Empirical Identification (aejmac-identification)

When to trigger

• The macro effect rests on a recursive (Cholesky) SVAR with no defense of the ordering
• A monetary/fiscal "shock" is plausibly anticipated or endogenous to the cycle
• Local projections are run but lag length, controls, and inference are ad hoc
• A narrative or high-frequency instrument is used but its exogeneity/relevance is unargued
• You are unsure the design clears AEJ: Macro's identified-empirical bar

The AEJ: Macro identification bar

AEJ: Macro publishes identified-empirical macro, so the mapping from data to the dynamic causal object (an impulse response, a multiplier, a pass-through) must be explicit and defended. The aggregate, time-series setting makes identification harder than in micro: few effective observations, anticipation, simultaneity, and structural breaks. State the shock you claim to identify, the assumption that delivers it, and the horizon and object you report. Report standard errors / confidence bands (the AEA house style; significance asterisks are conventional in AEA tables but the band/SE must carry the inference, not the stars).

Branch paths

Branch A: Structural VAR (SVAR)

• Recursive (Cholesky): defend the ordering as an economic timing assumption, not a default; show robustness to plausible reorderings.
• Sign restrictions: state the full set; acknowledge set-identification (report the identified set / median-target with a credible band, not a point as if point-identified); address the "multiple models" critique.
• Long-run / Blanchard–Quah: justify the long-run neutrality assumption.
• Proxy-VAR / external instruments (SVAR-IV): show instrument relevance (reliability/F) and defend exogeneity; report weak-instrument-robust bands where relevance is marginal.

Branch B: Local projections (LP)

• Report the horizon-by-horizon IRF with bands; state lag length and control set and show robustness to them.
• Use Newey–West / HAC or LP-specific inference; for panel LP cluster appropriately.
• Consider LP-IV when the shock needs instrumenting; report the first-stage strength.
• Address the LP-vs-VAR bias/variance trade-off explicitly if both are plausible.

Branch C: Narrative & high-frequency identification

• Narrative shocks (Romer–Romer style monetary/fiscal/tax): document the construction, the source record, and why the series is exogenous to the cycle; show it is unpredictable from macro history.
• High-frequency monetary surprises (event-window around announcements): defend the window, address the "Fed information effect" (orthogonalize against forecasts or use the information-robust instruments), report relevance.

Branch D: Micro-data macro / cross-sectional identification

• Cross-sectional or regional designs aggregated to a macro statement (e.g., regional multipliers): state the general-equilibrium vs. partial-equilibrium gap and how you map the cross-sectional elasticity to the aggregate.
• Use modern heterogeneity-robust estimators where staggered timing applies; cluster at the assignment level.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the identification claim, don't only argue it. Full map: execution-with-mcp. AEJ: Macro mixes empirical and structural work — local projections (local_projections / irf) are in StatsPAI, but DSGE / calibration estimation is outside this causal-inference toolchain.

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 shock and the data-to-IRF mapping stated in one sentence
• SVAR: ordering / sign set / long-run / proxy assumption defended, not defaulted
• LP: lag length, controls, HAC inference stated; robustness to them shown
• Narrative/HF: construction documented; exogeneity (unpredictability) demonstrated; relevance reported
• Cross-sectional-to-aggregate: PE-vs-GE gap addressed
• Inference: bands/SEs carry the conclusion; weak-instrument-robust where relevant
• The macro claim never exceeds the horizon/object the design identifies

Anti-patterns

• A Cholesky ordering presented as if it were innocuous, with no economic timing argument
• Sign-restricted IRFs reported as point estimates, hiding set-identification
• A "monetary shock" that is predictable from the prior quarter's data (anticipation not addressed)
• High-frequency surprises used without confronting the Fed information effect
• LP reported at a single cherry-picked horizon instead of the full response with bands
• Mapping a regional/cross-sectional elasticity straight to an aggregate multiplier with no GE caveat

Worked vignette: identifying a monetary shock (illustrative)

A paper estimates the output response to monetary policy via a recursive SVAR ordered output → prices → policy rate. A referee says the ordering is indefensible at high frequency. The AEJ: Macro fix: replace (or corroborate) the recursive shock with a high-frequency surprise from a tight window around FOMC announcements, purged of the information effect by orthogonalizing against Greenbook/SPF forecasts, then feed it as an external instrument in a proxy-VAR or as the shock in local projections. Suppose the peak output response stabilizes at -0.6% (90% band [-1.0, -0.2]) and is robust across the SVAR-IV and LP implementations — that cross-method agreement is the identification argument.

Output format

【Branch】SVAR / LP / narrative-HF / cross-sectional-macro
【Shock + data-to-IRF mapping】one sentence
【Identifying assumption】ordering / sign set / exogeneity / GE mapping
【Inference】bands/SEs; weak-IV-robust if relevant; HAC/cluster choice
【Object + horizon reported】...
【What it does NOT identify】...
【Next step】aejmac-robustness (then aejmac-theory-model if a model is matched to this)