Definition
Narrative identification recovers structural macroeconomic shocks from the historical record — policy documents, newspaper accounts, official transcripts — rather than from statistical restrictions imposed on a reduced-form model's covariance matrix. The analyst reads the archive to date, or to measure the size of, episodes plausibly exogenous to current macroeconomic conditions, then uses that externally constructed series either as a direct shock measure or as an instrument. The canonical cases are the Romer-Romer monetary series, built from Federal Open Market Committee (FOMC) records, and the Ramey-Shapiro military-news dates for fiscal policy.
Key Ideas
- Exogeneity from institutional knowledge, not from ordering. Where recursive (Cholesky) and sign-restriction schemes buy identification with assumptions on contemporaneous responses, narrative identification buys it with a claim that the dated events were not themselves reactions to the variables in the system.
- Two uses of a narrative series. (i) As a direct shock: enter the constructed series in a VAR and read off impulse responses. (ii) As an instrument (proxy SVAR / external instruments): use the narrative series as an instrument for the reduced-form innovation, which tolerates measurement error in the narrative measure.
- Romer-Romer monetary shocks. Romer and Romer (1989) identify contractionary episodes from FOMC minutes — dates when the Fed deliberately accepted higher unemployment to reduce inflation; their later (2004) series purges the intended funds-rate change of the Fed's own Greenbook forecasts, isolating the component orthogonal to the central bank's information set.
- Ramey-Shapiro military dates. For fiscal shocks, large defense build-ups (Korea, Vietnam, Carter–Reagan, 9/11) are dated from news accounts; being driven by geopolitics rather than the business cycle, they proxy exogenous government-spending news and help resolve the fiscal foresight / anticipation problem that defeats purely statistical timing.
- Cross-method validation. A key use is corroboration: monetary-policy responses under narrative identification are qualitatively similar to those from recursive and non-recursive VARs, which strengthens the causal reading of all three.
How It Works
- Construct the series from the archive. Read the documentary record and code an exogenous event series — dummy dates (Romer-Romer 1989; Ramey-Shapiro) or a continuous, forecast-purged shock measure (Romer-Romer 2004).
- Purge endogenous response (optional but important). Regress the raw policy change on the policymaker's real-time information (forecasts) and keep the residual, so the measure reflects only shifts unexplained by systematic responses to expected conditions.
- Insert into the empirical model. Either add the series to a VAR and identify the shock as its innovation, or use it as an external instrument for a reduced-form residual in a proxy-SVAR / local-projection design.
- Trace dynamics. Compute impulse responses and compare with restriction-based identifications as a robustness check.
Why It Matters
- Immune to invertibility failures. Because the shock is measured outside the VAR, narrative identification sidesteps the non-invertibility ("missing information") problem that can bias VARs whose variables do not span the agents' information set — a central concern in the fiscal foresight debate.
- Sharpens contested effects. Narrative monetary shocks deliver cleaner, more persistent output and price responses and help dampen the price puzzle; narrative fiscal dates are the backbone of much of the government-spending-multiplier literature.
- A discipline device. Anchoring shocks to documented events forces identification assumptions to be defended with historical evidence rather than asserted through an ordering.
Open Questions
- Measurement and subjectivity. Coding the archive is judgmental; different readings of the same record yield different series, and classical measurement error attenuates direct-shock estimates (motivating the instrument approach).
- Few events, weak instruments. Narrative series are sparse — a handful of episodes — so they can be weak instruments with wide confidence sets, especially for fiscal build-ups.
- External validity across regimes. Episodes drawn from particular historical periods (e.g. wartime defense spending) may not identify the effect of the routine policy changes analysts actually care about.
Related