Productivity developments have become increasingly uncertain in recent years due to a series of shocks, including the COVID-19 pandemic, the energy crisis, and rising geopolitical tensions. Despite this, attempts to nowcast recent trends in productivity growth have been limited, often focusing on micro-level productivity within specific occupations or industries. To date, no effort has been made to nowcast macroeconomic measures of labour productivity growth across a broad group of countries - a gap this paper seeks to address. It presents nowcasts of labour productivity growth over a panel of 40 OECD and accession countries. A key novelty of this paper is the integration of machine learning techniques with mixed-frequency models within a panel framework, enabling the optimal utilisation of higher-frequency data. The approach combines mixed-frequency setups with a diverse range of models, including dynamic factor models, penalised regressions (LASSO, Ridge, ElasticNet) and tree-based models (Gradient Boosted Trees, Random Forests) and accounts for publication lags. Performance gains compared to an autoregressive benchmark average around 35% across the 40 countries. Machine learning models, in particular Gradient Boosted Trees, are found to outperform alternatives in most countries. A MIDAS specification with estimated weight is found to bring additional information compared to an approach with imposed weights in 30 out of 40 countries.
