publishes its quarterly GDP growth forecast for the current quarter and the next quarter or two in the Conjoncture in France report each quarter. This forecast is based on those for each of the components of GDP such as household consumption or industrial production. The forecasts for these components are themselves based on short-term outlook indicators such as the business climate or the industrial production index. While only one forecast is published each quarter, new indicators are released almost daily and each new piece of information is likely to change the estimate of economic growth that appears most likely at a given date. New day-to-day or “nowcasting” forecasting models make it possible to take these frequent publications of new indicators into account for the quarterly growth forecast.

These models are developed through the use of statistical learning methods (known as “machine learning”) on the one hand, and through open access in real time to hundreds of cyclical indicators (“open data”) on the other hand. For example, since 2016, the Federal Reserve Bank (Fed) in Atlanta has published an updated growth forecast every week, based on a forecasting model of this type.

This brief presents a first proposal for continuous forecasting models for quarterly variations in French growth. The data used include the short-term outlook indicators published by the Banque de France, INSEE, OECD, Markit and various ministerial statistical offices. Several models are tested, including supervised statistical learning models such as random forest model and factor models.

The first results show that the forecast can vary significantly in the course of a quarter (between +0.2% and +0.4% for Q3 2019, for example), these variations following the publication of an indicator with a sharp rise or fall. The models used tend to converge at the end of the quarter and have an error, measured by the Root Mean Squared Forecast Error (RMSFE), of around 0.20 points. The forecast error ranges from 0.28 points at the beginning of the quarter to 0.20 points at the end of the quarter. The 80 % - confidence interval for Q3 2019 growth forecast thus rose from [–0.1; 0.6] in July to [0.0; 0.5] at the end of September.