GDP Nowcasting: US and Major Economies

Official GDP arrives with a lag of a month or more, so anyone allocating across economies is always working with stale numbers. This system nowcasts current-quarter real GDP growth in 10 major economies from point-in-time macro data, updating as indicators are released. Built as part of my research role at Versor Investments, the outputs directly informed global macro asset allocation decisions.

Approach

Input data comes from FRED, Eurostat, and the OECD, assembled point-in-time so the model only ever sees what was actually available on a given date. This matters: validating a nowcaster on revised data flatters its accuracy, since revised series contain information that did not exist in real time.

The core is an ensemble of three approaches: a mixed-frequency dynamic factor model, bridge regressions, and machine learning models, combined with inverse-variance weighting so that components with lower out-of-sample error carry more weight in the final estimate.

Results and caveats

Nowcasts were validated out-of-sample against standard time-series benchmarks and benchmarked against the Atlanta Fed's GDPNow for the US.

Data quality is the binding constraint outside the US. Some economies, particularly export-reliant ones, have less reliable or less frequent indicator data, and some publish key series only quarterly, arriving alongside GDP itself and therefore useless for nowcasting. In those cases the system falls back on survey data, which trades timeliness for precision.