intraday-microstructurelisted
Install: claude install-skill howard-lynn-ye/fin-skills
# Intraday microstructure
**Every microstructure estimator has a failure mode that returns a plausible number.** A trade
classifier that is 79% right looks fine until you learn it flips exactly the trades that moved the
price. So this skill scores each estimator on a synthetic stream where the answer was **planted**,
and reports the size of the failure as a number rather than as a warning.
✅ Every number below is printed by `scripts/microstructure_measures.py` (numpy 2.2.6, pandas
2.2.3, seed 20260908, about 20 s). It scores *estimators*, not markets: nothing here is a claim
about how real order flow behaves.
Getting tick data is `../market-data-sourcing/SKILL.md` (Databento MBO in
`../market-data-sourcing/references/databento.md`); joining quotes to trades without look-ahead
is `../market-data-engineering/SKILL.md` §2. This file starts where those stop.
## 1. Bars: activity time helps only when variance arrives with trades
Four bar types from one stream of 733,383 trades over 40 sessions, thresholds set so each type
yields about the same number of bars as 1-minute time bars (matched counts, or the comparison is
unfair). Two worlds, same price mechanism:
| World A - **94%** of daily variance arrives with trades | ex. kurt | \|z\|>3 | half-hour var ratio | bars/day |
|---|---|---|---|---|
| time 1-min | **7.13** | 1.794% | 2.4 | 388-390 |
| tick (47 trades) | 2.45 | 1.213% | 1.2 | 213-831 |
| volume (11,832 sh) | **1.87** | 0.911% | 1.3 | 209-824 |
| dollar ($1.25M) | **