topology (called $J_1$ by Skorokhod himself). However, we will see that in the fast case this convergence can not take place in neither of these topologies. So, the results concerning the MLE in the fast case were obtained by first extending the Ibragimov- Khasminskii method to use a weaker topology $M_1$ (also introduced by Skorokhod in his seminal paper of 1956, along with $J_1$ and two other topologies). (Joint work with Arij AMIRI)
Liudas Giraitis - Queen Mary University of London, UK
Title: A partially time varying regression model
Saturday, January 28, 2023 - 9:30 – 10:10
This paper explores a semiparametric version of a time varying regression, where a subset of the regressors have a fixed coefficient and the rest a time varying one. We provide an estimation method and establish associated theoretical properties of the estimates. In particular, we show that the estimator of the fixed regression coefficient preserves the parametric rate of convergence. Theoretical properties of the estimator are confirmed by Monte Carlo experiments and illustrated by an empirical example.
Anna Gottard - University of Florence, Italy
Title: Graphical models and circular variables
Friday, January 27, 2023 - 16:20 - 16:50
Circular variables, arising in several fields such as biology, medicine, geography or meteorology, are characterised by periodicity. Despite their potential, methods for analysing the dependence/independence structure of circular variables remain under-explored. We will discuss three multivariate circular distributions defined on the p-dimensional torus, investigate their key conditional independence characteristics and introduce suitable classes of graphical models. The usefulness of the proposal is shown by modelling the conditional independence among the dihedral angles characterising the three-dimensional structure of some proteins. This is a joint work with Agnese Panzera.
Maria Iannario - University of Naples Federico II, Italy
Title: Robust logistic regression for multinomial responses
Thursday, January 26, 2023 - 12:10 – 12:50
Baseline-category logit models and cumulative, adjacent-categories and continuation ratio models are applied in many fields to analyze unordered or ordered responses with respect to subjects’ profiles. They are typically fitted by maximum likelihood estimators, which unfortunately are sensitive to anomalous data. The contribution proposes robust M-type estimators based on the properties of the logistic link function and on a weighted likelihood approach. The M-estimators can be easily implemented numerically, provide reliable inference when data are contaminated and lead to an accurate model specification. Inference based on the M-estimators is illustrated in some case studies. (joint work with Anna Clara Monti)
Yury Kutoyants - University of Le Mans, France
Title: Hidden Markov processes and adaptive filtration
Friday, January 27, 2023 - 9:30 – 10:10
We consider a linear partially observed system of SDE. The coefficients of this system depend on some finite - dimensional unknown parameter. We study the problems of this parameter estimation and the construction of adaptive Kalman-Bucy filtration equations in the asymptotic of low noise in observations only. The properties of the MLE and BE are described. The adaptive filter is constructed in two steps. First we propose a method of moments preliminary estimator using observations on a vanishing interval. Then this estimator is used for construction of One-step MLE-process. Finally the last estimator allows us to construct an adaptive filter. The question of asymptotic efficiency of the proposed adaptive filter is discussed.
Yoichi Miyata - Takasaki City University of Economics, Japan
Title: Asymptotic properties of bridge estimators in linear models under heteroscedasticity
Saturday, January 28, 2023 - 10:10 – 10:50
We consider sparse linear models with heteroscedastic error terms when the number of explanatory variables might increase as the sample size grows. For the data generated from this process, we consider estimating the true model and its non-zero regression coefficients using bridge estimators, one of penalized least squares estimators with a nonconcave penalty term. In this talk, we will present some conditions for the bridge estimators to have consistency for estimating regression coefficients, consistency for model selection, and asymptotic normality, under the assumption that the disturbance term is heteroskedastic. Furthermore, we present a consistent estimator for the standard error of the bridge estimator and propose a simple test statistic with asymptotic normality under a null hypothesis. We also compare through simulations the difference in performance between the usual standard errors of the bridge estimator for the regression coefficients and our proposed standard errors.
Ilia Negri - University of Calabria, Italy
Title: Z-process method for change point problems: some applications in diffusion processes and in linear time series
Thursday, January 26, 2023 - 14:20 – 15:10
The Z-process method was introduced as a general unified approach based on partial estimation functions to construct test statistics for a broad spectrum of statistical change point problems. The method can test for change in any of the parameters of the model simultaneously. We study the asymptotic distribution of the test statistics under the null hypothesis and under a very general alternative. Applications of the method to change point problems for diffusion processes are presented, both for ergodic models and also for some models where the Fisher information matrix is random. Moreover, the problem of testing for parameter changes in linear time series models is also considered. Some simulated studies are presented for both the models.
Hiroaki Ogata - Tokyo Metropolitan University, Japan
Title: Periodicity for circular time series
Thursday, January 26, 2023 - 16:40 – 17:20
The purpose is to detect the periodicity for circular time series data. We try to define spectral density for cylindrical and circular time series data after introducing the complex-valued process. Simulation studies and real data analysis are also provided.
Domenico Piccolo - University of Naples, Federico II, Italy
Title: Contributions on discrete mixture models for ordered data
Friday, January 27, 2023 - 16:50 – 17:20
Mainly motivated by the psychological mechanism which generates discrete choices, a general framework for modelling ordinal data has been introduced since 2003. The rationale stems from the interpretation of the respondent’s final choice as a weighted combination of a personal feeling and some intrinsic uncertainty. A mixture of these components, specified by discrete random variables, has been defined CUB model possibly with the introduction of subjects’ and/or objects’ covariates. Several generalizations of this approach have been successfully explored, including the inclusion of inflated modality of response and/or a possible overdispersion. Easiness of interpretation, parsimony of parameters and immediate visualizations of estimated models are special features of this approach. Currently, open source software is available in R, STATA and GRETL languages. Noticeably, a significant and innovative contribution has been developed by Colombi and Giordano whose multivariate generalization of CUB class of models is a methodological advance that has proved very useful for ordinal data analysis.
David Preinerstorfer - University of St. Gallen, Switzerland
Title: Functional sequential treatment allocation with covariates
Thursday, January 26, 2023 - 10:20 – 11:00
We consider a multi-armed bandit problem with covariates. Given a realization of the covariate vector, instead of targeting the treatment with highest conditional expectation, the decision maker targets the treatment which maximizes a general functional of the conditional potential outcome distribution, e.g., a conditional quantile, trimmed mean, or a socio-economic functional such as an inequality, welfare or poverty measure. We develop expected regret lower bounds for this problem, and construct a near minimax optimal assignment policy. (Joint work with Bezirgen Veliyev and Anders Bredahl Kock).
Tommaso Proietti - University of Roma Tor Vergata, Italy
Title: Another look at dependence: the most predictable aspects of time series
Saturday, January 28, 2023 - 11:50 – 12:30
Serial dependence and predictability are two sides of the same coin. The literature has considered alternative measures of these two fundamental concepts. In this paper we aim at distilling the most predictable aspect of a univariate time series, i.e., the one for which predictability is optimized. Our target measure is the mutual information between past and future of a random process, a broad measure of predictability that takes into account all future forecast horizons, rather than focusing on the one-step-ahead prediction error mean square error. The first most predictable aspect is defined as the measurable transformation of the series for which the mutual information between past and future is a maximum. The proposed transformation arises from the linear combination of a set of basis functions localized at the quantiles of the unconditional distribution of the process. The mutual information is estimated as a function of the sample partial autocorrelations, by a semiparametric method which estimates an infinite sum by a regularized finite sum. The second most predictable aspect can also be defined, subject to suitable orthogonality restrictions. We also investigate the use of the most predictable aspect for testing the null of no predictability.
Tamas Rudas - Eötvös Loránd University (ELTE), Budapest, Hungary
Title: Analysis of data from sequential experiments with an unspecified number of observations
Friday, January 27, 2023 - 17:20 – 17:50
When in a multistage experiment, later stages depend on the outcomes of the earlier stages, the total number of observations is not known in advance. In the simplest such structures, the same experiment is or is not repeated at every stage. Examples include vaccination and then possible revaccination with the same vaccine, depending on the response to the first shot. The talk deals with the analysis of statistical models for such data. In maximum likelihood estimates (MLE) based on the model of constant response probability, the expected number of observations (ENO) plays a role similar to that of the actual number of observations (ANO) in other types of models. The sufficient statistics are reproduced by the MLE only in their respective proportions to the ENO, and in the variances, the ENO appears instead of the ANO. The relevant adjustments represent the effect of the observations not being independent from each other, thus the ENU may be interpreted similarly to the effective number of observations. These models are relational models without the overall effect. When at different stages of the experiment, different treatments are applied (e.g., different vaccines are administered), one obtains more general models, but the adjustment of the MLE pertains in these cases, too. (Joint work with Anna Klimova).
Takayuki Shiohama - Nanzan University, Department of Data Science, Japan
Title: Markov models for cylindrical time series data
Thursday, January 26, 2023 - 17:20 – 18:00
Recently, statistical analyses on the geometric manifold have received special attention in various fields of data science. These geometric manifolds include a circle, a cylinder, and a sphere. In response to the request of analyzing multiple regression and time series data on those manifolds, a multi-dimensional extension of statistical models is required. Here, we use a statistical model on a hyper cylinder to model a Markov process on that manifold. Several Monte Carlo simulations are performed to investigate a finite sample performance of a maximum likelihood estimator. In addition, real data analyses are performed to illustrate the applicability of the proposed models.
Hiroko Solvang - Institute of Marine Research, Bergen, Norway
Title: Recent temporal and geographical variation in blubber thickness of common minke whales (Balaenoptera acutorostrata acutorostrata) in the northeast Atlantic
Friday, January 27, 2023 - 12:30 – 13:10
The common minke whale (Balaenoptera acutorostrata acutorostrata) is a migratory species, and the summer period is generally characterized by intensive feeding and consequently seasonal fattening at high latitudes. The fat deposited is stored as energy reserves for overwintering at lower latitudes where feeding is supposed to be greatly reduced. It is therefore expected that their body condition on the summer feeding grounds will reflect foraging success during their most intensive feeding period and thus indicate how well the high latitude ecosystems can support the populations. During the commercial catch operations on feeding grounds in Norwegian waters, body condition data (blubber thickness and girth) have been collected from 13 937 common minke whales caught during the period 1993-2020. To investigate associations between body condition and area usage in minke whales, we applied three statistical approaches: varying coefficients (VC) model, canonical correlation procedure in VC estimation, and spatiotemporal effect estimation by the so-called fused lasso. The analyses revealed a significant negative trend in blubber thickness from 1993 until 2015. After 2015, the trend was reversed, and blubber thickness values increased significantly. It has previously been suggested that there may be a link between the decreased minke whale blubber thickness and the abundance of the Northeast Arctic cod (Gadus morhua) stock which increased to a record high level between 2006 and 2013. Recruitment to the cod stock in more recent years has been low with a subsequent and continuous decrease in the total stock after 2013 to a current level which is presumably approximately 60% of the 2013 level. Interestingly, the observed common minke whale body condition was at its lowest in 2015, after which it has increased. This may support a connection between cod abundance and common minke whale body condition.
This study was conducted by a joint work with Institute of Marine Research (Haug, T. and Øien, N) and Hiroshima University (Yamamura, M., Yanagihara, H., and Ohishi, M).
Gabriele Torri - University of Bergamo, Italy
Title: Risk measures for sustainable investing
Thursday, January 26, 2023 - 15:40 – 16:20
The growing interest for sustainable investing calls for an axiomatic approach to characterize risk and reward measures for investors that do not focus uniquely on financial returns, but also on environmental and social sustainability. The measurement of the sustainability of a company is typically done using ESG scores (Environmental, Social, and Governance) estimated by specialised rating agencies. We propose definitions for ESG-coherent risk and reward measures, as well as ESG risk- reward ratios. Our approach considers functions of bivariate random variables applied to the financial returns and ESG scores (a proxy for sustainability). We provide examples of such functions, and we describe approaches to extend traditional univariate risk and reward measures to the ESG case. We then show an empirical example in which we compute ESG-risk measures and ESG-risk-reward ratios, and we use them to rank stocks from the Dow Jones index according to investors with preferences to sustainable portfolios.
The talk is based on a joint work with R. Giacometti, D. Dentcheva, S. T. Rachev, and W. B. Lindquist
Masanobu Taniguchi - Waseda University, Tokyo, Japan
Title: In my statistical life
Thursday, January 26, 2023 - 9:40 – 10:20
This talk surveys my life research. The following topics will be delivered. (1) Introduction of spectral divergence and discussion on efficiency and robustness. (2) Development in high-order asymptotic theory of time series analysis. Beyond the simultaneous equation analysis. (3) Statistical analysis of “curved” stochastic models. (4) Foundation of time series discriminant analysis (5) Statistical theory based on integral functional of nonparametric spectral estimators. Semiparametric estimation for spectra. Introduction of high-order asymptotic theory for semiparametric time series estimators. (6) LAN based asymptotic theory for time series, including long memory ones. (7) Systematic approach for portmanteau tests. (8) Empirical likelihood approach for time series. (9) Non-regular estimation for time series, and Bartlett adjustment for nonstandard settings. (10) Asymptotic theory of shrinkage estimation for time series. (11) Asymptotic theory for portfolio estimation. (12) New look at circular distributions in view of high-order spectral distribution of stationary processes. (13) Analysis of variance for time series.
Frank Van Der Meulen - Vrije Universiteit Amsterdam, The Netherlands
Title: Likelihood representations for discretely observed stochastic processes
Friday, January 27, 2023 - 11:10 - 11:50
Consider parameter inference for a continuous-time stochastic processes that is only observed discretely in time. In this setting, the likelihood is typically intractable. From a computational perspective, likelihood based inference is usually based on a data- augmentation approach, where the latent paths get imputed. I will discuss a structured way for this imputation which leads to closed form expressions for transition densities. The expressions derived contain a term of the form 𝔼Ψ(𝑋◦), where 𝑋◦ is a tractable process. Such results can be used subsequently in MCMC, MCEM or variation inference, for example. I will include some numerical results. This concerns joint work with Moritz Schauer (Chalmers University of Technology - University of Gothenburg).