R programs to analyze revisions to Canada's national accounts
Canada's monthly, quarterly and annual national accounts estimates undergo a steady stream of revisions as additional source data become available. The initial estimates are based on monthly survey and administrative detail that have minimal detail and substantial sampling and other error. They are also gaps where no source data are available, so imputations are needed. As time passes, additional source data become available from survey revisions due to late-coming responses, and from annual surveys and administrative data sources that typically are more detailed and more accurate. Are these revisions unbiased or do they tend to be upward or downward? How large are the revision variances? Are the revisions efficient in terms of bringing a better estimate quickly into focus without diverging first in a wrong direction? This project seeks to explore the properties of the historical revisions track record.