30/10/2026 - 11:00 - LIDAM D.251
(Bocconi University)
will give a presentation on
Abstract:
We contribute to the literature on large multidimensional data by proposing a Matrix Mixed-Frequency Three-Pass Regression Filter (Matrix MF–TPRF) for nowcasting with high-dimensional, ragged-edge information sets. The method extends proxy-based factor extraction to matrix-valued predictors, preserving cross-sectional structure while accommodating mixed frequencies and missing observations. After discussing the procedure and its properties, Monte Carlo evidence shows that exploiting the matrix structure improves predictive-signal recovery and nowcasting performance relative to vector approaches, especially with persistent non-predictive components. A pseudo real-time application to GDP growth in eight major euro-area countries indicates sizeable gains at early vintages and in the post-Covid period, when cross-country co-movement is stronger. The results also highlight the importance of parsimonious information sets for stable real-time performance.
More information about the speaker