Jmcb Empirical Design

Empirical Design (jmcb-empirical-design)

When to trigger

• The dataset is assembled from Call Reports, Y-9C, supervisory, credit-register, or central-bank sources and the construction is under-documented
• The key variable (a "monetary shock," a "bank capital ratio," a "credit-supply" measure) is a constructed object whose definition matters for the result
• Sample period, window, or frequency choices are not motivated and could be driving the finding
• Restricted-access bank/central-bank data are involved and the access path is unstated
• A referee questioned whether the measurement, not the mechanism, produces the result

The JMCB measurement bar

JMCB carries a deep replication heritage — the journal's own 1980s–2000s Data Archive episodes (Dewald–Thursby–Anderson; the 2006 "Got Replicability?" audit) made it acutely aware that monetary/banking results often hinge on how series are spliced, deflated, and aligned. So referees scrutinize construction and timing: how a series is seasonally adjusted, how regulatory definitions changed mid-sample, how a bank merger reshapes a panel, and whether the announcement window for a monetary surprise is defensible. The standard is that a reader could rebuild the central variables from the description.

Construction craft by data type

Bank micro-data (Call Reports / Y-9C / credit registers)

• Identifier hygiene: track RSSD/entity IDs through mergers and acquisitions; state how you treat acquirers vs. targets so a merger does not masquerade as growth.
• Balance-sheet ratios: define capital, liquidity, and lending consistently across the sample; note Basel/regulatory regime changes that redefine the numerator or denominator mid-panel.
• Winsorize/trim extreme ratios and state the rule; report how many bank-quarters are dropped and why.

Monetary / macro series

• Real-time vs. revised data: for policy questions use the vintage the policymaker saw (ALFRED / real-time databases); say which and why.
• Frequency and alignment: state how high-frequency surprises are aggregated to the estimation frequency and how announcement timestamps map to observations.
• Splicing and deflation: document base years, deflators, and any series breaks (e.g., reserve-regime or reference-rate transitions).

Central-bank / supervisory / restricted data

• Access path: name the RDC / central-bank data room / register and the disclosure constraints; this scopes what the replication package can contain.
• Confidentiality: state aggregation/masking rules and how they affect inference.

Sample and specification hygiene

• Motivate the sample window from the institution/policy, not from where significance appears.
• Pre-specify the frequency (the local-projection horizon, the panel frequency) and show the headline survives nearby choices.
• Document missing-data and entry/exit handling so survivorship does not drive results.

The construction decisions referees probe most

A JMCB referee will mentally re-run your data build and ask where it could have gone wrong. The recurring pressure points:

• Seasonal adjustment and deflation. State the SA method and base year; an unstated SA choice can manufacture or erase a cyclical pattern.
• Reference-rate and regime transitions. LIBOR→SOFR, reserve-regime changes, and the move to ample reserves all break series; splice them explicitly and note the break date.
• Treatment timing. For a policy/regulation study, the exact effective date and any anticipation window matter; misdated treatment biases event studies.
• Aggregation level. Holding-company (Y-9C) vs. bank (Call Report) reporting answers different questions; pick the level that matches the mechanism and say why.
• Survivorship. Failed and acquired banks leaving the panel during a crisis is not random; show the result is not an artifact of who exits.

From measurement to a credible replication path

Measurement and reproducibility are the same discipline at JMCB. As you finalize the data build, write the construction in enough detail that the eventual replication package — or, for restricted data, the documented access path — lets someone rebuild the central variables. Note which inputs are public (Call Reports, Y-9C, FRED, ALFRED real-time vintages) and which require restricted access (supervisory panels, credit registers, RDC), since this determines what jmcb-internet-appendix can include. A measurement section that doubles as a reproduction recipe pre-empts the journal's signature replication concern.