qe-identification-strategy

Identification Strategy (qe-identification-strategy)

When to trigger

• A structural model's parameters are estimated but it is unclear what in the data identifies them
• An empirical causal claim rests on OLS + controls, or TWFE on staggered timing
• An experiment's estimand or its assumptions are not pinned down
• You are unsure the identification clears QE's quantitative, general-interest bar

The QE identification bar

QE is the Econometric Society's empirical/quantitative general-interest journal, so identification is judged through an ES lens: the mapping from data to the object of interest must be explicit and defended, whatever the method. Because QE spans empirical, structural/computational, experimental, and simulation work, "identification" means different things by branch — pick the branch and make the argument transparent. QE's house norms reinforce this: report standard errors and confidence/coverage sets (never significance asterisks), and make the strategy reproducible for the pre-acceptance ES Data Editor check.

Branch paths

Branch A: Structural / computational identification

• Name what identifies each parameter. Tie parameters to specific data features / moments; argue identification from the model's structure, not just "the estimator converged."
• Targeted vs. untargeted moments: report fit to targeted moments and show untargeted-moment validation as out-of-sample discipline.
• Sensitivity / informativeness: report parameter sensitivity to moments (e.g., a sensitivity matrix) so readers see which data move which parameters.
• Estimation regularity: state the objective (MLE / GMM / MSM / indirect inference), starting values, tolerances, and that the optimum is global enough (multi-start). Report Monte Carlo evidence that the procedure recovers known parameters.
• Counterfactual validity: argue the estimated parameters are policy-invariant enough for the counterfactual you run.

Branch B: Empirical causal design (applied micro / finance)

• DID / event study: with staggered adoption move beyond TWFE (Callaway–Sant'Anna, Sun–Abraham, de Chaisemartin–D'Haultfœuille); show a clean event-study with leads; report a Goodman-Bacon decomposition.
• IV: strong first stage; with weak instruments use Anderson–Rubin / weak-IV-robust sets; defend the exclusion restriction in theory, institutions, and falsification.
• RDD: McCrary / Cattaneo–Jansson–Ma density test; optimal bandwidth + robustness; covariate smoothness; bias-corrected CIs.
• Inference clustered at the assignment level; address few-cluster issues (wild-cluster bootstrap).

Branch C: Experimental

• Pre-registration in a recognized registry (AEA RCT Registry / AsPredicted / OSF) — required for own-data studies effective Jan 1, 2026; report deviations.
• Detailed instructions / survey transcripts included at initial submission.
• Randomization balance; attrition (Lee bounds if differential); multiple-hypothesis adjustment; explicit estimand and external-validity discussion.

Branch D: Simulation / measurement

• Documented data-generating process; seeds set and reported.
• Show the measured object is robust to grid/tuning choices and disciplined against measurement error and alternatives.

Checklist

• Branch chosen; the data-to-object mapping stated in one sentence
• Structural: each parameter tied to identifying moments; sensitivity + Monte Carlo recovery shown
• Empirical: design-appropriate diagnostics (pre-trends / density / first-stage / balance); modern estimator where TWFE would bias
• Experimental: pre-registered; instructions included; balance/attrition/MHT handled
• Inference reported as SEs / coverage sets (no asterisks); clustering/assignment level correct
• The claim never exceeds what the identification supports

Anti-patterns

• "The estimator converged" presented as if it were identification (structural)
• TWFE on staggered treatment with no heterogeneity-bias discussion (empirical)
• Calibrating parameters and running a counterfactual without arguing policy-invariance
• An experiment with no pre-registration or no reported estimand
• Reporting significance with asterisks instead of standard errors / coverage sets

Worked vignette: identifying a search-cost parameter (illustrative)

A labor-search model is estimated on matched employer–employee data. The referee asks what identifies the search-cost parameter. A weak answer points at the likelihood; a QE answer points at a data moment: the elasticity of the job-finding hazard to local vacancy density pins the cost, because a steeper hazard maps to a lower cost. Suppose the sensitivity matrix shows a 0.4 elasticity moves the estimate from 0.9 to 0.6 — that number makes identification visible. Pair it with Monte Carlo recovery (simulated panels return the true cost within 5%, illustrative).

Referee pushback mapped to the identification fix

• "Estimates are not credibly identified — calibration in disguise." → Show the sensitivity matrix and which moment moves which parameter; report untargeted fit.
• "Your counterfactual assumes policy-invariant parameters you never defend." → Argue invariance (Lucas critique); show parameters are not functions of the policy.
• "Staggered TWFE here is biased." → Re-estimate with Callaway–Sant'Anna or Sun–Abraham; show flat event-study leads.

Output format

【Branch】structural / empirical / experimental / simulation
【Data-to-object mapping】one sentence
【Identification evidence】[moments+sensitivity / pre-trends+density+first-stage / balance / DGP]
【Estimation/inference】objective + SEs/coverage (no asterisks); clustering if any
【What it does NOT identify】[...]
【Next step】qe-data-analysis