Business tendency surveys (BTS) carried out by the statistical institute INSEE are intensively used for the short-term forecasting of the French economic activity. In particular, the service BTS has been used together with the industry BTS for the short-term forecasting of GDP growth since Bouton and Erkel-Rousse (2003-2004) showed that the former survey contained a specific piece of information on GDP growth with respect to the latter survey. However, it remained to be demonstrated that this specific piece of information permits one to significantly improve the quality of short-term GDP forecasts with respect to models involving variables from the industry survey exclusively. More generally, the predictive accuracy of models based on the two surveys with respect to simpler autoregressive (AR) models deserved to be assessed. We, therefore, perform a real-time out-of-sample analysis that consists in estimating and, then, simulating miscellaneous kinds of models (VAR and univariate multistep models) aimed at the short-term forecasting of the quarterly GDP growth rate. Some BTS-based models encompass industry and service data, others either service or industry data exclusively. The predictive accuracy of these three kinds of models is compared to that of simple AR models; that of models including service and industry survey data is also compared to that of models based on data from one of the two surveys exclusively. Predictive accuracy tests (Harvey, Leybourne and Newbold, 1997, Clark-West, 2007) are performed up to four-quarter-forecast horizons. To assess the robustness of the results, we carry out both recursive and rolling estimations as well as three tests (differing by the method used to estimate the variance of the test statistics numerators) for each couple of competing forecasts. The results establish the usefulness of the two BTS, as well as the limited but significant contribution of the service survey in addition to the industry survey especially in the months when long enough service series are available (namely January, April, July, and October). The industry survey, nonetheless, appears to predominate over the service survey as a source of leading indicators of GDP growth.
