A procedure for short-term GDP forecasting

Characterising the conjunctural situation of the economy and projecting its future performance are particularly important tasks for a central bank. In general, short- and medium-term macroeconomic projections take an analytical approach based on the use of the most recent conjunctural information and on a structural knowledge of the economy within the framework of the National Accounts. This article describes a tool for forecasting short-term GDP growth, which takes its place alongside others used internally by the Banco de España. There is a wide range of quantitative techniques for forecasting macroeconomic variables of interest, among which GDP is of particular importance, each with its distinct advantages and limitations. One way of classifying the various techniques available for forecasting this variable in the short term consists of distinguishing the direct approaches (those which use short-term indicators to yield a result in the form of a GDP projection) from the indirect approaches (those in which projections of the various demand and supply-side components of GDP are generated for subsequent aggregation). This article summarises the main features of BEST (Banco de España Short-Term forecasting model), a GDP direct forecasting procedure. Specifically, a wide range of indicators is used to estimate a similarly high number of multivariate vector auto-regressive models which include GDP and a series of indicators chosen according to statistical criteria. The results of these models are averaged to give a GDP projection. The predictive power of the model is assessed for the period from 2008 Q1 to 2014 Q2, a span dominated by the double-dip recession of the Spanish economy which posed significant challenges for the obtainment of macroeconomic projections. Following this brief introduction, the structure of the article is as follows. The second section enumerates the indicators forming part of the database used. Next, the modelling strategy used is described. The fourth section analyses the predictive quality of the proposed procedure by comparing the projections obtained from the BEST model with those yielded by a simple statistical model. The last section of the article presents the main conclusions.