1 Imperfect Synthetic Controls

Citation. Powell [2026], "Imperfect synthetic controls," Journal of Applied Econometrics.

Question. How should we estimate and conduct inference when no perfect synthetic control a weighted donor average that matches the treated unit's untreated path exists, as is typical once outcomes contain non-vanishing transitory shocks?

Method. Builds a synthetic control for every unit, derives moment conditions valid in the presence of transitory noise (rather than fitting it), and introduces a weighting metric that asymptotically discards units lacking a valid synthetic match. The multiplicity of unit-level estimates underwrites inference even with few controls.

Result. Applied to the 2015 repeal of Wisconsin's 48-hour handgun-purchase waiting period and its effect on suicide, the estimator delivers calibrated inference that does not presume a perfect pre-treatment match, where classical SCM inference would be unreliable.

Takeaway. Excellent pre-treatment fit is not validity; honest synthetic-control inference must survive the realistic case where no perfect twin exists.

2 Extrapolating RDD with Comonotonicity

Citation. Kwon and Deaner [2025], "Extrapolation in regression discontinuity design using comonotonicity," arXiv:2507.00289.

Question. Can sharp RDD identify treatment effects away from the cutoff, escaping the design's notorious locality?

Method. Assumes the conditional treated and untreated potential-outcome surfaces are comonotonic in covariates covariate values associated with higher untreated outcomes are also associated with higher treated outcomes so the observed branch on each side of the cutoff identifies the missing counterfactual on the other. Estimation uses local linear regression with multiple covariates.

Result. Provides point identification and an estimator for treatment effects at running-variable values beyond the threshold under a single interpretable, order-preserving restriction.

Takeaway. A weak, interpretable monotonicity condition lets RDD speak to units away from the margin, upgrading internal validity into limited external validity.

3 Nonlinear DiD via Optimal Transport

Citation. Torous et al. [2024], "An optimal transport approach to estimating causal effects via nonlinear difference-in-differences," Journal of Causal Inference.

Question. How can the changes-in-changes idea a scale-invariant, distributional DiD- be extended to multivariate outcomes?

Method. Recognises the changes-in-changes quantile quantile map of Athey and Imbens [2006] as a one-dimensional optimal transport map, and replaces it with the multivariate optimal transport (Monge) map between control distributions, made computable via entropic regularisation.

Result. Identifies counterfactual distributions and distributional treatment effects for vector-valued outcomes under a common-transport-map generalisation of parallel trends, nesting univariate changes-in-changes as a special case.

Takeaway. Optimal transport gives distributional, scale-invariant DiD a tractable multivariate engine moving beyond effects on the mean of a single outcome.

4 Regression-Based Proximal Causal Inference

Citation. Ying et al. [2025], "Regression-based proximal causal inference," American Journal of Epidemiology.

Question. Can proximal causal inference identifying effects despite unmeasured confounding using negative-control proxies be made implementable with familiar regression tools?

Method. Recasts estimation of the outcome confounding bridge function as a sequence of regressions resembling two-stage least squares, lowering the technical barrier to using a negative control treatment and a negative control outcome, and extends the approach to right-censored survival data.

Result. Yields consistent treatment-effect estimates under unmeasured confounding using only standard regression software, with the proxy exclusion and completeness conditions doing the identifying work.

Takeaway. Two well-chosen proxies of a hidden confounder, plugged into a regression pipeline, can purge bias that no single covariate adjustment can.

5 Double Machine Learning for Panels with Fixed Effects

Citation. Clarke and Polselli [2026], "Double machine learning for static panel models with fixed effects," The Econometrics Journal.

Question. How can the Neyman-orthogonal, cross-fitted machinery of double machine learning be deployed in static panel models with unobserved unit heterogeneity (fixed effects)?

Method. Combines within/correlated-random-effects transformations with flexible machine-learning estimation of nonlinear nuisance functions of the covariates, preserving orthogonality and valid inference for the structural parameter under fixed effects.

Result. Delivers consistent, asymptotically normal estimates of treatment/structural effects in panels where covariates enter nuisance functions nonlinearly, outperforming linear fixed-effects specifications when that nonlinearity matters.

Takeaway. DML and panel fixed effects can be married, bringing flexible nuisance estimation to the workhorse setting of applied microeconometrics.

6 Synthesis

The thread running through these papers is the relaxation of assumptions that once seemed load-bearing: that a perfect synthetic match exists, that RDD speaks only at the cutoff, that DiD must be additive and univariate, that confounders must be measured, that panel nuisances must be linear. Each result trades an implausible convenience for a weaker, more defensible condition, and each pays for it with new estimation and inference machinery. Collectively they illustrate the maturing of causal inference from a catalogue of designs into a flexible engineering discipline.

References

1. Athey, S., and Imbens, G. W. (2006). Identification and inference in nonlinear difference-in-differences models. Econometrica, 74(2), 431-497.
2. Clarke, P. S., and Polselli, A. (2026). Double machine learning for static panel models with fixed effects. The Econometrics Journal, 29(1), 69-86.
3. Kwon, S., and Deaner, B. (2025). Extrapolation in regression discontinuity design using comonotonicity. arXiv preprint arXiv:2507.00289.
4. Powell, D. (2026). Imperfect synthetic controls. Journal of Applied Econometrics, forthcoming.
5. Torous, W., Gunsilius, F., and Rigollet, P. (2024). An optimal transport approach to estimating causal effects via nonlinear difference-in-differences. Journal of Causal Inference, 12(1), 20230004
6. Ying, A., Miao, W., Shi, X., and Tchetgen Tchetgen, E. J. (2025). Regression-based proximal causal inference. American Journal of Epidemiology, 194(7), 2030-2041.