What I actually do with these numbers

I built this to settle arguments I was having with myself. What follows is not advice and it is not a system. It is what I changed about my own routine once the numbers stopped agreeing with my instincts.

1. I stopped using one stop size for “a data release”

The median half hour range on the Nasdaq 100 is 0.71% for CPI and 0.27% for GDP (the quietest release here is retail sales, at 0.26%). That is a factor of 2.6 between CPI and GDP, and for a long time I used the same stop for both. On GDP it sat far away from anything that was going to happen, which sounds safe and mostly meant sitting through noise for no reason. On CPI it sat inside the normal range, which is not a stop, it is a donation.

The figure that changed my sizing more than the medians is the gap between a release’s median and its worst day. Retail sales has a median of 0.26% and a biggest reaction of 1.18%, a ratio of more than four. CPI is large but consistent. Retail sales is nothing until it is not, and the two need different treatment even though the calendar prints them in the same font.

2. I use a different clock for each release

Share of the thirty minute range already covered after sixty seconds: CPI 53%, PPI 36%, payrolls 35%, FOMC 29%, GDP 21%, core PCE 17%, JOLTS 16%, retail sales 16%.

About half of a typical CPI half hour has happened before I have finished reading the number. That is the opening, not the whole day: the distance from the release price at two hours is typically 1.74 times the thirty-minute one. Any plan of mine that involved seeing the print and then deciding was, in practice, a plan to take the second half of the first move. JOLTS at 16% is a different animal entirely: it builds across the whole half hour, and there is time to be wrong and change my mind.

FOMC deserves its own note. Its median distance from the release price at thirty minutes is 0.20% and at two hours it is 0.53%. The half hour mark, which is where most people including me stopped looking, is closer to the middle of that release than the end of it.

3. I stopped reading the surprise as a direction

The reflex is that a hotter number means the index falls. Across this record (releases that shared their minute left out) it holds for CPI, where the index fell on 7 of the 8 occasions the print came in above consensus and rose on 9 of the 11 it came in below. PPI lines up the same way, 10 of 14 and 10 of 14, and JOLTS more loosely.

Past those it falls apart. Payrolls fell on 10 of 21 hot prints. Retail sales on 8 of 18. GDP on 6 of 13. Those are coin flips wearing a narrative.

My guess at why, and it is a guess: CPI is one number with one obvious reading, while payrolls is four numbers that can contradict each other inside the same release. The headline can beat while the revision or the wage figure says the opposite, and the market has to decide which one it cares about that morning.

4. I resized for the era, not just for the release

Splitting each release’s record in half by date and comparing the earlier median against the later one, seven of the eight releases here move the index less than they used to. Payrolls went from 0.78% to 0.58%. GDP from 0.43% to 0.27%. Only PPI went the other way.

This is the one I treat most carefully. Two halves of a few dozen observations is a coarse instrument and a single violent print near either end moves the ratio. I include it because the direction is consistent across seven of eight, not because any one ratio settles anything. It did make me stop trusting range expectations I had built in a louder period.

What none of this tells me

Which way it goes. Whether any of it survives costs. Whether the next print behaves like the last thirty. These are counts of what happened, on samples of 25 to 39 releases per release page (fewer on the cross-release pages), measured on index CFDs rather than the CME contracts. Nothing here is investment advice.