Five things the data changed my mind about

I had opinions about all of these before I measured any of them. Writing them down is worth it because being wrong in a specific, checkable way is more useful than being vaguely right.

1. Payrolls is not the biggest mover

I would have said it was without hesitating. On the median half hour swing it moves 0.67% against CPI’s 0.71%, and its biggest single reaction of 1.58% falls well short of CPI’s 2.27%. On a typical day the two are close, and in the tail CPI wins clearly.

2. FOMC reverses far more often than most

Counting a flip as the direction right after the print not matching where price sat thirty minutes later: FOMC 32%, JOLTS 28%, and then everything else clusters between 12% and 19%.

I expected FOMC to be higher because of the press conference. I did not expect nearly double. And the persistence figures suggest some of what looks like a reversal is a move that simply has not finished, since the FOMC distance from the release price at two hours is 2.65 times the distance at thirty minutes.

3. JOLTS moves more before the print than after it

Every other release here is quieter in the hour before the number than in the thirty minutes after, usually by a lot. Payrolls is 3.2 times busier afterwards, CPI 3.3 times.

JOLTS is 0.8, meaning the hour leading up to it is busier than the half hour that follows. It is the only one that behaves this way and I do not have a confident explanation. My working guess is that whatever moves price at 10:00 on a JOLTS day is mostly not JOLTS, since other things land at the top of that hour. I have not proven it, and I am flagging it as the loose end it is.

4. The Nasdaq moves more than the S&P on every release

I assumed this was true on average and that some releases would go the other way. Ratio of the Nasdaq 100 median swing to the S&P 500 median swing over the same thirty minutes, each index over its own measured releases: PPI 1.45, core PCE 1.42, GDP 1.41, JOLTS 1.41, retail sales 1.41, payrolls 1.37, CPI 1.36, FOMC 1.20.

Eight out of eight, in a band between 1.20 and 1.45. Not one release where the S&P’s median was the larger, though on a handful of single days its range was the wider one. The tightest is FOMC, which fits: rates news reaches every part of the market, while a growth number lands harder on the longer duration index.

5. Shared minutes were inflating one of my numbers

GDP sometimes prints in the same minute as Core PCE (11 of its 35 releases), and retail sales sometimes prints with PPI. On those days I was recording a move that belonged to both releases under one release’s name, and nothing here can say how much of it each one caused. GDP looked about 1.49 times larger on its shared days than on the days it printed alone.

The asymmetry is the tell. GDP gains from sharing a minute with Core PCE, but Core PCE gains nothing from sharing with GDP, and PPI actually looks smaller on the days it shares with retail sales. It looks as if the quieter release borrows the louder one’s range. The cross-release figures on this site now leave shared-minute releases out (the release pages still count them, marked), and on those figures GDP is one of the two quietest things measured here.

None of that was a modelling error or a bad assumption. The sample itself was wrong, and I would have built conclusions on top of it and blamed the conclusions later when they stopped working.

The usual caveats

Samples run from 25 to 39 releases per release page, fewer on the cross-release pages. Prices are Nasdaq 100 and S&P 500 index CFDs on one minute bars, not the CME contracts. Everything above is a count of what has happened rather than a forecast of what will, and nothing here is investment advice.