Jfm Identification

Identification Strategy (jfm-identification)

When to trigger

• A causal claim about liquidity, spreads, depth, or price discovery rests on OLS + controls
• An event study uses a window so wide that confounding news or contemporaneous market-wide shocks contaminate it
• A market-structure change is exploited but the parallel-trends / no-anticipation logic is not argued
• Reverse causality is live: liquidity and the regressor (volume, volatility, ownership) are jointly determined
• A microstructure model is fit but it is unclear what moment in the data identifies the key parameter (PIN, lambda, adverse-selection share)

The JFM identification bar

JFM referees know that market-quality variables are endogenous to almost everything — volume, volatility, information arrival, and prices co-move mechanically. So the bar is high for any causal liquidity/price-impact claim, and the journal especially rewards designs built on exogenous changes in market structure. The credible JFM toolkit is design-based, not control-saturated.

Branch A: Market-structure natural experiments (the JFM sweet spot)

• Canonical shocks: decimalization, the SEC Tick-Size Pilot (2016-18), Reg NMS, MiFID/MiFID II, short-sale bans, circuit-breaker/LULD triggers, venue entry/exit, fee/rebate (maker-taker) changes, tick-size regime changes.
• Design: DiD across affected vs. unaffected stocks/venues; RDD at a price/market-cap threshold (e.g., tick-size eligibility); event study around a known implementation date.
• With staggered timing, move beyond TWFE (Callaway-Sant'Anna, Sun-Abraham, de Chaisemartin-D'Haultfœuille); show clean pre-trends in spreads/depth; cluster at the assignment level (stock/venue), not the observation.

Branch B: Intraday / high-frequency event studies

• Tight, pre-registered windows around a discrete event (an order-type launch, a latency upgrade, a halt). Justify the window from market mechanics, not from where the effect is biggest.
• Control for market-wide microstructure shocks (index moves, scheduled macro releases) and for the diurnal (U-shaped) intraday pattern in spreads/volume.

Branch C: Instruments for liquidity / order flow

• Honest about weak instruments: report first-stage F, use Anderson-Rubin / weak-IV-robust sets. Defend exclusion in market-mechanism terms (the instrument moves trading frictions only through the channel claimed).

Branch D: Structural microstructure models

• Tie each parameter to an identifying data feature: adverse-selection component from the permanent price impact / spread decomposition; PIN from the trade-imbalance distribution; Kyle's lambda from the price-impact regression. State estimator (MLE/GMM), and show the parameter is not just a fit artifact.

Referee pushback mapped to the identification fix

• "Liquidity and your regressor are jointly determined." → Replace the panel regression with an exogenous market-structure shock (DiD/RDD), or a defensible IV; show the result is not a mechanical co-movement with volume/volatility.
• "Your staggered TWFE is biased with heterogeneous effects." → Re-estimate with Callaway-Sant'Anna or Sun-Abraham; plot flat pre-event leads in spreads/depth.
• "The event window is cherry-picked." → Justify the window from market mechanics (settlement, implementation date), show robustness to nearby windows, and rule out contemporaneous macro releases.
• "What identifies PIN/lambda?" → Point to the data moment (trade-imbalance distribution; permanent price impact) and report estimation diagnostics, not just the point estimate.
• "Your control group is contaminated." → Show the control stocks/venues were not indirectly affected (e.g., order flow migrating from treated to control); report a clean, unaffected comparison or a spillover-robust design.

Separating mechanical from behavioral effects

A recurring identification subtlety in market-structure work is that a rule change has both a mechanical effect (a wider tick arithmetically widens the minimum quotable spread) and a behavioral effect (liquidity suppliers and informed traders re-optimize). A credible JFM design isolates the behavioral channel, because the mechanical one is not a finding. Show the effect on stocks where the tick does not bind, decompose the spread change into the binding-tick component and the residual, or condition on pre-period spread relative to the new tick. Conflating the two is a frequent reviewer catch.

Worked vignette: the tick-size pilot (illustrative)

The SEC Tick-Size Pilot widened the quoting/trading increment for a randomized set of small-cap stocks. This is close to an ideal JFM design: random treatment assignment, a discrete date, and a treated/control split. The clean identification statement is one sentence — the effect of a wider tick on depth is identified by the random assignment of stocks to the pilot's test groups. The credible version shows flat pre-pilot trends in depth, estimates with assignment-level clustering, and separates the mechanical (tick-binding) effect from the behavioral (liquidity-supply) response. A weak version regresses depth on a post-dummy with controls and calls it causal.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the identification claim, don't only argue 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.

1. detect_design → recommend → fit with as_handle=true → audit_result to list the checks the design still owes.
2. Staggered DiD: callaway_santanna / sun_abraham + bacon_decomposition + honest_did_from_result (the pre-trend test is low-power, Roth 2022).
3. IV: effective_f_test + an anderson_rubin_ci (valid under weak instruments), not a 2SLS t-stat alone.
4. RDD: rdrobust (bias-corrected) + rddensity / mccrary_test for manipulation.
5. OVB: oster_delta / sensemakr — how strong a confounder would have to be.

Report the economic magnitude; route the full battery to the appendix; keep every number reproducible. A run end-to-end (synthetic data, real returns) is in the JF execution walkthrough. If StatsPAI/Stata are not connected, adapt the vendored resources/code/ skeleton and flag any unverified number.

Checklist

• Branch chosen; the data-to-effect (or data-to-parameter) mapping stated in one sentence
• Endogeneity of the liquidity/flow variable explicitly addressed, not assumed away
• Market-structure design: clean pre-trends, modern staggered estimator, assignment-level clustering
• Intraday events: window justified from mechanics; diurnal pattern and market-wide shocks controlled
• IV: first-stage strength reported; exclusion defended in microstructure terms
• Structural: each parameter tied to an identifying moment; estimator and inference stated
• The causal claim never exceeds what the design supports

A catalog of clean market-structure shocks

Knowing the field's natural experiments speeds design. Commonly exploited exogenous changes, each with its own caveats: decimalization (2001) — tick size from sixteenths to pennies; the SEC Tick-Size Pilot (2016-18) — randomized, the cleanest assignment; Reg NMS (2007) — order protection and access fees; MiFID (2007) / MiFID II (2018) — European venue competition and transparency; short-sale bans (2008, and country-specific) — abrupt constraint changes; maker-taker / fee pilots — rebate structure; circuit breakers / LULD — discrete trading halts; index reconstitutions — forced, scheduled order flow; venue launches/closures and dark-pool entry. For each, the identification hinges on (a) whether assignment is plausibly exogenous to the stock's liquidity trajectory and (b) whether a clean control group exists. Argue both explicitly; a shock is not self-justifying.

Anti-patterns

• "Liquidity → outcome" from a panel regression with controls, called identification
• TWFE on a staggered tick-size / decimalization rollout with no heterogeneity-bias discussion
• An event window chosen to maximize significance rather than from market mechanics
• Ignoring the intraday U-shape so that time-of-day masquerades as the treatment effect
• Reporting a PIN/lambda estimate without saying what moment in the data moves it

Inference choices that travel with the design

Identification is not finished until inference matches the data structure. Microstructure panels are correlated in two dimensions — the same stock is autocorrelated over time and all stocks co-move on a given day — so single-clustered or plain OLS standard errors overstate precision. Default to two-way clustering by stock and by day; use Newey-West when the time-series autocorrelation is the dominant concern; use a wild-cluster bootstrap when the number of treated venues or events is small (few-cluster bias). For event studies, account for cross-sectional correlation in abnormal liquidity across the event window. State the choice and its rationale where the design is described, not as an afterthought — a referee reads the clustering as part of the identification claim.

Output format

【Journal】Journal of Financial Markets (JFM)
【Skill】jfm-identification
【Branch】market-structure NE / intraday event / IV / structural
【Data-to-effect mapping】one sentence
【Identifying variation】<shock / window / instrument / moment>
【Endogeneity handled】how liquidity/flow endogeneity is broken
【Inference】clustering level + (if needed) weak-IV-robust set
【What it does NOT identify】<…>
【Source status】verified URL / 待核实 / not asserted
【Next skill】jfm-empirical-design