Modern macroeconomic identification increasingly sharpens inference by combining multiple economically motivated restrictions. However, more restrictions are not a free lunch: conclusions based on them depend crucially on their credibility. This paper develops a robust Bayesian framework for modeling partial credibility in set-identified models. Because the likelihood does not update the unidentified structural component, qualitative prior information can be imposed by reweighting baseline structural draws. Posterior bounds are computed by linear programs restricted by prior classes. The framework links marginal-prior correction and full ambiguity robust Bayes inference, while allowing intermediate prior classes based on tail, moment, sign, or shape restrictions. The method shows that in oil-market SVARs, demand dominance is robust, while the near-irrelevance of supply shocks requires stringent elasticity-prior discipline.

Presented at:  Econometric Society European Winter Meeting  (December 2025), Macro Lunch Università Cattolica (September 2025), Sailing the Macro (Sicily, September 2025, regular session), Junior Milan Time Series Workshop (March 2025, poster), PRIN Workshop (June 2025, Poster) , Örebro Workshop on Macro- and Financial Econometrics  (November 2025), 16th European Seminar on Bayesian Econometrics (August 2026, oral session)

Researchers often confront regressions where key covariates are missing or only interval-observed. Common fixes, e.g imputation or auxiliary modeling assumptions, resolve ambiguity at the expense of credibility. Instead, we derive the sharp identified set, the smallest parameter set consistent with both the data and the maintained regression model. A computationally tractable (Hausdorff consistent) estimator for this set is provided with associated asymptotic (uniformly) valid confidence regions for the true parameter. Technically, this is achieved by extending a previous random set framework for finitely many moments to infinite-dimensional Polish (separable and complete) spaces.

Presented at: 31st International Panel Data Conference, University of Exeter (July 2026), ESIF Economics and AI+ML Meeting, Cornell University (June 2026), IAAE Lisbon (June 2026), Munich Econometrics Workshop (Poster, July 2026)

paper

Power priors modify the historical likelihood by raising it to the power of a discount factor, allowing researchers to downweight past or external data. This generalizes the standard Bayesian updating formula and provides a flexible framework for incorporating external information into macroeconomic models. We extend the theoretical foundations of power priors to vector autoregressions (VARs) and show that they are equivalent to simple hierarchical structures—while preserving conjugacy and compatibility with modern structural identification methods. Applications include cross-country borrowing to improve forecast accuracy, sharper estimation of long-run Phillips curves, and a novel sensitivity analysis tool based on observation-specific discounting.

Presented at: IAAE 2025 Turin (June 2025), 16th RCEA Bayesian Econometrics Workshop (May 2026), Nordic Econometric Meeting Helsinki (June 2026), Vienna Workshop on High-Dimensional Time Series in Macroeconomics and Finance (May 2026)

paper

This article introduces modelconfset, a Stata command for implementing the model confidence set procedure of Hansen, Lunde, and Nason (2011). The procedure starts from a user-specified set of competing models and returns a subset that contains the best model, or models, with a chosen level of confidence, where best is defined by a user-supplied loss function. Unlike procedures that select a single winning model, the model confidence set explicitly recognizes that the data may not be informative enough to separate closely performing alternatives. The command works directly with loss variables stored in a Stata dataset, supports the commonly used test statistics, implements bootstrap inference, reports the sequential elimination path, and stores the resulting superior set in returned results. The command is useful for forecast comparison, model selection, and any setting in which several competing procedures can be evaluated by a common loss criterion. We illustrate the syntax and output using reproducible examples and discuss practical choices concerning the confidence level, loss functions, and bootstrap settings.

paper

This paper shows the merit of one-sided conformal prediction methods for quantile forecasting. Our approach converts point forecasts of any kind of model into quantile predictions. Through simulations and an application in nowcasting US GDP tail risk using numerous high-frequency regressors, we demonstrate that conformal quantile forecasts are accurately calibrated for high dimensional problems, unlike bootstrap quantiles. Additionally, our algorithm produces comparable results to a leading nowcasting approach in a fraction of its computing time. In a second application we use our proposal to forecast percentiles of a distribution as quantile crossing is ruled out by construction.

paper | video

Presented at: 3rd International Econometrics PhD Conference (November 2023), VTSS Virtual Workshop for Junior Researchers in Time Series (March 2024)