A surprise z-score puts an economic surprise on one scale. Subtract the consensus forecast from the actual print, then divide by how much that indicator's surprises usually vary, which is its historical standard deviation. What comes out is how unusual the beat or miss is for that series, in a unit every release shares.
Raw surprises do not compare. Is payrolls missing by 60k worse than core CPI missing by 0.1pp? Run both through the same divisor and you can tell. Say NFP surprises have a standard deviation near 70k. A 60k miss is then about -0.9σ, worth a look and nothing more. If core CPI surprises have a standard deviation near 0.07pp, a 0.1pp miss is roughly -1.4σ, which is rare. The CPI miss is the bigger event, even though the number looks smaller.
A |z| under 1 is noise, and the market usually shrugs. Between 1 and 2 is a real surprise and usually a tradeable move. Anything above 2 is an outlier, and it can reprice the Fed path on its own.
Take retail sales. Consensus is +0.3% m/m and the print is +0.9%. Its surprises have a standard deviation of 0.4pp, so z = (0.9 - 0.3) / 0.4 = +1.5, and Helious marks it a strong beat. The 2-year cheapens 6bp as the market trims cut odds, about what a 1.5σ event is worth.
| DATE | SURPRISE | |
|---|---|---|
| Factory Orders (MoM): 0.1 | 0σ | |
| Average Weekly Hours: 34.4 | +1σ | |
| Average Hourly Earnings (YoY): 3.0% | -1.43σ | |
| Unemployment Rate: 4.2% | +0.77σ | |
| Nonfarm Payrolls: 29K | -0.85σ |
