The forecast for GDP growth in France for the current quarter can be made either using a “direct” approach or on a “bottomup” approach, based on detailed forecasts for the various supply and demand items. Both methods can be implemented at any time during the quarter and draw on the same set of short-term information: on the one hand, the available quantitative indicators (usually monthly) and, on the other hand, qualitative indicators derived mainly from business tendency surveys. However, these two approaches differ in the way this information is used for forecasting: in the direct approach, GDP growth for the current quarter is forecast directly from these short-term series, whereas the bottom-up approach involves reproducing the mechanism for constructing quarterly accounts, possibly in a simplified form.

The forecasts in the Economic Outlook, generally published in the middle of the third month of the quarter, use a bottomup approach, which allows for a precise short-term diagnosis, by analysing the behaviour of the various operations in the different sectors of activity. For each forecasting period, an internal accounting model used by INSEE’s Economic Outlook Department replicates the mechanisms of quarterly accounts by aggregating detailed forecasts for each item. However, the question arises as to how effective this approach is in forecasting GDP compared to direct approaches. In order to study the relative performance of the two approaches, a new bottom-up forecasting model for French GDP growth, inspired by the Federal Reserve Bank of Atlanta’s “GDPNow”, has been developed as part of this study. First, the missing months of the indicators used in the quarterly accounts for each accounting operation are estimated using statistical forecasting models based on business tendency surveys. Secondly, the basic accounting operations in the quarterly accounts are reconstructed using the forecast indicators: the quality of these reconstructions varies depending on the operations. Finally, the third step involves converting the accounting operations into GDP. Due to the simplification of the accounting framework used in forecasting compared to that used in the construction of quarterly accounts, this step is affected by a small but not negligible average aggregation error. The performance of this bottom-up approach is compared to that of direct approaches at different points during the quarter, using the same set of information. This comparison exercise is primarily educational in nature: in practice, the bottom-up approach used by INSEE for its Economic Outlook is more sophisticated than the one presented here, and is based on the use of business tendency surveys, a detailed analysis of multiple short-term economic indicators, and behavioural economic models.

At the beginning of the quarter, the direct approach performs slightly better than the bottom-up approach: few quantitative indicators are available at that time, and the added value of using a comprehensive accounting framework is limited compared to a direct model, which is naturally more parsimonious. From the end of the second month onwards, however, more quantitative indicators become available and the bottom-up approach performs slightly better than the direct approach estimated in this study. This is particularly the case when the Economic Outlook is published. Finally, the difference in performance between the two approaches is clear between the end of the quarter and the publication of the first estimates of the quarterly accounts thirty days later: during this period, it is considerably preferable to use the information available through a bottom-up approach rather than a direct approach.

It is also possible to study, over a large number of quarters, the changes in the growth forecast using a bottom-up approach throughout the quarter. In particular, this makes it possible to quantify the volatility of the growth forecast at different points during the quarter and to calculate the relative contributions of the available quantitative indicators and business tendency surveys to this volatility. At the beginning of the quarter, there is little volatility in the forecast because it is based mainly on business tendency surveys. This volatility increases over the course of the quarter as the available quantitative indicators are incorporated into the forecast. In particular, the industrial production index and turnover indices contribute to this volatility: on the one hand, these indicators are themselves volatile, and on the other hand, GDP growth is highly elastic to these indicators...