Reddit Posts
Mentions
Thanks, all three are fair. I ran them as a locked follow-up: the rules, the power check and both our predictions were committed before any returns were computed. Short version: the result holds at the order level, and the ETF line was a split-sample story that I'm correcting. \*\*(1) Prints vs orders.\*\* You're right about the unit. In aggTrades, one taker order sweeping several price levels shows up as several prints. \- I rebuilt the ledger from the same 84 days. Prints with the same timestamp on the same side, back to back, were merged into one taker order. Then I re-ranked the top 1% by order size. \- The distortion is real: \- 57% of the selected orders consist of more than one print; \- only 68% of the originally selected prints fall inside a selected order. \- Merging by timestamp is an approximation. Before 2025 the files have millisecond stamps, so two separate takers in the same millisecond can merge (72% multi-print then). From 2025 the stamps are microseconds (45% multi-print). So I report the two eras separately. \- \*\*Result at the order level:\*\* pierced minus absorbed over 15 minutes is \*\*−1.54 bp, t −3.82\*\* (null 95%: 2.11). In the original it was −1.47 bp, t −3.59. \- Pierced: −1.03 bp. Absorbed: +0.16 bp. \- All three coins negative. \- Both timestamp eras negative: millisecond −1.92 (t −3.4), microsecond −1.16 (t −2.1). \- Power check: with no real effect it fires 5.5% of the time. A 0.5 bp give-back is caught 82% of the time. \*\*(2) Which horizon.\*\* 15 minutes was the pre-registered primary. 60 minutes was listed as descriptive in the locked file, but the post didn't say so; my mistake in the write-up. With a Bonferroni over both horizons, p goes from 0.0003 to 0.0007. Widening the Newey–West bandwidth from 2 to 8 days leaves t at −3.6 to −3.8. \*\*(3) The ETF line.\*\* You're right that "clear after the ETFs" was a split narrative. \- With an interaction term on the pooled day-level model, post-ETF minus pre-ETF is −0.38 bp, HAC t −0.37. There's no evidence of a regime difference. "Clear after" was just 65 days vs 19. \- I also checked for sign flips across half-years: 7 of 8 are negative in both ledgers, and the eighth is about zero (+0.06 / +0.2). \- Daily volatility doesn't moderate it either (t −0.8). So the card's sentence stands at the order level: after a big order pierces, the price gives a little back on average. It's 1–2 bp, far below fees, so it's a description and not a signal.