High-Dimensional Inference for Panel and Network Data

Informations projet

PANEDA

N° de convention de subvention: 819086

DOI

10.3030/819086

Projet clôturé

Date de signature de la CE 18 Decembre 2018

Date de début 1 Août 2019

Date de fin 31 Juillet 2025

Financé au titre de

• EXCELLENT SCIENCE - European Research Council (ERC)

Coût total

€ 1 478 831,00

Contribution de l’UE

€ 1 478 831,00

Coordonné par

THE CHANCELLOR, MASTERS AND SCHOLARS OF THE UNIVERSITY OF OXFORD
United Kingdom

Periodic Reporting for period 4 - PANEDA (High-Dimensional Inference for Panel and Network Data)

Période du rapport: 2024-01-01 au 2025-07-31

Résumé du contexte et des objectifs généraux du projet

Classic data structures that economists and statisticians have traditionally worked with are cross-sectional data, time series, and panel data. The main statistical tools for these types of data were developed in the 20th century, often in response to new kinds of information becoming available. For example, stock market data, national accounts, household surveys, and longitudinal studies of families each gave rise to new methods.

Over the past two decades the amount and variety of data in the social sciences have grown very quickly. Much of this growth is due to administrative records, digital traces, and new measurement technologies, together with deliberate efforts to collect larger and richer datasets. Many of these modern datasets cannot be described simply as cross sections or panels. Instead, they often have a network structure, linking individuals, firms, or countries to one another. In such settings the precision of statistical inference depends on the shape of the network itself.

The objective of the project was to build a robust statistical framework for these modern forms of data. The project aimed to establish a mathematical way of representing the structure of the data, to connect that structure to the reliability of statistical inference, and to design new methods that correct bias and produce more reliable standard errors.

Conclusion: The project has delivered on these aims. It has produced a set of new identification strategies, bias correction methods, and robust inference tools that make the analysis of complex panel and network datasets more trustworthy and useful for applied researchers.

Travail effectué depuis le début du projet jusqu’à la fin de la période considérée dans le rapport et principaux résultats atteints jusqu’à présent

The project produced substantial advances in both theory and applications. One important breakthrough was the discovery of new moment conditions in dynamic panel logit models with fixed effects. These conditions allow researchers to estimate parameters in settings where it was previously thought to be impossible. We also developed bias correction methods for widely used nonlinear models with multiple fixed effects, including three-way gravity models in international trade. Further work provided new approaches to estimation in models with interactive fixed effects.

The research has resulted in a strong publication record. Several papers have been published or accepted in leading journals such as The Review of Economic Studies, The Journal of Political Economy, Quantitative Economics, and the Journal of Econometrics. Other papers are under review at top journals. A Stata software package that implements our bias corrections for gravity models has been released, and further code to support other methods is being prepared.

Dissemination has been extensive. Project results were presented in numerous invited talks and keynote lectures, and four major international workshops were held in Oxford in 2022, 2023, 2024, and 2025. These events gathered leading experts and young researchers from around the world. The project also supported postdocs, PhD students, and research assistants, many of whom have now moved on to faculty or senior professional positions.

Conclusion: The project has made its results widely available through publications, software, workshops, and presentations. Its outputs are already in use by applied researchers and will continue to influence empirical practice in economics and related fields.

Progrès au-delà de l’état des connaissances et impact potentiel prévu (y compris l’impact socio-économique et les conséquences sociétales plus larges du projet jusqu’à présent)

The project moved beyond the state of the art in several important ways. It showed that dynamic binary choice models with fixed effects admit useful moment conditions, overturning the long-standing view that no such conditions exist. This result provides new possibilities for estimation and inference in a class of models that are central in applied work.

It also provided practical solutions for bias in models with many fixed effects. For example, in three-way gravity models we derived explicit corrections that allow more accurate estimates of trade policy effects. In models with interactive fixed effects we designed new estimators that improve reliability when standard methods perform poorly. These contributions offer applied researchers concrete tools to improve the credibility of their findings.

Final outcome: The project has produced a toolkit of new methods for complex panel and network data. These methods advance econometric theory and also have clear practical value for applications in labor economics, international trade, and the study of networks. Through publications, software, workshops, and training, the results are now being taken up by the wider research community and will have lasting impact.

logo PANEDA