[nolank]

A statistical release is a best attempt at measuring some indicator using incomplete information. The longer one waits for data, the more complete the data is and the more accurate the indicator is. There is thus always a tradeoff between the speed of release and its accuracy. Sometimes, when a particular month's data is released, more complete information has significantly altered the calculation of the previous month's data and so the previous month's data is revised.

For example, if January's Non-Farm Payrolls (NFP) is 100,000, and February's comes out at 10,000 then this shows weakness in the US job market and is negative for the USD. Now, in March, NFP is 110,000 and this, by the same logic, is good news out of the US job market.

However, if at the same time as the March release, they revise February's data to 90,000, this shows that the reaction to what we thought was a much weaker reading in Feb was in actual fact, overdone, and the "recovery" from 10,000 to 110,000 in March is in fact a much smaller improvement from 90,000 to 110,000. There will be less reaction to the March release when you take into account the revision.