spb/wp15_uqo

1# LIMITATIONS.md — What this dataset can and cannot support23**Date:** 2026-07-13 (Stage A). To be revised after full-sample audit.45## 1. OHLCV is not microstructure data67The dataset contains one-minute OHLCV bars only. It does **not** contain quotes, bid–ask spreads, order-book depth, order imbalance, trade direction, hidden liquidity, venue identifiers, or execution costs. Consequences:89- Liquidity is never observed; only **OHLCV-based liquidity proxies** (Amihud-type ratios, high–low range, zero-return/stale-price frequency, volume shocks) are available. All claims must be phrased as proxies.10- Price discovery mechanisms (quote updates vs trades) cannot be separated.11- Trade direction and order-flow imbalance are unobservable; signed-flow contagion channels cannot be tested directly.12- Effective spreads / transaction costs are not measurable; any economic-value exercise must use assumed, conservative cost schedules.1314## 2. The panel is trade-conditional1516Zero-volume bars do not exist in the data: a missing minute means "no reported trade in that minute." Therefore:1718- Missingness is informative (illiquidity), not random; imputation would destroy the object of interest.19- "No recorded bar" must be distinguished from "zero return" throughout (flag, don't fill).20- Sparse names have few bars per day (2025-12 cross-ticker median: 5.5 RTH bars/day); minute-level propagation analysis is only meaningful within a liquidity-screened universe, with breadth denominators restricted to *eligible* (actively trading) securities.2122## 3. Identity and corporate actions2324- Securities are identified by **ticker only**. No CUSIP/PERMNO. Ticker reuse and renames mean long horizons cannot be treated as one security without a spell map; conservative analyses run within continuous ticker episodes.25- Prices are **unadjusted** (BRK.A at 771k confirms raw levels). Splits/dividends are not flagged; split detection near rational ratios, exclusion windows, and robustness to close-to-close vs open-to-close returns are mandatory.26- Security type (common stock, ETF, ADR, preferred, warrant, fund, OTC) is not labeled; the 2025 file plainly includes large OTC/foreign coverage (23k tickers). External classification or suffix heuristics required; residual misclassification will remain and must be bounded via robustness (ETF/ADR exclusions).2728## 4. Coverage is time-varying by construction2930- Universe: ~8.3k tickers (1992) → ~8.1k (2009) → ~23.1k (2025); the 2020s expansion largely reflects OTC coverage, not market growth. Secular comparisons must hold the universe definition fixed (e.g., top-N by lagged dollar volume) or results will mechanically reflect coverage changes. This is itself a falsification test (CLAUDE.md §16).31- Extended-hours coverage grows over time (0% pre-market rows in 1992-06 vs 4.9% in 2025-12); RTH-only is the only decade-comparable baseline.32- 1990s bar density for small caps is low; whether this is source coverage or true sparsity is unresolved (external validation planned).3334## 5. Provenance and licensing3536- Upstream is Finnhub via a third-party repackager; no checksums against the original archives, no revision history for corrections, and **no license on the card**. Data errors of the source (bad prints, missing halts, venue scope) cannot be independently ruled out; cross-validation against reference daily data (volume, close) is planned for a sample of names.37- Whether bars are consolidated-tape or single-venue is unknown (open question; affects volume levels and off-hours prints).3839## 6. Inference consequences4041- One-minute returns of sparse names are contaminated by bid–ask bounce and stale prices; robust/quantile methods and liquidity screens are required.42- Common-information contamination (macro news moving everything at once) is the central identification threat; OHLCV alone cannot label news — scheduled-announcement windows and placebo designs stand in (CLAUDE.md §12).43- Nothing in the data supports direct causal claims about algorithmic/high-frequency trading; only association with market-structure eras, with explicit competing explanations.4445## 7. Additions from the 2026-10-17 review (v1.1)4647- **Ambient-activity selection.** Real ±6σ events occur in minutes whose cross-sectional breadth is48  40×10⁻⁴ (4 % of base rate) higher than at matched placebo minutes, although the market factor is49  calm at t. The ΔΔ removes the level but inherits mean reversion (negative pooled estimates at long50  horizons come from co-arrival minutes); the pre-level-stratified estimator is an upper bound51  (+19×10⁻⁴). The market-wide response is therefore bracketed, not pinned at zero.52- **Co-arrivals.** Only 22 % of events are alone in their minute (2.1 source events per event minute).53  The network differential survives in singleton minutes (+51 vs +64 ×10⁻⁴).54- **Neighbour identification.** Neighbour sets come from one-minute residual correlations in month55  m−1; with the 1996 Epps effect (mean correlation 0.004) they are a noisy selection of economic56  peers. The 1990s→2020s rise of the network differential combines transmission and measurement57  precision; neighbour sets on 5/30-minute or Hayashi–Yoshida correlations are required to separate58  them (not executed; raw data).59- **Placebo network contrast.** ≈0 on average but +200 (1990s), −41 (2000s), −44 (2010s), +68 (2020s)60  ×10⁻⁴: every network statistic must be a real-minus-placebo difference.61- **Tick regime.** 52 % of top-1000 one-minute returns are exactly zero in 1996 (13 % in 2024);62  MAD=0 securities are dropped; threshold-based series (extreme minutes, co-jump minutes, severity)63  are not comparable across 2001. The early 1990s show more "severe" minutes than 2008 under the same64  standardisation.65- **Universe composition 2020–21.** Dollar-volume ranking admits leveraged ETPs and distressed66  issuers; the |r|>5 % spike in 2020 is partly compositional; ETFs cannot be identified.67- **Not executed.** Lower source thresholds, higher receiver threshold, top-1000/3000 universes,68  permuted networks, announcement-window propagation, pre-event network contrast, venue69  reconciliation, formal jump/co-jump tests (see paper §8 and Online Appendix §D).70