Jae methods

Research Design & Identification for JAE (jae-methods)

When to trigger

• You have a prediction but no credible way to rule out endogeneity or reverse causality
• A reviewer will ask "is this causal or just correlation?"
• You must choose between an archival quasi-experiment and an analytical model
• Your treatment (a disclosure, a standard, a contract feature) is plausibly chosen, not random

JAE's dominant methodology

JAE's workhorse is large-sample empirical archival research grounded in economics — observational capital-markets and contracting data analyzed with econometric, identification-focused designs — alongside analytical economic modeling. The journal favors economic analyses of accounting problems (capital-markets information content, contracting, disclosure, agency/monitoring) in the Watts-Zimmerman positive-accounting tradition. It does not publish normative prescriptions, behavioral lab experiments, or design-science artifacts; design accordingly.

Design for identification

Because accounting choices and disclosures are endogenous, a bare panel regression rarely survives review. Match the design to the prediction:

Setting / claim	Identification strategy
A regulation/standard changes for some firms	Difference-in-differences around the shock; staggered DiD
A continuous threshold (covenant, index, size cut)	Regression discontinuity
Endogenous regressor, valid instrument available	IV / 2SLS; defend exclusion restriction explicitly
Self-selection into disclosure/treatment	Heckman selection; propensity-score matching
Information event (earnings, 8-K, disclosure)	Short-window event study (CARs), market-reaction design
Pure mechanism / equilibrium claim	Analytical model with assumptions, propositions, proofs

State the identifying assumption in words (parallel trends, exclusion restriction, continuity at the cutoff) and show how the design satisfies it. A natural experiment from a regulatory shock (SOX, Reg FD, IFRS/ASU adoption, an enforcement change) is the most persuasive JAE design when available.

Sample and measurement design

• Population and sample waterfall: define the population, the merges (Compustat-CRSP-I/B/E/S-Execucomp-DealScan-Audit Analytics via WRDS), and every exclusion, with counts.
• Construct measurement: justify proxies (discretionary accruals, accrual quality, conditional conservatism, bid-ask spread for information asymmetry) and pre-register the expected sign.
• Control set: include the economic determinants the theory implies; avoid "bad controls" that absorb the mechanism.

Analytical-model design

If the contribution is the model: state primitives and the information structure, solve for equilibrium, present comparative statics as testable propositions, and put proofs in an appendix. Keep assumptions economically interpretable.

Referee pre-mortem

Before locking the design, answer the three questions an economics-trained JAE referee asks of every archival accounting paper:

1. Who chose the treatment? If the firm chose it (a disclosure, a covenant, an accounting method), what makes the variation you exploit exogenous to the outcome?
2. What else moved at the same time? Regulatory shocks bundle provisions; show that the mechanism-specific margin responds, not merely anything measured post-shock.
3. Would the estimate survive the firm's best response? Anticipation, selection into treatment, and contract renegotiation can each reverse a naive estimate — state which applies to your setting and how the design absorbs it.

If any answer takes more than three sentences, redesign before drafting.

Execution bridge (StatsPAI / Stata MCP)

For the empirical / causal lane, estimate and audit rather than only specify. Full map: execution-with-mcp. JAE is empirical accounting with an economics lens; treat identification and weak-IV-robust inference as the binding constraints.

• detect_design → recommend → fit with as_handle=true → audit_result to enumerate the checks the design owes.
• Panel / staggered DiD: callaway_santanna / sun_abraham + bacon_decomposition
  • honest_did_from_result. IV: effective_f_test + anderson_rubin_ci. RDD: rdrobust + mccrary_test.
• Experiments: randomization-based inference and romano_wolf for the many-outcome family-wise correction reviewers expect.

Match the toolchain to the reviewer pool, and report the effect size the venue wants. A run end-to-end (synthetic data, real returns) is in the JF execution walkthrough.

Checklist

• Identification strategy named and matched to the prediction
• Identifying assumption stated in words and defended
• A shock / instrument / discontinuity is exploited where feasible
• Sample-construction waterfall with counts is specified
• Proxies justified; control set avoids bad controls
• Analytical models state assumptions, equilibrium, and comparative statics

Anti-patterns

• "Kitchen-sink" panel OLS presented as if causal.
• Instruments with no credible exclusion restriction.
• Ignoring self-selection into disclosure/treatment.
• Bad controls that mechanically absorb the effect of interest.
• A normative or lab-experiment design that JAE does not publish.

Output format

【Design】DiD / RD / IV / matching / event study / analytical model
【Identifying assumption】parallel trends / exclusion / continuity ...
【Shock or instrument】...
【Sample waterfall】population → merges → exclusions → final N
【Key proxies & expected signs】...
【Threats to identification】... and how addressed
【Next step】jae-data-analysis