JFM is the journal where measurement is the contribution as often as identification is. Insiders know each liquidity construct embeds assumptions; the design must name them.
| Object | Common measures | The trap to disclose |
|---|---|---|
| Spread | quoted, effective, realized; %/cents | effective vs. quoted matters when trades execute inside the quote |
| Depth / quantity | quoted depth, order-book imbalance, Kyle's lambda | depth at touch vs. deeper levels; venue-fragmented depth |
| Price impact | permanent vs. temporary; 5-min realized | horizon choice drives the adverse-selection share |
| Trade direction | Lee-Ready, tick rule, BVC | sign-classification error biases PIN/impact |
| Informed trading | PIN, VPIN, adverse-selection component | PIN estimation is numerically fragile; report it honestly |
| Daily proxies | Amihud illiquidity, Roll, Corwin-Schultz | proxies for high-frequency claims must be validated |
jfm-robustness.jfm-internet-appendix).A study of intraday adverse selection on consolidated TAQ: (1) restrict to common shares on primary listings, drop ADRs and ETFs; (2) keep regular-hours trades, drop the opening/closing auctions and the first/last few minutes; (3) remove trades with corrected/cancelled condition codes and crossed/locked quotes; (4) sign trades with Lee-Ready against the prevailing NBBO with a defensible quote-staleness rule; (5) compute effective spread = 2·|p − m| and the 5-minute permanent price impact as the adverse-selection component; (6) aggregate to stock-day, controlling for the intraday U-shape. Every dropped record class and every parameter (staleness, horizon) is logged in the codebook. The headline is then shown to survive switching the impact horizon and the sign-classification rule — the two choices a referee will challenge first.
Run the asset-pricing battery, don't just specify it. Full map:
execution-with-mcp. JFM is market microstructure and asset pricing — liquidity, price discovery, and cross-sectional return tests where the factor-zoo multiple-testing haircut is salient.
feols with the right SEs (Newey–West /
clustered) — read the alpha and t off the return, not off a memory of the spec.romano_wolf / benjamini_hochberg and report the alpha that survives the threshold.mcp__stata-mcp__stata_do with the vendored resources/code/ (asreg / xtfmb).etable; hand formatting to the tables/figures skill.Report the economic magnitude (bps/month alpha, Sharpe gain); the full factor grid and all screened signals go to the appendix. JF execution walkthrough.
Intraday spreads, depth, and volume follow a pronounced U-shape — wide and active at the open, tightest midday, widening into the close — and this pattern is strong enough to swamp many effects if ignored. Any design comparing periods, events, or stocks at different times of day must neutralize it: include time-of-bin fixed effects, compare within the same intraday interval across treatment/control, or normalize each observation by its time-of-day average. The classic failure is an event that happens to cluster at the open or close, whose "effect" is really the diurnal level. State explicitly how the U-shape is handled; a microstructure referee will assume it contaminates the result until shown otherwise.
Report the sample as a funnel a referee can audit: start from the raw universe, then show each screen and the count it removes (e.g., 9,800 stocks → drop ADRs/ETFs → 6,200 → require ≥250 trading days → 5,400 → drop penny stocks → 5,100). This single table answers the perennial "are the results driven by sample selection?" before it is asked. Pair it with the period, venue, and frequency. Hidden or undocumented filters are the most common silent driver of a fragile microstructure result and the easiest reject to avoid.
In many JFM papers the contribution is a better measure: a sharper decomposition of the spread into adverse-selection vs. inventory vs. order-processing components, a price-impact estimator robust to microstructure noise, an order-book-based liquidity index, or a way to sign trades when the NBBO is stale. If that is your paper, hold the new measure to a higher standard: show it recovers the right answer in a setting with a known benchmark, show it correlates sensibly with established measures while capturing something they miss, and show the downstream result is not mechanically induced by the construction. A new measure that is only validated by "it gives the result we wanted" will not survive review.
【Journal】Journal of Financial Markets (JFM)
【Skill】jfm-empirical-design
【Sample】universe / period / venue / asset class / key filters
【Liquidity measure】<spread/depth/impact + construction + assumption>
【Granularity check】matches the claim? <quote/trade/order-book vs. daily>
【Pipeline】trade-quote match + diurnal handling + lineage documented? [Y/N]
【Measure validation】alternative construct agrees? [Y/N → jfm-robustness]
【Source status】verified URL / 待核实 / not asserted
【Next skill】jfm-robustness