The authors build a model for predicting current-quarter real gross domestic product (GDP) growth using anywhere from zero to three months of indicators from that quarter. Their equation links quarterly Canadian GDP growth with monthly data on retail sales, housing starts, consumer confidence, total hours worked, and U.S. industrial production. The authors use time-series methods to forecast missing observations of the monthly indicators; this allows them to assess the performance of the method under various amounts of monthly information.

The authors' model forecasts GDP growth as early as the first month of the reference quarter, and its accuracy generally improves with incremental monthly data releases. The final forecast from the model, available five to six weeks before the release of the National Income and Expenditure Accounts, delivers improved accuracy relative to those of several macroeconomic models used for short-term forecasting of Canadian output. The implications of real-time versus pseudo-real-time forecasting are investigated, and the authors find that the choice between real-time and latest-available data affects the performance ranking among alternative models.