GRETA provides research and development activities and data analysis for reporting in various areas of econometrics. In addition to traditional analysis and modelling techniques, GRETA offers support in the context of the latest Data Science and Machine Learning techniques and their combination with traditional econometric analyses.

By way of example, GRETA researchers are active on the following topics:

• Financial data analysis (monthly, weekly, daily and intra-daily frequency), and development of multivariate models (analysis of variances, correlation models, exchange rate analysis)
• Aggregation and estimation of statistics for aggregates of countries (Euro Area, European Union, etc.)
• Time disaggregation techniques for the estimation of macroeconomic variables at frequencies not currently available
• Analysis of the economic and financial cycle, identification of turning points, development of models and dating with parametric and non-parametric methods, development of multivariate VAR and switching-regime models
• Historical reconstruction of statistics based on partial information (back-recalculation, constrained interpolation)
• Interpretation and use of surveys for forecasting variables and analyzing the economic cycle
• Production of predictive models (medium and long term "flash" forecast, the combination of forecasts)
• Analysis of official statistics revisions; development and construction of composite indices.

Forecast analysis:

• Scoring rules applications for the analysis of forecasts generated by models or by surveys.
• Combination and calibration of point or probabilistic forecasts.

Data size reduction:

• Linear techniques such as Principal Component Analysis, Multidimensional Scaling and Random Projection.
• Nonlinear techniques such as Isomaps, Maximum Variance Unfolding, Local Linear Embedding; their applications to large econometric models (eg large VARs, large panel data) with a view to forecasting.

Data analysis with complex structure:

• Statistical analysis of networks (relational data) and temporal flows of data networks; extraction of networks through graphical models for the analysis of causal relationships between time series.
• Reduced rank approximations using tensor decomposition, such as Tucker and PARAFAC decomposition, for multilinear time series (for example data trade networks, financial flow networks).
• Space-time data analysis.
• Text mining and sentiment analysis.

Bayesian methods

• Classical Bayesian analysis of econometric models for time series (i.e. VAR, Markov-switching).
• Nonparametric Bayesian analysis (i.e., Dirichlet Process Prior, Pitman-Yor Prior).
• Sequential estimation and prediction with Bayesian update (Dynamic Linear Models).

Nonparametric methods

• Random Trees and Random Forests for prediction and classification and their combination with econometric models (i.e. Linear Models, GLM).
• Nonparametric Bayesian methods for the classification, hierarchical classification and robustness of econometric models.