Project

Back to overview

Accurate and reliable asymptotics for pooled cross-sectional and time series data

English title Accurate and reliable asymptotics for pooled cross-sectional and time series data
Applicant La Vecchia Davide Antonio
Number 169559
Funding scheme Project funding (Div. I-III)
Research institution Geneva Finance Research Institute Geneva School of Economics and Management University of Geneva
Institution of higher education University of Geneva - GE
Main discipline Economics
Start/End 01.08.2017 - 31.01.2021
Approved amount 188'112.00
Show all

Keywords (6)

Dynamic semiparametric factor models; Spatial statistics; Robust M-estimation; Asymptotic theory; Functional magnetic resonance imaging; Yield curve

Lay Summary (Italian)

Lead
Questo progetto di ricerca ha come scopo principale l'elaborazione di nuove tecniche statistiche per condurre delle analisi, accurate e veritiere, di dati osservati nel tempo e in luoghi diversi. Le tecniche sviluppate sono applicabili in economia e finanza (ed esempio, per lo studio e la previsione delle dinamiche dei tassi di interesse) o in medicina (ad esempio, per l’elaborazione di diagnosi che si avvalgono di dati acquisiti tramite risonanza magnetica). In primo luogo, il progetto definisce dei nuovi strumenti matematico-statistici che permettono di individuare le carenze delle procedure già in essere nella letteratura. In secondo luogo, vengono introdotte ed elaborate delle nuove procedure inferenziali, che suppliscono alle carenze individuate.
Lay summary

La necessità di analizzare dati dalla struttura sempre piu' complessa pone una grande sfida, cui la statistica moderna deve far fronte. In particolare, il raggiungimento dell’accuratezza e dell'affidabilità nell’analisi di dati complessi è uno degli obiettivi chiave, cui la ricerca aspira.

Nonostante la rilevanza della sfida, le comuni procedure inferenziali ignorano alcuni aspetti fondamentali e, a causa di cio', conducono ad analisi che perdono proprio quelle auspicate caratteristiche di affidabilità e di accuratezza. Uno dei contributi di questo progetto è quello di rimediare al problema, definendo una nuova classe di procedure statistiche che usano in maniera piu’ efficiente le informazioni contenute nei dati. Le potenziali applicazioni delle nuove procedure sono infinite. A scopo illustrativo, il progetto si focalizza su problemi di economia, finanza e medicina.

Si consideri, ad esempio, il problema di elaborare una diagnosi sulla funzionalità del cervello umano, esaminando milioni di immagini acquisite durante una risonanza magnetica. I dati sono registrati in vari punti del cervello (dimensione spaziale) e in diversi istanti (dimensione temporale). Le procedure statistiche comunemente applicate per l’analisi di questi dati si basano su alcune ipotesi semplificatrici (come, ad esempio, l’assunzione che le osservazioni anomale si manifestino con bassa probabilità). Tali ipotesi sono utili ai fini dell’elaborazione teorica, ma sono irrealisitiche (per esempio, molte osservazioni anomale sono causate da problemi tecnici verificatisi durante la risonanza magnetica). In prima istanza, il progetto mette in evidenza la questione della violazione delle assunzioni teoriche nei dati raccolti, mostrando che un trattamento inadueguato delle osservazioni atipiche puo' condurre a diagnosi errate.  Sulla base di questa consapevolezza, il progetto definisce una nuova metodologia, che segnala le osservazioni anomale e ne riduce l'impatto sulla diagnosi.

Direct link to Lay Summary Last update: 19.06.2017

Responsible applicant and co-applicants

Employees

Publications

Publication
Estimation of a nonparametric model for bond prices from cross-section and time series information
Koo Bonsoo, La Vecchia Davide, Linton Oliver (2021), Estimation of a nonparametric model for bond prices from cross-section and time series information, in Journal of Econometrics, 220(2), 562-588.
Center-Outward R-Estimation for Semiparametric VARMA Models
Hallin M., La Vecchia D., Liu H. (2020), Center-Outward R-Estimation for Semiparametric VARMA Models, in Journal of the American Statistical Association, 1-14.
A Simple R-estimation method for semiparametric duration models
HallinMarc, La VecchiaDavide (2020), A Simple R-estimation method for semiparametric duration models, in Journal of Econometrics, 218(2), 736-749.
Saddlepoint approximations for short and long memory time series: A frequency domain approach
La Vecchia Davide, Ronchetti Elvezio (2019), Saddlepoint approximations for short and long memory time series: A frequency domain approach, in Journal of Econometrics, 213(2), 578-592.
Semiparametric segment M-estimation for locally stationary diffusions
Deléamont P -Y, La Vecchia Davide (2019), Semiparametric segment M-estimation for locally stationary diffusions, in Biometrika, 106(4), 941-956.

Collaboration

Group / person Country
Types of collaboration
University of Cambridge Great Britain and Northern Ireland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
Monash University Australia (Oceania)
- in-depth/constructive exchanges on approaches, methods or results
- Publication

Scientific events

Active participation

Title Type of contribution Title of article or contribution Date Place Persons involved
Learning tools and applied quantitative methods for decisions making Talk given at a conference Estimation and testing for multivariate time series models: a measure transportation approach 09.12.2020 University of Bolzano-Bozen, Italy La Vecchia Davide Antonio;
25th International Panel Data Conference (IPDC 2019) Talk given at a conference Saddlepoint techniques for spatial panel data models 04.07.2019 Vilnius, Lithuania Jiang Chaonan;
Congress of the Swiss Society of Economics and Statistics (SSES 2019) Talk given at a conference Saddlepoint techniques for spatial panel data models 13.06.2019 Geneva, Switzerland Jiang Chaonan;
INTERNATIONAL WORKSHOP ON HUMAN VOICE AND EARLY DEVELOPMENT Talk given at a conference The statistical analysis of time series 05.12.2018 Geneva, Switzerland La Vecchia Davide Antonio;
2nd International Conference on Econometrics and Statistics (EcoSta 2018) Talk given at a conference Saddlepoint techniques for spatial panel data models 18.06.2018 Hong Kong, China Jiang Chaonan;
10th International Conference of the ERCIM WG on Computational and Methodological Statistics (CMStatistics 2017) Talk given at a conference Saddlepoint techniques for panel data 16.10.2017 London, Great Britain and Northern Ireland Jiang Chaonan;


Abstract

This project introduces new asymptotic techniques for pooled cross-sectional and time series data, as obtained when observations are drawn from a time series of cross-sections. This kind of data arises in many scientific fields, like, e.g., Economics, Management, Finance, Statistics and Neuroscience.In the statistical inference about pooled cross-sectional and time series data, decisions are made relying on asymptotic techniques (e.g., density approximations or stochastic expansions for statistics), as obtained letting the cross-sectional size and/or the time diverge to infinity. These techniques have a twofold purpose. First, from a practical standpoint, asymptotic theory defines an approximate behavior of statistical quantities (e.g., estimators) and allows the implementation in the statistical software of the inferential procedures needed in decision-making. Second, asymptotics defines helpful tools to study theoretically the quality (e.g., the efficient use of the information contained in the available data set) of the different inferential procedures that one can apply in the decision-making.The general aim of this project is the definition and the implementation of new, accurate, and reliable (small- and large-sample) asymptotics. Accuracy and reliability will be obtained taking into account (either separately or at the same time) several inferential aspects like, e.g., small-sample issues, spatial dependence, outliers in the data set, or time-varying features of the data. Ignoring these aspects can lead to wrong inference. Intuitively, the issue is related to the fact that many of the currently available asymptotic approximations for pooled cross-sectional and time series data are derived under some restrictive and simplifying assumptions (e.g., cross-sectional size diverging to infinity, absence of spatial correlation in the cross-section, sub-Gaussian tails, absence of anomalous records), which, sometimes, are only approximately true or, more often, are completely violated in many real data applications. Thus, inference based on the currently available asymptotics appears fragile and the need for new asymptotic methods, obtained using less restrictive, more flexible and more appropriate assumptions, is gaining rapidly attention. We are planning to satisfy this need, developing new asymptotic theory and methodologies.Due to the different characteristics of the analyzed inferential problems, we are going to consider different statistical procedures. More precisely: parametric methods and higher-order asymptotic techniques will come into play when the cross-sectional size and the time dimensions are small; semi/nonparametric methods and large-sample asymptotics will be suitable when large datasets are available, yielding either reliable dimensionality reduction (robust dynamic semiparametric factors models) or providing flexible and realistic modelling of functional data (nonparametric methods). Three examples motivate our research and illustrate some possible applications of thetheory and methods that we are going to develop. The examples are related to differentscientific areas (Economics, Neuroscience, Finance) and they have a common inferential problem: asymptotic approximations of some sort are needed to answer the underlying research questions, but the extant approximations are inadequate because they rely on strict and/or wrong simplifying assumptions. Specifically: the first example is about micro-panel data for the analysis of conflict spillover among 18 Ethiopian villages; the second example defines reliable inferential procedures for functional Magnetic Resonance Imaging (fMRI) data analysis, in the presence of outliers (e.g., due to motion artifacts and/or scanner failure); the third example considers the time-varying features of the yield curve, as estimated using coupon bonds.We hope that our new theoretical and methodological developments will open the door to a large number of applications in many scientific fields: our accurate and reliable asymptotics can potentially lead to more trustable decisions. We feel that this is a desirable goal, (i) given the unavoidable use of asymptotic methods for many relevant research problems in several scientific areas (e.g., Economics, Finance and Medicine), and (ii) because of the fragility and non appropriateness of most existing asymptotics.
-