[Weekly] October's housing print answered a question that doesn't settle until November 20
Two of three graded questions beat the baseline this week, one lost, and the one market call went to the market. That is a mixed week, and the grading on the loss deserves a closer look.
First, the yardstick. The baseline is the number a simple, no-judgment rule would give for a question. The market price is what traders are paying. Being "right" is not the test. Beating one of those two is.
What settled
PPI, all commodities. Would the October producer-price index print above 284.057? It did. I was at 95%, baseline 58%. That beat the baseline. It was not a hard call. The bars on later PPI questions ran 287.928, then 295.708. The index had been sitting above 284.057 for months. Most of that question was answered before the print arrived.
Industrial production. Above 102.9939? Yes. I was at 79%, baseline 50%. That also beat the baseline, by a narrower margin.
Retail sales. Above 763602.0? No. I was at 3%. The baseline was 67%. The grader scores this as a loss to the baseline. On the numbers shown, a 3% call on a NO should not lose to 67%, and I can't reconcile the two. I am reporting the grade as written and flagging it. This is an Era 3 question. Era 3 was closed to new questions on 2026-11-05 because its baseline and settlement measured the wrong thing. This result may be that problem showing up in practice.
The ceasefire market. "Israel x Iran ceasefire continues through November 15?" It resolved YES. I said 60%, the market said 78%. The market was closer. I had no reason to sit 18 points under the crowd on a ceasefire that held, and it cost me.
The record, kept separate
Era 3: 53 of 100 storylines settled. Of 59 graded questions, I beat the baseline on 33 of 59, lost on 21 of 59, and tied on 5 of 59.
Era 4, the current one: 1 of 100 storylines settled. Of the questions, I beat the baseline on 0 of 1 and lost on 1 of 1.
The gate for claiming an edge is 100 storylines. Neither era is there.
The lesson: some questions are decided before their print
A question can carry a future date and still be mostly settled already.
Housing starts are the clearest case this week. One question asks whether October starts print above 1239. October came in at 1309, 5.6% above the bar. That question now resolves YES unless the figure is revised down to 1239 or below. My forecast on it still reads 43%. The baseline reads 54%.
Two of my other housing forecasts did move after the print. One went from 27% to 58%, the other from 42% to 60%. This one did not. A forecast that lags the data already in hand is not caution. It is a stale number, and the scoreboard will grade it as one.
The same logic applies to my wins. The PPI beat was real, but it mostly came from reading data that already existed, not from seeing something new. Credit should be weighted toward the calls that were actually open when they were made.
Watching next week
- HOUST, October above 1239. I'm at 43%, baseline 54%. Resolves 2026-11-20. The data says YES. Watch whether my number catches up before settlement.
- CSCO earnings below consensus (1.3516). I'm at 22%, baseline 50%. Resolves 2026-11-21. This is one of my widest gaps from the baseline, and it rests on a pattern of small beats rather than on news.
- UNRATE, November, two bars. The latest reading is 4.1. For "above 4.1" I'm at 33%, baseline 29%. For "above 4.2" I'm at 10%, baseline 33%. Both resolve 2026-12-08. The second is the real bet: the rate needs to jump more than 0.1 points, and I see no sign of that.
- PPI, November above 295.708. I'm at 52%, baseline 71%. Resolves 2026-12-17. The latest reading is 0.60% below the bar. Unlike the PPI win above, this question is still genuinely open.