Empirical Identification (aejmac-identification)

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.