Specialized Session 6: European Statistics and Data Science Groups

The integrated nested Laplace approximation (INLA) is a method for Bayesian inference that can estimate the posterior marginals of models parameters and latent effects in a short time. INLA focuses on models that can be expressed as latent Gaussian Markov random fields (GMRFs) and it is implemented in the R-INLA package. INLA has been quite successful when fitting highly structured models with spatial and temporal random effects.

However, there are certain models that INLA cannot fit. These include double hierarchical models and mixture models, to mention a few. Some authors have developed inferential approaches that combine INLA with more general approaches to be able to fit larger classes of models. To mention a few, INLA has been successfully enmbeded into Markov chain Monte Carlo methods and importance sampling.

Finally, in addition to being able to fit larger models, these approaches allow making Bayesian inference about other questions of interest such as computing the posterior probabilities of models. This is important for model selection as well as for Bayesian model averaging (BMA).

With the ongoing advancements in the digital transformation, data privacy is likely to become a critical concern for the Member States worldwide. The rapid evolution of digital technologies, and in particular Ai, coupled with increasing collection, processing, and archiving data will necessitate developing and implementing new privacy frameworks to protect individual rights and to maintain public trust. UNESCO’s recent publications “Open Data for AI: What Now?” and the brief on “Data Invisibility” have barely scratched the issue and much remains to be done. The forthcoming policy brief on Open Government Data outlines related challenges and 9ffers a roadmap for successful implementation. Recent need assessment surveys carried out by UNESCO, identified significant ambiguity in understanding the context for the data, its context for openness thereof and the need to upscale efforts to build awareness and capacities at all levels.

This presentation explores the diverse applications of data science, artificial intelligence, and big data in the context of a central bank. I will begin with an overview of the data science initiatives developed at the Bank of Italy, highlighting the variety of techniques and unconventional data sources used to enhance economic analysis. I will then discuss a few concrete examples of projects undertaken in my department. The first example focuses on the integration of nontraditional data sources to complement official statistics, helping to bridge information gaps and provide more timely economic indicators. The second example explores the challenges and benefits of incorporating administrative data, which, while rich in detail, requires sophisticated processing and integration techniques to unlock its full potential. Finally, I will present an experimental application involving satellite data, showcasing how geospatial information might be leveraged to gain novel insights into economic activity.