How Much Should We Trust Staggered Difference-in-Differences Estimates?

Causal Question / Estimand

How reliable are staggered TWFE DiD estimates as used in accounting and corporate finance, and how often does the staggered-timing bias overturn published conclusions? The estimand of interest is the average treatment effect on the treated (Causal-Estimand).

Identification Strategy

Applies the Goodman-Bacon / de Chaisemartin diagnosis to the finance literature. Through the Bacon decomposition and Monte Carlo simulations calibrated to real corporate data, shows when generalized (staggered) DiD with TWFE is biased by already-treated-as-control comparisons under dynamic effects, and demonstrates how heterogeneity-robust estimators repair it.

Key Assumptions

Parallel-Trends plus, for TWFE validity, homogeneous/time-constant effects. Violations from Treatment-Effect-Heterogeneity generate Negative-Weighting. The robust alternatives instead rely on clean control cohorts and explicit aggregation. SUTVA throughout.

Threats to Validity

n/a — the paper audits a threat. It quantifies how staggered TWFE can mislead and shows that several influential published findings are sensitive to the estimator choice.

Setting / Data

Simulations calibrated to CRSP/Compustat-style corporate panels, plus replications of prominent accounting/finance DiD studies.

Key Claims

  • Staggered TWFE DiD is frequently biased in realistic finance settings; the sign and size of bias depend on the timing/effect-dynamics structure.
  • Heterogeneity-robust estimators (Callaway–Sant’Anna, Sun–Abraham, de Chaisemartin–D’Haultfœuille, stacked regression) materially change conclusions.
  • Practical guidance: diagnose with the Bacon decomposition, then adopt a robust estimator.

Connections

Citation

Baker, A. C., Larcker, D. F., & Wang, C. C. Y. (2022). How Much Should We Trust Staggered Difference-in-Differences Estimates? Journal of Financial Economics, 144(2), 370–395. https://doi.org/10.1016/j.jfineco.2022.01.004