European Central Bank working paper finds machine learning can detect euro area M1 turning points with a three-month lag and identify key drivers of M3 shifts

The European Central Bank (ECB) published a Working Paper presenting a machine learning approach to identifying turning points in euro area monetary aggregates M1 and M3 in near real time. Using Learning Vector Quantisation (LVQ), the paper finds that M1 turning points can be detected with a three-month delay, compared with the at least six-month confirmation lag required by the Bry–Boschan dating algorithm, while a distinction-sensitive extension (DSLVQ) provides interpretability for M3 by weighting the importance of different sources of broad money growth. The paper notes that the views expressed are those of the authors and do not necessarily reflect those of the ECB. Performance is benchmarked against the Bry–Boschan algorithm and standard classifiers including Gaussian Naive Bayes and logit regression, using balanced accuracy and a block bootstrap design. For M3, the analysis decomposes broad money into counterparts including lending to households, lending to firms, bank credit to government, net external monetary flows, banks’ long-term liabilities and Eurosystem net purchases, with DSLVQ weights indicating that lending to households and firms, and Eurosystem asset purchases when present, are the most influential drivers of M3 turning points. The results are reported as robust across parameter choices, bootstrap designs, alternative evaluation metrics and comparator models.