When the figure beats its forecast

What counts as above or below

Every release with a consensus forecast on record is split into the prints that came in above that forecast and the prints that came in below it. Dates where the figure matched the forecast exactly are left out of both columns, and a side with fewer than eight observations is dropped rather than shown, because three out of four is not a pattern, it is a Tuesday. What is counted is where the Nasdaq 100 closed the half hour after the print, relative to the price at the moment of release. Not how far it travelled, and not where it was at the open. The half-hour close, and only that.

Why inflation lines up and jobs does not

CPI is the cleanest row on the page. When it printed above forecast the index finished the half hour lower 7 times out of 8; when it printed below, it finished higher 9 of 11. PPI runs the same way at 10 of 14 and 10 of 14. A hot inflation print has one meaning to the market and it is the same meaning every month: rates stay higher for longer, and the Nasdaq is the index most sensitive to that.

The jobs report is different and the table shows it. A soft print has been read as friendly fairly consistently, 11 of 14 higher. A strong print has not been read as unfriendly with anything like the same consistency, 10 of 21 lower, which is close to a coin flip. Strong jobs can mean a healthy economy or a Fed that will not cut, and the market has to decide which one it cares about that morning. That ambiguity is not a flaw in the data. It is the data.

What I do with it

On an inflation print I do not fade the first move against the surprise. The count says the half hour tends to finish on the side the surprise points to, and the first-minute page says most of that travel is already done before I could react anyway, so fading it means fighting both the direction and the clock. On the jobs report I have no directional opinion at all, and I stopped pretending to. The below-forecast side has a real tilt; the above-forecast side does not, and a position that needs a direction from a strong payrolls print is a position that needs a coin.

What it cannot tell you

A figure one tenth above forecast and a figure a full point above it land in the same column, so this cannot tell you whether the size of the surprise matters. It almost certainly does, and the record is not yet large enough to split it. The samples on each side are small: 8 on CPI's above side, 8 on GDP's below side. A count of seven out of eight is a count, not a probability, and one different month would make it six of nine. Nothing here says which way the next print goes, only how the last few dozen resolved.