Why economic data gets revised
The number that moved the market on release is frequently not the number that ends up in the history books. This is a feature of how the data is made.
A quarterly GDP estimate published weeks after the quarter ends cannot be based on complete information, because the complete information does not exist yet. Tax records, company accounts and full survey returns arrive over the following months and years. The first estimate is an early read, and it is published early because users want it early.
The revision sequence
- Advance estimate. Published soonest, built from partial data and modelled assumptions for the gaps. The one that moves markets.
- Second and third estimates. More complete source data replaces the assumptions.
- Annual revisions. Incorporate more comprehensive sources, often tax-based.
- Benchmark revisions. Every few years, methods and classifications are updated and the entire historical series is recalculated.
It is entirely ordinary for a quarterly growth estimate to be revised by an amount comparable to the original estimate itself, and for employment figures to be revised by tens of thousands of jobs. The headline reaction on release day is therefore a reaction to a number with a wide and known margin of error, reported without one.
Where the uncertainty comes from
- Sampling. Most series come from surveys of a subset, carrying statistical error that is rarely quoted alongside the figure.
- Response timing. Firms that respond late are not a random subset of firms, and their data arrives after publication.
- Seasonal adjustment. Raw data is adjusted for predictable seasonal patterns using models estimated from history. When behaviour changes, the adjustment is estimated from a pattern that no longer holds, and it is re-estimated later.
- Business formation and closure. New and dying firms are hard to capture in real time and are modelled, then corrected against actual records much later.
The awkward implication
The models filling the gaps are calibrated on recent behaviour. When the economy changes direction, recent behaviour is precisely the wrong guide, and the errors are systematic rather than random — initial estimates tend to understate the change in both directions. Recessions have repeatedly been identified in the data only after they had been underway for months. See recessions.
How to read a release sensibly
- Treat a single month as noise. Three-month averages carry far more information than any individual print.
- Check whether the prior month was revised, and by how much. The revision is often more informative than the new figure and is rarely in the headline.
- Prefer series built from administrative records over survey-based ones where both exist.
- Note that a series being revised repeatedly is evidence of diligence, not incompetence.
What it means for markets
Prices move on release because traders are positioned against a consensus forecast, and the surprise relative to that forecast is what matters in the moment. That is a real and short-lived phenomenon. It is separate from the question of what the economy actually did, which is answered gradually over subsequent years and by then interests almost nobody.